Research On Fast Distributed Particle Filter Algorithm

Moving object detection and tracking in dynamic background are the keytechnologies in computer vision field. Substantial applications such as robotnavigation,car video, are closely related with moving object detection in dynamicbackground. This paper summarizes a better implementation method of moving objectdection and tracking in dynamic background in application of car video which ismainly based on DPF estimation acquired by the multi-sensor system.(1) This section describes the status of Particle Filter(PF) and movingobject,detection and tracking in dynamic background,including the concept andmethod of Distributed Particle Filter(DPF) in multi-sensor networks as well as theblock matching method widely used in dynamic situations,then demonstrates theapplication scope and demerits of each method.(2) In multi-sensor networks,due to the restraints of sensor node resource,theapplied algorithm can only obtain the position and path of the object from the localnode which requires a good cooperating ability to reach the target of lowtime-consuming and fast calculation.This paper presents a improved method based onthe traditional clustering estimation in which the allocation and fusion of the particlesare controlled by the node weight.The experiment shows that this method has theadvantage over the central and traditional distributed algorithm in the terms oftracking error,time-consuming.(3) This section introduces a widely-used approach of acquiring backgroundmoving vector based on block matching and demonstrates the possible errors andwasted calculation caused by the threshold choosing procedure in the traditional blockmatching.Hence,this paper presents a block matching method based on distributedblock Particle Filter Estimation in which move the former fame in negative directionacquired by the PF estimation in order to get the search center of the matching block as well as the search scope determined by the PF estimation error in the latter famewith a improvement of accuracy and rapidity in block matching.(4) Feature optical flow method widely-used in moving object detection in staticbackground is introduced in this section.According to its commodated conditions,suchas greyscale has to change consistantly in order to satisfy the optical flow basicequation and this method can not solve the problem effectively in dynamicbackground situations,this paper presents a optical algorithm based on DPF estimationin order to detect and track the moving object.Through the experiment in matlab andVisual Studio,this method proves to be effect to certain extent and anylized in terms ofthreshold choice and error analysis.