Research On Image Mosaic Technology Based On ORB Algorithm Under Fisheye Lens

Fisheye image has the advantages of wide field of view,but it will be very large deformation,which will hinder the image stitching.Therefore,if we want to make full use of fisheye image,that requires the distortion after image perspective transform into images without distortion,and then conduct the image stitching,so as to make mosaic fisheye images obtained by using the maximum value.This paper focuses on the research of image stitching technology based on fisheye lens.The stitching of fisheye image has the following components.In the image calibration,two different calibrations are introduced.The calibration algorithm based on spherical perspective projection constraint and the calibration method based on latitude and longitude are used in this paper.The latter is used to correct the distorted fisheye images.This algorithm has fewer pixels lost and can meet the needs of the experiment.In the aspect of feature points detection,improved ORB(Oriented BRIEF)algorithm is adopted.On the basis of SURF(Speed-Up Robust Features),the improved algorithm ORB has the dimensionality invariance and improves the correct rate of the feature point detection.Compared with the algorithms SIFT and SURF,the speed of matching has a big improvement while ensuring matching accuracy.Although the improved ORB algorithm has increased in time compared with the traditional ORB algorithm,it has scale invariance and detect more feature points and increase the matching accuracy.In the aspect of the purification of feature points,PROSAC(Progressive Sample Consensus)was used to remove the mismatched feature points.PROSAC is faster than RANSAC(Random Sample Consensus)in eliminating error matching.In the image fusion,because of the angle of view and change of illumination or other factors in the process of collecting images,it will lead to the phenomenon of ghosting,splicing and other phenomena.This paper adopts the gradual elimination algorithm to eliminate these phenomena,which in turn makes the spliced image more natural.This paper use Matlab R2015 b and Opencv,Visual Studio2015 for the above process experiments.In the calibration experiment,compare calibration algorithm which based on spherical perspective projection and algorithm which based on the latitude and longitude location,choose the latter as calibration algorithm in this paper.In the feature extration experimrnt,comparing the traditional ORB algorithm.The results of experiment show that the calibration algorithm based on the latitude and longitude location can retain the original fisheye image information when the image scale changes,the improved ORB algorithm can detect more feature points,and this is conducive to image mosaic.