Applied Machine Learning on Phase of Gait Classification and Joint-Moment Regression

Abstract

Traditionally, monitoring biomechanics parameters requires a significant amount of sensors to track exercises such as gait. Both research and clinical studies have relied on intricate motion capture studios to yield precise measurements of movement. We propose a method that captures motion independently of optical hardware with the specific goal of identifying the phases of gait using joint angle measurement approaches like IMU (inertial measurement units) sensors. We are proposing a machine learning approach to progressively reduce the feature number (joint angles) required to classify the phases of gait without a significant drop in accuracy. We found that reducing the feature number from six (every joint used) to three reduces the mean classification accuracy by only 4.04%, while reducing the feature number from three to two drops mean classification accuracy by 7.46%. We extended gait phase classification by using the biomechanics simulation package, OpenSim, to generalize a set of required maximum joint moments to transition between phases. We believe this method could be used for applications other than monitoring the phases of gait with direct application to medical and assistive technology fields.

Machine Learning for Optical Motion Capture-Driven Musculoskeletal Modelling from Inertial Motion Capture Data

“…Regardless, this 'field ready' option of IMC systems can be immensely valuable in low-resource settings and outdoor or real-world applications (such as sports and ergonomics). Wearable inertial sensors have evolved rapidly and are routinely used in different areas of clinical human movement analysis, e.g., gait analysis [9], instrumented clinical tests, stabilometry, daily-life activity monitoring, upper-extremity mobility assessment, and tremor assessment [10]. Such sensors have rapidly transitioned from use in constrained, lab-based practice to unsupervised and applied settings [11].…”