Research on Vehicle Trajectory Prediction and Warning Based on Mixed Neural Networks

Round 1

Reviewer 1 Report

Presented here article described implementation of the vehicle trajectory prediction system for making drive process more safety. Authors used for it neutral networks. The topic of the paper general fits to the journal topics. However the technical and scientific quality of the paper should be improved. There are general remarks:

• poor literature review
• no schematic of the control stand and view of the camera placed in vehicle
• there is insufficient emphasis on novelty in relation to existing work in this area.
• no statistical information.

Author Response

Dear reviewer,

Please see the attachment.

Thank you.

Author Response File: Author Response.docx

Reviewer 2 Report

This paper aims at detecting the vehicle trajectory prediction based on mixed neural networks.

Even though the application is of interest, the paper in its current form presents the following important issues to be addressed:

-The paper lacks the literature review that is essential to analyze the state of the art, provide the background and demonstrate the innovations introduced with respect to the large existing contributions on the topic. As a result, the Introduction is too short and poor.

-Moreover, the methodology is not well presented, and its description must be improved. To this purpose, the presented processes can be formalized as algorithms or flow-charts in order to better identify and define the different procedure steps. In addition, in the proposed models what is new and what is already existing? Authors do not propose for the first time YOLO and LSTM but they just use these tools for their purposes. What is the paper innovation? Is it just the use of these tools? The paper contribution and innovation must be clarified and put well in evidence.

- The structure described in Fig. 8 is not clear as well as parameters in Table1. The meaning of all the models parameters must be explained carefully along the paper; it is not sufficient to show the parameters names and values in the Tables. The meaning and the function of each layer in Fig. 12 should also be discussed.

- In addition, results demonstrate that the presence of other cars can affect the detection capacity. Is it a promising result to the purpose of driver safety? In table 11, how accuracy, loss and evaluation score are determined?

Other comments:

-why Fig. 7 is proposed again in the results section (Fig.15)? Please insert the figure only once.

- all the acronyms should be spelled out at the first mention. What is the meaning of FPS?

Author Response

Dear reviewer,

Please see the attachment.

Thank you.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Thank you for Your responses to reviews. I have no more comments.

Author Response

Dear reviewer,

Please see the attachment.

Thank you!!

Author Response File: Author Response.docx

Reviewer 2 Report

In the new version, the paper has been slightly improved.

Some of the previous concerns still exist:

-The literature review has been improved for sure but not covering the large existing contributions sufficiently. Here you can find just some examples that can be considered in the topic that have not been cited:

Leibe, K. Schindler, N. Cornelis and L. Van Gool, "Coupled Object Detection and Tracking from Static Cameras and Moving Vehicles," in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 30, no. 10, pp. 1683-1698, Oct. 2008, doi: 10.1109/TPAMI.2008.170.

Woo et al., "Lane-Change Detection Based on Vehicle-Trajectory Prediction," in IEEE Robotics and Automation Letters, vol. 2, no. 2, pp. 1109-1116, April 2017, doi: 10.1109/LRA.2017.2660543.

Xie, H. Gao, L. Qian, B. Huang, K. Li and J. Wang, "Vehicle Trajectory Prediction by Integrating Physics- and Maneuver-Based Approaches Using Interactive Multiple Models," in IEEE Transactions on Industrial Electronics, vol. 65, no. 7, pp. 5999-6008, July 2018, doi: 10.1109/TIE.2017.2782236.

Deo and M. M. Trivedi, "Multi-Modal Trajectory Prediction of Surrounding Vehicles with Maneuver based LSTMs," 2018 IEEE Intelligent Vehicles Symposium (IV), Changshu, 2018, pp. 1179-1184, doi: 10.1109/IVS.2018.8500493.

Park, S. H., Kim, B., Kang, C. M., Chung, C. C., & Choi, J. W. (2018, June). Sequence-to-sequence prediction of vehicle trajectory via LSTM encoder-decoder architecture. In 2018 IEEE Intelligent Vehicles Symposium (IV) (pp. 1672-1678). IEEE.

-Moreover, it is still not evident what is the novelty of the proposed structure and of the results. I would like to see clearly the difference with reference [6] and [7] that make also use of LSTM for car trajectory detection since it is the core of your proposal. How you improved the work done in [6] and [7]? Your results are better than results in [6] and [7]? Are there other references to be compared with your work (for instance, see suggested references)?  

-In addition, results demonstrate that the presence of other cars can affect the lane line detection capacity. Is it a promising result to the purpose of driver safety? This question has not been answered. If you have in memory the lane line area from previous frames, you should be able to keep this area detection even though a vehicle cross one of the two lines for few seconds. Is it reasonable and applicable in your model or you need to detect the line at each frame?

Author Response

Dear reviewer,   Please see the attachment.   Thank you!!

Author Response File: Author Response.docx