The Evolution of Video Forensics: Finding People in Real Time

Rony Vexelman, Campaign Management and Product Marketing Associate, Qognify // September 22, 2014
Share this
Facebook LinkedIn Twitter

The other day, we read about a manhunt in Phoenix, Arizona, in which two suspects from a gas station shooting fled to the Sky Harbor Airport. One was caught swiftly but the other eluded police. The busiest terminal was locked down for three hours while police searched for him, delaying 69 outbound flights, cancelling 13 others and inconveniencing thousands, to say nothing of the risk to all involved.

The airport probably has hundreds if not thousands of surveillance cameras, but it’s not surprising that police had to resort to a feet-on-the-ground search to find the suspect. What was the alternative – to watch all the recordings from all the cameras to try and locate him? Impossible – just 400 cameras and three hours of elapsed time would mean 1,200 hours, or 50 days of footage to plough through.

This is a situation familiar to all first responders, whether they are looking for a suspect, a witness to a crime, or even a missing child. So how can they leverage those cameras and recordings to find people faster when time is of the essence?

Here are six ways NICE Suspect Search uses video surveillance footage to help officers find out where suspects are right now, and also show where they were before and after the incident. Use the whole person, not just the face.  When searching for a given person it is tempting to use facial recognition techniques. But what if the person is walking away from the camera, or we are only catching a side profile, or they have a hat on? Working with information about the entire body, from head to foot (clothes, accessories, skin, hair) enables faster and more accurate matches.

Do the indexing work in advance, before you need to search. When every second counts you cannot waste time analyzing video, so Suspect Search does the analysis and tagging work as recording takes place. Searching in real-time then becomes a simple task of probing the database of tags, which by my measurements is about 3,000 times faster than trying to analyze the video. Know what you are looking for. To find someone, you obviously need a reference. Ideally this would be a clean photograph, or at least a snapshot from a video. But what if all you have is an eye witness account?

Suspect Search lets you create a composite of an entire person. This cartoonish character might have blue jeans, a white t-shirt, short black hair. This is enough to narrow down thousands of possible hits, where a human operator can easily pick out a match. Search cameras in parallel. Suspect Search processes multiple camera recordings in parallel, rather than one at a time, because it is designed to overcome the differences between the cameras, such as varying lighting conditions or different perspectives.

Search using different reference images. As you start to spot the suspect on different cameras, you build up a library of snapshots from different angles. The snapshots might also have different lighting, so what was a white stripe on a black shirt might be blue stripes on black in another. When you search recordings using different reference images, you have an even greater chance of getting a good match, which means finding the suspect faster. Tracking. As the search reveals snapshots of the suspect over time across different cameras, it evolves into a history of their movements before and potentially after the incident.

Timely situational awareness is critical to officers which is why every snapshot in Suspect Search is presented on an actual map. This answers questions like how did they get here? Where did they go? Did they meet up with anyone or were they alone?

During an unfolding incident Suspect Search tells you where a person is right now, according to whatever description is available. That is true, real-time insight based on fast forensic analysis of recorded video. And, it leverages your existing investment in video surveillance cameras.
If police in Phoenix had access to this technology, they would have been able to use a rough description of the suspects to search for them, and maybe even catch them as they entered the terminal.

A three-hour lockdown might have been just one hour or perhaps avoided altogether. If you can search quickly enough, you can often stop things from happening, like a lock down or subsequent exchange of gunfire, in the first place.

Share this
Facebook LinkedIn Twitter