Page 143 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

2�2 Benefits of use case

This system aims to reduce premature mortality from non-communicable diseases through
prevention and treatment. By enabling real-time exercise assessment and adaptive
rehabilitation, the system helps individuals recover from musculoskeletal disorders, post- 4.1-Healthcare
stroke conditions, and post-surgical rehabilitation. It ensures users perform exercises correctly,
reducing complications and improving health outcomes.

Traditional rehabilitation services are often inaccessible to rural populations and economically
disadvantaged individuals due to high costs and limited healthcare infrastructure. By leveraging
AI and automation, this system lowers the dependency on physical rehabilitation centres,
making quality rehabilitation and exercise guidance available remotely through televisions,
desktops, and mobile devices. This ensures that individuals, regardless of socioeconomic status
or geographical location, can receive equitable healthcare support. By making rehabilitation
cost-effective, scalable, and widely accessible, the system contributes to closing the health
disparity gap and fostering inclusivity in healthcare

2�3 Future Work
• To overcome occlusion issues, future iterations will explore the integration of 3D sensing
technologies, such as Microsoft Kinect. This will enable more accurate differentiation of
complex movements (e.g., forward lean vs. backwards bend) and subtle movements that
rely heavily on the trajectory rather than joint angles.
• Support for regional languages will also be extended, enhancing accessibility for non-
English-speaking users.

3 Use Case Requirements

• REQ 1: It is critical that the system is capable of tracking and analysing human body
movements in real time using a reliable pose estimation pipeline.
• REQ 2: It is critical that the system provides instant corrective feedback (audio and visual)
by comparing user movement with reference data and ensures low-latency performance
using optimised inference libraries such as XNNPACK.
• REQ 3: It is critical that the system is capable of utilizing an STGCN (Spatio-Temporal
Graph Convolutional Network) model to assess exercise correctness accurately.
• REQ 4: It is expected that the system is maintained in a cloud-based repository of
reference joint angles and movement patterns for comparison and dynamic updating.
• REQ 5: It is expected that the system will adapt feedback based on historical user
performance, age, and progress trends to personalize the rehabilitation experience.
• REQ 6: It is expected that the system will adapt to new exercises by identifying consistent
variations in joint angles and movement lengths over time.

4 Sequence Diagram

The sequence diagram in Figure 3 begins with the user performing an exercise after selecting
the type of exercise. The data is passed through MediaPipe to extract poses, and then the
reference data is compared with this to provide audio and visual cues. After completion of the
exercise, the data is sent to the cloud, where the STGCN model gives the assessment score
for the exercise. This performance report is sent to the user.

107