The Study Of Visual Tracking In The Inteligence Video Survelliance

In recent years, video monitoring technology has become the most popular area ofcomputer vision. It has been used in the fields of airports, banks, traffic, railway stations, andparks and so on. So, it is indispensable to public security and highly attended by academicsand management because of its benefit in the business and economic. Visual tracking is oneof key technology and essential to analyzing of target trajectory, understanding of behaviorand recognizing of target. The basic idea of it is to fix the target position and orientation inevery frame and estimate the location of the target in the next frame with the connection ofspace and time, and then generate target trajectories in the image sequence. It has animportant practical significance to the security of city in prevention and sudden affairs ofterrorist. Therefore, target tracking has attracted more and more research as an importantdirection of visual tracking, and become a research hotspot now.In this paper, both CamShift and Particle Filter tracking algorithms are improved.CamShift algorithm can not effectively solve the rotation, occlusion and so on, and is apt tolose target when the target is tracked in the complex background. The new approach ofKalman prediction imbedded into CamShift algorithm was supposed. Firstly, using theKalman filter to estimate where is the most likely location of the moving target in the nextframe. Secondly, using CamShift algorithm to search and match target in a lesser extent.Particle Filter tracking algorithms cam solve prediction of system in the case of nonlinearlyand non-Gaussianity, but it has a huge calculation due to its degradation, so that it is difficultto achieve real-time tracking. To solve above question, the improved method joined MeanShift algorithm in the stage of Important Sample can improve efficiency of particles andreduce the calculationTo verify the robustness and real-time of, two algorithms are simulated by OpenCV inthe Microsoft Visual C++. The result demonstrated that both of new methods can improvethe robustness and real-time of tracking, meanwhile can reduce the time of tracking.