This paper addresses the need for intelligent transportation systems (ITS) by proposing an innovative architecture for vehicle trajectory prediction. It integrates a Long Short-Term Memory (LSTM) module with a generative AI component, specifically the RoBERTa Transformer model. The LSTM network with a recursive decoder outperforms a teacher forcing decoder on clean datasets, demonstrating higher robustness in time-series predictions. When video data is partially missing, performance drops, but RoBERTa-based data reconstruction significantly improves results (from 37% to 92%). These findings emphasize the importance of data reconstruction in handling incomplete information in real-world ITS scenarios.
Oct 11, 2024
This paper explores urban changes necessitating digital transformation, proposing an Edge Intelligence (EI)-based Traffic Monitoring System (TMS) for smart cities. It advocates for placing intelligence near data sources, showcasing early benefits like enhanced performance and reduced resource usage compared to cloud-centric methods.
Oct 1, 2023