Dynamic Object Detection in Range data using Spatiotemporal Normals

Raphael Falque1, Cedric Le Gentil1, Fouad Sukkar1,
1University of Technology Sydney
Paper arXiv Code
Teaser.

Dynamic object detection in range data using spatiotemporal normals.

Abstract

On the journey to enable robots to interact with the real world where humans, animals, and unpredictable elements are acting as independent agents; it is crucial for robots to have the capability to detect dynamic objects. In this paper, we argue that the detection of dynamic objects can be solved by computing the spatiotemporal normals of a point cloud. In our experiments, we demonstrate that this simple method can be used robustly for LiDAR and depth cameras with performances similar to the state of the art while offering a significantly simpler method.

Experiments

Urban Dynamic Objects LiDAR (DOALS) Dataset

We tested our method on the Urban Dynamic Objects LiDAR (DOALS) Dataset and compared it to Dynablox.

Interpolation end reference image.

Ground truth

Interpolation end reference image.

Proposed method

Interpolate start reference image.

Dynablox


Dynamic scene and a moving camera (mounted on a robot arm)

We demonstrate how our method can be used to predict dynamic elements a scene with a moving camera.

Related Links

There's a lot of excellent work that was introduced around the same time as ours.

Dynablox solves the problem by leveraging a voxel map and analysing the free space from the voxels.

BibTeX

@inproceedings{falque2023dynamic,
      title={Dynamic Object Detection in Range data using Spatiotemporal Normals}, 
      author={Raphael Falque and Cedric Le Gentil and Fouad Sukkar},
      booktitle={Australasian Conference on Robotics and Automation, ACRA},
      year={2023}
  }