Sledenje objektov z generaliziranim Houghovim transformom

Muhovič, Jon Natanael (author), Kristan, Matej (mentor)

Sledenje objektov je zelo raznoliko in uporabno področje računalniškega vida. Pristopov k reševanju tega problema je ogromno, namen tega dela pa je predstaviti nekaj metod, ki se uporabljajo za implementacijo naprednih sledilnih algoritmov, analizirati konkreten algoritem, ki sledenje izvaja z uporabo generaliziranega Houghovega transforma ter načrtati in izvesti izboljšave, ki bodo koristile algoritmovi robustnosti in natančnosti. Predlagane izboljšave temeljijo na uporabi Harrisove detekcije oglišč, Kalmanovega filtra in segmentacijskega algoritma, ki deluje na podlagi Markovovih slučajnih polj. Rezultat diplomskega dela je tako izboljšan algoritem, implementiran v C++, z dodanimi metodami, ki mu omogočajo boljše delovanje. Praktični eksperimenti so bili izvedeni v okolju, namenjemu testiranju sledilnih algoritmov, z uporabo raznolikih in zahtevnih video sekvenc. Rezultati eksperimentov jasno prikazujejo izboljšave, ki so jih povzročile predlagane metode.

• Language: Slovenian
• Keywords: sledilni algoritem, generaliziran Houghov transform, sledenje netogih objektov, segmentacija.
• Work type: Bachelor thesis/paper
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2014
• PID: 20.500.12556/RUL-29526
• Publication date in RUL: 19.09.2014

Secondary language

• Language: English
• Title: Object tracking by a generalized Hough transform

Abstract

Visual object tracking is a very diverse and useful area of computer vision. There are many different approaches to solving this problem and the goal of the thesis is to first present some of the methods that are used for implementation of state-of-the-art tracking algorithms. Secondly, the analysis of a concrete algorithm that tracks the object by using generalized Hough transform and lastly to design and implement some enhancements that boost the algorithm's robustness and accuracy. The proposed enhancements are based on Harris corner detection, Kalman filter and a segmentation algorithm that uses Markov random fields. The result of the thesis is thus an improved algorithm, implemented in C++ with added methods that improve its performance. Practical experiments were carried out in a framework designed for testing tracking algorithms by using diverse and difficult video sequences. Experiment results clearly show the improvements caused by the proposed methods.

• Keywords: tracking algorithm, generalized Hough transform, non-rigid object tracking, segmentation