Bilkent University
Department of Computer Engineering


Human Motion Retrieval


Yağız Salor
MSc Student
Computer Engineering Department
Bilkent University

Nowadays, most of the video games and animation movies need different kind of realistic human motion data. However, the process of manually creating realistic human movements is a very difficult problem for animation artists. Therefore, the usage of previously recorded human motion capture data is inevitable. There are free to use motion capture databases on the internet which provide wide selection of motion types. As a consequence, searching similar segments on these sources emerged as an essential topic of computer animation research. In literature, this problem is solved efficiently by the use of binary geometric features and inverted indexes nonetheless they are not applicable for searches that require close similarity. There is another approach which uses kd-trees and shortest path search on directed graph to find closest motion clip. However, this one does not utilize any technique for fuzzy matches. In this research, we develop a hybrid approach of these previously mentioned approaches that aims to solve their shortcomings.


DATE: 27 April, 2015, Monday @ 15:40