Year: 2014
Type: dissertation
Abstract: In this thesis, an object detection system based on omnidirectional camera which \nhas the advantages of detecting a large view-field is introduced. Initially, the traditional \ncamera approach that uses sliding windows and Histogram of Gradients (HOG) features \nis adopted. Later on, how the feature extraction step of the conventional approach should \nbe modified is described. The aim is an efficient and mathematically correct use of HOG \nfeatures in omnidirectional images. Main steps are conversion of gradient orientations to \ncompose an omnidirectional sliding window and modification of gradient magnitudes by \nmeans of Riemannian metric. Owing to the proposed methods, object detection process \ncan be performed on the omnidirectional images without converting them to panoramic or \nperspective image. Experiments that are conducted with both synthetic and real images \ncompare the proposed approach with regular (unmodified) HOG computation on both \nomnidirectional and panoramic images. Results show that the performance of detection \nhas been improved by using the proposed method.
Author: İbrahim Çınaroğlu
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2275813896