Face Recognition Analysis Based on the YOLO Algorithm

Abstract

Algorithm YOLO consists of a discrete CNN method to achieve end-to-end detection of the target. The detected target is acquired by processing network prediction results that match up with the R-CNN algorithm; it is an integrated system with faster speed and an end-to-end mechanism for training. The first kind of method is slower but more accurate, and the second kind of algorithm is faster but less accurate. This paper introduces the Yolo algorithm. Face recognition analysis is an important task, basically summarizing the properties of this algorithm: Firstly, You Only Look Once which depicts that a single CNN operation is necessary, and Unified portrays that it is a combined system that gives end-to-end results of the prediction, while "real-time" says that the Yolo algorithm is quick. The YOLO 4 algorithm is slower than the SSD algorithm, but Yolo has continuously evolved, resulted in YOLO 9000. This paper basically portrays the principle of the Yolo3 algorithm, mainly the details related to the training, and detection, and finally gives how to use TensorFlow to realize the Yolo algorithm.