Use Case Title
Vehicle Type Detection - Detect bikes vs cars in campus parking or roads.
Team Members
- Rugvedi Khandekar
- Vrinda Khare
- Pranali Meher
- Vinay Kasar
- Himanshu Machhi
Use Case Importance
In smart city environments, managing parking efficiently is essential. By detecting whether a car or bike is occupying a parking slot, our system enables real-time slot monitoring, data-driven planning, and automation in parking systems. This project highlights how edge AI can improve public infrastructure and urban mobility.
Sample Image:
Data Collection and Annotation
Data Sources:
Roboflow (for annotation and dataset structure)
Kaggle (for diverse car and bike images)
Classes:
Car
Bike
Annotation Tool: Roboflow
Total Images: ~500
Train: 350
Validation: 100
Test: 50
We ensured variation in lighting (sunlight, shade), angles, and background clutter to improve model generalization in real-world parking areas.
Model Training and Validation
We trained a YOLOv8n model using Google Colab, aiming for efficient performance on mobile devices.
Metrics Monitored:
mAP@0.5
Precision
Recall
Loss
Observations:
Excellent detection performance for both classes
Minor confusion in overlapping vehicle areas
High confidence scores in open and clear frames
Model Deployment and Demo
Our best model (best.pt) was deployed using the YOLOvX mobile app for real-time classification of parked vehicles. The app was tested in live conditions and responded accurately with around 8–12 FPS on mid-range smartphones.
Strengths:
Real-time response
Robust detection in varying light and angles
Limitations:
Frame drops on low-end phones
Reduced accuracy with partially occluded vehicle
Demo Video
Click to Watch the Demo:
Conclusion
This project shows how YOLOv8n + YOLOvX can be used to create lightweight, efficient, real-world detection models. With just 500 images, we built a functional solution that performs well in smart parking and traffic surveillance environments.
