DRUIDE: a real-time system for robust multiple face detection, tracking and hand posture recognition in color video sequences

Abstract

"DRUIDE", which stands for "detection, recognition, unification, interpretation, decision, evolution", is a novel, real-time system primarily designed for the detection and tracking of multiple human faces as well as for the simultaneous recognition of multiple hand postures in color video sequences and in complex environments. The system relies on the three fundamental cues of color, shape and motion, and integrates three mutually complementary sub-systems, in order to achieve high rates of detection, tracking and recognition. Preliminary experiments yield an average correct face detection rate of about 90%, and the additional use of tracking increases significantly the robustness of the proposed system, particularly to illumination conditions and to partial occlusions. Although we first focus specifically on human faces and hand postures, the ultimate goal of DRUIDE extends beyond human-computer interactions, to encompass the adaptive detection, tracking and recognition of various objects under unconstrained scene conditions.