Скачать или смотреть DARTS Drone Based AI Powered Real Time Traffic Incident Detection System

Thursday, October 31, 2024 | 12:00 – 1:00 PM (ET)

Swift traffic incident detection is crucial for saving lives and property and for quickly responding to the incidents and restoring normal traffic conditions. This study addresses the gaps in traffic metrics-based detection methods, which often fail to promptly identify incidents, and CCTV video-based methods, which do not capture the impact range of incidents on upstream and downstream traffic. By utilizing drone thermal camera video and developing machine learning (ML) algorithms, this study innovatively segments live traffic monitoring videos into short clips using a sliding window approach, then vehicle trajectory images are generated from these clips, and a self-designed and trained ML algorithm determines whether these images indicate a traffic incident or non-recurrent congestion caused by an incident. This allows for the efficient detection of traffic incidents while accurately assessing their impact on upstream and downstream traffic conditions. To implement this research in practice, user-friendly Android and web-based interfaces were developed for real-time drone data transmission and visualization of detection results. Field tests confirmed that the system designed in this study can detect incidents 7 minutes earlier than traditional methods. Additionally, the system detected the upstream propagation of congestion caused by the incident. This not only validates the algorithm’s ability to promptly and accurately identify traffic incidents and assess their impact on traffic flow but also demonstrates the usability of the developed graphical user interface. In summary, the drone-based AI-powered real-time traffic incident detection system developed in this study provides an innovative approach to the efficient detection and management of traffic incidents.