Traffic-Sign Detection and Classification in the Wild

Zhe Zhu, Dun Liang, Songhai Zhang, Xiaolei Huang, Baoli Li, Shimin Hu

Although promising results have been achieved in the areas of traffic-sign detection and classification, few works have provided simultaneous solutions to these two tasks for realistic real world images. We make two contributions to this problem. Firstly, we have created a large traffic-sign benchmark from 100000 Tencent Street View panoramas, going beyond previous benchmarks. We call this benchmark Tsinghua-Tencent 100K. It provides 100000 images containing 30000 traffic-sign instances. These images cover large variations in illuminance and weather conditions. Each traffic-sign in the benchmark is annotated with a class label, its bounding box and pixel mask. Secondly, we demonstrate how a robust end-to-end convolutional neural network (CNN) can simultaneously detect and classify traffic-signs. Most previous CNN image processing solutions target objects that occupy a large proportion of an image, and such networks do not work well for target objects occupying only a small fraction of an image like the traffic-signs here. Experimental results show the robustness of our network and its superiority to alternatives. The benchmark, source code and the CNN model introduced in this paper is publicly available.

Paper  Supplemental Material

Fast R-CNN Model and Results

Tsinghua-Tencent 100K Tutorial

Before downloading, Please enter your information.

name:

Email:

Your orgnization:




If you use  find our code and  data useful, please cite our paper:

@InProceedings{Zhe_2016_CVPR,

author = {Zhu, Zhe and Liang, Dun and Zhang, Songhai and Huang, Xiaolei and Li, Baoli and Hu, Shimin},

title = {Traffic-Sign Detection and Classification in the Wild},

booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},

year = {2016}

}


If you have any questions about the code or dataset, please contact zhe zhu(ajex1988@gmail.com).