Metode za detekciju osoba u stvarnom vremenu iz niza slika

master's thesis

• Title (english): Real-Time Methods for People Detection from Video
• Author: Matej Peroš
• Mentor: Sven Lončarić (mentor)
• Committee member: Sven Lončarić (predsjednik povjerenstva)
• Committee member: Marko Subašić (član povjerenstva)
• Committee member: Tomislav Petković (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2018-07-03, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Metode za detekciju, pračenje i brojanje osoba i drugih objekata je zanimljivo polje istraživanja zahvaljujući svojim izazovima i važnosti i među krajnjim ciljevima unaprjeđenja i proširenja video nadzora i razumjevanja videozapisa. Metoda za detekciju se sastoji od otkrivanja i prepoznavanja objekata ili osoba u videu i za odeđene aplikacije i primjene kao nadzorna kamere, u stvarnom vremenu. Ovo zadnje prestavlja posebni izazov jer zahtjeva da algoritam radi jadnako brzo kao i brzina samog videozapisa. Iako posoje metode za detekciju objekata koje su u stanju se vrtiti brzinom videozapisa, ubacivanje metode pračenja tih objekata dodaje novu razinu kompleksnosti u cijelu priču. Za detekciju i prepoznavanje osoba i objekata koristimo metode strojnog učenja i neuronskih mreža, s obzirom da postaju sve naprednije i brže u prepoznavanju ljudi i/ili objekata, no u većini slučaja to ne rade u stvarnom vremenu.

Abstract (english)

Methods for detecting, tracking, and counting people and other objects is an interesting field of research thanks to their challenges and importance and as one of the ultimate goals of enhancing and expanding video surveillance and understanding of video. The detection method consists in detecting and recognizing objects or people in the video and for executed applications and applications as a surveillance camera in real-time. This last poses a special challenge because it requires the algorithm to run as fast as the speed of the video itself. Even though there are methods to detect objects that are able to spin at the video rate, inserting the tracking method of these objects adds a new level of complexity to the whole story. We use machine learning methods and neural networks to detect and recognize individuals and objects, since they are becoming more and more rapid in identifying people and / or objects, but in most cases they do not work in real time.

• Keywords: Detekcija osoba; strojno učenje; Kalman Filter; praćenje objekata; Neuronske mreže; YOLO.
• Keywords (english): Detection of persons; machine learning; Kalman Filter; Observation of Objects; Neural networks; YOLO
• Language: croatian
• URN:NBN: urn:nbn:hr:168:876863
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2019-03-13 23:50:52