Research And Implementing On Multitarget Tracking Involed In Dense Clutter

Multiple target tracking algorithm can be divided into two broad categories up to now: One is based on data association, such as Joint Probability Data Association(JPDA) and the other one is based on Random Finite Set(RFS) theory, such as Probability Hypothesis Density Filter(PHDF). Both of them have advantages as well as defects: Association based algorithms can receive high accuracy when they are implemented in dense clutter circumstance, while they require large computational cost; RFS based algorithms are association free and the computational cost is relatively small, while they usually can not obtain the continuous trajectory. Additionally, Multi Model(MM) and Interacting Multi Model(IMM) framework were proposed to track maneuvering targets and derived the maneuvering tracking versions of JPDA and PHDF: IMM-JPDA and MM-PHDF.Track Before Detect(TBD) framework can reach the good tracking effect in dense clutter, but RFS based tracking algorithms have not gotten a complete tracking model, so algorithm based tracking algorithms are usually used in dense clutter situation: First, tracks are initialized by accumulated several observation frames by joint maximum likelihood probability data association algorithm(JML-PDA), when targets’ initial state and covariance matrix are obtained, a JPDA filter will be followed up to maintain targets’ trajectory, which is called combined JML-PDA and JPDA algorithm(CJML-PDA).Dense clutter in monitoring area and an increase in the number of target will result in combination explosion problem in multiple target tracking algorithm based on correlation, and the multiple frame accumulation algorithm often need to use multiple observation data, which caused a large amount of calculation, it cannot be achieved in real time in serial operation platform processor, which obstructed their engineering application. Since the graphics processor unit(GPU) being used for general signal processing, a lot of algorithm which are high computational complexity in traditional architecture have a possibility of real-time implementation.This article in view of some defects existing in the multiple target tracking algorithm, multi-target tracking algorithm is studied and its implementation on the hardware platform, the main work is as follows:(1) Introduced the tracking model under the condition of multiple target, JPDA filter and its maneuvering tracking version;(2) Introduced tracking model under the framework of RFS theory, PHD filter, Particle Filter based PHDF, MM-PHD and MM-PHD under particle filter implementation.(3) Introduced multi frames accumulating based tarjectory initializing algorithm JML-PDA and proposed a new solution to maneuvering targets tracking in dense clutter circumstance.(4) Introduces the architecture of GPU, its characteristics and development platform, proposed implementation scheme of CJML-IMM-PDA on GPU. Additionally, some simulations are carried out to prove the accelerating rate on GPU compared to CPU, which laid a foundation on real-time application of multiple target track initiation and tracking algorithm.