Based On The Moving Platform Of Multi-objective Detection And Tracking

For most creatures, motion perception is one of the most important primitive and basic visual capabilities. That's because the importance of motion perception for survival. With the help of motion perception, the moving observer can increase its situation awareness. At the same time, motion information can drive visual attention. For creatures, motion targets are likely to be dangerous enemies or delicious prays. So, most creatures have most effectively motion perception capabilities. Because the importance of motion perception for most visual systems, constructing artificial systems which have the same motion perception capabilities is a dream for scientists for a long time. As safety and security become critical problems in the recent years, the demanding on autonomous surveillance systems increases dramatically. Independent moving target detection and tracking gain great attention of scientists. This thesis focuses on the independent moving target detection and tracking part for a smart surveillance system.Independent motion detection can be defined as detecting those objects which move independently of the observer in the 3D space. For independent moving target detection, the global motion caused by the moving of camera should be estimated first. After canceling the influence of global motion, independent motion estimation is simplified as the independent motion estimation in still background. Various salient target extraction and segmentation methods can be used to detect independent moving target. For moving target tracking, it is can be considered as a template matching algorithm. By choosing appropriate features and modeling these features, template matching can be used to tracking independent moving targets. Main achievements of this thesis are as follows:A robust statistics based global motion estimation method is studied in this thesis. Considering the importance of global motion estimation for the whole process of moving target detection and tracking , the need for real-time processing and nicety, a feature point based robust global motion estimation method is proposed. The process of recovery parameters from tracked feature points is a parameter estimation problem. A robust global motion estimation method based on tracking feature points is proposed in this thesis. Simulation results show that the method of tracking a series of feature points, using robust statistics based two step parameter estimation method is rapid, simple to implement and precise, and it is good for surveillance applications.Independent moving target detection and segmentation methods are studied. For...