Spatiotemporal Cleaning of PIR Sensor Data for Elderly Movement Monitoring

Round 1

Reviewer 1 Report (Previous Reviewer 3)

Comments and Suggestions for Authors

It is a resubmitted manuscript. Unfortunately, most of my concerns have not been addressed. Therefore, this paper remains unsuitable for publication at this time, especially in light of the following issues.

(1)  The spatiotemporal data cleaning method proposed in the article combines Non-Deterministic Tracking (NDT) and Late Binding Adjustment (LBA) algorithms. While both methods are effective, they do not demonstrate significant technical innovation. These techniques have been widely applied in other fields, and the article fails to showcase any substantial technological breakthroughs in the processing of PIR sensor data.

(2) The article uses PIR sensors for motion detection and proposes methods for handling false positives and missed detections. However, the robustness to environmental changes (such as temperature and lighting) is not adequately addressed. Complex indoor environments may lead to a significant number of false positives and missed detections, and the article lacks sufficient experimental data to demonstrate the stability and applicability of the proposed method in real-world scenarios.

(3) The detection range of PIR sensors is limited, and they are particularly prone to interference from obstacles in complex home environments. Although the article proposes reducing detection errors by obstructing the Fresnel lens, it fails to adequately address the issues of sensor blind spots and restricted detection range, which ultimately affects the overall performance of the system.

 

Previous comments:
(1) The authors' work is more like a case study, lacking generalized models, strategies and algorithms.

(2) Lack of quantitative metrics to verify that the method proposed in this paper performs better than previous literature.

(3) I note that the sensors involved in this manuscript are cheap and the arrangement scheme is simple, so why not carry out an experimental validation of the proposed method?

(4) How did the authors solve the problem of spatio-temporal synchronization between multiple sensors, which is a very important practical issue.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report (New Reviewer)

Comments and Suggestions for Authors

The authors present a topic of interest - assisting the elderly at home. The topic is interesting and quite well presented, and the authors solve the problem using a set of Non-Deterministic Tracking (NDT) and the Late-Binding Adjustment (LBA) algorithms.

To improve the article, I offer the authors some suggestions:

- to present why they chose the non-deterministic algorithm compared to the deterministic algorithms.

- what are the problems introduced by the use of non-deterministic algorithms.

- what are the future perspectives of the algorithms presented (The Future of Non-Deterministic Algorithms).

- to make a comparison of the obtained results with similar results reported in the specialized literature.

- what is the relationship between the reported accuracy and the detection distance of the PIR.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report (New Reviewer)

Comments and Suggestions for Authors

This study presents a novel spatiotemporal data-cleaning framework that integrates Non-Deterministic Tracking (NDT) and Late-Binding Adjustment (LBA). The following critique outlines specific recommendations for enhancing clarity, rigor, and overall impact.

Introduction:

1. Formatting Consistency: Line 80’s format should be adjusted for consistency with the rest of the document.

2. References for Background Information: Lines 52–64 should include relevant references to provide a stronger theoretical foundation for the discussion. Incorporating citations from seminal or recent works in sensor data cleaning and elderly movement monitoring will add depth and credibility.

3. Expansion of Literature Review: The literature review should be expanded to clearly delineate the research gap. Including studies on PIR sensor accuracy, elderly movement monitoring, and data-cleaning methodologies will help contextualize this study within the field. Highlighting previous work that addresses specific limitations of PIR data in elderly care environments, such as environmental noise or tracking in cluttered spaces, would further justify the novelty of the proposed framework.

Part 2: Framework Description and Methodology

1. Structural Flow: The structural flow of Part 2 would benefit from refinement. Presenting equations separately, rather than within the text, can improve readability and allow readers to focus on understanding the model step-by-step.

2. Detailed Method Explanation: Consider elaborating on the NDT and LBA processes. Explaining their respective roles within the framework with greater specificity can aid readers in understanding the integration of these methods.

3. Algorithm Pseudocode: Providing pseudocode or flow diagrams to illustrate the sequence and interaction of NDT and LBA steps could enhance clarity for readers and make the methodology more accessible to those aiming to replicate or build on this work.

