Unsupervised gait detection using biomechanical restrictions

Abstract

Quantification of human gait with sensors has enormous potential in health and rehabilitation applications. Objective measurement of gait features in the home and community can reveal the true nature of impact of disease on activities of daily living or response to interventions. Previously reported gait event detection methods have achieved good success, yet can produce errors in some irregular gait patterns. In this paper, we propose a novel unsupervised detection of gait events and gait duration by combining two exclusive processes: (i) exploration of gait event candidates based on iterative running of existing methods with changing parameters and, (ii) selection of the candidate which satisfies gait-specific biomechanical restrictions (e.g., when one leg is in swing, another leg is likely to be in stance). We evaluated this approach using data from a single-axis gyroscope on the left and right ankles in three experimental conditions. The proposed method decreased the timing error for detection of gait events (toe off and heel strike) in irregular gait patterns compared with the conventional method. It also improved the accuracy of measurement of gait duration in a longitudinal free-living dataset and distinguishing gait from non-gait actions.

Feature Identification With a Heuristic Algorithm and an Unsupervised Machine Learning Algorithm for Prior Knowledge of Gait Events

“…Developments in algorithmic methods, particularly machine learning, have improved the potential of mobile sensor technology for analysis of gait in real-world environments [8]- [10]. Of particular interest, unsupervised machine learning algorithms for the analysis of multiple time series are being developed and tested for gait event detection [11], [12]. While these algorithms have been shown to be accurate for gait event identification in the laboratory, they have not been tested on data collected from a real-world environment, nor have they been compared to simple heuristic identifiers with minimal sensor data from the same data set.…”

“…Additional options for mobile sensing are inertial sensors, such as accelerometers and gyroscopes. These sensors have been utilized extensively for the identification of gait events and locomotion mode, with varying success [1], [10], [11], [12], [13], [14], [15], [16]. One of the major limitations of accelerometers regards excess vibrations experienced during initial foot contact which may limit consistency of accelerometry data when compared to gyroscopic data.…”

“…In Feature Identification with a Heuristic Algorithm and an Unsupervised Machine Learning Algorithm for Prior Knowledge of Gait Events Seth R. Donahue, and Michael E. Hahn H contrast, unsupervised machine learning algorithms have the potential to input the data with minimal modification by the researcher. Indeed, there have been rare attempts to use unsupervised machine learning algorithms for identification of gait events [11], [12], which have performed at the same accuracy as the supervised and heuristic algorithms. These methods have been focused on determining precisely when a gait event is going to occur.…”