Mobile robot navigation map generation method based on ORB_SLAM2

Abstract

The invention discloses a mobile robot navigation map generation method based on ORB_SLAM2. The method comprises the following steps: step one, carrying out visual information collection; to be specific, enabling a mobile robot to move freely in a three-dimensional environment and collecting image information and depth information through a depth sensor; step two, on the basis of an ORB_SLAM2 algorithm, estimating pose information of the robot, carrying out local BA pose information optimization, screening key frames, selecting one key frame, outputting the pose information of the key frame, providing a sensor reference transformation matrix for creating a navigation map, and updating pose information of each key frame according to the global BA optimization under the condition of detecting a loop; and step three, carrying out map updating; to be specific, generating a three-dimensional map based on a skip list tree according to the pose information and carrying out updating continuously according to pose information of the new key frame. According to the invention, an efficient three-dimensional map is created by utilizing the depth image information; and the map uses the skip list tree structure as a data structure of the three-dimensional map, so that positioning, navigation and obstacle avoidance of the mobile robot can be realized in real time.

Description

technical field [0001] The invention relates to the technical field of visual SLAM and map creation, in particular to an ORB_SLAM2-based mobile robot navigation map generation method. Background technique [0002] Simultaneous Localization and Mapping (Simultaneous Localization and Mapping), referred to as SLAM, was first proposed by SmithSelf and Cheeseman in 1988. It is considered to be the key to realizing a truly fully autonomous mobile robot. The SLAM problem can be described as: the robot in an unknown environment Start moving from an unknown location, position itself according to position estimation and sensor data during the movement, and build an incremental map at the same time. The SLAM that only obtains environmental information through the camera is called visual SLAM. Because the camera has the characteristics of low price, low power consumption, light weight, small size, and rich image information, visual SLAM has been studied by scholars and experts at home a...

AI Technical Summary

Problems solved by technology

The maps created by current visual SLAM for mobile robot path planning and obstacle avoidance are usually 2D grid maps. Although 2D grid maps have a good navigation effect for mobile robots in a flat environment, they are not very good for 3D environments. Ground detection and establishment of maps cannot provide comprehensive environmental information for robots with more than three degrees of freedom such as drones

Therefore, it is of great significance to divide the three-dimensional space into 3D grid maps with voxels as the unit, but the 3D grid maps require a considerable amount of memory for storage. The current method for building 3D grid maps is to use octrees, A three-dimensional space is divided into eight subspaces, and the recursive method is used to continue to divide its subspaces, but the octree map is very unfavorable for query consumption, and it is not convenient for reducing the dimension to a 2D grid map, and it takes a lot of time to navigate the mobile robot. high real-time

Images