Semantic vision SLAM method based on probabilistic grid filtering

Abstract

The invention discloses a semantic vision SLAM method based on probability grid filtering, and the method comprises the steps: sequentially collecting RGB images of a scene through a camera sensor, and carrying out the ORB feature point extraction, super-point segmentation and semantic segmentation on the collected images; creating and initializing a probability grid; calculating matching information of the feature points between an upper frame and a lower frame, and using the matching information to propagate the probability of the grids in the upper frame to the probability grids of the corresponding lower frame to complete probability grid updating; performing motion consistency check on the matching points, and updating the motion state of the probability grid; updating the attribute of the current probability grid by using a Bayesian probability formula according to the updated probability grid, and creating a mask of a dynamic region; according to the extracted ORB feature points, performing filtering by using a mask of a dynamic region, and detecting the dynamic feature points with relatively high probability; and using the reserved feature points for tracking, local mappingand loopback detection, and finally realizing probability grid enhanced semantic vision SLAM.

Description

technical field [0001] The present invention relates to the field of robots, in particular to a semantic visual SLAM method based on probabilistic grid filtering. Background technique [0002] Visual Simultaneous Localization and Mapping (SLAM) is one of the key technologies in the field of robotics. Scenario Static assumptions are typical in SLAM algorithms. Such strong assumptions limit the use of most visual SLAM systems in densely populated real-world environments. Recently, semantic visual SLAM systems for dynamic scenes have gradually attracted more and more attention. Existing semantic visual SLAM systems in dynamic environments usually simply combine semantic information and motion inspection to obtain dynamic target contours, remove all feature points in the dynamic target contour, and only use static feature points to calculate camera poses to improve positioning accuracy . The specific method is: when the motion detection algorithm detects a dynamic feature po...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the problem that the traditional visual SLAM system is susceptible to dynamic target interference in dynamic scenes, the present invention provides a semantic visual SLAM method based on probabilistic grid filtering, which improves the robustness of SLAM in dynamic scenes and improves positioning accuracy

Images