Research On Visual Place Recognition Method Based On Panoramic Annular Imaging System

The positioning system is closely related to people's daily travel.For humans,it is not difficult to identify the location where they have been.But for robots,not only a suitable map,but also the "cognition" of location is required.Many aspects of the system are involved.The emergence of visual place recognition,to a certain extent,solved the problems that conventional satellite positioning cannot receive positioning information and that the positioning accuracy is not high enough in some cases.The panoramic annular imaging system is very suitable for visual place recognition because it has the advantages of a large field of view and constant horizontal rotation of the imaging scene.This paper proposes four methods for developing panoramic annular images,and chooses two of them for visual place recognition of panoramic annular images.This paper takes the advantages of the edge and color stability of the panoramic annular image to expand,and uses a simple but efficient image fingerprint method to coarsely localization the panoramic annular image indoors,and proposes a method to calculate the fingerprint features of the image.Since there is currently no feature point SLAM method that can be directly used for panoramic annular images,this article subsequently conducts visual SLAM exploration based on panoramic annular images,and finds a method for applying panoramic annular lens to the open source visual SLAM framework.The test results show that the panoramic ring zone is more robust than the same type of visual SLAM Fisheye camera.