Introduction
This research project aims to develop advanced algorithms for highly accurate human gait speed estimation using inertial measurement unit (IMU) data, with a specific focus on applications in clinical trials. We will leverage the actibelt platform, a world-leader in IMU devices for clinical trials, renowned for its robustness and reliability. With over two decades of use in clinical trials, actibelt provides an unparalleled foundation for our research.
Our team brings extensive experience in the field of IMU-based human motion analysis and a track record of developing speed estimation algorithms. By combining this expertise with recent advances in inertial odometry and integrating traditional filtering techniques with neural networks, you will seek to push the boundaries of speed estimation accuracy while maintaining strict privacy safeguards.
Key Objectives
1. Create a hybrid approach that fuses traditional filters (e.g., Kalman filters) with deep learning models (such as LSTMs or Transformers) to process IMU data from actibelt devices.
2. Optimise the algorithm to accurately estimate speed across various movement patterns and intensities typical in clinical trial scenarios.
3. Implement privacy-by-design principles to ensure that precise location information cannot be reconstructed from the speed data.
4. Develop techniques for on-device processing and data aggregation that minimise the risk of individual identification.
5. Conduct extensive testing to validate the accuracy of speed estimation against ground truth measurements.
6. Compare the performance of our hybrid approach against current state-of-the-art methods in inertial odometry.
Potential Impact
Accurate speed estimation without location tracking can provide valuable insights into patient mobility, drug efficacy, and disease progression across a wide range of conditions. The landscape of regulatory acceptance for clinical outcomes is evolving, becoming increasingly open to parameters based on daily human motion. This shift is exemplified by the recent acceptance of speed-related parameters for assessing Duchenne muscular dystrophy. Given this changing regulatory environment, now is the perfect time to develop advanced algorithms for human motion analysis in clinical trials. Our solution aims to capitalise on this momentum, offering researchers a powerful tool that combines the robustness of actibelt hardware with cutting-edge algorithmic approaches. This fusion of proven technology and innovative methods has the potential to revolutionise how mobility and functional capacity are assessed in clinical studies, providing more sensitive, objective, and ecologically valid measures of patient outcomes. By delivering highly accurate speed estimations while maintaining the highest standards of data privacy, our project is poised to meet the growing demand for sophisticated, real-world mobility assessments that align with evolving regulatory perspectives.
Expected Outcomes
The scope of the overarching project is extensive, and the final outcomes of the student subproject will be significantly influenced by the student's familiarity with various technologies, their understanding of existing research, and the choices made in collaboration with our team. Given the project's breadth, several potential outcomes are possible. These could include:
- a successful replication of an existing state-of-the-art research paper, adapted specifically for data from actibelt devices;
- an improvement to a current state-of-the-art algorithm that addresses the critical issue of privacy in clinical data processing;
- or an enhanced algorithm that incorporates data from an additional sensor to improve accuracy or robustness.
Regardless of the specific direction, the project aims to contribute meaningfully to the field of walking speed estimation for clinical applications, with outcomes that balance algorithmic innovation, practical applicability, and ethical considerations in data handling.
