To address the issue of large measurement errors in binocular vision devices and the inconsistent distribution of single-point measurement errors within the field of view, which leads to poor accuracy and stability in the measurement of 6D poses a measurement error compensation method has been proposed. To solve the problem of poor fitting effects of traditional regression models in the case of small samples and multiple inputs and outputs, an algorithm based on multi-output least squares support vector regression (MLSSVR) is adopted, which achieves accurate prediction of single-point measurement errors, and a genetic algorithm is used to optimize model hyperparameters globally. To realize the pose measurement error compensation of the tracking coordinate system, combined with the principle of pose measurement, the linear problem is transformed into an optimal estimation problem. By minimizing the coordinate transformation error function, multiple single-point prediction errors are converted into multi-point coupled errors, thereby achieving 6D pose measurement error compensation. Experimental results show that the proposed method reduces the average position measurement error of the tracking coordinate system by 63.4%, 45.2% and 75.0%, and the average attitude measurement error by 50%, 48.4% and 53.1%, respectively. Compared with other existing compensation methods, the method proposed can significantly improve the measurement accuracy of binocular vision devices and ensure the effectiveness of measurement error compensation from different observation angles.
ZHENGLianyu, FUQiang, FANWei, et al. A method for in-situ pose perception of large cylindrical components based on binocular vision and prior machining data[J]. Journal of Mechanical Engineering, 2023, 59(11): 129-146. (in Chinese)
ZHOUKejie, FENGChang. Three-dimensional measurement technology based on binocular vision[J]. Computer Measurement & Control, 2019, 27(1): 22-25. (in Chinese)
[6]
LIB, LIY, TIANW, et al. Pose accuracy improvement in robotic machining by visually-guided method and experimental investigation[J]. Robotics and Autonomous Systems, 2023, 164: 104416.
[7]
SHUT, GHARAATYS, XIEW, et al. Dynamic path tracking of industrial robots with high accuracy using photogrammetry sensor[J]. IEEE/ASME Transactions on Mechatronics, 2018, 23(3): 1159-1170.
WANGLonghua, ZENGGuohua, HUHuafeng, et al. Optimal design of binocular vision system for accurate 3D data measurement[J]. Electronic Measurement Technology, 2022, 45(7): 99-109. (in Chinese)
HUOJ, ZHANGG, CUIJ, et al. Corrected calibration algorithm with a fixed constraint relationship and an error compensation technique for a binocular vision measurement system[J]. Applied Optics, 2018, 57(19): 5492-5504.
[18]
HONGZ, LIZ, TONGX, et al. A high-precision recognition method of circular marks based on CMNet within complex scenes[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2022, 15: 7431-7443.
[19]
卫松. 面向复材铣边加工的机器人视觉伺服控制系统[D]. 南京: 南京航空航天大学, 2022.
[20]
SUYKENSJ A K, VANDEWALLEJ. Least squares support vector machine classifiers[J]. Neural Processing Letters, 1999, 9(3): 293-300.
DAIPeng, ZHOUPing, LIANGYanzhuo, et al. Multi-output least squares support vector regression modeling based adaptive nonlinear predictive control and its application[J]. Control Theory & Applications, 2019, 36(1): 43-52. (in Chinese)
[23]
ZHOUS. Sparse LSSVM in primal using cholesky factorization for large-scale problems[J]. IEEE Transactions on Neural Networks and Learning Systems, 2016, 27(4): 783-795.