Robot SLAM (simultaneous localization and mapping) method based on semantic segmentation technique

Abstract

The invention discloses a robot SLAM (simultaneous localization and mapping) method based on a semantic segmentation technique. The robot SLAM method comprises the following steps of S1, obtaining theshot image data in the movement process of a robot; S2, performing semantic segmentation treatment on the image data, so as to distinguish the plane area and the non-plane area in the image; detecting feature points in the non-plane area; S3, building an appearance description type of the feature points and plane area, so as to build the matching corresponding relationship of the feature points and the plane area between different image frames; S4, according to the corresponding relationship, building a likelihood function; S5, minimizing the likelihood function, so as to obtain a mixed three-dimensional map and camera movement parameters. The robot SLAM method has the advantages that the plane area and non-plane area in the image are distinguished by the semantic segmentation technique,and the feature points of the non-plane area are detected; when the feature points are difficult to extract in the image or the number of feature points is fewer, the stability in localization and scene rebuilding is improved by the matching of the plane area.

Description

technical field [0001] The invention relates to the technical field of robot navigation, in particular to a robot SLAM method based on semantic segmentation technology. Background technique [0002] For the problem of robot visual navigation, the main research is to restore the three-dimensional structure of the scene from multiple image frames (two or more images) and the corresponding position and attitude of the camera that captures each image frame. In the field of robot navigation, it is generally called Simultaneous Localization and Mapping (SLAM). Generally, the three-dimensional structure of the scene is expressed in a disordered point cloud. [0003] Early SLAM methods were based on matrix factorization techniques. This type of method first forms a matrix of image coordinates of feature points obtained by observation matching, and obtains the three-dimensional coordinates of feature points and the pose parameters of the camera through SVD decomposition technology. ...

AI Technical Summary

Problems solved by technology

Since the number of unknown variables to be optimized is proportional to the number of 3D scene points and image frames, when the image sequence scale is large, the complexity of the bundle adjustment algorithm is very high, which limits the application of this type of method in a large-scale environment

[0004] The existing SLAM algorithm is mainly realized by analyzing the feature points in the image frame, resulting in the final output result being a discrete three-dimensional point cloud. In other fields, it is necessary to know the structure and semantic information of the scene, and it is not enough to provide only 3D point cloud, thus limiting the scope of application of SLAM

In addition, the method based on feature points generally requires relatively rich texture in the image. For smooth and monotonous scenes, the SLAM method based on feature points is helpless.

Images