6DOF object attitude estimation method based on deep learning point cloud matching

Abstract

The invention discloses a 6DOF object attitude estimation method based on deep learning point cloud matching, and relates to the field of machine vision. The method comprises the following steps: 1, carrying out the key point detection of a to-be-recognized image according to an object key point detection network model based on deep learning, obtaining pixel coordinates of key points of an objectin the to-be-recognized image, and converting the pixel coordinates into image coordinates for a subsequent PNP algorithm; 2, solving the PNP algorithm according to the 3D coordinates and the image coordinates of the key points to obtain an initial attitude of the object in the to-be-recognized image, and using the initial attitude for an initial value of subsequent ICP algorithm registration; and3, performing object segmentation according to the key points of the object in the to-be-recognized image to obtain a local point cloud containing attitude information of the object, and registeringthe local point cloud with a template point cloud by using the ICP algorithm to obtain a pose of the object in the to-be-recognized image.

Description

technical field [0001] The invention relates to the field of machine vision, in particular to a 6DOF object pose estimation method based on deep learning point cloud matching. Background technique [0002] With the increasing application of machine vision and artificial intelligence, intelligent vision system has become one of the most critical systems in robots. Vision-based robot recognition and grasping research has attracted more and more attention at home and abroad, and in actual production such as feeding automation, welding robots, and logistics robots, intelligent robots must not only perceive the environment, but also interact with the environment. [0003] Object grasping is the most basic and important ability of robots. The main disadvantage of single 3-dimensional (3D) vision is that the amount of calculation is large, and the number of point clouds contained in the scene point cloud is huge, which is prone to mismatch; single 2-dimensional (2D) vision lacks d...

AI Technical Summary

Problems solved by technology

[0005] Therefore, those skilled in the art are committed to developing a 6DOF object pose estimation method based on deep learning point cloud matching, which solves the defects of large amount of calculation and lack of a good initial value of the ICP algorithm in the existing method

Images