Title: What Automated Vehicles need to learn? The dark side of sensor data

Abstract:

The situational awareness of Automated Vehicles is constructed from perception sensors’ data (e.g. camera, LiDAR, RADAR, etc.) and how data is used by machine learning and perception algorithms. However, in real-world driving conditions, produced data is far from ideal, and even automotive datasets might contain non ideal data. In this situation, it is of outmost importance to carefully consider and assess the quality of sensor data and of datasets in combination with the performance of the downstream algorithms consuming the data or using the data to train machine learning algorithms.

This talk will cover an introduction to the main challenges associated with the quality and variability of data produced by automotive perception sensors, their relationship with perception (e.g. detection, segmentation, de-noising), touching upon datasets quality, data augmentation, compression, use of ‘raw' data and more.

Bio: Prof Valentina Donzella received her BSc and MSc in Electronics Engineering from University of Pisa and Sant’Anna School of Advanced Studies (Pisa, Italy), and her PhD (2010) in Innovative Technologies for Information, Communication and Perception Engineering from Sant’Anna School of Advanced Studies. In 2009, she was a visiting graduate student at McMaster University (Hamilton, ON, Canada) in the Engineering Physics department.

She is currently Full Professor, head of the Sensors area in the Intelligent Vehicles group at WMG, University of Warwick, UK, and she has been awarded a Royal Academy of Engineering Industrial Fellowship on camera sensors (2020-22). She is currently leading the work package on perception sensor noise models as a part of the 4 year EU ROADVIEW project. Before joining WMG, she was a MITACS and SiEPIC postdoctoral fellow at the University of British Columbia (Vancouver, BC, Canada), in the Silicon Photonics group. She is first author, co-author, and last author of several journal papers on top tier sensors and intelligent transportation journals. Her research interests are: Machine Learning, automotive perception sensors, noise models, sensor fusion, data quality.

Dr Donzella is Full College member of EPSRC and Senior Fellow of Higher Education Academy.