---
tags:
- keypoint-detection
- sports
- football
- soccer
- roboflow
- yolov8
license: mit
---

# Football Field Keypoint Detection (Roboflow)

YOLOv8 keypoint detection model for identifying field lines and key points on a football/soccer pitch.

## Model Details

- **Model ID**: `football-field-detection-f07vi/14`
- **Platform**: Roboflow
- **Architecture**: YOLOv8 Pose
- **Task**: Keypoint detection

## Quick Start

### Installation

```bash
pip install inference-gpu  # or inference for CPU
pip install supervision
```

### Basic Usage

```python
from inference import get_model
import supervision as sv

# Initialize model
api_key = "YOUR_ROBOFLOW_API_KEY"
model = get_model(
    model_id="football-field-detection-f07vi/14",
    api_key=api_key
)

# Run inference
result = model.infer("image.jpg", confidence=0.3)[0]
key_points = sv.KeyPoints.from_inference(result)

print(f"Detected {len(key_points.xy[0])} keypoints")
```

### Using the Wrapper

```python
from field_inference import FootballFieldDetector
import os

detector = FootballFieldDetector(api_key=os.getenv("ROBOFLOW_API_KEY"))
results = detector.predict("image.jpg", confidence=0.3)
```

## Applications

- **Homography Estimation**: Map player positions to tactical view
- **Camera Calibration**: Align multiple camera views
- **Tactical Analysis**: Project player movements on 2D pitch
- **AR Overlays**: Augmented reality graphics on field
- **Voronoi Diagrams**: Team possession and control analysis

## Example: View Transformation

```python
import numpy as np
from sports.common.view import ViewTransformer
from sports.configs.soccer import SoccerPitchConfiguration

# Get field keypoints
result = model.infer(frame, confidence=0.3)[0]
key_points = sv.KeyPoints.from_inference(result)

# Filter confident keypoints
filter = key_points.confidence[0] > 0.5
frame_points = key_points.xy[0][filter]

# Create transformer
config = SoccerPitchConfiguration()
pitch_points = np.array(config.vertices)[filter]

transformer = ViewTransformer(
    source=frame_points,
    target=pitch_points
)

# Transform player positions to pitch coordinates
pitch_xy = transformer.transform_points(player_xy)
```

## API Key

You need a Roboflow API key to use this model. Get one at [roboflow.com](https://roboflow.com).

## Citation

```
@misc{football-field-detection,
    title={Football Field Detection},
    author={Roboflow},
    year={2024},
    url={https://universe.roboflow.com/roboflow-jvuqo/football-field-detection-f07vi}
}
```