基于视频图像的高速公路异常事件实时检测

Real time detection of expressway abnormal events based on video images

为提高高速公路异常事件图像实时检测的准确率,提出一种基于改进YOLOv3的高速公路异常事件实时检测方法。通过采用K均值聚类算法改进YOLOv3目标检测算法的先验框,然后采用HOG特征描述方法解决实时在线跟踪算法(英文简称DEEP-SORT)输入预测框过多而导致的实时性差问题,最后结合高速公路交通参数信息,实现车辆超速和车辆拥堵等异常事件的实时检测。结果表明,相较于改进前的YOLOv3算法,改进后的YOLOv3算法的mAP值均得到了不同程度的提升,平均mAP值提高3.6%;相较于改进前的实时在线跟踪算法,改进后的实时在线跟踪算法跟踪时间更短。由此得出,本改进方法满足高速公路实时检测的需求。

In order to improve the accuracy of real-time detection of expressway abnormal events,a real-time detection method of expressway abnormal events based on improved YOLOv3 is proposed.The K-means clustering algorithm is used to improve the prior box of YOLOv3 target detection algorithm,and then the HOG feature description method is used to solve the problem of poor real-time performance caused by too many input prediction boxes in the real-time online tracking algorithm(DEEP-SORT).Finally,combined with the highway traffic parameter information,the real-time detection of abnormal events such as vehicle overspeed and vehicle congestion is realized.The simulation results show that compared with the original YOLOv3 algorithm,the improved YOLOv3 algorithm has improved the mAP value to varying degrees,with an average increase of 3.6%;Compared with the improved real-time online tracking algorithm,the improved real-time online tracking algorithm has shorter tracking time.It is concluded that the improved method can meet the requirements of real-time detection of expressway.