Human identification in unconstrained scenarios using distributed camera networks

While traditional face recognition systems have to date been primarily designed to operate under controlled environments based on previously acquired photographs and videos, it is now becoming increasingly important to identify humans in real-time under unconstrained environments. Designing such a system is challenging because the acquisition conditions are subject to variable pose of the subject, low illumination and possible occlusions. Our goal is to utilize the robustness and parallel computational abilities of a distributed camera network to address these challenges and achieve accurate, real-time human identification.

Our main areas of research are:

Design of middleware services that utilize the geometry of the network to collaborate at run-time and enable the rapid extraction of multi-view face data and soft-biometric data (such as gait and human anthropometric features)

Camera configuration techniques that enhance the probability of acquiring suitable face images for recognition while minimizing the overall cost of deployment

Efficient fusion algorithms and multi-view face recognition techniques that are optimized for operation in a dynamic, continually streaming mode

Distributed camera network for face recognition

Recent publications

o S. Parupati, R. Bakkannagiri, S. Sankar and V. Kulathumani, “Collaborative acquisition of multi-view face images for real-time face recognition using a wireless camera network”, ICDSC 2011

o V. Kulathumani, S. Parupati, A. Ross and R. Jillela, "Collaborative face recognition using a network of embedded cameras”, Distributed Video Sensor Networks (DVSN), Editors: B. Bhanu, C.Ravishankar, A. Choudhary, D. Terzopoulos and H. Aghajan, Springer-Verlag, 2011

---