Abstract: COVID-19 is the greatest humanitarian challenge facing the world ever since World War II. World has been fighting the pandemic with great spirit, with the unlocking phases being in motion. Wearing mask properly has become an effective method to cut down the spread of virus transmission. In this work, we are targeting to build a system where we can detect fine-grained state of wearing mask. This work involves three major challenges: 1) different face orientations, 2) various mask types and patterns and 3) facial occlusions. To resolve these challenges we created a new practical dataset, which contains 4000 images. The proposed approach here uses OpenCV and Machine Learning with MobileNetV2 architecture as image classifier to perform mask detection in real time. This model achieves an accuracy of 97%. The new practical dataset created can be used for further advanced models for thermal screening and facial recognition.
Keywords: COVID-19, new practical dataset, OpenCV, Machine Learning, MobileNetV2.
|
DOI:
10.17148/IJARCCE.2021.10785
