Research On The Algorithm Of Electronic Image Stabilization Based On Infrared Image

The Electronic image stabilization is a video jitter prevention technique based on image processing.Compared with mechanical and optical stabilization,it has many advantages such as high accuracy and light weight.Nowadays,electronic image stabilization technology has been applied in many fields,such as military reconnaissance system,fire control system,guidance system,civil monitoring system,on-board system,hand-held camera system and etc.Based on the analysis of basic theory of image stabilization technology,the thesis further explores several common image stabilization methods and evaluation methods of electronic image stabilization effects.This article focuses on the topic of infrared video image stabilization and further studies the three major modules of image stabilization technology.The main research work and contributions are as follows.1.Firstly,the basic theory and imaging model of electronic image stabilization technology are introduced.Then,the theoretical analysis and algorithm introduction of the three modules of motion estimation,motion determination and motion compensation are described.Finally several evaluation indexes of the performance of image stabilization are presented.2.When there is small translational jitter in the video sequence,using block matching algorithm will result in more partitioned sub-blocks and calculation speed is too slow.To solve these problems,an electronic image stabilization algorithm is proposed based on partition block matching.The algorithm first divides the image into blocks,divides the sub-blocks with smaller gray levels,and combines the improved hexagonal search path to obtain local motion offsets.Finally,the largest sample statistic method is used to estimate the global motion vector.Experimental results show that the image stabilization accuracy of this algorithm is almost the same as that of the full search algorithm,the speed of motion detection is increased by about 3 times,and the number of search points is reduced by 13 times.3.When there is a local moving object in the video sequence,the traditional gray projection algorithm cannot accurately estimate the global motion vector.Therefore,this paper combines the multi-resolution and Sobel operator to improve the gray projection algorithm.Firstly the algorithm enhances the infrared video image sequence.Then,divides the low-resolution image into layers for the image.Sobel operator is used to calculate the gradient value of each sub-area and performs the row projection on the gradient value.Finally,the method obtains a global motion vector,which effectively avoids the influence of local motion objects on vector estimation according to the statistical maximum sample.The experimental results show that the proposed algorithm estimates the motion vector time by 0.6 times compared with the traditional gray projection algorithm,and the partition processing improves the accuracy of the global motion estimation.4.When there are complex motions in video sequences,the gray projection algorithm based on multi-resolution and Sobel operator cannot effectively estimate.So an algorithm based on the improved gray projection algorithm and Harris operator is optimized.The algorithm firstly enhances the infrared video image sequence,then the feature region in the reference frame is manually selected,and the gradient region projection method is used to obtain the rough motion vector for the feature region projection.The Harris operator extracts the corner points of the feature region and passes the rough motion vector matches the corners of the feature region in the current frame.The RANSAC algorithm is used to eliminate the false matching points.Finally,the global motion vectors are obtained by using the maximum sample statistic method.The experimental results show that this algorithm can effectively estimate the rotational motion and compensate with higher accuracy.Considering different application scenarios,three image stabilization algorithms are proposed to compensate small jitters in translation,jitter in local moving objects,and complicated jitter in the presence of rotational motion respectively.The experimental results show that these algorithms can reduce the computational complexity of motion estimation when compensating for different modes of jitter,and maintain a good matching result.