Title Comparative evaluation of deep learning architectures for environment and obstacle recognition in robotic lawn care /
Authors Siaulys, Julius ; Paulauskaite-Taraseviciene, Agne
Full Text Download
Is Part of CEUR workshop proceedings: IVUS 2023: Information society and university studies 2023: proceedings of the 28th international conference on information society and university studies (IVUS 2023) Kaunas, Lithuania, May 12, 2023 / edited by: A. Lopata, T. Krilavičius, I. Veitaitė, A. García-Holgado.. Aachen : CEUR-WS. 2023, vol. 3575, p. 55-63.. ISSN 1613-0073
Keywords [eng] robot mowers ; deep learning ; accuracy metrics ; image recognition ; semantic segmentation ; optimization
Abstract [eng] Robotic lawn mowers typically rely on boundary wires, which are installed around the perimeter of the lawn to define the mowing area. While boundary wires have been a reliable technology for robotic lawn mowers, there are certainly limitations and inefficiencies associated with them. AI technologies have the potential to improve the performance and capabilities of robotic lawn mowers, and reduce the reliance on boundary wires. By using computer vision technologies to recognize obstacles and environments in robotic lawn mower fields, it could potentially eliminate the need for boundary wires and reduce the risk of damage to objects on the field and the health risks to small animals. This study aims to evaluate different deep learning architectures for recognizing obstacles and environments in robotic lawn mower fields, using the Ade20k dataset. The differently merged datasets were evaluated on the full dataset of 150 classes, 21 classes and 3 merged subsets of classes. The results of the comparison of seven different deep learning models showed that merging classes is practical for this task and to improves semantic segmentation accuracy results by up to 1.4 times.
Published Aachen : CEUR-WS
Type Conference paper
Language English
Publication date 2023
CC license CC license description