Tracking Faces by Accumulation of Detection Probabilities

Dr. Ragini Choudhury

INRIA

In the talk I will present a fully automated method of face tracking in a video sequence which is robust to changes in pose and scale of the face. The method integrates detection and tracking in a unified probabilistic framework. Existing approaches either detect faces in every frame without using the temporal information or detect faces in the first frame and track it through the sequence using a separate algorithm. With the detection information incorporated at each time step, the method is able to handle the appearance and disappearance of faces as well as occlusion. The method also handles out-of-plane rotation of the face by representing the pose of the face as a combination of frontal and profile detection probabilities. The effectiveness of the method is illustrated through numerous video sequences.

Brief Bio :

Ragini Choudhury-Verma received her Ph.D. degree in Computer Vision in 2000, from the Dept. of Mathematics, Indian Institute of Technology, Delhi (India). She received a Masters degree in Mathematics and a Masters in Technology in Computer Applications from the same institute in 1994 and 1996, respectively. She has completed a two-year post doc with the MOVI team at INRIA Rhone-Alpes, France. Her research interests include tracking, reconstruction and recognition problems in vision and geometry.