Panorama Stitching For Low-light Image Feature Detection

The thesis uses low-light image stitching technology as the research background.Aiming at the problem that feature point extraction and matching processes are vulnerable to light interference during image registration,a better-performing image matching algorithm is re-searched and designed to improve the robustness of image stitching and accuracy.Then,the classic Harris,SIFT,SURF,FAST and ORB algorithms related to the above problems will be analyzed and compared,and the one with the best real-time performance will be selected for improvement.Because the scale invariant and rotation invariant of low-light images are not considered,and only the illumination invariant of low-light images is considered in the thesis.Therefore,the ORB algorithm without illumination invariant is not considered for improvement.Considering the real-time performance of the algorithm,the FAST algo-rithm is regarded as an algorithm to be improved.Hence,the thesis will improve it based on the FAST algorithm to ensure that it has fast registration time and improve the robust-ness.And the thesis proposes the EIRFT algorithm for panoramic image stitching.First,use a scale-invariant feature detection algorithm to extract feature points.Then,use the Surf feature descriptor to describe the feature points.And the method of lighting robust fea-ture transformation is also used to enhance the features of illumination invariant.Next,use both brute force matching and KNN Matching to perform rough matching,and then use the RANSAC performs precise matching and purification of matching points,further improving its accuracy,and finally completing image stitching.The experiments show that ERIRFAST proposed in the thesis has excellent robustness and stability under the conditions of image scaling,rotation,and illumination.EIRFT is the real-time,high-accuracy image registration method.Finally,the EIRFT algorithm is used in panoramic stitching.And it can be seen from the experimental results that the panoramic stitching using the EIRFT algorithm has the best performance,while the panoramic stitching using the SIFT algorithm has the worst performance than that using the EIRFT algorithm and than that using the SURF algorithm when considering the real-time and panoramic stitching quality.