结合YOLO检测和语义分割的驾驶员安全带检测

Driver Seat Belt Detection Based on YOLO Detection and Semantic Segmentation

为了通过交通监控自动检测驾驶员是否佩戴安全带,提出一种结合目标检测与语义分割的驾驶员安全带检测算法.首先通过设计轻量化目标检测算法完成驾驶员区域快速定位;然后利用经过剪枝加速的语义分割模型对驾驶员区域进行分割,得出安全带连通域;最后通过判断安全带连通域面积检测驾驶员是否佩戴安全带.在驾驶员区域定位和安全带检测2个数据集上进行训练和测试,实验结果表明,驾驶员区域定位算法在精准度为99.96%时速度为73帧/s,安全带检测算法在准确率为94.87%时速度为305帧/s;该算法在兼顾速度的同时具有较好的精准度.

In order to detect whether a driver wears a seat belt automatically through traffic monitor,a driver’s seat belt detection algorithm based on object detection and semantic segmentation was proposed.Firstly,a lightweight target detection algorithm was designed to locate the driver’s area quickly.Then,the driver’s area was segmented by the semantic segmentation model accelerated with pruning,and the connected area of the seat belt was obtained.Finally,the area of the connected area of the seat belt was judged to detect whether the driver worn the seat belt.The speed of the algorithm is 305 frames/s when the accuracy is 94.87%.The experimental results show that the algorithm has good accuracy while taking into account the speed.