Study Of Electronic Stabilization Techniques For Video Sequences

Electronic Image Stabilization(EIS) has been widely applied in many fields for its obvious talents, such as high accuracy, low power consumption, high intelligence, small volume, light weight and low cost, etc. As a new technology to stabilize video image sequence, EIS mainly use image processing to determine the motion between frames and then compensate it to eliminate the non-normal offset, such as translation, rotation and scaling. EIS System is mainly composed of three parts, such as Image Preprocessing, Motion Estimation, and Motion Compensation. Among them, Motion Estimation is most significant for directly determining the performance of system.In this paper, according to the different kinds of perturbations, the video are divided into four status, such as in the settled scene, in the dynamic scene with fixed-motion-object , in the scanning scene and in the scene with foreground. In order to deal with the dithering in the scene with multiple small moving targets, a new optimum motion estimation is presented to realize fast and robust image stabilization. In this algorithm, by using block absolute difference method, foreground objects is removed from the image, while useful reference blocks which are indispensable for accurate motion estimation are selected by analyzing gradient information. At the same time, an improved block matching algorithm is used to eliminate wrong motion vectors and the inter-frame motion parameters are obtained with Newton iteration algorithm. For the dithering in the dynamic scene with fixed-motion-object or in the scene without foreground, a multi-resolution tracking image stabilization algorithm is proposed based on spatial-temporal gradient in image sequence. Firstly, region of interest(ROI) in the video is selected for object tracking. Secondly, the images are down-sampled with Gaussian Pyramid method to reduce the computation complexity.Finally, the accurate affine transformation parameters are calculated with the method of error approximation. Aiming to deal with the dithering in the scanning scene, a filter compensation algorithm is used to get the steady video output. In addition, in order to reconstruct the lost information of images after processing, the video mosaic technique is proposed. The effectiveness of the algorithm is confirmed by extensive experiments over a wide variety of videos and the subjective effect is perfect. Compared with the initial video sequences, the PSNR of the stabled sequence is improved significantly.