Fast detection, tracking and classification of moving objects

A system was developed to detect, track and classify moving objects in natural environments with a video camera fixed on a mobile platform. The platform can freely pan horizontally and tilt a few degrees, but cannot move. The configuration is simple, but capable of monitoring a large field of view. The system assumed that the moving objects were indicated by outliers after compensating for the background motion due to the camera ego-motion, which conformed to the 2-D affine model in the system configuration. The analysis of a the motion model parameters used robust parameter estimation to eliminate the effect of the outliers. Experiments show that with a moving object buffer to track a moving object between frames, the system can focus on the object by controlling the camera motion.