Application Status, Challenges, and Development Prospects of Smart Technologies in Home-Based Elder Care

Abstract

The rapid growth of China’s aging population has made elderly care a pressing social issue. Due to an imperfect pension system, limited uptake of institutional care, and uneven regional economic development, most elderly people in China still rely on home-based care. Elderly people living at home are usually cared for by their family, partners, caregivers, or themselves. However, this often fails to meet their complex health, safety, and emotional needs. Artificial intelligence may provide promising solutions to improve home care experiences and address the multifaceted health and lifestyle challenges faced by homebound elderly people. This review explores the applications of artificial intelligence in home-based care from four main perspectives: home health care, home safety and security, smart life assistants, and psychological care and emotional support. We systematically searched PubMed, IEEE Xplore, CNKI, and Scopus databases, integrated the latest research published between 2015 and 2024, focused on peer-reviewed, practice-oriented research, and reviewed relevant technology development paths and the current status of the field. Unlike previous studies that focused on physiological monitoring, this study is the first to systematically and comprehensively evaluate the role of artificial intelligence in improving the convenience of daily life and mental health support for elderly people at home. By comprehensively reviewing and analyzing the basic principles and application background of artificial intelligence technology in this field, we summarize the current technical and ethical challenges and propose future research directions. This study aims to help readers gain a deeper understanding of the current status and emerging trends of artificial intelligence-enabled home-based elderly care, thereby providing valuable insights for continued innovation and application in this rapidly developing field.

1. Introduction

Global aging has reached unprecedented rates, with population aging emerging as an irreversible worldwide phenomenon. In China, recent statistics from the National Bureau of Statistics reveal that individuals aged 60+ exceeded 310 million (22.0% of the population) while those aged 65+ surpassed 220 million (15.6%) in 2024 [1]. According to international aging classification standards, China has progressed from mild aging and now approaches moderate aging. World Bank projections indicate that by 2035, over 25% of China’s population will be aged 60+, with 20% exceeding 65 years, signifying the nation’s transition into a super-aged society [2].

Despite these demographic shifts, traditional home-based elderly care models exhibit significant deficiencies across four critical domains: health monitoring, home safety assurance, daily convenience, and psychological support. Conventional health monitoring primarily relies on periodic clinical examinations and family supervision, a system shown to inadequately manage multiple chronic conditions. Research demonstrates that the absence of real-time monitoring technologies increases medical resource utilization by 35–42% and escalates associated health care costs for elderly patients [3]. In home safety management, delayed emergency response mechanisms elevate risks during critical incidents such as falls, where intervention delays exceeding 30 min correlate with 68% higher complication rates. Daily living assistance proves equally challenging, with age-related cognitive decline and mobility impairments reducing task completion efficiency by 40–55%, while insufficient lighting contributes to 28% of nocturnal falls among homebound seniors. Psychological well-being remains particularly neglected, as evidenced by a national survey of 5000 elderly individuals where 63.2% reported loneliness or depressive symptoms, yet fewer than 18% received professional psychological interventions [4].

Recent advancements in artificial intelligence (AI) and Internet of Things (IoT) technologies present transformative opportunities to address these systemic gaps. While existing reviews of smart elderly care systems predominantly emphasize physiological health monitoring, there remains insufficient scholarly attention to psychological support enhancement and daily living optimization. For instance, Sarfraz et al. [5] and Ma et al. [6] systematically analyzed intelligent technologies across institutional and community care settings, focusing on chronic disease management and behavioral pattern recognition. Thanos et al. [7] explored IoT applications in elderly care infrastructure, whereas Kye et al. [8] investigated video surveillance systems for motor disorder assessment. This review specifically examines home-based care scenarios, analyzing 103 peer-reviewed studies through a rigorous multi-database search strategy encompassing Springer, ScienceDirect, IEEE Xplore, MDPI, and Web of Science. Different from previous reviews such as physiological monitoring, our study evaluates the role of artificial intelligence in improving the convenience of daily life and mental health support for elderly people at home, while systematically exploring the challenges in the implementation process and proposing future development paths.

2. Overview of This Survey

The convergence and innovation of AI and home-based elderly care offer promising solutions to address the multifaceted challenges associated with aging-in-place. To provide a structured understanding of current developments, a conceptual framework illustrating the major AI application areas in home-based elderly care is presented in Figure 1. This framework outlines four domains where AI technologies are being actively applied: home health care, home safety and security, smart life assistance, and psychological care and emotional support. These categories correspond to the core needs of elderly individuals living at home, including medical monitoring, safety, daily living support, and mental well-being. They also represent the key directions in which AI technologies are transforming the landscape of home-based elderly care.

Figure 2 provides an overview of the structure and content of this survey. The applications of AI in home-based elderly care have been systematically categorized into four domains. Each section surveys associated AI technologies, the challenges, and future development prospects.

Table 1. Representative publications in the survey.

Applications	Representative Publications	Main Content
Home health care	Non-Wearable IoT-Based Smart Ambient Behavior Observation System [9]	A non-wearable intelligent environmental behavior observation system based on the IOT, which achieves seamless monitoring of daily activities of the elderly through environmental and behavior sensors.
	Smart patient monitoring and recommendation using cloud analytics and deep learning [10]	A chronic disease health monitoring framework that can be deployed in the cloud and on premises.
Home safety and security	Internet of things and deep learning enabled elderly fall detection model for smart homecare [11]	A fall detection model for elderly people based on the Internet of Things and deep learning.
	Arduino Based Smart Home Warning System [12]	A smart home warning system based on Arduino, capable of detecting abnormal situations such as fires, gas leaks, and intrusions, and sending notifications.
Smart life assistants	Design and evaluation of a smart home voice interface for the elderly: acceptability and objection aspects [13]	Through user evaluation, it has been found that voice technology has great potential in improving the security and comfort of smart homes, but there may be certain privacy issues.
	Biofeedback method for human–computer interaction to improve elder caring: Eye-gaze tracking [14]	A biofeedback method based on eye tracking, which utilizes the CNN-SVM model to achieve real-time classification of user gaze direction and improve human-computer interaction in elderly care.
	A Novel Approach to Elderly Care Robotics Enhanced With Leveraging Gesture Recognition and Voice Assistance [15]	A new type of elderly care robot aims to improve the safety, health, and quality of life of the elderly by integrating gesture recognition and voice assistance functions.
	AI-Enabled Elderly Care Robot [16]	An artificial intelligence-based elderly care robot that uses convolutional neural networks to recognize objects and people, helping elderly people with dementia recognize daily items and family and friends.
Psychological care and emotional support	Deep Learning Approaches for Speech Emotion Recognition Using CNNs and Self-Supervised Models [17]	The method of using convolutional neural networks and self-supervised models for speech emotion recognition achieved a recognition accuracy of over 75% on the dataset.
	IntelliJoyCare: A Realistic Interactive Audiovisual System for AI-based Elderly Care Companionship [18]	A smart application called IntelliJoyCare combines artificial intelligence, voice therapy, virtual portrait technology, and more to provide emotional support and mental health care for the elderly.
	Supporting Creativity in Aged Care: Lessons from Group Singing, Music Therapy, and Immersive Virtual Reality Programs [19]	The article explores the importance of creativity in elderly care, emphasizing the use of group singing, music therapy, and immersive virtual reality activities to meet the psychological and social needs of the elderly.

lists representative publications and introduces their main content.

