Structured Light High Precision Measurement And Workpiece Pose Estimation

Achieving stable three-dimensional measurement of workpieces and accurate pose recognition capabilities in the complex and variable environment is very important for robots in the field of bin-picking problem.This paper focuses on the high-precision structured light 3D measurement and the recognition of workpiece poses simultaneously.Improved technologies for surface structured light encoding and decoding are proposed firstly,and a perfect 3D surface structured light measurement system is developed.Then the paper builds a workpieces classification network using transfer learning based on the 3D point cloud data obtained from surface structured light measurement system.Finally,a 2D-3D integrated workpiece pose estimation method is proposed.The main contributions of this research are as follows:1.High-precision surface structured light 3D measurement:Aiming at the problems of low measurement efficiency and poor scalability in the existing 3D structured light measurement methods,an RGB-based Gray code Shifting and Positive/Negative coding method is proposed,in which the positive and negative coding technology is added to im-prove the measurement accuracy according to the traditional Gray code encoding method and the RGB channel coding method is used to improve the measurement efficiency.As for decoding method,a sub-pixel accuracy decoding method is proposed based on gray moment which breaks through the limitation of hardware from the aspect of algorithm.This de-coding method can improve the measurement accuracy of surface structured light system without any changes of the hardware conditions,thereby greatly reduces the measurement cost.An experimental platform for surface structured light 3D measurement and supporting soft-ware are developed.At the same time,related accuracy verification and application experiments are performed.The experiments show that the measurement accuracy of the surface structured light 3D measurement system can reach 0.01 mm.2.Workpiece pose recognition based on point cloud:This paper improves the PointNet network architecture and proposes a workpiece point cloud classification network based on transfer learning to overcome the difficulties caused by the disor-der,mutual correlation,and transformation invariance of point cloud data.The proposed network also has the ability to learn quickly and obtain a high classification accuracy with the support of a small amount of data.A workpiece point cloud dataset is constructed and is used to verify the effectiveness of the classification method.After classifying the point cloud,a 2D-3D integrated algorithm is proposed to estimate work-piece pose.Firstly,the SIFT features of the 2D image are used for rough matching,and the 3D point cloud template is transformed to a reasonable initial pose based on the results of the rough matching.Then,the Iterative Closest Point method is used for fine matching on the point clouds.This two-step process can speed up the convergence speed of the algorithm while effectively avoid-ing the algorithm to converge to a local optimal value,and can accurately estimate the pose of the object.