Research And Implementation Of Electronic Image Stabilization Algorithm Based On FPGA

Embedded cameras are widely used in intelligence detection,medical inspection,security monitoring and other fields.In some scenarios,the inevitable mechanical vibration will cause the video to be not clear enough and affect the quality of the information in the video.Therefore,the image stabilization technology of the camera video data is particularly important.Electronic image stabilization has become the mainstream image stabilization solution for today's embedded applications due to its small size,low power consumption,and good effects.Among them,electronic image stabilization based on SURF feature point extraction is valued due to its high robustness.However,the original SURF algorithm has high complexity and poor real-time software implementation,and there is room for improvement in the accuracy of its description and matching results.To solve above problems,this paper researches and improves FPGA video image stabilization technology based on the SURF algorithm,which effectively improves the video image stabilization effect and real-time performance.In this paper,the process and hardware implementation of SURF algorithm are firstly improved.The adaptive thresholding mechanism is introduced,so that a similar number of feature points can be extracted from each frame;the circuit implementation takes advantage of the abundant hardware resources of FPGA,which greatly improves the parallel processing capability of complex computing functions such as non-maximum suppression and precise positioning of feature points required for image stabilization;the main direction generation step is eliminated to speed up the description;the segmentation sub-region in the traditional descriptor generation is modified from 5×5 to 8×8 with overlapping sampling,performing Gaussian weighting after sub-regions accumulating completion,and introducing threshold mechanism during matching to enhance the discrimination of description and the accuracy of matching;In the estimation of the motion model,the time-consuming iterative part of the RANSAC algorithm is replaced by a parallel implementation designed for FPGA,which greatly reduces the processing time of the algorithm.For the above improvements,this paper then simulates the overall algorithm using MATLAB and designs and implements hardware circuits for each module.Finally,this paper uses vivado to simulate the function of each module of the circuit,and verifies the whole steadicam system on xilinx's Zynq-7000 series xc7z100 model FPGA platform,and uses the test set to compare with the algorithms proposed by other related work.The results show that the electronic image stabilization system designed in this paper can effectively improve the peak signal-to-noise ratio PSNR of video by 13.85%on average;meanwhile,with the parallelism advantage of FPGA,the circuit in this paper can process video with resolution of 640×480 at 125 fps,which has obvious advantages in real-time and image stabilization.