The Design of Preventive Automated Driving Systems Based on Convolutional Neural Network

Round 1

Reviewer 1 Report

The main topic of the paper is interesting and up to date; however, the paper has serious shortcomings.

The major criticism goes to unclear contribution and not clearly demonstrated comparison with other existing approaches for object detection and velocity estimation. If the contribution is in comparison of two considered CNN models, more details on assumptions, technical conditions, and mainly, realistic potential for possible use in automated driving systems should be provided. More detailed comments are given below.

For the reasons above, I cannot recommend the paper for publication in its present form.

Questions, comments and suggestions:

1. The authors should clearly explain their main contribution, is it in comparison of existing CNN models, or in methodology?
2. Basic problem formulation is missing, it is introduced only generally.
3. It seems that writing is not adequately balanced. Loss functions are introduced, however, without explanation of variables, the threshold definition in object detecting, accuracy definition and other modelling details are missing as well.
4. All abbreviations should be explained at their first appearance in the text.
5. Is the distance estimation dependent also on the velocity of the car (camera)? What is the estimation precision?
6. Is camera appropriate source of information to detect objects and estimate distances or should a kind of sensor fusion be considered? What are the limits of studied approach – considering weather condition, daylight, camera quality and position?
7. Can the authors discuss in more details the studied CNN approaches within the overall automated driving concept?

Author Response

Please see the attachment. (manuscrpit_revision + respond)

Author Response File: Author Response.pdf

Reviewer 2 Report

Automated vehicles have been regarded as one of the important trends in intelligent transportation systems as they have recently made rapid progress, and various research and development are being conducted to commercialize and enhance safety. In particular, as automated driving technology develops, the importance of technologies and infrastructure for The Design of Preventive Automated Driving systems, such as the detection of surrounding vehicles or the environment and estimation of the distance between vehicles, is increasing. Object detection is mainly performed through Cameras and LiDAR, but as the necessity of improving camera recognition technology increases due to LiDAR's cost and limits of recognition distance, it has made many advances based  on CNN models, one of the deep learning networks, by incorporating artificial intelligence technologies that have recently been utilized in various fields. In this study, we aim to compare the performance of Vehicle Detection and Classification by learning CNN-based Faster R-CNN and YOLO and estimate the distance from the surrounding vehicles through a model more suitable for the Automated Driving System. As a result of the analysis, Faster R-CNN had high accuracy for Object Classification, but it has difficulty detecting, and the processing speed was also slower than YOLO.  YOLO performed detection and classification of most vehicles on the frame with more than 80% accuracy, and the processing speed was also fast, so it was determined to be more suitable for the Automated Driving System, and thus further progressed with Distance Estimation between vehicles. The topic seems interesting, I have following concerns to enhance the quality of the work.

• The title of the paper is too long, Authors should think about it, Good paper has tile shorter not more than 8~10 words that’s depicts the theme of the paper.
• Why Faster R-CNN has better accuracy and why authors select this as compared to other AI algorithms.
• The literature Review work section should need to update so authors should need to update the structure of the paper. Such as AI>ML>DL>Then proposed approach

 Hu, J., Y. Sun, and S. Xiong. "Research on the Cascade Vehicle Detection Method Based on CNN. Electronics 2021, 10, 481." (2021). Khan, M. A. (2021). HCRNNIDS: Hybrid Convolutional Recurrent Neural Network-Based Network Intrusion Detection System. Processes, 9(5), 834.Rani, E., 2021. LittleYOLO-SPP: A delicate real-time vehicle detection algorithm. Optik, 225, p.165818. Avola, Danilo, Luigi Cinque, Anxhelo Diko, Alessio Fagioli, Gian Luca Foresti, Alessio Mecca, Daniele Pannone, and Claudio Piciarelli. "MS-Faster R-CNN: Multi-Stream Backbone for Improved Faster R-CNN Object Detection and Aerial Tracking from UAV Images." Remote Sensing 13, no. 9 (2021): 1670.Maity, M., Banerjee, S. and Chaudhuri, S.S., 2021, April. Faster R-CNN and YOLO based Vehicle detection: A Survey. In 2021 5th International Conference on Computing Methodologies and Communication (ICCMC) (pp. 1442-1447). IEEE.

