I work on multi-modal environment perception for intelligent vehicles, and mostly target crowded urban settings. My research applies deep learning and probabilistic models to automatically detect objects and predict traffic situations from multi-modal data, collected with on-vehicle sensors. Most of my research therefore relates to Computer Vision, but also addresses other sensing modalities, such as Lidar, Radar and Acoustics. To understand the environment, the autonomous vehicle needs to create useful representations of its environment, detect and identify dynamic and static obstacles, self-localize itself, and using all sensor information to anticipate future events. Applications include anticipating Vulnerable Road User (VRU) behaviour, path prediction, self-localizing the vehicle using satellite images, and fusing 3D depth and 2D image data for improved object detection.
Other Affiliations
- ELLIS Member, Delft unit
- I am co-director of the 3D Urban Understanding (3DUU) lab, which I started in 2020 with Dr. Liangliang Nan as part of the TU Delft AI Initiative.
Past projects
- “Epistemic AI (E-pi)”, EU Horizon 2020 FET Open, 2021-2026
- “Efficient Deep Learning (EDL)”, NWO Perspectief, 2018-2024
- “SafeVRU”, STW HTSM 2015, 2016-2020
- “i-CAVE”, STW Perspectief
- “PROSPECT”, EU FP8, 2015-2018
- “Technology in Motion (TIM)”, 2014 - 2016
- “ADABTS”, EU FP7, 2010-2014
- “CASSANDRA: Context-Aware SenSing for AggressioN Detection and Risk Assessment”, NWO, 2005-2010
Workshops
- Trust AD at IEEE Intelligent Vehicles Symposium.
- Unsupervised Learning for Automated Driving (ULAD) workshop 2022 at the IEEE Intelligent Vehicles Symposium, ULAD 2019, ULAD 2020.
Past affiliations
-
2015-2016: a PostDoc at the computer vision lab of the EWI Faculty of TU Delft. In this period, I worked on setting up the new Technology In Motion (TIM) lab at LUMC hospital in Leiden, and developed new signal processing techniques to detect subtle tremors in patients.
-
2010-2014: PhD at the University of Amsterdam. Topic: automated analysis of pedestrian tracks, using probabilistic graphical models for unsupervised learning and online predictive Bayesian inference. For the NWO Cassandra and EU-FP7 ADABTS projects, I developed graphical models for the integration of audio-visual cues, and for anomaly detection, in surveillance applications.
-
2013: Research internship at the Environment Perception group of Daimler AG in Ulm, Germany (Mercedes Benz). Worked on improved pedestrian path prediction for intelligent vehicles, exploiting various contextual cues such as pedestrian head orientation and location relative to the road.
Contact address
dr. J. F. P. Kooij
Faculty of Mechanical Engineering
Room 34 E-0-260, Mekelweg 2
2627 CD Delft, The Netherlands