Collision recognition from a video
System Outline
The system takes a video from a mobile device camera, with an angle to the direction of the movement.
The system analyzes the video according to the following stages:
- (1) Feature Tracking - First we are finding the interest points using Good Features To Track algorithm, then we use the KLT algorithm to track the points throughout the movie.
- (2) Recognition and Distinction Between Static and Moving Objects - Where we find the movement of each iterest point relative to the Center Of Mass (COM). Dynamic points are points that their movement doesn't correlate with the COM movement. Also we designed a feedback system that reduces the weight of the dynamic points in the calculation of the COM.
- (3) Evaluate Trajectory - where we estimate whether the dynamic points are moving towards us, using the size of the Circumscribed circle of those points. Collision is detected in case of expention of this circle.
Finally, if the third stage has consistently decided that there might be a collision, an alert is sounded.