SLAM loopback detection improvement method based on semantic segmentation

Abstract

The invention discloses a SLAM loopback detection improvement method based on semantic segmentation. The method comprises the following steps of 1, obtaining a general data set of road scene semanticsegmentation; 2, acquiring an environment image by using an RGB camera, and acquiring a road environment image under different environment conditions; 3, obtaining a classification result of each pixel predicted by semantic segmentation; 4, according to a classification result, obtaining an object feature vector containing a corresponding category in the semantic segmentation graph; 5, constructing a place model containing semantic information; 6, analyzing the images in the general data set through a place model to obtain a semantic place vector, creating an image library of places, and storing the semantic place vector into a vector containing place semantic information; 7, matching the similarity between the images to be detected and the images in the image library; and 8, carrying outclosed-loop detection, if a vector of which the similarity exceeds a threshold value is found, representing that the current image is a closed-loop node, and otherwise, adding the current image into an existing image library.

Description

technical field [0001] The invention relates to the technical fields of semantic segmentation and SLAM, in particular to an improved method for SLAM loop closure detection based on deep learning. Background technique [0002] Simultaneous localization and mapping (SLAM) refers to the process in which a vehicle relies on its own sensors to estimate its own position and build a map in an unfamiliar environment without human control. It is a prerequisite for many robot application scenarios such as environmental perception, obstacle avoidance navigation, etc. It is divided into laser SLAM and visual SLAM according to the sensors used. Because the cost of visual SLAM is low, and the picture carries rich texture information, it is widely used. By performing deep learning calculations on pictures, semantic information in the environment can be obtained. [0003] SLAM is divided into sensor acquisition data, visual odometer analysis and calculation, and back-end optimization whil...

AI Technical Summary

Problems solved by technology

The location description vector constructed by clustering with the k-nearest neighbor algorithm is an abstract representation of image information, which reflects the global information of the image to a certain extent, but this method cannot express the category information of the cluster center, and cannot distinguish whether the cluster points are is a reference

In the actual environment, there may be moving objects such as pedestrians, moving vehicles, etc., and using them as clustering points will obviously deviate the description of the scene

Images