Developing an agent-based model for pilgrim evacuation using visual intelligence: A case study of Ratha Yatra at Puri

Abstract Religious mass gatherings require proper movement planning for pilgrims in crowded situations. Most religious towns are old and have informal infrastructure with haphazard road networks, which results in improper connectivity. However, the sacredness of religious festivals calls for gatherings of millions of pilgrims within a limited space and duration. The annual Ratha Yatra at Puri is an event that is manually controlled by the logistics provided by the district authorities. Pedestrian crowds move along a 3-kilometre stretch of the road to witness the manual dragging of three chariots for around 4 h. The objective of this paper is to understand crowd characteristics at a particular location during the event and suggest for crowd evacuation in emergencies through identified exit routes. In Ratha Yatra 2015, a single-camera video recording of the event was performed from an overhead location. The individual video frames were perspective-rectified and analyzed to detect human shapes and extract parameters of crowd density, flow, and speed. The results were simulated through an agent-based model to trigger an evacuation process at critical crowd densities. A similar approach using multiple cameras at multiple locations in a large mobility network can provide a decision-making tool to adopt safety measures in mass gatherings.