Related Experiment Video
Updated: Jan 9, 2026
06:37
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
5.2K
Deep Learning-Based Adaptive Sitting Posture Recognition System
Summary
This study introduces a non-contact system for recognizing sitting postures using cameras and AI. It accurately monitors posture in real-time, offering a portable solution to reduce health risks associated with poor posture.
Area of Science:
- Biomedical Engineering
- Computer Science
- Ergonomics
Background:
- Prolonged poor sitting posture is a significant risk factor for musculoskeletal disorders and chronic diseases.
- Existing solutions like wearable devices or pressure sensors have limitations in portability and efficiency.
Purpose of the Study:
- To develop a portable and adaptive sitting posture recognition system.
- To provide real-time monitoring and feedback for improving workplace health and efficiency.
Main Methods:
- Integration of infrared wide-angle cameras, a deep learning model (YOLOv11), and edge computing.
- Implementation on a TB-RK3399ProX development board with an NPU for efficient processing.
- Development of a position calibration function for compatibility with various chairs and users.
Main Results:
- Accurate classification of 18 sitting posture combinations with 94.64% overall accuracy.
- Achieved low inference latency of 8.73 ms.
- Demonstrated stability and reliability, even with image occlusion.
Conclusions:
- The system offers an efficient, portable, non-contact solution for sitting posture recognition.
- It effectively addresses challenges like image occlusion and diverse environments.
- Provides an innovative option for enhancing health and productivity in modern work settings.
