Research On New Method For Positioning Of UAV Visible-light Image Sequence

Unmanned aerial vehicle(UAV)positioning is one of the key technologies.Using UAVs' high maneuverability to track military maneuvering targets,it is an indispensable task for information-based battlefield to carry out combat and quick response.However,the current UAV target location method relies on the airborne POS system,in many cases the POS system cannot provide high-precise positioning results.In this paper,we take the orthophoto with geographic information as a reference,and use UAV sequence images and orthophoto matching to achieve localization.However,the UAV cannot obtain enough control points to complete the matching,as the image range obtained by UAV is quite small.The image mosaic can expand the image but it will lead to the accumulation error,which cannot meet the positioning accuracy.The results of this study are as follows:1?This paper presents a method of sequence image localization by combining local transformation and global control points.This method first matches each adjacent two frames in the sequence image to establish the relative geometric relations between them,and then finds the pair of matching points from the sequence image and the reference image.Finally,according to the transformation matrix between all adjacent frames and several global control points,the indirect adjustment model is established,and the transformation model between each frame sequence image and the reference image is solved to complete the localization.This method not only solves the problem of insufficient control points caused by small image size,but also eliminates the cumulative error caused by continuous image matching,while maintaining a faster positioning speed.2?A fast matching method for adjacent frames of sequence images based on homography matrix decomposition is proposed.The relative changes of pose parameters of the first few states of the camera are obtained by decomposing the homography matrices of the previous pairs of neighboring images.Then,the relative change of parameters are predicted and the homography matrix to be estimated is obtained.The matching point is projected onto another image using the estimated homography matrix,and the exact matching is performed in its neighborhood.The method makes use of the correlation of image in time and space,and greatly reduces the matching range,and has better matching precision and real-time matching speed.3? A global control point extraction method based on SURF algorithm and adaptive threshold RANSAC algorithm is implemented.Taking into account the differences of the sequence image and the reference image in resolution,shooting angle,image gray-scale,etc.,this paper using the scale,rotation invariance SURF algorithm to extract and describe the feature points,and then the nearest neighbor algorithm feature points Finally,an adaptive thresholding RANSAC algorithm based on Otsu algorithm is used to eliminate the mismatching in matching.The experiment shows that this method can get exact match result.