Real-time computer vision application that detects whether someone is wearing a quarter zip using YOLOv8 pose estimation and OpenCV edge detection.

• Real-time Detection: Analyzes webcam feed in real-time
• YOLOv8 Pose Estimation: Uses state-of-the-art pose detection to locate the neck area
• V-Shape Recognition: Detects the characteristic V-shaped collar of quarter zips
• Temporal Stability: Smoothed detection to prevent false positives
• Visual Feedback: Shows detection area and V-shape lines when quarter zip is detected

The detector identifies the neck area and looks for two diagonal lines forming a V-shape - the signature feature of a quarter zip collar.

1. Pose Detection: YOLOv8 identifies key body points (nose, shoulders)
2. ROI Extraction: Calculates neck region based on facial and shoulder landmarks
3. Edge Detection: Uses Canny edge detection to find lines in the neck area
4. V-Shape Matching: Identifies two diagonal lines (25-85° angles) forming a V
5. Temporal Filtering: Applies exponential moving average for stable output

• Python 3.8+
• Webcam

1. Clone the repository

   git clone https://github.com/kapilcdave/camera_detect.git
   cd camera_detect

2. Create virtual environment

   python3 -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies

   pip install -r requirements.txt

Run the detector directly with OpenCV window:

python quarter_zip_detector.py

Or using the virtual environment directly:

./venv/bin/python quarter_zip_detector.py

Controls: Press q to quit.

Run the web interface:

streamlit run streamlit_app.py

You can adjust detection parameters in QuarterZipDetector.__init__:

• self.alpha: Smoothing factor (lower = smoother/slower)
• self.lock_threshold: Frames to keep showing detection after loss
• self.conf_threshold: Detection confidence threshold
• Edge detection parameters in detect_v_shape() function

• YOLOv8: Pose estimation for body keypoint detection
• OpenCV: Image processing and edge detection
• NumPy: Numerical operations
• Streamlit: Web interface for easy deployment

• Locate neck region using nose and shoulder keypoints
• Apply Gaussian blur and Canny edge detection
• Use Hough Line Transform to detect line segments
• Filter lines by angle mechanism
• Verify V-shape formation by intersection logic
• Apply temporal smoothing for stable results

• opencv-python-headless
• ultralytics
• numpy
• streamlit

• Requires good lighting conditions
• Quarter zip zipper must be visible
• Single person detection (analyzes detected people independently)
• May trigger on other V-neck clothing

• Distinguish between quarter zips and other V-neck garments
• Add clothing color detection
• Support for video file input
• Performance optimization for lower-end hardware

MIT License - feel free to use and modify!

Contributions welcome! Please feel free to submit a Pull Request.

• YOLOv8 by Ultralytics
• OpenCV community