Laser SLAM method integrating ORB closed-loop detection

Abstract

The invention discloses a laser SLAM method integrating ORB closed-loop detection. The laser SLAM method comprises a laser SLAM process and a visual closed-loop detection process which operate in parallel. In the laser SLAM process, when the change rate of the pose information of the previous key time point and the current time point is greater than a change rate threshold, a visual closed-loop detection process is triggered once; in the visual closed-loop detection process, an image key frame is extracted from the environment image, and ORB features of the current image key frame are extracted to obtain a word vector of the current image key frame; searching is carried out in a historical image key frame set according to the word vector of the current image key frame to detect a visual closed loop; and after the visual closed loop is detected, visual auxiliary laser closed loop detection is carried out in the laser SLAM process. After the visual closed-loop detection method and the laser SLAM are integrated, the laser SLAM can judge whether loopback exists or not by utilizing visual information, so that the mapping accuracy is improved.

Description

technical field [0001] The invention relates to the field of mobile robots, in particular to a laser SLAM method fused with ORB closed-loop detection. Background technique [0002] Simultaneous Localization and Mapping (SLAM) of mobile robots in unknown environments is a hotspot in mobile robot research. With the expansion of the exploration scale of mobile robots, the robustness of simultaneous positioning and mapping of robots in large-scale environments is particularly important, and reliable loop closure detection is one of the most important and critical issues in building robust SLAM. Closed-loop detection is the robot's judgment of whether its current location is in the environment area it has visited, and it is used as the basis for whether the map needs to be updated and corrected, which is of great significance for improving the robustness of SLAM in large-scale environments. The laser sensor is the most widely used sensor in SLAM. In a large-scale environment, th...

AI Technical Summary

Problems solved by technology

[0005] To sum up, the existing SLAM algorithms are mainly based on a single distance sensor: the pure laser SLAM method is prone to error matching in an environment with repeated geometric structures, and the visual SLAM is difficult to guarantee real-time performance. These solutions have no potential for general application

Images