Analysis of Human Accident Video Using Optical Flow With Physically-Based Constraints Available to Purchase

Videos that capture accidents are usually of poor quality: they are likely to be taken from a bad angle, with poor lighting, and contain occluded points. However, the motion data contained in such videos can be very valuable for understanding and preventing accidents. To extract the motion of a body from a video: 1) the points of interest must be identified and 2) point tracking from frame-to-frame must be accomplished. Accordingly, one logical approach is to focus on automated tracking, while allowing a human to identify important points of interest [1].