Road Intersection Recognition via Combining Classification Model and Clustering Algorithm Based on GPS Data
Round 1
Reviewer 1 Report
The paper reads well. Contribution is OK. However, the approach needs to be tested with recent DL methods. These are mentioned below. Show the efficiency compared to these approaches.
1. Vehicular trajectory classification and traffic anomaly detection in videos using a hybrid CNN-VAE Architecture, IEEE Transactions on Intelligent Transportation Systems, 2021
2. Temporal unknown incremental clustering model for analysis of traffic surveillance videos, IEEE Transactions on Intelligent Transportation Systems 20 (5), 1762-1773, 2018
3. Queuing theory guided intelligent traffic scheduling through video analysis using Dirichlet process mixture model, Expert systems with applications 118, 169-181, 2019.
Also refer some recent survey papers and discuss them.
1. A survey of real-time road detection techniques using visual color sensor , Journal of Multimedia Information System 5 (1), 9-14, 2018
2. Trajectory-based surveillance analysis: A survey, IEEE transactions on circuits and systems for video technology 29 (7), 1985-1997, 2018.
Author Response
RESPONSE LETTER
Manuscript ID: ijgi-1800372: “Road intersection recognition via combining classification model and clustering algorithm based on GPS data”
Authors: Yizhi Liu, Rutian Qing, Yijiang Zhao* and Zhuhua Liao
Dear editor and reviewers:
Thank you very much for the valuable comments and suggestions which benefit us much. We have carefully examined all the constructive comments and tried our best to refine the paper and explain the issues clearly.
We nearly modify all parts of the manuscript with blue pens, especially the abstract and Section 2-4. The followings are the comments, our responses, and the detailed revisions that we have made.
Finally, we would like to thank all of you very much for your constructive comments and positive support on this manuscript.
Yours sincerely,
Yizhi Liu, Rutian Qing, Yijiang Zhao and Zhuhua Liao
Aug 21, 2022
Response to Reviewer 1 Comments
Overall comments: The paper reads well. Contribution is OK.
Response: Thanks for your comments.
Point 1:However, the approach needs to be tested with recent DL methods. These are mentioned below. Show the efficiency compared to these approaches. 1. Vehicular trajectory classification and traffic anomaly detection in videos using a hybrid CNN-VAE Architecture, IEEE Transactions on Intelligent Transportation Systems, 2021. 2. Temporal unknown incremental clustering model for analysis of traffic surveillance videos, IEEE Transactions on Intelligent Transportation Systems 20 (5), 1762-1773, 2018. 3. Queuing theory guided intelligent traffic scheduling through video analysis using Dirichlet process mixture model, Expert systems with applications 118, 169-181, 2019. Also refer some recent survey papers and discuss them. 4. A survey of real-time road detection techniques using visual color sensor, Journal of Multimedia Information System 5 (1), 9-14, 2018.5. Trajectory-based surveillance analysis: A survey, IEEE transactions on circuits and systems for video technology 29 (7), 1985-1997, 2018.
Response 1: Thank you very much for your constructive comments and positive support on this manuscript. We append these papers in the references and discuss them in Section 4.4. We will compare their efficiency in the future work.
Author Response File: Author Response.pdf
Reviewer 2 Report
This paper presents a new method to recognize intersections from GPS data by combining two standard approaches. The paper is basically well written and organized. However, there is room for improvement in the abstract by modifying it to clear the limitations of the conventional method and the contribution of the proposed method.
The proposed method consists of a combination of several existing methods. If possible, add ablation studies of those sub-methods.
Detailed comments are shown below.
1.
> Abstract:
> (2) Only one algorithm is applied.
I understand that it means that one is selected from several possible algorithms. I understand that these algorithms have their advantages and disadvantages. If so, you should explain the strengths and weaknesses of those algorithms and clarify the reason for your choice.
2.
> Abstract
> Furthermore, the intersections’ radiuses are neglected.
What are the drawbacks of ignoring the intersection radius? If it's a performance degradation issue, explain intuitively and concisely why it happens.
3.
