Search Results (44)

Background/Objectives: Repetitive behaviors such as hand flapping, body rocking, and head shaking characterize Autism Spectrum Disorder (ASD) while functioning as early signs of neurodevelopmental variations. Traditional diagnostic procedures require extensive manual observation, which takes significant time, produces subjective results, and remains unavailable to many regions. The research introduces a real-time system for the detection of ASD-typical behaviors by analyzing body movements through the You Only Look Once (YOLOv11) deep learning model. Methods: The system’s multi-layered design integrates monitoring, network, cloud, and typical ASD behavior detection layers to facilitate real-time video acquisition, wireless data transfer, and cloud analysis along with ASD-typical behavior classification. We gathered and annotated our own dataset comprising 72 videos, yielding a total of 13,640 images representing four behavior classes that include hand flapping, body rocking, head shaking, and non_autistic. Results: YOLOv11 demonstrates superior performance compared to baseline models like the sub-sampling (CNN) (MobileNet-SSD) and Long Short-Term Memory (LSTM) by achieving 99% accuracy along with 96% precision and 97% in recall and the F1-score. Conclusions: The results indicate that our system provides a scalable solution for real-time ASD screening, which might help clinicians, educators, and caregivers with early intervention, as well as ongoing behavioral monitoring. Full article

Recent advances in fish transportation technologies and deep machine learning-based fish classification have created an opportunity for real-time, autonomous fish sorting through a selective passage mechanism. This research presents a case study of a novel application that utilizes deep machine learning to detect partially dewatered fish exiting an Archimedes Screw Fish Lift (ASFL). A MobileNet SSD model was trained on images of partially dewatered fish volitionally passing through an ASFL. Then, this model was integrated with a network video recorder to monitor video from the ASFL. Additional models were also trained using images from a similar fish scanning device to test the feasibility of this approach for fish classification. Open source software and edge computing design principles were employed to ensure that the system is capable of fast data processing. The findings from this research demonstrate that such a system integrated with an ASFL can support real-time fish detection. This research contributes to the goal of automated data collection in a selective fish passage system and presents a viable path towards realizing optical fish sorting. Full article

This study proposes a method for selecting suitable edge hardware and Artificial Intelligence (AI) models to be deployed on these edge devices. Edge AI, which enables devices at the network periphery to perform intelligent tasks locally, is rapidly expanding across various domains. However, selecting appropriate edge hardware and AI models is a multi-faceted challenge due to the wide range of available options, diverse application requirements, and the unique constraints of edge environments, such as limited computational power, strict energy constraints, and the need for real-time processing. Ad hoc approaches often lead to non-optimal solutions and inefficiency problems. Considering these issues, we propose a method based on the ISO/IEC 25010:2011 quality standard, integrating Multi-Criteria Decision Analysis (MCDA) techniques to assess both the hardware and software aspects of Edge AI applications systematically. For the proposed method, we conducted an experiment consisting of two stages: In the first stage of the experiment, to show the applicability of the method across different use cases, we tested the method with four scenarios on UAVs, each presenting distinct edge requirements. In the second stage of the experiment, guided by the method’s recommendations for Scenario I, where the STM32H7 series microcontrollers were identified as the suitable hardware and the object detection model with Single Shot Multi-Box Detector (SSD) architecture and MobileNet backbone as the suitable AI model, we developed a TensorFlow Lite model from scratch to enhance the efficiency and versatility of the model for object detection tasks across various categories. This additional TensorFlow Lite model is aimed to show how the proposed method can guide the further development of optimized AI models tailored to the constraints and requirements of specific edge hardware. Full article

Object detection in maritime environments is a challenging problem because of the continuously changing background and moving objects resulting in shearing, occlusion, noise, etc. Unluckily, this problem is of critical importance since such failure may result in significant loss of human lives and economic loss. The available object detection methods rely on radar and sonar sensors. Even with the advances in electro-optical sensors, their employment in maritime object detection is rarely considered. The proposed research aims to employ both electro-optical and near-infrared sensors for effective maritime object detection. For this, dedicated deep learning detection models are trained on electro-optical and near-infrared (NIR) sensor datasets. For this, (ResNet-50, ResNet-101, and SSD MobileNet) are utilized in both electro-optical and near-infrared space. Then, dedicated ensemble classifications are constructed on each collection of base learners from electro-optical and near-infrared spaces. After this, decisions about object detection from these spaces are combined using logical-disjunction-based final ensemble classification. This strategy is utilized to reduce false negatives effectively. To evaluate the performance of the proposed methodology, the publicly available standard Singapore Maritime Dataset is used and the results show that the proposed methodology outperforms the contemporary maritime object detection techniques with a significantly improved mean average precision. Full article

