Real-Time Face Tracking and Recognition System Using Kanade-Lucas-Tomasi and Two-Dimensional Principal Component Analysis

In this paper, a system for face tracking and recognition in a video sequence is proposed based on KLT (Kanade-Lucas-Tomasi) tracker and 2DPCA (Two-Dimensional principal Component Analysis). Before using KLT algorithm for tracking faces, Viola-Jones face- detection-algorithm is applied to-detect all faces in the image or video sequence. KLT tracks face objects after being detected in the consecutive frames and sustaining long-term- tracking when faces come in/out. Face features are captured and selected using 2DPCA technique which is applied as feature extraction in order to eliminate noise and recognize faces more efficiently using a distance classifier. Face94 dataset and images captured by computer webcam are-used-to test the proposed system. Experimental results-show-that Viola-Jones algorithm is efficient when detect front faces. The KLT algorithm is tested to track faces using ten different videos captured by computer webcam. KLT is successfully applied and is able to track multiple faces even when the detected face turns left or right. Finally, 2DPCA is successfully applied and is able to recognize faces in both Face94 dataset and computer webcam video sequence.