Research Of Object Tracking Method Based On Local Features And Particle Filter

Traditional video surveillance requires monitoring human to monitor scenario before thecamera, according to the video information monitoring human make the appropriate action,this will waste a lot of human and financial resources. With the development and progress of thetimes, intelligent monitor system has become the major direction of future monitoring system,the camera can automatically analyze surveillance video, then analyze and identify thebehavior of the moving object, the camera make timely alarm for the occurrence of abnormalevents, intelligent monitor system is convenient and easy, besides it do not need monitoringhuman, it has an important impact on day-to-day management and social security. Objecttracking as an important step for intelligent monitor system, obtain the object in themonitoring scene is the first step, then track the object accurately and real-time, theparameters of tracking object are recorded in the process. As the parameters is applied for thenext step of the analysis and processing, therefore, accurate tracking results will have animportant impact for the subsequent processing.At present, domestic and foreign research scholars have put forward many excellenttracking algorithm, object tracking algorithm can be divided into three categories: pointtracking, kernel track, silhouette tracking. Point tracking: the important step is to obtain thecentroid of the tracking target, through the point correspondence to achieve tracking. Kerneltracking: kernel refers to the object shape and appearance, for example, the kernel can be arectangular template or an elliptical shape with an associated histogram, objects are tracked bycomputing the motion of the kernel in consecutive frames. Silhouette tracking: tracking isperformed by estimating the object region in each frame, in the silhouette tracking methodtracking object can not normally be represented by simple geometrical shapes, silhouettetracking methods use the information encoded inside the object region, for example, objectedge and the contour, etc. Current object tracking methods commonly used global features tocreate the target model, global features are simple and feasible, but global features representthe global information of tracking object, the drawback is that can not handle occlusion well.This paper use the second type of tracking method and apply local features to establishtarget mode, local feature extract local information of the tracking target, this can effectivelyavoid the occlusion problem, shape context as the representatives of local feature, it can begood description of the tracking region, to establish the shape context histogram for eachpoint in polar coordinates, shape context histogram of all points represents the appearance of the target, then particle filter algorithm is applied, particle filter is the classic trackingalgorithm, the basic idea is to use the limited particles to approximate the posterior probabilitydensity, its advantages are non-linear and non-Gaussian. The tracker based on Shape contextfeatures and the particle filter algorithm can solve the partial occlusion problem very well inthe process of object tracking, due to shape context features are robust for illumination,tracker can deal well with illumination change. Template update method dynamically changethe target model every frame, the tracker can adapt to the changes of the object appearance.Different tracking scenarios have different requirements for the tracking method, for theapplications which occlusion occurs frequently, object model can be divided into multipleparts and extract the corresponding shape context features, this will effective solve theocclusion problem. For the semi-automatic tracking algorithm, moving object is first detectedby embedding the object detection model, and then tracks the object by tracking algorithm,this can achieve a fully automated object tracking method. Due to global and local features allhave corresponding defect in object tracking field, object tracking algorithm can be combinedwith the global and local features, and tracker not only can avoid the defects, but also improvethe tracking results.Different tracking object will generate different tracking application in the realenvironment; our algorithm can be applied well in the field of video calls and identification bytracking the face. The field of intelligent transportation is applicable for our trackingalgorithm, through the tracking and analysis of the vehicle, vehicle flow control is achievedeasily, the camera can make timely alarm for the abnormal behavior of the vehicle. Objecttracking is an important step for behavior recognition, tracking the moving people in the sceneand tracking results can be used as the input data of behavior recognition, and then completethe task of behavior analysis and recognition. With the rapid development of smart phones,object tracking algorithm can be embedded into the smart phone, the tracking program can beapplied anywhere and anytime.