Video Dataset for Predicting Crowd Congestion at Large Event Dismissals

In modern cities, large gatherings and events are frequently held, often leading to crowd congestion during dismissals, presenting various safety risks. Current crowd management solutions mostly rely on experience and simple sensor devices, making it difficult to predict and manage congestion in real-time and accurately. This dataset aims to provide high-quality video data to aid in the development of more precise crowd prediction and management systems. The data is collected using high-definition cameras at iconic event locations in multiple cities, with equipment installed at key positions such as entrances and main corridors of event venues. The video data has undergone multiple rounds of annotation, including reviews and consistency checks by numerous experts. The annotation team consists of professional video analysts, with a team of over 30 people. Data processing includes steps such as video compression and frame rate adjustment to ensure efficient storage and quick retrieval. The final data is organized in MP4 format, accompanied by detailed annotation information stored in JSON format.