Robot slam method and system for outdoor feature sparse environment

Abstract

The invention provides a visual inertial SLAM method and system for a mobile robot in an outdoor feature sparse environment, including: acquiring an environmental image in an outdoor feature sparse environment, preprocessing the environmental image, and obtaining the preprocessed environmental image ; Extract the sparse features of the pre-processed environment image by the block SIFT feature extraction algorithm; estimate the inter-frame motion of the robot through the pre-integration of the inertial unit IMU, and estimate the inter-frame displacement of the matching points, and the sparseness is determined by the size of the displacement. The importance of the feature is judged, and the visual reprojection error is calculated; the IMU error is accumulated according to the variance in the IMU pre-integration process; the loss function is jointly constructed according to the visual reprojection error and the IMU error; the loss function is minimized by the nonlinear optimization method, Solve the robot's pose transformation and map point space coordinates. The present invention can reduce the influence of difficulties such as drastic changes in illumination and sparse features on the positioning performance, and improve the autonomy of the mobile robot in a complex outdoor environment.

Description

technical field [0001] The invention relates to a real-time positioning and mapping algorithm of a robot, in particular, to a robot SLAM method and system for an outdoor feature sparse environment, and more particularly, to a robot visual inertia for an outdoor feature sparse environment Real-time positioning and mapping algorithm, using image preprocessing, feature detection, and optimizing the weight of visual error loss to improve the robot's positioning and mapping ability in outdoor scenes with sparse features. Background technique [0002] The general SLAM problem is aimed at scenes with relatively closed environments and rich environmental features such as indoors and streets. In outdoor environments with sparse features (such as deserts, beaches, rocky beaches, planets and lunar surfaces, etc.), the problem of real-time robot positioning and mapping is related to the autonomous positioning, safety, and mission planning of robots in such scenarios, especially When th...

AI Technical Summary

Problems solved by technology

[0003] The publication number is CN110309834A (application number: 201910393343.1), which discloses a SLAM algorithm for outdoor offline navigation. A certain number of feature points are obtained through a locally adaptive threshold adjustment algorithm, which overcomes the shadows and weak lighting in the images of wild scenes. The problem of insufficient accuracy of feature point extraction under the conditions of objective factors such as sudden noise and sudden noise

This method does not fundamentally solve the problems of difficult feature extraction and drastic illumination changes in the outdoor environment with sparse features. For the sparse environment, lowering the feature detection threshold will bring feature instability, which will cause greater errors; and use multi-scale detection. The method of the method is relatively vague for the description of visual features, and cannot accurately measure the weight of image features in optimization.

Images