Gait Phase Recognition from Angular Data Using Machine Learning and Kinovea-Based Motion Capture

Human gait measurement and analysis, particularly for estimating joint range of motion, play an essential role in clinical rehabilitation and sports biomechanics. This study investigates the feasibility of integrating Kinovea-based 2D kinematic analysis with machine learning techniques for automated gait phase recognition. Gait data were collected from 30 healthy young adults (20–25 years, normal body mass index) walking on a treadmill at three controlled speeds (3, 5, and 7 km/h). Joint angular trajectories were extracted using Kinovea and characterized using six time-domain features (variance, root mean square, multivariate alteration detection, waveform length, average amplitude change, and difference absolute standard deviation value). Four machine learning classifiers – decision tree, random forest, k -nearest neighbors, and support vector machine (SVM) – were evaluated. Subject-independent validation was employed to assess model generalization across participants. The results indicate that initial contact was consistently classified with high accuracy, while transitional phases, particularly toe-off, exhibited lower classification performance due to overlapping kinematic patterns. Among the evaluated models, SVM achieved the highest overall accuracy (93.7%), demonstrating superior robustness for non-linear angular data. These findings suggest that the proposed approach represents a feasible low-cost framework for gait phase recognition using 2D video-based motion capture, while also highlighting the need for enhanced temporal features to improve classification of transitional gait events.