Research On Electronic Image Stabilization Algorithm Based On Feature Point Matching And Its Hardware Implementation

Electronic image stabilization is a new,generation of image stabilization technology that combines electronic science and digital image processing.Compared with traditional mechanical and optical image stabilization,electronic image stabilization technology has gradually become the main research direction of image stabilization technology due to its advantages of small size,high precision,and low cost.It is widely used in airborne,shipboard,vehicle and other occasions.Because these camera carriers are easily disturbed by external factors during the shooting process,the output video sequence has irregular jitter,which seriously affects the effective use of video information.The main purpose of image stabilization technology is to eliminate video inter-frame jitter and improve video quality and observations.The paper analyzes the basic principle of electronic image stabilization based on feature point matching.The contrast of the whole image is calculated based on the gray value difference of 16 pixel points on the circumference of the image pixel to be measured and its radius is 3,and it is used as the FAST feature.The dynamic threshold of the point detection algorithm can detect suitable number of feature points for video images under different conditions.and then the non-maximal suppression method was used to screen the detected feature points,which effectively improved the phenomenon of feature point aggregation in the image.Next,the gray inferential centroid method is used to improve the rotation invariance of the BRIEF feature point description algorithm,so that it can describe the feature points of an arbitrary rotation angle image and match the feature points through the Hamming distance and 2 nearest neighbor methods.The PROSAC algorithm is used to further filter the matched feature point pairs,reducing the wrong matching point pairs,and calculating the global motion vector through the affine transformation model.Kalman filtering is used to separate the subjective motion vector and the random jitter component in the global motion vector.Finally,the dithered image is compensated by the bilinear interpolation method according to the dithering component,and the undefined pixel area of the compensated image is reconstructed by the reference frame image.The quality of the electronic image stabilization algorithm implemented in this paper was evaluated from both the peak signal-to-noise ratio(PSNR)and the image stabilization range.The traditional electronic image stabilization systems are implemented in the form of computer software.The inherent serial work mode makes it difficult for computers to meet the real-time requirements of electronic image stabilization systems that require excessive computational load.Therefore,the paper established a hardware platform for the electronic image stabilization system.The ALTERA STRATIXIV series FPGA was used as the control core to realize the hardware circuit design of the image stabilization system and enhance the real-time performance of the system.The system mainly includes image stabilization processing units,System bus and storage module,video input and output modules.The image stabilization system provides the input video source to the video acquisition module through the OV5642 camera,and then transmits the valid image information in the video to the image stabilization processing unit.Finally,the processed video sequence is output to the VGA display for display.The image stabilization processing unit includes image feature extraction,feature point description and matching,calculation of global motion vector,kalman filtering and image recovery and other modules,and data transmission and buffering between various functional modules are realized through the system bus and external storage.The modelsim software of Mentor Corporation was used to verify the function of the entire electronic image stabilization hardware system,and the experimental simulation results are given.Finally,board-level verification of the image stabilization system is performed on the FPGA development board to give the system performance and power consumption.Experimental results show that the image stabilization system platform can recover the translation and rotation jitter of a video image with a maximum resolution of 1024×768 and a frame rate of 28 fps.