Thursday, 11 August 2011

Crowdsourcing the London Riots Facial Recognition

Now first of all the Riots that have been happening across the UK have been awful, destorying communities and peoples lives, however as a community there are ways we can help. The police have begun releasing pictures of the muggers however they are usually low quality CCTV captures. In the day and age of DSLRs and hi-def videos on phones there are plenty of source footage on places such as YouTube.

So firstly we need to get at the footage, that's easy enough using the YouTube API and a select few search terms such as 'london riots' and then returning only new results. A great thing about this is the results can already be highly parrallised per video.

Next we need to use the videos, I would imagine querying a central database with video links that are still needing processing. Once a video has been processed and the faces confirmed, it can be removed from the list.

Actually using the video is a bit tricky, though there are lots of tools available to strip the video from a variety of media sites such as YouTube and save them as a usable file format such as mp4.
Now we have the video the fun begins, using such frameworks such as OpenCV we can scan through the video and recognise faces. For each face generate XML metadata for the position on screen, timestamp and video source and save thumbnail of the face to an image file.

Once processed all the data gets sent back to the server where it can be displayed on a webpage where people can submit information or vote on whether a picture has correctly identified looter.

I hope to make a proof of concept website soon but as you can see the concept behind it is simple, but powerful in helping bring community based justice.

No comments: