high performance computing on graphics processing units: hgpu.org

In the video sequence human faces have unlimited orientations and positions so face detection and clustering is very important. In this paper, I have proposed a method to cluster human faces from the video sequence based on Spatio Temporal method.The proposed method is based on three main stages. First I have used a face detector to localize faces in all frames of the video and extract various features of the detections.Face detection is done by using viola jones detector and feature extraction is done by using SURF detector. Optical flow estimation is used to compute the movement between any of two frames. In optical flow estimation pyramidal version of lucas kanade algorithm is used. This estimation is used to compute a dissimilarity matrix such as appearance dissimilarity,time dissimilarity. Clustering tracks the same faces from the video sequence it may have different pose,image orientation or it may occluded by objects. Finally, an optimization method involving clustering of faces.