• The authors did not discuss the data complexity issue. In the proposed method, authors need to provide a solid scientific reason why the traditional feature selection methods are not enough to deal with the problem? why the proposed method provides high accuracy??
• Figure 2. R-CNN structure. Should need to improve with layer-wise structure.
• The Authors need to explain how to handle class imbalance, PASCAL VOC data set. It must be added to the proposed method.
• Authors should draw a Table and compare it with previous approaches.
• All equations should be assigned numbers. And align with the text. One page 6,7 Equations no are missing.
• All figures should be redrawn with high resolutions and different colors.
• Authors should give all experiment parameters, Experimental setup, still few experiments paraments are missing??
• Conclusion and Future work must be updated.

Author Response

Please see the attachment.(manuscrpit_revision + respond)

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The paper was revised according to reviewers' comments.

There are some minor points which are recommended to be considered:

• All symbols (variables) appearing in (2) should be explained, some of them are not clear (e.g. w, h).
• Minor English correction should be made (e.g. lines 105-107 - the introductory paragraph in Section 2.

I recommend the paper for publication after minor revision.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors try to solve my previous comments but this paper still need to improve 

• Authors should need to explain more technically how to deal with class imbalance issues??
• The literature Review work section should need to update so authors should need to update the structure of the paper. Such as AI>ML>DL>Then proposed approach

 Hu, J., Y. Sun, and S. Xiong. "Research on the Cascade Vehicle Detection Method Based on CNN. Electronics 2021, 10, 481." (2021). Khan, M. A. (2021). HCRNNIDS: Hybrid Convolutional Recurrent Neural Network-Based Network Intrusion Detection System. Processes, 9(5), 834.Rani, E., 2021. LittleYOLO-SPP: A delicate real-time vehicle detection algorithm. Optik, 225, p.165818. Avola, Danilo, Luigi Cinque, Anxhelo Diko, Alessio Fagioli, Gian Luca Foresti, Alessio Mecca, Daniele Pannone, and Claudio Piciarelli. "MS-Faster R-CNN: Multi-Stream Backbone for Improved Faster R-CNN Object Detection and Aerial Tracking from UAV Images." Remote Sensing 13, no. 9 (2021): 1670.Maity, M., Banerjee, S. and Chaudhuri, S.S., 2021, April. Faster R-CNN and YOLO based Vehicle detection: A Survey. In 2021 5th International Conference on Computing Methodologies and Communication (ICCMC) (pp. 1442-1447). IEEE.

The conclusion part should need to improve.

There are few experimental parameters that are still missing authors should give all experiment parameters to readers.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 3

Reviewer 2 Report

This paper still needs to improve before further progress.

The authors should add these references in the related work section.

 Hu, J., Y. Sun, and S. Xiong. "Research on the Cascade Vehicle Detection Method Based on CNN. Electronics 2021, 10, 481." (2021). Khan, M. A. (2021). HCRNNIDS: Hybrid Convolutional Recurrent Neural Network-Based Network Intrusion Detection System. Processes, 9(5), 834.Rani, E., 2021. LittleYOLO-SPP: A delicate real-time vehicle detection algorithm. Optik, 225, p.165818. Avola, Danilo, Luigi Cinque, Anxhelo Diko, Alessio Fagioli, Gian Luca Foresti, Alessio Mecca, Daniele Pannone, and Claudio Piciarelli. "MS-Faster R-CNN: Multi-Stream Backbone for Improved Faster R-CNN Object Detection and Aerial Tracking from UAV Images." Remote Sensing 13, no. 9 (2021): 1670.Maity, M., Banerjee, S. and Chaudhuri, S.S., 2021, April. Faster R-CNN and YOLO based Vehicle detection: A Survey. In 2021 5th International Conference on Computing Methodologies and Communication (ICCMC) (pp. 1442-1447). IEEE

Authors should need to update conclusion work with possible future direction drawbacks of the prosed approach.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf