To prevent crowd crushes like the Itaewon disaster, technology is needed that goes beyond simply counting people to predict where crowds will flow in and how they will move. A domestic research team has developed a new technology that predicts crowd density based on this approach.
A research team led by Lee Jae-gil, a professor in the School of Computing at the Korea Advanced Institute of Science and Technology (KAIST), said on the 17th that it has developed a new artificial intelligence (AI) technology that can more accurately predict crowding conditions. The results were presented in August at the Association for Computing Machinery Special Interest Group on Knowledge Discovery and Data Mining (KDD) 2025, an international conference in the field of data mining.
The way crowds gather cannot be explained by changes in headcount alone. Even with the same number of people, the level of risk varies depending on where they enter and which direction they exit. Yet most previous studies on crowd density used only one type of information, such as how many people are gathered or which routes people are flocking to.
The researchers modeled these movements as a graph that changes over time. They represented how many people are in a particular area as nodes and the flow of people between areas as edges, concluding that accurate prediction is only possible when both types of information are analyzed simultaneously.
To that end, the team developed a bimodal learning approach that simultaneously considers current population counts and flows while learning both spatial connectivity and temporal change. They also implemented an AI that predicts where and when congestion will occur by introducing a three-dimensional contrastive learning method that adds time information to two-dimensional spatial data. When validated, the developed AI achieved prediction accuracy up to 76.1% higher than existing methods.
Lee said, "It is important to develop technologies that can have a social impact," adding, "We hope this technology will greatly help protect everyday safety by managing crowds at large events, easing traffic congestion in city centers, and curbing the spread of infectious diseases."
References
KDD 2025 (2025), DOI: https://doi.org/10.1145/3711896.3736856