Foot Progression Angle Estimation: An Algorithm for Foot-based Single-Sensor Application

Abstract

Background: Knee Osteoarthritis (KOA) is the predominant form of osteoarthritis, and its incidence is anticipated to rise due to an ageing and increasingly overweight population. Altering the Foot Progression Angle (FPA) has demonstrated potential in mitigating the progression of KOA. Wearable feedback methods, especially those utilising single IMU sensors, have gained prominence for their cost-effectiveness, time efficiency, and potential for real-world application in FPA alteration training. This study collaboratively advances the "Knee Wear project" with the objective of designing and validating an FPA estimating algorithm during straight-line walking using the Xsens DOT sensor.

Methods: This study proposed a calibration method for sensor-to-foot orientation by developing a sensor-to-shoe fixture for consistent sensor alignment with foot direction. While static calibration remained necessary, it eliminated the need for complete sensor fixation to the shoe or dynamic calibration before each use. The fixture's concept was verified by measuring sensor-to-shoe orientation post-fixture attachment to the shoe together with the sensor's fixating capability on the shoe. Additionally, the difference in FPA estimation utilising the dynamic and the fixture calibration method was examined. The FPA algorithm that estimated the FPA during the mid-stance phase, utilised the orientation estimation of the Xsens DOT sensor, in combination with a heading reset to compensate for magnetic disturbances. Two FPA estimation approaches were tested: one utilising the trajectory estimation of the stride and another the peak deceleration point during the late swing phase. Experiments on 13 participants (5 males, 8 females) aimed to validate the algorithm against an optical motion capture system under three walking conditions (natural, toe-in and toe-out walking) and the efficacy of the Xsens DOT in mitigating the effects of magnetic disturbances over prolonged use.

Results: The sensor's alignment to the foot after attachment of the fixture to the shoe provided consistency but had pronounced outliers ranging from -10° to 6,5°, yielding inaccurate FPA estimates utilising this calibration method. No change in the orientation of the fixture was noted during natural walking. However, fixture rotations of 1.1° externally and 0.6° internally were observed in toe-in and toe-out walking respectively. The dynamic calibration method and Peak Deceleration point algorithm yielded the most accurate FPA estimates, with mean absolute error (MAE) of less than 3.5° for all walking conditions. Notably, the FPA algorithm seemed to produce an overshooting FPA estimation, and its consistency decreased for higher FPA values. The FPA estimation over a period of 10 minutes suggested a potential minimal decline in accuracy over time, with an increase in 2,28° RMSE after 10 minutes.

Conclusion: This thesis contributes to the ongoing development of wearable technologies aimed at managing KOA. By the development of an FPA estimation algorithm, evaluating multiple calibration methods, and validating its accuracy and feasibility, this study is a step towards the “Knee Wear project’s” smart-wearable feedback device, simplifying gait retraining interventions for people with KOA. While the proposed fixture calibration method needs refining for foot direction alignment, the proposed FPA algorithm, especially when combined with the dynamic calibration method, provided accurate FPA estimations. While the accuracy of the method is not absolute, the MAE magnitude of less than 3.5° is accurate enough to provide feedback on the FPA. Additional research should be done on the use of the proposed FPA estimating method by the demographic intended for the FPA feedback device.