ABSTRACT
Transportation is not only an importantly strategic hub of a country, but also an indispensable part of social life. With the development of science and technology, smart transportation has gradually been changed from a concept only to a reality. In consequence, automatic driving technology has become a hot research topic that attracts a number of attentions. Autonomous driving covers fully automated driving and assisted driving. And automatic driving technology based on deep learning is becoming more maturing and advanced. One of the necessary conditions for the development of automatic driving is to recognize traffic signals firstly. And the application of automatic driving technology in highway traffic has begun to show great achievements. However, research in the railway field is still relatively weak, and railway transportation plays a pivotal role in both national strategy and people's livelihood. Based on this gap and need, this paper proposes a lightweight to improve YOLOR object detection algorithm for railway traffic signal recognition method. The method uses YOLOv4-tiny as the baseline model, selectively adds implicit knowledge used by YOLOR, and improves the loss function to enhance the model's ability to prevent overfitting and solve long tail problem adaptively. In addition, the number and size of anchor frames in the self-made data set are optimized. Compared with the same lightweight YOLOv4-tiny model, AP has increased by 2.44, and the model file size only occupies 31.4M.
- Yelal, M. R., Sasi, S., Shaffer, G. R., and Kumar, A. K. 2006. Color-based Signal Light Tracking in Real-time Video. IEEE International Conference on Video & Signal Based Surveillance, Sydney, NSW, Australia, pp. 67–67, November 2006.Google Scholar
- Kim, J., Cho, H., Hwangbo, M., 2018. Deep Traffic Light Detection for Self-driving Cars from a Large-scale Dataset. 2018 21st International Conference on Intelligent Transportation Systems (ITSC), Maui, HI, USA, pp. 280-285, November 2018.Google ScholarDigital Library
- Zhang, N., He, T., Gao, Z., and Huang, W. 2008. Traffic Light Detection and Recognition in Intersections Based on Intelligent Vehicle. Journal of Southeast University, vol. 4, pp. 517–521,2008.Google Scholar
- Wang, C. Y., Yeh, I. H., and Liao, H. 2021. You Only Learn One Representation: Unified Network for Multiple Tasks. Computer Vision and Pattern Recognition, May 2021Google Scholar
- Girshick, R., Donahue, J., Darrell, T., and Malik, J. 2014. Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation. 2014 IEEE Conference on Computer Vision and Pattern Recognition, Columbus, OH, USA, pp. 580-587, June 2014.Google Scholar
- Girshick, R. 2015. Fast R-CNN. In Proceedings of the IEEE International Conference on Computer Vision (ICCV), pp. 1440–1448, April 2015.Google Scholar
- Ren, S., He, K., Girshick, R., and Sun, J. 2017. Faster R-CNN: Towards real-time object detection with region proposal networks. in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 39, no. 6, pp. 1137-1149, 1 June 2017.Google ScholarDigital Library
- Redmon, J., and Farhadi, A. 2018. YOLOv3: An incremental improvement. April 2018.Google Scholar
- Bochkovskiy, A., Wang, C. Y., and Liao, H. M. 2020. YOLOv4: Optimal speed and accuracy of object detection. Computer Vision and Pattern Recognition, April 2020.Google Scholar
- Bochkovskiy, A., Wang, C. Y., and Liao, H. 2020. Scaled-YOLOv4: scaling cross stage partial network. IEEE Computer Vision and Pattern Recognition, November 2020.Google Scholar
- Srivastava, N., Hinton, G., Krizhevsky, A., 2014. Dropout: A simple way to prevent neural networks from overfitting. Journal of Machine Learning Research, vol. 15, pp. 1929-1958, January 2014.Google ScholarDigital Library
- Yun, S., Han, D., Chun, S., Oh, S. J., Yoo, Y., and Choe, J. 2019. CutMix: regularization strategy to train strong classifiers with localizable features. 2019 IEEE/CVF International Conference on Computer Vision (ICCV), Seoul, Korea (South), pp. 6022-6031, 2019.Google ScholarCross Ref
- Ishida, T., Yamane, I., Sakai, T., 2020. Do we need zero training loss after achieving zero training error? International Conference on Machine Learning (ICML), February 2020.Google Scholar
- Wang, C. Y., Liao, H., Wu, Y. H., 2020. CSPNet: A new backbone that can enhance learning capability of CNN. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), Seattle, WA, USA, pp. 1571-1580, June 2020.Google ScholarCross Ref
- Zhang, J., and Zhai, W. 2022. Blind attention geometric restraint neural network for single image dynamic/defocus deblurring. IEEE Transactions on Neural Networks and Learning Systems, doi: 10.1109/TNNLS.2022.3151099.Google Scholar
- Wang, T., Zhu, Y., Zhao, C., 2021. Adaptive Class Suppression Loss for Long-tail Object Detection. Computer Vision and Pattern Recognition, April 2021.Google ScholarCross Ref
Recommendations
VLS: Vehicle Tail Light Signal Detection Benchmark
ACAI '22: Proceedings of the 2022 5th International Conference on Algorithms, Computing and Artificial IntelligenceMany car accidents are caused by the driver’s failure to accurately and timely identify the driving state of the vehicle ahead. Therefore, it is very important to accurately and timely detect the driving state of the vehicle ahead in an automated ...
Nighttime Brake-Light Detection by Nakagami Imaging
Given the rapid expansion of car ownership worldwide, vehicle safety is an increasingly critical issue in the automobile industry. The reduced cost of cameras and optical devices has made it economically feasible to deploy front-mounted intelligent ...
Green wave-based virtual traffic light management scheme with VANETs
Green intelligent transportation systems GITS have received significant attention in recent years. Traffic congestion, which is an important research topic on intelligent transportation systems ITS, is rapidly becoming one of the most serious problems ...
Comments