• Home
  
  
• Übersicht
• Forschung
• Lehre
• Mitarbeitende
• Publikationen
• Auszeichnungen
• Sonstiges
  
  
• Aktuelles
• Kontakt

GET-Forschungsseminar Abstracts

Classification of Moving Objects Using Appearance and Optical Flow

Branko Dodic, GET Lab

17.08.2016, 16:30, P 1.4.17

Abstract:

The problem of extraction and classification of moving objects from video sequences, based on object's appearance and optical flow was investigated. The classes of pedestrians, bicycles and cars were considered. A classification algorithm, based on Support Vector Machines (SVM) and various moving object features and feature combinations was defined. A software environment was developed, which implements the algorithm and supports labeling and extraction of moving objects, calculation of their features and feature combinations, training of SVM, classification of objects and testing of the overall classification process. A video set with numerous moving pedestrians, bicycle riders and cars, filmed in a midsize city, was used for the investigation of the classification methodology and testing of the developed algorithm. The tests were performed with diverse features, feature combinations and two variants of SVM: linear and non-linear. Remarkable results were achieved on given video set, with certain features and feature combinations.