Mobile robot and method for the mobile robot to recognize an elevator

Elevator Recognition Technology for Mobile Robots Combining Deep Learning Recognition and Point Cloud Hough Transform

This technology identifies elevators by combining point cloud data collected via RGB-D sensors with deep learning-based image recognition models. It then accurately estimates the elevator's position and boundaries by analyzing linear data extracted through Hough transforms alongside point cloud distances.

Existing location recognition methods using laser range sensors or standard cameras often suffer from low accuracy due to reflections from metallic elevator surfaces, and they typically require additional environmental modifications, such as installing artificial markers.

This technology uses a deep learning model to identify elevators and processes RGB-D sensor point cloud data through noise reduction and Hough transforms to distinguish walls from elevators, subsequently calculating boundaries and positions through geometric analysis. It can be applied to indoor delivery and disinfection robots, enabling autonomous inter-floor movement without the need for separate markers.

‍

‍

Key Features:

• A movement phase where the mobile robot travels to a recognition point near the elevator to detect it.
• A phase where an estimated elevator image is acquired at the recognition point.
• A recognition phase where a model trained on elevator and non-elevator images determines the presence of an elevator.
• A phase where the elevator's position and boundaries are estimated using point cloud data and Hough transforms.

‍

This invention was developed with support from the Ministry of Science and ICT for the development of robotic hand manipulation intelligence, which learns methods and procedures for handling various objects using tactile-capable robotic hands.