Chinese Traffic Sign Text Detection Method Based on Improved EAST
Peng Yao, Ying Cuan · 2021
As an important content in the research of unmanned driving and traffic systems, traffic sign recognition has great theoretical value and application prospects. Especially text-based traffic signs, which contain rich high-level semantic information, can provide extremely rich road information. In order to realize the automatic detection of Chinese traffic sign text, an improved EAST text detection method is proposed. This method introduces BLSTM between the output layer and the feature merging layer to enhance the context relationship of the feature vector and improve the receptive field of the network. At the same time, in order to improve the detection accuracy of the algorithm for night images, an automatic image enhancement module is added at the head end of the network. It can detect the light of the input image and automatically improve the contrast of the image that is judged to be poorly lit. Effectively improve the image detection effect of the model in low-light environments. The experimental results show that the improved method achieves satisfactory results on the CTST-1600 data set.