改进的MIL跟踪算法代码

Readme for paper "Real-time Visual Tracking via Online Weighted Multiple Instance Learning," Kaihua Zhang, Huihui Song, Pattern Recognition. Author: Kaihua Zhang, HK POLYU Email: zhkhua@gmail.com Procedures >Put the video sequences into file ¡®\data¡¯; >Initialize the position in the first frame in ¡®\data\init.txt¡¯; The setup has the format [x y width height] where ¡°x,y ¡± are the coordinate of left top point of the rectangle. > run ¡°mexCompile.m¡± to generate ¡°mex¡± files > run ¡°Runtracker.m¡± The parameters in the main function ¡°Runtracker.m¡± can be tuned as follows 1.¡°trparams.init_postrainrad¡± is the search radius for the positive sample; This parameter can be set 4~8. If the object moves very fast, a large parameter should be used to contain more positive samples. 2.¡°trparams.srchwinsz¡± is the search radius for the search window at the new frame; This parameter can be set 15~35. If the object moves fast, a larger parameter should be used to contain the object. 3. ¡°lRate¡± is the learning rate parameter. This parameter can be set 0.7~0.95; If the object changes fast, a small ¡°lRate¡± should be used to weight more on the new frames. 4. ¡°ftrparams.maxNumRect¡± is the maximum number of rectangle features. The parameter can be set 4 or 6. If the appearance of the object varies much, 6 should be used to contain more discriminative features. Enjoy it!