Problem
Large distribution yards struggle with real-time visibility into vehicle locations and dock availability. Manual tracking leads to bottlenecks, wasted time, and missed appointments.
Approach
Built a computer vision pipeline using YOLO for vehicle detection and a custom tracking algorithm for persistent identification across camera feeds. The system integrates with existing warehouse management software through a REST API.
Key technical decisions:
- YOLO v8 for detection — best balance of speed and accuracy for this use case
- Custom re-identification model trained on yard-specific vehicle features
- Event-driven architecture for real-time updates
- PostgreSQL with PostGIS for spatial queries
Outcome
Reduced average vehicle dwell time by 35% in pilot deployment. The system processes 12 camera feeds simultaneously on a single GPU node with sub-second latency.