Real-Time Face Recognition from Surveillance Video - Intelligent Video Event Analysis and Understanding - page 163

Fig. 2 Distributed System Architecture: Multiple Front-Ends

Please see the Appendix to this chapter for more implementation details, includ-
ing how to obtain the source code.

3 Front-End

The front-end has three functions: to detect faces, to track detected faces and to
extract features for transmission to the back-end recognition system. The process is
summarised in figure 3.

3.1 Face Detection

Face Detection is carried out using the O PEN CV Face Detector [5]. This is an im-
plementation of the Viola-Jones Face Detector [31], which uses a boosted rejection
cascade based on Adaboost [8].
The theory of boosting is that many weak classifiers can be combined to make
a strong classifier. Adaboost is a supervised classifier, i.e. , it takes a set of labelled
samples as input. A tree of weak classifiers is created by applying a weak algo-
rithm to the labelled data. Each weak classifier is assigned a weight based on its
performance on the training data. Classifiers that return the wrong result are given
a higher weight, to concentrate attention on the places where errors are likely to be