Despite advancements in technology, safety equipment, and training within the construction industry over recent decades, the prevalence of fatal and nonfatal injuries and accidents remains a significant concern among construction workers. Hard hats and safety vests are crucial safety gear known to mitigate severe head trauma and other injuries. However, adherence to safety protocols, including the use of such gear, is often inadequate, posing potential risks to workers. Moreover, current manual safety monitoring systems are laborious and time-consuming. To address these challenges and enhance workplace safety, there is a pressing need to automate safety monitoring processes economically, with reduced processing times. This research proposes a deep learning-based pipeline for real-time identification of non-compliance with wearing hard hats and safety vests, enabling safety officers to preempt hazards and mitigate risks at construction sites. We evaluate various neural networks for edge deployment and find that the Single Shot Multibox Detector (SSD) MobileNet V2 model excels in computational efficiency, making it particularly suitable for this application-oriented task. The experiments and comparative analyses demonstrate the pipeline’s effectiveness in accurately identifying instances of non-compliance across different scenarios, underscoring its potential for improving safety outcomes. Full article

The classification of germinated pole sitao (Vigna unguiculata (L.) Walp.) seeds is important in seed germination tests. The automation of this process has been explored for different grain and legume seeds but is only limited to binary classification. This study aimed to develop a classifier system that can recognize three classes: normal, abnormal, and ungerminated. SSD MobileNet and Faster R-CNN models were trained to perform the classification. Both were trained using 1500 images of germinated seeds at fifth- and eighth-day observations. Each class had 500 images. The trained models were evaluated using 150 images per class. The SSD MobileNet model had an accuracy of 0.79 while the Faster R-CNN model had an accuracy of 0.75. The results showed that the average accuracies for the classes were significantly different from one another based on one-way ANOVA at a 95% confidence level with an F-critical value of 3.0159. The SSD MobileNet model outperformed the Faster R-CNN model in classifying pole sitao seeds, with improved precision in identifying abnormal and ungerminated seeds on the fifth day and normal and ungerminated seeds on the eighth day. The results confirm the potential of the SSD MobileNet model as a more reliable classifier in germination tests. Full article

Increasing numbers of public and private locations now have surveillance cameras installed to make those areas more secure. Even though many organizations still hire someone to monitor the cameras, the person hired is more likely to miss some unexpected events in the video feeds because of human error. Several researchers have worked on surveillance data and have presented a number of approaches for automatically detecting aberrant events. To keep track of all the video data that accumulate, a supervisor is often required. To analyze the video data automatically, we recommend using neural networks to identify the crimes happening in the real world. Through our approach, it will be easier for police agencies to discover and assess criminal activity more quickly using our method, which will reduce the burden on their staff. In this paper, we aim to provide anomaly detection using surveillance videos as input specifically for the crimes of arson, burglary, stealing, and vandalism. It will provide an efficient and adaptable crime-detection system if integrated across the smart city infrastructure. In our project, we trained multiple accurate deep learning models for object detection and crime classification for arson, burglary and vandalism. For arson, the videos were trained using YOLOv5. Similarly for burglary and vandalism, we trained using YOLOv7 and YOLOv6, respectively. When the models were compared, YOLOv7 performed better with the highest mAP of 87. In this, we could not compare the model’s performance based on crime type because all the datasets for each crime type varied. So, for arson YOLOv5 performed well with 80% mAP and for vandalism, YOLOv6 performed well with 86% mAP. This paper designed an automatic identification of crime types based on camera or surveillance video in the absence of a monitoring person, and alerts registered users about crimes such as arson, burglary, and vandalism through an SMS service. To detect the object of the crime in the video, we trained five different machine learning models: Improved YOLOv5 for arson, Faster RCNN and YOLOv7 for burglary, and SSD MobileNet and YOLOv6 for vandalism. Other than improved models, we innovated by building ensemble models of all three crime types. The main aim of the project is to provide security to the society without human involvement and make affordable surveillance cameras to detect and classify crimes. In addition, we implemented the Web system design using the built package in Python, which is Gradio. This helps the registered user of the Twilio communication tool to receive alert messages when any suspicious activity happens around their communities. Full article

