By Staff Correspondent
The recently released Ministry of Defence (MoD) report on ‘AI in Defence’ mentions the Facial Recognition System under Disguise (FRSD), developed by the Defence Research and Development Organisation (DRDO) to address the issue of facial recognition in the wild, where there are wide-ranging issues like low-resolution cameras.
According to the report, various agencies under the ministry developed FRSD and other facial recognition systems for the Indian Army. Some of these technologies are meant for civilian use too.
The report indicates that this technology will rely totally on algorithms which will be used to identify the person from low-resolution surveillance camera feeds, which are patchy too. The local security agencies can use it for searching across large repositories and can be deployed in restricted zones and security areas. DRDO has developed a technology that can penetrate under wigs, sunglasses, masks, disguises, and hats.
The FRSD system deployed by the Indian Army is geared toward surveillance, garrison security and population monitoring and can be set up remotely with a field-ready system, does not require internet connectivity and is capable of gathering intelligence from multiple sources. This state-of-the-art system can be deployed in disturbed areas and will help in continuous monitoring and surveillance.
The Facial solution has increased with the advancements in camera technology and computers. Presently, FRT uses various nifty algorithms to adjudge the similarity of two objects in two different pictures. These algorithms usually quantify a face using fundamental algorithms like edge detection, contouring and feature mapping to recognise where a face is in an image. It is how many phones put a box on their faces when you use the camera to capture a group photo.
According to experts, FRT looks further at the specifics of the face that generate multiple data points. In conjunction, these various data points are unique for each individual. The facial data gathered from specific demographics can be used in training FR algorithms of Artificial Neural networks. This is meant for ethnicity-based FR accuracy. Also, different social media platforms which gather users’ details based on the searches, as well as photos or comments posted, end up transferring vast amounts of personal information, which includes not only the location details but other personal data too.
At the same time, experts constantly have challenged the accuracy of FRT because FRT never actually produces a right or wrong statement. Instead, like any other recognition algorithm, it creates a confidence number, usually in percentage, indicating the match’s likelihood as a tool and leaving the reliability decision for the operator to decide.
India is expanding its FRT implementation with more states floating tenders for implementation. Although there is no integration yet, there are reports that an integrated nationwide system is in the pipeline. There are also projects like Seeker, which are standalone systems for disturbed areas that can identify individuals and aid surveillance.