Dealing with Low Quality Images in Railway Obstacle Detection System

Round 1

Reviewer 1 Report

The study is acceptable as it is.

Author Response

Thank you for your time to review our paper.

Reviewer 2 Report

The authors of the manuscript touch upon the important task of increasing the probability of detecting obstacles located on the railway track in front of the train. To solve the problem, the manuscript proposes a new method for processing blurry images.
Motion is cited as one of the reasons for blurry images.
It is not clear that only the movement of the camera was taken into account?
Was the movement of the observed object taken into account or not? For example, the movement of a car crossing a railway crossing in front of a train. Also, for example, in the article (DOI: 10.1504/IJCIS.2019.103015) the problem of preventing the collision of high-speed trains and drones operating on the railway, which could also cross the train path, was considered. This issue, in my opinion, needs to be clarified in the manuscript.

Author Response

We could not agree more with you that the movement of the observed object is one of the elements that can alter the appearance of blur in images. To test the proposed algorithm, we have used images that have blurriness regardless of the reason for which they were made (camera, train and observed object movement). Additional information is provided in fourth paragraph of Introduction, and in second paragraph of Section 2.

Thank you for pointing us to this paper which deals with the same topic presented in our paper. It is now included in position [5] as one of the publication providing an overview of the current situation.

Reviewer 3 Report

In this paper, an improved edge detection framework is proposed, which can be used to detect poor quality images taken during train driving, which is helpful to the collation of data sets. This paper introduces too much related work, occupying more than half of the space, it would be nice to have more information about what you do. After careful review, the following opinions are formed:

1) The key words of abstract is not accurate enough, and even edge detection is not mentioned.

2) The threshold value in Table 1 has not given sufficient reasons to explain its origin, so it is hoped that relevant calculation formula can be added.

3) The innovation points are not clear enough, please highlight the innovation of the framework of this paper.

4) In the experimental result display part, if the fuzzy graph screened in the part can be listed, the experimental results will be more intuitive.

5) In the comparison part of the experiment, the control group is set too little, so it is suggested to increase the comparison data of Sobel, Candy and other operators.

Author Response

First of all, we would like to express our gratitude for your insightful and helpful remarks on our article. Detailed responses to your suggestions are given below:

1) We added several new keywords.

2) The calculation formula for threshold value is now added in the revised version of our paper. It is labelled as equation (13).

3) The innovation points are now clearly highlighted with new paragraph (last one) in the Introduction part.

4) We have replaced the fuzzy graph in the experimental result display part with modified one, as you suggested.

5) We have expanded our control group in the comparison part of the experiment with additional algorithms, as you suggested.

Round 2

Reviewer 3 Report

My concerns have been adressed. Moderate English changes required.

Author Response

The paper is now proofread by a university English professor. All the changes are marked using Track changes option in attachment.

Author Response File: Author Response.pdf