Passenger counting is one of the most data-rich and underutilised capabilities in public transit. The technology is mature, accuracy is high, and the operational applications — from service frequency planning to crowd management — are direct and measurable. The gap is almost never in the technology. It is in the plan for how the data will be acted upon.
Why Transit Authorities Deploy Passenger Counting
Without accurate demand data, service planning relies on periodic manual surveys — expensive, infrequent, and producing point-in-time snapshots. Revenue modelling for flat-fare systems lacks ridership validation. Crowd management at peak periods is reactive rather than anticipatory. Automated passenger counting closes this information gap continuously and at scale.
Technology Options and Trade-offs
Infrared beam counters — the simplest and most established technology. Beams across a doorway are broken by passing passengers. Inexpensive and reliable in controlled environments (turnstile gates, single-file passages). Accuracy degrades significantly in crowded conditions with wide doorways. Practical accuracy: 85–92% at uncontrolled doors, 97–99% at single-file turnstiles.
3D depth sensors (time-of-flight) — mounted above doorways, creating a depth map of the space below. Individual people detected as 3D objects and tracked through the sensor field. More robust in crowded conditions — a group of three is counted as three, not as one interrupted beam. Accuracy: 93–97% at station doors, exceeding 98% in controlled evaluations. Higher cost than infrared.
Computer vision (AI-based) — cameras combined with deep learning object detection. Highly flexible — the same installation can support counting, crowd density estimation, abandoned item detection, and flow direction analysis. Accuracy competitive with depth sensors when the model is well-trained for the specific environment. Privacy considerations require explicit data governance documentation.
Wi-Fi and Bluetooth probe detection — counts mobile devices as a proxy for people. Useful as supplementary data for concourse crowd estimation but not as a primary counting mechanism — device-to-person ratio varies, multiple devices per person inflate counts.
Accuracy Requirements by Application
- Revenue and fare apportionment — requires accuracy above 95%, ideally above 98%
- Service planning and frequency optimisation — accuracy above 90% typically sufficient for directional operational decisions
- Real-time crowd management — depends on the consequence of count errors at the safety action threshold
Integration with AFC and Operations Control
Comparing passenger counting data with AFC transaction data validates fare evasion rates and cross-checks revenue assumptions. Real-time count data fed to the Operations Control Centre enables dynamic dwell time adjustment, early warning of station congestion, and evidence-based decisions about deploying additional vehicles. Both integrations require that data streams share common timestamps and station references — a data architecture requirement that must be planned at system design, not retrofitted.
Privacy and Data Governance
Systems using cameras or Wi-Fi probe detection must address data governance explicitly. In Singapore, the Personal Data Protection Act (PDPA) and PDPC guidance on video analytics apply where individuals could be identified.
For passenger counting, anonymisation at the point of capture — where computer vision processing happens on the device before any data is transmitted, and no identifiable imagery is stored — typically removes the system from PDPA's personal data provisions. This design approach should be confirmed with legal counsel before deployment.
The IoT and embedded systems architecture for a privacy-compliant passenger counting system processes counting and density data at the edge, transmits only aggregated counts and anonymised flow data to the central platform, and maintains no video archive. This is both a privacy best practice and a bandwidth and storage efficiency.
At Sirona Robotics, our embedded systems and IoT integration capability covers the sensor layer, edge processing, data transmission, and enterprise integration that passenger counting systems require.