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

This session discusses a first-of-its-kind study investigating the efficacy of novel multimodal techniques (computer vision/NLP) for analyzing body-worn camera (BWC) recordings to measure procedural justice.

The research addresses two primary questions:
(1) Are there differences in judgements of procedural justice among community members, university faculty, and police supervisors?
(2) Do procedural justice scores generated by automated video analytics (TrustStat) align with human raters?

Methods
The study involved further development and testing of the TrustStat software, creating a coding instrument for human evaluators, and comparing automated evaluations with human raters’ assessments. The research included 66 raters from diverse backgrounds and analyzed 10 selected BWC videos provided by the Dallas Police Department.

Findings
TrustStat’s evaluations were consistent with human coders’ assessments, demonstrating the tool’s accuracy and reliability. Differences in ratings among community members, police supervisors, and faculty members were minimal.

Significance
This study is the first to validate the reliability and accuracy of an automated tool for reviewing Body-Worn Camera content. The TrustStat technology shows the promise of scaling BWC review and turning raw video into data-driven officer performance management. This shift moves evaluations beyond anecdotal accounts to data-driven insights, emphasizing the quality of interactions over output metrics like arrests or tickets issued.