ML-2SN: A Hybrid Two-Stream System for Sitting Posture Detection


Kehang Jia (Suzhou University), Gaorui Zhang (Suzhou University), Yixuan Yang (Suzhou University), Guangwei Huang (Suzhou University), Penghuan Wang (Suzhou University), Cheng Cheng (Suzhou University)
The 35th British Machine Vision Conference
PDFPoster

Abstract

Abnormal sitting postures often lead to neck, shoulder, back, and lumbar disorders and are becoming more prevalent across all age groups. Therefore, it is crucial to investigate intelligent monitoring technologies that can accurately recognize sitting postures in real-time. However, most of the previous studies in this field have focused on a single spatial or temporal domain, and these methods do not yield full information. In this paper, we propose a system called ML-2SN. In that system, we use LSTM to capture the latent temporal features in a set of consecutive skeletal key points as the temporal domain features of the model. We use MobileNetV3 to learn the connections among skeletal key points in the black background image of the skeleton as the spatial features of the model. To enhance the model's attention to action changes, we design the Mapping Block to attenuate the influence of spatial features on the model. Additionally, the ResLSTM Block is designed to enhance the effect of temporal features on the model. During model training, we use a novel label smoothing method (Action Label Smoothing) to attenuate the effect of action boundaries on the model. The system improves accuracy by recognizing human joints and filtering out ambient noise, effectively reducing the model's reference time. The experimental results show that the average detection accuracy of the system is 0.8894, which is 0.0192 better than the best existing method.

Citation

@inproceedings{Jia_2024_BMVC,
author    = {Kehang Jia and Gaorui Zhang and Yixuan Yang and Guangwei Huang and Penghuan Wang and Cheng Cheng},
title     = {ML-2SN: A Hybrid Two-Stream System for Sitting Posture Detection},
booktitle = {35th British Machine Vision Conference 2024, {BMVC} 2024, Glasgow, UK, November 25-28, 2024},
publisher = {BMVA},
year      = {2024},
url       = {https://papers.bmvc2024.org/0510.pdf}
}


Copyright © 2024 The British Machine Vision Association and Society for Pattern Recognition
The British Machine Vision Conference is organised by The British Machine Vision Association and Society for Pattern Recognition. The Association is a Company limited by guarantee, No.2543446, and a non-profit-making body, registered in England and Wales as Charity No.1002307 (Registered Office: Dept. of Computer Science, Durham University, South Road, Durham, DH1 3LE, UK).

Imprint | Data Protection