I understand that it is your contribution to propose "a new method" for radius calculation. However, what are the advantages of these methods over existing methods? Once this becomes clear, the contribution of the proposed method will be clearer to the reader.
4.
> The state-of-art methods are divided into clustering-based methods and classifying-based methods.
This fact should be clarified in the Abstract section.
5.
> In this paper, we manually label 110 intersections in the experimental area of the OpenStreetMap, and then each trajectory point is labeled.
Do you think this number of labeled samples is sufficient? If so, explain the reason why it is sufficient.
6.
> In this paper, road intersection recognition involves detecting the center coordinate of the road intersections and computing the radius of the road intersection.
It is recommended to revise the "abstract" section to clear the motivation as described here.
Author Response
RESPONSE LETTER
Manuscript ID: ijgi-1800372: “Road intersection recognition via combining classification model and clustering algorithm based on GPS data”
Authors: Yizhi Liu, Rutian Qing, Yijiang Zhao* and Zhuhua Liao
Dear editor and reviewers:
Thank you very much for the valuable comments and suggestions which benefit us much. We have carefully examined all the constructive comments and tried our best to refine the paper and explain the issues clearly.
We nearly modify all parts of the manuscript with blue pens, especially the abstract and Section 2-4. The followings are the comments, our responses, and the detailed revisions that we have made.
Finally, we would like to thank all of you very much for your constructive comments and positive support on this manuscript.
Yours sincerely,
Yizhi Liu, Rutian Qing, Yijiang Zhao and Zhuhua Liao
Aug 21, 2022
Response to Reviewer 2 Comments
Overall comments: This paper presents a new method to recognize intersections from GPS data by combining two standard approaches. The paper is basically well written and organized. However, there is room for improvement in the abstract by modifying it to clear the limitations of the conventional method and the contribution of the proposed method. The proposed method consists of a combination of several existing methods. If possible, add ablation studies of those sub-methods. Detailed comments are shown below.
Response: Thanks for your good advices. We modify the manuscript according to your detailed comments.
Point 1: > Abstract:> (2) Only one algorithm is applied. I understand that it means that one is selected from several possible algorithms. I understand that these algorithms have their advantages and disadvantages. If so, you should explain the strengths and weaknesses of those algorithms and clarify the reason for your choice.
Response 1: It is a good advice. We have modified it with another sentence: “How to capture the points around intersections for clustering has great impact on the accurateness of intersection detection.” And we add a “Discussion” part in Section 4.4. In this section, we explain it with the following sentences: “Road intersection detection based on GPS data can divide into three phrases: feature point selection, feature extraction, and feature point clustering. Section 4.3 gives us three observations: (1) DBSCAN outperforms K-means and AHC; (2) Spatial features do better than geometric features; and (3) Combining geometric and spatial features can obtain further improvement. Therefore, feature point selection is worthy of much focuses. The existing methods always concentrate on turning points selection. Turning points are around intersections, but they are a little far from center coordinates of intersections. Last but not the least, the number of turning points after selection is too little.”
Point 2: > Abstract> Furthermore, the intersections’ radiuses are neglected. What are the drawbacks of ignoring the intersection radius? If it's a performance degradation issue, explain intuitively and concisely why it happens.
Response 2: Thanks for your good advices. We have modified it in the abstract with the following sentences. “Road intersection recognition contains detecting intersections and recognizing its scope. There are few works on intersections’ scope recognition. …” Recognizing intersections’ scope accurately is vital to create high-definition map.
Point 3: I understand that it is your contribution to propose "a new method" for radius calculation. However, what are the advantages of these methods over existing methods? Once this becomes clear, the contribution of the proposed method will be clearer to the reader.
Response 3: Thanks for your good advices. Existing methods use clustering algorithms to cluster GPS points and generate cluster centers. Then the distance between the cluster center and the furthest point in the cluster is calculated to calculate the intersection’s radius. For a cluster, there will be outliers far from other points in its interior, which is not conducive to calculating the radius of the intersection.
In this paper, Delaunay triangulation is used to measure the adjacent relationship between points in the cluster, and outliers are found through the adjacent relationship. Outliers are usually far away from other points in the cluster. We compute the edges of each adjacent point in the Delaunay triangulation. Mark edges with longer distances. Therefore, the outliers are first removed based on this distance. Then, the furthest distance between the cluster center and the remaining points is calculated to obtain the radius of the intersection.
