U-Turn Detection during Walking

Instrumented gait analysis is usually focused on the analysis of human locomotion along rectilinear trajectories. However, in the last years, quantitative analysis of curvilinear walking has found a great interest in different research areas, such as gait analysis, motor rehabilitation monitoring, and pedestrian mobility. Expert operators often manually perform the identification of turning/pivoting walking by looking at the video recording of walking tasks. However, this procedure is time-consuming and highly affected by intra- and inter-operator variability. This contribution aims at introducing and validating a k-means clustering approach for the automatic detection of the curvilinear trajectories based on gait features extracted from knee-joint kinematics and foot-floor contact signals. More specifically, two different k-means clustering approaches have been tested and compared against a common ground truth obtained by means of manual segmentations: (a) a single k- means classifier applied to gait features extracted from both left and right lower leg, and (b) two k-means classifiers applied to gait features extracted from left and right lower leg separately. Results revealed excellent performances of the tested approaches (Accuracy > 97.0
^%
, Precision > 95.2
^%
, Recall > 99.6%, and F1-score > 97.3%), suggesting that the k-means clustering approach can be successfully applied to all those applications requiring accurate and precise identification of U-turns.