Target Tracking And Positioning And Teaching-playback Navigation In Outdoor Environment Based On Fish-eye Lens

Vision-based navigation is a major research of AGV navigation owing to large amount of information.The traditional Vision-based navigation usually uses a lens with smaller perspective, so the information obtained is relatively less, and the system needs to rotate PTZ to obtain large field of view. It is hard to achieve the requirement of real-time, and also easily to lose targets. The omni-directional vision system built with a fish-eye lens is capable of collecting all the information within the hemisphere. So this system is suitable for vision navigation.Omni-directional vision system effectively overcome the loss of information during the PTZ rotates, and make targets'features more stable. Moreover, the system built with a single camera and a fish-eye lens is relatively easily to design and construct compared with camera panoramic stitching and reflective panoramic.Identification, tracking and positioning are the problems must be resolved for AGV navigation. Image distortion is an issue of fish-eye lens, so in order to accurately positioning the AGV, the coordinates of landmarks need to be corrected. The identification and tracking algorithm need to take into account some other problems, such as the large amount of information captured by the system, the complexity of the background and the illumination changes. In our project we choose the template matching algorithm for identification and particle filter algorithm for tracking. In order to test and verify the accuracy of the AGV navigation, some experiments based on Teaching-Playback have been carried out.In our project we build an omni-directional vision system with a single fish-eye lens, and fix it on the top of AGV. The experimental results proved that this omni-directional vision system can accurately track the targets in outdoor environment, and also achieve the purpose of self-positioning and teaching-playback navigation.