Znak Politechniki Warszawskiej

Artificial intelligence will take care of the safety of officers and inmates

Photo of surveillance cameras

fot. Unsplash / Paweł Czerwiński

Researchers from the Faculty of Electronics and Information Technology are working on a tool to detect undesirable behavior among people in pre-trial detention centers and penitentiaries. The solution is to be based on specially designed algorithms for automatic detection and prediction of events.

The effective monitoring of places where inmates are staying is one of the problems faced by penitentiary units. Currently, prison officers are monitoring the situation in specific locations through industrial IP cameras. Observing multiple locations at the same time is a challenge. A tool developed by researchers from the Warsaw University of Technology may be a relief for prison officers.

- Vision systems, which are currently implemented, are not automated. The officers themselves have to control a large number of cameras and observe what is happening at designated points - explains Piotr Bilski, PhD, DSc., project manager on behalf of WUT. - Our task is to design algorithms that predict and detect events such as fights, attacks on guards, escape attempts - the researcher explains.

Deep learning comes into play

Our researchers rely on the image captured by cameras, and in their research they use learning deep neural networks activated on graphics cards. Although systems for detecting specific people and behavior in films already exist, the task faced by researchers from the Faculty of Electronics and Information Technology is to modify them and adapt them to the specific conditions in penitentiary institutions.

- In order to minimize the number of false alarms, but also to avoid the risk of not capturing dangerous behavior, we must have a sufficiently large number of examples of individual events with a different course. An example may be inmates' fights - filmed from different positions, under changing lighting. A sufficient amount of training data must be in line with appropriately matched detection thresholds - Piotr Bilski, PhD, DSc.

Innovative solutions

Compared to other solutions of this type, the project of WUT specialists will stand out, among others, with its open architecture. Thanks to it, it will be possible to modify algorithms and analyze video, audio and thermal data. The system is also to be scalable, which means the ability to freely determine the number of simultaneously processed images.

As part of the project, a database of specific scenarios will also be prepared along with a description containing, among others, predictors of individual events.

 

The project "Intelligent decision support system based on algorithmic image analysis in the activities of justice services" is financed by the National Center for Research and Development. The leader is the Warsaw University of Technology, and the partners of the consortium are the University of Natural Sciences and Humanities in Siedlce and the JAS Technologie company. The Ministry of Justice is the administrator.