Hiking and cycling have become popular activities for promoting well-being and physical activity. Portugal has been investing in hiking and cycling trail infrastructures to boost sustainable tourism. However, the lack of reliable data on the use of these trails means that the times of greatest affluence or the type of user who makes the most use of them are not recorded. These data are of the utmost importance to the managing bodies, with which they can adjust their actions to improve the management, maintenance, promotion, and use of the infrastructures for which they are responsible. The aim of this work is to present a review study on projects, techniques, and methods that can be used to identify and count the different types of users on these trails. The most promising computer vision techniques are identified and described: YOLOv3-Tiny, MobileNet-SSD V2, and FasterRCNN with ResNet-50. Their performance is evaluated and compared. The results observed can be very useful for proposing future prototypes. The challenges, future directions, and research opportunities are also discussed. Full article

Pick-and-place operations are an integral part of robotic automation and smart manufacturing. By utilizing deep learning techniques on resource-constraint embedded devices, the pick-and-place operations can be made more accurate, efficient, and sustainable, compared to the high-powered computer solution. In this study, we propose a new technique for object detection on an embedded system using SSD Mobilenet V2 FPN Lite with the optimisation of the hyperparameter and image enhancement. By increasing the Red Green Blue (RGB) saturation level of the images, we gain a 7% increase in mean Average Precision (mAP) when compared to the control group and a 20% increase in mAP when compared to the COCO 2017 validation dataset. Using a Learning Rate of 0.08 with an Edge Tensor Processing Unit (TPU), we obtain high real-time detection scores of 97%. The high detection scores are important to the control algorithm, which uses the bounding box to send a signal to the collaborative robot for pick-and-place operation. Full article

Using neural networks on low-power mobile systems can aid in controlling pests while preserving beneficial species for crops. However, low-power devices require simplified neural networks, which may lead to reduced performance. This study was focused on developing an optimized deep-learning model for mobile devices for detecting corn pests. We propose a two-step transfer learning approach to enhance the accuracy of two versions of the MobileNet SSD network. Five beetle species (Coleoptera), including four harmful to corn crops (belonging to genera Anoxia, Diabrotica, Opatrum and Zabrus), and one beneficial (Coccinella sp.), were selected for preliminary testing. We employed two datasets. One for the first transfer learning procedure comprises 2605 images with general dataset classes ‘Beetle’ and ‘Ladybug’. It was used to recalibrate the networks’ trainable parameters for these two broader classes. Furthermore, the models were retrained on a second dataset of 2648 images of the five selected species. Performance was compared with a baseline model in terms of average accuracy per class and mean average precision (mAP). MobileNet-SSD-v2-Lite achieved an mAP of 0.8923, ranking second but close to the highest mAP (0.908) obtained by MobileNet-SSD-v1 and outperforming the baseline mAP by 6.06%. It demonstrated the highest accuracy for Opatrum (0.9514) and Diabrotica (0.8066). Anoxia it reached a third-place accuracy (0.9851), close to the top value of 0.9912. Zabrus achieved the second position (0.9053), while Coccinella was reliably distinguished from all other species, with an accuracy of 0.8939 and zero false positives; moreover, no pest species were mistakenly identified as Coccinella. Analyzing the errors in the MobileNet-SSD-v2-Lite model revealed good overall accuracy despite the reduced size of the training set, with one misclassification, 33 non-identifications, 7 double identifications and 1 false positive across the 266 images from the test set, yielding an overall relative error rate of 0.1579. The preliminary findings validated the two-step transfer learning procedure and placed the MobileNet-SSD-v2-Lite in the first place, showing high potential for using neural networks on real-time pest control while protecting beneficial species. Full article

Object detection for a pick-and-place system has been widely acknowledged as a significant research area in the field of computer vision. The integration of AI and machine vision with pick-and-place operations should be made affordable for Small and Medium Enterprises (SMEs) so they can leverage this technology. Therefore, the aim of this study is to develop a smart and lean pick-and-place solution for custom workpieces, which requires minimal computational resources. In this study, we evaluate the effectiveness of illumination and batch size to improve the Average Precision (AP) and detection score of an EfficientDet-Lite model. The addition of 8% optimized bright Alpha3 images results in an increase of 7.5% in AP and a 6.3% increase in F1-score as compared to the control dataset. Using a training batch size of 4, the AP is significantly improved to 66.8% as compared to a batch size of 16 at 57.4%. The detection scores are improved to 80% with a low variance of 1.65 using a uniform 135-angle lamp and 0 illumination level. The pick-and-place solution is validated using Single-Shot Detector (SSD) MobileNet V2 Feature Pyramid Network (FPN) Lite. Our experimental results clearly show that the proposed method has an increase of 5.19% in AP compared to SSD MobileNet V2 FPNLite. Full article

