Research On Feature Point Matching And Smoothing Filtering Algorithms In Electronic Image Stabilization

Video files play important role in our daily live.But in some cases,due to the limitations of shooting conditions,the video has some jitter,which affects our experience of watching video.Therefore,we need to reduce or even eliminate the jitter of the video by a certain means to obtain a smooth image sequence.If the video does not need to be played in real time,because there is sufficient post-processing time,and the information of the global video can be obtained,the optimal video output can be obtained through the optimized algorithm.However,the video information only has a prior knowledge without a posterior knowledge in the real-time broadcast situation,at this time,the speed and accuracy of the algorithm is particularly important in order to ensure the smooth and stable output of video.With the development of image processing technology and filtering algorithm,electronic image stabilization is becoming more and more powerful.The processing of electronic image stabilization mainly includes motion estimation and motion compensation.In this paper,two modules are studied,and the following work is done1)The workflow of electronic image stabilization technology is briefly introduced.The core idea of classical motion estimation algorithm and common motion compensation algorithm are introduced,and their limitations are analyzed.2)Analyzing the advantages and disadvantages of classical algorithms and studying the characteristics of scenarios,a feature point matching algorithm based on circular template is proposed on the basis of classical feature point matching algorithm.Based on the rotation invariance of some feature vectors of circular template and the more abundant information of edge pixel distribution provided by edge detection operator,the searching range of the second layer feature matching points is greatly reduced on the basis of determining the first layer feature matching points.Geometric moment invariants are added to correct the first feature points of matching errors,which not only ensures the accuracy,but also does not affect the real-time performance.The matching algorithm takes advantage of some rotation invariant characteristics of the circle,and can be applied to video stabilization with parallel and rotational motion,with less computational complexity,and can meet the needs of real-time detection3)According to the discrete distribution of motion vectors,a smoothing filter algorithm based on adaptive weights of first derivative is proposed.According to the different performance of the first derivative of normal motion and jitter on the two-dimensional discrete curve,the adaptive weights are obtained.The correlation coefficient is introduced.After the first filtering,the values in the neighborhood are weighted averaged again to eliminate the peaks on the curve.The final curve is very close to Kalman filter,but for this jitter scenario,its operation efficiency is higher.4)The program of feature point matching and smoothing filter module in motion estimation and motion compensation is compiled in the development environment of VC++.Experiments are made on the result shaky car.avi video of electronic image stabilization common video library.Relevant data are obtained and analyzedExperiments show that the proposed feature point matching algorithm based on circular template can effectively reduce the number of matching points and the amount of computation on the basis of a sufficient number of successful matching points when the motion vectors of translation and rotation are dominant in the jitter scene and the scaling motion is not obvious.Moreover,the smoothing filtering algorithm based on the first derivative adaptive weights proposed in this paper not only satisfies the better filtering effect,but also has less computational complexity.It can improve the efficiency and meet the requirements of real-time image stabilization to a certain extent.