基于运动检测的多车辆跟踪方法研究

针对交通监控场景中多目标粘连造成跟踪上的困难和前后两帧车辆关联困难,提出了区域运动相似性分割方法和相似度关联矩阵的解决方案；在运动目标检测过程中, 首先使用背景差分法提取运动区域,经过消除缺口、空洞和分离等处理,在运动区域所在范围内进行块匹配搜索和局部光流计算区域运动矢量,然后使用模糊聚类方法对运动矢量区域融合,完整的分割出粘连运动目标；在目标跟踪部分,目标跟踪建立在目标关联的基础上,提出建立连续两帧目标间距离和局部二元模式相似度关联矩阵的方法进行运动目标标定,从而实现多目标关联；使用公共视频库的图像序列进行测试,所提算法都能实现连续的跟踪和准确的运动目标分割,且处理速度快,表明了算法具有鲁棒性和适用性。

Research on Multiple Moving Vehicles Tracking Method Based on Motion Detection

In view of the difficulties in tracking the overlapping targets in traffic surveillance scene and the difficulty of the vehicle correlation in two consecutive frames, the solutions of the region motion similarity segmentation and the similarity correlation matrix are proposed,background substraction is used for detecting targets,after eliminating gaps, holes and separation, etc. ,motion vector is calculated in the motion region by block matching search and local optical flow method, and then the fuzzy clustering method is used to implement the regional integration of motion vectors, which completes adhesion segmentation of the moving target, and reduces the amount of computation.In the part of tracking, target tracking is established on the basis of the target correlation, and the target matching is completed by using similarity correlation matrix of distance and local binary pattern features between frames , so as to realize the multiple target correlation. By using image sequences in PETS to test, The proposed algorithm can achieve continuous tracking and accurate target segmentation,and the processing speed is fast,which shows that the algorithm is robust and applicable.