An Adaptive Group of Density Outlier Removal Filter: Snow Particle Removal from LiDAR Data

Round 1

Reviewer 1 Report

First of all, I would like to thank Editorial Board of Electronics for the opportunity to participate in the review of this paper. The paper addresses a topic of great interest (in particular, practical interest in terms of the applications that can be derived from it) as is the filtering -in real time- of snow particles in LiDAR scenes for use in the guidance of non-driving vehicles. The most relevant contribution is the proposal of a new filtering algorithm, which provides highly satisfactory results and can be applied at the speed that this application requires.

As for the paper, I must emphasize that I consider that it has a correct structure and length, and its reading is very easy. The work includes a correct analysis of the problem and of the contributions made in particular by the work (Introduction), a review of the current methods and a proposal of a new method, as well as an application on standard datasets used for this type of studies and a complete analysis for the validation of the results. In summary, the structure can be considered to be correct. On the other hand, the design of the experiment, the development of the method and the application and analysis of results and conclusions are also correct.

In this sense, I consider that the work is susceptible to be published in the current format in the journal Electronics, however, I would like to pose, on the one hand, a question to the authors and, on the other hand, make a suggestion. The question would be related to what would be the behavior of the methodology and the application in low illumination conditions, or rather, how the illumination would affect the test since in the end a very large component is linked to the intensity of the echo received, and on the other hand, I would like to make the suggestion of the incorporation of the tools that allow the application of this methodology, which obviously I consider that not only has application to this particular field.

Finally, I would just like to suggest the revision of some typographical errors that I have been able to locate in my review of the work:
L38 LiDAR
L51 However
L137 between
L322 Please revise (and complete, if the case) it
L330 Please revise (and complete, if the case) it

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

I reviewed the article deeply and although the paper is promising there are a few but important points that should be revised before publication.

1. Please extend the end of the abstract by providing information about how much accuracy, precision, and recall increased with respect to the state-of-the-art methods.

2. The first two paragraphs of the paper seem non- scientific, please revise these paragraphs in a more informative way and with more citations.

3. Most importantly, I can only see one LIDAR cloud in Figure 6. Authors should visualize their findings on more datasets to satisfy the performance stability of the proposed method. Moreover, they should add some zoomed regions to show where their methods succeeded and others fail.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

In this paper, the authors propose an Adaptive Group of Density Outlier Removal filter that can remove snow particles in raw LiDAR point clouds. The proposed method has two filtering steps. First, an intensity threshold is applied to find all low-intensity points in a point cloud for further filtering. Second, low-intensity points with high neighbor point density are filtered out as snow noise.

Even though the work is interesting exportation in snow noise removal, the proposed approach shares great similarities with DDIOR [21] without much non-trivial contribution. The performance comparison also agrees with this concern: DDIOR has a 14.64% better precision, while this method shows a 16.32% higher recall. One would suspect that tweaking the hyper-parameters will negate the performance difference. I suggest the authors make substantial improvements on the method to be different from [21] or on experiments to show that their performance is superior to [21]. Another specific suggestion is that in Table 2: DSOR and DDIOR execution times are n/a. If possible, contact authors or reimplement to get this information.

The writing needs to be improved. Some examples: on page 1, line 5, the full names of ROR and SOR are not provided. Page 1, line 21 LiDAR first appeared, but full name is provided later. Page 1, line 17: “The vehicles can know where they are going and immediately avoid objects, give right decision.” can use some improvements. Page 2, line 51 How-ever. Eq (5): p0 is not defined.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Paper improved through revision. However, there is a minor point that should be revised. 

1. In Figure 6 please add closer looks where improvement achieved.

2. In Figure 7 closer look still not very in formative. Maybe much more zoom and mark where improvements achived.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The authors did meaningful exploration in snow noise removal. After the revision, more experiments are conducted and missing data are added. Experiments show that the proposed method is faster while providing similar performance when compared with other methods, such as the DDIOR. This method may provide a method that has different de-noise characteristic that is useful in some scenarios.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf