Multi-Modal Analysis of Body Worn Camera Recordings: Evaluating Novel Methods for Measuring Police Implementation of Procedural Justice

The wider potential of body-worn cameras (BWCs) to improve practices and outcomes in policing has gone largely unrealized.

Project Overview

Although the use of body-worn cameras (BWCs) has expanded rapidly, the capacity to efficiently analyze the enormous amount of data collected by BWCs lags far behind. As a result, the wider potential of BWCs to improve practices and outcomes of policing has gone largely unrealized. The purpose of this project is two-fold: (1) develop novel techniques to automate analysis of BWC recordings of police-community interactions and evaluate officers’ adherence to principles of procedural justice and; (2) use a randomized controlled trial to assess the accuracy of those techniques by systematically comparing them to evaluations of BWCs recordings done manually by human raters under conditions of high and low procedural justice.

This project has significant potential to transform the way police agencies use body-worn camera (BWC) data as a training tool that will help to promote more constructive police-citizen interactions. The tools that we propose to develop and test will make it possible for agencies to evaluate large amounts of BWC footage in far more accurate, efficient, and cost-effective ways than the status quo of random sampling and unstructured supervisory review. The tools can also be used by agencies to develop department-wide measures of procedural justice to support the rating and evaluation of officers’ encounters with citizens and to track their performance over time across the entire agency, as well as within particular working groups. Once fully implemented, agencies that use COMPSTAT could use the tools we will develop to systematically track trends in procedural justice in the same way that they now track crime statistics.

Methodology

In phase 1 of the project, we will focus on the development of a multi-modal system for analysis of BWC footage. The system will integrate techniques from Natural Language Processing (NLP) and Computer Vision (CV) with well-established models of social interaction to automatically extract and measure key linguistic and behavioral features of officer-community encounters. In phase 2, once the automated measurement process is sufficiently refined, we will test its accuracy by comparing its evaluation of BWC footage to evaluations of the same police-civilian interactions done by human raters. We will obtain footage from officers receiving a specialized procedural justice training and officers not in the training condition to fully evaluate the system’s performance. The project team is led by Dr. Travis Taniguchi, Director of Research, (National Policing Institute) and Jonathan Wender, Ph.D. (Polis Solutions), in collaboration with GE Research and the Caruth Police Institute at the University of North Texas at Dallas.