Design and implementation of virtual scene dynamic generation system based on multi-modal AIGC
Data and Metadata · 13.04.2026 · 5 · 811
DOI: 10.56294/dm2026811 ↗Kurz gefasst
Fachartikel in der Zeitschrift Data and Metadata, online erschienen am 13.04.2026. Verfasst von Feng Wang. Die Arbeit nennt 16 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Introduction
A key element of Digital Twins (DT) is real-time interactive 4D traffic scene production in Intelligent Transportation Systems (ITS); however, existing generative approaches often exhibit limited temporal coherence, weak interactivity, and insufficient adaptability to active traffic environments.
Objective
This research suggests a cloud-edge-terminal collaboration framework for creating visual traffic scenes in real time using Artificial Intelligence Generated Content (AIGC) in vehicular networks to overcome these constraints.
Method
A Big Bang-Big Crunch-driven Wasserstein Generative Adversarial Network (BBBC-Wasserstein GAN) is developed to support temporally consistent and controllable 4D scene synthesis. The BBBC optimization is employed to adaptively tune GAN parameters for faster convergence and improved stability, while the Wasserstein GAN ensures, enhanced training robustness and temporal coherence by minimizing distributional divergence in complex 4D traffic scenes. The proposed framework integrates Multimodal Traffic Scene Interaction Dataset with 2000 data using sensors and images through modality-specific processing pipelines. Sensor data are denoised and temporally aligned using Adaptive Kalman Filtering (AKF), followed by multi-resolution feature extraction using Discrete Wavelet Transform (DWT). Image data undergo Histogram Equalization (HE) and are processed using Convolutional Neural Networks (CNNs) to separate semantic and structural features related to vehicles, pedestrians, road infrastructure and environmental context. The proposed framework integrates multimodal features into a single representation, facilitating realistic and interactive scene generation with real-time updates. It utilizes microservice-based on fine-tuning for adaptable model evolution across various traffic scenarios.
Result
Experimental results in Python demonstrate a significant accuracy improvement (90,34 %) over traditional centralized methods.
Conclusions
Enabling feasible real-time 4D traffic scene production in vehicle networks.
Zitieren
APA
Wang, F. (2026). Design and implementation of virtual scene dynamic generation system based on multi-modal AIGC. Data and Metadata, 5, 811. https://doi.org/10.56294/dm2026811
Vancouver
Wang F. Design and implementation of virtual scene dynamic generation system based on multi-modal AIGC. Data and Metadata. 2026;5:811. doi:10.56294/dm2026811
BibTeX
@article{Wang2026,
author = {Wang, Feng},
title = {Design and implementation of virtual scene dynamic generation system based on multi-modal AIGC},
journal = {Data and Metadata},
year = {2026},
volume = {5},
pages = {811},
doi = {10.56294/dm2026811}
}
EUVIMED