An Application-Driven Conceptualization of Corner Cases for Perception in Highly Automated Driving

September 2021

tl;dr: Multi-sensor expansion of previous work on corner cases.

Overall impression

An essential task of ML-based perception models are to reliably detect and interpret unusual new and potentially dangerous situations. A complication for the development of corner case detectors is the lack of consistent definitions, terms and corner case descriptions.

Corner cases introduced not only by vision system, but also radar/lidar and sensor fusion. This is an extension from CC-vision and CC-vision (IV version).

Key ideas

Technical details