Raspoznavanje fizičke aktivnosti zasnovano na strojnom učenju

master's thesis

• Title (english): Physical Activity Recognition Based on Machine Learning
• Author: Krunoslav Jurčić
• Mentor: Ratko Magjarević (mentor)
• Committee member: Ratko Magjarević (predsjednik povjerenstva)
• Committee member: Tomislav Pribanić (član povjerenstva)
• Committee member: Dinko Oletić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2021-07-15, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Motivacija za razvoj algoritma za raspoznavanje fizičke aktivnosti uključuje praćenje dnevne aktivnosti korisnika u stvarnom vremenu, te moguću dijagnostiku neželjenih i neočekivanih pokreta.
Algoritam za raspoznavanje razvijen je koristeći dva skupa podataka koji sadrže signale dobivene iz akcelerometra, magnetometra i žiroskopa. Prvi otvoreni skup podataka, dostupan svim korisnicima, sadrži signale koji opisuju pet fizičkih aktivnosti. Drugi skup opisuje sveukupno 16 aktivnosti, koje su nakon grupiranja svrstane u 12 klasa. Od toga 9 aktivnosti pripada svakodnevnim aktivnostima, a tri pripadaju neočekivanim pokretima, točnije padovima.
Najveću točnost klasifikacije oba skupa ostvarile su metode slučajne šume (96.45%, 87.10%) i stroja potpornih vektora (95.12%, 81.72%). Osim raspoznavanja svakodnevnih aktivnosti, algoritmi su izvrsne rezultate postigli i pri detekciji i razlikovanju padova (97.83% i 86.96%), što upućuje na njihovu široku upotrebu u biomedicinskom inženjerstvu.

Abstract (english)

The motivation for the development of physical activity recognition algorithm includes keeping track of user's activities in real time, and possible diagnostics of unwanted and unexpected movement.
Activity recognition algorithm was trained on two datasets containing signals from an accelerometer, a magnetometer and a gyroscope. The first open access dataset contains signals describing five physical activities. The other dataset describes a total of 16 physical activities, which were clustered into 12 classes. A total of nine classes describe activities of daily living, while the other three belong to unexpected movements, more precisely falls.
The best results of classification on both datasets were achieved by Random forest (RF) algorithm (96.45%, 87.10%) and Support vector machine (SVM) algorithm (95.12%, 81.72%). Aside from recognition of daily activities, the algorithms also achieved excellent results when detecting and distinguishing fall types (97.83% and 86.96%), which suggests of their wide use in biomedical engineering.

• Keywords: Biomedicinsko inženjerstvo; Fizička aktivnost; Strojno učenje; Umjetna inteligencija; Akcelerometar; Detekcija pada
• Keywords (english): Biomedical engineering; Physical activity; Machine learning; Artificial intelligence; Accelerometer; Fall detection
• Language: croatian
• URN:NBN: urn:nbn:hr:168:006090
• Study programme: Title: Information and Communication Technology; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka informacijske i komunikacijske tehnologije (magistar/magistra inženjer/inženjerka informacijske i komunikacijske tehnologije)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-01-13 12:33:43