Background Invariant Faster Motion Modeling for Drone Action Recognition

Reviewer-1

Sr.

No.

Comments

Response

1.      

Research on the use of AI for the purpose of recognizing the image obtained from a drone is valuable and very necessary. They already prove the usefulness of this technology. The authors focused on adapting the image analysis technology to the quality and specificity of the material obtained from the drone, which gave impressive results.

Thank you so much for this appreciation.

2.      

Taking into account the labor-intensive nature of the development and implementation of the technology developed by the authors, it is necessary to consider whether the development of appropriate flight performance standards will allow for obtaining data whose quality will be similar to that obtained from the ground.

Thank you so much for this suggestion, although we have experimented with such variety, but somehow it was not included in the previous version of the paper.

We have added one paragraph describing the effect of drone flying height. and other parameters related to the drone. Same can be seen in the line no-498 which is a part of Discussion section of the paper.

Reviewer-1 additional comments and response

S.No

Comments

Response

1.      

Please correct “Aerial and drone surveillance29can”

Corrected as 29 is not required here, we removed it. Changes can be seen at line no-38.

2.      

Please check “EPE loss1”

We have added the full form of EPE loss, now I think it will make sense for the reader.

3.      

Line 154 - Please change the layout of Figure 1. Current is uncomfortable for the reader

We have slightly changed the layout and orientation of the figure-1 for clarity.

4.      

Line 280 - Please check the subsection numbering

For subsection 4.2, incorrectly listed as 4.1, We have updated the number and heading. Changes can be seen at line no-288

5.      

Line 466 - Please check the figure numbering

Numbering of figure 3 and figure 4 is updated as previously it was figure 4 and figure 5 which was not in order.

6.      

Line 471 - Please check the figure numbering

Numbering of figure 3 and figure 4 is updated as previously it was figure 4 and figure 5 which was not in order.

Reviewer-2

Sr.

No.

Comments

Response

1.      

The authors highlighted a novel architecture that eliminates the dependency on Optical flow. It is good work, however there are few comments to improve paper

Thank you so much for this appreciation.

2.      

Activity recognition is mainly discussed in "Multi-user activity recognition: Challenges and opportunities", therefore, I suggest authors to read and discuss this concept with more details with the help of this paper. 

Thanks for this suggestion, this paper is a well-organized summary of various challenges and opportunity in muli-human action recognition in same frame of videos. We read this paper, and added various lines in our introduction and literature section wherever required. Some added lines are.

1. Various approaches for multi-human action recognition is summarized in [56], with the detailed discussion on opportunity and challenges.

 At line no - 47

2. However, applying these models in multi-human present in same scene especially in unstructured environment is little complicated and article [56] gives the detailed discussion on complexity and opportunity of this task.

At line no - 99

3.      

I suggest authors to create table to comparison the existing work in section literature.

Thank you so much for this suggestion, we have added on table in literature section, which compares some key research and dataset proposed in literature for drone-based surveillance.

The changes can be seen in the paper in

Table 1. at page number 3.

Reviewer-3

Sr. No.

Comments

Response

1.      

Do not use first person such as we, us, etc.

Thank you so much for this appreciation. We have reviewed the paper and corrected all the mistake of such type.

For example:

" Therefore, we come up with a novel architecture proposed in this paper. "  à " Therefore, this paper proposed a novel architecture for this."

2.      

 Why is the whole of section 4.1 in italics. Also 5.1 - maybe just a selection that made the whole paragraph instead of the heading...

Thank you for your suggestion,

We have corrected the subheading which were italic before as nonitalic throughout the paper.

3.      

In the discussion and the conclusion (these two could actually be joined), it would be nice to have some more discussion. Not only a summary of the paper, but also of the results and comparison of the performance. What was the failure rate and error in the different model observations? Some statistical analysis. How does this compare to systems and models out there not mentioned in this paper? These are some ideas of things to discuss and add.

We have added on paragraph for the failure cases and limitations of the proposed approach and it can be seen in the paper from line no. 491 to 497. The added paragraph is "However, these performances have certain limitations, as the height of the drone will increase, the human feature changes especially in drone footage captured from top. These experiments were performed on the drone footage captured from our DJI Mavic Pro drone, with camera having 60 frame per second rate. Images and videos were captured from 10 meter to 50-meter height. Some important failure cases arise when we have tested with the dataset published in aerial image of Okutama dataset. Since, the height from where those images were captured are more than 60 meters."

Reviewer-4

Sr.

No.

Comments

Response

1.      

create a formal notation section and include all the definitions.

Section and subsection headers are improperly capitalized

Thank you for the suggestion.

We have corrected the issue related to the section and subsection header throughout the paper. The changes can be seen in the revised paper.

2.      

Improper capitalization of words in the middle of the sentence (ex: Besides, For in line 435)

missing words, incomplete or incorrect statements

Thank you for this suggestion. We have corrected the issue. However, somewhere, it is required as the new line was beginning.

3.      

Line 82 A unique five-class-action dataset for aerial drone surveillance was introduced, with each image containing a single set of actions.  Suggestion: Give the number of 480 clips having 14,400        bframes.  Also, give the distribution of number of objects in each class.  Will this dataset be made public?

Thank you for the suggestion.

We have added one line describing the total number of video clips in the proposed dataset. Added line can be seen at line no – 84. However, the other details of this dataset are described in the dataset section.

A unique five-class-action dataset for aerial drone surveillance was introduced, with approximately 480 number of video clips distributed in five different classes.

Yes, this dataset will be made public, we are planning to create a website where all our dataset will be available, mostly captured from drone.

4.      

Line 190 The dataset contains a total of 480 clips having 14,400 frames. Did you manually label these frames? If yes, this should be mentioned.

We have added the suggestion, as we have manually labeled this dataset into five different classes. Now this statement is clearer.

The added change can be seen at line no – 197.

The dataset contains a total of 480 clips having 14400 frames. The dataset is labeled in five different classes manually, and the number of videos is equally divided between all five classes