Detekcija objekata u videu snimljenom iz vozila u pokretu

master's thesis

• Title (english): Object Detection in Video Taken from a Moving Vehicle
• Author: Tomislav Lokotar
• Mentor: Zoran Kalafatić (mentor)
• Committee member: Zoran Kalafatić (predsjednik povjerenstva)
• Committee member: Siniša Šegvić (član povjerenstva)
• Committee member: Tomislav Hrkać (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2019-09-27, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Detekcija objekata u prometnim situacijama je velik izazov i ključan dio razvoja autonomnih vozila. U posljednje vrijeme, detekcija objekata vrši se gotovo isključivo dubokim učenjem. U okviru rada su proučavani modeli za detekciju objekata, pristupi njihovog učenja te detaljno opisane dvije arhitekture detektora objekata YOLO i SSD. YOLO i SSD implementirani su u nekoliko biblioteka, a proučavana implementacija je izrađena Python bibliotekom PyTorch. U eksperimentalnom dijelu, dva modela dubokog učenja su trenirani na javno dostupnom skupu podataka BDD100K s označenim objektima u prometu. Na kraju, prezentirane se su uspješnosti detektora na skupovima podataka koje modeli nisu vidjeli tokom učenja, te njihova usporedba. Evaluirane su performanse modela na hardveru modernog računala te demonstriran rad detekcijom na videu snimanom iz vozila.

Abstract (english)

Detection of objects in traffic situations is a great challenge and a key part of the development of autonomous vehicles. Lately, object detection has been done almost exclusively by deep learning. As a part of thesis, object detection models and their learning approaches were studied, with two deep learning based detectors, YOLO and SSD, described in detail. YOLO and SSD have been implemented in several libraries. The studied implementation has been created with the Python PyTorch library. In the experimental part, two deep learning models were trained on a publicly available dataset with labeled objects in BDD100K traffic. Finally, the performance of the detectors on part of dataset that the models did not see during learning are presented and a comparison is made. The performance of the models on modern computer hardware was evaluated and detection on real-life video shot in traffic conditions was demonstrated.

• Keywords: duboko učenje; detekcija objekata; računalni vid; yolo; SSD
• Keywords (english): deep learning; object detection; computer vision; yolo; SSD
• Language: croatian
• URN:NBN: urn:nbn:hr:168:678318
• 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: 2020-01-07 19:44:14