The tripleB-ID project

An increase in visual surveillance systems together with an increased inquiry for security and efficiency leads to the need of efficient systems which are able to process and interpret video data automatically. Based on developments in the area of machine vision, pattern recognition and visual sensors this project deals with the development of a system for automatic event recognition in videostreams, under special consideration of social and cultural aspects. By integrating social-scientific aspects from the beginning a close collaboration with the technical content is granted.

Motivation

CCTV cameras in bank foyers are a common view. Up to now most of the surveillance is accomplished by recording and monitoring systems based on analogous camera technology. In case of an incident, the security staff has to browse through the recorded material manually in order to get video material of the crime scene.
It is well known that robbers are investigating the bank they intend to rob in advance. Therefore it would make sense to sift through the recorded videos in order to look for persons behaving suspiciously. However, the huge amount of data renders such a search practically impossible. State-of-the-art digital video recorders (DVRs) support this cumbersome task by providing elementary functionality like basic motion detection and camera tamper detection (i.e. the system detects if a camera is vandalised). However, if the scene is explored by the criminals during the regular business hours, these features are of little help.
This paper presents the recently granted tripleB-ID project which aims to improve this situation by a smart system able to recognise abnormal behaviour and set the appropriate actions (e.g. informing the responsible members of staff but also automatically operating a pan-tilt-zoom (PTZ) camera in order to get higher resolution images of a suspicious activity). Furthermore, the event information will be archived, so that in case of an incident the screening of the video data can be supported..

Scope of this project

The goal of the project is the development of a system for automated event recognition or so called "abnormal behaviour" detection within security-critical infrastructure. Recognition of certain activities in a scene occurs through the analysis of the acquired image sequences. The planned system concentrates on object tracking with several cameras followed by an event detection, as well as optimised data compression with efficient archiving. Existing methods are adapted and enhanced. The methods investigated are evaluated under realworld conditions, therefore the consumers define certain scenarios in order to install a prototype system: an operational real time processing prototype at the facilities of the end users (Erste Bank sOM Objektmangement) will be implemented. tripleB ID will be implemented on off the shelf IP cameras and computing hardware which will ensure that the technologies and algorithm can be adopted more easily later on. The project will be a key enabler for all end-users to utilize their surveillance technology far beyond the existing possibilities of data storage and/or monitoring by human operator.

 

This project is supported by:

Bundesministerium_für_Verkehr,_Innovation_und_Technologie_(BMVIT)

Österreichische Forrschungsförderungsgesellschafft mbH (FFG)