Research On Fast Feature Matching And Camera Orientation Estimation Method Based On UAV Aerial Video

UAV aerial photography is an important way to obtain spatial data,and is widely used in military and civilian fields.Three-dimensional reconstruction technology based on UAV aerial video plays an important role in the city planning,change detection,disaster assessment and other applications.In the three-dimensional reconstruction of video images,feature matching is the basic step that provides reliable input information for camera orientation and parameter estimation.Camera orientation estimation is the key link of three-dimensional reconstruction,and its estimation accuracy is closely related to the effect of three-dimensional reconstruction.Therefore,how to improve the speed of feature matching and the estimation of camera orientation is a research focus in the field of image processing and three-dimensional reconstruction.For this,this paper aims at the characteristics of UAV aerial video images and focuses on the fast feature matching and camera orientation estimation problem.The main research work is as follows:(1)Aiming at the feature points extraction and matching speed problem of aerial video images,this paper proposes a feature points tracking algorithm combined UKF and KLT,and realizes the azimuth prediction and fast matching of feature points in the adjacent frames.Firstly,the algorithm uses UKF to predict feature points in the adjacent frames to determine the scope of the match according to the features of the aerial video.Secondly,the algorithm tracks feature points according to KLT matching algorithm.The results are seen as observed value.After all,we could get the accurate position of the feature points through the kalman gain correction.The comparative experiments prove that the algorithm is not only efficient,but also better than KLT algorithm in matching accuracy.(2)Aiming at the characteristics of small parallax variation in aerial video frame,this paper puts forward a keyframe selection algorithm to reduce the cumulative error caused by frequent camera orientation estimation.The algorithm takes the number of feature points and the size of motion as the evaluation criterion.The algorithm firstly uses the feature point pairs to calculate the translation and rotation between the image frames.Then uses the weighted synthesis to acquire the degree of dissimilarity between images.Finally selects keyframes by setting threshold.The comparative experiments prove that the camera orientation of the key frame obtained by this algorithm is higher than that of the key frame obtained by ORB-SLAM.This paper firstly proposes the UKF and KLT optical flow combination algorithm,achieves the fast matching of the feature points and provides reliable data for the camera orientation estimation.Then,puts forward the keyframe selection algorithm and gets the accurate estimation of the camera orientation of the keyframes.The related results can provide the basic theory for the fast three-dimensional reconstruction method.