Gerçek zamanlı görsel nesne tanıma
Küçük Resim Yok
Tarih
2000
Yazarlar
Dergi Başlığı
Dergi ISSN
Cilt Başlığı
Yayıncı
Ege Üniversitesi
Erişim Hakkı
info:eu-repo/semantics/closedAccess
Özet
Ill ÖZET GÖRSEL NESNE TANIMA Erünver, M.Öncel Yüksek Lisans Tezi, Elektronik Mühendisliği Bölümü Tez Yöneticisi: Prof.Dr.Erol Uyar Haziran 2000, 125 sayfa Bu tezde, endüstriyel uygulamalarda kullanılan otomatik sistemlere, görsel nesne tanıma yeteneği kazandırılması amacıyla gerçek zaman şartlarında çalışabilecek ve diğer sistemlere entegre olabilecek bir görsel nesne tanıma sistemi geliştirilmesi amaçlanmıştır. Bu amaç doğrultusunda genel bir görsel nesne tanıma sistemi tasarlanmıştır. Geliştirilen sistem hem hız olarak, hem de doğruluk yüzdesi olarak endüstriyel kullanıma uygundur. Sistem görüş alanına giren nesneleri farkederek onların fotoğraflarını çeker ve çekilen fotoğraftan nesneyi ayırır. Nesne aymldıktan sonra nesneye ait görsel bilgiyle, altı tane dönme, öteleme, büyüklük değişiminden bağımsız özellik, ağırlık merkezi, asal eksen açısı vb. değerler bilgisayar tarafından hesaplanır. Hesaplanan bu altı özellik nesnenin sınıflandırılması için kullanılır. Akan görüntüde oluşan ortalama parlaklık değişiminden nesnelerin belirlenmesi, farklı renk ve yansıma katsayılarına sahip nesnelerin adaptif olarak artalandan aymlması, nesnelerin görüntülerinden, HU moment sabitleri kullanılarak dönme, öteleme ve
V ABSTRACT VISUAL OBJECT RECOGNITION Erünver, M. Öncel M.Sc. in Electronic Engineering Department Supervisor: Prof.Dr.Erol Uyar June 2000,125 pages In this thesis; a system that could make the automatic systems, capable of visual object recognition, is aimed to be designed and for this purpose a general visual object recognizer is designed. One of the priori constraints was that the designed system must also be integratable to other systems. The developed system is suitable for many industrial applications in many aspects; such as speed and ratio of right classifications. The system detects the object that enters its field of view, takes a photo of it and then separates the object from the background. After segmenting the object the computer calculates six rotation, translation, scale invariant features, centre of gravity, angle of principle axis etc. Afterwards, these six features are used for classification of the object. For detection of objects that enter the field of view of camera, brightness level changes in images is used. As there exists different colour objects in the set that will be recognized, adaptive thresholding is used for segmentation. For obtaining rotation, translation and scaleVI invariant feature vector representation Hu moments are used and for the classification, Euclidean distance measure is used. As an application; the recognition of moving objects on a conveyor belt is performed. This application is a live working example of the system. Because of the generality of the methods applied in the application; this application can also be used for many two dimensional visual codes, like alphabetical characters and numbers, with a modification on the database of prototype set. Keywords: Object recognition, feature vector, visual object segmentation, Hu moment invariants, Euclidean distance measure.
V ABSTRACT VISUAL OBJECT RECOGNITION Erünver, M. Öncel M.Sc. in Electronic Engineering Department Supervisor: Prof.Dr.Erol Uyar June 2000,125 pages In this thesis; a system that could make the automatic systems, capable of visual object recognition, is aimed to be designed and for this purpose a general visual object recognizer is designed. One of the priori constraints was that the designed system must also be integratable to other systems. The developed system is suitable for many industrial applications in many aspects; such as speed and ratio of right classifications. The system detects the object that enters its field of view, takes a photo of it and then separates the object from the background. After segmenting the object the computer calculates six rotation, translation, scale invariant features, centre of gravity, angle of principle axis etc. Afterwards, these six features are used for classification of the object. For detection of objects that enter the field of view of camera, brightness level changes in images is used. As there exists different colour objects in the set that will be recognized, adaptive thresholding is used for segmentation. For obtaining rotation, translation and scaleVI invariant feature vector representation Hu moments are used and for the classification, Euclidean distance measure is used. As an application; the recognition of moving objects on a conveyor belt is performed. This application is a live working example of the system. Because of the generality of the methods applied in the application; this application can also be used for many two dimensional visual codes, like alphabetical characters and numbers, with a modification on the database of prototype set. Keywords: Object recognition, feature vector, visual object segmentation, Hu moment invariants, Euclidean distance measure.
Açıklama
Bu tezin, veri tabanı üzerinden yayınlanma izni bulunmamaktadır. Yayınlanma izni olmayan tezlerin basılı kopyalarına Üniversite kütüphaneniz aracılığıyla (TÜBESS üzerinden) erişebilirsiniz.
Anahtar Kelimeler
Elektrik ve Elektronik Mühendisliği, Electrical and Electronics Engineering, Görsel nesne tanıma, Visual object recognition, Nesne tanıma, Object recognition