Mummolo Receives Microsoft Funding for Police Body-Worn Camera Project

Aug 24 2021
By Riis L. Williams
Source Princeton University

Microsoft has awarded $250,000 in funding to professor Jonathan Mummolo and a team of fellow social scientists to fund the development of a novel system to computationally analyze police body-worn camera (BWC) footage.

Mummolo, assistant professor of politics and public affairs at the Princeton School of Public and International Affairs, is creating a system that will automate the process of summarizing body-worn camera content. He is working with Olga Russakovsky and Brandon Stewart of Princeton University, as well as Dean Knox and Rachel Mariman of the University of Pennsylvania.

Body-worn cameras are ubiquitous across American police agencies, but the thousands of hours of audio and video they record are expensive to store and time-intensive to review. These challenges may explain why some research has shown that BWCs have minimal effects on police behavior.

The platform that Mummolo’s team is designing will use computational techniques for audiovisual analysis to convert body-worn camera footage into written timelines, akin to movie scripts, providing detailed, moment-to-moment accounts of police-civilian encounters. The new system will provide an entirely new perspective on police-civilian interactions beyond traditional officer-generated reports, which the researchers hope will enable more effective supervision and accountability for misconduct.

“Policing is an inherently complex, multi-stage process. Currently, officer reports provide only a limited, possibly inaccurate snapshot of that process,” Mummolo said. “Harnessing the information contained in body-worn camera footage would provide a valuable record of police-civilian interactions, allowing for new approaches to studying classic questions such as de-escalation and racial bias.”

Mummolo and his collaborators already have a group of annotators compiling and working through body-worn camera footage gathered from multiple police jurisdictions. The team’s system will use these annotations to train a statistical model to identify visual and auditory features from bodycam recordings corresponding to objects, circumstances, and behaviors, which will then be assembled into navigable transcripts of the filmed encounter. These timelines could be stored cheaply in lieu of the footage itself, which is routinely destroyed by many police agencies who cannot pay for storage costs.

Mummolo’s previous work studying police tactics and behaviors has highlighted troubling disparities between officers of color and their white counterparts, and it has discredited inaccurate studies that understated levels of racial bias in police behavior due to statistical errors. These errors involved ignoring the influence of race in portions of police-civilian interactions that are not documented in traditional administrative records, a gap the body-worn camera project can help to fill.

The project is being conducted by Mummolo’s multi-disciplinary research group, Research on Policing Reform and Accountability, co-founded with Knox, and the team hopes to deploy and evaluate the system in police agencies by 2023.