Home // VEHICULAR 2023, The Twelfth International Conference on Advances in Vehicular Systems, Technologies and Applications // View article

Federated Monocular 3D Object Detection for Autonomous Driving

Authors:
Fangyuan Chi
Yixiao Wang
Panos Nasiopoulos
Victor C.M. Leung
Mahsa T. Pourazad

Keywords: monocular, 3D object detection, federated learning, autonomous driving.

Abstract:
In this paper, we propose and implement a novel method for 3D object detection in autonomous driving by applying federated mechanism to a monocular camera-based network. Our approach has several advantages over traditional 3D object detection methods that rely on lidar or other sensors, as it is more cost-effective and can be more easily integrated into existing autonomous driving systems. We use a federated learning framework, which allows us to train the model on a large amount of data covering a variety of scenarios without having to share the raw data with a central server. This allows us to reduce transmission bandwidth requirements and preserve the privacy of the data contributors, while still achieving high accuracy in 3D object detection. In our experiments, we evaluate our method on a variety of challenging real-world driving scenarios and show that it is able to accurately detect objects in 3D from a monocular camera view. Our results demonstrate the effectiveness of our approach and show its potential for use in autonomous driving systems.

Pages: 20 to 23

Copyright: Copyright (c) IARIA, 2023

Publication date: March 13, 2023

Published in: conference

ISSN: 2327-2058

ISBN: 978-1-68558-061-2

Location: Barcelona, Spain

Dates: from March 13, 2023 to March 17, 2023