Research On Video Image Electronic Derotation Algorithm

Missile imaging is the key to TV terminal guidance,which is because the shell needs to rotate during the flight to keep the shot stable,so the video image needs to be subjected to the eliminate rotation process.Video electronic derotation has a wide range of use in both military and civilian.In this paper,a pure software electronic racemization is proposed for the shortcomings of the current electronic eliminate rotation system,obtaining the rotation vector to achieve the purpose of derotation according to the characteristic relationship between the video image frames.The work of this paper is mainly divided into the following three parts:The video image is pre-processed to better detect the characteristic of the image.Firstly,the image is enhanced by histogram bidirectional equalization.For the fog interference often encountered in the missile video image,combined with the atmospheric scattering model and the steering filter to defog the image,the experiment shows that it has a good defogging effect.In the case where the rotation axis is stable,which means only the rotation and the scale transformation exist between the image frames.We introduce the log polar transformation turning the rotation and scale transformation in the original image into translational motion.We study on the common motion vector estimation,and then compare and analyze the time and accuracy of each method.On this basis,this paper proposes a peak grayscale projection derotation algorithm based on log polar coordinate system.The experimental results show that the derotation time of a frame of video image on the host computer is 19 ms and the accuracy is high,which can meet the real-time performance.For an on-board imaging device,there is a rotational motion while there is also a jitter parallel to the image plane.For such a rotation axis is unstable,that is,when there are rotation,translation and scale transformation between image frames,it cannot be directly converted to log polar coordinates for processing.In view of this situation,this paper first analyzes the traditional phase correlation method and the Fourier-Mellin algorithm.The experimental results show that although the rotation vector can be obtained,the algorithm complexity is not as high as real-time requirements.In this paper,we propose a motion vector estimation algorithm based on rotation invariant feature points.We analyze and compare the SIFT、SURF and ORB feature point extraction and description matching algorithms.Experiments show that the SURF algorithm is the best in the derotation effect and the ORB algorithm is faster.Aiming at the lack of accuracy of the ORB feature point derotation algorithm,the judgment of the Euclidean distance between feature points is added.The experimental results show that it takes 87 ms to process a video image and is suitable for large rotation angle.Finally,we use the algorithm to test on a set of video frames whose rotation axis is unstable.The video frame subtraction processing can be basically meet the accuracy and real-time requirements of the system.