Part 3: Experimental Design and Data Collection

1. Figure Labeling: Ensure that key elements such as the PIR sensor Field of View (FOV) and detection range illustrated in Figure 2 are also labeled in Figure 7 for consistency and easy cross-referencing.

2. Sensor Specifications and Rationale: Clarify the specific PIR sensor models used in this study and the rationale for their selection. Consider discussing factors such as sensitivity, detection range, cost, and suitability for elderly movement monitoring. This information will strengthen the technical rigor of the study and help readers assess the scalability of your framework.

3. Document Revision History: The presence of red font in Line 416 suggests prior edits. If this was intended as a revision marker, consider updating it to standard font for the final draft to maintain a polished and professional presentation.

Analysis of Results:

1. Justification for Software Selection: In Lines 514–519, please provide a rationale for the choice of the three software systems utilized in the analysis. Explain how each system contributed to the data-cleaning process or offered unique analytical benefits. This context will help readers evaluate the robustness of the results and the flexibility of the framework across different platforms.

2. Result Visualization: Adding visual representations of the cleaned data, such as before-and-after examples of PIR sensor tracks, would further illustrate the effectiveness of the NDT and LBA methods. Visual comparisons or error metrics could reinforce claims about accuracy improvements.

Research Limitations and Comparison with Other Studies:

1. Limitations Discussion: Including a comprehensive discussion on research limitations will enhance the study’s transparency. Address constraints such as the controlled nature of the simulation environment, limitations in the dataset size or diversity, and the potential impact of computational power on the framework’s real-time applicability. These acknowledgments will not only increase the study’s credibility but also guide future researchers in addressing these challenges.

2. Comparative Analysis with Existing Studies: Consider comparing the study’s findings with those of other PIR sensor-based data-cleaning frameworks or similar applications in elderly movement monitoring. Highlighting areas where your approach excels or diverges from existing methods will provide context and emphasize the unique contributions of your framework.

3. Future Directions: Briefly outline potential future directions that can build on the limitations and comparative analysis discussed. For example, potential adaptations of the framework to other types of sensors, integration into larger elderly care systems, or application to different settings (e.g., clinical vs. residential environments) could inspire further research.

Implementing these recommendations would enhance the clarity, structure, and academic rigor of the study, making it more accessible and impactful for both practitioners and researchers in the field of elderly movement monitoring and spatiotemporal data analysis.

Comments on the Quality of English Language

The English could be improved to more clearly express the research.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report (Previous Reviewer 3)

Comments and Suggestions for Authors

No more comments.

Reviewer 2 Report (New Reviewer)

Comments and Suggestions for Authors

The authors responded to the suggestions.

Reviewer 3 Report (New Reviewer)

Comments and Suggestions for Authors

The authors have revised manuscript accordingly

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The article aims to introduce a framework that utilizes spatio-temporal cleaning and sensor network topology to improve the quality of movement data and derive the walking speed and the distance of an individual in their home.

General problems to be solved:

A)     The captions are too long; they should describe exclusively what is shown in the figure. Sometimes they repeat part of the text. They need to be radically modified.

B)     The acronyms GASS and OBAS represent the same thing. They are acronyms introduced by the authors and not found in the literature. Since they essentially describe a PIR (as a transducer), the authors are requested to use PIR to avoid confusing the reader.

C)     The text is redundant in some parts.

D)    The method and experimental results are conflated. The method should not be explained through an example, although an example can clarify it. This section needs to be substantially revised and rewritten.

E)      The experimental results are exclusively simulations. It is suggested to reference a real case. Artificial data, not derived from real cases, often do not highlight unexpected problems.

2 - PREVIOUS WORKS

These references are not relevant to the proposed activity (unobtrusive localization and tracking) and are unnecessary. Replace them with others related to indoor localization and tracking without the use of wearable devices.

1)     Danklang et al. employed Wi-Fi as a tracking methodology for autonomous robots [15]. They proposed a method in which the robot receives signals from a Wi-Fi access point and compares the received RSSI with the RSSI predicted by accelerometer mounted on the robot. The results demonstrated that this approach reduced positioning errors compared to using RSSI alone.

