Virtual Grounding Point Concept for Detecting Abnormal and Normal Events in Home Care Monitoring Systems

Round 1

Reviewer 1 Report

The authors have chosen a very timely and important topic for their study, and I’m glad to see this area of research being discussed. I congratulate the authors for a well-conducted study and very nice written paper. Nevertheless, I would like to highlight some points with potential for further improvement.

(1) The title “... in Home Care Monitoring System" is somewhat far-reaching. The analysis was performed on a dataset of standardized, controlled and "staged" fall situations and is therefore difficult to equate with a real-world scenario in a "home care" facility.

(2) With respect to the characteristics selected for the decision-making process, did you check to what extent their selection affects the prediction accuracy of the test data set? Overall, a strong data reduction from a full body description to single projection parameters is performed. Copy-Paste Note: When single time-discrete variables (e.g., Peak) are extracted from time-continuous variables (e.g., 3D acceleration signal-gait stride curve), a large amount of data are discarded. In many cases it remains unclear, if and to what degree single pre-selected variables are capable to represent a sufficient description of a whole-body movement like human locomotion [e.g., Federolf, P., Tecante, K. & Nigg, B. A holistic approach to study the temporal variability in gait. Journal of Biomechanics 45, 1127-1132 (2012). ]. An a priori selection of single time-discrete movement variables may lead to a certain risk of investigator bias and single pre-selected variables might miss potentially meaningful information that are represented by - or in combination with - other (not selected) variables. If I am not wrong, this could be a limitation of the used analysis approach. I think this aspect should be discussed in the paper.

(3) Another aspect that should be discussed is the use or comparison to model-based procedures that place body models in the silhouettes and allow calculation of biomechanical variables such as joint angles. This would have the great advantage that knowledge from biomechanical research on fall risk or the detection of disease patterns could be used.

(4) Another point I would like to mention is the evidence about unique movement patterns to individual persons [e.g., Connor, P., & Ross, A. Biometric recognition by gait: A survey of modalities and features. Computer Vision and Image Understanding, 167, 1-27 (2018).]. Furthermore, a call for an individualised diagnose and therapy in gait analysis is raised and different authors suggest that the analysis and treatment of human gait should take individual needs into consideration rather than trying to fit the individual into the frame of stereotypes and normal reference data. Considering those findings from research on unique individual gait patterns, could you comment and discuss, how the described approach can enable to respect individual needs during analysis of human locomotion.

(5) And in the context of this point, did you ensure that data from one person are either used for training or for testing? Halilaj et al. (2018): “When developing a model, include all the data from one subject (e.g., different trials) in only one set (training, validation, or test dataset) to ensure that the model generalizes well to new data” [Halilaj, E., Rajagopal, A., Fiterau, M., Hicks, J. L., Hastie, T. J., & Delp, S. L. Machine learning in human movement biomechanics: best practices, common pitfalls, and new opportunities. Journal of biomechanics, 81, 1-11 (2018).].

(6) I would also like to see a somewhat extensive discussion of the results of the experiments. Compare the results presented in Table 1 and highlight the advantages and disadvantages of the different approaches. And add a subchapter where you discuss the limitations of the presented approach.

Specific Comments:

(7) Check the consistent use and introduction of abbreviations (e.g., KNN in line 100, R-CNN in line 103, SVM and AdaBoost in line 118, … ).

(8) Rename section from “Some Related Works” to “Related Works”.

(9) Move the aim of the study (described in lines 125-132) to the end of the introduction section and merge the content with the description of the aims described there.

(10) I suggest changing the wording of the third main component (e.g., 3.3. Abnormal and Normal Event Analysis) of the presented system throughout the document, as I find it misleading. The term used suggests to me that a classification into "normal" and "non-normal" gait events is already made here. If I have understood it correctly, however, this is performed using the procedure described in section 3.4?

(11) And the follow-up question is, does abnormal gait event mean a fall? This point should be named consistently and clearly throughout the document.

(12) Use consistent frame ranges in Figures 7-9.

(13) In line 303-304: “In the dataset, 20 videos were randomly…”. What exactly were these 20 trials used for and how do these 20 trials relate to the cross-validation described later?

(14) The authors should specify what is meant by scenario 1, 2, 3, ….

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

This is a good research paper and this paper discusses a novel home care video monitoring system for dependent people. The authors have made a significant contribution. This paper successfully shows the results of the proposed new approach. Overall a good paper, only in a few places there are grammatical errors and that is a very minor error like “missing an article”.

Author Response

Please see the attachment.

Author Response File: Author Response.docx