一种结合特征点匹配与仿射变换的电子稳像算法  

An Electronic Image Stabilization Algorithm Combining Feature Point Matching and Affine Transformation

作　　者：何涛[1,2] 柯黎明 何嘉奇 王少东 HE Tao;KE Liming;HE Jiaqi;WANG Shaodong(School of Mechanical Engineering,Hubei Univ.of Tech.,Wuhan 430068,China;Hubei Key Lab of Modern Manufacture Quality Engineering,Wuhan 430068,China)

机构地区：[1]湖北工业大学机械工程学院,湖北武汉430068 [2]现代制造质量工程湖北省重点实验室,湖北武汉430068

出　　处：《湖北工业大学学报》2022年第2期1-5,共5页Journal of Hubei University of Technology

基　　金：湖北省自然科学基金(51275158)。

摘　　要：车载视频具有背景运动复杂及光照不均匀等特点,造成现有的电子稳像算法在对车载视频进行实时电子稳像处理时,存在图像特征点提取困难,表现出较差的稳像鲁棒性及稳像实时性。针对上述问题,提出一种结合特征点匹配与仿射变换的电子稳像算法,即先通过分区的特征点提取策略,结合优化后的RANSAC算法,提取出正确的匹配点对,再利用异加权值的运动估计方法及仿射变换模型估计运动参数,并采用卡尔曼滤波器对估计的运动参数进行平滑处理,最后用平滑后的运动参数对待处理视频帧序列进行图像补偿,得到稳定的视频帧序列。实验结果表明,该算法在实时电子稳像上有良好的效果,相比原始视频,稳像后视频的PSNR值约提升32.77%,SSIM值约提升57.83%。Due to the complex background motion and uneven illumination characteristics of on board video,existing electronic image stabilization algorithms have difficulties in extracting feature points in real time electronic image stabilization processing of on board video,which show poor image stabilization robustness and real time performance.To solve these problems,an electronic image stabilization algorithm combining feature point matching and affine transformation is proposed.First,the correct matching point pairs were extracted through the partitioned feature point extraction strategy combined with the optimized RANSAC algorithm,then the motion estimation model of difference weighted value and affine transformation model were used to estimate the motion parameters,and the Kalman filter was used to smooth the estimated motion parameters.Finally,the smoothed motion parameters were used to compensate the video frame sequence to obtain a stable video frame sequence.Experimental results show that the algorithm has a good effect on real time electronic image stabilization,in which the PSNR value of the video after image stabilization is about 32.77%higher than that of the original video,and the SSIM value of the original video is about 57.83%higher than that of the original video,which is successfully applied in the real time electronic image stabilization application of vehicle video.

关 键 词：特征点提取 优化RANSAC算法 运动估计 图像补偿 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]