2)      Koutris et al. received signals emitted from BLE tags at multiple indoor anchor points and proposed using the received signal RSSI value, along with the in-phase and quadrature-phase components of the received BLE signals at a single time instance, to simultaneously estimate the angle of arrival at all anchor points using machine learning [16]. With this method, a localization accuracy of 70 centimeters was achieved in simulations. Since the accuracy of UWB localization relies on distance estimation based on the arrival time from the anchor point to the UWB tag, Poulose et al. proposed a new location prediction model that employs a deep learning approach using Long Short-Term Memory (LSTM) networks to predict user location [17]. Simulation results indicate that the proposed model achieves an average location error of seven centimeters, compared to conventional UWB localization approaches.

 

There are other articles more closely related to the work done. These are just examples:

-        PIR Probability Model for a Cost/Reliability Tradeoff Unobtrusive Indoor Monitoring System (10.1007/978-3-319-61949-1_7)

-        Passive Infrared Sensor-Based Occupancy Monitoring in Smart Buildings: A Review of Methodologies and Machine Learning Approaches (10.20944/preprints202401.1924.v1)

-        PIR Sensor-Based Arduino Home Security System (e-ISSN: 2456-9860)

-        PIR sensors (10.1145/2737095.2742561) – technological

-        A Multi-Resident Number Estimation Method for Smart Homes (10.3390/s22134823)

-        PIR sensors: characterization and novel localization technique (10.1145/2737095.2742561)  

 

This section is redundant as it refers again to systems that use wearable devices, unless specified otherwise. My suggestion is to remove it and expand the section on previous works that utilize PIR devices: Tracking with radio wave technologies such as Wi-Fi, BLE, UWB, offers high accuracy in locating subjects but requires them to constantly wear a tag that emits signals. Additionally, these tags need regular recharging or battery replacement, which can be burdensome for elderly individuals. In contrast, ambient sensor tracking is device-free and less burdensome for the elderly. It employs more sophisticated methods such as deep learning, Hidden Markov Models (HMM) [21, 22], Kalman filters [23, 24], and particle filters [25-27] to achieve high accuracy. These methods estimate the next coordinates based on the detected direction and velocity of motion, repeating the correction process with each detection. This approach often requires a dense arrangement of sensors or the integration of multiple sensors into a single module.

This section is redundant and is already part of the introduction (and should be placed at the end of the introduction). My suggestion is to remove it and to “revisit” the introduction. This research aims to enable the implementation of the system with minor modifications to existing equipment, such as motion sensor lights. Therefore, instead of using special sensor arrays or densely arranged sensors, we envision a scenario where single sensors are sparsely placed at locations such as doorways, kitchens, or room entrances. The 167 sensor data consists of binary information indicating the presence or absence of a subject. Since the data volume is small, it can be processed by a computer with limited processing capabilities such an edge computer. Due to potential false positives and omissions caused by blocking time during sensing, we propose a method to filter out incorrect information using the walking speed and the connectivity relationship between sensors.

3 – SYSTEM

GENERAL NOTE: Captions are descriptions of figures. The text of a caption should be limited to a few words. If there is more information to convey, it should be included in the text with reference to the figure.

 

The caption for Figure 1 is too long. It should not describe what should be in the text (e.g., "This data can then be shared with family members, caregivers, and healthcare providers to facilitate the early detection of changes in physical activity levels.")

 

The acronym GASS does not exist. Why include it? For the HW module? This module has nothing innovative and it doesn't make much sense to design a system that can be found in many versions commercially (from 868/915/950 MHz wireless to WiFi to Bluetooth). Describing the module does not add value to the work. It is suggested to use existing acronyms in the literature.

 

This is not the place for: “This study employs PIR sensors, but in the future, we plan to enhance functionality by integrating various sensors such as ultrasonic, proximity, and color sensors”. The article uses PIR. That in the future we want to add more should be put in the future developments and not in the architecture.

 

Caption 2 is too long and contains text that should not be included here. Furthermore, it is unnecessarily redundant.

