Читать книгу Pedestrian Inertial Navigation with Self-Contained Aiding - Doruk Senkal, Andrei M. Shkel - Страница 21

Another category of aiding techniques is self‐contained aiding. Instead of fusing external signals into the system, self‐contained aiding takes advantage of the system's patterns of motion to compensate for navigation errors. Therefore, self‐contained aiding techniques vary for different navigation applications due to different dynamics of motions.

For example, in ground vehicle navigation, the wheels can be assumed to be rolling without slipping. Thus, IMU can be mounted on the wheel of the vehicle to take advantage of the rotational motion of the wheel. In this architecture, the velocity of the vehicle can be measured by multiplying the rotation rate of the wheel by the circumference of the tire [28]. In addition, carouseling motion of the IMU provides the system more observability of the IMU errors, especially the error of yaw gyroscope, which is typically nonobservable in most navigation scenarios [29]. Besides, low frequency noise and drift can also be reduced by algorithms taking advantage of the motion of the IMU [30].

Another approach is to take advantage of biomechanical model of human gait instead of just the motion of the foot during walking. This approach typically requires multiple IMUs fixed on different parts of human body and relate the recorded motions of different parts through some known relationships derived from the biomechanical model. In this approach, a more accurate description of the human gait is available, through for example, human activity classification and human gait reconstruction. Recognition of gait pattern can help to reduce the navigation error obtained from a single IMU.

Machine Learning (ML) has also been applied to pedestrian inertial navigation. ML has mostly been explored in the field of Human Activity Recognition (HAR) [31], stride length estimation [32], and stance phase detection [33]. However, few studies used the ML approach to directly solve the pedestrian navigation problem. Commonly used techniques include Decision Trees (DT) [34], Artificial Neural Network (ANN) [35], Convolutional Neural Network (CNN) [36], Support Vector Machine (SVM) [37], and Long Short‐Term Memory (LSTM) [38].

Sensor fusion method can be used in a self‐contained way, where multiple self‐contained sensors are used in a single system, and their readouts are fused in the system to obtain a navigation result. For example, self‐contained ranging technique is one possibility. In this technique, the transmitter sends out a signal (can be ultrasonic wave or electromagnetic wave) which is received by the receiver. This technique can be categorized as self‐contained if both the transmitter and the receiver are within the system whose state is to be estimated. For example, in the foot‐to‐foot ranging, the transmitter and receiver are placed on two feet of a person to keep track of the distance between them [24]. In the cooperative localization, ranging technique is applied to measure the distance between multiple agents as a network to improve the overall navigation accuracy of each of the agent [39].