Friday 28 February 2025
A new approach has been developed to estimate the collective volume of crowds in a scene, which could have significant implications for event management and public safety. The technique uses artificial intelligence to analyze images and predict the total volume occupied by groups of people.
The researchers created a dataset called ANTHROPOS-V, which contains synthetic videos featuring crowds in diverse urban environments. Each frame of these videos is labeled with information about the individuals present, including their volume, shape parameters, and key points such as the location of their heads and shoulders.
To develop the crowd volume estimation (CVE) task, the researchers adapted several existing models from the fields of human mesh recovery and crowd counting. They also introduced a novel approach that combines features from both domains to improve performance.
The results show that the new model, called STEERER-V, outperforms previous approaches in estimating crowd volume. The researchers tested the model on a variety of scenarios, including scenes with multiple people, varying lighting conditions, and occlusions.
One key advantage of STEERER-V is its ability to handle challenging situations such as low-quality images or severe occlusions. This is because the model uses a combination of features from different sources, including the shape and location of individual bodies, to make its predictions.
The researchers also explored the potential benefits of using temporal information in CVE. They modified STEERER-V to leverage context frames, which are neighboring frames that provide additional information about the scene. This approach improved the model’s performance by 5.27% in terms of mean absolute error and 4.22% in terms of per-person mean absolute error.
The development of STEERER-V has significant implications for a range of applications, from event management to public safety. By accurately estimating crowd volume, authorities can better plan and respond to events, reducing the risk of overcrowding or other hazards.
The researchers are continuing to improve their approach, with plans to test it on real-world data in the future. As the technology continues to evolve, it could potentially be used in a wide range of scenarios, from concerts and sporting events to emergency response situations.
Cite this article: “Estimating Crowd Volume with Artificial Intelligence”, The Science Archive, 2025.
Artificial Intelligence, Crowd Volume Estimation, Event Management, Public Safety, Image Analysis, Machine Learning, Human Mesh Recovery, Crowd Counting, Steerer-V, Anthropos-V
