Research based on improved YOLOv5 algorithm in unmanned driving technology

Abstract

With the rapid development of modern society, people in order to meet their travel needs, the demand for cars increased rapidly. In recent years, traffic accidents have occurred frequently, accounting for the highest death toll in China's workplace safety. According to statistics, in our country, the average number of people who died in traffic accidents always keeps at about one hundred thousand, countless families bear the pain and pain of parting loved ones. Road traffic accidents have become an extremely prominent problem that must be faced squarely. Ninety-three percent of road traffic accidents are caused by improper operation and immature driving technology. To solve this problem, in recent years, more and more scholars began to engage in the research of unmanned driving technology. This can greatly improve the safety of car driving, so as to reduce the occurrence of traffic accidents to a certain extent, to avoid the occurrence of tragedy. This paper mainly describes and improves the target detection technology algorithm in unmanned driving, and proposes an improved NAM-SIOU yolov5 algorithm based on the yolov5 algorithm. The original yolov5 algorithm incorporates the NAM attention mechanism and uses a better SIOU loss function. The experimental results show that the mAP and accuracy rate of the improved algorithm is improved by about 1 percentage point, the accuracy is up to 94.6%, the recall rate is up to 94.2%, the detection time of single vehicle pedestrian road image on GPU is 15.2ms, the detection speed is increased by about 23% compared with the original algorithm, and the real-time detection speed is up to 43 frames/s. The NAM-SIOU yolov5 algorithm can well meet the requirements of high accuracy, fast speed, high real-time performance and other indicators in unmanned vehicle technology target detection.