• 
• 
• 
• 
• 
• 

Building 3D Models of the Human Head

The goal of this project is to generate 3D models of the human head. The illustration shows video images of a subject at top. These images are used to automatically construct a 3D model, and views of the model (synthetically generated views of the face) are shown at bottom. This 3D data is useful for a variety of purposes. It can be used for recognition. It can provide the basis for a system that tracks the motion of the subject's head. It can be used to identify expression change. It can be used to generate photo-realistic avatars. Combinations of these technologies underpin significant applications. For example, transmitting a 3D model of a subject's head, and subsequently transmitting just the head motion and expression change, offers significant compression of the data compared to transmitting a full video stream.

Background & Objective:  3D modeling of the human head underpins multiple applications.  A prime objective is to investigate the use of 3D models for face recognition from oblique views of a subject (this contrasts with face recognition from fronto-parallel views of a subject, which is now a well advanced area). An example scenario would involve a subject's face being scanned on entrance to an Intelligent Building, and subsequent identification of the subject from elevated cameras at locations inside the building.

Technical Discussion:  The current system works with a single camera, and utilizes a prior model of a generic human head. The extraction process takes about two minutes and requires controlled head motions by the subject. Newer work is utilizing a three-camera stereo system and has the goal of avoiding the requirement for controlled head motion.

Technology Areas:
Computer Vision
Graphics

Modification Date:  June 13, 2008