Multiple Moving Objects Tracking under Complex Scenes

This paper presents a method of tracking multiple objects under complex scenes with a stationary camera for real time surveillance. The adaptive background is used for subtracting foreground objects based on chromaticity and gradient. In tracking module, the motion parameters are predicted by Kalman filter, and a "tracking matrix" is built to determine whether the objects occur occlusion, camouflage, etc., if not, the results of scanning the matrix can be used for object clustering and matching. If objects are tracked under occlusion, their color appearance is acquired by the adaptive binning color histogram. Their color histogram and predicted parameters are utilized to form a matching score which is the element of "tracking matrix", the best solution for objects matching is got by the improved Hungarian algorithm. The method has been tested on image sequences to show the validity of the approach