Robot RGB-D SLAM method based on geometric and motion constraints in dynamic environment

Abstract

The invention provides a robot RGB-D SLAM method based on geometric and motion constraints in a dynamic environment. The robot RGB-D SLAM method comprises the steps that image data are acquired through an RGB-D camera, wherein the image data comprise an RGB image and a depth image; the coordinates of feature points in a camera coordinate system are acquired through an ORB feature point detection algorithm, the RGB image and the depth image; the pose of a camera is calculated by using a g2o optimization algorithm; the feature points in the camera coordinate system are converted into a world coordinate system, and the coordinates of the same point in the world coordinate system are continuously tracked to obtain the multiple observation speeds of the point; whether the point is on a real moving object or a static object is known by analyzing the law of multiple observation speeds; after analyzing that the point is a real motion, the motion speed of the moving point and the coordinates under the world coordinate system are filtered through a Kalman filtering algorithm; and finally the moving point and the static point are added into g2o pose optimization. The method is effectively suitable for a dynamic environment, and a more accurate camera pose is obtained.

Description

technical field [0001] The invention relates to a robot positioning method in an indoor dynamic environment. Background technique [0002] In the related research on intelligent navigation technology of autonomous mobile robots, the simultaneous localization and map construction (SLAM, SLAM, for short) technology of robots in unknown environments is a key technology, which has both application and academic value. It has become a research hotspot in this field in the past two decades. Under this trend, scholars have proposed a variety of methods to solve the SLAM problem, and also applied a variety of sensors to solve the environmental perception problem in SLAM. [0003] As an inexpensive sensor with a large amount of information, the camera has excellent application prospects in the field of autonomous mobile robots. Visual SLAM mainly solves the pose transformation of the camera by matching the same feature points in multiple frames of images. However, there are still f...

AI Technical Summary

Problems solved by technology

For the visual SLAM system of sparse maps, the challenge of dynamic scenes mainly comes from the inability of traditional visual SLAM to distinguish motion feature points from static feature points. When calculating the camera pose, the motion feature points are mistakenly calculated as static feature points, resulting in There is a huge deviation between the calculated camera pose and the actual pose, which then leads to incorrect calculation of map points, further expanding the positioning deviation of visual SLAM

Images