Autonomous Boundary Detection Using Image Recognition for Robotic Lawn Mower

Abstract

Most robotic lawnmowers adopt buried metal wires to define their working area, which is labor and time-consuming. This study applies the fusion of multiple sensors, including an inertial measurement unit (IMU), wheel encoders, light detection and ranging (LiDAR), and an RGB-D camera, to automatically detect the boundary of the lawn and construct the navigation map. Visual simultaneous localization and mapping (vSLAM) are adopted based on an RGB-D camera. For the boundary pavement or obstacles lower or similar to grass height, a support vector machine (SVM) is applied to identify boundaries and obstacles and establish the augmented Octomap with virtual walls. A set of image processing methods is proposed to lower the computing load of the embedded system and transfer the image data into point cloud data to construct the navigation map. The robot operating system (ROS)-based control system of the prototype was built using the embedded system (Jetson Nano, NVIDIA; Santa Clara, CA, USA) and control boards (ARM STM32, STMicroelectronics; Geneva, Switzerland). The proposed method could be further used in the process of map construction and navigation.