Eos

Project Information

Contemporary investigations increasingly rely on video material collected from sources such as CCTV systems, body-worn cameras, and smartphones. While such material can be crucial for identifying suspects and reconstructing events, its sheer scale presents major challenges in terms of time, cognitive load, and sensemaking.

The Europol Innovation Lab develops digital tools to support law enforcement investigators across its member states. Through the Europol Tool Repository (ETR), it provides free and downloadable tools for evidence analysis tasks. Recently, the Europol Innovation Lab developed a video analysis tool that uses computer vision (CV) to enable keyword or image searches within video material, using configurable confidence thresholds. Europol collaborated with Umeå Institute of Design (UID) to explore a design project focused on improving the usability of this video analysis tool, with students using insights gathered from field research with the Swedish Police during the design process.

The brief for this project was to design a user interface that supports law enforcement investigators to efficiently, adaptively, and responsibly process large amounts of videos to find critical information.

Methods

The project followed a research-through-design approach, combining field research, embodied methods, and desk research to inform concept development and interface design. Over two days of fieldwork, interviews were conducted with ten law enforcement professionals across two locations in Sweden. Participants included intelligence officers, image analysts, operational analysts, and investigators working with youth crime, hate crime, and online abuse cases. Presentations and interviews provided insight into investigative workflows, constraints, and existing tool use.

Interviews were conducted in small groups to encourage in-depth discussion. During these sessions, insights were drawn and annotated live on large sheets of brown paper in front of participants. This method allowed investigators to validate interpretations, emphasise what they found most relevant, and make connections across anecdotes. The visual and collaborative nature of the process supported shared reflection and surfaced aspects of investigative work that might otherwise remain implicit.

To further understand investigative work, bodystorming and role-playing were used to recreate scenarios described in interviews. These were translated into “oneshot” videos depicting realistic investigative situations. This method helped surface the tacit expertise involved in investigative reasoning, such as judgment calls, iteration, and hypothesis testing. Desk research complemented these methods by examining the technical foundations of computer vision, retrieval-augmented generation, and large language models, as well as benchmarking existing video analysis tools to understand current interaction patterns and limitations.

Result

The outcome of the project is Eos, a conceptual AI-supported video analysis tool designed to support key stages of video evidence analysis: scanning datasets, sorting and organising material, sensemaking, and deep analysis. Research revealed that these stages are nonlinear and often overlap, requiring investigators to repeatedly diverge and converge as new insights emerge. Very comparable to a design process. The design of Eos reflects this iterative reality rather than imposing a rigid, linear workflow.

Eos is organised into three interconnected workspaces. The data workspace provides visibility into large video datasets, allowing investigators to filter, cluster, and gain an overview of available material. The canvas workspace supports sensemaking by enabling investigators to bring selected videos into a dedicated space where they can compare, annotate, and connect evidence – without watching a vast amount of video frames. This workspace was a primary design focus, as it aligns closely with investigators’ need to externalise thinking and work through complex relationships. The video playback workspace supports detailed review and close inspection of individual clips.

A defining feature of Eos is context-aware prompting, which acts as a connective thread across workspaces. Investigators can interact with the system using multimodal (text, audio, image, video) prompts grounded in their investigative goals, refining prompts over time as their understanding evolves. To promote trust in AI, we design interactions based on our explainability approach, which emphasizes visibility, traceability, and understandability. Every prompt is accompanied by a reasoning panel that explains how the system processed the request, what data was used, and the steps taken to produce the output.

Rather than aiming to replace investigative judgment, Eos is designed to amplify human expertise by providing transparent, flexible tools for exploration and sensemaking. The result is a design concept that foregrounds professional practice, ethical responsibility, and the role of interaction design in shaping how AI systems are experienced and trusted in high-stakes investigative contexts.