Traffic Sign Detection using Deep Random Mapping Autoencoder Network

This paper proposes an effective method for traffic sign detection by employing deep random mapping autoencoder network. The architecture is composed of three modules: coarse detection, fine detection, and candidates clustering. The method utilizes histogram of oriented gradient and color histogram to express the features of traffic signs. Our method is simple and extensible. Results are indicated on both German traffic sign detection benchmark and Belgium traffic sign detection dataset. Our method achieves 99.27% area under the precision-recall curve (AUC) for all categories of traffic signs on German traffic sign detection benchmark, and 93.34% AUC for all categories on Belgium traffic sign detection dataset.