A portable sitting posture monitoring system based on a pressure sensor array and machine learning

Abstract

Poor sitting posture is one of the main inducements that lead to a series of skeletal muscle diseases. Sitting posture monitoring system can remind the user to maintain the correct sitting posture to prevent the harm of poor sitting posture to the body. In this paper, we proposed a portable sitting posture monitoring system to recognize the user's sitting posture and feedback the results in real time. A pressure sensor array is used to collect sitting postures related information, while the collected data can be displayed on a computer. The proposed system was designed to recognize seven types of sitting postures, including sitting upright, leaning forward, leaning backward, leaning left, leaning right, cross left leg, and cross right leg. Seven machine learning algorithms were implemented for comparation. The results showed that a five-layer Artificial Neural Network could achieve the highest accuracy of 97.07 %. To enhance system performance and reduce hardware cost, we further optimized the size of the sensor array. An 11 × 13 sensor array combined with Random Forest algorithm realized the optimal balance between classification accuracy (96.26 %) and hardware resource consumption. The final system prediction time is 19 us on the Raspberry Pi, which could satisfy the practical application requirement on the embedded platform.