Robot connector six-degree-of-freedom pose estimation system based on deep learning

Abstract

The invention relates to the technical field of deep learning and visual robots, in particular to a robot connector six-degree-of-freedom pose estimation system based on deep learning. The system comprises the following steps: making a data set by utilizing a virtual reality technology; performing 6D pose estimation on the connector by improving a DOPE algorithm; and building a robot connector pose estimation grabbing platform. According to the robot connector six-degree-of-freedom pose estimation system based on deep learning, the data set is improved by using the virtual reality technology, so that the background information of the data set is diversified, and the precision is not affected even in a new environment; a specific data set is made for a specific industrial scene, and the virtual reality technology is used for making the data set, so that the workload of manually marking objects is greatly alleviated; and meanwhile, in a monocular vision robot connector sorting scene, a random mask local processing method is used for improving the data set for the shielding problem, and thus the accuracy of the network during object shielding processing is improved.

Description

technical field [0001] The invention relates to the technical field of deep learning and visual robots, in particular to a six-degree-of-freedom pose estimation system for robot connectors based on deep learning. Background technique [0002] Facing the challenge of manufacturing upgrading in recent years, my country has successively proposed a series of smart manufacturing strategies. According to the Industry 4.0 white paper issued by China, it is proposed to focus on intelligent manufacturing as the main direction to create an internationally competitive Chinese manufacturing industry. In the field of intelligent manufacturing, machine vision technology, artificial intelligence technology, virtual reality technology, and robot technology are undoubtedly the key content and core content of intelligent manufacturing, and the automatic recognition and grasping of robots and flexible assembly technology that integrate these technologies have become A hot research direction i...

AI Technical Summary

Problems solved by technology

[0006] However, in actual engineering, the original DOPE algorithm has the problem that the recognition speed is too slow, resulting in the estimation of the object pose slower than the feeding speed of the feeding car or the conveyor belt.

At the same time, in an industrial environment, the parts in the collected images may have large scale changes, thereby reducing the accuracy of pose estimation

In addition, the original DOPE network only uses RGB images when estimating the object pose, and it is difficult to accurately identify occluded parts, resulting in low accuracy and robustness of the robot grasping system

[0007] At present, there are still the following technical problems in the field: 1) One of the difficulties that restricts the practical application of 6D pose estimation is that it is very difficult to make data sets by hand, and most of the existing 6D pose estimation methods use LINEMOD or YCB- Datasets such as Video, but due to the particularity of robot connectors, the network with good test results on these data sets is not suitable for some low-textured artifacts in practical applications

Images