Research on Automatic Driving Safety Image Recognition Based on Deep Learning

The research is based on the data set BDDI00K, which can well reflect the complex traffic scenarios in the real situation, and build YOLOv4 based on the uncertainty regression of the prediction box to establish the reliability relationship mapping between the prediction box and the real box, so that the model can better detect complex traffic scenarios. In order to overcome the shortcomings of traditional detection methods and meet the actual driving environment, a traffic sign and pedestrian detection method based on YOLOv4 is proposed to provide data support for intelligent analysis and decision-making of safe operation of automatic driving, and provide reliable guarantee for eliminating hidden dangers of safe operation of automatic driving.