The face mask detector focuses on the partial facial features, so the public needs no stopping for the detection. (Picture provided by NCTU)

During the COVID-19 pandemic, Taiwan Centers for Disease Control (CDC) encourages the public to wear masks. Some crowded places, such as the MRT stations, campus, or commercial buildings, require massive human resources to monitor face masks.

Acknowledging the disease control needs, Prof. Wen-Huang Cheng and Prof. Hong-Han Shuai work together on the "Large-Scale Face Mask Detecting System," a leading-edge technology in Taiwan, to detect whether people wear masks. The system only relies on one camera and has an accuracy of 95% in a crowded environment.

Cheng said that hands are the medium of virus transmission, so the emergence of COVID-19 urges the rise of contactless technology. His research team has devoted to prospective studies such as AI, machine learning, and computer vision for a long time. He applied the existing technologies to the contactless mask detection system. He spent two weeks to complete the system, which only relies on one camera to determine whether the crowd wears masks.

His research team did the preliminary examination in crowded environments, such as cafeterias and classes at National Chao Tong University. When people appear in the camera frame, the system points out the ones with masks in green rectangles and others without masks in red rectangles. The accuracy is up to 95%, no matter if people show their side face, phubbing with their phones or resting their head and arms on the table.

According to Cheng, the face-mask detector''s novelty is its focus on the partial facial features, such as the edge of eyes. In this manner, once the system detects the critical part of human faces, it can capture the face and determine whether the person is wearing a mask, no matter what sizes or postures of faces. Moreover, the detection can be done within 1/30 seconds in a large-scare crowded environment.

Shuai points out that conventional face-detection technology applies machine learning, relying on high-resolution photos with only one person in the frame and complete information. If the face in the photos is too small, blocked, or not the front face, the detection might fail. People need to approach the camera to get accurate detection. However, the newly proposed face-mask detector doesn''t require people to approach the machine.

During the pandemic, the government applies the epidemic prevention measures at all costs: basic ear/forehead thermometer, infrared camera, face masks, and guards who check the wearing of a mask. Shuai believes that it is possible to automate epidemic prevention with masks and thermometers, reducing the need for human resources for guarding the entries. With the system, human resource cost can be reduced, but people also need no stopping for measuring body temperature. The efficiency of the measurement will be increased.

However, according to Cheng and Shuai, the face mask detector can only supervise the mask''s wearing. It still requires the public to wear the masks spontaneously to protect themselves during the pandemic.

The accuracy of face mask detection can reach up to 95%. (Picture provided by NCTU)The automation of epidemic prevention can reduce the cost of human resources while increasing the measurement''s efficiency. (Picture provided by NCTU)