Research On Image Stitching Based On Feature Points And Position-Pose Measurement Of Single-Camera

The rapid development of image processing technology makes the image stitching technology become an important content in the field of image processing.The main problem of image stitching is that high-definition images are not obtained due to the limitation of equipment conditions.The core problem is to find the overlapping regions of two adjacent images,and to carry out efficient feature registration for overlapping regions.Pose estimation problem plays an important role in mobile robot and tracking,photogrammetry,target recognition and so on.Pose measurement is mainly achieved by image acquisition,image preprocessing,feature extraction and pose solution.The level of industrial automation and the rapid development of machine vision,people put forward higher requirements for posture estimation.Therefore,it is a hot research to study an accurate and efficient image mosaic algorithm and pose solution algorithm.In the process of visual navigation of mobile robots,monocular vision can not meet the needs of wide field of vision,in order to increase the visual field of mobile robot,the use of binocular acquisition images for panoramic image stitching to achieve the purpose of obtaining large field of view image,For visual navigation providing convenience.Therefore,this paper has carried on the thorough research to the image stitching technology and the monocular visual pose measurement technology.This paper first studies the principle of image mosaic technology,and adopt the robustness of the SIFT and SURF feature extraction algorithm,the RANSAC registration method based on single matrix and the fusion method of multi-band fusion are well completed two image stitching;Gaussian filtering of the noise in the image,SIFT,SURP,ORB three feature extraction algorithm in different image transformation model of the feature point matching accuracy and efficiency of the experimental verification,improve the accuracy of image registration.Secondly,a camera pose estimation method is studied,based on the PnP visual measurement method,the unique solution of the P3P problem is discussed,the algorithm of calculating the pose of the camera with the feature points extracted by the target is designed,the posture information can meet the design requirements in terms of speed and precision.Then,the principle of the geometric constraint is discussed,and the world coordinates of a certain point in the three-dimensional space are calculated by using the method of single-camera posture transformation;Finally,the whole system is designed with Qt.