Closed-loop detection method for indoor scene recognition

Abstract

The invention relates to a closed-loop detection method for indoor scene recognition. The closed-loop detection method comprises the following steps that: a current scene image is collected, the FAST algorithm is utilized to extract the feature points of the current scene image, and the LSD algorithm is utilized to extract the feature lines of the current scene image; the ORB is adopted to generate the descriptor vectors of the feature points, and the BRLD is adopted to generate the descriptor vectors of the feature lines; the K-means clustering algorithm is adopted to cluster all the descriptor vectors to generate a visual vocabulary, and the BOW vector of the current scene image is generated by using the visual vocabulary; and the similarity of the BOW vector of the current scene image and the BOW vector of a stored historical scene image is calculated, the consistency of the current scene image is detected, so that whether a closed loop occurs on the current scene image is judged. According to the closed-loop detection method of the invention, the feature lines are additionally adopted in closed-loop detection, sites which have been visited are identified in an indoor environment where feature points are scarce; and since most of feature lines in the indoor environment scene are static, and after the feature lines are introduced, influence on the closed-loop detection algorithm by the change of dynamic objects in the scene is small, and the problem of scene confusion can be better solved.

Description

technical field [0001] The invention relates to the technical field of mobile robot navigation, in particular to a closed-loop detection method for indoor scene recognition. Background technique [0002] Closed-loop detection is a very important module in the SLAM system. The purpose is to allow the robot to identify the places it has visited, correct the cumulative error of SLAM, and ensure the consistency of the map. Loop closure detection is actually an image recognition problem. The mainstream algorithm is generally based on the BoW (Bag of Words) framework, using different types of feature points as a solution. [0003] Most of the current closed-loop detection algorithms are optimized for outdoor scenes, and there are relatively few studies on indoor environments. Indoor scenes are quite different from outdoor scenes in both the number of features and the type of features. It is very obvious that outdoor scenes generally have rich feature points, while the number of f...

AI Technical Summary

Problems solved by technology

[0003] Most of the current closed-loop detection algorithms are optimized for outdoor scenes, and there are relatively few studies on indoor environments.

Images