A deep learning based 2-dimensional hip pressure signals analysis method for sitting posture recognition. (March 2022)

• Record Type: Journal Article
• Authors: Fan, Zhe; Hu, Xing; Chen, Wen-Ming; Zhang, Da-Wei; Ma, Xin
• Is Part Of: Biomedical signal processing and control. Volume 73(2022)
• Publication Date: 2022-03
• Subjects: Sitting posture -- Hip pressure -- 2D pressure sensor array -- Convolution neural network; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2021.103432
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: An automatic teenagers' poor posture recognition method is proposed. Hip pressure is analyzed using both traditional and deep learning-based methods. Extensive Experiments verify the advantages of the proposed method. Abstract: Abnormal sitting postures usually cause adolescents' myopia, scoliosis, and degenerative diseases. Therefore, research on intelligent monitoring technology that can quickly and accurately identify irregular sitting postures is of profound significance to the healthy development of adolescents. Existing methods mostly use computer vision to recognize sitting posture, but the model is not only complicated but also easily interfered with by problems such as occlusion and light. This paper proposes a method based on the analysis of the pressure on the hip interface to identify the sitting postures. An array pressure sensor placed on the cushion collects the tester's hip pressure and obtains a pressure heat map. This paper uses traditional feature extraction and shallow classifier methods and popular end-to-end deep convolutional neural network (CNN) methods to identify different types of sitting postures. The method in this paper is verified on the data of multiple testers of different body types. Experimental results show that the classification accuracy based on CNN reaches 99.82%, which proves the effectiveness of the method in sitting posture recognition. The study indicated hip pressure distribution is closely related to the sitting posture, and compared with computer vision, it is less disturbed and easier to recognize. The time efficiency of feature extraction using CNN is nearly 30% higher than traditional methods. Therefore, in the practical application of real scenes, with the increase of data volume, the time benefit brought by CNN can be more considerable and our system can be embedded in the cushion and do real-time detection.