Using video analysis to detect unusual and risky behavior for public safety
Erkut Akdag defended his PhD thesis at the Department of Electrical Engineering on June 26th.
One of the key areas addressed in this research is road safety. Detecting distractions or unsafe behavior from drivers on the road can help prevent accidents and save lives. The other key area in this research is urban security. Identifying unusual events, such as traffic violations, disturbances, or dangerous actions, in cities can support safer and more efficient urban environments. To achieve this, developed new methods that enable AI models to learn from video footage. The research explores how to detect subtle patterns in driver behavior using multiple camera views and motion analysis. It also uses information about human body posture to improve the recognition of actions.
Understanding complex settings
In this research, a new dataset is introduced for city environments which is focused on detecting throwing actions, an often overlooked but potentially dangerous behavior in traffic. It also proposes techniques for better understanding when and how anomalies happen over time, allowing the system to identify a wide variety of unusual situations more effectively. Next to that, different types of data are combined, such as video motion, depth information, and even text-based understanding, to create a more complete picture of what鈥檚 happening in a scene. These multi-layered systems improve the ability to detect and respond to unexpected events in complex real-world settings.
Application in the real world
Finally, the technology has been successfully deployed in real urban conditions through field experiments in the Helmond field lab. Roadside cameras and a digital system are used there to monitor traffic in real time, map anomalies as they occur, and support more informed, rapid responses by city authorities. Overall, this research offers practical and scalable solutions for detecting dangerous or abnormal behavior in both vehicles and public spaces. By combining intelligent video analysis with behavioral insight, it contributes to building safer, smarter, and more responsive communities.
This research was carried out as part of the NWO TTW EDL and ITEA SMART projects. It has received recognition through the ITEA SMART project, which was awarded the ITEA Award of Excellence 2024 at the ITEA PO Days in Antwerp. This award highlighted the project鈥檚 impact on improving urban traffic through 4D spatial technology and real-time vehicle data analytics.
Title of PhD thesis: . Supervisors: Prof. Peter de With, Prof. Henk Corporaal and Associate Prof. Egor Bondarau.