VISION ARCHITECTURE TO DESCRIBE FEATURES OF PERSONS

摘要

A machine vision system is described that allows the description of persons appearing in video images. The system integrates vision routines that detect heads and hands with modules that perform pattern recognition to analyze the heads, faces and hands in fine detail. Head and hand detection makes use of a broad integration of different visual pathways such as motion, color and stereo vision as well as modules that extract topological and structural cues. Pattern recognition for fine analysis makes use of the technique known as elastic bunch graph matching.