Gait Classification With Gait Inherent Attribute Identification From Ankle's Kinematics.

The human ankle joint interacts with the environment during ambulation to provide mobility and maintain stability. This association changes depending on the different gait patterns of day-to-day life. In this study, we investigated this interaction and extracted kinematic information to classify human walking mode into upstairs, downstairs, treadmill, overground and stationary in real-time using a single-DoF IMU axis. The proposed algorithm's uniqueness is twofold - it encompasses components of the ankle's biomechanics and subject-specificity through the extraction of inherent walking attributes and user calibration. The performance analysis with forty healthy participants (mean age: 26.8 ± 5.6 years yielded an accuracy of 89.57% and 87.55% in the left and right sensors, respectively. The study, also, portrays the implementation of heuristics to combine predictions from sensors at both feet to yield a single conclusive decision with better performance measures. The simplicity yet reliability of the algorithm in healthy participants and the observation of inherent multimodal walking features, similar to young adults, in elderly participants through a case study, demonstrate our proposed algorithm's potential as a high-level automatic switching framework in robotic gait interventions for multimodal walking.