Inertial sensor-based step length estimation model

Vezočnik, Melanija (author), Jurič, Matjaž Branko (mentor)

Pedestrian dead reckoning (PDR) using inertial sensors has paved the way for developing several approaches to step length estimation. In particular, emerging step length estimation models are readily available to be utilized on smartphones, yet they are seldom formulated considering the kinematics of the human body during walking in combination with measured step lengths. Besides, there is an absence of performance evaluation protocols when dealing with the analysis and comparison of the models. The main scientific contributions of this doctoral dissertation encompass a new approach to performance evaluation of the models and two improved inertial-sensor-based step length estimation models that estimated step length more accurately than related models selected for comparison. Both models were designed considering measured stride lengths and kinematics of the human body during walking. We present two step length estimation models herein. The first model utilizes acceleration magnitude as the predominant input, whereas the second model also includes step frequency. To the best of our knowledge, we were the first to employ principal component analysis and canonical correlation analysis to characterize the acquired experimental data that included spatial positions of anatomical landmarks on the human body during walking, tracked by an optical measurement system. We evaluated the performance of the proposed models for four common smartphone positions and walking on a treadmill and a rectangular-shaped test polygon. Both models yielded promising results, i.e., overall mean absolute stride length estimation errors of 6.44 cm and 5.64 cm, respectively. On average, the first model achieved a mean absolute error (MAE) of stride length estimation approximately 27% less than the average MAEs produced by the related models included in the comparison. Whereas the second model achieved an MAE of stride length estimation approximately 26% less than the MAEs of the related models included in the comparison on average. Both models are unaffected by smartphone orientation, having the advantage that no special care regarding orientation being needed when attaching the smartphone to a particular body segment. Due to promising results and favorable characteristics, both models could present an appealing alternative for step length estimation in PDR-based approaches. During this research, we also started setting the basis for standardizing the performance evaluation procedure by dealing with an in-depth analysis and comparison of step length estimation models, proposing the following categories of models: step-frequency-based, acceleration-based, angle-based, and multiparameter. Furthermore, we investigated the evaluation approaches of step length estimation models and extracted the evaluation guidelines considering several criteria. In the scope of this work, we also established an open benchmark repository including over 70 km of gait measurements obtained from a group of healthy adults. This repository fosters the comparability of the evaluation results and simplifies the benchmarking of new models. To the best of our knowledge, we were the first to introduce this way of comparison of the models, which has the potential to become a generalized and accepted way of evaluating and comparing performances of step length estimation models.

• Language: English
• Keywords: accelerometer, gait model, inertial sensing, inertial sensors, open-source dataset, step length estimation, smartphone, step length estimation model
• Work type: Doctoral dissertation
• Typology: 2.08 - Doctoral Dissertation
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2023
• PID: 20.500.12556/RUL-152330
• COBISS.SI-ID: 174455555
• Publication date in RUL: 20.11.2023

Secondary language

• Language: Slovenian
• Title: Model za oceno dolžine koraka z inercijskimi senzorji

Abstract

Računska navigacija z inercijskimi senzorji uporablja številne pristope za oceno dolžine koraka, zlasti modele, ki so namenjeni oceni dolžine koraka na pametnih telefonih. Vendar pa avtorji modelov pri njihovi zasnovi redko uporabijo izmerjene dolžine korakov in kinematiko človeškega telesa med hojo. Prav tako ni uveljavljenega protokola za ovrednotenje modelov, kar še posebej pride do izraza pri njihovi analizi in primerjavi. Glavni znanstveno raziskovalni doprinosi te doktorske disertacije zajemajo nov pristop za ovrednotenje modelov ter dva nova izboljšana modela za oceno dolžine koraka z inercijskimi senzorji, ki dosegata boljše rezultate od primerljivih modelov ter sta zasnovana upoštevajoč izmerjene dolžine ciklov korakov in kinematiko gibanja človeškega telesa med hojo. Prvi predlagani model temelji na magnitudi pospeška, drugi pa na magnitudi pospeška in frekvenci korakov. Za karakterizacijo zbranih eksperimentalnih podatkov pri izpeljavi modelov smo kot prvi uporabili analizo glavnih komponent in kanonično korelacijsko analizo, pri čemer smo se osredotočili na prostorske položaje referenčnih točk na človeškem telesu med hojo, ki jih je spremljal optični merilni sistem. Predlagana modela smo ovrednotili za štiri tipične položaje pametnega telefona ter hojo po tekalni stezi in pravokotnem testnem poligonu. Oba modela sta dosegla obetavne rezultate. Skupna povprečna absolutna napaka ocene dolžine ciklov korakov je znašala 6.44 cm za prvi model oziroma 5.64 cm za drugi model. Prvi model je na enaki množici podatkov v povprečju dosegel približno 27% manjšo skupno povprečno absolutno napako ocene dolžine ciklov korakov kot modeli, ki smo jih vključili v njegovo primerjalno analizo. Drugi model pa je v povprečju dosegel približno 26% manjšo skupno povprečno absolutno napako ocene dolžine ciklov korakov kot modeli, ki smo jih vključili v njegovo primerjalno analizo. Posledično predlagana modela predstavljata atraktivno alternativo za oceno dolžine koraka pri računski navigaciji, saj orientacija pametnega telefona na njiju ne vpliva in zato ni potrebno posebne pozornosti nameniti usmerjenosti pametnega telefona pri namestitvi na določen del telesa, kar predstavlja pomembno prednost v primerjavi s številnimi drugimi modeli. Med to raziskavo smo začeli postavljati tudi osnove za standardizacijo ovrednotenja modelov za oceno dolžine koraka. Na podlagi poglobljene analize in primerjave smo predlagali nove kategorije modelov: modeli, zasnovani na frekvenci koraka, modeli, zasnovani na pospešku, modeli, zasnovani na kotu, in multiparametrični modeli. Poleg tega smo raziskali obstoječe pristope ovrednotenja modelov za oceno dolžine koraka in izluščili smernice za ovrednotenje ob upoštevanju več kriterijev. V okviru tega dela smo za primerjavo vzpostavili javno dostopen referenčni repozitorij z več kot 70 km meritev hoje zdravih odraslih oseb. Ta repozitorij spodbuja primerljivost rezultatov ovrednotenja in poenostavlja primerjalno analizo novih modelov. Kot prvi smo začeli z aktivnostmi za vzpostavitev takega načina primerjave, ki ima potencial, da postane splošno sprejet način za ovrednotenje in primerjavo performans modelov za oceno dolžine koraka.

• Keywords: pospeškometer, model hoje, inercijsko zaznavanje, inercijski senzorji, javna podatkovna množica, ocena dolžine koraka, pametni telefon, model za oceno dolžine koraka