Still visualization of object motion in compressed video

Visualizing motion of objects in a scene is important in video analytics, summarization, sports, entertainment, and security. In this paper, we propose an efficient framework for constructing a sprite-based visualization of the object's trajectory from compressed video. A background sprite is generated by using a limited amount of information from a compressed video stream - motion vectors from each frame, and pixel values from a select subset of frames. Global motion parameters are estimated using a combined motion vector-based and pixel-based approach, and then used to map decoded pixel values to the sprite. Meanwhile, the trajectory of a moving object is estimated using an efficient method based solely on decoded motion vectors and coding modes, and overlaid onto the sprite to indicate the path traversed by the object. Experimental results demonstrate both efficiency and effectiveness of the proposed method.