Читать книгу Pedestrian Inertial Navigation with Self-Contained Aiding - Doruk Senkal, Andrei M. Shkel - Страница 14

There are two general approaches in the pedestrian inertial navigation. One is the strapdown inertial navigation as introduced in Section 1.2, where IMU readouts are integrated into position and orientation. This approach is universally applicable, but the integral step makes the algorithm computationally expensive and the navigation error accumulates as time cubed due to the gyroscope bias. In order to limit the error propagation, the most commonly used method is to apply the Zero‐Velocity Updates (ZUPTs) when the velocity of the foot is close to zero (the foot is stationary on the ground) [12]. The stationary state can be used to limit the long‐term velocity and angular rate drift, thus greatly reduce the navigation error. In this implementation, IMU is fixed on the foot to perform the navigation and to detect the stance phase at the same time. Whenever the stance phase is detected, the zero‐velocity information of the foot is fed into the Extended Kalman Filter (EKF) as a pseudo‐measurement to compensate for IMU biases, thus reducing the navigation error growth in the system. In this architecture, not only the navigation errors but also IMU errors can be estimated by the EKF. The limitation of this approach is that the IMU needs to be mounted on the foot.

In order to avoid the integral step in the pedestrian inertial navigation and also relax the requirement of IMU mounting position, a Step‐and‐Heading System (SHS) is an alternative. It is composed of three main parts: step detection, step length estimation, and step heading angle estimation [13]. Unlike the first approach, this approach can only be applied in the pedestrian inertial navigation. In this approach, the step length of each stride is first estimated based on some features of motion obtained from the IMU readouts. Methods based on biomechanical models and statistical regression methods are popular for the estimation. Some commonly used features include the gait frequency, magnitude of angular rate, vertical acceleration, and variance of angular rate. Then, the heading angle is estimated by the gyroscope readout, which is typically mounted at the head. This step can also be aided by magnetometers to improve the accuracy. In this way, the total displacement can be estimated combining the traveled distance and the heading angle. However, two major challenges exist for this approach. First, the gazing direction needs to be aligned with the traveling direction, implying that the subject needs to look at the traveling direction all the time, which is not practical. Second, the step length estimation remains difficult. The average value of the estimated step length may be accurate when median value generally less than 2%, but the estimate precision is generally low, with the Root Mean Square Error (RMSE) about 5% [14]. With a wide adaption of hand‐held and fitness devices, this is currently an active area of research.