Research On Electronic Image Stabilization Technology

The instability of the platform results in the jitter between the video sequences of the photographic equipment.This jitter will increase the difficulty of subsequent video processing,and also affect the visual effect and even cause erroneous judgement.With the development of computer and the development of industrial technology,the electronic image stabilization technology integrates digital image processing technology and electronic technology,which has high precision and low cost,has become a hot spot of research and discussion in various countries.This paper analyzes three modules in the electronic image stabilization technology.Because the time-consuming and precision of the electronic image stabilization algorithm is the motion estimation,this paper mainly analyzes the key module of the motion estimation,and selects the motion estimation algorithm based on the feature matching.After studying the algorithm of motion estimation based on feature matching.We found that:(1)the feature points detected by the FAST algorithm has small scale aggregation phenomenon,which increases the computation amount of subsequent motion estimation.(2)there are a lot of mismatches in feature point matching,which will affect the accuracy of motion estimation and reduce the effect of image stabilization.Considering the above problems,it is very important to study the motion estimation algorithm based on feature matching.Aiming at the problem of small scale aggregation of feature points in feature matching motion estimation algorithm,this paper proposes an improved algorithm.The algorithm divides the image into multiple equal-sized pixel areas and uses the FAST algorithm to detect the feature points using the area as the basic unit.Then,the feature points detected in the area are screened by comparing the size of pixel gray values.At the same time,the FREAK descriptor with rotation invariance is used to describe the feature vector.In order to solve the problem of mismatching of feature points,this paper first makes coarse matching for feature points and then performs fine matching.The coarse matching firstly introduces the nearest neighbor ratio,and the feature points are filtered and matched by combining with the Hamming distance.Then the RANSAC algorithm is used to further match the precision.The two step matching greatly reduces the effect of mismatch and local motion on the motion estimation,and it obtains a higher precision global motion vector,thus it can establish an accurate motion model to calculate more accurate motion parameters.In view of the problem that the image appears undefined area because of the image transformation,this paper uses adjacent frames to compensate the undefined regions of the transformed images,and the image after the compensation is basically complete.Aiming at the algorithm proposed in this paper,we set up the experimental simulation platform,and we also compares several image and video sequences.The results show that our algorithm can get more robust feature points and reduce the number of mismatches effectively.In order to improve the processing performance of the original algorithm,a stable image quality factor with higher image quality is obtained.