Object Detection for Vehicles with Yolo

Urban traffic management needs more than generic vehicle counting. Emergency vehicles such as ambulances and fire trucks should be recognized separately so that intelligent traffic-control systems can prioritize them when necessary.

Dataset design

The study presents a vehicle dataset with 29,759 images and seven classes:

1. Ambulance
2. Bus
3. Car
4. Motorcycle
5. Truck
6. Van
7. Fire truck

The dataset contains 55,925 annotations and is split into training, validation and testing subsets. Data augmentation—including HSV transformations, translation, scaling, flipping and mosaic augmentation—was used to improve robustness and address class imbalance.

Comparing YOLO versions

YOLOv5s, YOLOv6-s and YOLOv7 were trained and evaluated. Among them, YOLOv7 produced the strongest overall result with approximately 85% precision, 76% recall and 85% mAP@0.5.

Why emergency-vehicle classes matter

Generic datasets frequently map ambulances to vans and fire trucks to trucks. That is acceptable for general-purpose detection, but not for a smart intersection that needs to grant priority to emergency traffic.

A dedicated class structure allows the perception layer to feed more meaningful information into a control layer—such as a reinforcement-learning traffic agent.

From perception to control

text Copy
Camera → Vehicle detector → Traffic-state estimator
       → Priority logic / RL agent → Signal timing action

This separation between perception and decision-making creates a practical foundation for intelligent transportation systems.