Learning Visual Models of Social Engagement

Introduction

The wearable face recognition and detection problem differs significantly from the classical problem and its solutions. This is due to large variations in lighting, scene, and camera positions that occur in everyday situations. The problem is currently compounded by computational limitations induced by useability and mobility requirements of such a device. Prior work has only proved the feasibility of constructing such a device.  We propose a formal study of the space of data acquireable from such a device and the interfaces for its acquisition.

Publications

Hardware

An earlier capture vest that enables mobile audio/video capture

Short example movie (windows/linux DV encoded AVI - 60MB) that a user captured using the vest.
A Montage of several extracted faces  
A peek at the mean face and first 20 color Eigenfaces computed from that set of faces after cropping them by the social constraint method and reducing them to 31x22x3.

Links