Home // AIMEDIA 2025, The First International Conference on AI-based Media Innovation // View article

A Novel Synthetic Dataset for Broadcast Motorsports Scene Understanding

Authors:
Luca Francesco Rossi
Andrea Sanna
Federico Manuri
Mattia Donna Bianco

Keywords: computer vision; augmented reality; synthetic data generation; transfer learning

Abstract:
The paper introduces a foundational approach to motorsports scene understanding by investigating the role of synthetic data generation in advancing scene understanding for high-speed broadcast scenarios. Utilizing the CARLA (Car Learning to Act) simulation environment, the study constructs a high fidelity dataset incorporating diverse lighting conditions, occlusions, and dynamic camera perspectives to enhance model generalization. A multistage data refinement pipeline is introduced to mitigate the impact of extreme occlusions and irrelevant samples while preserving the complexity of real-world challenges. Possible applications include 3D real-world understanding from a single monocular 2D image, which could open up interesting possibilities for augmented reality in broadcast media by allowing seamless integration of virtual elements, interactive graphics and dynamic visual effects, enhancing storytelling, audience engagement, and production flexibility. The efficacy of the dataset is further evaluated via transfer learning to the real-world domain, with the model pretrained on synthetic data demonstrating a significantly superior performance compared to its counterpart.

Pages: 48 to 56

Copyright: Copyright (c) IARIA, 2025

Publication date: July 6, 2025

Published in: conference

ISBN: 978-1-68558-330-9

Location: Venice, Italy

Dates: from July 6, 2025 to July 10, 2025