Extraction and Analysis of Highway On-Ramp Merging Scenarios from Naturalistic Trajectory Data

TL;DR

This paper presents a methodology for extracting and analyzing highway on-ramp merging scenarios from naturalistic trajectory data using machine learning techniques, aiding scenario-based safety validation of connected and automated vehicles.

Contribution

It introduces a novel approach combining HMM, DTW, and decision trees with safety metrics for scenario extraction and categorization from large naturalistic datasets.

Findings

Effective extraction of on-ramp merging scenarios demonstrated

Method enables scenario categorization and safety assessment

Validated on data from the TFNDS

Abstract

Connected and Automated Vehicles (CAVs) are envisioned to transform the future industrial and private transportation sectors. However, due to the system's enormous complexity, functional verification and validation of safety aspects are essential before the technology merges into the public domain. Therefore, in recent years, a scenario-driven approach has gained acceptance, emphasizing the requirement of a solid data basis of scenarios. The large-scale research facility Test Bed Lower Saxony (TFNDS) enables the provision of ample information for a database of scenarios on highways. For that purpose, however, the scenarios of interest must be identified and extracted from the collected Naturalistic Trajectory Data (NTD). This work addresses this problem and proposes a methodology for onramp scenario extraction, enabling scenario categorization and assessment. An Hidden Markov Model (HMM) and Dynamic Time Warping (DTW) is utilized for extraction and a decision tree with the Surrogate Measure of Safety (SMoS) Post Enroachment Time (PET) for categorization and assessment. The efficacy of the approach is shown with a dataset of NTD collected on the TFNDS.