The Research On Detection And Tracking Of Moving Target Based On Omni-directional Vision

The Omni-directional vision has greatly developed it has been researched by experts. The detecting and tracking of moving objects is a hot issue in computer vision field. These two technologies have been applied in intelligent robotics and intelligentized weapons, and the combination of them provides much broader prospects of application.This paper mainly studies the detecting and tracking of moving targets based on Omni-directional vision and presents the improved algorithm of different motion detection and tracking algorithms. Then they are applied in Omni-directional vision.Firstly, by researching the imaging principle of Omni-directional vision, and in order to overcome the impact of Omni-directional image distortion on target detection and tracking, a cylinder unwrapping algorithm based on geometry principle of Omni-directional image is proposed which can obtain 360 degree image. For the surveillance of important scene, another perspective unwrapping method that is based on coordinate transformation is put forward.The common detection and tracking of moving target is studied. Difference is simple and convenient to operate and it has good real-time quality. But it needs better background and is sensitive to the change of illumination condition such as light, weather. Afterward, the optical flow theory is introduced which is complex and can not meet the real-time requirement. And the difference is applied to Omni-directional images to detect the moving objects.Correlation-based method is applied widely in tracking of moving target. The general template matching which is slow is insensitive to template changes and is not suitable for real-time tracking. Then the SSDA algorithm is proposed which is increases the speed. Some improved measures are brought forward. The update of templates is resolved by Maximum Close Distance. And template matching and SSDA are applied in cylinder unwrapped Omni-directional images. Mean-Shift is an algorithm based on color that is small amount but also meets the complicated changes of objects and background. And it not only meets requirements of real-time, but also has good efficiency and tracking. Considering background color and adaptive combinational color features are put forward in order to resolve the disadvantage of Mean-Shift which often misses the target when the target is similar to background. It can be used in Omni-directional images without unwrapping because of its low template requirements.