Study On The Methods Of Video Stabilization Based On Block-matching And Line Feature

With the popularity of video equipment, video file can be filmed everywhere, but the video equipment generally small volume, light quality, is vulnerable to outside interference when shooting, load to video screen jitter and vague. Dithering video screen caused further difficulties in intelligent video analysis and processing, but also affect the appreciation of comfort, it is necessary for the maintenance of stability on video. Video stabilization technology has undergone mechanical image stabilization, optical image stabilization and digital image stabilization in three stages, with the rapid development of electronic technology, digital image stabilization technology in the military and civilian aspects had widespread application.This paper summarizes the existing motion vector estimation and motion compensation methods after reading a lot of domestic and foreign literature about video image stabilization, for translational and rotational jitter, this paper gives two methods of image stabilization, the main work is as follows:(1) The theory of existing digital image stabilization technology method has been researched and summarized. The first part introduces the form of jitter and establishes mathematical models for different jitter form. The next part detail introduces several existing motion vector estimation method; Then, explains the motion compensation methods, and compares each other; Finally, gives the common image stabilization technology evaluation criteria.(2) In view of the existing translation and rotation complex jitter video, this paper first puts forward a kind of video stabilization method of hierarchical block matching method for motion estimation and Kalman filter method for motion compensation. Firstly, this paper construct pyramid structure based on the size of video image, judge the information of blocks in the bottom and remove the float areas block; Secondly, effectively blocks down from the top to search the best matching position for matching point pairs; Then, using the random sample consensus to remove error matching pairs, the remaining points, using the least squares method to estimate the motion vector; Lastly, using Kalman filter to obtain jitter component, suppresses the jitter as the final compensation component and correct the image.(3) In view of the existing translation and rotation complex jitter video, this paper also gives video stabilization method based on linear feature matching for motion estimate and Kalman filter method for motion compensation, Firstly, using the gradient information to extracted straight line to constructor linear feature set and describe the line using gradient vector in local coordinates; Secondly, using the Euclidean distance criterion to match adjacent frames straight linear, so get the motion vector based on the matching linear; Lastly, using Kalman filter to get jitter component, thus suppresses the jitter as the final compensation component and correct the image.Using different scenes video to make experiments for testing two different image stabilization method presented in this paper and compared with existing algorithms, Computer simulation results show that the method presented in this paper can effectively remove jitter to get stable video output.