Artificial Intelligence in the Penal System

Ensuring security and order in Austrian prisons is a core task of the penal system. In particular, the protection of persons directly involved in the prison system, i.e. inmates and staff, against physical and psychological violence is of essential importance. In 2019, there was an average of one physical assault on staff in Austrian prisons almost every day and more than two reported criminal acts between inmates per day.

The aim of the proposed project is to investigate how new technologies from the field of artificial intelligence can be used to reduce the workload of prison staff and better protect both staff and inmates. To this end, an innovative multimodal approach is being taken, applying insights from the emerging research area of Privacy-Preserving Machine Learning (PPML). Already during the acquisition of sensor data, privacy preservation is ensured by increasingly relying on anonymizing sensor technologies. Specifically, 3d and thermal sensors, as well as “wearables” will be used; conventional RGB cameras will only be activated when needed. The designed system is open; a future extension by other modalities (for example audio or any IoT devices) is possible. The various data sources will be combined into a consistent fusion model to capture complex behavioral patterns.  In addition to critical – for example health-related – events that require immediate intervention, other events that appear inconspicuous when viewed in isolation are also to be recorded by the system. These include simple non-verbal interactions or physical contacts which will be stored in an event graph and allow the analysis of (among other things, aggression-related) behavior patterns over longer periods of time. Furthermore, this approach establishes a link to the concept of “Explainable Artificial Intelligence (XAI)”, which is becoming increasingly important, and enables the development of an overall system that presents the end user with not only useful but also comprehensible conclusions. In addition, the technology in question is subjected to a comprehensive impact assessment. This includes an empirical analysis of practical demands and risks as well as a legal examination and critical ethical discussion of the permissibility of such applications in terms of fundamental and human rights.

The main goals of this project are:

1. Development of a comprehensive system to support staff using 3D, thermal and wearable sensor as well as specialized intelligent models for human behavior analysis.
2. Guaranteeing the fundamental and human rights of the detainees.
3. Comprehensive and critical assessment of the developed technologies along central legal and ethical principles and the interdisciplinary analysis of potential consequences and risks, including the perspective of those affected and their representatives.

Project Partners

• Bundesministerium für Justiz
• CogVis Software und Consulting GmbH
• PKE Holding AG
• Research Institute AG & Co KG

This project is funded by the Austrian security research program KIRAS of the Austrian Research Promotion Agency  FFG (grant 879744 ).

 

 

Publications

T. Heitzinger and M. Kampel “A Fast Unified System for 3D Object Detection and Tracking”, accepted at the International Conference on Computer Vision (ICCV), October 2023, Paris, France

C. Stippel and T. Heitzinger and M. Kampel “A Trimodal Dataset: RGB, Thermal, and Depth for Human Segmentation and Action Recognition“, accepted at the German Conference on Pattern Recognition (GCPR), September 2023, Heidelberg, Germany

Mucha W., Kampel M. “Hands, Objects, Action! Egocentric 2D Hand-based Action Recognition”, Accepted in the 14th International Conference on Computer Vision Systems (ICVS), September 2023, Vienna, Austria.

Strohmayer, J., Kampel, M. (2023). WiFi CSI-Based Long-Range Through-Wall Human Activity Recognition with the ESP32. In Computer Vision Systems. ICVS 2023. Lecture Notes in Computer Science, vol 14253. Springer, Cham. https://doi.org/10.1007/978-3-031-44137-0_4