基于梯度特征与彩色特征相融合的mean shift跟踪方法

Mean shift tracking algorithm based on gradient feature and color feature fusion

针对mean shift跟踪方法中存在的光照变化不稳定问题,提出了基于梯度特征与彩色特征相融合的mean shift跟踪方法。首先分别提取目标的梯度特征和彩色特征,利用多尺度的相似度计算方法进行特征的匹配,然后通过最大化相似度对目标进行跟踪。通过物体和人体等运动目标的跟踪,验证了改进的跟踪算法在光照变化情况下的鲁棒性优于原有的算法,显著降低了跟踪位置误差。

A novel mean shift algorithm based on object tracking gradient feature and color image fusion is proposed for the illumination unstable problem in traditional mean shift method.Firstly,gradient features and color features of the target are extracted separately,the features are matched using multi-scale similarity calculation method.Then,the target can be tracked by maximizing the similarity.Experiments on the tracking of moving targets such as object and human demonstrate that the proposed algorithm has more robust than the original algorithm under the situation of illumination changes and reduces the tracking position error obviously.