Attention-enhanced graph convolutional network for assessing rehabilitation exercises

Physical rehabilitation exercises are crucial in the post-operative recovery and treatment of various musculoskeletal conditions. An automated vision-based rehabilitation exercise assessment is a portable and cost-effective way of evaluating the patients' performance in a home-based setting and predicting a performance score by analyzing the correct and incorrect exercise sequences performed by the patient. Recent works have shown that exploring spatial and temporal features of the skeleton data is vital for this task. However, most of the methods treat all body joints equally and fail to capture the correlation information between all joints. Hence, to address this limitation, we have proposed an attention-enhanced spatial-temporal graph convolutional network that captures the spatial and temporal dependencies among the body joints incorporating an attention mechanism to learn global information for the joint-specific roles to provide better assessment results.