Adaptive Polar-Grid Gaussian-Mixture Model for Foreground Segmentation Using Roadside LiDAR
Round 1
Reviewer 1 Report
The paper is interesting, but it needs to be improved further.
Main remarks
Introduction
L30-59-Consider adding more references at international level.
The objetives are well explained
Methods
Well explained and the equations are explicit.
L339-348The authors must justify why the sample used isn't low.
Author Response
Thank you for reviewing our manuscript and offering valuable advice. In accordance with your suggestions, we have made the following revisions to our manuscript:
1. L30-59: Consider adding more references at international level.
Response: Thank you very much for your advice, we added more references at international level, such as references [2-7].
2. L339-348: The authors must justify why the sample used isn't low.
Response: Thank you for your constructive comments, we added in the manuscript that common road scenes in cities are two-way lanes and intersections. We selected a typical location for data collection for each urban road scene, which can cover the common conditions of urban roads. The selected scene contains complex traffic facilities and background environments, including buildings, trees, and pole-like traffic facilities. In addition, we selected a time period with large traffic flow, and a total of 8191 frames of data samples were collected. The time period of a traffic light on the two roads is about 2-3 minutes, and 4152 frames and 4039 frames are collected in the scenes, which is about 6 minutes. The collected data samples can cover a traffic light cycle of urban roads, so the sample used isn't low.
Reviewer 2 Report
It is important to detect objects such as vehicles and pedestrians for intelligent transportation systems. This paper proposed an adaptive polar grid Gaussian mixture model (APG-GMM) in point cloud method to improve the accuracy of foreground and background segmentation based on roadside LiDAR. The experimental results show that the proposed algorithm can segment the foreground and background well while reducing the amount of calculation and time complexity. However, there are some errors and shortcomings that need revision shown as the following.
1. The aims of this study should be clearly explained and your research hypothesis could be explicitly proposed in the Introduction section. The contributions 1 and 2 are similar?
2. There are a lot of grammar and expression errors needed to have their paper language edited and proofread.
3. Each parameter in formulas should be explained. Equations (12) and (13) are repeated.
4. Some abbreviations should be given their full spelling words when first used, such as APG-GMM.
5. Lines 306, 329, 330, 335, and 350: lidar to LiDAR?; Line 347: Lidar to LiDAR?
6. Line 343 and 347: grammar errors, and so on.
7. Figure 15: The subtitles for (c), (d), (e), and (f) are missing.
8. The structure of this manuscript needs to be modified. That is, Section 3. Data Acquisition and 4.1. Evaluating Indexes should be moved to place in Section 2. Materials and Methods.
9. What are your results for clustering vehicles and pedestrians?
10. It's better to analyze the shortcomings of your current study and future work to improve them.
Author Response
We really appreciate you for your carefulness and conscientiousness. Your suggestions are really valuable and helpful for revising and improving our paper. According to your suggestions, we have made the following revisions on this manuscript:
1. The aims of this study should be clearly explained and your research hypothesis could be explicitly proposed in the Introduction section. The contributions 1 and 2 are similar?
Response: We have added the purpose of the study and research hypotheses in the Introduction section. Contribution 1 and contribution 2 are dissimilar. Contribution 1 proposes a new polar grid for dividing the data of roadside LiDAR, while contribution 2 proposes an adaptive polar grid Gaussian mixture model in each polar grid. The language of contribution 1 has been revised to make the expression more accurate.
2. There are a lot of grammar and expression errors needed to have their paper language edited and proofread.
Response: We have edited and proofread for grammar and expression.
3. Each parameter in formulas should be explained. Equations (12) and (13) are repeated.
Response: We explained parameters in formulas. Error in Equations (13) have been corrected.
4. Some abbreviations should be given their full spelling words when first used, such as APG-GMM.
Response: We checked and modified the abbreviations in Line33, Line51 and Line 220.
5. Lines 306, 329, 330, 335, and 350: lidar to LiDAR?; Line 347: Lidar to LiDAR?
Response: Both lidar and Lidar of the manuscript are uniformly modified to LiDAR. Line 347 Lidar has been modified to LiDAR.
6. Line 343 and 347: grammar errors, and so on.
Response: Line 343 and 347: grammar errors have been corrected, and we checked other errors.
7. Figure 15: The subtitles for (c), (d), (e), and (f) are missing.
Response: We have added subtitles for Figure 15.
8. The structure of this manuscript needs to be modified. That is, Section 3. Data Acquisition and 4.1. Evaluating Indexes should be moved to place in Section 2. Materials and Methods.
