ELLIS Delft Talk: New perception challenges driven by the intelligent vehicle use-case

01 juni 2021 16:00 | Zet in mijn agenda

Speaker: Julian Kooij


Abstract:
Machine Learning has been key to the success of many state-of-the-art methods for perception tasks in Intelligent Vehicles (IV). Well-known tasks include object detection, segmentation, and visual localization. As a result, several benchmarks that are popular in CV/ML are built from camera images and lidar point clouds from the IV domain, which can give the impression that 'solving' these benchmarks also addresses the challenges for vehicle perception. While these benchmarks indeed play an important role, I will present several examples from my research that illustrate how consideration of the target application can lead to new challenges and research opportunities that the current benchmarks do not properly capture. Examples will include the development of a data-driven method to detect unseen approaching around corners using vehicle-mounted microphones [Schulz,RAL21], and vehicle localization by cross-view matching between camera and satellite images [Xia,ECCVW'20] [Xia,RAL21].

 To join this event, please contact Frans Oliehoek.