Facial Recognition on the Raspberry Pi: The Future Today

By Mike Cook, Jonathan Evans, Brock Craft

Want your Raspberry Pi to do something cool? The human eye can instantly identify characteristics of an individual that tells us many things about that person, at a glance. Within seconds of looking at someone, you typically know that person’s gender, approximate age, and height, and you may be able to identify who the person is (if you’ve met him before). You form impressions about people by the way they look, what they wear, the way they move, and their facial expressions. The human brain processes all these variables in a matter of seconds. Teaching a computer to do all this has been a focal point of image processing research for decades.

Image processing has evolved significantly. Today, a computer can identify facial features like eyes, nose, ears, and the size and shape of the face. These facial metrics (known as biometric data) form a fingerprint-like identifier that is unique to every individual.

The creation of facial biometric data was the first step in creating complete recognition. The second step was being able to match the data to a database of biometrics and associate it with an individual’s identity. The human eye can quickly process facial characteristics, but the human brain can store only a few hundred faces reliably. After a while, we tend to forget people’s names and often need to relearn information about people. Computers, on the other hand, excel at storing and matching data. Facial recognition software has evolved to the point where computers can process an image and match it against a database of millions of people in seconds.

Law enforcement has led the way with the development of facial recognition systems that can identify criminals against a watch list in real-time. If you’ve traveled through an airport recently, chances are, your facial biometric data has been captured and matched against a watch list.

Having your biometric data stored in a database has raised privacy concerns. Storing biometric data without consent has been a topic of discussion and debate for privacy groups for years and, in some cases, had led to the creation of policies to protect a person’s identity. In addition to privacy concerns, there are also fraud concerns. Having your facial fingerprint matched to what is known as metadata (name, address, and Social Security number, for example) is a major identity theft risk. In order to combat this risk, software vendors have created biometric encryption algorithms to encrypt the data within the database and also provide an almost unbreakable link between the biometric data and the metadata.

More recently, image processing research has made advances that will open up information contained in images on the Internet. Search engines are starting to be able to identify the content of a picture and not have to rely on text-based metadata tags that need to be manually created by the image owner. Using image processing, a search engine can now search every image on the Internet. Using pattern recognition software, it can identify the content of a picture and automatically create captions like “male child playing with a dog in a park.”

Consumer products have also gotten in on the act. Cellphone applications have been developed to be able to take a picture of a scene and instantly match it to a database of scenes that can provide more information on what you’re looking at, as well as overlay information like street name or shop name, for example.

Embedded electronic boards, like the Raspberry Pi, now have the processing power required to process facial images in real-time and match them against a watch list. This has made the accessibility of this technology available to hobbyists who can quickly build their own image-processing systems using the Raspberry Pi, a webcam, and some open-source software. Having computer vision on a computer as small as the Raspberry Pi dramatically reduces the footprint of the hardware and opens up new possibilities for how this technology can be deployed.

Wearable technology like glasses with cameras and built-in screens are also starting to come to the fore. Who knows? Perhaps someday you’ll be able to walk down the office corridor and not have to remember the name of the person walking toward you. Just by looking at the person, her name could pop up on the glasses screen so you could greet her with her name. Maybe “someday” is today.