source code for Context-aware 3-D Mean-shift with Occlusion Handling for Robust Object Tracking in RGB-D Videos
dependency: OpenCV (2.4.9 other versions should also be OK)
To run the code, you should provide two folders named as "rgb" and "depth" which contains the rgb and depth images for tracking. The images are named as 0.png, 1.png, 2.png...
you should also provide a "init.txt" which contains the initial boulding box.
"param.txt" contains the camera parameters: fx, fy, cx, cy
the tracking results are saved as "box.txt", which is formated as:
left_top_corner_x,left_top_corner_y,right_bottom_corner_x,right_bottom_corner_y,occlusion_flag
This code is tested with videos capture with OpenNI for Kinect 1.0, it should also be suitable for Kinect 2.0 or other types of RGBD sensors with camera parameters in "param.txt" modified, but it has not been tested.