Machine Learning Based Adaptive Gait Phase Estimation Using Inertial Measurement Sensors

This paper presents a portable inertial measurement unit (IMU)-based motion sensing system and proposed an adaptive gait phase detection approach for non-steady state walking and multiple activities (walking, running, stair ascent, stair descent, squat) monitoring. The algorithm aims to overcome the limitation of existing gait detection methods that are timedomain thresholding based for steady-state motion and are not versatile to detect gait during different activities or different gait patterns of the same activity. The portable sensing suit is composed of three IMU sensors (wearable sensors for gait phase detection) and two footswitches (ground truth measurement and not needed for gait detection of the proposed algorithm). The acceleration, angular velocity, Euler angle, resultant acceleration, and resultant angular velocity from three IMUs are used as the input training data and the data of two footswitches used as the training label data (single support, double support, swing phase). Three methods 1) Logistic Regression (LR), 2) Random Forest Classifier (RF), and 3) Artificial Neural Network (NN) are used to build the gait phase detection models. The result shows our proposed gait phase detection with Random Forest Classifier can achieve 98.94% accuracy in walking, 98.45% in running, 99.15% in stair-ascent, 99.00% in stair-descent, and 99.63% in squatting. It demonstrates that our sensing suit can not only detect the gait status in any transient state but also generalize to multiple activities. Therefore, it can be implemented in real-time monitoring of human gait and control of assistive devices. INTRODUCTION In the last two decades, wearable devices, exoskeletons, and rehabilitation robots emerge as a new approach to prevent injuries and augment human capabilities [1]. In those applications, to determine the gait phase is crucial to monitor gait patterns, generate assistive torque profile, position profiles, and prevent injuries. Prior work studied foot switches [2], gyroscope [3], accelerometer [4], electromyography (EMG) [5] to detect gait cycles and used the information to trigger the assistance of wearable devices to augment the human walking. Typically, the movements of a limb in a gait cycle can be divided into 1) initial contact, 2) loading response, 3) mid-stance, 4) terminal stance, 5) pre-swing, 6) initial swing, 7) mid-swing, and 8) terminal swing [6]. Among those phases of the gait cycle, the most important events to separate the gait phase are heel-strike and toe-off [7]. Researchers have explored the foot pressure sensing system [8-10] and optical motion capture system [11-14] to obtain accurate gait cycle detection. Although these systems provide accurate gait detection, optical motion capture systems are heavy and not portable. Pappas et al. used the angular velocity of the foot and three force sensitive resistors to detect stance, heel-off, swing, and heel-strike in real-time and they achieve 99% accurate in climbing and walking in non-steady state walking and 96% for the subjects with impaired gait [10]. A foot pressure sensing system is not easily integrated into the wearable robots or may be unreliable due to its resistive sensing nature. Therefore, an IMU-based gait detection algorithm has been developed [15]. This algorithm uses the IMU sensor mounted on the foot and heuristic threshold to detect the heelstrike and toe-off event, then derive the gait cycle from 0 to 100% in time series with good accuracy in a steady state walking. But, if the subject suddenly changes their walking speed, stops walking, or changes activities, the gait cycle detection will become inaccurate. The novelty of our algorithm overcomes the limitation of existing gait detection methods that are time-domain thresholding based for steady-state motion and are not versatile to detect gait during different activities or different gait patterns of the same activity. We developed a portable IMU-based motion sensing system and proposed a novel gait phase detection approach for non-steady state walking and multiple activities (walking, running, stair-ascent, stair-descent, squatting). The proposed approach not only detects the gait status in any transient state but also generalizes to multiple activities across different users without retraining the new user’s data. Therefore,