Open-source tools for ethical analysis of body-worn camera footage

Accurately interpret Police BWC footage by employing open source ML, CV, & NLP technologies, resulting in identifiable behavioral patterns among police & civilians, analyzed in an interdisciplinary setting.

Improve policing through evidence-based recommendations sourced from BWC footage by leveraging AI in an ethical and unbiased manner.

99% of Body-Worn Camera footage is never reviewed. The vast majority of policing captures officers simply doing their jobs—a reality that stays hidden due to privacy laws and massive data volume.

This creates a critical gap: without systematic analysis, we miss opportunities to understand effective policing practices, identify training needs, and build evidence-based policy.

OpenBWC is a research initiative developing open-source artificial intelligence tools to analyze police body-worn camera (BWC) footage. We combine computer science, social science, and criminal justice expertise to create interpretable, bias-audited analysis systems that serve the public interest.

• Open Source - Transparent methods, auditable code, no vendor lock-in
• Ethics First - Bias minimization built into every layer
• Research-Grade - Confidence scoring and uncertainty quantification
• Interdisciplinary - Computer science + social science + criminal justice
• Academic Rigor - Peer-reviewed methods and findings

1. Behavioral Patterns - What communication and interaction patterns emerge in police-civilian encounters?
2. De-escalation - What verbal and non-verbal techniques correlate with successful de-escalation?
3. Use of Force - What precursors can be identified before force escalation?
4. Training & Policy - How can AI-assisted analysis inform evidence-based training and policy recommendations?

Our interdisciplinary approach integrates:

• Machine Learning & NLP - Multi-model transcription with confidence scoring
• Computer Vision - Visual analysis of BWC footage
• Speaker Diarization - Identifying who spoke when, even with overlapping speech
• Social Science Methods - Qualitative and quantitative analysis frameworks
• Continuous Bias Auditing - Algorithmic fairness monitoring throughout the pipeline

Our core technology is TrustScript, a multi-model pipeline for processing BWC footage with interpretable confidence scores.

• Multi-Model Fusion - Combines ASR systems for robust transcription
• Confidence Quantification - Every word tagged with multi-source confidence scores
• Overlap Handling - Speaker re-identification for concurrent speech
• Research-Grade Output - Detailed metadata for reproducibility and auditing
• Human-in-the-Loop - Prioritized review based on confidence scores

We take algorithmic fairness seriously. Our multi-level bias mitigation strategy includes:

• Continuous Auditing - Regular fairness testing across demographic groups
• Performance Monitoring - Track model accuracy and calibration by subgroup
• Transparency - All confidence sources documented and interpretable

• Interdisciplinary Review - CS + social science + ethics oversight
• Privacy Protection - Anonymization and secure data handling
• Stakeholder Engagement - Input from law enforcement, communities, and experts

• Stratified Sampling - Ensure diverse representation in analysis
• Counterfactual Reasoning - Test alternative explanations
• Open Documentation - Publish methods, limitations, and findings

We do not build predictive policing tools. Our work is descriptive and analytical, focused on understanding what happened, not predicting future behavior.

We collaborate with the Rochester Police Department (RPD) in Rochester, New York, through a research partnership that balances transparency with privacy protections.

Our partnership is governed by:

• Institutional Review Board (IRB) oversight
• Memorandum of Understanding with clear data use terms
• Privacy-first data handling protocols
• Community stakeholder engagement

• Technical Papers - TrustScript methodology and benchmarks
• Ethics Framework - Bias minimization in policing AI research
• Policy Recommendations - Evidence-based guidance for law enforcement
• Open Datasets - Annotated data for research community (privacy-protected)

• Training future researchers in ethical AI development
• Advancing interdisciplinary research methods
• Contributing to public understanding of AI in criminal justice

Current Phase: Active Research & Development

• TrustScript core pipeline developed
• Multi-model transcription with confidence scoring
• Speaker re-identification for overlapping speech
• Bias auditing framework established
• GroundTruth annotation interface prototype

(This section will be populated as we release code)

openbwc/
├── trustscript/          # Core processing pipeline
├── groundtruth/          # Annotation interface
├── bias-auditing/        # Fairness monitoring tools
├── notebooks/            # Research notebooks and examples
├── docs/                 # Documentation
└── papers/               # Research publications

We welcome contributions from researchers, developers, and domain experts. Areas where we need help:

• Audio Processing - Improving robustness to noisy BWC environments
• Bias Auditing - Developing fairness metrics for policing contexts
• Social Science - Qualitative analysis frameworks
• Documentation - Making our work accessible
• Use Cases - Identifying ethical applications

Coming Soon: Contribution guidelines and code of conduct

Open-source transparency is a core pillar of OpenBWC. We believe that tools used in the analysis of public infrastructure, particularly in criminal justice, must be auditable, reproducible, and transparent to maintain public trust.

Every tool, model, and framework produced by the OpenBWC initiative will be released under an Open Source Initiative (OSI) approved license. This ensures that our work remains a public good, free from vendor lock-in and accessible for independent audit.

Because OpenBWC is a multi-disciplinary toolkit—ranging from high-level research interfaces to low-level systems code—we may utilize different licenses to best suit each component's use case:

• Core Logic & Frameworks: Components like our pipeline will be licensed under the AGPL, prioritizing maximum permissive reuse for the research community.
• Infrastructure & Tooling: Supporting tools, such as the OpenBWC Project Guide or GroundTruth annotation prototypes, may be released under licenses that encourage community contribution while protecting the integrity of the core research.
• Research Outputs: While code is open-source, specific Open Datasets and Peer-Reviewed Methodologies will be released under appropriate open-access licenses (such as Creative Commons) that balance transparency with strict privacy and Institutional Review Board (IRB) protections.

Note: As we transition from active R&D toward a stable release in October 2026, specific license files will be added to each sub-repository within the openbwc/ organization to provide definitive legal clarity for each module.

• Website: openbwc.org
• Publications: openbwc.org/publications
• GitHub: github.com/OpenBWC

• Rochester Police Department - Partnership and data access
• RIT Faculty & Students - Research collaboration
• Department of Justice - Grant support
• Open Source Community - Tools and frameworks that make this work possible

• Research tool for understanding police-civilian interactions
• Open-source alternative to proprietary BWC analysis systems
• Framework for ethical AI in criminal justice contexts

• Predictive policing or risk assessment tool
• Surveillance technology
• Replacement for human judgment
• Production-ready commercial software (yet)

We are committed to ensuring our tools are used ethically. Our research is designed to:

• Improve police training and accountability
• Inform evidence-based policy
• Advance academic understanding
• Protect civil rights and privacy

Misuse Notice: This technology should not be used for mass surveillance, predictive policing, or any application that violates civil rights.

• Visit our website - Learn about our mission and approach
• Read our publications - Academic papers and research findings
• Technical Documentation - Coming soon
• Ethics Framework - Coming soon

If you're interested in supporting ethical AI research in criminal justice:

• Star this repository to follow our progress
• Share our work with researchers and policymakers
• Engage in discussions about AI ethics
• Collaborate - reach out if you have relevant expertise