Removing Personally Identifiable Information From Video Streams

With the advancement in technology platforms and digital tools, like cell phones, cameras, the internet, e-commerce and social media, there has been an explosion in the supply of all kinds of data. However, with the rise of such ‘big data’, there have been concerns over how sensitive information is properly managed by companies.  Individuals are looking for ways to stay anonymous, and regulatory bodies are seeking new laws to protect such Personally Identifiable Information (PII) – any information that can be used to distinguish or trace an individual’s identity.  General Data Protection Regulation (GDPR) in the EU governs the way in which we can use, process, and store personal data.  Now we need a way to easily implement solutions when capturing visual data across technology platforms and digital tools.  

While other forms of technology can intrude into our privacy, computer vision can help protect our privacy by obscuring objects of interests. With computer vision, the very mechanism to detect objects, people, and sensitive information, can also be used to obfuscate that information in a digital image or video.

There are a variety of elements within an image or video that can reveal personal information, besides identity.  For example, a computer or smartphone screen may display sensitive information that needs to be obscured. But the most common use case for obfuscation is for faces and license plates.

To protect and properly manage PII, you can easily use Matroid’s API to obfuscate one or more objects within an image, live stream, or video. A very useful and powerful tool for protecting people’s privacy and it’s simple to use. In a nutshell, any person or object(s) that Matroid detects, can also be obfuscated by using a simple script (https://gist.github.com/rwong-matroid/9b634e2b5aebf3c837f927dbc49f6983).

The script takes an input video, a detector, a set of labels, and corresponding thresholds.  The script then classifies all the frames using the Matroid API and generates a new video with all the desired labels obfuscated.

Below are examples of running the script to obfuscate faces and license plates. You can use any color to obscure the object of interest; we chose green for better viewing.

The ability to obscure PII personal identifiable information is a great tool to have in your arsenal. Go ahead and try it out.  If you need assistance, email us at help@matroid.com. We’ll be happy to help!

Team Coco (2012, April 24). When Simon Helberg met Stephen Hawking – Conan on TBS. Retrieved from https://www.youtube.com/watch?v=dKmoMr8E4ZY