The performance of mean shift algorithm using kernel histograms as tracking cues is always affected by illumination

visual angle and camera parameters. Kernel co-occurrence matrices (KCM) constructed on the concept of gray level co-occurrence matrix (GLCM) were used to represent the target model and the target candidate. Then those matrices were employed as the tracking features in mean shift tracking framework. Some improvements were made in the implementation of the algorithm. First

pixels on the opposite position of the current pixel were treated discriminately to depict the asymmetric characteristics of the object. Second

the KCMs of the target model and the target candidate were normalized to a same integer to improve calculation accuracy. Third

the computation of each pixel weight was modified to improve operation speed. The tracking results of several real world sequences with dark or changing illumination and partial occlusion show that the proposed algorithm can track the target effectively.