2.1. Home Health Care

The complexity and variability of the elderly’s health conditions, coupled with high rates of chronic diseases, mobility impairments, and cognitive disabilities, underscore the urgency of real-time health status monitoring. This subsection examines AI-integrated home care applications for the elderly, reviewing IoT-based systems for health data collection and analysis. We summarize current challenges and project future trends in these domains. Figure 3 shows the typical technical architecture of an intelligent health monitoring system. Wearable devices, environmental sensors, and cameras continuously capture physiological and behavioral metrics from older adults. These data are transmitted via secure networks to a cloud-deployed health analysis platform, which integrates advanced visualization techniques and artificial intelligence models to process and interpret incoming signals. The system provides clinicians and caregivers with real-time dashboards of the individual’s health status. In the event of any detected anomalies—such as aberrant vital signs or unusual activity patterns—the platform immediately issues alerts to designated medical personnel and family members, thereby facilitating timely intervention.

2.1.1. Data Collection

The elderly often suffer multiple comorbidities, including diabetes, hypertension, sleep disorders, and mobility impairments. Multiple IoT devices can monitor these comorbidities by collecting data on blood sugar, blood pressure, sleep quality, and daily behavior of elderly people. Chang et al. [20] designed a non-invasive watch equipped with a flexible sensor patch that can continuously sample interstitial fluid to monitor blood glucose data in elderly individuals. Li et al. [21] introduced a fiber-optic wearable wristband that captures precise pulse waveforms and derives pulse transit time to estimate blood pressure in real time. Das et al. [22] developed a comprehensive system for assessing psychological and cognitive states. Their platform records brain activity through electroencephalography, tracks the full body movements of elderly people with motion sensors, and recognizes gestures and facial expressions. Afterwards, transfer-learning convolutional neural network (CNN) [23] was used to classify the psychological cognitive states of the elderly. In the sleep domain, Kim et al. [24] proposed a non-invasive monitoring setup that uses environmental sensors to gather sleep data. They designed a feature-extraction algorithm and validated its performance—on expert-labeled datasets—achieving over 90% on the overall performance index. Moreover, Salvade et al. [25] presented a sleep-environment monitoring system that records ambient noise, light levels, and room temperature, uploads these metrics to the cloud, and processes them on a remote analysis platform.

Home-deployed sensors offer an effective means to capture the elderly’s daily behaviors. Irfan et al. [9] introduced SABOS, an IoT-based Smart Ambient Behavioral Observation System. SABOS gathers data on daily activities, appliance usage, and environmental comfort of elderly people, then transmits it to a cloud platform for visualization and analysis. In order to improve data transmission efficiency, they designed a data reduction algorithm that compresses the transmitted data by over 90% while allowing for complete reconstruction of the data in the cloud. Dahou et al. [26] advanced a Human Activity Recognition (HAR) model by combining CNN-based feature extraction with the Binary Arithmetic Optimization Algorithm (BAOA)—a binary variant of the Arithmetic Optimization Algorithm [27]—to select salient features. Using Support Vector Machines (SVMs) [28] to classify these features, the classification accuracy exceeds 95% on three benchmark datasets. Bibbo et al. [29] developed a behavior recognition system that uses motion sensors to collect motion data from elderly people on mobile phones and determines their location through ultrasonic localization technology. Finally, CNN is used to classify the activities of the elderly. Furthermore, Shanmugathashan et al. [30] proposed another framework that applies K-Means [31] clustering to sensor data, uses SVMs to flag abnormal behaviors of the elderly, and employs CNN to detect facial positions in image streams; the system automatically alerts caregivers when assistance is needed of the elderly people. Finally, Lingmei et al. [32] presented a smart care solution that integrates trilateral wearable sensor localization with video monitoring and leverages a Long Short-Term Memory (LSTM) model [33] to predict the elderly’s daily activities.

2.1.2. Health Analysis Systems

The health analysis system processes data collected by IoT devices, including elderly individuals’ body data and environmental data. It monitors their physical condition in real time and predicts potential health issues. The system promptly informs guardians and doctors of these insights, ensuring the elderly receive timely medical interventions. Zhang et al. [34] proposed a health monitoring framework for the elderly. It includes data acquisition, signal transmission, remote interaction, and diagnostic functions. The framework uses machine learning to assess heart disease risk. Experiments show the diagnostic accuracy exceeds 95%. Yazici et al. [35] proposed a smart e-health framework that monitors the health of elderly patients and chronically ill patients. The framework collects real-time data from inertial sensors, ECGs, and video sensors. Edge computing processes the data using Random Forest (RF) [36] and CNN for HAR and detection of cardiac irregularities. Huang et al. [37] proposed a smart service demand prediction Neural Network Boost Model (NNB), deployed on a cloud platform. The model deeply mines data on elderly users’ physical condition, lifestyle habits, and residential environment safety to accurately predict and quickly respond to potential health risk events. NNB is based on artificial neural networks and integrates the AdaBoost algorithm [38] to prevent overfitting in individual neural network models. NNB models are trained and deployed in cloud platforms. Sundas et al. [10] proposed the Smart Patient Monitoring and Recommendation (SPMR) framework. SPMR continuously monitors and predicts the true health status of the elderly using their physical data, collected via ambient assisted living devices. The framework processes data using deep learning models like CNN and Recurrent Neural Network (RNN) [39], optimizing predictions with categorical cross-entropy. SPMR operates on the cloud and includes a local intelligent processing layer. This layer enables real-time monitoring and emergency treatment via local models, even during cloud-service interruptions or service unavailability.

Hassen et al. [40] proposed an IoT-, fog-, and cloud-based home hospitalization system. The system allows patients to undergo rehabilitation and treatment at home. Vital-sign-sensing units monitor patient health, while environmental sensing units track room conditions. Doctors can access these data to monitor patient health and provide advice to patients and their families. Iranpak et al. [41] proposed a remote patient monitoring and classification system based on IoT and cloud computing. The system uses sensors to collect vital-sign data, such as heart rate, temperature, and blood pressure, and transmits them to a cloud platform. An LSTM deep neural network classifies and monitors patient health status. If the patient is in critical condition, the system immediately alerts the patient, doctors, and family. This simulation result and comparison with other methods showed a health-condition categorization accuracy of 97.13%, an average improvement of 10.41% over other approaches. Liyakathunisa et al. [42] proposed an ambient-assisted living system based on the Internet of Medical Things. The system uses sensors to collect physiological data from the elderly and employs Bidirectional GRU (BiGRU), a bidirectional variant of the Gated Recurrent Unit (GRU) [43], for feature selection. The proposed method achieved an accuracy of 98.14% on private datasets and 99.26% on publicly available ambient-assisted living datasets. Wu et al. [44] proposed FedHome, a cloud-edge federated learning framework for family health monitoring. FedHome preserves privacy by keeping user data local. It aggregates data from multiple families to train a global model, then transfers knowledge to achieve personalized model learning without compromising user privacy. Wu et al. also designed a generative convolutional autoencoder. This autoencoder generates class-balanced datasets from personal user data. This enhances the model for accurate, personalized health monitoring.

