Researches On Rapid Multiple Target Detection And Tracking In Video Surveillance

In recent years, the video monitoring system has been widely used in trafficscene and the public places. The intelligent monitoring technique has been widelyconcerned due to the unaffordable needs that required in the traditional artificialmonitoring technique. As an indispensable part of intelligent monitoring technique,the multiple objects detection and tracking technology has always been a hot spotresearch problem due to the definite link between the performance of the technologyand the ability of the subsequent intelligent processing.In simple situation, the proposed multiple objects detection and trackingapproaches achieve high accuracy rate with real-time operating. However, incomplex circumstances, such as bad detection situation, enormous and dense objects,existing algorithms could not taking the accuracy rate and the real-time operatinginto account simultaneously. Therefore, multiple objects detection and trackingapproach with high accuracy rate and real-time operating Is still faced with manyproblems and challenges.This work in this thesis focus on two aspects: multiple objects detection andmultiple objects tracking, aiming at the real-time operating in complex situationswith shape change background, colossal number of objects and short-time occlusion.Throughing the improvement and optimization of methods, runing time is reduced onthe basis of high accuracy rate. The main work is described as follows:1. Multiple objects detection based on ICF-ASFE: aiming at shape changebackground, long running time, multiple size targets in multiple objects detection,sliding-window objects detection framework is firstly chosen to avoid the influenceof shape change background, secondly, for fast extracting the characteristics ofobjects, Integral Channel Features(ICF) is introduced, thirdly, throughing theextracting of common prior knowledge, the running time is further reduced on thebasis of high detection accuracy rate. Experimental results in INRIA pedestriansdatabase and TownCentre pedestrians monitoring video show that the proposedalgorithm is of good performance in large size image(19201080) with real-timeoperating.2. Multiple objects tracking based on PDO-MCMCDA: aiming at tremendousnumber of objects, targets occlusion, miss detection and false alarm in multiple objects tracking, firstly, the multiple objects tracking is formulated as Maximum aPosterior Probability (MAP) estimation rule; then Markov chain Monte Carlo DataAssociation(MCMCDA) approach is introduced to find the solution in polynomialtime; at the same time, using the scene and object prior knowledge, the proposaldistribution is optimized to further reduce running time. Experimental results insimulation scene, NGSIM intersection vehicles monitoring video and TownCentrepedestrians monitoring video show that the proposed algorithm is of goodperformance in complex circumstances of50more objects, short-time occlusionsand low detection rate and high false alarm with real-time operating.