Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Fast maneuver recovery from aerial observation: trajectory clustering and outliers rejection

Nelson de Moura, Augustin Gervreau-Mercier, Fernando Garrido, Fawzi Nashashibi

TL;DR

This work tackles fast, map-free trajectory clustering for autonomous vehicle simulations by using a Dynamic Time Warping (DTW) distance to compare trajectories of different lengths. It introduces a two-stage pipeline: a DTW-based clustering (hierarchical, k-medoids, and dissimilarity-matrix variants) followed by post-processing that reorganizes clusters using initial and final points and removes outliers via mean-shift refinements (A2MS). The A2MS method, together with a Spread-on-cluster quality metric, consistently yields tight, outlier-robust clusters across pedestrians, cyclists, and vehicles on the inD and roundD UAV datasets, outperforming PAM and dissimilarity-based baselines in many cases. The approach enables scalable generation of maneuver catalogs for learning, prediction, and planning tasks without relying on maps, with potential integration into longitudinal driver behavior extraction and zone-level traffic analysis.

Abstract

The implementation of road user models that realistically reproduce a credible behavior in a multi-agentsimulation is still an open problem. A data-driven approach consists on to deduce behaviors that may exist in real situation to obtain different types of trajectories from a large set of observations. The data, and its classification, could then be used to train models capable to extrapolate such behavior. Cars and two different types of Vulnerable Road Users (VRU) will be considered by the trajectory clustering methods proposed: pedestrians and cyclists. The results reported here evaluate methods to extract well-defined trajectory classes from raw data without the use of map information while also separating ''eccentric'' or incomplete trajectories from the ones that are complete and representative in any scenario. Two environments will serve as test for the methods develop, three different intersections and one roundabout. The resulting clusters of trajectories can then be used for prediction or learning tasks or discarded if it is composed by outliers.

Fast maneuver recovery from aerial observation: trajectory clustering and outliers rejection

TL;DR

This work tackles fast, map-free trajectory clustering for autonomous vehicle simulations by using a Dynamic Time Warping (DTW) distance to compare trajectories of different lengths. It introduces a two-stage pipeline: a DTW-based clustering (hierarchical, k-medoids, and dissimilarity-matrix variants) followed by post-processing that reorganizes clusters using initial and final points and removes outliers via mean-shift refinements (A2MS). The A2MS method, together with a Spread-on-cluster quality metric, consistently yields tight, outlier-robust clusters across pedestrians, cyclists, and vehicles on the inD and roundD UAV datasets, outperforming PAM and dissimilarity-based baselines in many cases. The approach enables scalable generation of maneuver catalogs for learning, prediction, and planning tasks without relying on maps, with potential integration into longitudinal driver behavior extraction and zone-level traffic analysis.

Abstract

The implementation of road user models that realistically reproduce a credible behavior in a multi-agentsimulation is still an open problem. A data-driven approach consists on to deduce behaviors that may exist in real situation to obtain different types of trajectories from a large set of observations. The data, and its classification, could then be used to train models capable to extrapolate such behavior. Cars and two different types of Vulnerable Road Users (VRU) will be considered by the trajectory clustering methods proposed: pedestrians and cyclists. The results reported here evaluate methods to extract well-defined trajectory classes from raw data without the use of map information while also separating ''eccentric'' or incomplete trajectories from the ones that are complete and representative in any scenario. Two environments will serve as test for the methods develop, three different intersections and one roundabout. The resulting clusters of trajectories can then be used for prediction or learning tasks or discarded if it is composed by outliers.
Paper Structure (19 sections, 12 equations, 9 figures, 11 tables, 2 algorithms)

This paper contains 19 sections, 12 equations, 9 figures, 11 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Separating trajectories of interest
  3. Dynamic time warp (DTW)
  4. Clustering methods
  5. Hierarchical clustering
  6. Partition around medoids (or k-medoids)
  7. Dissimilarity matrix clustering
  8. Methods evaluated
  9. Reorganization using initial and final points
  10. Evaluating cluster partition quality
  11. Davies-Bouldin index (DB)
  12. Silhouette score (Slh.)
  13. Spread on cluster (Spr.)
  14. Results
  15. Methodology
  16. ...and 4 more sections

Figures (9)

  • Figure 1: Trajectory cluster for different road users
  • Figure 2: One eccentric trajectory and a target cluster (background image from bock2020)
  • Figure 3: Differences between methods for scenario 0
  • Figure 4: Differences between methods for scenario 2
  • Figure 5: Clustering differences for scenario 2
  • ...and 4 more figures