Doctoral Dissertation “Development and Analysis of Algorithms for Object Identification Based on Image”

Author, institution: Saulius Sinkevičius, Kaunas University of Technology

Science area, field: Technological Sciences, Informatics Engineering

The Doctoral Dissertation is available at the library of Kaunas University of Technology (K. Donelaičio St. 20, Kaunas).

Scientific Supervisor: Prof. Dr. Arūnas LIPNICKAS (Kaunas University of Technology, Technological Sciences, Informatics Engineering – 07T).

Dissertation Defence Board of Informatics Engineering Science Field:

Prof. Dr. Robertas DAMAŠEVIČIUS (Kaunas University of Technology, Technological Sciences, Informatics Engineering – 07T) – Chairman;

Prof. Dr. Adas GELŽINIS (Kaunas University of Technology, Technological Sciences, Informatics Engineering – 07T);

Assoc. Prof. Dr. Arnas KAČENIAUSKAS (Vilnius Gediminas Technical University, Technological Sciences, Informatics Engineering – 07T);

Prof. Dr. Kurosh MADANI (University PARIS-EST Créteil, Technological Sciences, Informatics Engineering – 07T);

Assoc. Prof. Dr. Rytis MASKELIŪNAS (Kaunas University of Technology, Technological Sciences, Informatics Engineering – 07T).

Annotation:

The aim is to offer algorithms that could effectively sort industrial products by visual properties based on examples provided by the expert.
The objective of the research is to create computer vision algorithms for multiclass identification of amber nuggets by their color, texture, and shape.
Research uses databases of amber nuggets and electrical connectors. Algorithms have been initially developed for identification of amber nuggets but later adapted and applied for identification of electrical connectors. The main difference is that amber nuggets are natural objects featuring infinite diversity, while electrical connectors are man-made and their diversity in terms of color, texture, and shape is limited.
Scientific novelty of the study: the new system for multiclass identification of amber nuggets by their color and texture applying the proposed methods has been developed; the algorithm for multiclass identification of natural objects by their similarity to simple geometrical shapes has been developed; the method of using SIFT features for identification of rotation angle of the man-made objects has been proposed.
The developed algorithms have been applied in the automated amber sorting line at MAX SOLAR ENERGY private limited liability company.