Abstract
Traffic sign recognition (TSR) is a challenging task for unmanned systems, especially because the traffic signs are small in the road view image. In order to ensure the real-time and robustness of traffic sign detection in automated driving systems, we present a single level detection model for TSR which consists of three core components. The first is we use channel shuffle residual network structure to ensure the real-time performance of the system, which mainly uses low-level features to enhance the representation of small target feature information. Secondly, we use dilated convolution residual block to enhance the receptive field to detect multi-scale targets. Thirdly, we propose a dynamic and adaptive matching method for the anchor frame selection problem of small traffic signs. The experimental surface on TsinghuaTencent 100k Dataset and Chinese Traffic Sign Dataset benchmark has better accuracy and robustness compared with existing detection networks. With an image size of 800 × 800, the proposed model achieves 92.9 running at 120 FPS on 2080Ti.
Publisher
Academy and Industry Research Collaboration Center (AIRCC)
Reference41 articles.
1. [1] A H U , A N K , B S Y . A Safety Knowledge Representation of the Automatic Driving System[J]. Procedia Computer Science, 2016, 96:869-878.
2. [2] Takeuchi E , Yoshihara Y , Ninomiya Y . Blind Area Traffic Prediction Using High Definition Maps and LiDAR for Safe Driving Assist[C]// IEEE International Conference on Intelligent Transportation Systems. IEEE, 2015.
3. [3] Muhammad K , Ullah A , Lloret J , et al. Deep Learning for Safe Autonomous Driving: Current Challenges and Future Directions[J]. IEEE Transactions on Intelligent Transportation Systems, 2020, PP(99):1-21.
4. [4] Zhe Z , Liang D , Zhang S , et al. Traffic-Sign Detection and Classification in the Wild[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2016.
5. [5] Girshick R. Fast R-CNN[J]. Computer Science, 2015.