Artificial Intelligence and workplace safety measures monitoring are two essential elements for the prevention of COVID-19 contagion.
Reopening of offices, restaurants, sports centers, hairdressers and others can only take place on the basis of rigorous safety measures, aimed at promoting personal hygiene and avoiding the propagation of droplets, which are the main means of contagion. To this aim, companies and shops are installing devices such as temperature detectors, disinfectant gel dispensers, plexiglass separators to ensure safety and operability.
The measures described are useful but not decisive in the fight against the spread of the virus, and the only truly effective measures are represented by the maintenance of social distancing as well as by the use of protective facial masks. It is therefore necessary to make available, at affordable costs, technologies for the automatic detection of the use of the mask and the measurement of the social distance between individuals in indoor environments.
Artificial intelligence for COVID-19 safety measures monitoring
To do everything possible and ensure maximum prevention from COVID-19 in the company, Artificial Intelligence solutions represent the answer to a very common problem today: respect for the precautions that we are all invited to use for common safety.
In fact, in this Phase 2, Artificial Intelligence techniques for computer vision can have a significant impact in the emergency management from COVID-19, which provides for a gradual recovery of economic and production activities.
iGuard: the Revelis solution for the prevention and control of infections in the context of the covid-19 emergency
To prevent COVID-19 contagion, Revelis developed a solution to eliminate the risk of the presence of persons without facial masks or that are not respecting social distance. This solution is called iGuard, and is available as a software platform and as a “all-in-one” device that, based on a neural network for analyzing video in real time, is able to warn the failure in safety measures compliance.
The iGuard neural network
By Computer Vision we mean an artificial intelligence field that trains computers to interpret and understand the visual world.
The neural network developed by Revelis is a convolutional network capable of:
- analyze in real time the individual video frames of a surveillance camera
- identify the presence of faces with and without a mask, (this detection, for privacy reasons, does not provide for the identification of a specific subject, but simply the presence of human faces)
- measure the distance between people in the room
iGuard neural network development
The neural network model is based on the Faster R-CNN architecture which uses convolutional networks and a set of anchor boxes to locate faces with or without the mask. The choice fell on this type of architecture due to the excellent compromise offered between performances and inference speed.
To train the model it was necessary to find examples of images with individuals with and without masks. For this purpose, two datasets have been merged: WIDER used in literature for Face Recognition and MAFA, containing images of faces with and without mask. Once sufficient data was obtained, a Faster R-CNN was trained using Python and the Deep Learning framework PyTorch.
In order to measure the performance of the trained model, it has been used the Test Set provided by the MAFA dataset (excluded from our training) and it has been compared to other models used for the mask/no-mask detection. The metric used is the mean average precision with an intersection of at least 50% with the ground truth object. The benchmark models are the ones released by Baidu and Aizoo.
|Baidu||0.445||Custom Single Stage|
Results must be analyzed in the light of the hardness of the Test Set, as explained by the paper’s authors, and that the Revelis system is based on a Multi-Stage architecture which favours precision despite of the inference time (~1s on GPU).
How iGuard works
iGuard is a software platform, also available as an “all-in-one” device, that prevent the covid-19 contagion by monitoring the respect of security measures within companies.
The “all-in-one” device is equipped with a camera and allows:
- monitoring of working environments by identifying the presence of people – but without proceeding with the identification of individual subjects for privacy reasons
- recognition of failure to comply with security measures, and in particular:
- failure to use the facial mask
- failure to respect social distancing
- Sending an acoustic warning in the face of potentially dangerous situations
Using iGuard device is very simple, because it does not need complex installations or internet connection. It is powerful and extensible and also supports external cameras.
iGuard is also available in server mode, for large organizations that have video surveillance systems with many cameras and intend to monitor compliance with security measures to avoid infection by covid-19.
For more information on iGuard please contact us at firstname.lastname@example.org
(1) Images used in concession by:
Artificial intelligence by Tomi Triyana from Project Noun <a href=”https://it.freepik.com/free-vector- photos/people”> Aerial people created by pikisuperstar – en.freepik.com </a>
Neural Network by David Christensen of the Noun Project Oleksandr Panasovskyi’s convolutional neural network of the Noun Project