Structure From Motion Based On Panoramic Camera

3D reconstruction is one of the most important technologies in the fields of mapping,robotics and military.Thanks to the low-cost and high-precision features,image-based 3D reconstruction,or structure from motion technology is applied to many areas such as drones,robots and augumented reality.Self-driving cars also use image-based 3D reconstruction technics to rebuild and predict the environment around it.In order to obtain the environment information more efficiently,avoid the common low texture problem in 3D reconstruction field and make the device more portable,this paper studies the 3D reconstruction technic based on panoramic camera.This paper first studies the principle of 3D reconstruction of ordinary camera.A flexible 3D reconstruction method is proposed.And then the paper studies the imaging principle of panoramic camera and develops a panoramic camera which can stitch images in real time.Finally,the perspective geometry in 3D reconstruction is extended to panoramic images to implements the 3D reconstruction of panoramic camera.All the results of this paper are integrated into a structure-from-motion framework developed by C++.The framework has a high degree of flexibility with the supporting of different type of cameras(include panoramic camera)and algorithoms.At the same time,many algorithms in the framework is paralleled so that the compute capabilities of not only CPUs,but also GPUs are fully utilized and boost the framework a lot.The framework is tested by many data set downloaded from the Internet.The results show that the framework can reconstruct the object's points cloud and camera's locations accurately and rapidly.As the lack of panoramic data set on the Internet currently,this paper collected a number of images for testing by the device made in this project.The results also reach the expected.