Name
Abstract | ||
|---|---|---|
This paper describes a system capable of realizing a face detection and tracking in video sequences. In develop- ing this system, we have used a RBFneural network to locate and categorize faces of different dimensions. The face tracker can be applied to a video communication system which allows the users to move freely in front of the camera while communicating. The system works at several stages. At first, we extract useful parameters by a low-pass filtering to compress data and we compose our codebook vectors. Then, the RBFneural network realizes a face detection and tracking on a specific board. 1 Introduction A system capable of doing face localization and recog- nition in real time has many applications in intelligent man-machine interfaces and in other domains such as very low bandwidth video conferencing, virtual actor and video e-mail. We describe a system capable of de- tecting and to track faces in video sequences using a RBFneural network. The Radial Basis Function (RBF) allows to make learning in neural networks. This function makes it possible to design a network with a good generaliza- tion ability and a minimum number of nodes to avoid unnecessary computational time. The RBFmethod is a technique for interpolation in a high dimensional space. RBFclassifiers belong to the category of kernels classi- fiers. They use an overlapping formed by simple kernel functions to create complex decision regions. RBFnet- works are a recent addition to the face tracking and analysis model because their main advantages are com- putational simplicity and robust generalization. Mark Rosenblum and al.(1) have developed a system of hu- man expressions recognition from motion based on a RBFnetwork architecture. Howell and Buxton have performed a learning identification with RBFmethod(2). Our aim is to elaborate a quite efficient algorithm us- ing a RBFnetwork which can track and recognize faces of different dimensions in real time in natural video se- quences in any background. In the second section, we present the RBFnetwork model. Learning process and node reduction are described. We show the system of face tracking developed in the third section. We ex- hibit the chosen method : extraction of useful param- eters, RBFnetwork architecture used. We display the results and performances we have obtained with a video sequence. Finally, we discuss about our hardware im- plementation. 2 Architecture of the network |
Year | Venue | Keywords |
|---|---|---|
2002 | EUSIPCO | data compression,face recognition,image sequences,low-pass filters,neural nets,object tracking,radial basis function networks,video cameras,video communication,video signal processing,rbf neural network,camera front,codebook vector,face detection,face tracking real time implementation,low-pass filtering,video communication system,video sequences |
Field | DocType | ISSN |
Computer vision,Video processing,Object-class detection,Computer science,Filter (signal processing),Video tracking,Artificial intelligence,Face detection,Video compression picture types,Facial motion capture,Codebook | Conference | 2219-5491 |
Citations | PageRank | References |
1 | 0.38 | 6 |
Authors | ||
3 | ||
Name | Order | Citations | PageRank |
|---|---|---|---|
| Malasne, N. | 1 | 1 | 0.38 |
| F Yang | 2 | 86 | 22.90 |
| Michel Paindavoine | 3 | 115 | 21.70 |