The preceding sections systematically review two facets of elderly home-based health care. We select several representative studies to show their technical implementations. The fiber-optic wristband device described in [21] continuously emits light toward the skin, detecting minute cutaneous vibrations induced by arterial pulses via the fiber-optic end face. By analyzing phase shifts in the reflected light, it reconstructs high-fidelity pulse waveforms and derives pulse timing to inform a blood pressure estimation model. Irfan M. et al. [9] introduces an elderly activity monitoring system that gathers quotidian behavioral data using non-imaging sensors—such as touch, temperature, humidity, current, and motion detectors—and applies redundant filtering alongside periodic storage techniques for lossless data compression, thereby minimizing transmission overhead while ensuring full recoverability of raw signals. These compressed data are relayed to a cloud platform, which generates visual analytics to map daily activity trajectories and appliance usage patterns. In the event of anomalous readings, the system automatically dispatches multi-channel alerts. The home-based care framework proposed in [34] is structured as a three-tier IoT–intelligent terminal–cloud architecture: the IoT layer integrates wearable biosensors (e.g., for blood pressure and heart rate monitoring), FMCW millimeter-wave radar for non-contact respiratory detection, and cameras for video communication. The intelligent terminal layer encompasses data storage and a main control processor to facilitate local decision-making and emergency response. The cloud platform supports multi-device connectivity, visual displays of health metrics, and the deployment of machine learning algorithms to assess cardiovascular risk. Huang X et al. [37] presents an intelligent elderly care system built on a cloud platform employing an NNB algorithm. Multidimensional data are captured via IoT devices and processed in-depth on the cloud, where AdaBoost-enhanced artificial neural networks—each serving as a “weak learner”—are iteratively trained with dynamic reweighting of challenging samples and subsequently aggregated into a robust ensemble model. Leveraging the nonlinear representational power of ANNs, this approach uncovers latent demand patterns within elderly physiological and lifestyle datasets to generate personalized service recommendations, which are delivered through mobile applications and smart speakers.

Table 2. Data used in the studies for home health care.

Publications	Physiological Data	Physical Activity (Video)	Physical Activity (Non-Video)	Position	Environmental Data	Household Appliance Data	Lifestyle Habits	Medical History and Symptom Data
[20]	✔
[21]	✔
[22]	✔	✔
[24]			✔		✔
[25]					✔
[9]			✔	✔	✔
[26]		✔
[29]			✔	✔	✔
[30]		✔
[32]		✔		✔
[34]	✔	✔	✔	✔	✔
[35]	✔	✔	✔
[37]	✔	✔	✔	✔	✔	✔	✔	✔
[10]		✔	✔	✔	✔			✔
[40]	✔				✔			✔
[41]	✔		✔		✔
[42]	✔	✔	✔
[44]	✔		✔

summarizes the data types utilized in the study described in home health care for the elderly.

2.1.3. Challenges: Data Reliability Issues and Performance of Health Analysis Systems

The collection of health data from the elderly primarily occurs non-invasively through IoT devices equipped with multiple sensors. However, the accuracy of data collection is often compromised due to various factors. Additionally, health analytics systems that process these data face challenges related to insufficient analytical capabilities.

(1) Reliability of data

The quality, quantity, and other parameters of collected elderly individuals’ physical data heavily depend on IoT device performance. Daily activities of the elderly and environmental changes can interfere with data collection. Raw data require preprocessing, often using machine learning or deep learning models, which increases processing time and places demands on device performance. Additionally, obtaining high-quality data annotation necessitates specialized knowledge [45]. Collecting physical data from older adults faces challenges of balancing device accuracy with portability and ease of use.

(2) Inadequate performance of the health analysis system

IoT devices collect various data types, such as elderly individuals’ physical parameters, daily-activity trajectories, and home environment data. Differences in data formats and sampling frequencies across devices pose challenges. Sundas et al. [10] highlights underutilization of existing data and model overfitting in health analysis systems. Achieving multi-modal analysis through synchronization and feature alignment of multi-source data remains a challenge [46]. Additionally, health analytics systems deployed in cloud environments must ensure secure transfer of sensitive data, such as IoT-collected data and analytical outputs, between home devices and the cloud servers.

2.1.4. Future Prospects: Convenient Data Collection and Multifunctional Home Health

Non-invasive biomarker-based data collection devices are expected to advance rapidly. These devices will monitor physiological changes in the elderly and track the progression of chronic conditions such as hypertension, hyperlipidemia, diabetes, cardiovascular disease, and osteoporosis. They will achieve this by analyzing biomarkers found in sweat, tears, urine, and saliva [47].

The health analysis system will evolve into a comprehensive home health system. Beyond real-time analysis of elderly individuals’ physical data, the system will integrate with smart homes, home robots, and wearable assistive therapy devices to support daily living and chronic disease rehabilitation. For example, based on health analysis results, intelligent robots can assist the elderly in preparing healthy meals. Meanwhile, flexible exoskeleton-based wearable devices can aid in the rehabilitation of stroke and muscle atrophy rehabilitation.

2.2. Home Safety and Security

In the home care scenario, intelligent technology applications focus on both real-time monitoring of elderly individuals’ physical data and managing emergencies and ensuring daily safety. Security risks for elderly people living at home include personal accidents and domestic safety hazards. Personal accidents often involve falls, while home risks encompass unauthorized entry, fire, scam calls, and privacy breaches. This chapter outlines the technology applications for fall detection and home risk protection, while analyzing challenges and future trends in these areas. Figure 4 illustrates a representative technical architecture for safety and security in home-based elder care. In home-based elder care settings, security incidents—including falls, fires, unauthorized entry, cyberattacks, and internet fraud—are continuously recorded by distributed sensors cameras and mobile applications. These data streams are transmitted over secure networks to AI-driven analytical platforms, which automatically classify the nature of each event and determine the optimal response protocol. Depending on the identified threat, the system can dispatch alerts to emergency medical services, law enforcement, or fire departments, or activate cybersecurity countermeasures—such as filtering fraudulent communications or isolating compromised devices—to rapidly mitigate risk and protect older adults.

