摘要
Linemod⁃2D是一种基于特征匹配的快速目标识别算法,其像素级的定位精度,使其不能直接用于亚像素级高精度要求的工业检测场景。针对上述问题进行改进,提出一种基于改进Linemod⁃2D的快速形状匹配算法。采用网格化方式提取模板特征,并用Zernike矩对边缘点进行亚像素校正,以获取精确的位置和角度作为匹配特征。在图像金字塔的上层采用改进的Linemod⁃2D进行快速的粗匹配,其角度量化方向扩展为16个。在图像金字塔底层,采用角度量化扩展为360个方向的匹配策略进行精匹配,得到像素级的匹配位置和小于0.5°的角度误差,然后采用最小二乘平差法得到亚像素级的匹配位置和精确的角度。实验研究结果表明,该算法的匹配位置误差小于±0.04像素,角度误差小于±0.015°,其匹配精度和速度优于CkVisionBuilder的形状匹配算法,接近Halcon的形状匹配算法。
As the Linemod⁃2D is a fast target recognition algorithm based on feature matching,and its pixel⁃level positioning accuracy cannot be directly used in the industrial detection scene with subpixel⁃level high precision requirements,a fast shape matching algorithm based on improved Linemod⁃2D is proposed for above problems.The gridding manner is used to extract the template features,and the Zernike moment is used to perform the subpixel correction for the edge points,so as to obtain accurate position and angle as the matching features.The improved Linemod⁃2D is used in the upper layer of the image pyramid for fast coarse matching,and its angle quantization direction is expanded to 16.In the bottom layers of the image pyramid,the matching strategy that the angle quantization is extended to 360 directions is used for the fine matching,to get the pixel⁃level matching position and the angle error that is less than 0.5°,and then the least square adjustment method is used to get the subpixel⁃level matching position and the accurate angle.The experimental results show that the matching position error of the algorithm is less than±0.04 pixels and the angle error is less than±0.015°.The matching accuracy and speed of the proposed algorithm are better than those of the shape matching algorithm of CkVisionBuilder,and close to those of the shape matching algorithm of Halcon.
作者
陈全义
陈三风
胡涛
刘祚时
CHEN Quanyi;CHEN Sanfeng;HU Tao;LIU Zuoshi(Jiangxi University of Science and Technology,Ganzhou 341000,China;Institute of Information Technology,Shenzhen Institute of Information Technology,Shenzhen 518172,China)
出处
《现代电子技术》
2021年第6期159-164,共6页
Modern Electronics Technique
基金
国家自然科学基金:基于内容的智能视频编码优化研究(61871154)
深圳信息职业技术学院平台培育项目(PT201704)。
