I. Introduction
Face recognition is a crucial way to achieve non-intrusive identity authentication. It can obtain identity information from a long distance without being perceived. Model performance under ideal conditions has achieved excellent results, and mature algorithms have been commercialized to provide a secure identity authentication manner. However, general technologies are hard to achieve optimal effect in unconstrained environments, where objects may encounter problems such as expression change, extreme illumination, large pose, and occlusion Error! Reference source not found.. Face detection directly affects the subsequent recognition task. Obstructions such as glasses, scarves, and masks increase the probability of missed detection, which is not conducive to epidemic prevention and control.