AI traffic management on Google Cloud
Camera feeds to adaptive signal timing, with a human override.
- Status
- PrototypeComplete proof of concept with deployment scripts
- Period
- Sep 2025
- Role
- Architecture and full implementation
- Brand
- Tuchopini
Cloud Functions extract frames from traffic cameras into Pub/Sub; Dataflow streams them through Vertex AI vehicle detection into BigQuery; a signal controller adapts timing per approach with emergency-vehicle and manual override; a FastAPI service on Cloud Run feeds a live React dashboard. Nine subsystems including a simulation harness and automated GCP setup.
What it does
- Pub/Sub ingestion that absorbs camera and network failures
- Vertex AI detection and classification
- Adaptive signal timing with emergency override
- BigQuery analytics, monitoring, simulation
An AI that cannot be overridden at a real junction is a liability, not a feature. The override path is designed in, not bolted on.
More ai, vision and iot
Built ▶ 1 AI, vision and IoT
AI traffic survey and vehicle counting
Survey-grade counts from video: 14 categories, by direction, in 15-minute slots.
Python · YOLOv8 · OpenCV · SQLite
Prototype ▶ 1 AI, vision and IoT
Smart traffic lights (ESP8266 + Raspberry Pi)
Two coordinated intersections on commodity hardware, and a YOLO-driven signal demo.
ESP8266 · Raspberry Pi · Python · YOLOv8
AI, vision and IoT
Smart parking occupancy
Two Raspberry Pis, two sensors, one live bay count you can trust.
Raspberry Pi 4 · Python · OpenCV · YOLO