Safer autonomous driving thanks to cooperating sensors

(31-05-2023) Martin Dimitrevski's PhD examines how cars can safely drive autonomously even in difficult traffic and weather conditions.

In autonomous driving, correctly observing the traffic environment is obviously crucial. Based on that observed data, the autonomous vehicle can make informed decisions to drive safely. Correctly perceiving the traffic environment becomes more difficult when an object partially obscures another object and when the weather is bad or when it’s dark. For example, raindrops on the camera lens allowing the AI to make wrong interpretations.

In current systems, autonomous cars use sensor fusion. This involves combining data from different sensors (camera, radar, distance sensor,...).

"In my research, I go a step further and the data from sensors are not only combined but the sensors actually start cooperating," Martin says.

In this cooperative sensor fusion, individual sensors of the system no longer take individual decisions on road users presence  (for example, a pedestrian detected or not) but they also start exchanging information, thus improving their decisions and therefore reaching an overall better concensus.

"This is especially effective in situations where one of the sensors is very resistant to certain environmental conditions (e.g., radar in fog) while some or all of the other sensors are not. Based on the information from the resilient sensor, the other sensors can adjust their parameters and better perceive the traffic situation," Martin explains.

The proposed cooperative fusion sensing system was thoroughly tested on public datasets of autonomous driving and validated on traffic in several cities in Flanders.

"This showed that the cooperative fusion perception system significantly outperforms traditional sensor fusion in difficult traffic and weather conditions. This will allow autonomous driving in difficult conditions to be a lot safer in the future," Martin concludes.

Read a more detailed summary or the entire PhD

Martin Dimitrevski Portretfoto

Having completed my PhD, I can reflect on the personal benefits and societal impact of my research. On a personal level, the pursuit of a PhD in this field has provided me with a profound sense of accomplishment. The opportunity to explore cutting-edge technologies at Ghent University, work alongside renowned experts in the field, and contribute to the development of a groundbreaking field of study has been immensely rewarding. Additionally, my PhD has opened doors to new and exciting career opportunities in both academia and industry.
Furthermore, the development of accurate and reliable perception systems for self-driving cars has the potential to revolutionize transportation, making it safer, more efficient, and accessible to everyone. By contributing to this field, I have had the privilege of playing a part in advancing the state-of-the-art in autonomous driving technology, bringing us one step closer to a future where transportation is both sustainable and reliable. Ultimately, my PhD research has not only provided me with personal growth and career opportunities but also contributed to the improvement of society as a whole.

Martin Dimitrievski was born on March 21, 1988, in the Socialist Republic of Macedonia, which was formerly a part of the Socialist Federal Republic of Yugoslavia and is now the Republic of North Macedonia. He lived there and attended school there until 2014, when he graduated with a master's degree in electrical engineering and information technologies in the area of video signal quality assessment.

Under the direction of professors Wilfried Philips and Peter Veelaert at Ghent University's department of Telecommunications and Information Processing, research group Image Processing and Interpretation, his continued interest in signal processing and artificial intelligence led him to pursue a PhD in computer science and engineering.

He participated in a multiple research projects over the course of his doctoral studies, focusing on sensor fusion topics for object detection and tracking in autonomous vehicles. As a first author, he published 12 papers in the proceedings of international conferences and 2 papers in international journals.

PhD title: Cooperative Sensor Fusion for Autonomous Driving

Contact: Martin Dimitrevski, Wilfried Philips, Peter Veelaert

 -

Editor: Jeroen Ongenae - Illustrator: Roger Van Hecke