JR Centre for Multimedia analysis in the mobility domain

This JR Centre deals with efficient tracking of multiple objects and subjects based on multiple sensors and with the analysis of the movement and route information obtained in use cases as diverse as lane changes on motorways or livestock behaviour in agriculture, which can be used for traffic safety and animal welfare, for example.

Multimedia sensor systems, which allow machines and algorithms to "see" our real world in the form of an approximated model that they can understand ("computer vision") and react to it, are a highly challenging research topic per se: if not all aspects of the environment recognised for the respective application are taken into account, the system makes decisions based on an incomplete model, which can have serious consequences in safety-critical situations. However, if the models generated are too complex, they will be inefficient and cost-intensive at best and, at worst, incompatible with the (limited) hardware environments of, for example, traffic control systems in which they are to be used due to their complexity.

With over a decade and a half of research experience in the field of multimedia information systems, Mario Döller is ideally equipped to head the Josef Ressel Centre for Multimedia Analysis in Mobility, which is dedicated to three main areas of research: Firstly, the development of a "data factory" for the generation of synthetic data that supports the training (and thus the continuous improvement) of computer vision systems, especially in the area of particularly important special cases. Secondly, novel AI-based algorithms and architectures for tracking objects across multiple sensor systems ("multi-object multi-sensor tracking") are to be researched. By networking existing multimedia sensor units in this way, complex interrelated motion sequences can be better captured and corresponding hazardous conditions can be detected (near misses, etc.). The third research objective is the optimisation of AI systems for efficient processing on limited hardware resources.

In all of these objectives, a strategic exchange and the integration of innovations for the purpose of practicability through standardisation is planned with the Motion Picture Experts Group (MPEG) and the Joint Photographic Experts Group (JPEG), for example. Mario Döller is both a long-standing member and editor of several standards in these committees and co-founder of the Austrian branch of the Austrian Standards Organisation.

The centre researches basic methods of trajectory detection and analysis of objects for transfer to different areas of application. These areas range from analysing traffic situations (e.g. fast lane changes or near misses) to counting objects and monitoring spaces. Parking space management, for example, is one use case: not only to combat crime (detection of car break-ins or suspicious persons staying in car parks for an excessive amount of time), but also for emergency medical assistance, for example when a person lying permanently on the ground is detected or senior citizens who have fallen are recognised using an additional method for estimating body postures. And finally, the multimedia sensor systems and algorithms developed should even be able to track and monitor herds of cattle and their individuals, for example to detect predatory animal attacks in good time and enable rapid intervention.