Object Tracking And Localization Based On Active Machine Vision System And Scenes Understanding

Machine vision is a new research field that focuses on the cognition process of visualcues at the level of information processing and the concerned calculating theory, algorithmand representation on the basis of digital image processing. Active machine vision isderived from machine vision and emphasizes the imitation of human beings'activeadaptability, which is to simulate the human "head-eye"function and enable the system toselect, gaze at and track the object automatically.This thesis mainly focuses on the following topics:(1) Extend the visual information processing architecture of active machine vision.(2) Regarding the moving object's hue feature as its characteristics, realize object-tracking algorithm based on the Active Machine Vision System AMVS-JLU I.(3) Propose a binocular 3D localization method based on active machine vision system.(4) The description, representation and reasoning of the linguistic knowledge concerned with scenes understanding.A valid active machine vision system should contain the modules and functions asfollows: a test-bed composed of cameras, motors, communication devices and controllers;3D information acquirement, object localization and recognition; motion informationextraction from image sequence; selective attention; stereo vision; gazing and trackingsystem; and scenes reconstruction and understanding. Furthermore, the tasks from thehigher system and knowledge are necessary for the active machine vision system to adjustthe relevant parameters to improve the whole system performance.Color has good invariability and is suitable to describe the moving object for trackingapplication. There are many color representations corresponding to different color space. Inthis article, we adapt the HSI (Hue, intensity and saturation) color space which is close tohuman visual psychology to make a extended hue component. The extended huecomponent regarding the gray and the chromatic as colorful uniformly is derived from hueand intensity component of HSI color space, which makes the algorithm of moving objectcolor feature extraction simple and general for both gray and chromatic object.To reduce the time cost of image processing and speed up the whole system, we adoptthe variable window technique to handle the image region that the object may appears in.For sake of this, the constant acceleration kinematical model of moving object isestablished and the Kalman filter is used to predict the object's position in next frame.