Research And Realization Of Vehicle Detection And Tracking Based On Online Learning In The Surveillance Video

In recent years, the rising number of vehicles and illegal traffic problems increasedsimultaneously, it put forward higher requirements for the management of road traffic. Due to theunique advantages of intelligent video surveillance as well as continuous research and developmentof machine learning, computer vision and image processing technology makes this field gain moreand more attention.The thesis mainly uses the online learning theory to detect and track the vehicles in the urbantraffic and community surveillance video. Finally detect the illegal vehicle parking and reversedriving. The main work is completed as follows:Moving target detection section: compare with multiple background modeling algorithm andproposes a general framework for background modeling in surveillance video system, theframework can automatically remove the noise and detect the foreground objects, then put them asoriginal traning objects in online learning system. The thesis put forward the shadow detectionalgorithm based on online learning.Static target detection section: It proposes multiscale detection method based on color andtexture features, the weight of each feature is learned using online learning. Experiment resultsdemonstrate the method can effectively solve the problem of illumination vary.Tracking and classification section: Introduce the mainstream online learning tracking algorithmand then put forward the tracking and classification algorithm in this thesis: using the moving andstatic detected targets as the initial templates for the classifier and then using the each frame onlinelearning to renew the classifier simultaneously. Using the sparse coding theory solve the falsedetection and missed problem. Get the target vehicle and its status information and combine withthe rules of real scenes can effectively complete the illegal parking and reverse driving dectction.