Validity Of An Image-based Posture Recognition System For Monitoring Daily Physical Activity

Smart home care environments and monitoring systems are gaining increased attention. It is essential to monitor the activity status, detect potential dangers, and provide early warnings for individuals living independently, especially the elderly or patients. Observing and identifying posture and movement behavior can provide insights into their activities. Typically, posture estimations rely on data acquired from conventional motion capturing system. However, many existing systems are often impractical for everyday use due to the complexity of setup and the need to attach special devices to people’s bodies. Recently, advanced techniques such as pose estimation algorithms have shown effectiveness in real-time assessment of human movement. Leveraging advancements in computer vision, these algorithms can automatically track human motion based on simple images captured by common household devices equipped with relatively inexpensive cameras. PURPOSE: An image-based posture recognition method, capable of identifying human body landmark locations using only a single depth sensor, was introduced in a previous study. This study aims to examine the validity of this method for posture recognition during daily activities. METHODS: Twenty participants were recruited to perform two postures (upright standing and sitting) in front of a depth sensor within a laboratory setting. Simultaneously, a marker-based optical motion tracking system, as a reference system, recorded twenty anatomical body landmark locations. For each landmark, differences between the 3D landmark localizations generated by the image-based posture recognition method and their counterparts identified by the reference system were determined through the calculation of Euclidean distance. Across both tested postures, these differences in regard to those landmarks were utilized to quantify the tracking error. RESULTS: This study found that the average tracking errors were 6.65 ± 2.14 cm for the upright standing posture, and 6.58 ± 2.13 cm for the sitting posture, respectively. CONCLUSIONS: The image-based posture recognition system offers a potential alternative, enabling continuous monitoring of an individual’s physical activity at home without interrupting their daily routines.