Research On The Adaptive Digital Video Stabilization Algorithm

With the development of artificial intelligence,video is playing an increasingly important role in daily life,industrial development,intelligent medical treatment and other fields.However,due to various reasons in the shooting process,the picture is unstable,which will not only cause a bad viewing experience for the video viewers,but also affect the subsequent video image analysis and processing and recognition process.Therefore,the research of digital image stabilization method has important application value.For the video jitter with complex motion state,in order to get a better estimation of the motion model,obtain the appropriate filtering trajectory,and ensure the video image stabilization effect,this paper mainly does the following aspects:(1)This paper first introduces the process of digital video stabilization algorithm and the current commonly used evaluation methods,and summarizes the research status of digital video stabilization algorithm at home and abroad,including 2D algorithm,3D algorithm and the emerging 2.5D algorithm.And it analyzes some technical difficulties existing in the current digital video stabilization algorithm(2)This paper presents a video image stabilization algorithm based on multi-path optimization.In the motion estimation stage,video images are divided into uniform sub-blocks,and the FAST feature points extracted respectively based on machine learning.Then,the KLT algorithm is used to track the motion of each sub-block,and the motion trajectory of each sub-block is calculated respectively.In the phase of motion smoothing,the optimal global smoothing trajectory can be obtained by the improved energy function with adaptive adjustment parameters for multi-path optimization,according to the current motion state,and the final image stabilizing video can be obtained through compensation.(3)This paper presents a digital video image stabilization algorithm based on improved adaptive Kalman filter.In the motion estimation stage,the Shi-Tomasi corner point with distance constraint and the Pyramid Lucas Kanade Optical Flow method are combined to calculate the motion trajectory.In the process of motion smoothing,the traditional Kalman filter algorithm is improved,and the deviation probability distribution is added to adjust the parameters to carry out real-time tracking.Finally,the motion compensated image stabilization video is carried out.A large number of computer simulation experiments are carried out for different scenes,different categories and different motion states of the video.And we compared these experimental results with some widely used and effective image stabilization methods at present.Experimental results show that the proposed method has a good image stabilization effect.