0

The Use of Facial Recognition Glasses by Law Enforcement3 min read

The use of facial recognition technology in wearables by law enforcement is more common than one may think. Brazilian police officers have been using facial recognition glasses to monitor citizens at large events since the World Cup 2014. And as reported yesterday (7 February 2018) by the Wall Street Journal, Chinese law enforcement officers have been testing facial recognition sunglasses to track travelers during the Lunar New Year migration.

The facial recognition technology allows the police to instantly match a tracked person’s face against the database of photos of criminals or people wanted by the police. For example, according to LLVision, the Beijing-based manufacturer of the facial recognition sunglasses for the Chinese law enforcement, the time it usually takes to perform this matching against a pre-loaded database of 10 thousand records is around 100 milliseconds. The technology performs well at distance and in crowded areas such as a busy street, subway station, or airport.

How does facial recognition work?

The facial recognition technology is based on machine learning, a technique within the AI domain making it possible to design an algorithm (data classification model) and train it on a set of labeled data so that it picks up patterns, associates these patterns with the labels, and classifies new unseen piece of data by assigning a label to it. To exemplify, imagine we have a database of 10000 photos comprising same-size collections of photos of individuals that we are interested in detecting, e.g. criminals: let’s say, 1000 collections of 10 photos of a criminal, each collection and photo labeled with the name of the corresponding criminal.

The task is to use the collected data to train a data classification model that detects whether a new photo is of a criminal from the set and, if so, of which criminal. The model is trained on the complete database of labeled photos, during which the patterns associated with a specific person are discovered and remembered by the model. The learning process is repeated a number of times, from hundreds to thousands to hundreds of thousands depending on the complexity of the underlying model selected. During every subsequent iteration, the accuracy of the classification is slightly improved since the patterns picked up by the previous iteration are passed on to the next one. The process results in a trained data classification model that can, with a certain accuracy, detect whether a new (not seen by the model before) photo is of a criminal in the database and, if so, of which criminal.

Given the recent advancements in machine learning-based computer vision, the facial recognition technology is getting better at filtering out the “noise” such as disguise, be it a hat, scarf or even a mask, making it harder for criminals to circumvent detection.

Embedding a facial recognition model into a wearable device such as police sunglasses makes the technology more efficient. Real-time detection feedback is instantly delivered directly to the law enforcement official who is in close proximity to the target, which is contrary to CCTV tracking where the possibility of the criminal leaving the area by the time he is identified and the officer is sent to the location is higher. In addition, the wearable’s inherent mobility enables surveillance in areas where fixed security cameras are not installed. According to the Chinese law enforcement, the use of facial recognition sunglasses by the police has already helped to apprehend 7 suspects and 26 individuals traveling under false identities.

The use of the facial recognition technology by law enforcement in wearables such as eyeglasses raises privacy concerns. While such use may be justified by national security interests, a proper regulatory oversight must exist to ensure the compliance of the processing of the collected personal data with data protection principles and non-infringement of the data subjects’ rights.

Sergii Shcherbak

CTO @ Maigon.io

Leave a Reply

Your email address will not be published. Required fields are marked *