Research And Realization Of Image Sequence Mosaic For UAV

UAV remote sensing technology has become a n important access for disaster monitoring, environmental surveys and other fields to get remote sensing data. It can timely and accurately obtain high-resolution images of the target area. With affected by the altitude, the camera focal length and so on, a smaller target area is covered by the scope of a single image. This paper expands the monitoring scope by researching the UAV image sequence mosaic technology.1) For UAV image sequences with many deformations of translation, rotation, scale, noise, etc., SURF algorithm is researched to solve the problem of feature points extracting. By the method of comparative analysis, this paper focus on researching the reasons which the SURF algorithm is faster than the SIFT algorithm. The feature extracting experimental results of Harris, SIFT and SURF indicate that the feature points with extracted by SURF is stable and well-distributed reasonably, and the SURF algorithm is robust highly and relatively faster.2) Nearest neighbor algorithm is used for feature points coarse matching. Perspective transformation is analysed and chosen as the transformation model of image mosaic. Dual-4-RANSAC algorithm is proposed to calculate the homography parameters and to exclude the mismatching points in order to achieve precise matching. Dual-4-RANSAC algorithm is a kind of improved RANSAC algorithm. It firstly chooses 4 pairs of matched points to calculate the model parameters and then increases 4 pairs of matched points to test the model. If there is a pair of matched points is non interior point, the algorithm re select the points and calculation model. The theoretical analysis and experimental results indicate that Dual-4-RANSAC algorithm with proposed in this paper is faster than the RANSAC algorithm.3) Image fusion algorithm based on wavelet transform is researched and Improved for two frame and multi-frame UAV image sequence fusion and mosaic. Firstly, the direct average method and fade-out method are analysed, and then the paper focus on the wavelet transform fusion method. The improved weighted fusion algorithm based on normalized Cross-Correlation is proposed to fusion the low frequency coefficients of wavelet decomposition in this paper. The algorithm expands the pixel to its m×n neighborhood firstly, and then calculate the NCC value of the corresponding neighborhood. If the NCC value i s greater than a predetermined threshold, the fusion coefficient is the average value of input coefficients. If the NCC value is smaller than the predetermined threshold, the fusion coefficient is the maximum value of input coefficients.The analysis of fusion effect from the two aspects of subjective evaluation and objective evaluation indicate that the improved weighted fusion algorithm based on NCC is better for low frequency coefficient fusion and the improved wavelet transform image fusion and mosaic is better in this paper.