by Lucy Pan, Prateek Kulkarni, Biddappa Muthappa, Stephanie Pinto, Ashutosh Kumar, Diptarka Roy, Anshul Thapa, Meenakshi Deenadayalan, Aishwarya Chandrashekar
Every year, we host a global hackathon in our Bengaluru office. Teams of 8-10 are pitted against each other in friendly competition, and it’s an exhilarating process to see how much we can build when we combine our different skill sets to focus as a unit on a single problem.
In a usual year, we’d have teams from across the world flying in to collaborate intensively for 24 hours to build innovative technology solutions.
But this year was far from normal and that meant this year’s hack was special for two reasons. First, due to COVID-19, we had to make the event virtual, with teams all working together across multiple time zones. (Fortunately, our MiQ day jobs mean we’re already pretty good at this!)
Second, and more important, was this year’s theme. In response to the anti-racism movement, we decided the theme would be ‘Using Technology to Fight Social Injustice’. We were absolutely thrilled (cheering-in-front-of-our-computers thrilled) at this opportunity to use our expertise in data and technology to try and tackle an issue which means so much to all of us.
Our problem statement
Our team, ‘The Quaranteam’ (because, of course), decided to get to the root of the injustice we are currently seeing: police brutality in the US.
And yet there is little to no police accountability. 99% of police killings from 2013-2016 have not resulted in any criminal charge3.
We wanted to design a technological solution that could strengthen law enforcement accountability. But what exists already?
Well, the leading police accountability solution is body cameras, which have been shown to reduce police use of force by 60% and citizen complaints by 88% in a year long study4.
But they’re not without their problems. Continuous video recording raises a host of privacy concerns for both citizens and officers alike, and the high costs of storing so much footage inhibits law enforcement agencies from adopting body cams in the first place. Agencies that do adopt body cams must implement strict guidelines for which encounters officers should record, yet there is no way to enforce those guidelines. Officers are ultimately able to choose when to record or not record – which is far from ideal as a solution for holding police officers to account.
To solve these issues, we built a proof of concept for ACCELEROCOP, a wearable device for police officers that uses accelerometers and Artificial Intelligence to detect violent actions with >99% accuracy.
ACCELEROCOP is able to identify actions such as walking, sitting, running, punching, and tackling in real-time and classify them as green (not violent), yellow (becoming violent), and red (violent) actions. Yellow actions will trigger body cam recording to start, and red actions will alert central authorities and medics, creating much more visibility as well as accountability when violence occurs.
There are many other benefits of supplementing body cams with ACCELEROCOP:
- Cameras aren’t built for dark areas and have a limited field of vision
- Most cameras don’t come with real-time streaming, which quickly drains the battery
- Video storage is expensive, but ACCELEROCOP makes sure only the most relevant video footage is stored.
- Privacy concerns are reduced, since cameras do not have to record as long when non-violent actions are being observed.
Below is our demo of what a central monitoring UI powered by ACCELEROCOP could look like. Officer 1’s actions are tracked in real-time, along with time and classification of each action.
How did we do it?
To build this proof of concept, our team first needed to create sample data. We each downloaded an accelerometer app on our phones and tracked our motions as we performed a number of activities ranging from non-violent to violent. Once the sample data was ready, we exported the data from our phones to AWS (Amazon Web Services’ cloud), where we cleaned the data, processed it, and used it to train our model using a machine learning technique to identify the actions performed.
The hackathon started at 6pm Bengaluru time on July 16. By 5am, we had successfully built a multiclass classification model that predicted actions with >99% accuracy. By 7am, the UI demo was done. As our Bengaluru team members got a few hours of sleep, team members in New York and London took over the presentation preparation. With our team distributed globally, we were all able to get more sleep than we would have in a typical hackathon!
By 4:30pm the next day, we were ready to deliver our three minute, first round presentation. Adrenaline was high as we advanced to the final round, and then went through the roof when the judges announced our team for 1st place. The best part of winning was that each team member gets $500 to donate to a charity of their choice.
What we learned
The virtual nature of this year’s hackathon wasn’t without its challenges. It took some creative coordination to find times that worked for all three time zones to sync over Zoom and start developing our plan. However, we quickly learned the benefits of having multiple time zones working together, such as around the clock coverage and increased productivity, as long as we communicated regularly with each other over Slack and Zoom. It was fun to feel so connected on a global level and admire others’ expertise or discover their hidden talents along the way.
We also learned a lot more about police accountability and the challenges that agencies and communities face when trying to implement solutions. We learned that there is a lot more work to be done when it comes to social injustice – but we strongly believe that technology can play a huge role in contributing to progress on that front.
Joincampaignzero.org was an incredibly valuable resource to us during our research, and we hope our work in this hackathon will inspire others and further contribute to the fight against social injustice.