This paper introduces a multidisciplinary system for home-based stroke rehabilitation. The proposed system uses a wearable 6-DOF motion capture unit combining inertial measurement units and ultrasonic acoustic sensors to capture 3D position and orientation during exercise. It performs automated assessment of exercise quality and delivers real-time corrective audio-visual feedback, enabling task-specific motor relearning without continuous therapist supervision.
Stroke Rehabilitation in Ireland and the Technology Gap
Stroke is the main cause of adult acquired disability worldwide. In Ireland at the time of this work, over 30,000 survivors were living with residual disability. Of those, 48% had hemiparesis and 22% could not walk independently. A national audit had recently found that stroke services were overburdened and average inpatient stays were being cut short, pushing the need for effective home rehabilitation.
The clinical collaborator on this paper, Olive Lennon, worked at the Stroke Rehabilitation Unit at Baggot Street Community Hospital in Dublin 4. That grounding in clinical reality shaped what the system needed to do. It could not require a lab environment. It had to be wearable and ambulatory. It needed to adapt to patients with very different levels of impairment. And it had to be safe and understandable enough for a patient to use at home without a physiotherapist in the room.
Existing commercial options at the time were either too expensive, required optical cameras with line-of-sight constraints, or relied on magnetometers that performed poorly near the ferrous building materials common in Irish domestic environments.
Four Functional Units Working Together
Wearable unit per body segment: one IMU plus one ultrasonic receiver. IMU provides dead reckoning; ultrasonic provides absolute position when line of sight exists. Together: 6-DOF capture of 3D position and orientation.
Clinician defines exercise programme including repetition counts, difficulty levels and target motion templates. System stores a state-space model of the correctly performed exercise for comparison.
Kinematic parameters (joint angles, limb positions) are extracted in real time and compared against the stored template using a scale-invariant state space model.
Audio beep increases in tone as the limb deviates from the target trajectory. Animation highlights the specific body part responsible for the deviation. Patient adjusts in next trial.
How Position and Orientation Are Captured
The sensor placement follows a 15-segment Hanavan body model. A belt around the waist carries three ultrasonic transmitters and inertial units, forming the fixed global frame of reference. Sensor units on limbs each contain an IMU and an ultrasonic receiver.
Calibration
Limb lengths measured and sensors calibrated for axis alignment relative to limb local frame. Orientation of inter-sensor vector computed and stored.
Ultrasonic Ranging
FHSS signals from waist transmitters allow time-of-flight measurement to each limb receiver. Frequency hopping reduces multipath and reverberation errors. Range accuracy below 2 mm.
3D Position from Ranges
When all three transmitters have line of sight, trilateration gives 3D coordinates of each sensor board relative to the belt frame.
Orientation via Inverse Kinematics
Known sensor positions on limbs and known segment lengths allow Euler angles (roll, pitch, yaw) to be computed via the rotation matrix that maps sensor positions to limb axes.
Occlusion Fallback
During non-line-of-sight periods, accelerometer data is double-integrated from the last known position. Effective for short occlusions; errors grow for longer gaps.
Simulation Accuracy for Forearm Tracking
Elbow flexion-extension and forearm pronation-supination were simulated in MATLAB for 1000 different positions and orientations. A sensitivity analysis included 10 degrees of error in sensor placement.
The system shows the expected degradation with sensor placement error, but remains clinically useful. For rehabilitation exercises, an error of five degrees in orientation is considered acceptable, and the results are well within that for the large majority of positions tested.
Technology for Clinical Access
This is the foundational paper of the PhD research programme. It defines why the technical work matters. Every subsequent paper on Kalman filtering, sensor fusion and magnetic distortion compensation exists to make this system reliable enough to work in a patient's home, not just in a UCD laboratory.
The clinical insight from Baggot Street gave the engineering a target. A system accurate to 5 mm and 1.5 degrees, running without GPS or an optical lab setup, on hardware small enough to wear, with feedback a patient can act on in real time. That specification drove the choice of ultrasonic ranging over other technologies, the focus on IMU dead reckoning quality during occlusion, and the need for robust orientation estimation in magnetically distorted indoor environments.
Sit-to-stand: the target exercise
The system was designed around the sit-to-stand exercise. This movement is critical to functional independence after stroke. It requires coordinated timing between hip and knee extension, and existing literature shows that temporal characteristics of these angular displacements are clinically meaningful indicators of recovery. The system extracts exactly these features and compares them to a template captured under therapist supervision.
The clinical trial for sit-to-stand training was planned as future work. The motion tracking accuracy reported in simulation was intended to be validated with actual patients in collaboration with the Baggot Street team.
UCD Research Repository: hdl.handle.net/10197/3861