Made with 😷 by Hideo Daikoku, Sooji Kim, Kishino Watanabe, Koki Serada, Kohnosuke Yamada
We invented a mask classifier that calls on those who do not wear masks to wear them. COVID-19 pandemic still continues to spread and is concerned about a second wave. However, specific medicines and vaccines are currently under development so the only way to control infections is to wash hands, disinfect with an alcohol-based sanitizer and wear masks.
Some people say that wearing masks is ineffective, but previous researches proved the effect of them. A systemic analysis of published studies found strong evidence during the 2003 SARS epidemic in support of wearing masks. One study of community transmission in Beijing found that consistently wearing a mask in public was associated with a 70% reduction in the risk of catching SARS. Wearing masks is thought to be effective for COVID-19, the same respiratory infection as SARS. The US Centers for Disease Control and Prevention recommends that everyone wear a cloth face covering in public, especially where there is a high degree of community-based transmission. Meanwhile some do not wear masks in spite of this situation because they feel physically uncomfortable. In particular, those who have no custom of wearing masks hesitate to do it. For that reason, we made this system which aims to minimize the risk of an infection by alerting people to wear masks.
We took 10-15 second videos of each type of mask we have available at home. (Pitta, Cloth, Abenomask, Surgical, N95) to use as training data for our teachable machine AI. Teachable machine then uses the training data to detect what is being shown through the webcam and whether it matches the images in the training data.
Our group members posted 10-15 second videos of themselves wearing different kinds of masks on the respective Google drive folder for each mask. Our group leader then exported them as a jpg sequence and input them into the system to train the model. Because we are all in different locations, we had the advantage of taking videos from different background set-ups.
Our goal was to train and develop a mask classifier that accurately warns individuals who are not wearing masks. In the beginning, the system only recognized a few images. We improved the classifier by implementing different images and videos into the system to understand more variations.