Abstract：In order to solve the problems of low degree of automation, poor real-time performance and weak portability of traditional sorting robots when they complete designated actions through manual instruction, an experimental platform of sorting robots based on machine vision is designed, and a high-precision 6D pose estimation method based on the mixed progressive registration strategy of coarse registration and enhanced fine registration is proposed. Firstly, the 3D point cloud information collected by the vision sensor is preprocessed, such as background information removal, target region clipping, ROI extraction, etc., and the sampling consistency initial registration algorithm (SAC-IA) is used for pose rough registration. Then the iterative closest point (ICP) algorithm is used for fine registration and the normal distribution transformation (NDT) is used to enhance the fine registration to achieve a high precision 6D position. The experimental results show that the integration of coarse registration and enhanced fine registration can quickly and accurately obtain the 6D position and attitude of the target to be captured, with errors controlled within 1.5mm and 2 ° compared to the theoretical position and attitude, respectively, and meet the practical needs of sorting robots. The proposed method can be applied to the machine vision based robot assembly, grinding and other occasions with high precision 6D pose estimation.
Rusu, R B, Blodow N, Beetz M. Fast Point Feature Histograms (FPFH) for 3D registration［C］//The 2009 IEEE International Conference on Robotics and Automation. Kobe, Japan, 2009.
［5］
Zeng A, Yu K T, Song S, et al. Multi-view Self-supervised Deep Learning for 6D Pose Estimation in the Amazon Picking Challenge［C］//IEEE International Conference on Robotics and Automation. Singapore, 2017.
Rusu R, Blodow N, Beetz M. Fast point feature histograms (FPFH) for 3d registration［C］//2009 IEEE international conference on robotics and automation. 2009:3212-3217.
Kehl W, Manhardt F, Tombari F, et al. SSD-6D: Making rgb-based 3d detection and 6d pose estimation great again［C］// IEEE international conference on computer vision(ICCV). Venice, Italy, 2017.
Lu H, Xu X S, Wang S G, et al. Blade Profile Profile Error Evaluation Based on Improved ICP Algorithm ［J］. Acta Metrologica Sinica, 2022, 43(8): 1015-1020.
Li F F, Xu D F, Wang C, et al. Densefusion: 6d object pose estimation by iterative dense fusion［C］// Computer Vision and Pattern Recognition. Long Beach, USA, 2019.
Fu G Q, Corradi P, Menciassi A, et al. An integrated triangulation laser scanner for obstacle detection of miniature mobile robots in indoor environment［J］. IEEE/ASME Transactions on Mechatronics, 2011, 16(4): 778-783.
Xu G Y, Dong H W, Qian J H, et al. Pose estimation algorithm for random bins based on point pair features［J］. Laser & Optoelectronics Progress, 2020, 57(18): 342-350.
Yue Y, Zhao Z C. Calculation method of work piece pose based on regional point cloud［J］. Chinese Journal of Construction Machinery, 2021, 19(4): 295-301.
［11］
Song K T, Wu C H, Jiang S Y. CAD-based pose estimation design for random bin picking using a RGB-D camera［J］. Journal and Robotic Systems, 2017, 87(3-4): 455-470.
［14］
Magnusson M. The three-dimensional normal-distributions transform: an efficient representation for registration, surface analysis, and loop detection［D］. rebro: rebro University, 2009.
Xu P, Xu F Y, Chen H. Large Scene Segmentation of Outdoor Laser Point Cloud Based on Fusion and Registration ［J］. Acta Metrologica Sinica, 2022, 43(3): 325-330.
Hao W, Wang Y H, Ning X J, et al. Survey of 3D object recognition for point clouds［J］. Computer Science, 2017, 44(9): 11-16.
Zhang T, Xiao Z, Zou Y B, et al. Workpiece Recognition and Pose Estimation Based on 3D Point Cloud Features［J］. Machinery Design & Manufacture, 2022, 372(2): 252-256.
Wen X L , Kang C S, Song A G, et al. Study on Robot Accuracy Based on Full Pose Measurement and Optimization ［J］. Chinese Journal of Scientific Instrument, 2019, 40(7): 81-89.
2023, Vol. 44 
