Study Of Algorithms For Inertial Pedestrian Navigation System

In environments where GPS or beacon signals are degraded or unavailable, tracking and locating an object is a challenging issue. As self-contained solutions, inertial pedestrian navi-gation systems (IPNS) have been developed for this purpose, which have the capability to achieve stable, non-radiating and anti-jamming navigation performance. Such systems can work in arbitrary unfamiliar and unprepared, indoor and outdoor environments, and have been attracting increasing interests in the field of pedestrian navigation. However, the strapdown inertial navigation systems (SINS) feature a rapid error growth when no aiding measurements are available, thus it is necessary to provide an effective means for bounding the navigation errors, so as to improve the accuracy of long-term navigation. Based on a small-size, low-cost inertial measurement unit (IMU) fabricated by micro-electro-mechanical systems (MEMS) technology, the study of algorithms has been done for applying the zero velocity updates (ZUPT)-aided SINS to pedestrian navigation, by taking advantage of the periodic nature of foot motion. The main contents of this paper are as follows:1. The initial alignment of SINS is studied. Initial alignment is one of the key technologies of SINS, the main purpose of which is to determine the initial conditions of the system. In this paper, Euler platform error angles (EPEA) are adopted to describe the misalignment angles from the theoretical navigation coordinate system to the computational navigation coordinate system, based on which the linear and nonlinear error models of SINS are derived for different initial misalignment angles, so as to establish the theoretical basis for the error correction methods proposed in this paper.2. The gait detection with dynamic thresholds is studied based on a clustering algorithm. The existing detection methods either do not examine the false or interrupted gait phases at all, or directly filter out all the short gait phases indiscriminately, which makes the detection results sensitive to measurement fluctuations and detection parameters, thereby affecting the validity and efficiency of the ZUPT technique. By analyzing the roles of the detection parameters and the relationship between them, an adaptive gait detection method is proposed based on an ef-ficient clustering algorithm. The proposed detection method is a modified flat-zone detection method, which is supposed to overcome the limitations of the existing detection methods and achieve optimal detection results over a wide parameter space, so as to improve the accuracy and reliability of the detection results.3. The estimation of navigation errors is studied based on a Kalman-type filter. The Kalman-type filter can achieve additional benefit of estimating other navigation errors that are correlated with the velocity error, such as position error and attitude error. In order to reduce the computational cost and higher-order truncation error, as well as avoid the extra modeling error and restrain the filtering divergence, the Kalman filter is implemented as an indirect filter by decomposing the original navigation equations into two subequations, and the full error model can be simplified by ignoring some physical effects, considering the specific application of pedestrian navigation. By analyzing the quality of the stance-based zero-velocity measure-ments, no sensor measurement error is estimated during the filtering process.4. The estimation of navigation errors is studied based on a fixed-interval smoother. During the ZUPT period, the Kalman-type filter can only estimate the navigation errors that are present in the stance phase, which can lead to abrupt state changes at the transition from swing to stance. The fixed-interval smoother can estimate the navigation errors that are propagated over the whole gait cycle, and thus helps to realize smooth transition between swing and stance, so as to improve the accuracy and stability of the system. In order to reduce the modeling error and computational cost, no sensor measurement error and no position error are estimated during the filtering and smoothing processes, under the condition of not degrading the system per-formance. The navigation errors are estimated, smoothed and corrected in a stepwise manner, and thereby near real-time performance can be achieved for online applications.