Real-Time Human Movement Recognition Using Ultra-Wideband Sensors

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This is a careful study and the manuscript discusses a methodology for real-time human movement recognition based on two legs using Ultra-Wideband (UWB) sensors

Firstly, this paper defines 4 Human Movement State as the base of following model. Secondly, it develops the model of Human Walking Sequence and analyzes the human movement pattern.Then, the methodology for classifying individual states is presented. Finally ,experiments are carried out across four distinct human movement scenarios.

However, there are few problems with the manuscript, which must be solved. In my opinion, revision should be made before it is considered for publication.

(1) The experimental content needs to be expanded. Experiments discuss each state separately, but the actual situation is often comprehensive and complex, and the experiment should consider this more practical situation to verify the recognition accuracy in the case of multiple state combinations.

(2) The article mentions that this method can save computing power more than the deep learning method, but there is no data in the experiment to prove this conclusion, and adding this data will make the experiment more complete and convincing.

(3) The introduction part does not introduce the overall structure and arrangement of the article, so that readers will be very confused when reading it for the first time, and they cannot quickly grasp the structure of the article. Mind map is recommended as a leader of the article.

(4) In figure 8, there are two “step1”, and some similar minor errors in the article that need to be checked again.

(5) In the beginning, the introduction only mentions the shortcomings of other methods, but does not indicate the advancement of this method compared with existing sensor methods.

(6) The figure quality in the paper is not high enough. The work and results of this manuscript are not clearly shown by the figure. For example, the accuracy results of the experiment are not clearly illustrated in figure. And the richness and legibility of figures should be improved, too many figures of the same form appear monotonous. For example, figure 2 is not intuitive enough.

(7) And there are also some presentation questions. The coherence and logic of the language need to be strengthened, so as to enhance the readability of the article. And more mathematical languages are recommended, such as formulas, during the modeling.

Comments on the Quality of English Language

Need to be improved.

Author Response

Dear Reviewer,

First and foremost, I would like to extend my gratitude for your review. Your insights have significantly contributed to enhancing the quality of our paper. Along with this letter, I have attached a PDF file that includes the review comments, as well as a revised version. The modified sections have been highlighted, and all comments have been clearly annotated for easy reference. Thank you once again for your valuable feedback.

Best regards,

Sang C. Lee

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This paper presents a model-based human motion classifier using UWB sensors. The authors propose several efficient handcrafted features, extract from thresholds and FFT, for classifying four types of common motion. The research topic is of decent interest to sensors and human action recognition community and explanation of implementation steps and experiments are clear. The paper is generally easy to follow.

 

It falls slightly short in justifying the usability of the proposed method. It is said the proposal is efficient compared to vision/laser/LiDAR based methods. However, these method do require human to be affixed with receivers, thus, non-obstructive. I trust the UWB only works within a certain range, and would be curious how robust the method is in occlusion and through-wall conditions.

 

There are statements explaining the rationale for targeting the four motion classes: stop, walk, linger, and transition. Many existing works have explored a greater spectrum of classified activities like jumping, jogging, falling, sitting, lying etc. It would be necessary to justify use cases of the said four motion classes.

 

In Figure 6 (b), it seems to focus on walking right away (towards) the UWB anchors. I wonder how the system perform when user is walking in different, random directions with respect to the anchors. Is the proposed method still working if user walks in a tangent direction to the anchors?

 

In Section 5, a series of thresholds and algorithms are specified. Could you provide some ablative evaluation by changing these values slightly? Especially, I wonder how the Cosine similarity method perform when dealing with human agents of different heights and walking speeds.

 

I think the paper fails to address the sampling window requirement for practical classification. If a user lingers for 1s then starting walking for 1s, how would the method segment such sequential activities?

Comments on the Quality of English Language

I think axises are necessary in Figure 2 to allow understanding. Otherwise, a detailed caption should explain what the authors really mean.

Author Response

Dear Reviewer,

First and foremost, I would like to extend my gratitude for your review. Your insights have significantly contributed to enhancing the quality of our paper. Along with this letter, I have attached a PDF file that includes the review comments, as well as a revised version. The modified sections have been highlighted, and all comments have been clearly annotated for easy reference. Thank you once again for your valuable feedback.

Best regards,

Sang C. Lee

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The authors presents investigations of human movement recognition techniques based on UWB sensor modules. The presented work has merits but still needs further improvement. My concerns are summarized as follows:

1. The purpose of the investigation of human motion recognition is for human-robot interactions. More specific application scenarios should provided to justify the significance of the presented work. 

2. The recognition movements of this study is stopped, walking, lingering, and transition between sitting and standing. The selection of movement states should be selected according to specific applications scenarios. 

3. For the UWB modules mounted to human legs, there might be None-Line-of-Sight occlusions, which jeopardize the quality of signal and add to the difficulty of movement recognition. How to deal with the NLoS occlusion, multi-path effect of UWB radio wave signals should be mentioned. 

Author Response

Dear Reviewer,

First and foremost, I would like to extend my gratitude for your review. Your insights have significantly contributed to enhancing the quality of our paper. Along with this letter, I have attached a PDF file that includes the review comments, as well as a revised version. The modified sections have been highlighted, and all comments have been clearly annotated for easy reference. Thank you once again for your valuable feedback.

Best regards,

Sang C. Lee

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Comments and Suggestions for Authors

The authors have made substantial changes to the paper, and all my concerns are properly addressed. The paper can be accepted in the present form.