Although the intersection radius calculation method proposed in this paper is more complex than the traditional method, the accuracy is higher than the traditional method.
Point 4: > The state-of-art methods are divided into clustering-based methods and classifying-based methods. This fact should be clarified in the Abstract section.
Response 4: It is a good advice. We rewrite the Abstract section.
Point 5: > In this paper, we manually label 110 intersections in the experimental area of the OpenStreetMap, and then each trajectory point is labeled. Do you think this number of labeled samples is sufficient? If so, explain the reason why it is sufficient.
Response 5: Thanks for your good advices. We expand the existing area. The number of marked intersections is increased to 221. Table 1 compares the parameters of the two experimental areas. Table 2 gives experimental results in different regions. Seen from Table 2, the performance reduces less with increasing the number of intersections.
Table 1. The parameters of the experimental areas
Parameters |
the original area |
the expanded area |
|
The number of intersections (manually marked) |
110 |
221 |
|
The time period of GPS data |
2016.11.01~07 (7 days) |
2016.11.01~07 (7 days) |
2016.11.01~14 (14 days) |
The longitude and latitude of the area’s lower left corner |
(104.0661, 30.6625) |
(104.06, 30.66) |
|
The longitude and latitude of the area’s lower left corner |
(104.09, 30.680) |
(104.11376, 30.72009) |
|
The range of an area |
2.3km * 1.9 km |
5.2km * 6.7km |
Seen from Table 2, we observe two aspects:
(1) When the number of intersections grows bigger, the Precision improves but the Recall grows down, and the F-measure reduces less.
(2) If we increase the time period of GPS data from 7 days to 14 days, the number of GPS points increases too. But the performance of the proposed method varies little. Therefore, the proposed method is robustness to some changes on the numbers of intersections or GPS points. In the future work, we will increase the magnitude of changes and further observe its robustness.
Table 2. Comparisons of the proposed method in different areas
Matching area (m) |
Evaluation metrics |
the original area (7 days data) |
the expanded area (7 days data) |
the expanded area (14 days data) |
5 |
Precision (%) |
66 |
68.6 |
66.2 |
Recall (%) |
61.8 |
58.4 |
60.2 |
|
F-measure (%) |
63.8 |
63.1 |
63 |
|
10 |
Precision (%) |
91.3 |
95.7 |
96 |
Recall (%) |
85.5 |
81.4 |
87.3 |
|
F-measure (%) |
88.3 |
88 |
91.5 |
|
20 |
Precision (%) |
97.1 |
99.5 |
99.5 |
Recall (%) |
90.9 |
84.6 |
90.5 |
|
F-measure (%) |
93.9 |
91.4 |
94.8 |
|
30 |
Precision (%) |
99 |
100 |
100 |
Recall (%) |
92.7 |
85.1 |
91 |
|
F-measure (%) |
95.8 |
91.9 |
95.3 |
Point 6: > In this paper, road intersection recognition involves detecting the center coordinate of the road intersections and computing the radius of the road intersection. It is recommended to revise the "abstract" section to clear the motivation as described here.
Response 6: Thanks for your good advices. We revise the “abstract” section.
Author Response File: Author Response.pdf
Reviewer 3 Report
The manuscript is about the development of a new approach to recognizing road intersections. It is an interesting study. The authors should consider the following issues in improving the manuscript.
1. The related works section seems like a compilation of the methods used in the literature without further explanation of the lessons learned from the studies. Table 1 partially complements the discussion by showing the precisions of the previous works. At what distances were those precisions obtained? The precisions obtained by the authors' method were not better than the others at 10 meters.
2. The organization of the manuscript is not according to the journal's recommendation. Section 3 and the first paragraph of section 4 should be under Material and Methods while section 4 should be under Results and Discussion.
3. Why was Tang's method selected for the comparison of center coordinate detection and radius computing?
4. Section 4 should be expanded with more discussion in the context of the existing literature. Are there any limitations in this study?
