Dissertation > Excellent graduate degree dissertation topics show

Targets Tracking Based on Joint Algorithm of Motion Compensation and RJ-MCMC in Video Sequence

Author: LiuPengWei
Tutor: WangHuiYuan
School: Shandong University
Course: Signal and Information Processing
Keywords: Tracking Motion compensation Background update Markov Monte Carlo RJ-MCMC
CLC: TP391.41
Type: Master's thesis
Year: 2010
Downloads: 99
Quote: 2
Read: Download Dissertation

Abstract

Video moving target tracking field of computer vision research is an important issue, by domestic and foreign scholars' attention. It visually guided missiles, and unmanned autonomous robot navigation, positioning military target tracking and recognition, auto autopilot, security monitoring, intelligent transportation, virtual reality has broad application prospects. In reality, the presence of interference due to various video increases the difficulty of target detection and tracking. Video object detection is the basis for target tracking, detection effect is not good then he will follow a big impact. Because light, shelter, and the various camera sensor noise makes target detection becomes very complex, especially when the camera motion, background and foreground are moving along, but not good to separate the foreground and background. This paper takes a background based on motion compensation and adaptive dynamic updates to detect moving objects in the scene. Before and after the two first calculate the optical flow field between the pyramids, the optical flow field data in both a target and the background noise of the optical flow values ??also contain optical flow. Because I do not know in advance how many goals can not predict the interference caused by noise, so can not use a fixed number of class k-means clustering. This paper uses a technique called leader-follow online clustering algorithm, the algorithm can adaptively according to input data for classification, without having to know in advance how much you want to be divided into classes. Optical flow of the background value is obtained after separation of the motion compensation value using the compensation value compensation on an image, then the previous frame and the current frame image background aligned, so that the application can be used in a static scene, Background update method to get a dynamic background and foreground of the scene. In target tracking, usually from the perspective of recursive Bayesian target tracking problem, that is based on a series of observations to estimate the state of a target. Since the target motion without the law is difficult to describe with mathematical formulas, and the particle filter does not need to make any prior assumptions on the state transition, so many scholars in the field of research regarded it used for target tracking. Ordinary particle filter particle degradation problems severely limit the development of its basic approach. MCMC as a particle filter which can effectively solve the problem of particle degradation is causing concern. For multi-target tracking ordinary MCMC is unable to cope with, because when the target out of the scene will cause the solution state variable spatial variation, while the RJ-MCMC is able to effectively deal with the situation. For multi-target tracking dynamic scenes, we propose a model based on quadratic observation RJ-MCMC particle filter first observation of moving through motion compensation model for real-time correction value so close to the real equation of motion, the second observation that is, RJ-MCMC particle filtering step. The time-varying motion model can improve the efficiency of the method RJ-MCMC, points to reduce its inactive particles, so that it can more quickly converge to the true value. To not obvious or target area is too small target tracking, the system observation model proposed in this thesis based on color histogram matching combined with the prospect of observation model. This observation model the effective application of the detector in the foreground foreground information. Experiments show that it can effectively track the target area was not obvious or too small goals.

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