2.2.1. Fall Recognition

Falls are the most critical accidental situation in the lives of the elderly. Factors such as acute stroke, sudden heart attack, epilepsy, and mobility issues can lead to falls. As a result, the recognition and processing of accidental falls in the elderly have become a key focus of research in home intelligence technology. Vaiyapuri et al. [11] proposed the IMEFD-ODCNN, an IoT- and deep learning-based elderly fall detection model. The model detects falls using video data from IoT devices. It employs the SqueezeNet model [48] for feature extraction and optimizes hyperparameters via the SSO algorithm [49]. When a fall is detected, the model sends alerts to caregivers and hospitals through mobile devices. The IMEFD-ODCNN model achieves a maximum accuracy exceeding 99% on fall detection datasets. Villegas-Ch et al. [50] developed an easy-to-deploy fall detection model using a computer vision approach. The system uses a camera to monitor elderly individuals’ daily activities and employs a neural network to analyze images in real time. It classifies postures and identifies falls, achieving an average accuracy of 94.8%. The model’s key advantages include low cost and ease of deployment. Papan et al. [51] first utilized the You Only Look Once (YOLOv8) [52] model for elderly fall detection. They designed a cloud-based intelligent fall detection and alert system. The system gathers video and behavioral data of the elderly via IoT devices. Data are analyzed using the YOLOv8 model, which activates an alert mechanism upon detecting a fall. Yao et al. [53] combined Unmanned Aerial Vehicle (UAV) mobility with the Dlib HOG algorithm [54] and intelligent fall-posture analysis. This approach aims for real-time outdoor elderly behavior tracking and reducing risks for the elderly living alone. The system uses UAV intelligent tracking. Fall detection integrates Dlib face detection with the OpenPose deep learning algorithm [55]. It determines fall direction based on the care recipient’s posture and records fall details, including direction, timestamp, severity, and real-time images.

2.2.2. Home Security

In daily home life, the elderly face various security risks, including personal safety, fire hazards, and digital security, in addition to physical emergencies. Taiwo et al. [56] proposed a smart home automation system capable of monitoring human behavior patterns and distinguishing between intruders and occupants. The system also controls home appliances and monitors environmental factors. An improved CNN deep learning model is used for classifying and detecting home intruders, achieving 99.8% accuracy in tests for distinguishing family members from intruders. Akhmetzhanov et al. [57] developed a smart home control system. It features automatic lighting control and voice-command functionality. The system also monitors elderly individuals’ physical data. Additionally, it includes an integrated alarm system that automatically triggers alerts during emergencies such as intrusion or fire. Sarhan et al. [12] proposed a smart home alarm system based on sensors and controllers. The system detects fire, gas leaks, and intrusions, and notifies guardians of emergencies. Additionally, it can mitigate disaster effects by actions such as activating sprinklers and reducing harmful gas concentrations through ventilation.

The elderly face significant digital security risks, including phone fraud, privacy data breaches, and unauthorized access to smart home devices. Malhotra et al. [58] introduced a novel method for detecting fraudulent calls using SVM and RNN techniques for speech processing and fraud classification. Evaluated on a dataset of 1,000 real and fraudulent calls, this method achieved 95% accuracy and 97% precision, surpassing existing approaches. Elahi et al. [59] proposed an AI approach to protect the privacy of applications for elderly users. The study introduced two algorithms: participatory privacy protection Algorithm 1 and Algorithm 2. Algorithm 1 determines optimal privacy settings for an app, while Algorithm 2 manages runtime permission requests. These algorithms significantly enhance privacy protection. The method reduces the cognitive burden on older users during complex decision-making and improves user experience by automating privacy management. Bhatia M. [60] proposed a security framework for monitoring physical activities of elderly people, which collects behavioral data of elderly people through IoT devices and analyzes their behavior using CNNs. The framework protects elderly health care data through Digital Twin (DT) technology [61] and blockchain. DT technology creates a digital copy of the patient to enable the real-time monitoring, anomaly detection, and prediction of patients’ health status while ensuring data security and privacy protection. The decentralized and tamper-proof nature of blockchain ensures the security of health data of the elderly in transmission and storage. Liu et al. [62] proposed a personal information protection system. This system authenticates and protects personal information through a trusted user agent. Personal information is stored and managed solely by this agent and can only be verified by authorized information verification services after user approval. The scheme safeguards sensitive details like identity information and financial account passwords from leakage. Yu et al. [63] designed and implemented a smart home security analysis system based on home routers, aiming at effectively detecting and defending smart homes from possible contactless attacks. Experiments show that the smart home security analysis system can effectively detect and defend against attacks and can use plugins to mask possible future vulnerabilities.

The following paragraphs describe several representative studies on the safety and security of elderly households, with a focus on their technical implementations. Vaiyapuri T. et al. [11] proposed the IMEFD-ODCNN model for elderly fall detection, which consists of four modules: data preprocessing, feature extraction, hyperparameter optimization, and classification. The data preprocessing module standardizes video-frame dimensions, performs data augmentation, and applies normalization. Feature extraction is accomplished via SqueezeNet—a lightweight convolutional neural network—comprising an input layer, an initial convolutional layer, multiple hidden modules, and an output layer to derive salient features. Hyperparameter optimization leverages the Sparrow Search Optimization (SSO) [64] algorithm, which emulates the foraging behavior of sparrow populations and employs a leader–follower mechanism to fine-tune network parameters. Finally, classification is performed by an SSO-enhanced Variational Autoencoder (VAE) [65]. The VAE learns the latent distribution of input features and generates robust representations, while the SSO algorithm iteratively optimizes the encoder and decoder parameters to maximize classification accuracy. Taiwo O. et al. [56] proposes an enhanced smart home security system that integrates deep learning for intruder detection. A motion sensor initially detects anomalous movement. Upon activation, a camera captures video frames, which undergo preprocessing (including resizing, denoising, and normalization). A convolutional neural network then classifies human activity patterns—such as walking or jumping—to differentiate residents from a potential intruder. The system further combines IoT hardware with a cloud-based platform to enable remote home control and environmental monitoring. With respect to information security, Elahi H. [59] introduces a “privacy shared-responsibility” framework to supplant traditional, user-managed permission models. Expert users’ privacy preferences are modeled via soft set theory, yielding two algorithms: PPPA-I and PPPA-II. PPPA-I analyzes historical decisions of expert users to automatically suppress superfluous permission requests. PPPA-II addresses high-risk, runtime permission requests by comparing their subjective necessity—calculated through pre-stored expert-preference weights—with predefined thresholds, thus automating decision-making and mitigating the cognitive burden on end users.

Table 3. Data used in the studies on home safety and security for the elderly.

Publications	Physiological Data	Physical Activity (Video)	Physical Activity (Non Video)	Position	Environmental Data	Household Appliance Data	Digital Information
[11]		✔
[50]		✔
[51]		✔	✔
[53]		✔	✔
[56]		✔	✔			✔
[57]	✔	✔	✔	✔	✔	✔
[12]		✔			✔	✔
[58]							✔
[59]
[60]							✔
[62]							✔
[63]						✔	✔

summarizes the data types utilized in the study described in home safety and security for the elderly.

2.2.3. Challenges: Insufficient Detection Accuracy and Difficulties in Synergizing Multiple Devices

The smart home security system must analyze elderly individuals’ daily activities and home-environmental changes in real-time, which demands high detection accuracy for quick emergency responses. Meanwhile, the system, comprising multiple IoT devices, faces challenges in ensuring inter-device collaboration and security.

