Mobile robot loopback detection method based on deep learning

Abstract

The invention discloses a mobile robot loopback detection method based on deep learning. The method comprises the following steps: forming an airport terminal actual scene data set and an airport terminal actual scene enhanced data set; obtaining a region generation model, a global description model and a loopback detection model; obtaining a loopback candidate frame; and performing loopback verification to obtain a final loopback frame. Compared with the prior art, the deep learning technology is introduced to learn the global feature descriptor of the local area of the image frame, and compared with an artificially designed descriptor, the method can learn deeper semantic information, can better adapt to scene changes, and combines the advantages of the global descriptor and the local descriptor, and improves the image quality. On the basis that the global descriptor has good appearance invariance, the robustness of the descriptor to viewpoint change is enhanced.

Description

technical field [0001] The invention belongs to the technical field of Simultaneous Localization and Mapping (SLAM), in particular to a method for loop detection of a mobile robot based on deep learning. Background technique [0002] Visual SLAM is a synchronous positioning and mapping technology based on visual sensors. In an unknown environment, it solves the pose and three-dimensional space map between camera frames by tracking the extracted image features. It has been widely used in robots, unmanned machines and self-driving car platforms. [0003] As an important part of visual SLAM, loop detection is to identify the places that have been reached through image data, which is an important part of visual SLAM technology. In the long-term working process, the visual SLAM system will inevitably have cumulative errors. Global optimization after correctly identifying loop closures can eliminate accumulated errors. A true positive loop closure (a true loop predicted to be a...

AI Technical Summary

Problems solved by technology

[0004] The early bag-of-words-based loopback detection method can still run normally in a stable environment, but its recall rate and accuracy rate will decline in the terminal building scene affected by factors such as dynamic targets, illumination changes, and perspective changes. , so this has become an important problem that many researchers are vying to solve

Images