[ML/CV/ENG] Face Mask Detection

Intro

---

Based on Face Recognition Network(FRN), we can easily detect human face. We can do many things with FRN such as human tracking, unlocking cellular phone, and etc. Moreover, Covid-19 brings new era, world with mask. So I try to make new open-source SW, face mask detector. I know there are already many face mask detector SW, but I think face mask detection problem is not that simple which means it's not 'yes or no' problem. People should wear mask properly. They should cover their nose and mouse at the same time. Special thanks to many wonderful FRN open-source SW, I can reduce development time for face recognition.

 

Flow

---

 

Flow Chart

FRN is based on CNN, so they make several feature maps(f-map) every layers. One of them might be 'nose' f-map, 'mouse' f-map, and 'chin' f-map.

 

So, I can get two additional inputs for second classification by analyzing every f-map.

 

Before analyzing each f-maps, it would be better to compare Face Recognition scores between no mask, mask without nose cover, and mask without mouse cover. Maybe, the order will be

 

No Mask > Mask without chin> Mask without mouse > Mask without nose > Mask with nose cover

 

Which means mask hide some features of face, so that the score should be lower.

 

Reference

---

• Test SW

동영상 프레임 추출(파이썬 코드)

 

• Face Recognition Network

COVID-19: Face Mask Detector with OpenCV, Keras/TensorFlow, and Deep Learning - PyImageSearch

 

• Issue for 2nd evaluation : Training, Evaluation by f-map score, just multiply