Reduced support vector machines applied to real-time face tracking

The paper presents the implementation of a real-time face tracker to study the integration of support vector machines (SVM) classifiers into a visual real-time tracking architecture. Face tracking has a large number of applications, especially in the fields of surveillance and human-computer interaction, which requires real-time performance. Even though SVM have previously been applied to face detection, their use in real-time applications is a challenge due to the computational cost implied in the SVM's evaluation stage. We address this problem by reducing the number of support vectors with almost no loss in accuracy of the classifier. Experiments showed that classification performed by the original SVM without reducing the number of support vectors took 42% of the total computation time of the face tracker and less than 2% after the reduction was performed.