利用二值描述符的实时目标跟踪算法

Real-time visual object tracking based on binary descriptors

针对目标跟踪过程中的速率低和存储量大的问题,提出了一种新的利用二值描述符特征的快速稳定的目标跟踪算法.该算法首先在保持目标结构信息的情况下,通过寻找最优正交矩阵对样本进行旋转聚类,将样本从欧式空间投影到汉明空间,生成二值描述符.然后在粒子滤波采样的框架下,通过计算目标与候选样本的汉明距离确定目标跟踪位置.实验结果表明,当发生光照、姿态变化和快速移动时,该算法跟踪速度较快,并且能够实现稳定跟踪.

Object tracking often has the problems of low rate and high storage.Therefore,a tracking algorithm based on binary descriptors is proposed.The algorithm retains the original construction information on the samples and projects the samples from Euclidean space to Hamming space in order to generate binary descriptors by searching the optimal orthogonal matrix for rotating cluster.Then under the frame of particle filtering,it is necessary to determine the tracking position by computing the hamming distance.Analysis and experiment show that the proposed tracking algorithm performs rapidly and favorably when the target objects undergo large illumination,pose changes and fast movement.