摘要
结构光三维测量技术是获得物体三维信息的重要途径,激光条纹中心线提取是影响结构光三维测量精度和速度的关键因素。提出了一种适用于多场景下结构光三维测量的激光条纹中心线提取方法,充分利用图像中激光条纹的几何信息和相关性生成自适应卷积模板,实现激光条纹图像的滤波和增强处理,使激光条纹横截面灰度值满足高斯分布;经灰度加权法实现激光条纹中心线的亚像素精度定位与提取。实验测试结果表明:该方法可实现多场景下形状、材质各异物体的条纹中心线提取,有效克服了激光条纹亮度分布不均、噪声干扰等影响,单幅图像处理时间缩短为0.107 s且相对误差减少到0.076 5%,有效提高了激光条纹中心线的提取精度和速度。
3D measurement of structured light technology is an extremely vital approach to obtain 3D information of objects. Extraction of the centerline of laser stripe is a key factor that could affect the accuracy and speed of 3D measurement of structured light in the meantime. A method of extracting centerline of laser stripe for 3D measurement of structured light that adaptive to multiple scenarios was proposed. The adaptive convolution template was generated by making full use of the geometric information and correlation of laser stripe in the image, which can filter and enhance the image quality of laser stripe and enable the gray value of cross-section of laser stripe satisfy the Gauss distribution. The sub-pixel accurate localization and extraction of laser stripe centerline were realized by gray weighted algorithm. The experimental results show that the proposed adaptive convolutional algorithm can extract the laser stripe centerlines of the objects with different shapes and materials based on multi-scenarios and overcome the influence of uneven brightness and noise at the same time. Based on the algorithm, the extraction time of single frame is shortened to 0.107 s and the relative error is reduced to 0.076 5%,which improves the extraction accuracy and speed of laser stripe centerline effectively.
作者
宋晓凤
李居朋
陈后金
李丰
万成凯
Song Xiaofeng;Li Jupeng;Chen Houjin;Li Feng;Wan Chengkai(School of Electronics and Information Engineering,Beijing Jiaotong University,Beijing 100044,China;Beijing Century Real Technology Co.,Ltd.,Beijing 100085,China)
出处
《红外与激光工程》
EI
CSCD
北大核心
2020年第1期205-212,共8页
Infrared and Laser Engineering
基金
国家自然科学基金(81671034)
北京交通大学教育基金会重点项目(9907005501)。
关键词
三维测量
线结构光
条纹中心提取
自适应卷积
灰度加权
3D measurement
line structured light
stripe centerline extraction
self-adaptive convolution
gray weighted
