Selected peer-reviewed publications from doctoral research at UCD, working on a wearable motion capture system for post-stroke rehabilitation. Each paper solves one specific failure mode that would prevent the system from working in ambulatory settings.
The founding paper of the research programme. Proposes the complete system architecture: a wearable 6-DOF motion capture unit combining IMU and ultrasonic sensors, an automated exercise assessment module using scale-invariant state space comparison, and a real-time audio-visual feedback loop. Clinical collaboration with the Stroke Rehabilitation Unit at Baggot Street Community Hospital, Dublin 4. Validates position accuracy below 5 mm and orientation error below 1.5 degrees in simulation for forearm tracking.
When the ultrasonic signal is blocked, the system must estimate position from IMU dead reckoning alone. This paper improves that fallback by placing two IMUs on the same rigid body and using the fixed known separation between them as a geometric constraint inside an EKF. A first stage estimates orientation from gyroscope and accelerometer. A second stage uses the rotation-projected inter-sensor separation vector as an observation, tightening the position estimate. Around 30% improvement over standard dead reckoning across circular, linear and rest motion profiles.
Not all occlusion events are total. During arm movements, one or two ultrasonic ranges may remain available even when full trilateration is impossible. This paper introduces a hybrid filter that selectively uses an EKF (single hypothesis) when ranges are sufficient and a particle filter with Gaussian mixture likelihoods (multi-hypothesis) when they are not. The Gaussian mixture correctly models the ring-shaped uncertainty from partial ranging, unlike a single Gaussian. Up to 10% improvement over SCAAT and 24% improvement over EKF in the hardest occlusion scenarios.