Dissertation > Excellent graduate degree dissertation topics show

Intelligent Vehicle Detection and Tracking Based on Particle Filter

Author: LiJun
Tutor: LiHengChao
School: Southwest Jiaotong University
Course: Signal and Information Processing
Keywords: GGM GFMM Particle Algorithm Parallel Computing
CLC: TP391.41
Type: Master's thesis
Year: 2012
Downloads: 96
Quote: 1
Read: Download Dissertation

Abstract

Video based vehicle detection and tracking technology is a part of pattern recognition and computer vision research, many scholars have made great effort to do an in-depth study. The detection and tracking technology based on video has the advantages of low cost and high speed and etc. It has become an important part of the Intelligent Traffic System (ITS). Because of such as the background noise, illumination changes of the video signal, there are too many random factors to lead that, the accuracy and convergence of different algorithm is different. This paper used the generalized GAMMA mixture model instead of Gauss mixture model for background modeling of the video sequences, to vehicle detection. Then, particle filter algorithm is used to track the vehicle which is detected in the last step. At last, the GPU parallel algorithm is used to improve the operating speed of the particle filter algorithm. It can get better results after improving the overall algorithm efficiency.In the intelligent vehicle detection system which is based on video analysis, the background difference method is usually used for detection. This method requires modeling the vehicle and background, the most commonly used to model is the Gauss mixture model. In this paper, based on studying the Gauss mixture model of vehicle detection, generalized GAMMA mixture model is used instead of the Gauss mixture model, and EM algorithm is used for parameter estimation of the generalized GAMMA mixture model. This method can overcome the poor performance of Gauss mixture model in shadow recognition; it can get better detection result. Finally, the simulation experiments prove the reliability and superiority of the algorithm.After detected the moving vehicles, the next step is usually vehicle tracking. Because of the commonly used of Kalman filter theory and its improvement for vehicle tracking algorithm, first of all, it is necessary to model the motion of the vehicle, and set up the state transfer equation and observation equation, then track the vehicle by the application of filtering algorithm. In1993, Gordon put a resampling step after the sequential importance sampling, this step made an improvement on the particle filter algorithm, so the particle filter algorithm became a better algorithm of target tracking in a non-Gauss and nonlinear environment, which is closer to the actual application situation. This article firstly studies the basic motion modeling and particle filter theory, and then realizes vehicle tracking based on particle filtering algorithm through the experiment simulations.However, the particle algorithm needs to imply the sequential importance sampling for too many particles, and the operation speed should be slow, so it has poor performance in real-time tracking. Therefore, this article studies the parallel algorithm based on GPU programming, and then uses it to realize the particle filter algorithm in parallel computing, the performance of particle filter is improved. Finally, the performance of the whole algorithm is improved.

Related Dissertations

CLC: > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device
© 2012 www.DissertationTopic.Net Mobile