This paper tackles the challenge of automatically assessing physical
rehabilitation exercises for patients who perform the exercises without
clinician supervision. The objective is to provide a quality score to ensure
correct performance and achieve desired results. To achieve this goal, a new
graph-based model, the Dense Spatio-Temporal Graph Conv-GRU Network with
Transformer, is introduced. This model combines a modified version of STGCN and
transformer architectures for efficient handling of spatio-temporal data. The
key idea is to consider skeleton data respecting its non-linear structure as a
graph and detecting joints playing the main role in each rehabilitation
exercise. Dense connections and GRU mechanisms are used to rapidly process
large 3D skeleton inputs and effectively model temporal dynamics. The
transformer encoder’s attention mechanism focuses on relevant parts of the
input sequence, making it useful for evaluating rehabilitation exercises. The
evaluation of our proposed approach on the KIMORE and UI-PRMD datasets
highlighted its potential, surpassing state-of-the-art methods in terms of
accuracy and computational time. This resulted in faster and more accurate
learning and assessment of rehabilitation exercises. Additionally, our model
provides valuable feedback through qualitative illustrations, effectively
highlighting the significance of joints in specific exercises.
This paper presents a novel approach to automatically assessing physical rehabilitation exercises for patients who perform the exercises without clinician supervision. The goal is to provide a quality score to ensure correct performance and achieve desired results. The authors introduce a new graph-based model, called the Dense Spatio-Temporal Graph Conv-GRU Network with Transformer, which combines modified versions of the STGCN and transformer architectures to handle spatio-temporal data efficiently.
The multi-disciplinary nature of this approach is worth highlighting. By treating the skeleton data as a graph and incorporating graph convolutional networks, the model integrates concepts from graph theory and deep learning. Furthermore, the use of the GRU mechanism and dense connections enables rapid processing of large 3D skeleton inputs, while effectively modeling temporal dynamics. The transformer encoder’s attention mechanism focuses on relevant parts of the input sequence, making it valuable for evaluating rehabilitation exercises.
The evaluation of the proposed approach on the KIMORE and UI-PRMD datasets demonstrates its potential and superiority over state-of-the-art methods in terms of accuracy and computational time. This is significant as it enables faster and more accurate learning and assessment of rehabilitation exercises, ultimately leading to improved patient outcomes.
One notable aspect of this model is its ability to provide valuable feedback through qualitative illustrations. By effectively highlighting the significance of joints in specific exercises, patients can better understand their performance and make necessary adjustments. This feedback mechanism adds an additional level of personalization and engagement to the rehabilitation process, enhancing patient motivation and adherence to the prescribed exercises.
In conclusion, the Dense Spatio-Temporal Graph Conv-GRU Network with Transformer presents a promising solution for automatically assessing physical rehabilitation exercises. By leveraging the concepts from multiple disciplines such as graph theory, deep learning, and attention mechanisms, this model offers a comprehensive approach to evaluating exercise performance. The demonstrated improvements in accuracy, computational time, and feedback mechanism make it a valuable tool for both patients and clinicians in the field of physical rehabilitation.