Texture-Based Mean Shift Target Tracking Algorithm

In this paper, a texture-based Mean Shift target tracking algorithm is proposed. The MATLAB source code has been publicly shared. The algorithm's core concept utilizes a joint color-texture histogram to achieve robust target tracking. The implementation includes texture feature extraction using gradient orientation histograms combined with color information from the target region. The tracking process involves calculating similarity measures between the target model and candidate regions through Bhattacharyya coefficient comparisons. The Mean Shift procedure iteratively adjusts the candidate location to maximize similarity, with kernel density estimation for spatial weighting. This algorithm demonstrates improved robustness against illumination changes and partial occlusions compared to traditional color-based methods. This research, published in the International Journal of Pattern Recognition and Artificial Intelligence in 2009, provides significant contributions to computer vision by introducing novel approaches for target tracking that combine multiple feature modalities. The open-source MATLAB implementation allows researchers to study the algorithm's kernel functions, histogram computation, and optimization procedures for real-time tracking applications.