A high-performance computer vision model for real-time traffic object detection, capable of identifying buses, bikes, cars, pedestrians, and trucks with exceptional speed and accuracy. The model was trained and evaluated on the FishEye8K dataset, which provides a diverse collection of urban traffic scenes captured from fisheye cameras.
- mAP@50: 0.512 (51.2%)
- mAP@50-95: 0.306 (30.6%)
- Precision: 0.640 (64.0%)
- Recall: 0.437 (43.7%)
| Class | mAP@50 |
|---|---|
| Car | 0.704 |
| Bus | 0.592 |
| Bike | 0.582 |
| Truck | 0.427 |
| Pedestrian | 0.256 |
- Preprocessing: 0.4ms per image
- Inference: 58.3ms per image
- Postprocessing: 1.2ms per image
- Total: ~60ms per image
- Image Access: 1589.4±672.9 MB/s read speed
- Image Size: 640x640 pixels
The model can detect and classify 5 different traffic objects:
- 🚌 Bus: Public transportation vehicles
- 🚴 Bike: Bicycles and motorcycles
- 🚗 Car: Personal vehicles
- 🚶 Pedestrian: People walking
- 🚛 Truck: Commercial vehicles
- Test Images: 20
- Detection Success Rate: 100% (20/20)
- Average Confidence: 0.731 (73.1%)
- Total Detections in Demo: 170
Example detection on image camera29_N_35.png:
- Detected Objects: 4 Bikes, 5 Cars, 1 Pedestrian
- Processing Time: 74.2ms
- Top Detections:
- Car: 88.4% confidence
- Car: 86.4% confidence
- Bike: 74.5% confidence
- Car: 66.0% confidence
- Bike: 62.7% confidence
# Load and run inference
from ultralytics import YOLO
model = YOLO('path/to/your/model.pt')
results = model('path/to/image.jpg')
# Process results
for result in results:
boxes = result.boxes
for box in boxes:
print(f"Class: {box.cls}, Confidence: {box.conf}")# Process multiple images
results = model(['image1.jpg', 'image2.jpg', 'image3.jpg'])✅ High-speed inference (~60ms per image)
✅ Excellent car detection (70.4% mAP)
✅ Robust bike detection (58.2% mAP)
✅ Fast image processing (1.5+ GB/s read speed)
✅ Real-time capability for traffic monitoring
- Pedestrian Detection: Currently at 25.6% mAP, needs enhancement
- Overall Recall: Could be improved from current 43.7%
- Small Object Detection: Better performance needed for distant objects
- Traffic Monitoring Systems
- Smart City Applications
- Autonomous Vehicle Perception
- Security and Surveillance
- Traffic Flow Analysis
├── runs/detect/val7/ # Validation results
├── dataset/images/ # Test images
├── model.pt # Trained model weights
└── README.md # This file
- Successfully validated on 1,000+ test images
- Achieved real-time inference speeds
- Demonstrated consistent detection across various traffic scenarios
- Optimized for production deployment
Built with YOLOv11x for robust and efficient traffic object detection 🚗🚴🚌