A Novel Multi-Class State Detection Algorithm
Skip to main content
eScholarship
Open Access Publications from the University of California

UC Irvine

UC Irvine Electronic Theses and Dissertations bannerUC Irvine

A Novel Multi-Class State Detection Algorithm

Abstract

One popular example of state detection and correction is the Zero Velocity Update (ZUPT), which is often used in pedestrian navigation to limit error due to sensor noise and biases. Existing methods, while they have become both efficient and effective, are often either computationally heavy or are not designed to handle the detection of a variety of states.This study proposes an expression to be used in a state detection algorithm, produced in part by the information provided by machine learning models. The intent for the state detection algorithm is the classification of inertial sensor data into various important trajectory states. In this study, the goal was to distinguish zero-velocity states and left- and right-handed orbital trajectories out of sets of IMU data using the simplest effective methods. First, some existing methods are evaluated in their utility as candidates as a singular data feature of a classifier for the expected navigational trajectories. None of the methods examined were suitable candidates for implementing a piecewise classifier for the trajectory states. In turn, machine learning techniques are employed to identify important data features for the proposed classifier. To assist in the implementation of a method using these key data features, a novel test statistic for a likelihood test similar to the Stance Hypothesis Optimal Detection method is proposed. Finally, the proposed classifier is tested using collected IMU data, and the potential of incorporating such a state detector is explored. The results show that the novel state detector is able to classify the expected trajectory states with over 95% accuracy. In the data used to test the sensor, it did not experience any false classifications or missed detections. While there is still work to be done to improve the robustness and applicability of the algorithm, this study acts as a proof of concept, providing a starting point for future work.

Main Content
For improved accessibility of PDF content, download the file to your device.
Current View