Traffic Monitoring System Based on Deep Learning and Seismometer Data

Round 1

Reviewer 1 Report

This paper investigates the use of seismic methods for traffic monitoring and security purposes. In particular, different machine learning techniques (namely, deep neural network, convolutional neural network, and recurrent neural network) are analysed to extract features of three different sized vehicles (buses, cars, motorcycles) and seismic noise. A data set using geophones to obtain seismic data for different vehicles at the Kyushu University in July 2020 is used to test the methods, finding that the CNN is the most satisfactory approach.

The paper is interesting, well written, and overall well-structured. The aim of the contribution is clear and I think the work has a practical relevance and can stimulate a discussion on an important topic.

1) The introduction is concise but effective enough. More details on other approaches for traffic monitoring should be added to provide a complete overview on this field. In particular, as the authors correctly highlight nowadays it is well recognized that camera-based systems for traffic monitoring and congestions avoidance have many limitations and they appear inadequate to properly handle the traffic control (since they depend on weather conditions and require huge investments on infrastructures to perform a complete coverage of road networks). Moreover, while traffic information provided by cameras is currently available for uninterrupted flow facilities such as freeways, the analysis on urban interrupted traffic flows is more challenging and requires innovative solutions. Therefore, I think the Introduction should (briefly) recall other methodologies on traffic monitoring which are not strictly related with camera-based systems. As an example, see the following contributions:

- R. Carli, M. Dotoli, N. Epicoco, B. Angelico, A. Vinciullo (2015), “Automated evaluation of urban traffic congestion using bus as a probe”. 11th IEEE International Conference on Automation Science and Engineering (CASE 2015), August 24-28, 2015, Gothenburg (Sweden). Art. n. 7294224, pp. 967-972.

- S.H. Ahmed, S.H. Bouk, M.A. Yaqub, D. Kim, H. Song, J. Lloret (2016), “CODIE: COntrolled Data and Interest Evaluation in vehicular named data networks”, IEEE Trans Veh Technol, 65(6): 3954-3963, 2016.

- R. Carli, M. Dotoli, N. Epicoco (2017), “Monitoring Traffic Congestion in Urban Areas through Probe Vehicles: A Case Study Analysis”. Internet Technology Letters. vol. 1 (4), 1:e5, 7 pp. DOI: 10.1002/itl2.5.

- S. Wang, X. Zhang, J. Cao, L. He, L. Stenneth, P.S. Yu, Z. Li, Z. Huang (2017), “Computing urban traffic congestions by incorporating sparse GPS probe data and social media data”, ACM Trans Inform Syst, 35(4): art. 40, 30 p.

2) The methodology seems to be correct, without relevant flaws I can notice.

3) I think much more insights could arise from the analysis. The paper would gain a lot from deepening other aspects on the traffic monitoring. As an example, since the collected traffic data are useful to provide direct information on flow, density, average speeds, and traffic congestion, suitable traffic indicators could also be proposed to properly assess the traffic monitoring. See, e.g., the contribution in Litman (2007) and the already recalled work in Carli et al. (2017).

- T. Litman (2017), “Developing indicators for comprehensive and sustainable transport planning”, Trans Res Rec, 2017(1): 10-15, 2007.

4) As a minor remark, although the reference list is complete enough, more recent works could be added.

Author Response

Thank you very much for your comments and suggestions.

Please find our response in the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Thank you for valuable studies. The paper is well organized and the experiment was designed and performed correctly. Please consider my following remarks and questions:

1. 12 –The vehicle classification techniques which are currently used mainly based on inductive loops and magnetic characteristics of vehicles. The statement in the abstract needs to be corrected.

2. Introduction: you should describe the inductive loop techniques in more details in term of the accuracy of vehicle class detection. In the paper you should also present the comparison of accuracies obtained from different techniques to justify the meaning of presented studies.

3. How did you recognized which waveform match to a given vehicle class? How did you collect the waveforms and how the tests were performed – you should explain this aspects in more details.

4. I recommend to discuss and compare briefly the accuracy of the newly developed method with the automatic visual classification. This aspect should be described in discussion section (no. 5.)

5. You currently describe in discussion section (no.5) the aspect of signal overlapping. The aspect is very significant however I recommend to shift it into the section 4 because it is one of element of results analysis.

6. I think you should compare visual (manual) measurement of traffic flow (numbers and vehicle class distribution) with the results of processing of geophone measurements (your new system). Such comparison shows the real accuracy of the system. I believe that you have all necessary data to fulfill this. It should be included in discussion section.

7. According to figure 8 it is not obvious how many vehicles occurred, eg. On d) there are three peaks for probability obtained for bus while there was one bus. What is the algorithm to identify the number of vehicles on the basis of probability obtained from neural network?

8. How sensitive is the method on vehicle speed? This effect is not mention in the paper, please add some comments.

9. Do you consider to combine all of three considered algorithms DNN, CNN and RNN to improve the accuracy of vehicle class detection? Maybe it is a good way for future studies?

Author Response

Thank you very much for your comments and suggestions.

Please find our response in the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

Thank you for responses. The work is interesting. I wish you good luck in introducing it in practice.