HYDERABAD: Although it is proving to be a challenge to identify people who wear masks using facial recognition cameras, the police said it has not been impossible and that other parameters have been helping them in identifying people.
According to high-ranking officials, the facial recognition cameras are identifying people using other key characteristics like their physicality, forehead features and others in the Police, Home and IT Department.
Larsen and Toubro has been helping the government with various smart solutions, including facial recognition, AI-based crowd control and others. Speaking to Express on condition of anonymity, an L&T official said, “Any facial recognition camera uses deep-learning techniques. It works on probability, which means that algorithm has to be trained to detect such faces.”
“Detecting faces in the present scenario is easier in airports where there are cameras at the human height level, but with CCTVs, it gets harder. In airports, the lighting is always optimum, but with CCTVs across Hyderabad, they are at a distance and the lighting changes constantly. So the accuracy dips,” the official added. A senior official in the Home Department said, “The accuracy will definitely decrease because of the masks, but we can still manage.”
The technology to identify persons in disguise (masks), more accurately than others, was first explored in a 2017 research done by Amarjot Singh, a Cambridge University student. The study notes that in facial recognition, there are 14 key points on the face. “They are the eye region: left eyebrow outer corner, left eyebrow inner corner, right eyebrow inner corner, right eyebrow outer corner, left eye outer corner, left eye center, left eye inner corner, right eye inner corner, right eye center, right eye outer corner; nose; and lip region: lip left corner, lip centre, lip right corner.”
The study showed that when a person is in disguise or is wearing a mask, it detects the exposed facial key points and aligns them, and then compares with available dataset to arrive to an identification.