基于多尺度感知哈希特征的目标跟踪算法研究

The research of a target tracking algorithm based on multi-scale hash feature

研究了目标跟踪与检测。考虑到其应用环境复杂,造成对光照变化鲁棒性不强,目标跟踪过程中数据丢失,实时跟踪时检测速度不够的问题,结合哈希指纹特征对光照变化鲁棒性强的特点,提出了一种基于多尺度(multi-scale)感知哈希(Hash)特征的目标跟踪算法,即MHash算法。该算法采用多种尺度扫描窗口对待检测图像进行扫描,利用双线性内插值的方法缩小扫描窗口提取哈希指纹特征;同时利用汉明距离与高斯分布相结合的方法对扫描窗口进行评价,评分最高的作为目标窗口。该算法已应用于人脸、运动车辆及快速和缓慢移动物体的跟踪与检测,并分析了跟踪与检测结果。结果表明,MHash算法具有速度快、跟踪稳定及对光照、目标尺度变化鲁棒性强的特点,能够长期有效地对目标进行跟踪。

Target tracking and detection are studied. Considering that the complex application environments bring the problems of low robustness to illumination variation,data loss due to the change of target scale,and low detecting speed during real-time tracking,a new object tracking algorithm using multi-scale perceptual Hash feature,called the MHash algorithm for short,is proposed based on the characteristic that Hash fingerprint features are robust to illumination changes. The algorithm uses multi-scale scanning windows to detect objects and reduces the dimension of scanning window to access perceptual Hash fingerprint characteristics with the bilinear interpolation method. And it combines the Hamming distance with the Gaussian distribution method to evaluate scanning windows. The window of the highest score is as the target window. The method is applied to tracking and detection of face,sports cars and fast and slow moving objects,and the experiments are analysed. The experiments show that this algorithm is robust against illumination changes and target scale change. It can be effectively used for long-term target tracking.