Research On Method Of Bridge Surface Image Stitching Based On UAV Remote Sensing

Scientifically keeping the bridge safe is a major national need.It is imperative to closely combine emerging technologies such as UAV and remote sensing with bridge inspection to promote modern and intelligent upgrading of bridges.There are many bridges and large inspection demands in China,traditional bridge inspection methods can hardly meet the increasing inspection demands.Therefore,it is urgent need to apply inspection technology based on UAV remote sensing to the daily maintenance and monitoring of bridge pavements.Due to the limitation of sensing equipment,the UAV cannot directly obtain high-resolution panoramic images of the bridge road surface,so it needs to use stitching technology for image synthesis.At present,there are still problems of alignment error and stitching distortion in stitching technology of remote sensing image.so,it is theoretically and practically valuable to study a stitching method that can maintain the original bridge pavement details and adapt to the characteristics of low-altitude remote sensing images.This paper takes the bridge pavement images collected by UAV as the research object,and proposes a multi-feature joint constrained image stitching method for the problems involved in the image stitching process,such as low feature detection efficiency,insufficient matching features,and inaccurate image transformation model.The main research contents are as follows.(1)Bridge pavement image acquisition and image pre-processing.According to the overlap rate and resolution requirements of bridge image stitching,the flight parameters of UAV remote sensing acquisition are optimized,the acquisition scheme of bridge pavement images is formulated,the flight path is planned,and the real bridge images are taken.The image pre-processing method of histogram equalization and geometric correction was adopted to eliminate the effects of inconsistent color brightness and geometric distortion of bridge pavement images.(2)Improvement of image feature extraction algorithm.The SIFT algorithm is improved by reducing the dimensionality of the algorithm feature descriptors and replaying Euclidean distance with cosine similarity to match and filter feature information.The improved algorithm is experimentally validated,the time is reduced by 40% in detecting and matching feature,efficiency of detection is significantly improved.The bridge pavement is a low-texture image,which is difficult to provide sufficient and reliable matching features.Therefore,this paper introduces the structural features of angle features and line features in the feature detection model to solve the problem of insufficient bridge pavement features.(3)Research on the image stitching method with joint constraints of multiple features.According to the transformation characteristics of matching points,angles and lines in the image,an image single-strain transformation model is established to prealign the bridge images.Based on the Content-Preserving Warping(CPW)model,the line feature and angle feature constraint are proposed,the mesh optimization model of the image is established,and the constraint function is solved.The mesh optimization of the pre-aligned images is achieved by texture mapping.In this paper,we use image stitching method to perform double image and multiple image stitching experiments on bridge pavement images The experimental results show that the visual effect of the stitched images is good and the stitching error is within 10 cm,which meets the demand of detection of bridge pavement.