Abstract:

In recent years, with the application of the face detection gradually towards realistic scenes, the occlusion face detection has become one of the hot topics in the field of computer vision. There are some difficult problems to be faced and solved urgently due to the feature damage and noise aliasing caused by occlusion. This paper reviews the research process of the face detection under the occlusion conditions, and divides detection methods into two series according to the different feature construction. One is the classical method based on hand crafted features and the other is the modern method based on deep learning. Their basic principles, model performance properties and existing problems of different algorithms are compared and analyzed. Furthermore, the research directions in the future are discussed.

Key words: occluded face detection, machine learning, deep learning, deformation model

摘要：

近年来，随着人脸检测逐步面向现实场景应用，遮挡条件下的人脸检测成为计算机视觉领域研究的热门课题之一。遮挡所造成的特征损坏和噪声混叠，是人脸检测中亟待面对和解决的难点问题。综合分析了有遮挡人脸检测方法的研究进展，依据特征构造方法的不同将遮挡人脸检测分为基于手工设计特征的经典方法和基于深度学习的现代方法两大系列；对比分析了不同算法的基本原理，模型性能和存在的问题；探讨了未来可能的研究方向。

关键词: 遮挡人脸检测, 机器学习, 深度学习, 形变模型