Stereo Image Rectification And Image Stitching Based On Fisheye Lens

Binocular stereo vision is research hot spot in computer vision. Due to the limitation of the field of view of the perspective camera, the application of large field of view camera with wide angle lens in computer vision draws more and more attention. Due to sharp decrease in the number of images, image stitching based on images captured by large angle camera has a good application prospect.This thesis is mainly about stereo vision algorithms and image stitching based on fisheye camera.We proposed fisheye stereo based on the Taylor model firstly. The Taylor series model is used to calibrate the stereo camera. Based on this model, the mapping relationship between perspective model and Taylor model is obtained, which can be used to undistort the fisheye image. The interior parameters of the virtual perspective camera can be set directly. The exterior parameters of the binocular cameras calibrated by using calibration pattern are used in epipolar rectification of the stereo image pair. Derivation of the depth information of stereo pair is also performed before 3D reconstruction.Image Stitching based on Taylor model is researched aimed to the application of vehicle surrounding monitoring. Four overlapped fisheye images are undistorted into perspective ones. Then we projected them into one common plane and fused them into one monitoring image.