Research On The Key Technique Of Vision Computing Platform

As a new generation of computing model, the goal of ubiquitous computing is to embed computing into the environment, making the environment with computing power, the ability to achieve active perception of the needs of people and the ability to provide people with resources and services. To achieve this computing model which is human-centered and ubiquitous, the key depends on sensor networks which are composed of various sensors deployed in the environment. Visual sensor network plays an important role in ubiquitous computing, which has the characteristics of rich aware content, activeness and non-restricted. The research on the perception and understanding of the human visual behavior, which is based on visual sensor networks, becomes one of the hot areas in ubiquitous computing.Distributed vision computing is based on visual sensor networks, including network infrastructure, vision computing and network coordination. Vision computing is the basis of the implement of distributed vision computing. Its main task is to implement the detection and tracking of visual target, as well as the perception and understanding of the visual behavior. This thesis studies the visual tracking technology belonging to vision computing and camera calibration technology belonging to network coordination. And also we design and develop a system for vision computing platform oriented to vision computing application.The content of the thesis includes the following aspects:(1) This thesis studies the SIFT feature matching algorithm. A key step of camera calibration technology is to extract feature points between different images. This thesis uses SIFT feature matching algorithm for feature point detection.(2) This thesis also studies the human motion tracking based on the particle filter algorithm. Particle filter is the result of the combination of the Monte Carlo method and Bayesian estimation theory. This thesis studies the basic principles and algorithm of particle filter because of its adaptability and strong tracking performance.(3) This thesis designs and implements a system for vision computing platform. It mainly introduces the basic features, system architecture and subsystems. Meanwhile a prototype system has been developed for vision computing platform, which is demonstrated at the end of this thesis.