Elevator, Escalator or Neither? Estimating Pedestrian Conveyor State Using Inertial Navigation System

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering

Final Year Thesis Oral Defense

Title: "Elevator, Escalator or Neither? Estimating Pedestrian Conveyor State 
Using Inertial Navigation System"

by

XIA Zhiqiu

Abstract:

To determine if a phone user is taking an elevator, escalator, or neither, we 
study to estimate the pedestrian conveyor state using inertial navigation 
system (INS) in terms of accelerometer, gyroscope and magnetometer. Despite 
abundant studies on INS-based human action recognition, they cannot extend to 
this problem because human action is orthogonal to the pedestrian's conveyor 
state and can cause disturbances in INS signals. To overcome it, we propose 
ELESON, which utilizes a motion feature extractor to decouple pedestrian 
conveyor state and human action in feature space and a magnetic feature 
extractor to enhance the estimation accuracy using the temporal variation 
feature of magnetic fields. Additionally, it introduces an evidential state 
classifier to classify conveyor state and reflect the level of classification 
errors. Based on extensive experiments on twenty hours of data, ELESON has 
outperformed state-of-the-art approaches by significantly reducing 37% 
estimation errors and improving 13% accuracy of uncertainty measurement with a 
minimal computational cost on smartphones.


Date            : 3 May 2024 (Friday)

Time            : 11:15 - 11:55

Venue           : Room 2611 (near lifts 31/32), HKUST

Advisor         : Prof. CHAN Gary Shueng-Han

2nd Reader      : Dr. MAK Brian Kan-Wing