5. The manuscript needs moderate copy-editing. For example, in line 14, "There are more spatial features of GPS data should be helpful" should be rewritten.
Also, in line 16, "Addressing at these issues" should be rewritten.
In line 52, "it is not accuracy enough" should be checked.
In line 54, "As we all known" should be checked.
In line 75, "better the geographic features" should be rewritten.
The list is not exhaustive, other sections of the manuscripts should pass through an editorial oversight.
Author Response
RESPONSE LETTER
Manuscript ID: ijgi-1800372: “Road intersection recognition via combining classification model and clustering algorithm based on GPS data”
Authors: Yizhi Liu, Rutian Qing, Yijiang Zhao* and Zhuhua Liao
Dear editor and reviewers:
Thank you very much for the valuable comments and suggestions which benefit us much. We have carefully examined all the constructive comments and tried our best to refine the paper and explain the issues clearly.
We nearly modify all parts of the manuscript with blue pens, especially the abstract and Section 2-4. The followings are the comments, our responses, and the detailed revisions that we have made.
Finally, we would like to thank all of you very much for your constructive comments and positive support on this manuscript.
Yours sincerely,
Yizhi Liu, Rutian Qing, Yijiang Zhao and Zhuhua Liao
Aug 21, 2022
Response to Reviewer 3 Comments
Overall comments: The manuscript is about the development of a new approach to recognizing road intersections. It is an interesting study. The authors should consider the following issues in improving the manuscript.
Response: Thank you very much for your good advices. We modify the manuscript step by step.
Point 1: The related works section seems like a compilation of the methods used in the literature without further explanation of the lessons learned from the studies. Table 1 partially complements the discussion by showing the precisions of the previous works. At what distances were those precisions obtained? The precisions obtained by the authors' method were not better than the others at 10 meters.
Response 1: It is a good advice. We modify the “Related work” section, and append a column of “The matching area’s radius” in Table 1.
Point 2: The organization of the manuscript is not according to the journal's recommendation. Section 3 and the first paragraph of section 4 should be under Material and Methods while section 4 should be under Results and Discussion.
Response 2: Thanks for your good advices. We modify it as you said.
Point 3: Why was Tang's method selected for the comparison of center coordinate detection and radius computing?
Response 3: Thanks for your good advices. We add a sentence “Tang's method is one of the typical methods in the field.” in line 77. Tang's method uses geometric features of GPS points to detect intersection center coordinates (no spatial features are used). The method in this paper uses the geometric features and spatial features of GPS points, and the experimental results show the importance of spatial features for the detection of intersection center position. Tang's intersection center location detection and intersection range calculation are continuous work. In addition, both Tang's method and the method in this paper use "the distance between the cluster center and the farthest point in the cluster" as the radius of the intersection. Therefore, this paper chooses Tang's work as the comparison method.
Point 4: Section 4 should be expanded with more discussion in the context of the existing literature. Are there any limitations in this study?
Response 4: It is a good advice. We add the “Discussion” section after Section 4.3.
Point 5: The manuscript needs moderate copy-editing. For example, in line 14, "There are more spatial features of GPS data should be helpful" should be rewritten.
Also, in line 16, "Addressing at these issues" should be rewritten.
In line 52, "it is not accuracy enough" should be checked.
In line 54, "As we all known" should be checked.
In line 75, "better the geographic features" should be rewritten.
The list is not exhaustive, other sections of the manuscripts should pass through an editorial oversight.
Response 5: Thanks for your good advices. We nearly modify all parts of the manuscript with blue pens, especially the abstract and Section 2-4. These parts include the following sentences.
- We replace “There are more spatial features of GPS data should be helpful” by “Besides geometric features, spatial features explored from GPS data and the interactions among all features are also important to represent intersections’ semantics more accurately”.
- We replace "Addressing at these issues" by “To solve the preceding problems”.
- We replace "it is not accuracy enough" by “it still has improvement room”.
- We have deleted "As we all known" as you mentioned.
- We replace "better the geographic features" by “better than the geographic features”.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
The authors have revised the manuscript according to the comments. They just need to take a second look and copy-edit the manuscript.