基于改进Harris的消音壁亚像素级角点检测算法  

Subpixel-level Corner Detection Algorithm of Sound-Absorbing Wall Based on Improved Harris

作　　者：马韵琪 田明[1] 刘阳 王雨萌 李国旺 MA Yunqi;TIAN Ming;LIU Yang;WANG Yumeng;LI Guowang(School of Opto-Electronic Engineering,Changchun University of Science and Technology,Changchun 130022)

机构地区：[1]长春理工大学光电工程学院,长春130022

出　　处：《长春理工大学学报（自然科学版）》2023年第1期44-51,共8页Journal of Changchun University of Science and Technology(Natural Science Edition)

基　　金：科技部国家重点研发计划项目(KYC-JC-XM-2018-157)。

摘　　要：在计算机视觉领域,角点检测作为图像拼接、三维重建等算法的关键,能够直接影响视觉处理的最终效果。为了进一步提高角点检测在消音壁视觉检测应用场景下的精度,提出了一种亚像素级角点检测算法,首先以Harris算法获得像素级角点为中心角点,然后通过最小二乘法迭代计算不断逼近,最后计算得出亚像素级角点坐标。实验结果表明,算法角点检测的平均偏移量为0.18像素,相较Harris角点检测算法的偏移量降低0.56像素,在确保正确率的基础上更加接近真实值,能够满足在消音壁视觉检测应用场景下的精度需求。In the field of computer vision,corner point detection,as the key to algorithms such as image stitching and 3D reconstruction,can directly affect the final result of vision processing.In order to further improve the accuracy and precision of corner point detection in the visual inspection scenario of silenced walls,this paper proposes a sub-pixel level corner point detection algorithm,which firstly obtains the pixel level corner point as the central corner point by Harris algorithm,and then iteratively approximates it by the least squares method,and finally calculates the sub-pixel level corner point coordinates.Experimental results show that the average offset of this algorithm is 0.18 pixels,which is 0.56 pixels lower than that of the Harris algorithm,and is closer to the true value on the basis of ensuring the correct rate.

关 键 词：亚像素 特征提取 角点检测 HARRIS算法 消音壁 

分 类 号：TN911.73[电子电信—通信与信息系统]