Contextual and Multi-Scale Feature Fusion Network for Traffic Sign Detection

The traffic sign detection, as an important part of the automatic driving system, requires high accuracy. In this paper, we proposed an end-to-end deep learning network, named the Contextual and Multi-Scale Feature Fusion Network, for traffic sign detection. The model consists of two sub-networks: the Weighted Multi-scale Feature Learning network (W-net) and the Contextual-Attention Learning network (C-net). The W-net extracts multi-scale features and calculates the weights of each feature map layer to detect traffic signs under different scales. The C-net learns the contextual attention mask of interference items and concatenates it with the multi-scale feature, which reduce the detection false efficiently. Compared with several state-of-the-art traffic sign detection methods, our proposed model outperforms others on extensive quantitative and qualitative experiments.