基于特征点匹配和广义Hough变换的扩展目标跟踪方法

Extended Object Tracking Method Based on Feature Point Matching and Generalized Hough Transform

摘要:

针对扩展目标在跟踪过程中存在的因旋转、尺度变化和目标部分区域离开视场时引起的漂移等问题,提出了一种在特征点匹配的基础上结合广义Hough变换的扩展目标跟踪方法。采用ORB特征处理相邻两帧图像,得到初始匹配特征点,然后再用随机采样一致处理筛选出匹配正确率高的特征点作为基础特征点,利用这些基础特征点进行广义Hough变换,提取出一个相对位置固定的跟踪点对目标进行跟踪。实验表明,该方法具有能保持跟踪点的相对位置固定的特性。

Abstract:

In terms of the existing drift problems in the process of extended object tracking caused by rotation, scale transformation and the partial region of the object leaving the field of view, an extended object tracking method is proposed based on feature point matching and generalized Hough transform. Two adjacent frames are processed by ORB feature, and the initial matching feature points are obtained, then the feature points with high matching accuracy are selected as basic feature points by random sampling and consistent processing. These initial feature points are used to carry out generalized Hough transform, and a relative position fixed tracking point is extracted to track the object. Test results show that the proposed method can keep the tracking point relative position fixed for the extended object with rotation, scale transformation and the partial region of the object leaving the field of view.