Human gait phase recognition based on thigh movement computed using IMUs

Abstract

Abstract-Human gait analysis is a major topic in pedestrian navigation and geriatric care. Identifying gait phases is important in using human gait for pedestrian navigation and tracking. Most of existing gait phase identification techniques use multiple sensor modules attached to each section of the lower body. This paper discusses the feasibility of recognizing gait phases using a single inertial measurement unit (IMU) placed in a trouser pocket of the subject. The movement of the thigh is computed by fusing accelerometer and the gyroscopic data gathered from the of the IMU. Experimental results indicated that most of the major gait phases such as Initial Contact, Load Response, Mid Stance, Terminal Stance, Pre-Swing and Swing, can be identified by the movement of one thigh tracked by an IMU. It was also noted that the movement of the offside leg can also be estimated from the fused IMU data. This paper presents a method to recognize all major phases of human stride cycle during walking from movement of one thigh.

2-Point Error Estimation Algorithm for 3-D Thigh and Shank Angles Estimation Using IMU

“…It must be noted that the ideas of eqs. (3-4) are taken from [10], but the researchers used these equations to decide whether to use accelerometer or gyroscope data to estimate the flexion/extension angle, while the application of these two equations in this proposed method is to identify the starting and stopping time of walking.…”

“…Ԑ , = , − , for = , = pitch, roll (10) As accelerometer cannot estimate the yaw angle, the initial yaw angle at time A is assumed to be 0º because the person is facing straight initially. Since the heading before and after a person walks straight should ideally remain the same, the final yaw angle at time B is also assumed to be 0º.…”

“…The challenge of estimating angles using IMU is that the inertial sensors are prone to drift, noise and surrounding interference such as temperature [7]. Most researchers have performed normalization or filtration of the sensor data to estimate the angles accurately [8], [10]- [14]. Some researchers tilted the accelerometer to read the reference value of each axes at 1g, i.e.…”

“…The estimated flexion/extension angle had a correlation coefficient of 0.97, and a RMSE of 4.17°. On the other hand, Abhayasinghe and Murray [10] compensated the gyroscope integration drift by updating the angle using accelerometer data when the gyroscope reads small angular velocity and when the accelerometer readings are reliable, i.e. its magnitude is close to 1g.…”

“…The proposed algorithm estimates the errors of the gyroscope angles at the starting and stopping time of a walking motion, and linearly distribute these two errors to compensate the gyroscope drift. The key merit of the proposed algorithm is that it is more computationally efficient than most other methods [10]- [16], while still achieving low average RMSE of 3.6º for pitch, roll, and yaw angles.…”