Vision based pose recognition in video game

Abstract

We present a vision based HCI system which exploits background subtraction comparing local orientation histograms. As a new virtual input device for game control, we focus on extracting coarse pose of the foreground object and its application to video game. The captured image is divided into the cells where the local orientation histogram with Gaussian kernel is computed and compared with the corresponding one using Bhattacharyya distance measure. The orientation histogram-based method is partially robust against illumination change and small moving objects in background. We also propose a vision-based interfacing system to existing game engines and appropriate modules that includes recognition process using neural network. The real-time 3D video games are implemented as a test-bed with the proposed system to prove the presented vision based system is highly applicable to let users control virtual environment without any hard-wired input devices. © 2008 Springer-Verlag Berlin Heidelberg.