Anvesha: Autonomous Railway CCTV Computer Vision & Crowd Safety
An intelligent surveillance engine designed to run over existing railway station CCTV networks, detecting hazardous crowd surge densities, platform bottlenecks, and track trespass incidents in real time.
1. The Problem
Major railway hubs across India handle hundreds of thousands of commuters daily. During peak festival seasons, unannounced platform changes, or sudden train delays, localized crowd density on narrow footbridges and platform edges can escalate into life-threatening surges within seconds.
Station control rooms already have hundreds of CCTV cameras, but human operators cannot monitor 50+ video screens simultaneously without severe attention degradation. An automated, zero-hardware-overhaul computer vision solution is critically needed.
2. Target Operators & Real-World Stakeholders
Station Master & RPF (Railway Protection Force)
Receives instant audio-visual alerts when platform sector crowd density exceeds safety limits (e.g., >3 persons/m²).
Platform Safety Staff
Identifies localized bottlenecks before stampede risks develop, allowing proactive crowd redirection before trains arrive.
Operations Dispatchers
Historical density trends inform platform allocation schedules to avoid simultaneous high-density arrivals on adjacent tracks.
3. Architecture & Vision Pipeline
Anvesha ingests standard RTSP IP video streams directly from existing fixed station cameras. The image pipeline applies perspective projection correction to compensate for camera mounting tilt, separates background infrastructure, and generates a real-time crowd heat density map.
4. Neel's Role & Technical Choices
- Designed the OpenCV video frame processing loop, implementing background subtraction and optical flow tracking to gauge movement direction.
- Formulated threshold alert logic: when crowd density in any bounding quadrilateral exceeds safety tolerance for >5 seconds, alert signals are triggered.
- Built the demonstration prototype presented during institute-level internal SIH evaluations.
5. Challenges & Trade-offs
Severe Occlusion in Crowds
In tightly packed platforms, human bodies overlap heavily, making standard bounding boxes inaccurate. Prioritized head detection and blob density centroids rather than full-body silhouette tracking.
Varying Illumination & Camera Angles
Station cameras face direct sunlight during morning and sodium-vapor yellow lighting at night. Implemented adaptive histogram equalization (CLAHE) to normalize contrast across light cycles.
6. Current Status & Substantiated Evidence
Honest Status Disclosure
Anvesha was developed as a solution concept for the internal institute-level Smart India Hackathon problem statement at CHARUSAT. The core detection and density alert scripts exist in the verified GitHub repository. It is a proof-of-concept prototype rather than a production-deployed railway software.