Research On The Algorithm Of Power Line Aerial Image Matching For Power Line Patrol UAV

With the development of UAV aerial photography technology,it has become a mainstream to construct power line corridor map,power line three-dimensional topographic map and power line patrol inspection with UAV aerial photography.Image matching algorithm based on local features is the most important link to realize the above process.The image matching algorithm based on local feature points is mainly divided into three steps: scale space construction,feature point detection and descriptor construction.Aiming at the traditional matching construction of scale space will lead to the loss of image edge information or low efficiency,this paper proposes an aerial image matching algorithm based on Gaussian curvature filtering.Firstly,the analysis of common scale space methods shows that Gaussian filtering is fast but cannot protect image details,while nonlinear scale space can protect image details but is slow.Therefore,Gaussian curvature filtering is proposed to construct scale space,which can preserve image edge information with high efficiency.Then analyze the feature point extraction method,and use FAST algorithm to extract the feature points.Finally,the binary and non-binary descriptors were analyzed.The binary descriptors were constructed and matched faster but expressed less information,while the non-binary descriptors were constructed slower but had better robustness.Therefore,ASV-LIOP descriptors were proposed in this paper.The scale space of the feature sampling region was established and the LIOP descriptors in the scale space layer were extracted.Then the difference values of two layers of LIOP descriptors were binarized,and finally the difference descriptors after binarization were accumulated to get ASV-LIOP descriptors.On aerial images,SIFT,ORB,KAZE,AKAZE and other algorithms are used to compare with the proposed experiment.The experimental results show that the proposed algorithm improves the correct matching rate by about 5% on average,and reduces the matching efficiency by about 50%.The proposed algorithm can be applied to scenes with high stability requirements and low real-time performance.The thesis has 49 pictures,11 tables,and 52 references.