A local desktop application for tracking your posture while working on a computer with a webcam. The goals are accuracy, low-resource usage, and intuitive UI.
- Made entirely in Rust, mainly using the ort library for inference and iced for the UI.
- The pose detection is the YOLO pose model by Ultralytics (version may change in future updates).
Claude Code was used in the development of this application. Architecture decisions and code reviews are strictly non-AI.
UI Rough Draft in LibreOffice
Rust Library Test Project Link: rust-webcam-model-bench Used this simple project to make sure that the libraries such as ort and ccap-rs could be integrated with Iced. Also, it allowed me to test multiple models on my laptop for efficency in a similar environment.
POC Project Link: PostureTracker_PoC I built this to see if it would be feasible to capture posture data using RF-DETR and ViTPose computer vision models through a laptop webcamera. Another goal of the experiment was using a low frame rate of images passed into the model allowing it to run on an edge device without taking up too many resources.