基于数据关联融合KCF与Kalman滤波的车辆多目标跟踪

Multi-vehicle Tracking Combing KCF and Kalman Based on Data Association

在复杂交通场景下的车辆多目标跟踪,由于车辆之间较高的相似性和交互性,跟踪算法为了保证精度一般都较为复杂,无法满足智能分析应用需求。为此,结合简单有效的数据关联算法和快速精准的单目标跟踪算法,提出在线数据关联的多目标跟踪新方法。方法利用目标检测算法获得的当前目标集,通过关联算法建立目标与已形成轨迹集的关联矩阵,并通过行列耦合原则选出最佳关联对作为关联结果,针对不同的关联结果尤其是漏检和严重遮挡的情况,引入KCF算法与Kalman滤波联合完成目标轨迹的持续更新。实验表明,本文算法可以很好地解决目标误检、漏检以及严重遮挡情况,并且对目标轨迹的实时准确获取,可以为交通视频智能分析提供可靠的轨迹数据。

Due to the high similarity between vehicles,tracking algorithm under complex traffic scenes is time-consuming,which cannot meet the application requirements of intelligent analysis.To deal with it,a multi-target tracking method of online data association was proposed by combining simple and effective data association algorithm with fast and accurate single-target tracking algorithm.Firstly,target detection algorithm was used to obtain the current detection target set.Then,the corresponding association algorithm was used to establish the association matrix of the target and the formed trajectory set.On this basis,the optimal correlation pairs were selected as the association results through the principle of row and col optimization.Finally,in case of target missing and serious occlusion,KCF and Kalman filtering were introduced to achieve continuous tracking.The actual test shows that this tracking method can effectively deal with the problem of target error detection,missed detection and serious occlusion,achieve real-time and accurate acquisition of target trajectory.The research can provide reliable trajectory information data for intelligent traffic video analysis.