The movement of vehicles in and out of the predefined enclosure is an important security protocol that we encounter daily. Identification of vehicles is a very important factor for security surveillance. In a smart campus concept, thousands of vehicles access the campus every day, resulting in massive carbon emissions. Automated monitoring of both aspects (pollution and security) are an essential element for an academic institution. Among the reported methods, the automated identification of number plates is the best way to streamline vehicles. The performances of most of the previously designed similar solutions suffer in the context of light exposure, stationary backgrounds, indoor area, specific driveways, etc. We propose a new hybrid single-shot object detector architecture based on the Haar cascade and MobileNet-SSD. In addition, we adopt a new optical character reader mechanism for character identification on number plates. We prove that the proposed hybrid approach is robust and works well on live object detection. The existing research focused on the prediction accuracy, which in most state-of-the-art methods (SOTA) is very similar. Thus, the precision among several use cases is also a good evaluation measure that was ignored in the existing research. It is evident that the performance of prediction systems suffers due to adverse weather conditions stated earlier. In such cases, the precision between events of detection may result in high variance that impacts the prediction of vehicles in unfavorable circumstances. The performance assessment of the proposed solution yields a precision of 98% on real-time data for Malaysian number plates, which can be generalized in the future to all sorts of vehicles around the globe. Full article

Nowadays, Convolution Neural Network (CNN) based deep learning methods are widely used in detecting and classifying fruits from faults, color and size characteristics. In this study, two different neural network model estimators are employed to detect apples using the Single-Shot Multibox Detection (SSD) Mobilenet and Faster Region-CNN (Faster R-CNN) model architectures, with the custom dataset generated from the red apple species. Each neural network model is trained with created dataset using 4000 apple images. With the trained model, apples are detected and counted autonomously using the developed Flying Robotic System (FRS) in a commercially produced apple orchard. In this way, it is aimed that producers make accurate yield forecasts before commercial agreements. In this paper, SSD-Mobilenet and Faster R-CNN architecture models trained with COCO datasets referenced in many studies, and SSD-Mobilenet and Faster R-CNN models trained with a learning rate ranging from 0.015–0.04 using the custom dataset are compared experimentally in terms of performance. In the experiments implemented, it is observed that the accuracy rates of the proposed models increased to the level of 93%. Consequently, it has been observed that the Faster R-CNN model, which is developed, makes extremely successful determinations by lowering the loss value below 0.1. Full article

This research proposes a study on two-way communication between deaf/mute and normal people using an Android application. Despite advancements in technology, there is still a lack of mobile applications that facilitate two-way communication between deaf/mute and normal people, especially by using Bahasa Isyarat Malaysia (BIM). This project consists of three parts: First, we use BIM letters, which enables the recognition of BIM letters and BIM combined letters to form a word. In this part, a MobileNet pre-trained model is implemented to train the model with a total of 87,000 images for 29 classes, with a 10% test size and a 90% training size. The second part is BIM word hand gestures, which consists of five classes that are trained with the SSD-MobileNet-V2 FPNLite 320 × 320 pre-trained model with a speed of 22 s/frame rate and COCO mAP of 22.2, with a total of 500 images for all five classes and first-time training set to 2000 steps, while the second- and third-time training are set to 2500 steps. The third part is Android application development using Android Studio, which contains the features of the BIM letters and BIM word hand gestures, with the trained models converted into TensorFlow Lite. This feature also includes the conversion of speech to text, whereby this feature allows converting speech to text through the Android application. Thus, BIM letters obtain 99.75% accuracy after training the models, while BIM word hand gestures obtain 61.60% accuracy. The suggested system is validated as a result of these simulations and tests. Full article

As one of the core applications of computer vision, object detection has become more important in scenarios requiring high accuracy but with limited computational resources such as robotics and autonomous vehicles. Object detection using machine learning running on embedded device such as Raspberry Pi provides the high possibility to detect any custom objects without the recalibration of camera. In this work, we developed a smart and lean object detection model for shipping containers by using the state-of-the-art deep learning TensorFlow model and deployed it to a Raspberry Pi. Using EfficientDet-Lite2, we explored the different cross-validation strategies (Hold-out and K-Fold). The experimental results show that compared with the baseline EfficientDet-Lite2 algorithm, our model improved the mean average precision (mAP) by 44.73% for the Hold-out dataset and 6.26% for K-Fold cross-validation. We achieved Average Precision (AP) of more than 80% and best detection scores of more than 93% for the Hold-out dataset. For the 5-Fold lean dataset, the results show the Average Precision across the three lightweight models are generally high as the models achieved more than 50% average precision, with YOLOv4 Tiny performing better than EfficientDet-Lite2 and Single Shot Detector (SSD) MobileNet V2 Feature Pyramid Network (FPN) 320 as a lightweight model. Full article