Non-rigid Objects Detection And Tracking In Video Survillance

The research of moving objects detection and tracking is an important branch of thecomputer vision. When the non-rigid targets are moving, it involes deformationsincluding rotation, variation of appearance, etc. The research of the non-rigid objects isimportant for the daily security, abnormal behaviors and accidents. When non-rigidobject deforms, the tracking window changes. Taking the human body for example, thetext studies the situation of non-rigid objects motion based on the color characteristics.First of all, the background subtraction is used for the human detection in this text,and the background which has been built is updated timely to adapt to the change ofbackground. Then for tracking of the human body in the next frames, it marks thedetected human in the original image.Secondly, the template matching method is used for human tracking. The featuretemplate is established based on the color characteristics. In order to reduce the searchrgion of the template in the next frame when template would be matched, the centroidcharacteristcis of human body is extracted, then the position in next frame can bepredicted using Kalman Filtering. And the template could be updated when the templateis compared to the human body which has been tracked. Then information which isextracted form feature points is introduced for restraining the tracking window. So whenthe information of the human body changes, the tracking window changes.Finally, Camshift algorithm is used for tracking human to reduce the computationalcomplexity. The initial window of Camshift algorithm is the marked window from thebackground subtraction automatically. When calculating the color probability distribution,the paper makes use of a different-weights method. Camshift algorithm can adaptive thechange of the window by the method of back project. And in the experiment it has beenproved to be efficient when non-rigid objects are be tracked.