High quality anonymization for autonomous driving systems

Autonomous vehicles are the future of road transport. There are estimated to be over 30 million cars currently on the road globally that have some level of autonomy, and this number is expected to continue to grow steadily. But this doesn’t mean they are all completely self-driving.
An autonomous vehicle can be anything from basic assistance at level 1 through to full automation at level 5, as the image below illustrates.

The long-term goal for many manufacturers is to reach level 5. But what does it take to develop fully automated advanced driving systems, and when can we expect to see them on the roads?

Data, data, and more data

There are, of course, many different factors that go into the creation improvement of autonomous driving systems, but one thing that is absolutely essential is lots and lots of data.
These systems aren’t simply built, they’re trained over long periods of time. Yes, you need to create the initial model and give it various rules to follow, but then you need to hand it over to the neural networks and machine learning algorithms that fine-tune the performance to the point where it meets all safety standards.
And it’s vital to use real-world data because these systems will be operating in the real-world, not a simulation.

Capture, anonymize, learn

Video footage is key, as it allows the systems to understand the behavior of other vehicles and pedestrians to improve its performance in any given scenario. Most importantly, it helps the systems learn to prioritize human safety in any situation.
But herein lies a problem.
This data contains vast amounts of personal identification information. Namely individual’s faces and vehicle number plates
Regulations across the world, such as GDPR in Europe, ban the storage of this data without user consent. Tracking down all these people and asking them for permission is obviously not possible, so instead this information will have to be blurred out, right?
Wrong.

Simple blurring severely reduces the value of the data for machine learning purposes. Without facial expressions you can’t see which way pedestrians are looking, whether they are aware of the vehicle, and many other factors which influence the quality of the autonomous system. And if number plates are blurred that will affect the system’s ability to correctly recognize other vehicles and act accordingly.

This is why Simcenter Autonomy incorporates cutting-edge Deep Natural Anonymization from brighter AI. This software replaces faces and license plates to prevent identification while retaining maintaining data quality, such as facial expressions, for machine learning.

It’s fully GDPR compliant, meaning data can be stored without any legal concerns and used at any time for developing new autonomous driving systems.
You can read more about why Deep Natural Anonymization by downloading the white paper here .