'AI made crowd mgmt effective during Maha Kumbh'

Prayagraj: Optimum use of Artificial Intelligence (AI) helped the police to redefine crowd monitoring density during the Maha Kumbh 2025 in Prayagraj that saw participation of over 66 crore devotees from across the country and world.

The cutting-edge technology was leveraged on the instruction of Chief Minister Yogi Adityanath who had directed DGP Prashant Kumar to effectively use AI to ensure the safety and security of pilgrims, and for better traffic and crowd management. The state police, and the media authorities, along with implementing agencies, did extensive research and planning, followed by a series of discussions with domain experts on using AI.

Monitoring crowd density during the world's biggest religious and spiritual congregation was critical for ensuring safety. This system was aimed at estimating the number of people present in a fixed region at a given time using computer vision-based crowd detection techniques. ADG (Prayagraj zone) Bhanu Bhaskar said, "Crowd density estimation involved validation and consistency check which generated alerts according to the crowd level as low\medium\high. It was revalidated to suitably modify the alerts. The approach leverages deep learning models trained on real world datasets to handle varying crowd densities." He added that real-time processing enabled authorities to take proactive measures to prevent overcrowding.

Integration with surveillance systems ensured continuous monitoring and timely alerts, Bhaskar said. IG (Prayagraj range) Prem Gautam told TOI that an accurate head count at entry and exit points of the mela area was essential for safety and crowd management. "This was effectively done with the help of the system that used computer vision-based techniques to detect and count individuals crossing designated entry and exit points in real-time. Deep learning models, combined with object tracking algorithms, helped in differentiating between incoming and outgoing individuals even in dense and overlapping scenarios," Gautam said.

The IG further said: "The system, after analysing the entry and exit rate of pilgrims, could tell the authorities about resultant crowd density. The data was used to analyse crowd trends, predict congestion and optimise resource development." Pointedly, live dashboards and automated alerts enabled officials at ICCC (Integrated Control & Command Centre) to take timely actions to prevent bottlenecks.

"This AI-generated data, turnaround time, along with data from the facial recognition system (FRS), was used in an Area-based Headcount Modelling model to obtain data-backed estimation of pilgrims visiting Sangam," the official said.

A network of 550 AI-enabled cameras kept a constant watch over the sprawling tent city and entry-exit points offering real-time updates. Advanced AI-driven crowd density algorithms and mathematical modelling delivered 95% accuracy in headcounts, replacing traditional guesswork with precise data. "This integration of modern technology with traditional faith has set a global benchmark,"- said IG. Also, person and vehicle attribute recognition enhanced situational awareness and security monitoring during Maha Kumbh. This system utilised deep learning models to extract key attributes like clothing colour, bag detection, vehicle type and licence plate characteristics from CCTV footage. The analytics system operated in real-time, ensuring seamless integration with law enforcement and emergency response workflows.

The system could alert any mismatch in vehicle and number plate regarding any change in number plate on any vehicle, indicating suspicion on them. This capacity significantly improved event safety and operational efficiency in high-density environments, officials said.

Similarly, technology was used for effectively managing parking. After correctly estimating the exact situation of occupancy in a parking area, the system could tell the estimated time in which the parking would be full based on the rate of inflow and outflow of vehicles. This was used to predict when the next available parking should start taking vehicles. This also helped with efficiently using parking spaces to avoid traffic jams.