Jennifer kitesurfing in Aitutaki
🏄 Kitesurfer Detection & Tracking System
A practical computer vision project that detects and tracks kitesurfers in beach video footage. It uses a lightweight YOLO model for object detection and Meta’s SAM 2 to help annotate training data, making it easier to build a custom detector for riders and kites in real-world coastal scenes.
Designed for sports analysis and video-based experiments, the system can process recorded beach footage, including iPhone video, and follow riders across frames over time. It supports automated annotation, object tracking, and visual overlays, creating a useful workflow for both dataset creation and applied detection tasks.
✨ Key Features
- Detects kitesurfers and kites in beach footage
- Uses SAM 2 to assist with video and image annotation
- Tracks riders across frames for continuous analysis
- Supports recorded video, including iPhone footage
- Creates training data for custom YOLO detection models
- Displays visual overlays for detection and tracking results
🔍 How It Works
The workflow begins with video or photo data collected from kitesurfing sessions. SAM 2 is used to speed up annotation by propagating object masks or labels across frames, and the resulting dataset is then used to train a lightweight YOLO model. Once trained, the detector can process new footage, identify riders and kites, and track movement across the scene in real time or during playback.
🎯 Practical Uses
This project is useful for kitesurfing analysis, beach activity monitoring, sports footage review, and testing custom computer vision pipelines in outdoor conditions. It provides a practical example of how annotation tools, object detection, and tracking can be combined into a lightweight end-to-end workflow.
Short tech stack line:
Tech stack: Python, YOLO11 nano, Meta SAM 2, OpenCV, PyTorch, and Ultralytics tracking tools.