 

Limiting the range of PIR has two disadvantages: the first is that it must be increased in density to avoid uncovered areas, the second is in the dynamics of detection (if the subject passes too quickly, they may not be detected) [see: PIR Probability Model for a Cost/Reliability Tradeoff Unobtrusive Indoor Monitoring System]. Neither of these issues is addressed. Also, what effect does covering with aluminum foil have compared to the Fresnel lens used in PIR and the discretization of detection (the zones, as indicated in the specification, are 64 and change when acting on the lens)? No impact?

 

WARNING: Figure 3b shows that something is wrong. This sensor should have a center of symmetry and as many axes of symmetry as there are sides of the polygon. However, symmetric points with respect to the center should have similar characteristics.

 

Once again, the caption is too long and contains information that should be in the text.

 

Caption 4 is not good and, furthermore, replicates what is written in the text.

 

In section 3.2, it talks about criticality in communication latency: "A significant issue for detection time is the time lag between when the sensor detects an event and when the edge computer collects and timestamps the data." These times should be on the order of tens of milliseconds. Assuming a movement speed of 60 cm per second (reasonable for elderly people without motor difficulties), a transmission delay of 30 msec (congested network) means admitting an error of 1.8 cm. Even with poor temporal alignment, it should not be a problem unless there are errors in the transmission architecture. I don't think the assertion made in the relationship between detection granularity and subject movement dynamics is reasonable. Radically change or eliminate.

 

In section 3.3, the statement "It has been found that the walking speed measurements obtained with the GASS module are highly accurate, as demonstrated in previous work [28]." is not sufficient to claim that it works well. The work is by the same authors. Self-referencing is not enough to make a claim credible.

 

The acronym Open Binary Area Sensor (OBAS) is another acronym from the authors. Again, it is not clear why it is needed and what difference there is from a PIR to which a computing and transmission platform has been added. My opinion is that it is completely useless: doing edge or local computation changes nothing given the sampling speed. It's a wasted resource.

 

The figures confirm the fragmentation of the detected areas and that the system may fail to recognize due to the crossing speed of the monitored area, and there are three obvious problems. 1) PIRs have different parameters (sensitivity, blind time, Pulse counter, window time) that change the effectiveness of detection; they may not if the person touches the visibility area or passes too quickly. 2) The subject may accidentally avoid sensitive areas and not be detected for a long time, 3) when moving from one zone to the next, the person can take any route and the times expand. Talking about accuracy and precision, in these situations, is not credible.

 

Caption 6, 7, 8, 9 are too long.

 

The statement. "Figure 9(b) visualizes the movement trajectory by estimating and plotting the subject’s location every 0.1 seconds along with the shortest path between two OBAS units." … but what is it? Is it a simulation where moving an agent generates the activation sequence? Or is it an estimated path based on the state of the PIR? It talks about simulation and shows an obvious choice by the authors that is not justified by any real data. When a PIR is activated, it makes no sense to trace a segment under it (unless it is a moving agent). The person could have arrived at any angle, could have simply brushed the area, or passed under it.

4 – LIBIS

The house is very narrow, and the walls are not clear (simple scenario). Moreover, it is not clear whether the region to the right of the figure, as well as that on the left, is reachable by the person. Why these spaces has been included if not useful? And, if they are useful, how is the person moving there detected? Excluding these zones, the house is 8 meters by 6 meters, with a corridor of 1 meter (the detection area). In some areas of the house (such as the living room), it is quite possible that a person may not pass under the covered areas. Seeking accuracy in measurement by limiting the PIR's field of view requires an increase in the number of PIRs: this does not seem to be the case. There are many empty areas: even if the crossing probability is operated on, this approach is specific to the analyzed case.

 

The statement: "the transit time between OBAS units can be accurately measured, and walking speed can be determined from the distance and transit time between OBAS units" raises many doubts. The detection system has its latency, which can be on the order of tenths of seconds and can worsen if the user passes in the peripheral areas of the sensing area (they might not be detected at all). Accurately measuring the space between transit units can only be an estimate; for example, from inside to center. Since Velocity is a composite quantity, this too is an estimate and can be significantly wrong.

