Multi-Modal Analysis of Body-Worn Camera Recordings

Overview

Body-worn cameras (BWCs) generate petabytes of video and audio data in public safety agencies, yet manual review is prohibitively labor-intensive. As Principal Investigator on this National Institute of Justice (NIJ) funded initiative (Award No. 2020-R2-CX-0010), our research team evaluated state-of-the-art multi-modal machine learning pipelines to automatically detect and measure procedural justice during police-citizen encounters.

Research & Architecture

• Multi-Modal Modeling: Evaluated fused deep learning pipelines combining automatic speech recognition (ASR), natural language processing (NLP), and computer vision features from encounter footage.
• Procedural Justice Metrics: Developed quantitative indicators measuring key procedural justice dimensions, including neutrality, respect, citizen voice, and trustworthiness.
• Scalable Evaluation: Tested model reliability, potential algorithmic bias, and acoustic noise sensitivity across diverse real-world field recordings.

Findings & Publications

Published the final project report with the National Policing Institute: Multi-modal analysis of body-worn camera recordings: Evaluating novel methods for measuring police implementation of procedural justice (2024).