A multi-segment modelling approach for foot trajectory estimation using inertial sensors

Background: Kinematic gait analysis employing multi-segment foot models has been mainly conducted in laboratories by means of optical motion capture systems. This type of process requires considerable setup time and is constrained by a limited capture space. A procedure involving the use of multiple inertial measurement units (IMUs) is proposed to overcome these limitations. Research question: This study presents a new approach for the estimation of the trajectories of a multi-segment foot model by means of multiple IMUs. Methods: To test the proposed method, a system consisting of four IMUs attached to the shank, heel, dorsum and toes segments of the foot, was considered. The performance of the proposed method was compared to that of a conventional method using IMUs adopted from the literature. In addition, an optical motion capture system was used as a reference to assess the performance of the implemented methods. Results: Employing the suggested method, all trajectory directions of the shank, heel and dorsum segments, as well as the Z (yaw) direction of the toes segment, have exhibited an error reduction varying between 8% and 55%. However, X (roll) and Y (pitch) direction of the toes segment presented an error increase of 17% and 26%, respectively. The estimation of the vertical displacement, corresponding to the foot clearance, was improved for all segments, resulting in a final mean accuracy and precision of 3.5 +/- 2.8 cm, 2.7 +/- 2.1 cm, 0.8 +/- 0.7 cm and 1.1 +/- 0.9 cm for the shank, heel, dorsum and toes segments, respectively. Significance: It has been demonstrated that as an alternative to tracking each foot segment separately, the fusion of multiple IMU measurements using kinematic equations, considerably improves the estimated trajectories, especially when considering vertical foot displacements. The proposed method could complement the use of smaller and cheaper sensors, while still matching the same performance of other published methods, making the suggested approach very attractive for real life applications.