 

The statement: "Previous studies have reported an average indoor walking speed of 1.16 meters per second for individuals around 60 years old [31]. Another study has also reported a maximum walking speed of 1.4 to 2.2 meters per second for the same age group [32]. Additionally, walking speed decreases with age. Therefore, in this study, a threshold of 2.0 meters per second is used. If the walking speed between OBAS units exceeds this threshold, it is judged as a false positive" does not correspond to data in the literature nor to my decade-long experience, both in home support and in nursing homes. The reported speed (which is approximately 5-6 km/h) is plausible outdoors and for younger people. The speed indoors decreases significantly due to continuous starts and stops, curves, and slowdowns in crossings doors. People over 65 have an average speed between 0.8 and 1.2 meters per second. The experiments conducted at home, however, show speeds between 0.6 and 0.8 meters per second. 2 meters per second (7 km/h) is objectively too much.

 

The algorithm works on a graph of adjacencies: the authors should use known terminology. The correction mechanism is very simple and not very functional. The authors could take inspiration from convolutional algorithms for information decoding. When formula (2) is not satisfied, the detection by OBASB is judged as a false positive. If a false positive is detected, the state change table is modified to adjust the detected time and blocking time from 'ON' to 'OFF.' This case is simple and does not involve sequences of evaluations. In fact, it may happen that the moving person performs a series of passages that do not activate all the PIRs, or that the PIRs are in a blocked state, etc., determining a series of situations not interpretable with simple formula 2. The sequence is a Markovian process since the probability of being in a state depends on the previous state; the path is the one probabilistically most likely compared to all admissible ones. The example of 4.2 Implementing Late Binding to Refine Trajectory Data Corrections is not supported by a method formally described in the respective section. The authors should describe the method very accurately and use an example to clarify it. Using an example instead of the method, in my opinion, is not scientifically appropriate. Finally, the authors do not describe well the admissible cases and the possible failures (for example, data transmission failure, temporary failure, etc.). This could be limitations that the authors intentionally did not address, but there is no trace of a systematic discussion of events and possible situations..

5 - SIMULATIONS

 

The simulation environment generates synthetic data that may not be credible. It is not clear how the authors generated anomalous and critical situations. There is no real evidence.

Reviewer 2 Report

Comments and Suggestions for Authors

1. Lines 37-38: "However, few existing services aim to constantly monitor the health status…"

This description is subjective and improper. A detailed investigation of related works is necessary.

2. The introduction did not clarify the technical issue addressed in this paper or provide a focused research goal. Simply using a different sensor or a sensor array based approach without requiring the devices to be worn or addressing privacy concerns, is insufficient.

3. Lines 181-183: "Such monitoring enables the early detection of changes in physical activity levels, which leads to highly accurate health monitoring."

I disagree with this point. First, while the sensor used here can detect the speeds of moving objects, it cannot ensure the data collected is significant if the user is not living alone. Second, although the back-end system can calculate changes in walking speed, it cannot guarantee "highly accurate" monitoring unless the user consistently walks at the same speed at home.

4. Lines 192-194: "This study employs PIR sensors, but in the future, we plan to enhance functionality by integrating various sensors such as ultrasonic, proximity, and color sensors."

This seems to contradict your previous research motivation of favoring simple, inexpensive sensors. 

5. Section 3.1.1: Limitation of The PIR Sensor’s Detection Range

An accuracy verifying mechanism or error reports are needed here. 

6. A section comparing the proposed method with other existing devices or methods, along with an experimental results report, is necessary.

 

Reviewer 3 Report

Comments and Suggestions for Authors

This manuscript's work on home monitoring of the elderly is an important topic. However, the organization of the manuscript is more like an engineering project report than an academic paper.

(1) The authors' work is more like a case study, lacking generalized models, strategies and algorithms.

(2) Lack of quantitative metrics to verify that the method proposed in this paper performs better than previous literature.

(3) I note that the sensors involved in this manuscript are cheap and the arrangement scheme is simple, so why not carry out an experimental validation of the proposed method?

(4) How did the authors solve the problem of spatio-temporal synchronization between multiple sensors, which is a very important practical issue.