Three-dimensional object pose estimation method based on PVFH characteristics

Abstract

The invention discloses a three-dimensional object pose estimation method based on PVFH characteristics, and relates to a three-dimensional point cloud object pose estimation method. The invention provides an improved global feature descriptor PVFH. The descriptor is used for achieving three-dimensional point cloud object pose estimation. The method comprises the steps of in an offline stage, rendering from the CAD model to obtain point clouds of a plurality of visual angles, and extracting PVFH features of a part of the point clouds to establish a model library, segmenting a point cloud object from a scene in an online identification stage, then extracting PVFH features of the identified object, carrying out feature matching with a model library to obtain a roughly matched pose, and finally carrying out optimization by using an ICP algorithm to obtain an accurate pose. According to the method, the defect that a traditional global feature descriptor cannot identify a rotationally symmetrical or mirrored symmetrical object is overcome, and the robustness is high.

Description

technical field [0001] The invention belongs to the field of three-dimensional point cloud object recognition, in particular to a three-dimensional object pose estimation method based on global feature matching, which can be applied to a Bin-picking system. Background technique [0002] In the Bin-picking system, the pose recognition method of the object is required to have the characteristics of high recognition accuracy and fast recognition speed, so as to meet the working requirements of the fast and real-time grasping of the robot arm. Although in the field of 3D point cloud object recognition, there are many methods for object pose recognition (based on local feature matching, based on template matching), but the recognition accuracy and recognition speed of most of these methods cannot meet the fast real-time grasping of the robot arm. demand. [0003] The global feature is a descriptor that characterizes the overall characteristics of a point cloud object, which can ...

AI Technical Summary

Problems solved by technology

Although these descriptors are used for 3D point cloud object recognition, although the recognition speed is fast, there are the following defects: cannot recognize rotationally symmetrical or mirror symmetrical objects

Therefore, the existing object recognition method based on global feature matching has low recognition accuracy and low robustness for such objects, and it is difficult to be widely used in Bin-picking systems.

Images