Monocular vision-assisted laser radar odometer method based on ground plane

Abstract

The invention relates to a monocular vision-assisted laser radar odometer method based on a ground plane. The method comprises the steps of efficiently extracting ground feature points in an image through the ground point cloud in a laser odometer, achieving efficient absolute-scale camera motion estimation based on homography transformation, and then using motion estimation for correcting self-motion point cloud distortion and pose optimization in a laser odometer. Compared with the prior art, the tight coupling scheme provided by the invention efficiently utilizes the monocular vision imageand the ground plane information in the laser radar point cloud, and avoids the problems that the existing vision and laser radar fusion algorithm is high in calculation complexity, and the system precision and stability are influenced by wrong depth matching.

Description

technical field [0001] The invention relates to the technical field of robot identification, in particular to a ground plane-based monocular vision-assisted laser radar odometer method. Background technique [0002] Simultaneous localization and mapping (SLAM, simultaneous localization and mapping) is a key technology in the field of robotics and a basic requirement for robots to operate autonomously. Lidar and vision sensors are two mainstream sensors in SLAM. In recent years, SLAM algorithms based on these two sensors have been widely researched and applied. In terms of visual SLAM, excellent solutions represented by ORB-SLAM2, DSO, VINS, etc. have emerged. In laser SLAM, frameworks such as LOAM, IMLS-SLAM, and SegMatch have been formed. However, a single sensor always has certain deficiencies, such as vision sensors are sensitive to environmental textures and lighting conditions, while lidar has point cloud distortion problems caused by body motion and insufficient loop...

AI Technical Summary

Problems solved by technology

However, simultaneous odometry and mapping based on two sensors requires a lot of system overhead, and is not suitable for low-computing occasions such as embedded platforms

[0005] Therefore, it is necessary to study how to allocate a small amount of system resources to use visual sensors to assist lidar in simultaneous positioning and mapping, to make up for the shortcomings of lidar in terms of self-motion distortion and loop closure detection, and to avoid the high computational complexity and poor system stability of existing methods. issues are of great importance

Images