Binocular vision positioning method and binocular vision positioning device for robots, and storage medium

Abstract

The invention discloses a binocular vision positioning method and a binocular vision positioning device for robots, and a storage medium. The binocular vision positioning method includes acquiring current binocular images and current positions and posture of the robots; acquiring historical key images from key libraries if the current binocular images are key images; splicing the current binocularimages and the historical key images according to the current positions and posture and historical key positions and posture to obtain vision point cloud maps; acquiring preliminarily built laser point cloud maps; optimizing the current positions and posture according to the vision point cloud maps and the laser point cloud maps. The historical key images acquired from the key libraries and the current binocular images have overlapped visual fields, and the historical key images are related to the historical key positions and posture. The binocular vision positioning method, the binocular vision positioning device and the storage medium have the advantages that position and posture estimated values are optimized by the aid of the preliminarily built laser point cloud maps at moments corresponding to key frames, and accordingly position and posture estimation cumulative errors can be continuously corrected in long-term robot operation procedures; information of the accurate laser pointcloud maps is imported in optimization procedures, and accordingly the binocular vision positioning method and the binocular vision positioning device are high in positioning accuracy.

Description

technical field [0001] The invention relates to robot navigation technology, in particular to a robot binocular vision positioning method, device and storage medium. Background technique [0002] At present, more and more different types of robots appear in all aspects of production and life. For warehousing logistics, inspection and monitoring and other fields, the work requires robots to be able to achieve long-term stable operation in a relatively fixed environment, and to achieve accurate self-positioning. At this stage, the positioning method based on 3D laser radar and prior laser map can meet this demand, but the cost of 3D laser is relatively high, which is not suitable for the widespread application of robots. The cost of visual sensors is low, and the amount of information obtained is large. If the visual sensor is used to realize this function, the production cost can be reduced to a large extent. However, the existing visual odometry calculation method will mak...

AI Technical Summary

Problems solved by technology

However, the existing visual odometry calculation method will make errors due to accumulated errors during long-term operation, and the visual positioning method is easily affected by ambient light and climate change, making it difficult to achieve a stable association with the visual map or prior image, resulting in positioning failure. , which in turn affects the completion of the robot task

There is also a method of using closed-loop detection to achieve pose optimization. However, when the scope of the working environment is large and there are few scenes where closed-loops can be detected, the robot may already have relatively large gaps before successfully detecting closed-loops for global optimization and correction of accumulated errors. Large pose drift, resulting in failure of closed-loop optimization and inaccurate robot pose estimation

Images