摘要：

在交通视频场景中,由于环境、设备安装角度等因素的影响,可能会引起车辆相互遮挡的情况,从而造成车辆检测及跟踪的误差。为此,基于轮廓特征点,提出一种重叠车辆检测与分割方法。利用背景差分法获得目标区域,运用Freeman链码检测目标区域的边缘轮廓点,通过链码对确定特征点。根据目标区域的特征点数及占空比进行重叠车辆的判断,若重叠,则对目标区域进行凸包分析,寻找最优分割点分割重叠车辆。实验结果表明,与基于椭圆拟合的方法及基于凹性分析的方法相比,该方法不需要车辆形状以外的先验知识,能较为准确地分割重叠车辆,具有较好的适应性。

关键词: Freeman链码, 特征点, 运动矢量, 车辆分割, 重叠检测

Abstract:

In traffic videos,the impact of environment,installation angle of equipment and other factors may cause vehicle occlusion,resulting in the error of vehicle detecting and tracking.Hence,this paper presents a method based on contour feature points to detect and segment the overlapped vehicles.It uses the background difference method to obtain the target areas,detects the contour points of target area by Freeman chain code,and identifies the feature points by chain code pairs.The number of feature points and the occupancy ratio of target area are used to judge the overlapped vehicles.If they are overlapped,convex hull analysis is carried out in the target area to find the optimal segmentation points and segment the overlapped vehicles.Experimental results show that compared with related methods based on concavity analysis and ellipse fitting,this method can segment the overlapped vehicles more precisely and has better adaptability without any prior knowledge except the vehicle shape.

Key words: Freeman chain code, feature point, motion vector, vehicle segmentation, overlap detection

中图分类号: 

• TP391.41