Response:
Thank you for your valuable comments. After careful attempts, some of the structures have been modified. 4.1 Evaluating Indexes has been moved to place in Section 2. Materials and Methods as 2.5. Evaluating Indexes. Section 3. Data Acquisition introduces the parameters of roadside LiDAR sensors we used, the location of data acquisition, and how the data was collected and preprocessed, etc. And Section 2. Materials and Methods mainly expound on the proposed method in this paper. We tried to move Section 3 into Section 2 as 2.6, but the revised article structure was unsatisfactory.
9. What are your results for clustering vehicles and pedestrians?
Response: Thanks for your valuable suggestion. We have added the clustering results of vehicles and pedestrians in Section 5. In this paper, we segment vehicles and pedestrians as foreground, and cluster the volume of vehicles and pedestrians based on the segmentation. The main contribution of density adaptive DBSCAN for target clustering is to improve the speed of clustering without losing too much clustering accuracy. Further, we compare the results of density adaptive DBSCAN clustering proposed in this paper with traditional 3D DBSCAN.
10. It's better to analyze the shortcomings of your current study and future work to improve them.
Response: In Section 6. Conclusions, we have analyzed the shortcomings of our study and added future work.
We hope the revised manuscript is now acceptable to you. If not, we are glad to receive any further feedback which we shall continue to apply our best effort to address. Thank you.
Reviewer 3 Report
Dear authors,
I’ve read your paper focusing on the application of a new segmentation method based on a scanning lidar sensor in a roadside context. On my opinion the paper lacks on several aspects. I suggest few modifications which are listed below.
- the introduction must be improved with other relevant references. In this section research activities, such as field activities, should not to be presented.
- In materials and methods authors figures must be cited in numerical order (e.g: fig 5 first and the fig. 6).
- Sub-sections of Section 5 could be merged
- There is a lack on the discussion section: the absence of a comparison with other relevant references to highlight the paper innovation solution. Please add it.
- The references are poor.
Author Response
Thank you for handling our submission and offering us an opportunity for revision. We have addressed all the comments and detailed all the changes made to the manuscript.
1. The introduction must be improved with other relevant references. In this section research activities, such as field activities, should not to be presented.
Response: Line 92 – Line 100: We have improved the introduction and added relevant references, such as references [38 - 24]. We have removed some field activities.
2. In materials and methods authors figures must be cited in numerical order (e.g: fig 5 first and the fig. 6).
Response: Line 189: We have adjusted the order of the figures and added a cite of Figure 5.
3. Sub-sections of Section 5 could be merged.
Response: Line 511 and Line 528: Sub-sections of Section 5 have been merged.
4. There is a lack on the discussion section: the absence of a comparison with other relevant references to highlight the paper innovation solution. Please add it.
Response: Line 500 – Line 505, and Figure 22: We have added comparisons with other references in Section5. Discussion to highlight the paper innovation solution.
5. The references are poor.
Response: We have edited and proofread the references.
Reviewer 4 Report
Roadside LiDAR-type systems are sensors for detecting moving objects, such as vehicles and pedestrians, and have many advantages for uses and accuracy. And its effective applicability for future intelligent transportation is critical. The experiments are very interesting and appropriate, although a greater number of cases would be reasonable to reinforce the investigation since the two exposed are scarce for a full reliability of the system. It would also be interesting to place the sensor on fixed signaling elements on highways and urban roads, such as the arms of traffic lights, and on signals suspended from bridge structures. Being able to monitor from these control points in different urban environments and in different time slots throughout a day, depending on the intensity of traffic, would reinforce the studies. The exposed method is only tested at intersections and urban roads with heavy traffic flow.
In reference to the methodology, I recommend to reinforce and expose more clearly the dynamic updating of the background point cloud image.
Finally, it should be noted in the conclusions that the experimental results show that the proposed algorithm has improved the segmentation of the foreground and background to group vehicles and pedestrians; that is, in comparison with other cases and methods applied in previous investigations.
Attend to the annotations indicated by the reviewer in the review document - pdf file-.
Comments for author File: 
Comments.pdf
Author Response
We really appreciate you for your carefulness and conscientiousness. Your suggestions are really valuable and helpful for revising and improving our paper. According to your suggestions, we have made the following revisions on this manuscript:
（The number of lines such as Line 274 is the revision mode of the Word.）
In reference to the methodology, I recommend to reinforce and expose more clearly the dynamic updating of the background point cloud image.
Response: Line 274 – Line 276, Line 304 – Line 305: We added a description of dynamic updating of the background. We update the parameters of Gaussian distribution based on the polar grid to complete the dynamic updating of the background.
Finally, it should be noted in the conclusions that the experimental results show that the proposed algorithm has improved the segmentation of the foreground and background to group vehicles and pedestrians; that is, in comparison with other cases and methods applied in previous investigations.
