Research Of Panorama Stitching Technology Based On Monocular Vision And The Realization On Android Platform

With the development of image processing technology, the panoramic image stitchingtechnology has become a research hot topic, more and more military and civilian facilitieshave implemented this technology instead of expensive professional wide-angle camera,and have achieved better results. Panoramic image stitching technology solves thecontradiction between wide-field and high-resolution, it has strong practical backgroundsand high commercial values.In this paper, we have studied image stitching technology based on monocular visionsystem and the implemention on Android platform. We have proposed an automaticreal-time panoramic stitching algorithm. First we found the relationship between theoverlapping images through image registrations, and then the image sequence to be stitchedare projected to the same reference coordinate system by adopting transformation model,further we got visual effects of the image sequence through image fusion algorithmn,finally we have implemented this algorithm on the Android platform. SURF algorithm isadopted to detect and describe the feature points in the image registration procedure, andwe employed FLANN algorithm and RANSAC algorithm to match and purificate thesefeature points respectively. After the completion of image registration, we adoptedcylindrical projection model to project the images to be stitched to the same cylindricalsurface, then we integrated the overlapping areas informations of the images effectively byadopting the multi-resolution image fusion algorithm in order to generate a panoramicimage which meets the visual consistency. Finally we developed mobile applicationsoftware, completed the design of panoramic camera software and sititching algorithm, andimplement our methods on Android platform, and the performance is better than similarsoftwares on the market.The algorithms mentioned above are developed with C/C++in VS2010andOpenCV2.4.6development environment. We carried a large number of simulationexperiments with the indoor scenes of our laboratory. Then we finished the software designand algorithms transplant in Java language on Eclipse and OpenCV for Android2.4.6environment. We take many experiments under some distinct scenes such as the mountains,rivers, buildings, sky, prairie, sheep and Night scenes. The algorithms show great stability,real-time quality, and preferable stiching effect in these experiments, meeting the actual application requirements which enable the Anroid users to capture high quality panoramicimages easily whenever and wherever.