How biometric identity verification protects passengers using
ride-hailing apps

Viktor Bielko

Product manager at Innovatrics

To get home one night, you order a taxi through a ride-hailing app. You don’t check who your driver is because you don’t have to. With driver face verification and liveness check algorithms, you know that the ride-hailing app continuously verifies thousands of drivers, making all the rides you take safe.

Ride-hailing apps made earning money as a driver and saving money as a passenger easy, and all just a few clicks away. But with the growing popularity of these services and increasing numbers of registered drivers, ensuring passenger safety has become a priority for several of Innovatrics’ clients: companies that offer ride-sharing through smartphone apps, and who are prioritising reliable driver verification worldwide through the use of biometrics.

Boosting trust by verifying drivers

Consider this scenario: It’s late at night and you need a ride home. You step into a taxi, and it speeds off. Do you ever pause to think about the identity and trustworthiness of your driver? These are the critical concerns that responsible ride-hailing apps centre their businesses around. They work hard to ensure you feel safe, because their business relies on your trust.

This trust is not to be underestimated. These companies expend significant resources in terms of time and finances to ensure passenger safety. Driver verification, i.e. making sure you get a ride from the same person displayed in the app, is one of the most important aspects of building and maintaining trust.

However, with the prevalence of fake accounts and instances of driver impersonation, how do these apps enhance their security and deepen trust in their services?

“The problem is that some drivers don’t follow the rules properly and can get banned. Then, some try to find a way around the ban by creating new accounts.”
Viktor Bielko, Innovatrics

Test run

Ride-hailing apps that collaborate with Innovatrics always test how well Innovatrics’ algorithms work. They do this by sending a huge dataset of pictures to see how the algorithms handle three tasks: checking if a selfie matches an ID, comparing one selfie to another, and identifying photos in a database.

Uncovering fake accounts with biometrics

Ride-hailing companies check their drivers regularly and ban them from working with the service if they fail to provide an up-to-date ID, have committed a crime in recent years, behave inappropriately, or receive persistently low ratings from passengers. The problem comes when these banned drivers try to side-step the ban:

“Some drivers don’t follow all the rules properly, which leads to them getting banned. They then often create new accounts to find a way around the ban, impersonating their family or friends. We consider this to be a serious safety issue that can be solved with biometric face verification,” explains Viktor Bielko, product manager of the DOT (Digital Onboarding Toolkit) product in Innovatrics.

A reliable and powerful remote identity verification system enables the ride-hailing apps to safely verify the identity of each driver at any point in time. The verification process works as follows:

any driver can receive a notification at any given moment asking them to safely stop and take a selfie,
the selfie is then compared to the driver’s ID,
at the same time, the system performs a liveness check.

The selfie to ID verification and liveness detection is part of Innovatrics’ digital onboarding tool DOT, which is a software that enables facial registration, extracts data from IDs, performs facial comparisons and verifies the liveness of each registered individual.

Liveness detection – the easiest way to avoid spoofing

The role of a liveness check is to make sure the system is dealing with actual people, not fake images. To illustrate its importance, this is a typical scenario where a liveness check proves critical: imagine a driver who is banned because of consistently bad reviews. They don’t want to lose their income, so they create a new account in their friend’s name and ID. They also have a few photos of this friend on their phone. When they’re prompted by the system to verify their identity by taking a selfie, they simply use the friend’s photo, and that’s it, right? Well, not with a proper liveness check.

“A presentation attack occurs when someone tries to trick the system by using a fake representation instead of their real face. To detect such attacks, we train our detection algorithm on huge amounts of data to uncover the use of 2D or 3D masks, paper photos or images of screens. That way, we can stop potential fraud or identity thefts,” explains Viktor.

Proof of reliability

Innovatrics’ liveness detection algorithm, which is trained to detect identity fraud, has passed tests by iBeta – an independent biometrics testing lab. The submission has attained the highest possible level of accuracy in iBeta testing, correctly detecting all 1,500 attempted spoofs and accepting all of the genuine logins. Simply put, the algorithm rejects 100% of false presentations and doesn’t mistakenly reject anyone who is indeed present.

The most common presentation attacks that
ride-hailing apps face

It’s crucial for ride-hailing services to tackle these common tricks to keep their platform secure and make sure everyone has a safe and trustworthy experience, whether you’re a passenger or a driver.

Active and passive liveness checks

There are two types of liveness detection:

Active check is based on action. You ask a person to perform a certain task, for example, smile, follow the dot with their eyes, etc.

Passive check evaluates liveness without the customer ever noticing it. It makes the interaction simple and hard to spoof as the attacker does not know if and when liveness verification has transpired.

The passive approach to liveness detection considers hundreds of variables, such as light, shadows, the texture of the skin, and other significant factors. Active check on top of it uses multiple photos, where a difference between the photos has to correspond with the action required from the user.

The biggest challenge for verification – photo quality

One of the issues that makes it difficult, and sometimes even impossible, for the algorithm to perform the liveness check and verify the driver, is low photo quality. “Comparing selfies with IDs is not the problem, photo quality is. Drivers usually take selfies inside the car in bad lighting conditions, and often during the night. Sometimes they take extremely close-up photos of their face. Other times it’s the opposite; the face is too small,” explains Viktor.

However, thanks to the close cooperation with the ride-hailing apps, Innovatrics identified these problems at the very beginning of the project. The R&D team created an analysis that summarised acceptable thresholds of quality to ensure reliable verification. The team analysed the photos’ brightness, contrast, sharpness and other attributes.

Another issue ride-hailing companies struggle with is quick feedback. The goal is to keep the app as small as possible so it doesn’t take up a lot of space on the driver’s device. For this reason, the verification process occurs on remote servers, not in the app, meaning that feedback on the quality of the photo is not in real-time. If incorrectly denied, drivers must wait for information about what went wrong. There is an update in implementation that makes this process smoother and more user-friendly.

“Comparing selfies with IDs is not the problem, photo quality is. Drivers usually take selfies inside the car in bad lighting conditions, and often during the night. Sometimes, they take extremely close-up photos of their face. Other times it’s the opposite; the face is too small.”
Viktor Bielko, Innovatrics

Servers are key to good performance

As many ride-hailing companies use servers for the verification process, the right server configuration is essential. “It is important to have the hardware set up for high availability, so in case of any errors, other servers can cover for the lost power,” explains Viktor. It is also important to know how many operations the servers need to be able to manage.

In this scenario, verification is evenly distributed over time, as companies manage notifications independently. Presently, the system handles approximately a hundred driver verifications per minute.

Making the product better

Innovatrics offers ID document reading and OCR algorithm that reads text and extracts faces from approximately 300 different types of official documents and IDs. However, the solution faced a challenge due to the diverse range of nationalities and documentation of registered drivers. “In response, we delivered the algorithm to only look for the face portrait and omit the document reading. It makes the whole process easier and faster,” adds Viktor.

Thanks to partnering with ride-hailing companies, Innovatrics improved its face recognition and liveness testing product. Many of these partnerships are still ongoing, with the completion of the verification and liveness check marking the initial phase, and the collaborations progressing into the second phase with 1:1 (selfie vs. selfie) verification, and subsequently the third phase, where the identification is based on database information, ensuring continuous improvement.

AUTHOR: Jana Nováková
COVER: Elizabeth Bugyiová
COVER ANIMATION: Dominik Prokop
INFOGRAPHICS: Aleksandar Perišić

Sources

How biometric identity verification protects passengers usingride-hailing apps