computersees·main
dindigul · tamil nadu · in
▶ Scope a project
busnet/README.md×
back to products
// products/busnet/
Hardware + Software Edge AI

Passenger counting
camera system

A complete edge AI system we engineered for transit operators — custom hardware, on-device inference, and a reporting pipeline that runs 24/7 in extreme conditions.

Start a conversation →
BusNet edge-compute camera enclosure — weatherproof housing with embedded Jetson for passenger counting
The challenge

Counting passengers in Indian buses

Transit operators had no reliable way to audit actual ridership against ticket sales. Manual counts are inaccurate. Existing solutions weren't built for the crowding patterns, lighting conditions, and harsh environments of Indian public transit.

/ accurate count reports

Trip-by-trip data for ticket reconciliation

Every boarding and alighting captured and logged. Operators audit actual ridership against ticket sales.

/ revenue confidence

Fill more passengers, know exact counts

Operators can confidently fill more passengers on-route, knowing independent counts are being tracked.

/ operational insights

Driving patterns, stop-level data, peaks

Stop-level ridership data and peak analysis to optimise routes and scheduling.

What we engineered

Full-stack edge AI system

/ CPU

Custom hardware design

Purpose-built camera unit with embedded compute. Ruggedised for heat, dust, and vibration. Day and night operation with IR capability.

/ EYE

On-device CV models

Person detection and tracking optimised for overhead camera angles. Handles crowding, dim lighting, and crew filtering — all inference on the edge device.

/ CLOUD

Live reporting pipeline

Real-time data upload over cellular. Trip reports with video clips for each boarding/alighting event. Cloud dashboard for fleet-level analytics.

/ GLOBE

Made for Indian transit

Trained on Indian boarding patterns — crowded doorways, simultaneous entry/exit, non-standard stops. Adapts to global markets.

BusNet passenger detection in action — overhead view with bounding boxes on boarding passengers
The hard problem

If it works here, it works anywhere

A bus fitted with BusNet cameras — deployed passenger counting in the field

Anyone can build CV for a clean, controlled environment. We made ours work in the hardest conditions — 50°C heat, dust, vibration, pitch darkness, patchy connectivity, and crowds pushing through a single door simultaneously. That's the engineering challenge we solved.

  • Accurate detection in zero-light and overcrowded doorways
  • On-device inference at 50°C with no active cooling
  • Offline-first architecture — stores and syncs when connectivity returns
  • Proven in the field, adaptable to any transit system globally
In the field

Two clips from deployed buses

The hardware and models running on live routes. No re-edits, no lab setup.

/ Full debug overlay

Passenger counting · full debug

Detection + part-segmentation + re-id + path tracking overlaid on the doorway feed. What the pipeline sees, frame by frame.

/ Pitch-dark deployment

IR night vision · 24/7

Same pipeline, IR sensor, zero ambient light. On-device inference runs unchanged through the night shift.

Need an edge AI system?

We've proven we can go from concept to deployed hardware running real-time CV models. Let's talk about your use case.

Start a conversation →
end of file · © 2026 computersees · helping computers see
main 12 pipelines live parity · 99.94%
py 3.11 cuda 12.4 trt 10.0 IST ⛶ Focus mode