Study On Low-Latency Video Stabilization Methods

Video stabilization has always been one of the focuses in the field of image and video processing for a long time.Many of the traditional post-processing methods of video stabilization usually have long processing delays.However,many application scenarios need more efficient and low latency video stabilization methods,such as real-time video transmission of unmanned aerial vehicles,making live online using a handheld camera and so on.The low-latency video stabilization method can get stable videos with a short delay while shooting video.It has no impact on users to watch online,and also helps video's subsequent storage and transmission.Therefore,the low delay video stabilization method gets more and more attention.In this paper,a low-latency video stabilization method based on Kalman filter is proposed to solve the problem that the stabilization effect of low-latency video stabilization methods is still not ideal and they rely on high performance processors.We improve the prediction equation when using Kalman filtering method to remove the shake of the camera path.In the case of fully considering the continuity of the motion between video frames,which is the motion "inertia" between the two frames of a video.This develops a more stable video movement.The Kalman filter only needs the video frame information of the previous state and its recursive estimation process has low computational complexity and spend less time.Therefore,our method is very fast and requires low processing and storage performance of the processor.We can obtain the stable video with only one frame latency.The experiments demonstrate our effect is similarly to the other method.In order to improve the stabilization quality of low-latency video stabilization method,a video stabilization method based on the on-the-fly total variation optimization is proposed.Different from the filtering method,we use the energy function to optimize the shaking camera path.We use the on-the-fly total variation minimization method to smooth the shaky path.We reduce the total variation of the shaking camera path to obtain a stable path.Compared with the filtering-based method,we can utilize the motion information of more frames before the current frame.So we can get a more stable path.Because we utilize an on-the-fly method for total variation optimization,so we can obtain the stable video with only one frame latency and get pleasing stabilization results.