Doktora Tezi Görüntüleme

Öğrenci: Hasan KARAL
Danışman: Assoc. Prof. Dr. Rıfat YAZICI
Anabilim Dalı: Elektrik-Elektronik Müh.
Enstitü: Fen Bilimleri Enstitüsü
Üniversite: Karadeniz Teknik Üniversitesi
Tezin Türü: Doktora
Kabul Tarihi: 16/6/2000
Sayfa Sayısı: 190
Tez No: di267

      The aim of this thesis is to design an autonomous car which moves by observing its surrounding environment with a camera and processing the image taken. The car can move safely under every road condition on either asphalt road or stabilized roads. Until now the edges of road has been extracted from the images using conventional techniques such as gradient, Laplacian or moment methods. The success of these methods in finding find out road edges depends upon the road conditions, including road markers and/or contrast between the colours of the road surface and the road bank. In addition, the car is able to navigate on a stabilized road as well using only the texture primitive feature to effectively define the edges of such a road.

      Investigations on texture classification began many years ago, but none of the class methods have been succesfuly indetermining road edges, due to high complexity in texture calculations and their inefficient processing algorithms. In this thesis a new method based on texture dissimilarity (TDM) is proposed to extract edges from road images without any constraint on road types.

      The proposed method is capable of dividing an area into subarea which differ significantly in texture primitives and their distributions. The method has been successful for not only textured surface but also coloured surfaces with no texture. Furthermore it also overcomes the shadowing and wetness problems which cause misunderstanding of a road environment.

      Another important feature included is the forecast learning technique, whereby that the car is controlled with fuzzy logic by processing far and near image segments. The fuzzy rule st is updated if necessary so that the car will be able to make appropriate decisions to easily avoid obstacles. Recognition of obstacles is implemented by a novel technique called leaf method which is a type of artificial neural network. The method when applies to character classification correctly classified 98% of previously unseen handwritten digits. This rate is greather that of its competitors.


Keywords: Texture, Texture Dissimilarity, Autonomous Car, Fuzzy Logic, Forcast Learning, Neural Networks, Associative Memory