Download Full Text (1.0 MB)
Real-time tracking and recognition of people in complex environments has been a widely researched area in computer vision as it has a huge potential in efficient security automation and surveillance. We propose a real time system for detection and recognition of individuals in a scene by detecting, recognizing and tracking faces. The system integrates the multi-view face detection algorithm, the multi-pose face recognition algorithm and the extended multi-pose Kalman face tracker. The multi-view face detection algorithm contains the frontal face and profile face detectors which extract the Haar-like features and detect faces at any pose by a cascade of boosted classifiers. The pose of the face is inherently determined from the face detection algorithm and is used in the multi-pose face recognition module where depending on the pose, the detected face is compared with a particular set of trained faces having the same pose range. The pose range of the trained faces is divided into bins onto which the faces are sorted and each bin is trained separately to have its own Eigenspace. The human faces are recognized by projecting them onto a suitable Eigenspace corresponding to the determined pose using Weighted Modular Principal Component Analysis (WMPCA) technique and then, are tracked using the proposed multiple face tracker. This tracker is implemented by extracting suitable face features which are represented by a variant of WMPCA and then tracking these features across the scene using the Kalman filter. This low-level system is created using the same face database of twenty unrelated people trained using WMPCA and classification is performed using a feature correlation metric. This system has the advantage of recognizing and tracking an individual in a cluttered environment with varying pose variations.
Vijayan K. Asari
Primary Advisor's Department
Electrical and Computer Engineering
Stander Symposium poster
Cui, Chen; Diskin, Yakov; and Nair, Binu M., "Pose Invariant Face Recognition and Tracking for Human Identification" (2013). Stander Symposium Posters. 241.