The Realization For Image Sequence Mosaics

With the rapid development of Image Processing, Image Mosaics has become a hot branch in image processing field. Image Mosaics have many affiliations with many subjects such as Computer Vision, Computer Graphics and Calculation Geometry and so on, at the same time, the technology has important applications in many fields.In the thesis, some efforts on it are performed. The realization of Image Mosaics is successful by Perspective Transformation Algorithm. The improved image Mosaics algorithm keep eye on the main steps, such as Feature Points Matching, Nonlinear Optimization Algorithm, and Image Registration and so on.This paper expatiates the course of Image Mosaics with 8-parameter perspective transformation algorithm in detail. In the step of feature points matching, because of the practical traffic accident spot, we adopt and improve traditional the RANSAC algorithm to remove the false feature points and degenerate combination. In the step of parameter optimization, we labor the advantages and disadvantages of DLT algorithm and the traditional nonlinear solution, which base on the cost function of intensity or color error. To overcome the disadvantage of traditional solution, we introduce a cost function basing on the symmetry shift error, and make use of the Levenberg-Marquardt algorithm to get the exact transformation matrix. In the Image Registration step, Bilinear Interpolation method and median filter are used to get a consecutive surface. In the end of the paper, residual parallax magnitude model is presented to illustrate the solution of image Mosaics with big parallax.Steps of the realization of algorithms are produced in the paper with examples in detail. Substantial evidences show that the new method significantly improves the accuracy of model parameters, and work effectively in the traffic accident investigation system.