Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DynORecon: Dynamic Object Reconstruction for Navigation

Yiduo Wang, Jesse Morris, Lan Wu, Teresa Vidal-Calleja, Viorela Ila

Abstract

This paper presents DynORecon, a Dynamic Object Reconstruction system that leverages the information provided by Dynamic SLAM to simultaneously generate a volumetric map of observed moving entities while estimating free space to support navigation. By capitalising on the motion estimations provided by Dynamic SLAM, DynORecon continuously refines the representation of dynamic objects to eliminate residual artefacts from past observations and incrementally reconstructs each object, seamlessly integrating new observations to capture previously unseen structures. Our system is highly efficient (~20 FPS) and produces accurate (~10 cm) reconstructions of dynamic objects using simulated and real-world outdoor datasets.

DynORecon: Dynamic Object Reconstruction for Navigation

Abstract

This paper presents DynORecon, a Dynamic Object Reconstruction system that leverages the information provided by Dynamic SLAM to simultaneously generate a volumetric map of observed moving entities while estimating free space to support navigation. By capitalising on the motion estimations provided by Dynamic SLAM, DynORecon continuously refines the representation of dynamic objects to eliminate residual artefacts from past observations and incrementally reconstructs each object, seamlessly integrating new observations to capture previously unseen structures. Our system is highly efficient (~20 FPS) and produces accurate (~10 cm) reconstructions of dynamic objects using simulated and real-world outdoor datasets.
Paper Structure (13 sections, 5 equations, 6 figures, 2 tables, 1 algorithm)

This paper contains 13 sections, 5 equations, 6 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Method
  4. Notations and Background
  5. Scan Integration
  6. Map Structure
  7. Object Local Reconstruction
  8. ESDF Query
  9. Experiments
  10. Incremental Reconstruction of Dynamic Objects
  11. Computation Efficiency
  12. Free Space
  13. Conclusion and Future Work

Figures (6)

  • Figure 1: DynORecon constructing an approaching vehicle incrementally based on a Dynamic SLAM framework. (a): We visualise the volumetric reconstruction of a dynamic object in addition to the object trajectory, camera pose and camera trajectory to better represent the volumes of reconstructed obstacles. We also visualise the dense reconstruction of the static environment. (b): Initial observation of the object. (c--d): Object reconstruction being completed as more observations are integrated into it.
  • Figure 2: A 2D example of ESDF update via VDB-GPDF for (a) static scene and (b) dynamic object. Green points represent surface measurements and inform the surface voxels. Within the sensor frustum, we sample testing points (orange) to update known free voxels. Within a truncation distance from the surface along surface normals, voxels have their ESDF updated as occupied voxels.
  • Figure 3: Incremental reconstruction results in the Simulation experiment. (a) - ground truth mesh; (b-f) - the reconstruction of the vehicle is accumulated consistently with more observations.
  • Figure 4: Incremental reconstruction results on 3 cars from the Outdoor Cluster dataset Huang2019iccv and a cube in OMD Judd19ral. The incremental reconstructions are shown at three different phases. Outdoor Cluster reconstructions are aligned with the ground truth point cloud (shown in black) for visual comparison.
  • Figure 5: Per-frame reconstruction percentage coverage on the Simulation (top) and Outdoor Cluster (bottom) datasets. The frame index shows the number of frames from the first observation until the object goes out of view.
  • ...and 1 more figures