Design And Implementation Of The High Speed Space Multi-target Tracking System Based On PHD Filter

In target tracking, video surveillance and wireless sensor network systems, it often need to track multiple targets and obtain the targets location, shape and other information of the target analysis. The probability hypothesis density(PHD) filter algorithm is suitable for the unknown target number of time-varying tracking scenario, At the same time, PHD filter can avoid data associations, so it is more effective for multiple target tracking. In probability hypothesis density filter algorithm implementation, because the PHD filter based on particle filter algorithm has good performance in nonlinear non-Gaussian environments, it is widely used in engineering application. But the PHD filter algorithm based on particle filter requires a lot of particles to reach a certain precision. Therefore, its low computational efficiency limits the practical application. This article mainly aims at the shortcomings of particle PHD filter real-time, from the measurement step, distributed structure and parallel three point to optimize the PHD filtering, then design and implementation of a space simulation platform for multi-target tracking. The main content of this paper is as follows:Firstly, for the phenomenon of the clutter in the particle PHD filtering, a clustering PHD filtering algorithm based on meanshift filter is proposed. The meanshift cluster algorithm is applied to particle classification, and our approach only uses the nearby measurements within a specified threshold to update the particle, thereby saving the large amount of computation time. This algorithm is more suitable for high clutter situations.Secondly, PHD particle filter algorithm for resampling can not parallel problems, presents two parallel resampling algorithms, distributed computing PHD particle filter algorithm and GPU-based parallel particle PHD filter. Acceleration of the particle PHD filtering algorithm with different degrees of parallelism.At last, using 3D simulation technology, using 3D Open Scene Graph engine to visualize the multi object movement, the 3D visual simulation system of target motion is constructed. The main realization of the three-dimensional visualization of the particle cloud, the target trajectory visualization, and provide good user interactivity.This paper started with describes the origin and background of subject, analyzes the research status at home and abroad, and then analyzes the function of the project requirements, design particle PHD filter algorithm, meanshift filtering PHD filter algorithm, parallel PHD filter and so on, after describing the implementation process of the system, and finally the system is tested on simulation data and summarizes the main points of the study.