A Method Of Geometric Correction And Mosaic Of Unmanned Aerial Vehicle Remote Sensing Image Without Ground Control Points

A method of geometric correction and mosaic of UAV (Unmanned Aerial Vehicle) remote sensing images without control points is proposed to meet relative requirements from the project "Research on Technology of Dynamic Monitoring and Supervision of Land"(Grant NO:201011015-1, Scientific Research Special Sponsored by Special Funds for Public Welfare Industry, from Ministry of Land and Resource of P.R.C.).UAV remote sensing has shown a promising application in the field of land monitoring and supervision in recent years, with its advantage of low cost, flexibility and rapid response. One of the key points is to obtain a full image covering the interested area in a fast way based on UAV remote sensing images. Technology of geometric correction and mosaic without ground control points is receiving more and more concern and attention as the mosaic technology based on image matching and the traditional photogrammetry method can’t completely meet relative requirement. With the emergence and rapid development of POS (Position and Orientation System), georeferencing technology becomes mature than before. The integration of UAV remote sensing and POS together construct the basement, and speed up the application of Geometric Correction and Mosaic of UAV remote sensing image without ground control points. UAV remote sensing images can be integrated into a seamless mosaic image with spatial orientation by geometric correction supported by POS information, and by image matching based on some same features among images. The main content and conclusion of the research is listed below:(1) Correction of image distortion and calibration of POS data. The correction of image distortion takes two steps. The first is to establish the mapping relationship between distorted image and undistorted image according to the interior position elements and optical distortion coefficient acquired from earlier camera calibration. The second is to resample for eliminating the distortion of image edge caused by camera lens distortion. Calibration of POS data takes two steps too. The first is to acquire the accurate collimation axis deviation through the integrated sensor orientation of the specific calibration field by using INPHO software. The second is to calibrate by using the above collimation axis deviation and geometric rotation relationship. In this way, POS data is calibrated as the real exterior elements of image.(2) Geometric correction of image based on POS data. The pre-processed image is geometrically corrected, using calibrated POS data as the exterior elements of image, and using collinear equation and the indirect method of image correction. In this way, even if no ground control points provided, the plane of UAV remote sensing image, with geometric deformation, can be transformed onto horizontal plane, and the geometry coordinate information of the image acquired as well.(3) Image matching based on feature. A method of image matching based on improved Harris-Laplace and SIFT descriptor is proposed. Because of noise and illumination change, some feature points extracted by SIFT algorithm are instable, which will affect the match result in a way. With the proposed method, improved Harris-Laplace can be used to detect key points. And then the main orientation of key points is determined to form feature points, which are described by SIFT descriptor, and matched by BBF algorithm and RANSAC algorithm. The experimental results show that the proposed method possesses higher matching accuracy at the same matching speed compared with SIFT algorithm. (4) Seamless image mosaic based on geometry coordinates. Geometry coordinates adjustment of adjacent images is carried out by using the proposed method of image matching and the method of geometry coordinates adjustment. Further, all images are adjusted to the unified geometry coordinate system to avoid the mismatch problem. A specific algorithm is designed to search the best stitching line between images. Every two images located aside of this line is chosen to participate in the heading mosaic or side mosaic. In this way,"ghost" phenomenon can be avoided effectively. A buffer area is built along the stitching line with a calculated width. The overlap area of images in the buffer is fused by using the weighted average method, which can effectively weaken the obvious seam-line on mosaic image.The proposed method is applied for processing the UAV remote sensing images in the test area covering about40KM2. A seamless mosaic image with good spatial orientation is generated, based on897original UAV remote sensing images distributed over13flight lines.The proposed method can save much manpower and material cost, mainly caused by laying ground control points. The proposed method can achieve the goal of the project for quick acquisition of complete image of the monitoring area.