Study Of Image Corner Detecting And Matching Method

Image feature detection and matching technology is widely used in target tracking, image registration, and 3D reconstruction. Among the most kind of features in the image, the position of the corners is very clear, there is no ambiguity in their location. Moreover, the corner preserves the features of the original image, it can greatly reduce the amount of data. This thesis focuses on the study of corner detecting and matching method.This thesis mainly includes three aspects, study of corner detection algorithm, study of feature description and matching method, image acceleration algorithm based on GPU.In the aspect of the corner detection method, after analysis the advantages and disadvantages of the conventional corner detection algorithm, this thesis proposes a new corner detection algorithm. In addition to detecting the corners accurately, the proposed corner detector preserves the edge direction of the corner, which provides more information to the corners. Most of corner detectors will detect round corners in image, to solve this problem. The round-corner is removed by using the change rate of curvature. Finally, the proposed corner detector is compared with five well-known algorithms. Experimental results reveal that the proposed method outperforms the state-of-the-art corner detectors in ACU, location error criterion. Meanwhile, the proposed algorithm is competitive in computation time and the proposed algorithm reaches a good balance between accuracy and computation time.In the aspect of feature description and matching method, this thesis proposes a binary corner descriptor based on the edge direction. This binary descriptor preserves the edge direction information of the corners and this binary descriptor can match the same type of corners between two images. Moreover, the proposed descriptor can exploited to eliminate the mismatching points. The proposed descriptor combine with BRIEF can help to improve the matching correct rate without increase the computation time.In the aspect of image acceleration method based on GPU, this thesis focus on the parallel processing technology based on CUDA platform, especially on the aspect of image parallel processing method based on CUDA, including de-noising algorithm, edge extraction algorithm and corner detecting algorithm based on CUDA. After analysis the parallelism of the proposed corner algorithm, we finally ported the corner algorithm to the CUDA platform, the speed has been improved.