# Efficient Human Posture Recognition and Assessment in Visual Sensor Systems: An Experimental Study

## TL;DR

This paper introduces a new system for recognizing and assessing human posture using visual sensors, achieving high accuracy and real-time performance.

## Contribution

A novel architecture with four subsystems for efficient human posture recognition and assessment using visual sensors.

## Key findings

- The proposed system achieves an overall accuracy exceeding 96%.
- The architecture demonstrates excellent real-time performance and scalability.
- The system was validated using pull-up and push-up exercises in an experimental testbed.

## Abstract

Currently, recognition and assessment of human posture have become significant topics of interest, particularly through the use of visual sensor systems. These approaches can effectively address the drawbacks associated with traditional manual assessments, which include fatigue, variations in experience, and inconsistent judgment criteria. However, systems based on visual sensors encounter substantial implementation challenges when a large number of such sensors are used. To address these issues, we propose a human posture recognition and assessment system architecture, which comprises four distinct subsystems. Specifically, these subsystems include a Visual Sensor Subsystem (VSS), a Posture Assessment Subsystem (PAS), a Control-Display Subsystem, and a Storage Management Subsystem. Through the cooperation of subsystems, the architecture has achieved support for parallel data processing. Furthermore, the proposed architecture has been implemented by building an experimental testbed, which effectively verifies the rationality and feasibility of this architecture. In the experiments, the proposed architecture was evaluated by using pull-up and push-up exercises. The results demonstrate that the proposed architecture achieves an overall accuracy exceeding 96%, while exhibiting excellent real-time performance and scalability in different assessment scenarios.