基於FPGA的即時性人臉擷取

Abstract

人臉偵測在人臉辨識系統中為重要且首先必要的步驟，因此在系統上要求必須快速而準確。故近年來，已有許多有關人臉偵測的方法被提出，大部分都是基於演算法方面的改善，在硬體上的實現比較少見，故本論文引用的兩種膚色演算法實做在嵌入式開發版並加以改善來使整個系統硬體加速，並以型態學中的去雜訊法來使膚色區域更完整，並額外加上移動偵測，使系統能在個別的情況下，根據物體移動的程度來做即時性的人臉擷取

Face detection is the first and a required step in a face recognition system. For real-time systems, the step should work quickly and accurately. Recently there have been numerous methods for human face detection proposed for real-time imaging systems. Among them, a class of methods is based on some color space features to detect face. However, few of them considered the implementation it for the hardware system. The thesis uses two widely used skin detection algorithms and aim to implement the algorithms in a hardware platform, which is an Altera FPGA embedded boards. The proposed face detection system makes use of noise removal methods to make the skin region more complete. In addition, the system also uses motion detection to help determine an object's motion, which in turn is used as a condition to account for varying circumstances. The experiment results show that the proposed face detection can achieve the real-time face detection task with 60 frames per second.