The extracted points or features are tracked in real time
video by facial feature recognition algorithm of Kanade-LucasTomasi
(KLT) [1, 2, 3]. This algorithm provides the locus of
the variation of recognized features. The algorithm tracks the
most perfect match to the facial detected features provided the
motion of the face is slow. It computes the Eigen values of the
gradients and compares with threshold values in order to
recognize the detected points. The points which will be
detected in the video, must be inside the bidirectional
threshold. The real time video must be initialized properly to
track the extracted features from the already stored image. A 2-
D geometric transform object has been created to confine each
matched points or features from the source image and the real
time video.