Police in Tamil Nadu’s Tiruvallur district is using an app that uses facial recognition and geo fencing to keep track of people placed under quarantine, Superintendent of Police for Tiruvallur district, P. Aravindhan, told MediaNama. Called CoBuddy, the Android-only app validates a person’s location using GPS coordinates, and as an added layer of authentication, uses face verification to ascertain their identity. Tiruvallur police has been using the app since March 24, Aravindhan said.

About 150 people currently placed under quarantine in Tiruvallur are already using the app, Aravindhan said. They have to provide the app access to their GPS location and enter an image of their face. After that, the police sends them prompts at “random hours, 2-3 times a day” to verify their faces, he informed us. The facial recognition helps the police in making sure that it is indeed the person who has been asked to quarantine, he said. The app continuously checks for the location of the phone and sends an alert to the police if a person steps outside of the geofence, which can be set from anywhere between 10 metres to 100 metres, according to promotional material of the app which was shared with MediaNama. The police can also see a heat map of all quarantined/affected people on a centralised map by district, by area on a centralised web dashboard.

Apart from Tiruvallur police, police in Thiruppattur, Kancheepuram and Chengapet are also using the app, while police in Tirunelvelli district is also planning to use it, we were told. The app has been developed by NotionTag Technologies, the same company that developed the FaceTagr software — a facial recognition app that the Tamil Nadu police uses to identity criminals. We’ve reached out to other police stations for more details.

CoBuddy uses GPS geofencing and facial recognition to validate that a person is actually at a particular location, Vijay Gnanadesikan, co-founder and CEO of FaceTagr told MediaNama. The reason for using face verification on top of tracking a person’s GPS is that “people [who’ve been asked to quarantine themselves] can, and are leaving phones at home. [CoBuddy] will make random alerts to the quarantined to verify themselves with face recognition. Basically someone has to be with their phone to verify and they have to be in their zone to get verified,” he said.

A close look at the app

At the time of publication, the app had been downloaded over 1,000 times. The police provides a link to download the app to the quarantined people along with an OTP to set up their profile on it. We were, however, able to test the app using the download link that Aravindhan sent us, and here’s what we found:

App didn’t open until phone’s location was turned on: After installing the app, it did not open until we turned our phone’s location. It then asked for permission to take pictures and record videos, before asking for location access. It also asked permission to access photos and videos on the device. After allowing access to all these things, we were provided with the option to sign in to the app using a mobile number and an OTP. After signing in, the app asked us to input an image, although we could do so only using an image from the phone’s gallery. It did not have an option to click an image using the phone’s camera.

L to R: App didn’t open till GPS was turned on | The request for access to photos | Inputting an image to the app

Denying access to track GPS will “become a flag”: The app asked for access to “physical activity”. Gnanadesikan told us that this means that the app will continuously track the phone’s GPS, and if anyone denies permission to this request, “it will become a flag”. “We are not basing our triggers on continuous GPS coordinates. [It] might become handy when needed, if someone is out and the police have to go after them,” he said.

Home screen is minimal: The home screen of the app has three options — to open the front and rear camera along with a “I need help button”. Aravindhan said that by using the latter button, a person can get in touch with the police in case they face any problems, and so far, the police had helped 2 people who used that button. There’s also a settings button on the home screen, but that needs a password to access, which we didn’t have access to.

The app’s home screen

Accuracy of the facial recognition system: I tried checking how accurate the face recognition system was, and it seemed to verify my face well. It showed a “not recognised” message once in several tries. Vijay told us that it was likely because the app might not have seen my face. However, I was able to trick the face verification by showing it an image of my face from another screen. Also, the verification process happens automatically, meaning that I just had to hold my face in front of the camera and not press any buttons.

  • When we asked about incorrect face matches, Aravindhan said that they are yet to find a case of the app making an incorrect match. “If there is an incorrect match, then the app won’t allow to take a pic[ture],” he said.

L to R: How the face verification interface looks | When there’s a successful match | When there’s an unsuccessful match

Developer claims app doesn’t store any personal data

The app is end-to-end encrypted, Gnanadesikan claimed. The app’s promotional material says that the app uses 1:1 verification method, instead of the 1:n verification method, meaning that it only authenticates a person against a photo provided by them, instead of verifying it against an entire database. Gnanadesikan said that as of now, they only have the 1:1 verification, and have not included support to verify multiple faces using the same app on the same device.

Gnanadesikan also said that the photo uploaded to the app while first setting it up is converted into a “face template — mathematical model” and then “destroyed”. We couldn’t find a terms of service or privacy policy in the app, but there is a message on the app’s home screen which says: “The photos are not stored”.

“The police doesn’t have access to a person’s private information. They only have their unique identifier and GPS coordinates, using which they can verify the identity of a person,” Gnanadesikan told us. The “mathematical model” itself is stored on the device, and the verification part happens over the cloud, he added.

The platform is easily scalable: Gnanadesikan

The company never had to “demo” the app to the authorities because it worked closely with all SPs of Tamil Nadu to understand what kind of an app they wanted,” Gnanadesikan said. “It was more of an iterative approach”. He told us that the company already has a face verification based enterprise attendance solution, and it only had to tweak it slightly to develop this platform. The development itself, took them around 3-4 days. Vijay told us that they can easily scale the operations of the app us since their solution is based on a standard platform. “If we can use the app in one district, we can very easily distribute to other districts”, he said.