The Warning System for Speed Cameras on the Road by Deep Learning
In order to reduce traffic accidents, government authorities would install speed cameras for identifying vehicles travelling over the legal speed limit. However, drivers often ignored warnings and were handed tickets, and meanwhile it did not achieve the purpose of setting up the cameras. This paper took the road in Taiwan as an example. By installing a camera in front of the vehicle, it could capture the image of the road ahead. This system was implemented on the NVIDIA Jetson TX2 and used YOLO V3 as the architecture for CNN classification. The images were divided into six categories: “police patrol car”, “warning sign for front speed camera”, “warning slogan for front speed camera”, “front view of radar speed camera”, “back view of radar speed camera”, “traffic sign of speed limit”. When the first five categories were detected, the system would issue a warning, and when “the traffic sign of speed limit” was detected, the number in the middle part would be identified to obtain and update the speed limit. The result of the experiment on the YOLO V3 when outputting 6 categories was that Mean AP was 81.25%, and Average Recall was 85.5%. The result of the digital recognition for speed limit had a precision of 83.5% and a recall of 80.2%.
KeywordsConvolutional neural network Embedded system Speed limit detection Speed cameras alert YOLO CNN
This work was supported by the Taiwan Ministry of Science and Technology MOST 107-2221-E-218-023-MY2.
- 1.Wu, C.C., Wu, S.L.: A novel method for traffic sign detection and recognition. In: Proceedings of the 2017 International Symposium on Novel and Sustainable Technology, pp. B-12–B-13 (2017)Google Scholar
- 2.Yavas, B., Ozbilen, M.M., Yilmaz, Y.: Traffic speed sign recognition with RFID support. In: Proceedings of the 2014 IEEE 8th International Conference on Application of Information and Communication Technologies, pp. 1–4 (2014)Google Scholar
- 3.Luo, H., Yang, Y., Tong, B., Wu, F., Fan, B.: Traffic sign recognition using a multi-task convolutional neural network. IEEE Trans. Intell. Transp. Syst. 19, 1–12 (2017)Google Scholar
- 4.Peemen, M., Shi, R., Lal, S, Juurlink, B., Mesman, B., Corporaal, H.: The neuro vector engine: flexibility to improve convolutional net efficiency for wearable vision. In: Proceedings of the 2016 Design, Automation & Test in Europe Conference & Exhibition, pp. 1604–1609 (2016)Google Scholar
- 5.Road Traffic Safety Portal Site in Taiwan. http://168.motc.gov.tw/News_Photo.aspx?n=n1oOkV$gy2YXqPriGaYUcQ@@&sms=Qck7tL6R3WJmM6srNWCCKA@@
- 6.Tzutalin, D.: labelImg GitHub site. https://github.com/tzutalin/labelImg
- 7.Redmon, J., Divvala, S., Girshick, R., Farhadi, A.: You only look once: unified, real-time object detection. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 779–788 (2016)Google Scholar
- 8.Redmon, J.: Darknet GitHub site. https://github.com/pjreddie/darknet
- 9.Redmon, J.: Installing Darknet website. https://pjreddie.com/darknet/install/