(1) Insufficient accuracy in behavioral detection

Two main challenges affect elderly daily behavior detection models. First, home-environmental factors such as lighting, obstacles, and activity-specific postures of the elderly can compromise system detection accuracy, leading to higher false alarm rates [50]. Second, the absence of representative datasets capturing elderly individuals’ daily behaviors hinders the generalization and accuracy of detection models [66].

(2) Difficulty in coordinating multiple devices

Home security systems face challenges in synchronizing and integrating various devices like cameras and smart home components to ensure seamless operation [67]. Additionally, smart IoT devices are vulnerable to cyberattacks, information theft, and hardware failures [68], which threaten the privacy and safety of the elderly. This requires the home security system to have robust computational capabilities, such as selecting appropriate AI technology for multi-device coordination of multiple devices.

2.2.4. Future Prospects: Multifunctional Integrated Home Security Protection System

The safety system will not only recognize and respond to physical emergencies of the elderly but also monitor food, water, and air quality, sending alerts when food items at home are not fresh. Additionally, the system will integrate with mobile service robots such as intelligent robots and robotic arms. In case of falls, fires, electrical short circuits, earthquakes, or other emergencies, these devices can provide support, deliver emergency medications, and even escort the elderly to safety. Furthermore, the security system will incorporate psychological analysis to monitor the mental health of the elderly, to help prevent self-harm due to depression or loneliness.

2.3. Smart Life Assistants

The application of AI technology in the field of smart life assistants has become an important research direction, particularly in enhancing the autonomy and quality of life for the elderly. Figure 5 presents the architecture of an AI-based smart life assistant system designed for home-based elderly care. The system receives user commands through multi-modal input, performs intelligent analysis and decision-making, and coordinates responses via a central control unit. It facilitates interaction with both smart home devices and functional robotic assistants, thereby supporting daily activities and promoting independent living.

2.3.1. Voice Interaction

In the field of voice interaction for intelligent home assistance for the elderly, breakthroughs have been made in natural language processing technology based on deep learning. Large-scale pre-trained language models centered on the Transformer architecture have significantly improved the system’s ability to parse non-standard elderly speech. Pradhan et al. [69] demonstrated that models employing self-attention mechanisms and multi-task learning effectively address ambiguous references and local accents prevalent among elderly users, significantly reducing the word error rate. Specifically, an end-to-end speech recognition system combining a bidirectional LSTM network [70] achieved an intent recognition accuracy of 91.7% on an elderly speech dataset, and especially excelled in dealing with the dysarthria of people with Parkinson’s disease [71].

Contextual comprehension is made possible by innovations in multi-modal fusion architecture. When an elderly user gives the command “It’s a little dark here”, the system integrates speech features, ambient light sensor data and historical behavioral patterns through a gated attention mechanism, driving a reinforcement learning-based decision module to select the optimal response strategy [72]. Moreover, Jie Gu et al. [73] proposed a personalized adjustment strategy based on deep Q-network [74], which led to an increase of 38% in the satisfaction of elderly users in a three-month field test.

At the same time, the researchers utilized AI technology in a bid to address ethical issues. Many elderly users are concerned that the system is overly controlling, and they prefer systems that improve comfort and safety without affecting their daily activities or sense of independence [75]. However, the introduction of federated learning frameworks can protect speech data privacy through distributed training, which can degrade model performance by 12–15 percentage points [76].

2.3.2. Visual Interaction

In the field of visual interaction, the application of computer vision technology has enabled gesture recognition. The elderly can control household appliances through simple hand gestures, such as waving and pointing. This is particularly beneficial for those with mobility impairments or reduced hand dexterity. The french SWEET-HOME project [13] integrates a three-tier system combining touch-sensitive panels, physical switches, and AI-enhanced visual cues, achieving an 84% operational fluency rate among elderly users. Data indicate that in security monitoring scenarios, 89% of users approve of contactless interaction.

Beyond gesture recognition, biofeedback-based interaction technologies further expand the boundaries of visual interaction in recent years. Eye-tracking technology captures users’ unconscious behavioral intentions by analyzing physiological signals such as gaze trajectories and blink frequencies. It can determine a user’s needs for specific household areas—for instance, preheating a smart stove by detecting sustained attention towards the kitchen. Traditional approaches rely on infrared devices and high-definition cameras to ensure accuracy. However, advances in mobile computing have made low-cost, non-intrusive monitoring feasible. For example, Madhusanka et al. [14] used a CNN-SVM hybrid model to classify eye movement directions in real-time video streams (30 frames per second). The system achieves an average accuracy of 90.83%. It precisely recognizes commands such as “central gaze”, “left/right gaze”, and “blink”, enabling timely responses to the needs of elderly users.

2.3.3. Smart Home Assistant

Smart home assistants have emerged as a transformative application in elderly care, leveraging automation and intelligent interaction to enhance safety, convenience, and independent living capabilities. These systems typically adopt a three-tier technical architecture: perception layer (sensors, cameras, and IoT devices for collecting environmental/behavioral data); edge computing layer (local processing units for real-time decision-making); interaction layer (voice, touch, or gestural interfaces for user control).

A representative implementation by S.K. Sooraj et al. [77] demonstrates this framework through an IoT-based system integrating AI speech recognition with Raspberry Pi, PIR motion sensors, and Pi cameras. Divya Ganesh et al. [78] propose the application of Google Duplex AI technology as a personal assistant for home-based elderly care, utilizing Body Sensor Network [79] for real-time health monitoring and analysis. The edge computing architecture enables localized processing for safety monitoring and appliance control, significantly reducing latency while ensuring privacy. Crucially, these systems employ AI-driven device scheduling—optimizing energy consumption while maintaining thermal comfort, thereby minimizing manual intervention.

Recent advancements focus on multi-modal interaction to address age-related impairments. G. Ashwini et al. [15] developed an elderly care robot integrating voice assistance, gesture recognition, and tactile interfaces. This hybrid approach accommodates diverse needs. Voice commands assist mobility-impaired users. Gesture controls support those with speech disorders, while touch interfaces provide redundancy for cognitive decline scenarios. Notably, their medication management subsystem implements a dual verification mechanism—combining temporal reminders with ResNet-based [80] pill morphology image recognition–reducing dosage errors by 63% in trials. Collectively, these innovations demonstrate how smart home assistants transcend basic automation. By embedding contextual awareness and predictive analytics, they proactively mitigate risks while preserving elderly autonomy—a critical factor in home-based elderly care.

2.3.4. Functional Robotic Assistants

The development and application of functional robots can significantly enhance the convenience and well-being of the elderly in home-based living. By integrating AI technologies with automation, these systems not only assist in daily activities but also provide caregiving services, fulfilling multifaceted roles in elderly care [81,82]. The application of AI in elderly care robots is primarily manifested in three core domains: environmental perception, human–robot interaction, and task execution.

In environmental perception, multi-sensor fusion has become the mainstream methodology [83]. Scholars further emphasize that by integrating temperature, pressure, motion, and other multi-modal sensor networks, robots can monitor the vital-sign parameters of the elderly in real-time. For example, the QTrobot described by Andrea Antonio Cantone et al. [84] utilizes medical-grade sensors to continuously collect data such as blood pressure and blood oxygen saturation, combined with a modified early warning wcore system [85] to evaluate health conditions. Additionally, advanced algorithmic implementations are employed. The CHARMIE robot designed by Ribeiro T. et al. [86] exemplifies this through the collaborative operation of 2D LiDAR and RGB-D cameras to construct dynamic 3D environmental maps, achieving millimeter level real-time localization and obstacle recognition. Its Adaptive Monte Carlo Localization algorithm [87] simultaneously processes odometry and LiDAR data to maintain navigation stability in unstructured environments.

In human–robot interaction, the AI-enhanced robot demonstrated by Noreen et al. [16] adopts a CNN architecture, achieving 95% accuracy in daily object recognition based on the ImageNet dataset [88], with a personalized facial database supporting family member identification. The care robot designed by Zwilling et al. [89] employs a multi-directional microphone array and deep learning-based speech models, enabling natural conversation and interaction in real-world nursing facility tests. Its tactile feedback system captures contact signals via force-sensitive resistive sensors embedded in flexible synthetic leather surfaces. Chaolong Qin et al. [90] proposed a multi-modal home service robot interaction system integrating five interaction modalities: touch, voice, electromyographic gestures, visual gestures, and haptic feedback. By designing a dynamically weighted multi-modal fusion algorithm, the system achieves a 93.62% intent recognition rate in complex environments, effectively addressing the vulnerability of single-modality systems. These technological breakthroughs enable robots to detect emotional fluctuations in the elderly’s voice commands, automatically initiating soothing protocols or contacting relatives via video calls when anxiety or depression is detected. In task execution, AI technologies empower robots with dynamic planning and adaptive capabilities. The system developed by Noreen et al.’s team [16] implements daily task scheduling, such as medication reminders and cognitive training, through time-series modeling, with its object-tracking algorithm maintaining over 90% recognition stability in mobile scenarios. Linkun Zhou et al. [91] proposed a generalized AI architecture for elderly care robots named AoECR, fine-tuned from ChatGLM-9B [92] using low-rank adaptation technology [93] to infuse caregiving knowledge. The robotic platform developed by Zwilling et al. [89] rapidly deployed a UV disinfection module during the COVID-19 pandemic, with its thermal imaging temperature detection system achieving non-contact fever screening within an error margin of 0.3 °C. The Heterogeneous Intelligent Multi-Task Assistant Ecosystem proposed by R Barber et al. [94] combines physiologically driven emotion prediction models with proactive activity recommendations. Through dynamic path planning, semantic environmental modeling, and task collaboration, the system enhances the mental health of homebound elderly individuals and assists in daily tasks. These innovative practices demonstrate that AI technologies are driving elderly care robots toward autonomous decision-making, contextual adaptation, and predictive care paradigms.

To further elaborate on the system illustrated in Figure 6, which shows a real-world deployment of an intelligent home assistant, the smart home assistant architecture integrates multiple technical modules previously discussed in this section. The perception layer collects multi-source data through microphones, Pi cameras, and PIR sensors, enabling the recognition of elderly voice commands and movement patterns [13]. Speech input is processed using bidirectional LSTM networks combined with Connected Timing Classification algorithms [70]. These models are particularly effective in handling speech impairment conditions such as dysarthria in Parkinson’s disease [71].

The decision-making module fuses context-aware data, including speech signals, ambient light levels, and historical usage, via gated attention mechanisms [72], and triggers adaptive actions using reinforcement learning strategies like deep Q-networks to optimize user satisfaction [73]. This allows the system to interpret indirect commands (e.g., “I’m cold”) and respond appropriately. The system architecture also embeds federated learning frameworks to enhance data privacy during model training [76], a critical consideration for elderly users who are often sensitive to surveillance and control [75]. These features ensure the system does not only automate tasks but also respects personal boundaries while improving comfort, safety, and autonomy in home-based elderly care.

Table 4. Data used in the studies on smart life assistants.

Publications	Physiological Data	Physical Activity (Video)	Physical Activity (Non Video)	Voice Data	Environmental Data	Household Appliance Data	Lifestyle Habits	Medical History and Symptom Data
[70]		✔
[71]				✔		✔	✔
[72]				✔		✔
[73]				✔		✔	✔	✔
[74]				✔	✔
[75]				✔	✔	✔		✔
[76]				✔		✔		✔
[13]	✔		✔	✔	✔	✔
[14]			✔		✔
[77]		✔		✔	✔		✔
[78]		✔
[15]		✔		✔			✔
[80]		✔
[81]	✔	✔		✔	✔		✔	✔
[82]	✔		✔		✔	✔
[84]	✔	✔		✔	✔		✔	✔
[85]	✔							✔
[86]		✔		✔	✔
[87]	✔	✔		✔	✔
[16]	✔			✔	✔
[89]		✔		✔	✔
[90]			✔	✔		✔
[91]		✔		✔
[94]			✔	✔	✔	✔	✔

summarizes the data types utilized in the study described in smart life assistants.

2.3.5. Challenges: User Acceptance and Technological Compatibility

The application of AI technologies in life-assisting devices faces dual challenges of technological adaptability and user acceptance, particularly among elderly populations. While smart home devices can significantly enhance convenience and safety for seniors, their designs often fail to adequately consider the physiological and psychological characteristics of elderly users. For instance, age-related challenges such as hearing loss, visual impairment, and reduced manual dexterity can hinder device operation or voice interaction. In addition, complex setup procedures and the prevalence of technical terminology frequently discourage the elderly from engaging with AI technologies.

(1) User acceptance

From a psychological perspective, the elderly encounter a steep learning curve when adopting new technologies, leading to significant barriers to adaptation. Surveys reveal that approximately 60% of elderly users feel confused or uncomfortable when using smart devices, with some voicing concerns about potential privacy breaches or security risks [95]. A pervasive lack of trust in smart technologies, especially in their reliability during emergencies, further discourages adoption and sustained use.

(2) Technological adaptability

In terms of technological adaptability, most smart devices are designed primarily with younger users in mind, offering limited optimization for the elderly. For example, many voice assistants do not effectively accommodate regional dialects or imprecise pronunciations, which are common among older users. Furthermore, the slower speech patterns typical of older people often result in increased recognition errors and latency in system responses. In addition, reliance on smartphones or tablets for control functions, combined with limited digital literacy among older populations, significantly undermines the accessibility and usability of smart life-assisting devices.

2.3.6. Future Prospect: Smarter Home Service

First of all, smart home environments will provide safer and more convenient living experiences for the elderly. Future smart home systems will use AI algorithms to learn the elderly’s daily habits and automatically adjust lighting, temperature, and air quality, creating a more comfortable and health-conscious living environment. Additionally, devices such as smart locks, surveillance cameras, and emergency call systems will further enhance home security, offering seniors greater peace of mind. Furthermore, advancements in smart technologies will drive the adoption of intelligent assistants and robots in home-based elderly-care. Progress in voice recognition and natural language processing will enable seniors to interact with smart devices through voice commands for tasks such as playing music or checking the weather. In the future, smart robots are expected to undertake more caregiving responsibilities, such as reminding the elderly to take medication, engaging in conversations, assisting with daily activities, and even providing emergency aid. These intelligent assistants and robots will serve as “smart companions” for seniors, ensuring they feel cared for and supported in their homes.

