Research On Object Tracking In Video Surveillance

Moving object tracking is a hot research field of computer vision. The mostintuitive way of human getting to know the world is visual information. Sequentialframes often contain more information than single static image. Moving objectdetection, firstly is to detect object, then to track in sequential frames, to lay thefoundation for trajectory analysis, events judging, action recognition.This paper is dedicated to research on object tracking algorithm, mainlycompleted the following innovative work：1. propose a particle filter tracking algorithm based on adaptive fusion ofmultiple feature, the algorithm concerns the global representation and the partialrepresentation of object appearance, and combines with adaptive fusion based onrepresentation and discrimination of feature models, that against different scenes anddifferent objects, playing their own role and implementing the complementaritybetween each other. So the algorithm becomes more robust.2. Aiming at varying object appearance in noisy scenes, the original featuremodel could not represent the current tracking object accurately. Updating modelsmay cause drift problems easily. This paper proposes an object tracking algorithmbased on online learning, to a certain extent, overcoming the problem.3. As cameras are widely used in traffic monitoring, vast amounts of video datahave not been able to get to be processed. As a result, there is an urgent need forintelligent transportation system. This paper constructs the intelligent transportationmonitoring system based on the crossroads. The system could monitor the trafficviolation automatically. The system implements vehicle detection, vehicle tracking,license plate recognition, traffic light recognition, violation judging, capturescreenshots and so on. These functions meet the users’ need basically.