Realtime ONNX Handtracking

A hand tracking system built with ONNX models and OpenCV that detects hands and tracks 21 landmarks, color-coded by depth. To keep tracking stable across varying distances, I implemented a dynamic bounding box mechanism.

Unlike most approaches, this one uses standalone ONNX models in a single Python script with no MediaPipe libraries or Python bridges. That streamlined design makes it suitable for Unity integration via the Sentis package, and I plan to deploy it into an AR application while keeping it free of external dependencies.