Video Management System (VMS)
Central live view of 50 cameras on a multi-screen video wall, with person bounding boxes overlaid on each feed.
Case Study · Crowd Management Video Analytics
AI Crowd Intelligence for One of South India's Largest Dasara Festivals
Over 3 days, Katomaran turned a festival CCTV network at a renowned coastal temple in South Tamil Nadu into a live AI control room. It logged 77.8 lakh person detections, counted 2.5 lakh+ temple entries, and surfaced 814 face-recognition watchlist alerts for officer verification, leading to 10 verified matches.
- 77.8 lakhperson detections
- ~23 lakhindicative unique visitors*
- 4.58 lakhdetections in the peak hour
- 814FRS watchlist alerts
- 50cameras
- 3 days24x7
The challenge
Each year, a major Dasara festival at a renowned coastal temple in South Tamil Nadu draws lakhs of devotees over a few days. Crowds spread across the temple gates and queue lines, the beach, a procession route and the town's junctions. The busiest moments come in short, intense surges, including a 12-hour main night event.
The temple administration and district authorities needed to:
- See crowd build-up across the whole festival area in real time, not after the fact
- Estimate how many people were entering and leaving the temple, without double counting
- Know which zones and hours would peak, so police and volunteers could be deployed ahead of time
- Get early leads on persons on an official watchlist moving through dense crowds
- Do all of this on the existing festival CCTV network, set up within 1 day
The solution
Katomaran deployed its AI video intelligence stack in an integrated AI CCTV control room, with a large video wall and a live analytics dashboard.
Video Management System →
Video Analytics, crowd detection
Person detection on 18 crowd cameras covering the beach, procession route, temple surroundings and junctions.
People counting, entry and exit
3 temple-entry cameras (main entrance, free-darshan queue, children & elderly queue) and 1 exit camera, with normalization logic to avoid double counting.
Live dashboard
Total events, people count, crowd detection, busiest location and a time-of-day heat pattern for command decisions.
Face Recognition System (FRS)
Watchlist (blacklist) matching on live camera feeds. Each alert shows the full frame, the cropped face and the registered reference image, plus confidence, camera and timestamp.
Automated reporting
Daily and cumulative camera-traffic reports and an FRS evidence report for the authorities.
Deployment and architecture
- CamerasExisting festival CCTV
- Network
- Katomaran VMS
- AI AnalyticsCrowd · People Count · FRS
- Video Wall + Dashboard + Reports
- Police / Temple administration
Cameras: the existing festival CCTV network across the temple, beach, procession route and town junctions. 50 cameras in total, a mix of 8MP and 5MP varifocal cameras.
Ingest and management: Katomaran VMS aggregating all feeds into one control room. Deployed fully on-premise.
AI layer: person detection and crowd analytics, entry/exit counting, and FRS watchlist matching with a configurable confidence threshold (above 45% for the evidence report).
Presentation: multi-screen video wall, analytics dashboard with hourly trends, and FRS alert pop-ups for operators.
Operations: police and officials briefed directly in the control room. 6 operators across 3 shifts, round the clock.
Coverage window: continuous monitoring with reports for 19, 20 and 21 Oct 2026 (00:00-23:59 IST each day), including the 12-hour main night event.
Results
| Metric | Result |
|---|---|
| Person detections, 19-21 Oct 2026 | 77,83,516 (7.78 million) |
| Indicative unique visitors (estimate) | ~23.35 lakh (2.3 million) over 3 days |
| Busiest day | 35,65,192 detections (~10.7 lakh indicative unique) |
| Peak hour | 4,58,049 detections (17:00-18:00, peak day) |
| Main night event (12 hours) | 18,82,743 detections; ~5.65 lakh indicative unique; peak 19:00-20:00 |
| Approx. temple entries / exits counted | 2.5 lakh+ entries; 2.09 lakh+ exits |
| FRS watchlist alerts | 814 alerts in total: 253 in the 20 Oct report and 561 in the 21 Oct report |
| Evidence completeness | 100% of alerts with full-frame, face crop and reference image (0 failures) |
| Operational outcomes | 10 verified matches (persons) via face recognition; incident response time improved by 40% |
Key insights delivered to the authorities
- Surges were short and sharp. The peak day's busiest hour carried 4,58,049 detections.
- Hotspots shifted between day and night. During the main night event, the busiest cameras moved to the beach and procession route, and the day's busiest junction dropped to 32,488 detections.
- A second crowd surge came after midnight (2,71,855 detections, 00:00-01:00), which informed late-night staffing.
- Normalized entry counting avoided the roughly threefold overstatement that raw multi-camera sums would have produced.
Note: person detections are not unique individuals. Unique-visitor figures are indicative estimates. FRS results are investigative leads that authorised personnel verify.
Planning a festival, a large public event or a city control room? See how Katomaran turns your existing CCTV into real-time crowd intelligence.
