基于分区融合特征点提取的车载视频稳像方法  

Vehicle Video Image Stabilization Method Based on Partition Fusion Feature Point Extraction

作　　者：刘城逍 吴春林 姚思威 何飞宇 LIU Chengxiao;WU Chunlin;YAO Siwei;HE Feiyu(School of Mechanical Engineering,Hubei Univ.of Tech.,Wuhan 430068,China;Detroit Green Industry College,Hubei Univ.of Tech.,Wuhan 430068,China;Hubei Institute of Product Quality Supervision and Inspection,Ezhou 436000,China)

机构地区：[1]湖北工业大学机械工程学院,湖北武汉430068 [2]湖北工业大学底特律绿色工业学院,湖北武汉430068 [3]湖北省产品质量监督检验研究院,湖北鄂州436000

出　　处：《湖北工业大学学报》2025年第2期11-16,共6页Journal of Hubei University of Technology

基　　金：国家自然科学基金(51575164)。

摘　　要：车载视频在动态背景下往往会出现视频图像抖动的问题,而常用的Harris特征点提取算法在处理此类图像时,所提取的特征点常存在分布不均匀、易产生伪特征点等缺陷,进而影响电子稳像效果,由此提出一种基于分区融合特征点提取的稳像算法。首先利用Prewitt边缘检测提取图像的边缘信息,并对车载视频的边缘提取图像进行分区处理;再使用Harris和AKAZE算法对不同区域图像进行特征点提取;在完成分区特征匹配后,通过计算运动参数并进行加权融合,得到全局运动参数,使用卡尔曼滤波器平滑运动参数;最后,根据得到的运动参数对车载视频进行补偿,从而得到稳像后的视频。实验结果表明,改进后的算法显著提升了稳像效果,稳像后视频的PSNR值约提升了5.2%,且具有较好的实时性。In vehicle videos,there used to be video image jitter in dynamic backgrounds,and the commonly used Harris feature point extraction algorithm,when applied to extract feature points from such images,exhibited uneven distribution and produced pseudo-feature points.This affects the effectiveness of electronic stabilization.Consequently,this paper introduced a stabilization algorithm based on partitioned fusion feature point extraction.Initially,edge information was extracted from the images using Prewitt edge detection,and then in vehicle video frames were partitioned based on their edge content.Subsequently,the Harris and AKAZE algorithms were applied to extract feature points from different regions within the images.After completing the partitioned feature matching,motion parameters were computed and weighted,resulting in a fused global motion parameter.A Kalman filter was then utilized to smooth these motion parameters.Finally,compensation was applied to the in vehicle video frames to obtain stabilized video output,Experimental results demonstrate that the improved algorithm effectively enhances the stabilization performance,with an approximately 5.2%increase in the PSNR value of the stabilized video.Furthermore,it exhibits good real-time performance.

关 键 词：电子图像稳定 Harris特征提取算法 AKAZE特征检测算法 边缘检测技术 特征点提取 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]