Robot visual feature matching and positioning method and device based on planar motion

Abstract

The invention discloses a robot visual feature matching and positioning method and device based on planar motion, and the robustness of feature matching and pose estimation is improved through the planar motion characteristics of a wheeled mobile robot, so that the end-to-end system performance of visual positioning is improved. For feature matching, local attributes of image appearance are analyzed, and an enhancement strategy is provided by combining planar motion characteristics to suppress changes caused by view angle changes. In combination with learning-based feature extraction, the proposed technology can increase a considerable number of feature matches and obtain a feature matching set conforming to affine invariance. For pose estimation, a two-point-based minimum model is provided, and map feature points with depth or feature points without depth are utilized, so that dense feature matching is fully utilized. By embedding the model, the RANSAC can identify more outliers from a feature matching set with a relatively high outlier proportion, so that visual positioning with relatively high robustness is realized.

Description

technical field [0001] The invention belongs to the technical field of robot positioning, and in particular relates to a robot visual feature matching and positioning method and device based on planar motion. Background technique [0002] With the development of mobile robot technology, mobile robots perform more and more types of tasks, and the connection with human life is getting closer. At present, the application scope of mobile robots includes logistics handling, express delivery, cleaning, inspection and unmanned driving, etc., which has greatly promoted the transformation and unmanned process of these industries. Achieving high-precision and long-term stable self-localization is a prerequisite for autonomous mobile robots to complete tasks. [0003] Although the visual positioning technology is less mature than the distance sensor, the camera is an indispensable sensor unit to provide intuitive environmental information for human-computer interaction; the cost of th...

AI Technical Summary

Problems solved by technology

[0007] However, due to the large number of non-textured scenes in the indoor environment, such as walls, windows, ground, etc.; narrow visual field of view, poor robustness to problems such as occlusion of dynamic objects and changes in the placement of objects; changes in the appearance of the environment caused by complex indoor lighting conditions will directly affect the visual information

The characteristics of the above indoor environment lead to errors in the depth estimation of 3D feature points in the process of visual mapping, especially in non-textured scenes, where the error is larger, which directly affects the accuracy of pose estimation.

[0008] In summary, the current visual positioning technology still has some key problems to be solved, and the technology maturity has not yet reached the stability of laser positioning technology, and cannot reliably meet the needs of applications.

The core of these key problems can be summarized as insufficient robustness due to insufficient field of view and sensitivity to environmental changes; especially the problem of vision-based global positioning in changing environments is far from being solved, and the core of the problem is how to obtain reliable feature matching

In the long-term seasonal change scenario, the existing mature methods will give a large number of false matches, mainly because the existing visual feature extraction and description methods have poor repeatability and invariance, and the feature matching method lacks robustness

Images