H/E Tracking Mouse

Abstract

Waiting for final completion

Introduction

• Background: I use computer equipment for a long time and use a regular mouse. This causes my right wrist to feel uncomfortable all the time.
• Challenge Description: Waiting for final completion

Hardware

UM790 Pro

USB Camera

Software

RyzenAI-SW

Python

Mediapipe - Face landmark detection

Building

In order to achieve eyeball and face detection, I used the Python module MediaPipe Face Landmarker. It can output the score of eye status.

Through a simple training process, an ONNX is output and run on the NPU.

Results

The accuracy of the output results is very low. Unexpected cursor movement can be seen.

Don’t know why the NPU shows working at 0%. But it does work, aybe the model is too simple.

Future Work

I haven't tested out a suitable model yet. I hope that one day when inspiration comes, that can build a more accurate model.

Conclusion

Using NPU to run the model can reduce the loading on the CPU, but building the model still requires people to complete it. The camera plays an important role in this project, with high resolution and correct placement having a huge impact on the results.

References

1. Cover image created from Copilot (microsoft.com) Designer