Research And Implementation Of Unmanned Aerial Vehicle Image Processing Technology

During recent years,UAV(Unmanned Aerial Vehicle)with its advantages of convenience and low cost,has become the "star" in many fields,such as environmental monitoring,media report,electricity supplier delivery,etc.As the natural environment is complex,UAV's GPS signal usually lost or interrupted,so visual navigation method is particularly significant.At present,there are many researches on vision navigation,but there are few examples of real application.In this paper,based on the UAV autonomous flight project,one of the key works is to study the image distortion correction method and the image stitching technology.The other key work is designing and identifying the landmark during the visual landing phase.UAV uses aerial camera to take photos that are easily distorted.This paper analyzed the imaging principle of camera firstly,and used the Zhang Zhengyou's calibration method including making calibration checkerboard plate,collecting images at different angles,extracting Harris corner points,and using the maximum likelihood estimation to optimize the parameters to calculate the internal and external parameters matrix and the distortion coefficient of the camera.Finally,the distortion coefficient was used to correct fisheye image,which obtained good results.In the process of stitching images,we used the image stitching technique based on SIFT feature.Firstly,we extracted the SIFT feature points of images,and generated 128-dimensional feature vectors,and then calculated the similarity between the feature points of the two images.To filter the mismatching points,we used the ratio of the minimum distance and the minor distance to filter mismatching points,and then used the RANSAC algorithm to optimize the candidate matching points,which decreased mismatching points rate to 74.74%.The homography matrix was calculated according to the selected matching point pair,and the stitching image pixel was transformed into the reference image coordinate system by the homography matrix.In this paper,we proposed the average fusion method and the weighted average fusion method to carry on the pixel fusion on the image overlap area,and got the good sptitching result.Finally,this paper implemented automatic stitching algorithm for many maps.In the autonomous landing part,we designed a simple and easy-identifed H landmark.We preprocessed the captured image to obtain the gray image,and then applyed the adaptive threshold method to get binary image which was used to extract the contours.According to the length of the contour and area information,we filtered out non-landmark contours.Finally,we used the shape of the Hu invariant moment to determine which contour is landing mark contour.The project improved the H landmark with adding the contour edge,and applying the adaptive threshold method to get binary image,which enhanced the robustness of the landmark in various environments.This paper put forward a set of software and hardware system for the visual landing of the UAV in low altitude and small range,and verified the feasibility,stability,accuracy of the system in the University of Electronic Science and Technology of China campus.Experiments showed that the landing's precision can meet the design requirements of 1 meter.