Abstract: With the rapid development of artificial intelligence, the application of deep learning in the field of unmanned driving is gradually mature, but the traffic sign detection task still has a lot of room for improvement as a difficult problem. Traffic sign detection under urban road has the characteristics of complex environment, many small targets, many kinds of targets and unbalanced quantity. Aiming at these problems, an improved scheme based on Libra R-CNN is proposed. The target detection network Libra R-CNN is proposed based on balance, which can better deal with the problem of many kinds of targets and unbalanced quantity. In the anchor extraction stage of Libra R-CNN network, GA-RPN is used to generate the anchor. It can produce more accurate and diversified samples during the training period, thus reduce the background influence and deal with the problem that small targets difficult to locate, and improve the detection accuracy. The effectiveness of the method is verified by experiments. The experiment was conducted on MS COCO 2017 and traffic sign dataset. The mAP of the improved Libra R-CNN increased by more than 2.7 percentage points. The experimental results show that the performance of the improved network is significantly improved compared with the original target detection network.

Key words: high-speed on-off valve, pulse width modulation, duty cycle, linear regulation