Closed-loop detection system and method for multi-scale feature fusion

Abstract

The invention discloses a closed-loop detection system and method for multi-scale feature fusion, and the system comprises a basic feature extraction module and a multi-scale feature extraction-fusion module which are used for processing a similarity to-be-matched image composed of a current collection image and any image collected historically. And the multi-dimensional similarity measurement module is used for acquiring the similarity score of the image to be matched. According to the method, a unified end-to-end network system is constructed in consideration of different receptive field information of images, different dimension information of features and the like, a closed-loop detection result can be directly output, meanwhile, high accuracy and relatively high recall rate are achieved, the estimated pose of the robot in the SLAM system can be effectively corrected, and the detection accuracy of the robot in the SLAM system is improved. According to the method and the system, the robot is assisted in repositioning when the pose is lost, the track accumulative error of the robot along with time increase is corrected, and a global consistency map is constructed, so that the robot obtains more accurate positioning and a better mapping effect.

Description

technical field [0001] The invention belongs to the technical field of robot navigation, and relates to closed-loop detection in robot navigation, in particular to a closed-loop detection technology based on multi-scale feature fusion. Background technique [0002] Loop closure detection has become a key issue and research hotspot in the field of mobile robot navigation, especially in simultaneous localization and mapping (SLAM), because it can reduce the cumulative error of robot pose estimation and construct a globally consistent map, which is crucial for autonomous Robot positioning, mapping, navigation and obstacle avoidance in large scenes are particularly important. Proper loop closure detection can add edge constraints in the pose map to help further optimize robot motion estimation and build consistency maps. Wrong loop closure detection will lead to failure of map construction. Therefore, a good loop closure detection algorithm is very important for the consistenc...

AI Technical Summary

Problems solved by technology

[0010] Aiming at the problems of poor robustness and poor mobility in the current closed-loop detection method suitable for robot navigation, a closed-loop detection method based on multi-scale feature fusion is provided. Using multi-scale features and a single feature similarity measure, etc., it can extract The high-level, abstract and global features contained in it are measured by multi-dimensional feature similarity, aiming to improve the accuracy and robustness of visual loop closure detection

Images