| Summary: | This project aims to develop a lightweight, real-time, single-user-focused sit-posture monitoring system for the Final Year Project. The goal of this project is to address health concerns related to improper sitting habits in office settings. The uniqueness of this system design is the implementation of a dual-camera setup, which enables the system to capture the user’s posture from both front and side perspectives at the same time, allowing it to analyze the frames and estimate the sitting posture immediately.
The major core of this project is three different types of machine learning models: MediaPipe Pose Detector for determining the position of user posture facing forward; YOLO for determining the position of user posture from a side view, including whether the user leans forward or not; and EfficientNet acting as a classifier to detect the signs of drowsiness from the user's eyes. Each model has played a vital role in system development. The system program script is written using Python and trained on Google Colab as it possesses powerful cloud GPU service. Further developments are taken in Jupyter Notebook, which included combining all the models and system UI.
This system offers an option for the user to customize preferred settings of the system. Besides, it will generate a feedback report after the user closes the program, and the user will understand his or her sitting quality through this report. In this project, the Cobb angle will be served as the measuring criterion for sitting posture. The journey of this project is to prioritize the integration of modern solutions with an intuitive interface. Through this sit-posture monitoring system, it is hoped to promote physical health for office users.
|
|---|