Research On Observed Images Motion Transformation Method

In recent years, the technology of target automatic recognition is widely used in military, industrial and agricultural field. The performance test of these algorithms has become one of the key problems. Because of the limitation of time, the shortage of fund the non-cooperation of the target, and the complex backgrounds, only parts of image sequence under imaging conditions. Therefore, it is necessary to develop a software system to simulate images acquired under various conditions.The goal of this paper is to design and develop an observed images motion transformation system, which can simulate images captured on the specifited paths according to a set of images acquired on the other flight path.Based on the literatures, the observed images motion transformation algorithm is proposed. The main contents of this paper are as follows:(1)The framework of the observed images motion transformation method is proposed.All the images which are processed and simulated in the paper are aerial images. Considering the impact of environment, imaging equipments and shooting gesture, the process of observed images motion transformation method consists of the preprocessing of observed images, estimation of motion parameters, generation of panorama and novel view synthesis, etc.(2)An improved SIFT and RANSAC based image registration algorithm is proposed.Based on the in-depth study of SIFT and RANSAC method, and aiming on the increasing threshold robustness of RANSAC algorithm when it eliminates mismatching points, an adaptive threshold RANSAC algorithm is designed to improve the robustness of the algorithm. In order to make the distribution of matching points more uniformly and reasonably which can improve the accuracy of registration, the feature points evenly distributed algorithm based on errors and contrast is proposed, and a theoretical argument is provided. The experimental verifications of simulation data and actual data are conducted.(3)An image sequence panorama generation algorithm is proposed Images are layered by height, and the least images to generate new motion parameter image are chosen before panorama generation. If the optical angle changes largely, image correction is needed to conduct, which can reduce the error accumulation of the transformation.(4)The observed images motion transformation software system based on the proposed algorithms is developedThe software system based on the proposed algorithms is implemented in the developing environment of VC 6.0, and the OpenCV library is used to increase the developing speed. The tests have shown that each sub-module of the system can complete the expected functions and has stable performance.