Automated Vehicle Classification and Counting in Toll Plazas Using LiDAR-Based Point Cloud Processing and Machine Learning Techniques

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

1. It is suggested to revise the title, since besides lidar, camera and radar are also used.
2. Abbreviations should be defined before using, such as FPFH, BoW and etc.
3. As mentioned in the abstract that the system also can detect the license plate, then why not get the vehicle’s information by the detected license plate directly? Moreover, the detection results are used to determine the fees. Unfortunately, the accuracy is not 100%, which leads wrong charges. Then how to deal with this problem in practice?
4. It seems that piezoelectric sensor doesn’t belong to intrusive technologies.
5. It seems that lidars, cameras are more sensitive to weather than the methods mentioned in line 35-36.
6. Such sensors as lidar, camera and lidar are also embedded into the infrastructure. Why they are classified into the non-intrusive solutions?
7. It is suggested to only list the related studies in Tab. 1. It seems that there many un-related works.
8. The manuscript only introduces the detection and measurement system, but the contribution of this study is not very clear and most contents are basic idea, concept, process, technology and method. It is suggested to focus on the contribution, not the development work.
9. In Tab. 4, what does “No-vehicle”, “Unidentified” and “Distorted” mean? Why not label these objects manually? Generally, all objective truth should be labeled, not just ones that can be detected by the system.
10. The structure of this manuscript is very confusing. It is suggested to re-arrange the content according to the research work. For example, why put “Automatic License Plate Recognition” behind the dataset, while other detection tasks before the dataset?
11. Why not give out the overall accuracy? Is there any special information or conclusion can be achieved from Fig. 29 and 30? Moreover, the results in Fig. 29 and 30 are obtained by the test dataset or application data in practice?
12. It is suggested to compare with the latest methods.
13. There many format errors in references.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The article is a significant contribution to the field of Intelligent Transportation Systems (ITS), offering an innovative solution for the automatic classification and counting of vehicles at toll stations using LiDAR and machine learning. The work is characterized by high technical sophistication, experimental validation in real-world conditions, and compliance with the regulatory requirements of Colombia.

However, I have a number of questions and comments that will help improve and understand the study….

Answer the following questions: 

1) How does the system handle cases where multiple vehicles are in the scanning area at the same time (e.g., in a traffic jam)? Are additional clustering or tracking algorithms used?

2) What modifications would be required for high-speed highways (e.g., >80 km/h)? Has the use of LiDAR with a higher scanning frequency been considered?

3) How does the system classify vehicles with unusual geometry (e.g., construction equipment, trains with a non-standard number of axles)?

4) Explain why the authors chose SVM instead of modern neural network architectures (e.g., PointNet, VoxelNet) that can work better with 3D data. Have you conducted comparative experiments?

5) How were the key parameters determined (e.g., the radius *r* for FPFH, the number of visual words in BoW)? Did you perform grid search or optimization?

6) What is the processing time for a single vehicle on an embedded platform? Does it meet the requirements for real-time processing?

7) Have you conducted direct comparisons of the system's accuracy with purely camera-based solutions (such as YOLO + OCR) under identical conditions?

8) Why is the accuracy for 3-axle vehicles significantly lower than for 2-axle vehicles (F1 60.2% vs 96.9%)? Is it possible to improve the result by combining LiDAR with radar data on the distance between axles?

9) There is no understanding of how the nighttime ALPR issue (accuracy drops to 60%) will be addressed. Have you considered using infrared lighting or thermal cameras?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The papers concerns the use of LIDAR technology for individuating and classifying road vehicles. The research issue is relevant and the paper is interesting. However the current version presents some limits.

The main limit regards the field of application of the proposed methodology.

1) There are specifical insights about tools and procedures for individuating road vehicles in an automated way; despite this strength, the paper does not report a comparative analysis with the other methods for counting vehicles; I suggest the authors to provide a comparative framework with available tools in order to identify the main benefits in adopting the proposed approach;

2) Existing literature proposes the use of different data sources and data fusion methodologies for increasing the level of knowledge of the mobility phenomena; I suggest the authors to discuss about possible integration of the proposed approach with existing data sources (e.g. Floating car data; (see for instance the case of the "Estimation of a Fundamental Diagram with Heterogeneous Data Sources" some results of experimentation are provided for the city of Santander in Spain.

3) I propose the authors to more underline the possible application of the proposed approach for the analysis of transport system; the paper should be considering the data supports of emerging ICT for Transport System Models; see for instance the case of "Estimation of Travel Demand Models with Limited Information" by using Floating Car Data  

4) please provide some quantitative supports for the conclusion section

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The author has addressed all my concerns.

Reviewer 3 Report

Comments and Suggestions for Authors

The new version increases the paper's quality. In this form, the paper can be accepted.