2.4. Psychological Care and Emotional Support

The application of smart technology in the field of the elderly not only involves assistance on the physical level, but also goes deeper into the psychological care for the elderly. With the aggravation of population aging and the growth of the scale of home care, there are more and more empty-nested elderly people, whose psychological health and emotional needs should not be ignored.

Figure 7 illustrates a typical AI-driven system architecture designed for psychological care and emotional support in home-based elderly care. The system collects emotional cues through multi-modal inputs and conducts emotion recognition. Based on affective interaction, AI enables mental health intervention strategies, which are delivered through multi-modal outputs to support psychological well-being.

In the field of emotion recognition, the application of AI technology has made breakthrough progress. Based on the results of psychological research, this field not only recognizes human facial expressions through facial muscle activity but also leverages deep learning networks to extract deep and abstract features for accurate expression recognition. For example, in addition to visual analysis of facial expressions, emotion recognition often incorporates vocal features such as tone, pitch, and speech rate to improve overall accuracy and broaden its applicability.

Early emotion recognition systems widely used machine learning techniques such as SVM and KNN [96]. These methods performed well in processing physiological signals like EEG, and electrodermal responses. In recent years, deep learning techniques have become dominant. For example, CNNs are widely used in image and facial recognition, while LSTM networks are effective in processing temporal data such as speech and emotional sequences.

2.4.1. Emotion Recognition

In the field of emotion recognition, recent advancements demonstrate that leveraging machine learning and deep learning techniques to analyze facial expressions can significantly enhance the detection of the elderly’s emotional states [97], supporting mental health monitoring, emotional intervention, and social engagement. Emotion recognition systems, particularly those based on facial expressions, have been deployed in home care settings, such as intelligent care robots, affective interaction systems, and remote psychological counseling platforms. Wiam FADEL et al. [17] achieved significant progress in speech emotion recognition by combining CNNs with self-supervised learning techniques. They proposed a dual-path CNN architecture: a 1D-CNN for raw waveform processing and a 2D-CNN with three-layer spatial convolutions for analyzing Mel-spectrograms. Their model achieved over 75% accuracy in an eight-class classification task on the RAVDESS dataset. Furthermore, their fine-tuned HuBERT model [98] demonstrated strong generalization, reaching an AUC of 0.87 on cross-age validation sets. These systems can track emotional fluctuations in the elderly and, when integrated with voice interaction and NLP technologies, provide timely emotional support and mood regulation recommendations [99]. The CareTaker.ai system proposed by Ankur Gupta et al. [100] employs NLP techniques for emotional communication with patients. However, despite these advancements, challenges persist in cross-age emotion recognition, data privacy protection, and real-time processing. Aging-induced changes in facial features may compromise the accuracy of recognition systems.

Recently, the integration of multi-modal data has improved the capabilities of emotion recognition systems. Chirag Dalvi et al. [101] developed a notable multi-modal emotion recognition system that integrates facial expression recognition, speech emotion analysis, and physiological signal monitoring using deep learning techniques. They employed SVM for processing physiological data and CNN for facial emotion recognition. This approach not only achieved high accuracy but also significantly reduced processing time from 40 min down to 2 min. By capturing comprehensive emotional data, the system enables timely interventions and delivers more precise emotional support for the elderly.

2.4.2. Emotional Interaction

In the field of emotional interaction, the innovative application of intelligent technology is reshaping the mental health service model for the elderly, and the AHOBO care robot developed by Yamazaki’s team [102] is a pioneering fusion of “memory coloring therapy” and multi-modal interaction technology, which is equipped with a touch-control painting interface and a natural language processing module, realizing real-time art generation and emotional communication. Its core technology lies in the image style migration algorithm based on CNN, which can instantly transform the user’s drawings into watercolor effects, and with the dynamic feedback of the bionic-designed “AHOGE” emotional tentacles, it can form a bidirectional emotion regulation mechanism. Although clinical trials showed limited improvement in language fluency, the system significantly outperformed traditional interventions in terms of ease of operation and clarity of instructions, confirming the feasibility of non-pharmacological intervention pathways.

A breakthrough in this field came when Kiran et al. [103] developed the Intelligent Aging Companion System, which constructed a closed-loop mental health prediction-intervention system by integrating the RF algorithm with deep neural networks. Its innovativeness is reflected in the three-level architecture: the data layer integrates wearable device physiological signals with environmental sensor data; the algorithm layer adopts XGBoost [18] for loneliness index prediction and combines with the BERT model to achieve emotional intent recognition; the application layer provides personalized intervention programs through an adaptive interface. Notably, the system introduces a federated learning mechanism to achieve cross-agency model optimization under the premise of privacy assurance, which provides a new idea to solve the problem of health care data silos.

During the same period, the IntelliJoyCare system developed by Ma Wenjie’s team [104] took a different approach, creating a “digital humanities” technology paradigm. The system builds a knowledge graph containing 200,000 elderly corpus, adopts WaveGlow vocoder to generate personalized audio content, and realizes cross-cultural context adaptation through transfer learning. The core breakthrough lies in the development of a dynamic assessment matrix of psychological state, which combines speech emotion recognition and micro-expression analysis to realize minute-by-minute monitoring of mood fluctuations. Practical application data show that after 8 weeks of continuous use, the experimental group showed significant improvement in social engagement and subjective well-being, which verified the effectiveness of the human–computer symbiotic intervention.

2.4.3. Mental Health Intervention

In terms of entertainment and socialization, the application of virtual reality and augmented reality technologies creates an immersive experience for the elderly. Real-time video chatting and virtual social platforms effectively reduce the risk of loneliness and depression among the elderly. Waycott et al. [19] have shown that immersive experiences, such as virtual concerts and theater performances, significantly increase engagement and interaction among older adults. These technologies not only provide rich entertainment options for the elderly, but also enhance their social connections through virtual social platforms, effectively alleviating their loneliness.

The application of AI technology in psychological support and health monitoring for the elderly offers distinct advantages. First of all, it provides real-time and continuous monitoring capabilities, enabling 24/7 surveillance to ensure timely interventions during health emergencies. Secondly, AI systems deliver personalized support by adapting services to individual needs through machine learning algorithms. Thirdly, multi-modal data fusion enhances recognition accuracy by integrating visual, auditory, and physiological signals from multiple sources. Additionally, interaction-friendly interfaces improve user engagement through natural language processing and affective computing technologies. Finally, AI facilitates preventive interventions by identifying early risk indicators through predictive analytics and initiating proactive measures. Collectively, these capabilities allow AI-driven systems to comprehensively address the health care and psychological needs of the elderly, significantly improving their quality of life.

