Performance evaluation of YOLOv2 and modified YOLOv2 using face mask detection

Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) also known as COVID-19, the recent pandemic, brought about anxiety in the past few years due to its contagious nature. This virus can spread from human to human through droplets and could be airborne too. A detailed study suggested that using face masks is the foremost containment measure to prevent COVID-19. Following this research, many countries have made it mandatory to wear masks in all public areas which encouraged people to explore face mask detection as a potential application to monitor people in public places in order to reduce manpower. Object detection is an extensively used Image Processing Technique used to detect the objects of a certain class in an image or video and Face mask detection is one such Object detection application. Some of the recent and advanced face mask detection approaches may be designed using Deep learning Technology. One such state-of-the-art object detection algorithm is the You Only Look Once (YOLO) algorithm. YOLOv2 is faster and accurate than original YOLO. Along with implementing mask detection using YOLOv2, it is also implemented on the new proposed modified YOLO. The proposed Modified YOLO algorithm is implemented on a network whose architecture is less complicated than the original YOLOv2 with its convolutional layers being reduced to (1/6)th the original. A comparison in terms of accuracy and detection time is done between Modified YOLO and the YOLOv2 algorithm. Also, a performance evaluation of the modified YOLO with a variation of certain hyperparameters is done.

Automated summary

Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2), also known as COVID-19, has caused anxiety in recent years due to its contagious nature. Using face masks is considered the primary containment measure to prevent the spread of COVID-19. Face mask detection using object detection techniques, such as the YOLOv2 algorithm, provides an effective way to monitor people in public places. The modified YOLO algorithm, with a simplified architecture compared to YOLOv2, shows promising results in terms of accuracy and detection time.