Response: Line 583-Line 589: we have added in Sectionv6. Conclusions that proposed algorithm has improved the segmentation of the foreground and background to group vehicles and pedestrians, and compared with other methods such as 3D-DSF and 3D DBSCAN to show our improvements.
Attend to the annotations indicated by the reviewer in the review document - pdf file-.
Response: We have revised the manuscript based on the comments in the pdf file.
(1) Figure 2: For a better understanding, it should be reinforced with a graphic scheme that indicates the position of the teams in the two cases (a) (b).
Response: Line 80: we added the position of LiDAR in Figure2(a), (b).
(2) Figure 4: A brief explanation of the structure of the work is required, that is, the three indicated phases: input - model - segmentation.
Response: Line 184: we added a brief explanation of the schematic block diagram.
(3) A better explanation is convenient, clarification based on Figure 7.
Response: Line 248-Line 252: We modified the explanation based on to Figure 7.
(4) Line 470 – Line 474: Better indicate why the choice of the number of points per frame (100?).
In general, a clarification of the process is required.
Response: Line 470 – Line 474: We modify the expression, calculating the number of points per frame in order to find the appropriate polar grid parameters.
(5) Figure 14: This comment requires an exposition in the section with contrasted arguments.
Response: Line 473 – Line 474: We have added exposition to illustrate this issue.
(6) Finish the sentence appropriately.
Response: Line 581: This sentence was redundant and we removed it.
(7) Therefore, it should be noted that the improvement is with respect to the 3D-DSF comparison algorithm, since there are still difficulties in segmentation with complex backgrounds.
Response: Line 585 – Line 590: We added the comparison of the proposed method and the 3D-DSF method in our experiments, showing the improvement of the proposed algorithm.
Thank you.
Author Response File: 
Author Response.docx
Round 2
Reviewer 2 Report
The authors have addressed all comments I concerned.
In addition, are Fig.1, 2, and 3 yours? If so, the should be moved to Section 2, or please add citation references.
The paragraphs below Eqs (1) and (2) should be left typesetting with no space.
Author Response
Thank you for handling our submission and offering us an opportunity for revision. We have addressed all the comments and detailed all the changes made to the manuscript.
1. In addition, are Fig.1, 2, and 3 yours? If so, the should be moved to Section 2, or please add citation references.
Response: Fig.1, 2, and 3 are all ours, and we moved Fig. 2 and 3 into Section 2. Since Fig. 1 illustrates the advantages of LiDAR for image sensors, we did not find a suitable place to place it in Section 2, so Fig. 1 was deleted.
2. The paragraphs below Eqs (1) and (2) should be left typesetting with no space.
Response: We have modified the format of the paragraphs below Eqs (1) and (2), and checked the other formats.
Besides, we added more references on roadside LiDAR background filtering methods in recent years and removed some references that were not particularly relevant.
Reviewer 3 Report
Dear authors,
I have read your revised paper and here below are reported some suggestions:
- In the introduction are still present figure 1,2 and 3 which referred to some activities or study step. Please move such figures in the Mat/Meth section.
- In Discussion section there is still the absence of a comparison with other relevant references. No references are present in this section: the comparison and the discussion with other solution adopted in other papers is important to highlight your research innovation. Lines 500-505 doesn’t provide any references. The discussion section must be expanded.
- On my opinion the references used in this paper are not sufficient
Author Response
Thank you for reviewing our manuscript and offering valuable advice. In accordance with your suggestions, we have made the following revisions to our manuscript:
1. In the introduction are still present figure 1,2 and 3 which referred to some activities or study step. Please move such figures in the Mat/Meth section.
Response: We have moved Figures 2 and 3 into Section 2. Since Figure 1 illustrates the advantages of LiDAR for image sensors, we did not find a suitable place to place it in Section 2, so it was deleted.
2. In Discussion section there is still the absence of a comparison with other relevant references. No references are present in this section: the comparison and the discussion with other solution adopted in other papers is important to highlight your research innovation. Lines 500-505 doesn’t provide any references. The discussion section must be expanded.
Response: Thank you for your valuable comments. We have added references to Section 5. Discussion and added more content. Our method is mainly compared with the 3D-DSF method, and experiments show that our method has better segmentation accuracy on complex urban roads. Besides, our method is compared with 3D DBSCAN, and the experimental results show that our method improves the running speed. In the original Line 500 -505, now Line 524 has added references.
3. On my opinion the references used in this paper are not sufficient
Response: Thanks for your valuable suggestion. We added more references on roadside LiDAR background filtering methods in recent years and removed some references that were not particularly relevant.
