A Fully Automated Active Shape Model for Segmentation and Tracking of Unknown Objects in a Cluttered Environment

Abstract

The segmentation of shapes is automated using a new objective function to deform and move a contour toward the actual shape. New profile modeling and optimization criterion to automatically find corresponding points are also applied for segmentation and tracking of people in cluttered backgrounds. The proposed framework utilizes a Pan-Tilt-Zoom (PTZ) camera and automatically performs the initial target acquisition through motion and color-based segmentation. Successful results are presented for within and between frame segmentation and tracking. This algorithm presents a major extension to the state of the art and the original active shape model (ASM) which was designed for known objects in smooth non-changing backgrounds and where the landmark points need to be manually picked off-line. This is a fully automated, real time ASM that deals with changing backgrounds and does not require prior knowledge of the object to be segmented and tracked.