MSc Thesis Defense by Yue Ben


Railroad Catenary Detection


This thesis introduces and implements two methods to detect railway catenaries in train videos. The first method detects vertical edges by using Sobel operator and determines whether these vertical edges belong to catenaries. The second method customizes HOG(Histogram of Oriented Gradients) algorithm to detect catenaries and uses linear SVM(Support Vector Machine) as classifier. Sobel operator algorithm is tested on Danish railway images and HOG algorithm is tested on Danish railway images as well as Ethiopian railway images. These railway images are captured by a recording camera installed on the locomotive. By accurately detecting catenaries in railway videos, railway catenary database can be generated, which could be used to simulate the railway network.

Supervisor: Søren Olsen

Censor: Søren Overgaard