Home -> Projects ->Robust 3D Head Tracking by View-based Feature Point Registration Member Login  

Jump to: Demos | Publications | Releases

 Description

A set of point features is registered and updated for each reference pose serving a multi-view head detector. Since tracking-by-detection approach is used in our method, the proposed tracker utilizes view-based model more efficiently than previous view-based model approaches. The view-based feature point registration rectifies error accumulation and provides fast recovery after occlusion has ended, while preventing divergence problem which frequently occurs in conventional frame-to-frame tracking methods. Kalman filter is used to incorporate motion between successive frames and the estimated pose with the view-based head model. Although we focus on the robustness of tracker, the accuracy of the proposed tracker is comparable to previous studies (Face Tracking using AAM and Cylinder Head Tracker etc.).



Overall architecture of this 3D head tracker

 Demos

Head and Face Orientaiton Result 1

Download: media1.avi [20.8 MB]
Head and Face Orientaiton Result 2

Download: media2.avi [5.4 MB]
Head and Face Orientaiton Result 3

Download: media3.avi [4.8 MB]

 Related Publications

 Releases [ Members Only ]