Moving Target Visual Tracking System Design And Implementation

Motion targets tracking is an advanced technology of comprehensive subjects,which integrates areas of artificial intelligence, computer vision, auto control, patternrecognition, image processing and so on. Recently, researchers are devoting to studytargets tracking so that tremendous achievements have been developed in this field. Yetnew challenges exist in this domain such as the disturbance of illumination variation,targets reappearance, shadow interference, and object transformation. Therefore, theissue is practical and worthy whatever in academic or in application.This thesis is centered on the targets tracking by moving robot and designs a set ofpragmatic program. The whole projects include four modules, as vision detect, imageprocessing, motion analysis and robot implement modules respectively.Vision detecting module: the main task is to obtain high quality images forsubsequent processing. This module consists of camera selection and cameracalibration. Image processing module: abstract effective foreground objects. Firstly,modeling in the background is necessary. Afterwards, foreground objects have to bedetected. Finally, foreground and background information is separated on edge. Motionanalysis and decision-making module: motion targets detecting and tracking. Detectingtargets is one of the most important steps. Sampling set and classification are thefoundation of detecting. After detecting the objects, the objects have to be locatedprecisely by updating the classifier. In some cases, the targets may be sheltered ordisappear for a long time so that tracking could be very difficult or might be failed.This module is a good solution to handle it. The robot’s execution module switches theresults from the module which reasons and execute the policy to digital signal for therobot controller so as to controls the rotate speed and orientation of the robot’s motor.Eventually, this dissertation provides an experimental platform includinghardware selection and software architecture. This experiment is valid to track movingobjects.