Detection And Tracking Of Moving Object Based On Video Image Sequences

The research of moving object detection and tracking based on video sequenceshas always been one of the basic and hot topics in the field of computer vision. Itconcerns the technology of many fields such as computer application, imageprocessing, probability and mathematical statistics, and it has been successfullyapplied to intelligent transportation, video surveillance, robotic vision, battlefieldsurveillance and other occasions.Based on the mode of single camera with fixed focal length, this paper makes acertain degree of improvement and perfection to some commonly domestic algorithmsof moving object detection and tracking with the research of basic algorithms ofmoving object detection and tracking.The detection of moving object is to obtain the pixels that represent movingobject area in this frame. Traditional methods of detecting moving object containoptical flow, interframe distraction, and background distraction. Optical flow mehodhas good accuracy of detection, but it has some problems that its morbid equations aredifficult to solve and it has large computation. Interframe distraction doesn’t need toconstruct the background model, and detects the moving object quickly, however, ithas poor accuracy of detection when there is little change in the external environment.Background distraction focuses on the construction of the background model, analysesthe two common methods of background construction: Gaussian mixture backgroundmodel and W4 background model, and proposes the improved algorithm of detectionof moving object based on W4 theory, which can extract more complete outline ofmoving object.The tracking of moving object is to determine the process of moving objectlocation of the current frame. The commonly algorithms used for tracking videosequences contain kalman filter tracking , mean shift tracking and particle filtertracking. Kalman filter tracking needs the model of moving object, and this algorithmbecomes ineffective when the moving object is in the non-linear, non-Gaussian system; Mean shift tracking has poor accuracy of tracking when the object position changes alot in the interframe images; Particle filter tracking can be effectively applied to thenon-linear,non-Gaussian motion system, but its particles are susceptible todegradation,mean while it has large computation. This paper realizes a fast trackingalgorithm of moving object based on the region covariance matrix which is the latestproduct in the computer vision, and through experimental simulation, this algorithmhas good results.