3D FACIAL SCANNING SYSTEM BASED ON POINT CLOUD FOR BIOMETRIC IDENTIFICATION

• Main Author: Firmansyah Putra, Aidil
• Format: Final Project
• Language: Indonesia
• Subjects: Teknik (Rekayasa, enjinering dan kegiatan berkaitan)
• Online Access: https://digilib.itb.ac.id/gdl/view/48043
• Institution: Institut Teknologi Bandung

2D facial recognition as a biometric identification system, nowadays is widely used because of it ease of use and highly accurate on identification process. Regardless of the benefits, there are still weakness on 2D facial recognition that are used with RGB camera while scanning can cause security problem. For that, we are proposed a facial recognition system design which made with 3D scanning system based on point cloud using stereo vision camera (kinect model 1414) along with facial detection algorithm that made from Python programming language with Deep learning method. Specifically, in this TA’s book will be discussed about 3D facial scanning program design based on point cloud using Processing IDE. The program designed to fullfill design requirements such as, facial scanning on multiple illumination condition, facial scanning on certain range of distance, data visualization of 3D point cloud face in multiple pose, scanning experiment with spoofed face. From using the scanning program, we will create a dataset for train the detection algorithm based on deep learning. Data will be prepared in 2D images with PNG format that was converted from 3D point cloud. The datasets created in four type that contain from 2 – 5 subject’s faces. From the experiment result obtained several aspect as follow, system sensitivity was stable at different illumination, scanning range can be specified based on depth map data, able to visualize every angle of face’s poses by rotated the 3D face, system can avoid 2D spoofing face. From all four data, the loss value obtained between 2,6 – 3,5. This value was affected not only by quantity of dataset, but also the amount subject per dataset, input sample choosed for dataset and training method. The system was launch to be running on Windows OS with Java supported.