MOXA
A deep learning based unmanned approach for real-time monitoring of people wearing medical masks using the pre-installed CCTV cameras in public areas.
Created on 12th April 2020
•
MOXA
A deep learning based unmanned approach for real-time monitoring of people wearing medical masks using the pre-installed CCTV cameras in public areas.
The problem MOXA solves
As suggested by medical practitioners, it is a necessity to wear a mask whenever you are at public places or places with probability of people gathering to prevent the spread of the virus if you are a carrier and even if not, it is safe to wear the mask.
Till now the only way of monitoring the people if they are wearing the mask was manual.Hence to overcome this cumbersome process of manual monitoring of people wearing masks. We designed a real time monitoring technology to detect people wearing medical masks which can be done with the help of image processing, and deep learning. We have come up with an easily adaptable model which can detect the people wearing masks from the images or live feed fed to it. We designed and trained the YOLO model on a custom dataset of medical masks. The training was done on datasets labeled with people wearing masks and not wearing masks, thereby enabling the model to distinguish between the people wearing masks from the ones not wearing them.This can be used to classify places with higher or lower chances of transmission in case an active carrier enters that place and hence required measures can be taken to aware people of the emergency.
It is humanly not possible to monitor thousands of people at all times to ensure whether they are wearing a mask or not. Even to achieve 10% of this goal we would need huge manpower, which is not feasible to arrange in this situation. But with our system, we can continuously monitor large volumes of people with even a small team working remotely, and store the demographic information (storing information with many details like timestamps to increase the accuracy of the model) which could be very helpful in mitigation of this situation.
Our model is extremely cost-effective. As it is primarily dependent on the cameras that are pre-installed so the hardware requirement of this model is minimal.
Challenges we ran into
The main challenge we faced was collecting a proper datasets for training our model as it required classified datasets of people wearing masks and not wearing masks that too from realiable sources. The datasets of images of people wearing masks from kaggle helped a lot in this problem and batch downloading from google images was also a way to overcome this hurdle in the project which was the foundation for the entire project.
The second challenge we faced were to find a perfect algorithm keeping in mind the accuracy and the time taken. So, testing with several algorithms YOLO v3 algorithm with the darknet Neural networks proved as the perfect one for the project.
The Final challlenge was the testing as in the current condition of lockdown people are not allowed to go out of their houses. So, getting such conditions were tough. But, we managed to create such circumsatnces through online conferences, video footages available and through edited image structures depicting various conditions of people wearing and not wearing masks.
Technologies used