A Lightweight IoT Device-friendly Anomaly Sitting Posture Detector for Protecting Adolescent Bone Development

In recent years, people have tended to spend a lot of time sitting while unconsciously having bad posture. Research has shown that bad sitting poses, for instance, sitting cross-legged, could lead to serious spinal disorders, including thoracic spine tilting and pelvic posterior tilting. Correspondingly, correcting bad sitting posture is essential to prevent people, especially adolescents, from those problems. Inspired by the recent prevalence of Internet-of-Things (IoT) health monitoring systems, we proposed a lightweight vision-based bad sitting pose detector, which can be applied to IoT devices with a camera on. This method consists of two main parts: 1) generating human pose landmarks of the subjects and 2) integrating pose landmarks with shoe detection data into a pose classifier. To verify the reliability of our method, we also created a dataset containing a total of 5042 labeled sitting postures. The proposed method achieves high accuracy in detecting bad sitting postures. Besides, adding the shoe’s position as a supplementary attribute was found to be effective in detecting lower body bad poses. This paper demonstrates the feasibility of applying a pose detection model on IoT devices to effectively correct bad sitting postures in daily lives.