Implementation of Fisherface Algorithm for Eye and Mouth Recognition in Face-Tracking Mobile Robot

Abstract

Facial recognition is an artificial intelligence algorithm that distinguishes one face from another by capturing facial patterns visually. This recognition specifically detects and identifies individuals based on facial features by scanning the entire face. Several methods are used for facial detection, including facial landmarks points, Local Binary Patterns Histograms (LBPH), and Fisherface. In the context of this research, Fisherface is used to reduce the dimensionality of facial space in order to obtain image features. The method is insensitive to changes in expression and lighting, leading to better pattern classification and making it suitable for implementation on mobile devices such as robot vision. Therefore, this research aimed to measure the response time speed and accuracy level of pattern recognition when implemented on mobile robot devices. The results obtained from the accuracy testing showed that the highest accuracy for face detection process was 90%, while the lowest was 78.3%. In addition, the average execution time (AET) for the fastest process was 1.63 seconds and the slowest was 1.72 seconds. For pattern recognition, the statistics showed 90% accuracy, 100% precision, 81.81% recall, and F-1 score of 89.5%. Meanwhile, the longest execution time was 0.084 seconds and the fastest was 0.064 seconds. In face tracking process, the mobile robot movement was based on real-time pixel sizes, determining x and y values to produce the center of face region.