Detekcija biciklista u video snimci dubokim neuronskim mrežama

master's thesis

• Title (english): Cyclist Detection from Video Using Deep Neural Networks
• Author: Iva Radić
• 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: 2017-07-14, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

U radu je predstavljen problem detekcije objekata u slici, te problem detekcije i praćenja objekata kroz slijedne slike u videosnimci. Uz povjesni pregled, predstavljena su i dosadašnja dostignuća u brzoj i točnoj detekciji objekata. Model korišten u implementaciji sustava za detekciju je potpuno konvolucijski SqueezeDet model. Za korištenje temporalne informacije kao pomoć detekciji objekata u videu razmotrena su dva algoritma, MedianFlow i MIL praćenje. Sustav provodi detekciju s preciznošću od 62.3%

Abstract (english)

This paper presents the problem of object detection in image, and the problem of detecting and tracking objects through sequential images in video clips. Along with the historic review, the paper presents recent achievements in fast and accurate object detection. The model used in the detection system implementation is the fully convolutional SqueezeDet model. Two tracking algorithms, MedianFlow and MIL tracking were considered. Tracking is used to improve accuracy of detection in video.
The system performs detection with accuracy of 62.3%

• Keywords: Detekcija objekata; Konvolucijske neuronske mreže; Propozicija regija; Fast-RCNN; Faster-RCNN; YOLO; Single Shot Detection; SqueezeDet; Potpuno konvolucijske mreže; Praćenje objekata
• Keywords (english): Object detection; Convolutional neural networks; Region proposal; Fast- RCNN; Faster-RCNN; YOLO; Single Shot Detection; SqueezeDet; Fully-Convolutional networks; Object tracking
• Language: croatian
• URN:NBN: urn:nbn:hr:168:045241
• 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: 2019-03-14 00:17:43