光学学报, 2018, 38 (8): 0815012, 网络出版: 2018-09-06
面向机器人位姿测量的大视场变焦测量方法 下载: 893次
Large Field of View Vision Method for Robot Pose Measurement Based on Zoom Lens
机器视觉 变焦测量 变焦控制 位姿检测 大视场 machine vision zoom measurement zoom control pose detection large field of view
摘要
针对机器人大范围位姿精准测量问题,提出一种大视场位姿测量的变焦测量方法。利用变焦图像匹配点的单应矩阵进行变焦内参动态计算,根据变焦前后靶标位姿求解相机坐标系变换,给出与变焦参数相关的Pn P算法,考虑畸变后进行参数优化,并提出缩放和对焦两步变焦控制策略。提出一种基于图像模板的靶标板位姿测量算法,在靶标可视姿态均能稳定检测并区分标志点。对变焦测量系统进行误差测量实验和大场景跟踪实验,结果表明,不同变焦参数下均具有较高的单点测量精度,平均位置精度最高21.8 μm,在400~1600 mm范围误差0.09 mm,精度较高,使用的变焦相机可实现178.7461~9022.31 mm测量。该方法实现了相机活动范围受限而测量范围可拓展的位姿测量。
Abstract
In order to acquire high precision robot pose measurement when robot moves in large range, we propose a large field of view vision method based on zoom lens. Homography matrix between image matching points is used to dynamically calculate internal parameters while zooming. Camera coordinate system transformation is united according to target pose before and after zooming. In addition, by taking distortion into consideration, we introduce PnP algorithm which is related to zoom parameters and implement parameters optimization. We also propose a two-step zooming control strategy based on zoom and focus for dynamic pose measurement, as well as a target plate position measurement algorithm based on image template, which can detect and distinguish the mark points in any visible effective position of the target. Large number of measurement experiments including signal point error experiment and large scene tracking experiment are carried out to verify the effectiveness of the system. The experiment data show good performance in both pose measurement error and measurement range. Under different zoom parameters, the precision of single point measurement is high, as the average position precision is up to 21.8 μm. In the range of 400-1600 mm, position error is 0.09 mm which cannot be realized with ordinary fixed focus camera. Multiple experiments show that zoom camera can be used to measure during the range of 178.7461 to 9022.31 mm. The proposed method has an extensible measurement space and an advantage of position and orientation measurement in the case of variable measurement range with limited camera motion.
姜涛, 程筱胜, 崔海华, 田威. 面向机器人位姿测量的大视场变焦测量方法[J]. 光学学报, 2018, 38(8): 0815012. Tao Jiang, Xiaosheng Cheng, Haihua Cui, Wei Tian. Large Field of View Vision Method for Robot Pose Measurement Based on Zoom Lens[J]. Acta Optica Sinica, 2018, 38(8): 0815012.