Skeleton-based Motion Analysis and Nursing Care Posture Assessment Using Spatial Temporal Graph Convolutional Networks

時空間グラフ畳み込みネットワークを用いた骨格ベースの動作解析と介護姿勢評価

Skeleton-based Motion Analysis and Nursing Care Posture Assessment Using Spatial Temporal Graph Convolutional Networks

Degree Grantors Yamaguchi University

As the population ages, the demand for elderly care services will continue to increase, which includes providing specialized care, daily life support, and medical health services. As a result, informal caregiving provided by non-professionals such as family, friends, neighbors, and volunteers is becoming more prevalent. Injuries that occur during caregiving can affect the caregiving’s life, especially their mental and physical health. Therefore, the correct positioning and posture during caregiving are crucial to prevent musculoskeletal disorders among caregivers. Although training programs are useful to reduce the risk of musculoskeletal disorders for informal caregivers, many of them express that it is still difficult for them to grasp the correct caregiving postures. Moreover, they struggle to obtain professional advice to correct their posture through long-term practice. Therefore, finding a targeted ergonomic posture risk assessment and guidance method is crucial to improve caregivers' posture-related risks, enhance work efficiency, and safeguard their physical health.
Rapid Entire Body Assessment (REBA) is a postural risk assessment method based on ergonomics that has been attracting attention recently, and it basically evaluates the risk from the angle of each joint of the body. However, in caregiving movements, the way of load placed on the caregiver and the time to maintain the movements vary greatly depending on the weight and posture of the cared person, so the current risk assessment using REBA is insufficient for caregiving movements. Additionally, posture recognition algorithms such as OpenPose are often used to extract skeletons. With these techniques, problems such as missing skeletons or misrecognition often occur due to image conditions or the overlapping of multiple people, and skeleton extraction may sometimes fail.
In this research, the Spatial Temporal Graph Convolution Network (ST-GCN) is applied to develop a technique for complementing missing skeletons based on behavioral features and a technique for correcting skeletons that are misrecognized due to overlapping people, and to improve the accuracy of calculating skeletal joint angles. In order to evaluate caregiving posture risk more appropriately, some parameters such as center of gravity trajectory, load duration, asymmetric load during caregiving movements are investigated and a new REBA method is proposed.
This paper consists of six chapters.
In Chapter 2, to solve the problems of skeleton misidentification and missing information by OpenPose an improved skeleton reconstruction method based on ST-GCN is propose. The method compensates for missing skeletons in terms of behavioral features and corrects incorrectly identified skeletons based on skeleton weight features. This approach improves the accuracy and robustness of pose recognition and allows more accurate estimation of skeletal joint angles and its REBA score.
In Chapter 3, to address the issue of REBA evaluation scores being too high for caregiving scenarios, a postural risk assessment method (C-REBA) is proposed by considering the characteristics of caregiving task. Customize the traditional REBA method and add parameters such as center of gravity trajectory, load duration, and asymmetric loading to the evaluation score. the caregiving movements to assist in transferring from a bed to a wheelchair on a group of experienced nurses and a group of inexperienced caregivers are analyzed and the effectiveness of the C-REBA method is verified.
In Chapter 4, a method that combines the ST-GCN framework and C-REBA for postural risk assessment is proposed. The deep neural network algorism is applied to learn motion features and additional features such as load duration, motion frequency, center of gravity variation, and asymmetric load. So that all evaluation parameters for C-REBA rules can be obtained automatically. With this method, postural risk assessment processes in caregiving operations can be performed automatically.
In Chapter 5, "Behavior Analysis and Posture Assessment System" (BAPAS) is developed. BAPAS is a system aimed at assessing the risk of musculoskeletal disorders related to working postures in medical support work. This chapter introduces the functions and usefulness of this system and demonstrates how this system can be extended to other medical fields easily by setting parameter is settings.
Chapter 6 provides a summary of the paper as a whole and future prospect.

Creators Han Xin