The situation
Construction sites need to know which vehicles are where, when they entered and left, and whether they stay in their designated zones. Manual logs were slow and error-prone.
Custom vehicle detection at construction sites
Real-time detection of construction vehicles to monitor movement, log entries and exits, and keep vehicles inside designated work zones.
A leading construction company
01 · Context
02 · Challenge
What had to be true
Detect site-specific vehicles reliably in real time, on low-cost edge hardware at the site.
03 · What we built
The system
- Pre-defined classes (excavators) combined with custom training for tippers, concrete mixers and boom pumps
- A dataset of site vehicles in varied orientations and backgrounds, randomly resized and skewed for robustness
- Annotations in labelImg, an 80/20 train–validation split and a custom YAML label file
- YOLOv5 trained on the custom dataset, then converted from PyTorch to a TensorRT engine for NVIDIA Jetson Nano
- Automated entry and exit logging, and alerts for unauthorised movement
04 Architecture
How it works, step by step.
Step through the system, or let it play.
- Site camera streams video
- YOLOv5 detects vehicles
- Zones and movement checked
- Entries and exits logged
- Alerts on unauthorised use
05 · Outcome
Real-time monitoring that cuts manual logging errors and labour, and a pattern for other sites that need to track equipment.
06 · Stack
Next case · Retail · Computer visionCustom retail item detection in store