A camera-based AI system that automatically counts snow transport trucks, identifies them via license plate recognition, retrieves vehicle dimensions from a registry database, and estimates both the volume and weight of transported snow — enabling accurate billing and operational analytics.
End-to-end automated pipeline from camera feed to billing data
Cameras positioned at the snow dumping polygon capture arriving trucks in real-time.
YOLO v12 deep learning model identifies and localizes trucks in every frame with high confidence.
ALPR extracts license plate text from the detected truck region with OCR confidence scoring.
The plate is matched against the vehicle registry to retrieve truck dimensions and cargo bed volume.
Volume × SWE density = estimated snow weight. Results are logged for billing & analytics.
Modular design for flexible deployment
Built on proven, state-of-the-art components
Real-time truck detection using the latest YOLO architecture. The medium model balances accuracy and speed for production deployments. Trained on COCO dataset with truck (class 7) specialization.
Fast-ALPR provides both plate localization and OCR in a single pipeline. Confidence scoring enables multi-frame confirmation, reducing false reads and ensuring reliable truck identification.
Snow Water Equivalent (SWE) data from meteorological sources, combined with known truck cargo volume, enables weight estimation without physical scales. SWE varies by snow type and transport conditions.
From truck dimensions to billing-ready weight data
The cargo bed dimensions (Length × Width × Height) are retrieved from the vehicle registry database after license plate identification. For known trucks, the maximum cargo volume is a fixed, reliable value.
Vcargo = L × W × H
Computer vision can estimate the fill percentage of the truck bed by analyzing the snow surface relative to the truck walls. For the current prototype, a default 100% fill is assumed.
Vsnow = Vcargo × fill%
Snow Water Equivalent provides the density of the transported snow. Combined with volume, it yields the estimated weight.
Wsnow = Vsnow × SWE (kg/m³)
| Snow Type | SWE (kg/m³) | Context |
|---|---|---|
| Fresh powder | 50 – 100 | Newly fallen, dry |
| Settled snow | 100 – 200 | Days-old, compacting |
| Wind-packed | 200 – 350 | Wind-driven compaction |
| Transported | 300 – 450 | Loaded by machinery |
| Wet / Spring | 350 – 500 | Melting, saturated |
| Glacial / Ice | 500 – 917 | Refrozen, dense |
Phased approach to a production-ready system
YOLOv12-based truck detection, ALPR integration, multi-frame confirmation logic, web-based processing interface with real-time video feed.
SQLite vehicle registry with truck dimensions, automatic plate-to-vehicle matching, snow event logging, enriched results with vehicle specs and volume data.
Integration with FMI/SYKE API for real-time SWE data. Date-aware snow density calculation. CV-based fill level estimation for more accurate volume readings.
Connect to Traficom's open vehicle registry API for automatic retrieval of truck specifications by license plate. Requires GDPR-compliant data access agreement.
Live RTSP camera feed processing, real-time dashboard, automated billing report generation, multi-site support, and API access for third-party integration.
What is needed to take the system from prototype to production