Most law enforcement agencies in the Unites States, and nearly all medium and large agencies, have tested or deployed body-worn cameras (BWCs). This has resulted in millions of hours of BWC footage stored on agency computers and cloud-based servers across the country. However, relative to its cost, BWC footage is vastly underutilized. While the overall goal of BWCs is to promote accountability, transparency and legitimacy, law enforcement executives and researchers have primarily focused on the degree to which these devices affect and inform investigations outcomes like uses of force, assaults against officers, and complaints.
To a lesser extent, law enforcement agencies also use BWC footage for more proactive purposes. Supervisors can manually pull officers' BWC footage, assess the quality of their interactions with community members and provide feedback or intervene as necessary. However, this is typically only done on an ad hoc basis or in response to a specific incident ; few agencies have the means or capacity to review footage in a more comprehensive manner. In response, several companies are developing and testing new analytic tools that would allow agencies to systematically review BWC footage, hold officers accountable for their behaviors and improve performance. This post offers an overview of these analytic tools and their promise for maximizing the utility of BWC footage.
The science behind such analytic tools is rooted in two complementary fields of artificial intelligence (AI). Natural language processing (NLP) is a branch of computer science that takes language in its natural state of spoken and written words and processes it into usable data through language translation, speech-to-text conversion, and automatic summation of large volumes of text. Recent advances in NLP can further determine some of the underlying concepts, thoughts, emotions, and social contexts associated with speech (e.g., analyzing online reviews to determine how the public feels about a certain product or company). Researchers have already applied NLP methods to BWC footage to measure respectfulness and identify key features in community interactions, such as whether officers introduced themselves, provided explanations for their stops or asked for permission to conduct a search. Some companies have also begun developing and marketing NLP-based analytic tools for law enforcement agencies that can identify measures of professionalism during community interactions from BWCs.
A related field of study, computer vision (CV), relies on AI to extract meaningful data from visual inputs, rather than written and speech-based communication. CV detects and classifies objects from images or videos, identifies distances between objects and determines movement patterns. CV is increasingly used in several important sectors of society, including transportation , healthcare, manufacturing and agriculture. As applied to BWCs, CV can identify relevant visual markers during police-community interactions, such as physical distance, facial expressions, body posture and other movements. When combined with audio extraction from NLP, these markers can enhance the identification of critical performance metrics related to aggression, confrontation, de-escalation, respectfulness and professionalism. Polis Solutions is one company that is currently developing a multi-modal analytic tool that uses both NLP and CV to processes BWC footage of police-community interactions and generate measures related to rapport, respect, trust, cooperation, conflict and coordination.
Data generated through NLP and CV-based analytic tools have the potential to provide agencies with up-to-date information on officer performance, which could then be distributed to the agency's sergeants to inform their supervisory duties. Sergeants typically serve as the front-line supervisors within an agency and play a crucial role in holding officers accountable for their behaviors, evaluating their performance and implementing various corrective interventions. Sergeants, therefore, would greatly benefit from the systematic analysis and classification of their officers' BWC footage. An example of how this process might work is detailed in the figure below.
As noted in the figure, an analytic tool would automatically extract and process BWC footage to identify critical metrics of officers' performance, such as effective communication, displays of procedural justice or the application of appropriate or inappropriate de-escalation techniques during community interactions. This information would be provided to sergeants through a dashboard or other easy-to-use interface that would summarize their officers' performance. Based on an officer's overall performance, and on flags generated through the dashboard, sergeants could then pull up specific incidents captured on that officer's BWC. Whenever sergeants identify problematic behavior, they could take corrective actions as needed. Likewise, when an officer displays optimal or above-average performance, sergeants could offer commendations.
Despite their near-ubiquitous presence in law enforcement agencies, we know very little about how to leverage BWCs in a way that supports transparency and accountability. Recent advances in AI-based analytic tools have the potential to help agencies maximize BWCs by extracting critical, actionable metrics from camera footage on police-community interactions. Such analytic tools would position supervisors to respond in near-real-time to problematic behaviors before they lead to more serious incidents and identify positive interactions to encourage and reinforce good behavior. If implemented effectively, these tools could increase public perceptions of legitimacy, build trust and help mend fractured police-community relations.