Please use this identifier to cite or link to this item: http://hdl.handle.net/10609/99226
Title: Detecció d'objectes a seqüències de vídeo
Author: Bonnín Hernández, Joan
Director: Casas-Roma, Jordi  
Tutor: Moyà Alcover, Gabriel
Abstract: The project consists in the identification, selection and evaluation of different methods and systems for solving two current problems in the computer's vision field: object detection and object tracking. To solve both tasks, we've studied classical solutions with a well-known good performance and the latest approaches based on machine learning and deep learning. In order to make a comparison between models, a set of experiments has been done. Those experiments are built over the dataset of MOTChallenge, specifically 2017 edition. For the detection task the studied models are: DPM, SDP, Mask-RCNN and YOLOv3. For the tracking task the studied models are: CamShift, correlation filters and SORT. The combination between different systems to solve both tasks, aims to the fact we actually have the required techniques to automatize of the tasks. Even that, the characteristics of the images to process directly affect the results' quality. To sum up, we define the best models for general scenes, but it's crystal clear that there exists the need of evaluating the context and characteristics of the scene to decide which model to use.
Keywords: computer vision
artificial intelligence
machine learning
object detection
object tracking
Document type: info:eu-repo/semantics/masterThesis
Issue Date: 9-Jun-2019
Publication license: http://creativecommons.org/licenses/by-nc-nd/3.0/es/  
Appears in Collections:Bachelor thesis, research projects, etc.

Files in This Item:
File Description SizeFormat 
joanbonninTFM0619memòria.pdfMemòria del TFM8,67 MBAdobe PDFThumbnail
View/Open