A Clustering-Based Data Reduction for the Large Automotive Datasets

Patryk Siwek, Pawel Skruch, Marek Dlugosz

Abstract: Large datasets used in automotive consist of a set of recorded sequences that represent possible road scenarios. Such scenarios are mainly utilized as test scenarios to verify developed driver assistance systems. Another application of the dataset is the training and verification of machine learning-based algorithms. As the number of possible road scenarios is, in fact, infinite, the process of selecting representative and meaningful sequences is a difficult and challenging task. This article presents an approach based on various clustering techniques for data reduction for large datasets that are used in the automotive industry to evaluate environmental perception algorithms. The approach is supported by the results obtained on representative datasets.