Research On Multiple Human Targets Detection And Tracking Algorithm Based On Video

Human-oriented multi-target detectionand trackingtechnologybased onvideo is the basisofcomputer visionresearch. Motiontrajectories are not onlythe first-hand objects ofstudy, butalso act as input objects of follow-up high level of visual analysis, such as motion patternunderstanding, behavior analysis and so on. Video-based human multi-target detection andtracking technology is mainly used in video surveillance, defense, industrial control, criticalinfrastructure protection, and advanced human-computer interaction system.The paper presents a real-time multiple targets detectionand tracking algorithm. Firstly, inthe moving targets detection section, we design a dual-adaptive target detection algorithmbased on bilateral filtering, this method issues the influence of outdoor illumination on targetdetection. The bilateral filtering not only can filter noise, but also save a good edge. Especiallyfor the deformable moving human objects, it can provide better edge information to thesubsequent target tracking. The double adaptation refers adaptive background updating andadaptive threshold. Taking the use of the three consecutive frame-difference means todetermine the time of background updating, and then the changes of illumination need to bequantified which can be classified. When the light changes in the global picture dramatically,we extract the foreground moving region by the use of background updating method; Weakillumination happens, the dynamic threshold updating method is applicable to extract the targetarea. We can get the targets collection with the numbered rectangular boxes, at the same time,the pretreatment needs implement for the next muti-target tracking. Experiments show that thisapproach can extract the moving object accurately under different illumination conditions.Our multi-target tracking technology is mainly based on the realization of motiondetection. The characteristic parameters could be extractedthroughsteadyand perfect detectionresults, whichprovide goodinput parameters for the implementationof multi-target tracking. Inthis paper, we utilize a multi-target tracking based on detection. A target state matrix isdesigned to analysis the state of moving targets, and a search box based on movementinformation is programmed to shorten the overall searching time and accelerate the trackingspeed. When the moving targets occlusion occurs, the central region feature matching trackingalgorithm based on adaptive searching box would be used. This method can extract the Hcomponents of the central target region based on HSV color space. The H components canovercome the certain effects of illumination. According to the level of occlusion, the centralarea of motion is scaled on a limited extent. When the quadratic scaling feature matching fails,the serious occlusion of targets occurs, the occlusion targets are mergered. In other way, theocclusion doesn’t happen, all targets are isolate, and the nearest neighbor algorithm based on adaptive searching box is used to tract targets. Experimental results show that the proposedalgorithm can be applied to indoor or out door fixed camera monitoring environment, with wellreal-time and robustness performance.