Three-dimensional matching method for self-reconfiguration butt joint of reconfigurable spherical robot

Abstract

The invention provides a stereo matching method for a spherical robot, and particularly relates to a stereo matching method for self-reconstruction docking based on a reconfigurable spherical robot, which comprises the following steps of: using an aggregation mechanism of recognition features proposed based on an attention mechanism; and a graph neural network capable of carrying out feature matching and sensing a 3D scene in real time is created based on an attention mechanism, feature matching processing is carried out on the spherical robot, an inter-frame pose is solved through feature matching, and a pose estimation task in the indoor and outdoor environments of the reconfigurable spherical robot is completed. According to the method, the process that two images are browsed back and forth in the human feature matching process to screen matching key points and repeatedly conduct context comparison is simulated, the attention-based graph neural network is applied to the stereo matching process of the reconfigurable spherical robot, the method can operate in real time, and meanwhile classic and deep learning features are achieved; the method greatly improves the feature matchingspeed and precision in the real-time stereo matching of the spherical robot, can achieve the quick matching of a sparse feature image under the large-view-angle transfer, is highly suitable for the indoor and outdoor working conditions, and can provide the high-precision relative pose estimation for the spherical robot.

Description

technical field [0001] The invention relates to the technical field of robots, in particular to a stereo matching method for self-reconfiguration docking of a reconfigurable spherical robot. Background technique [0002] Because of its unique geometric characteristics, spherical robots have the advantages of good balance, anti-overturning, and rolling to resolve impact forces. Therefore, they have significant advantages and development potential in the fields of military reconnaissance, aerospace exploration, and disaster environment detection. a research hotspot. However, the more complex the environment and task conditions, the more demanding the spherical robot to respond quickly and accurately, and to be flexible, and a single spherical robot has environmental adaptability and working ability due to its closedness and lack of suitable sensors and corresponding fusion methods. limited issues. The modular self-reconfigurable robot has a strong adaptability to complex env...

AI Technical Summary

Problems solved by technology

[0003] In the current existing technology, only the spherical robot with the application number 201610842050.3 mentions the reconstruction ability, the technology with the application number CN201921442053.3 mainly revolves around the motion control problem of the spherical robot, and the spherical robot with the application number CN201310067858.5 has stereo vision However, it is only used for visual positioning, and there is no technical reference to the autonomous docking capability of spherical robots and related stereo matching methods to solve docking and other detection problems.

Images