Research On The Image Sequence Stabilization Technology Based On Inertial Navigation Information

With the background of terminal guidance using image information on low cost strapdown seeker, this thesis makes a research on the stabilization of image sequence which is get from the camera mounted on a missile. The inevitable random vibration of the missile results in the dithering of the got image, which severely deteriorates the observation quality and next processing steps. The purpose of the paper is to eliminate the influence of the image dithering and export a stable image sequence.The thesis is mainly concerned with the follows:1. A distorting image correction method using inertial navigation information is presented. This method achieves the geometric correction by transforming the image collected by the homing head into the one collected at the standard position using the attitude and position information of the missile.2. The current motion estimation technique for image sequence is investigated. The basic ideas of the main algorithms in each category are expatiated; the merits and defects of each algorithm are compared.3. The current corner detection methods are analyzed. On which a binary Moravec+SUSAN combinatorial corner detection algorithm is presented, which markedly reduces the numbers of potential corners and gets a high execution efficiency.4. A set of feature points correspondence methods are presented. First, a correspondence algorithm based on centroid is presented on condition that there is no noise and no local motion point. Second, a correspondence algorithm based on partial Hausdorff distance is presented to mend the disadvantage of the first algorithm. At last, for the complicated condition such as noise and local motion, a correspondence algorithm based on the first pair of matching points and partial Hausdorff distance is presented. And the suppression effect of the last algorithm on noise and local motion points is demonstrated.5. A coarse method and a delicate method for motion vector estimation based on feature points correspondence are presented. And, based on the motion estimation methods, a image stabilization method is developed. On the occasion of no intentional movement, this method will effectively eliminate the displacement vibration and rotation vibration; on the occasion of existing intentional movement, on the aid of inertial navigation information, this method will effectively eliminate the random vibration and totally preserve the intentional movement.The simulation results indicate that, the geometric correction algorithm effectively eliminates the geometric distortion of images, and the image stabilization algorithm eliminates the random dithering exactly and exports a stable image sequence.