Study On Multiple Target Detection And Tracking In Video Surveillance

With the rapid development of modern science and technology and the continuousimprovement of the material requirements, the application of the Video Surveillance systemshas penetrated into people’s work and all aspects of the daily life. Multiple target detectionand tracking of Video Surveillance systems is one of the hot issues in computer vision field.Itincludes pattern recognition, image processing, probability theory, computer application andso on. There still remains some key problems to be solved, so the research for this technologyhas very important practical significance and application value.In object detection, we select the improved MRBM background model in backgroundsubtraction, which is a kind of non-parameters modeling method, without designated thedistribution of samples, and has better adaptability. Considering that the original algorithmonly considered the individual pixels time information and ignored its space correlation, thispaper uses the area related samples to model background which can avoid the influence of thesudden movements goal for the background to some extent.The experimental results showthis improved algorithm can better detect moving target information.In multiple object tracking,the main research is the objects of stationary camera.Wefocus on the Camshift algorithm and the particle filter.Camshift bases on meanshft iteration tofind target position, and has adaptive tracking window,small amount of calculation and lowcomputing speed.But when heavy occlusion the tracking effect is not ideal.Particle filter canadapt to the system which is nonlinear and non-gaussian and have steady trackingperformance in complex video scene.But it’s running time grows along with the increasenumber of samples, which can not meet the requirement of real-time tracking. In order tosolve this problem,we embed Camshift in particle filter to process the multi-objective trackingin occlusion condition.According to the aggregation of Camshift,this algorithm can greatlyreduce the particles which have no contribution.So the computational cost particle filter willbe significantly reduced.When the goals are independent we still use Camshift algorithm totrack,otherwise use the particle filter based on Camshift. Experimental results show that themodified algorithm can improve the problem of large calculaed amount which doesn’tinfluence the tracking effect, significantly decrease response time delay.