Use a mugshot to automatically identify suspects in all bodycam video

Police departments around the country are generating more and more bodycam footage, and companies like Axon (Taser) and other manufacturers are building software to help law enforcement wrangle what is essentially a big data problem with video.

Bodycam footage is often used as evidence, or at least as proof of a certain set of actions. But it can also be used to help identify suspects, piece together more evidence of a crime, and generally act as a source of knowledge. Today, the idea of going through evey second of body cam footage looking for known suspects is near impossible. That is where some machine learning can help.

Let’s look at a quick and easy way to search through a massive collection of videos with a face.

Let’s take this mugshot as an example.

http://www.tmz.com/2017/07/11/shia-labeouf-arrest-bodycam-footage-savannah/

I’d like my software to search all the bodycam and security camera video I have for instances where this man appears. For developers of this kind of system, here’s a great, simple way to get started. The rest of us can try it pretty easily as well.

1. Head over to http://machinebox.io and sign up. Follow the simple instructions to download and run Facebox (state-of-the-art facial recognition). ~ 10 minutes
2. Use the console or the RESTful API to teach Facebox that this is Shia LaBeouf. ~5 seconds
3. Find some bodycam footage you want to test. ~ 3 minutes
4. Use the Facebox console or RESTful API to post a frame (or the entire video). ~ 1 minute
5. Behold the results:

A screenshot of a UI I built to test images with Facebox

Using Facebox and a bit of integration work, you can build software that lets law enforcement agencies quickly find footage in which a given suspect appears. They might want to search by the person’s name, or by a sample image using visual similartiy search capabilities. They may even want to stream live footage to Facebox to be immediately notified if a suspect is within field of view of a security camera.

With state-of-the-art machine learning and facial recognition, a whole world of possibilities opens up.