Small robot indoor passable area obtaining method and device

Abstract

The invention discloses a small robot indoor passable area modeling method, and the method comprises the following steps: (10) monocular SLAM: obtaining an indoor environment image through a monocularcamera on the robot, inputting the indoor environment image into a monocular SLAM system, and obtaining the pose and sparse feature point cloud of the camera; (20) image segmentation: segmenting theindoor environment image to obtain a ground area; (30) ground plane fitting: extracting points in the ground area, performing filtering and fitting to obtain a ground plane; (40) segmented image screening: screening the segmented ground area by using the ground plane position to obtain a ground area meeting the requirement; and (50) dense ground modeling: projecting the ground segmentation image to the ground plane to obtain a dense ground point cloud model. The method and the device are low in cost, good in real-time performance and good in modeling effect.

Description

technical field [0001] The invention belongs to the technical field of scene modeling, and is a method and device for modeling a small robot indoor passable area with low cost, good real-time performance and good modeling effect. Background technique [0002] In recent years, with the advancement of computer and sensor technology, robotics has also developed vigorously. Vision-based simultaneous localization and mapping (SLAM) is one of the important technologies in the field of robotics. SLAM can use the input continuous images to solve the motion trajectory of the camera in real time, and build a map of the scene to guide the robot's progress. [0003] Indoor environments are currently the main application areas for small robots. In application scenarios including warehousing logistics, factory material handling, and home services, small robots have great application potential. For small robots whose main working environment is indoors, the monocular camera vision senso...

AI Technical Summary

Problems solved by technology

[0004] However, due to the small size, low viewing angle, weak computing power, and limited field of view of indoor small robots, it is difficult to apply visual SLAM methods to understand the scene in real time

The SLAM method that can build dense maps has high requirements for computing resources and is not suitable for small robots with weak computing power.

Moreover, since the main part of its field of view is the ground and there are few texture features, the SLAM method based on features and gradients can only obtain extremely sparse maps, and its ability to express the scene is poor, so it cannot guide the robot's actions.

[0005] Generally speaking, when the existing mainstream monocular SLAM system runs on a small indoor robot, due to the characteristics of the platform itself, or the map construction is too sparse; or even though the map is dense, it cannot realize real-time operation.

Generally does not achieve satisfactory results in terms of building scenes

However, in the existing low-cost modeling methods, there are not many studies on the characteristics of small ground robots, and there is no suitable low-cost modeling method for ground areas.

Images