This study explores the importance of traffic incident detection in intelligent transportation systems and introduces a hybrid model combining GANs and Transformer models. The proposed model addresses challenges related to data imbalance and scarcity, demonstrating improved detection accuracy and reduced false positive rates across various datasets. The study highlights the significance of balanced datasets for effective traffic incident detection.
The content delves into the historical background of traffic incident detection algorithms, from traditional methods like the California algorithm to modern deep learning approaches such as CNNs, GANs, and Transformers. It emphasizes the importance of precise incident detection, reduction of false alarms, and real-time detection for efficient urban traffic management.
Furthermore, the study details experiments conducted on four real datasets - PeMS, I-880, Whitemud Drive, and NGSIM - evaluating different models' performances in detecting traffic incidents. The results showcase that the proposed hybrid model outperforms baseline models by effectively addressing data imbalance issues through GANs while utilizing Transformers for feature extraction.
The study concludes by discussing future research directions to enhance model generalizability by including additional variables like holidays and adverse weather conditions. It also suggests applying the model to diverse urban systems for broader insights beyond highway datasets.
翻譯成其他語言
從原文內容
arxiv.org
深入探究