Associating sensor data and reference truth labels: A step towards SOTIF validation of perception sensors
Kryda M, Qiu M, Berk M, Buschardt B, Antesberger T, German R, Straub D (2021)
Publication Language: English
Publication Type: Conference contribution, Conference Contribution
Publication year: 2021
Abstract
A main requirement for safe automated vehicles is a reliable environment perception in terms of the intended functionality. Perception sensors are intended to detect all surrounding objects. However, uncertainties in the measurement principles and the underlying machine learning algorithms makes it difficult to describe the intended functionality of perception sensors mathematically which is necessary prior to the reliability analysis of those. We approach this problem by, first, applying a multi-object-tracking algorithm and, second, using Density-Based Spatial Clustering of Application with Noise (DBSCAN) to associate tracked objects from multiple sensors. Finally, we transform the clustered object data to a binary format which allows a quantification of the sensor reliabilities.
Authors with CRIS profile
Minhao Qiu
Computer Science 7 (Computer Networks and Communication Systems)
Reinhard German
Computer Science 7 (Computer Networks and Communication Systems)
Related research project(s)
Involved external institutions
Germany (DE)How to cite
APA:
Kryda, M., Qiu, M., Berk, M., Buschardt, B., Antesberger, T., German, R., & Straub, D. (2021). Associating sensor data and reference truth labels: A step towards SOTIF validation of perception sensors. In Proceedings of the Sixth IEEE International Workshop on Automotive Reliability, Test and Safety (ARTS).
MLA:
Kryda, Marco, et al. "Associating sensor data and reference truth labels: A step towards SOTIF validation of perception sensors." Proceedings of the Sixth IEEE International Workshop on Automotive Reliability, Test and Safety (ARTS) 2021.
BibTeX: Download