Based On The Mutual Information Of The Three-dimensional Moving Object Identification And Tracking

Visual object tracking is a problem that has been studied by the computer vision community for decades.Computational stereo refers to the problem of determining 3-dimentional structure of a scene from two or more images taken from distinct viewpoints.Detection and tracking refers to the problem of determining real position and state of an object from images taken from cameras.The primary problems to be solved in tracking stereo are occlusions and light changing.Since stereo vision provides extra information of the depth of the object,we focus our work on stereo-moving object tracking.An approach to detect stereo- moving objects with bio-cameras is presented in this thesis.Stereo and motion are detected simultaneously under a modified formulation of energy function.We extend the application from static stereo match and motion mono-camera detection at a time to handle the whole problem together,which results in a reduction of computation time and ability to handle more difficult situation. When we address visual correspondence problems,we always assume that scene reflectance and structure never changes,however it does not always exist.In this thesis we add mutual information to help with the system to overcome the lighting changes.Experiments on several video-images,that shown convictive object detecting performances demonstrate its strong adaptability to lighting changes and shadows.