yolo_digital_screen_detector

yolo_digital_screen_detector is an open-source Python tool for detecting and localizing digital screens (monitors, TVs, laptops, tablets, smartphones) in images using YOLO-based deep learning. The project provides fast, accurate detection and can automatically crop detected screen regions for further analysis, privacy masking, or OCR.

---

Features

• YOLO-based detection: Fast, robust, and accurate screen detection using state-of-the-art object detection models.
• Easy CLI interface: Run detection on images with a single command.
• Automatic cropping: Saves output images with detected screens cropped or highlighted.
• Flexible input/output: Works with individual files or output directories.
• Production-ready: Minimal setup with requirements.txt and default config values.
• Ready for video support: Extendable for video and webcam detection.

---

Installation

Clone this repository and install all dependencies:

git clone https://github.com/emirbaycan/yolo_digital_screen_detector.git
cd yolo_digital_screen_detector
pip install -r requirements.txt

---

Usage

Basic usage with default values:

python yolo_screen_model_test.py

• Runs detection using the default model and test image.

Custom usage:

python yolo_screen_model_test.py \
  --model path/to/your_model.pt \
  --input path/to/your_image.png \
  --output results/ \
  --conf 0.6

Arguments:

• --model: Path to the trained YOLO model .pt file (default: runs/detect/train10/weights/best.pt)
• --input: Input image file (default: camera_images/frame_1738193401.png)
• --output: Output file or directory (default: output.png)
• --conf: Detection confidence threshold (default: 0.5)

Example:
Detect and save output to a directory:

python yolo_screen_model_test.py --input my_image.jpg --output ./results/

---

Example

Input image:

Detection result:

---

Model Training

To train your own digital screen detection model:

1. Annotate your images

   • Use LabelImg (open-source, GUI-based) to draw bounding boxes around digital screens in your dataset.
   • Save labels in YOLO format (.txt files with each image).
   • Alternative tools: makesense.ai, Roboflow.

2. Organize your dataset

   • Images and labels should be in folders (e.g. images/, labels/).
   • Each label file should have the same name as its image, but with .txt extension.

3. Train with Ultralytics YOLO

   • Follow Ultralytics YOLOv8 documentation for custom object detection training.
   • Example command:

     yolo detect train data=data.yaml model=yolov8n.pt epochs=100 imgsz=640

---

Requirements

All required packages are listed in requirements.txt:

ultralytics>=8.0.0
torch>=1.8.0
opencv-python>=4.5.0
PyYAML>=5.4.0
av>=10.0.0
matplotlib>=3.6.0

Install with:

pip install -r requirements.txt

---

Annotation

Bounding box annotation for screens was performed using LabelImg.

• To label your own dataset, install LabelImg, draw boxes around screens, and export in YOLO format.
• Example workflow:
  1. Install LabelImg: pip install labelImg
  2. Run: labelImg
  3. Load images, annotate, and save labels as YOLO .txt files.

---

Contributing

Contributions, bug reports, and feature requests are welcome!

• Fork this repo, make your changes, and open a pull request.
• For suggestions or questions, open an issue.

---

License

MIT License

---

Contact

Developed by Emir Baycan
GitHub

For questions, open an issue or contact via GitHub.

---