World
India
The Maha Kumbh Mela in Prayagraj has always been one of the world’s largest religious gatherings. This year, the estimated number of recorded attendees is approaching a staggering 64 crore. Despite the overwhelming influx of people, authorities have successfully managed the crowds with the help of cutting-edge Artificial Intelligence (AI) technology. This unprecedented integration of AI has allowed for real-time crowd monitoring and enhanced safety measures, ensuring a smooth experience for pilgrims and visitors alike.
The Preparation Behind AI Deployment
Although the Maha Kumbh Mela officially began on January 13, 2025, preparations for AI integration started almost a year in advance. The AI models used to estimate crowd density and movement were trained using footage from the Magh Mela 2024, a similar religious gathering, to understand human behavior in such large congregations.
“Maha Kumbh is a unique event where nothing can be predetermined, unlike any other event across the globe… we were supported by the previous Kumbh data as well. Our team sitting on the ground and the development team at the command center at Prayagraj did a lot of training, retraining, and annotation of data which helped move towards a matured and stable system,” said Shailendra Kumar Singh, Business Unit Head at Vehant Technologies, the company responsible for implementing AI at Maha Kumbh, while speaking to India Today.
The AI models required significant refinement before they were ready for deployment on such a large scale. Approximately three months before the Mela, ground teams conducted surveys to determine optimal camera placements. Ultimately, 1,700 cameras were strategically installed, including 500 AI-powered cameras specifically for real-time crowd analytics.
How AI Works at Maha Kumbh
The AI system at Maha Kumbh operates through an intricate network of cameras and servers, all interconnected via an optical fiber network. The live footage feeds into two AI models that work in tandem—one estimates crowd density, while the other counts individuals at key entry and exit points. The data from these AI models is further refined using inputs from facial recognition cameras and processed through a sophisticated business algorithm. The final crowd numbers are displayed on a dashboard accessible at the Integrated Command and Control Center (ICCC) in Prayagraj.
“One is for getting the crowd density data and another AI model which is installed at various entry and exit points of the Ghats and the routes. So now we have two pieces of information, one where people approach the Ghats, the other where people are already there at any particular point of time. So both these two data are being correlated and, with the help of an algorithm business tool, we come out with more logical and conclusive information,” Singh explained.
For better accuracy, virtual parameters have been defined at various locations, ensuring that each Ghat has an entry and exit count for monitoring purposes. The system is designed to continuously refine the final figures by cross-referencing multiple data points.
Challenges and Limitations
While AI has significantly improved crowd management, there are still some limitations. Since crowd density mapping is based on periodic scans, the same individual could be counted multiple times if they visit the Ghats on different days. Additionally, environmental factors such as poor lighting, weather conditions, and obstructed views can impact the accuracy of AI-generated data.
However, authorities emphasize that the system remains highly efficient. “If a person visits twice, they will be counted twice. This is the basis of headcount… but if there is a snapshot of the Mela, if we talk about right now, it will show the current number, so it uses different parameters and it cross-correlates the numbers to provide the final figure,” said Amit Kumar, IPS and In-charge of the ICCC at Prayagraj.
Beyond Crowd Counting: Other AI Applications
Apart from monitoring crowd numbers, AI-equipped cameras have been used for several additional purposes, including:
- One-way movement monitoring: AI detects if someone is moving against the designated flow of the crowd.
- Barricade tracking: AI identifies people attempting to cross barricades unlawfully.
- Parking management: AI-based vehicle tracking records license plate numbers to monitor parking areas effectively.
- Security enhancements: AI helps law enforcement track unusual behavior or unattended objects that could pose security threats.
The Future of AI in Large-Scale Events
The successful implementation of AI at Maha Kumbh Mela 2025 has set a benchmark for future large-scale events worldwide. AI models are expected to evolve further, integrating features such as anomaly detection for identifying potential safety hazards and behavioral analytics for predicting crowd movement trends.
“AI in large-scale events is rapidly evolving, with models becoming more adaptive, accurate, and intelligent over time. Future deployments may integrate anomaly detection to identify potential safety hazards, such as unattended objects or unusual movement patterns. Behavioral analytics could help predict crowd movement trends, optimizing traffic flow and security deployment, while automated incident response would enable AI-powered alerts for medical emergencies, lost individuals, and law enforcement interventions,” Singh added.
Despite the rapid advancements, AI still has room for improvement. Current accuracy levels range between 90-95%, influenced by environmental conditions and real-time adjustments. However, this implementation has provided AI systems with an invaluable dataset, allowing them to improve for future events of this magnitude.