The technical architecture presented in Figure 7 reflects a multi-stage emotion recognition and psychological care framework, which supports intelligent emotional monitoring and intervention for elderly users. This framework is realized through layered integration of multi-modal perception, feature extraction, intent understanding, and responsive feedback. As shown in the CAER-Net architecture in Figure 8, the system employs a dual-branch network that combines facial features with contextual scene information to enhance emotion recognition accuracy in real-world scenarios [105]. At the sensing level, emotional cues are collected from facial expressions, speech signals, and physiological data. CNN-based models are applied to extract facial features for emotion classification [97,101], while dual-path CNN architectures process both raw audio and Mel-spectrograms to detect speech emotions with high accuracy [17]. These inputs are further enhanced by Transformer-based self-supervised learning models such as HuBERT, which demonstrate strong generalization across age groups [98].

For intent recognition, systems leverage advanced natural language processing techniques. Transformer-based models like BERT are employed to infer emotional intent from user utterances, while context is dynamically integrated using multi-modal fusion approaches that align audio-visual data streams with behavioral histories [72]. The decision-making layer includes machine learning classifiers such as XGBoost, used to predict loneliness indices and mood trends based on time-series physiological and environmental data [18,103]. To ensure personalization while maintaining data privacy, federated learning frameworks are incorporated [76], enabling distributed model training across devices or agencies without transferring raw data.

Table 5. Data used in the studies on psychological care and emotional support.

Publications	Physiological Data	Physical Activity (Video)	Physical Activity (Non-Video)	Voice Data	Environmental Data	Household Appliance Data	Lifestyle Habits	Medical History and Symptom Data
[17]	✔			✔
[98]	✔			✔
[99]		✔
[100]	✔	✔			✔		✔	✔
[101]		✔
[102]	✔		✔				✔	✔
[103]
[104]								✔
[18]			✔	✔			✔	✔
[19]				✔
[106]				✔				✔

summarizes the data types utilized in the study described in psychological care and emotional support.

2.4.4. Challenges: Emotional Authenticity and Ethical Concerns

In the field of psychological care and emotional support, AI technologies face challenges related to the authenticity of emotional interaction and ethical concerns. Chatbots, through voice, text, and visual analysis, can partially perceive and respond to the emotional states of the elderly, offering companionship and emotional support. However, the authenticity and depth of such interactions remain insufficient to provide meaningful emotional companionship, falling short of replacing human emotional connections and raising potential privacy and ethical risks.

(1) Emotional authenticity

A. Rodríguez-Martínez et al. [106] conducted a study involving interviews and focus group discussions with 42 elderly participants. Their findings revealed that 80% of the respondents hoped chatbots could not only recognize emotions but also demonstrate “empathy” and “understanding” to foster deeper emotional connections. This finding indicates that elderly users expect more from the technology than just functional support; they seek deeper emotional resonance. However, most current chatbots provide surface-level emotional responses, often limited to predefined rules or simple emotional models, without a nuanced understanding of user contexts or complex emotions. Such shallow emotional interactions may lead users to feel increasingly disconnected and could even result in emotional dissatisfaction or disappointment.

(2) Ethical concerns

From an ethical perspective, privacy protection and data security are prominent issues. The aforementioned study also showed that 65% of elderly participants expressed concerns about chatbots’ ability to protect their personal information. These concerns stem from the use of sensitive data, such as voice, images, and facial expressions, in emotion computing technologies. These data are vulnerable to risks of leakage, misuse, or unauthorized secondary use during collection, storage, and application. Furthermore, emotional companion devices, such as chatbots, may introduce ethical dilemmas by influencing the decision-making processes of elderly users through overly personalized suggestions or responses. For example, the elderly living alone for extended periods may develop excessive dependence on robots, even perceiving them as genuine emotional attachments, potentially neglecting real-life interactions with family or society. This may have an adverse effect on their psychological health.

To address these challenges, AI technologies must balance emotional interaction authenticity with ethical compliance. Technically, more sophisticated emotional modeling and natural language processing can enhance chatbots’ emotional understanding. Integrating multi-modal emotion computing models could yield responses more akin to human interactions. In terms of privacy and security, it is essential to strengthen data encryption and anonymization techniques to safeguard sensitive user data. Transparency mechanisms should also be introduced, such as clearly labeling and informing users about how their data are collected and used, as well as granting users control over their data. These measures can enhance user trust in the technology and ensure ethical and secure applications in emotional interaction systems for elderly care.

2.4.5. Future Prospect: Affective Computing and Personalized Care

Affective computing represents a cutting-edge domain in AI that aims to equip machines with the ability to understand, interpret, and respond to human emotions. In the context of home-based elderly care, the integration of affective computing with personalized companion services is emerging as a pivotal direction for enhancing AI’s capacity to provide emotional support. This convergence not only imbues AI technology with greater human-like qualities but also offers new opportunities for improving the psychological well-being and quality of life of the elderly.

The core of affective computing lies in its ability to deeply perceive and interpret multi-modal data, including vocal tone, facial expressions, body movements, and textual content. By leveraging the collaborative application of natural language processing, computer vision, and sensor technologies, AI systems can detect the emotional states of the elderly during daily interactions. For example, an intelligent voice assistant can analyze voice pitch and rhythm to identify signs of emotional distress, while facial expression analysis can detect fatigue or irritability. These perception capabilities provide a robust feedback mechanism for affective companion robots, enabling them to respond more effectively to emotional needs.

Building on affective computing, personalized companion services can dynamically adapt to the emotional states and behavioral patterns of elderly users. Companion robots and intelligent systems can not only offer emotional support through conversation and interaction but also create customized care plans based on the elderly’s daily behavior. For instance, if signs of emotional distress are detected, the robot might play the user’s favorite music, recommend relaxation activities, or employ humor to uplift their mood. Over time, these companion robots can accumulate data through sustained interactions, gradually learning the user’s preferences and habits. This enables the robots to optimize caregiving strategies and deliver more attentive and precise services. The integration of affective computing with companion services can also compensate for the lack of social interaction in the lives of the elderly, enhancing their sense of belonging and overall happiness.

3. Conclusions

In this survey, we examined the applications of AI technologies in home-based elderly care from four main domains: smart health monitoring, intelligent safety and security, smart life assistance, and psychological care and emotional support. AI-driven health systems, such as wearables and remote diagnostic tools, help track vital signs, detect early symptoms, and manage chronic conditions more effectively. In terms of safety, AI-enabled fall detection, emergency alerts, and hazard monitoring reduce risks and support faster response. Smart life assistance tools, like home automation, voice control, and care robots, improve daily routines and promote independent living. Psychological care, supported by emotion-sensing systems and virtual companions, helps reduce loneliness and enhance mental well-being. Despite the benefits, challenges remain. Challenges include privacy risks, limited access to technology, high costs, and the need to connect AI with current care systems. Bringing AI into elderly care is more than innovation—it is a new way to imagine care at home. With responsible use, AI can help build a safer and more comfortable environment for seniors. We call on both researchers and industry leaders to work together to improve home-based elderly care services.