基于计算机视觉的视距不良路段会车碰撞预警系统

打开文本图片集

中图分类号：X951;TP391.41 文献标志码：A 文章编号：1671-8755（2025）03-0108-07

Abstract：In order to reduce the incidence of traffic accidents in the road section with bad visual distance，a collision warning system based on computer vision was proposed.Firstly，the OpenCV technology was used for image preprocessing，and the basis of YOLOv6 data set was constructed. Then the coordinate attention mechanism was introduced into the YOLOv6 backbone network，and the deep separable convolutional neural network was applied to train the model to detect the vehicle targets.In older to improve accuracy and stability of the system，the DeepSORT algorithm was integrated into the YOLOv6 to locate and track the fast-moving vehicles.At the same time，the safety time threshold was set and the safety distance threshold was calculated. Then the collsion warning strategy was determined based on this analysis.Finaly，the software and hardware system was co-designed. Then the field tests were carried out to verify the eectiveness of the system，and the test set was applied to evaluate the advantages and disadvantages of the algorithm.The results show that the effective early warning rate of the system is

95% ，and the precision，recall rate，average precision of the algorithm are 94.3% ， 86.6% and 91.2% ，respectively. Compared with YOLOv6 algorithm and YOLOv6 CA algorithm，the precision has increased by 3.40% and 1.62% respectively，the recall rate has increased by 1.05% and 0.35% （20号 respectively，and the average precision has increased by 2.82% and 0.88% respectively. This system can effctively achieve collision warning for vehicles on sections with bad visual distance，and is superior to YOLOv6 algorithm and YOLOv6 CA algorithm. It is expected to be applied in assisted driving.

Keywords： Car collsion warning system； Safety distance threshold； Road with poor visual distance; Computer vision

道路视距不良使驾驶员难以判断对向车辆的距离和速度，增加了驾驶风险，成为会车碰撞事故的主要原因之一。（剩余6876字）