In the field of forensic analysis, the reliable (re-)identification of individuals involved in criminal activities based on video data represents a central challenge. Particularly under conditions of limited image quality or occluded identity cues, movement-based characteristics—such as individual gait patterns—are becoming increasingly important. At the same time, for use in legal contexts, not only the performance but also the interpretability and transparency of the employed AI models are crucial.

Objectives and Approach

The subproject within REGAiT focuses on the development of interactive visualization approaches for the explainability and validation of AI-based identification models. At its core is the analysis of person-specific movement characteristics extracted from video data, in particular through gait analysis. The models are designed such that their decision-making processes can be made transparent using methods from Explainable Artificial Intelligence (XAI).

To this end, visual analytics techniques are being developed that enable the interactive exploration of model decisions and provide comprehensible representations of the underlying features. In addition to visualizing the basis of decisions, particular emphasis is placed on communicating uncertainties in order to avoid misinterpretations and to increase the trustworthiness of the systems.

Innovations and Perspectives

The combination of AI-based movement analysis and interactive visual explainability provides a foundation for improved admissibility of such systems in legal contexts. Transparent model architectures and comprehensible decision-making processes not only facilitate technical validation but also support communication between technical experts and legal practitioners.

In the long term, this approach opens up new possibilities for the responsible deployment of AI in safety-critical domains by linking performance with explainable evidence, thereby strengthening the acceptance of such technologies in legal environments.

Project Partners

Fundings

Federal Ministry of Research, Technology and Space of Germany (BMFTR, Germany) in the project REGAiT (project number 13N17424 to 13N174247) under the program “Forschung für die zivile Sicherheit – gemeinsam für ein sicheres Leben in einer resilienten Gesellschaft” and its announcement “Anwendungen in der zivilen Sicherheit”.