RIT scientists develop technology to analyze police body-cam footage

ROCHESTER, N.Y. – A new million dollar project focused on providing insight on what police officers do in their day to day work is being spearheaded by RIT scientists.

News10NBC’s Marsha Augustin spoke to one of the RIT professors about the machine learning techniques they have developed to analyze body worn camera footage.

RIT scientists say RPD produces about 70,000 hours of footage every day- too much body worn video for officers or supervisors to review it all.

So, the hope is with this new machine learning it will help flag videos, monitor the entire video footage, and help the public better understand what officers encounter on a daily basis.

“There’s many interactions that police officers have with citizens and we don’t really know about the ones that don’t turn into a use of force,” says RIT scientist, John McCluskey.

John McCluskey says university scientists have developed machine learning techniques that will analyze police body camera video. The project has been three years in the making.

Some of the goals is the use of algorithms to transcribe body-worn camera audio, flag specific words and algorithms to detect biased behavior by RPD officers, and identify effective de-escalations tactics used by officers.

“I think it’s fantastic that the Rochester police department is partnering  with Rochester Institute of technology,” says McCluskey. “I think its a good sign for maybe the way that public policing will be informed by sort of civilian insights and scientific method.”

McCluskey explains how the technology works

“The machine learning that is applied to the video will allow us to look at the audio and video channel and think about cases that encounters that start at very similar places but wind up at different outcomes,” says McCluskey.

RPD began implementing its body-worn camera initiative in 2016 to improve police community relations. They currently have about 500 body worn cameras assigned to officers.

“We want to ultimately have people code videos that have the beginning that is very similar but generate different outcomes to figure out what branches conversations might take or physical actions might take that generate outcomes that are deescalated verses escalated,” says McCluskey. 

One of the challenges will be the computer vision, because the camera is worn on the body, 

“Normally if its a fixed camera it’s a little easier for machine learning to disentangle what’s going on with the jitter,” McCluskey explains. “I think that’s going to be a harder set of algorithms.” 

RPD says they are lucky that its location allows for easy access to world-class research universities.

To learn more about RPD’s body worn camera initiative, click here.