AI-Driven Surveillance Auditing to Police Ethically

The growth of Body-Worn Cameras (BWCs) and CCTV systems in law enforcement has generated unprecedented volumes of surveillance data, presenting new opportunities for enhancing transparency, accountability, and public trust. However, manual review of this footage remains slow, inconsistent, and often significantly delayed, creating substantial barriers to timely oversight. This study addresses this gap by designing an AI-driven surveillance auditing system capable of detecting unethical practices in policing — including acts of aggression, racial profiling, and procedural violations — to strengthen transparency, accountability, and public trust. Multimodal AI techniques are employed, integrating computer vision and Natural Language Processing (NLP). Computer vision detects aggression, facial expressions, and nonconforming gestures from video feeds, while NLP models analyse spoken language for hostile, biased, or inappropriate tones. These modalities are fused in a scoring engine to flag questionable behaviour and generate structured compliance reports, while a secure dashboard allows supervisors and external oversight bodies to visualise audit results in real-time or batch mode. The research targets a scalable, precise, and ethically aligned AI system for real-world policing. All datasets will be anonymised, and the system will be audited for fairness across race, gender, and contextual variables. Expected outcomes include a working prototype, peer-reviewed contributions to AI ethics in law enforcement, and a deployment-ready implementation framework. By automating surveillance audits, this research aims to accelerate accountability processes, standardise behavioural assessments, and strengthen public trust in policing practices.