A Method For Tracking of Facial Action Points Using Pyramidal Lucas Kanade Algorithm for Expression Recognition

This paper presents a systematic methodology to analyze displacement metrics to be used in recognizing facial expressions with the help of FAP or landmarks on a face particularly on the mouth, nose tip and eye and eye brows. The proposed system tries to automatically perform human face detection, feature point extraction for the further purpose of facial expression recognition stably and robustly over live webcam input of real time sequences of frames and yield promising performance in low-resolution video sequences captured in real-world environments. Thus a sequential three-stage approach is taken, firstly Face Detection, secondly Feature Extraction and finally tracking of the extracted features. In real time, a human face is detected using an improved Haar based algorithm. Then facial features are extracted using a combination of appearance and geometrical approach depending on the ROI extracted by Haar based algorithm. The tracking is done with the help of pyramidal Lucas Kanade algorithm. Several experiments conducted under relatively non-controlled conditions demonstrate the accuracy and robustness of the approach.