Robust Moving Objects Detection in Lidar Data Exploiting Visual Cues

Gheorghii Postica, Andrea Romanoni, Matteo Matteucci

Abstract

Detecting moving objects in dynamic scenes from sequences of lidar scans is an important task in object tracking, mapping, localization, and navigation. Many works focus on changes detection in previously observed scenes, while a very limited amount of literature addresses moving objects detection. The state-of-the-art method exploits Dempster-Shafer Theory to evaluate the occupancy of a lidar scan and to discriminate points belonging to the static scene from moving ones. In this paper we improve both speed and accuracy of this method by discretizing the occupancy representation, and by removing false positives through visual cues. Many false positives lying on the ground plane are also removed thanks to a novel ground plane removal algorithm. Efficiency is improved through an octree indexing strategy. Experimental evaluation against the KITTI public dataset shows the effectiveness of our approach, both qualitatively and quantitatively with respect to the state-of-the-art.

1. Introduction

Detecting moving objects (MOD) is a fundamental requirement for autonomous vehicles and mobile robots. However, it is notoriously difficult when the robot itself is moving. Egomotion makes the entire world appear to be in motion relative to the sensors. While many researchers focus on "change detection" (finding differences between a new scan and a pre-existing map), true MOD requires identifying moving entities in an unknown environment in real-time.

Lidar beams on a slope illustrating the challenge of ground removal on non-flat terrain.

Existing LiDAR-only methods often struggle with false positives from sensor noise, misalignments, and the ground plane. Standard algorithms often mistake the ground for a moving object, especially on non-flat surfaces or hills. Furthermore, processing millions of points in a 3D point cloud is computationally expensive, often exceeding real-time constraints.

2. Proposed Methodology

The authors propose a multi-stage pipeline to tackle these challenges by combining occupancy grid analysis with visual validation from cameras.

System architecture showing the integration of laser-based detection and image-based validation.

2.1 Smart Ground Removal

Instead of assuming a flat ground plane, the system models the ground as a Markov Random Field (MRF). Belief propagation is used to estimate ground height across a 2D grid, allowing the system to follow terrain variations and accurately filter out ground points.

2.2 Discretized Occupancy Grids and Octree Efficiency

The method employs Dempster-Shafer Theory (DST) to manage environmental uncertainty. By comparing current scans against a temporal window of previous and future scans, "conflicting" voxels are identified. To optimize performance, the point cloud is indexed using an octree. MOD tests are performed on a random subset of points within each octree leaf; if a threshold is met, the entire leaf is flagged, significantly reducing processing time.

2.3 Visual Validation

To improve robustness, LiDAR-flagged moving points are projected into corresponding 2D camera images. Image patches are compared across frames using Normalized Cross Correlation (NCC). If visual patches match, the LiDAR detection is discarded as a false positive, effectively using the camera as a sanity check.

3. Experimental Results

The approach was evaluated using the KITTI public dataset (sequences 0091, 0095, and 0104). The proposed method showed a marked improvement over the state-of-the-art (Vallet et al.):

  • Precision Boost: In one KITTI sequence, precision increased from 0.11 to 0.26 following visual validation. While recall dropped slightly, the "cleanliness" of the detections was much higher.
  • Efficiency: The algorithm processes a point cloud in approximately 0.6 seconds, compared to 4.9 seconds for the baseline method.
  • Detection Quality: The system detected a higher total number of moving objects across all tested sequences.

4. Conclusion

This paper demonstrates that treating ground removal as a probabilistic propagation problem and using cameras to validate LiDAR detections drastically improves MOD performance. While the visual validation was implemented as an off-line prototype, its potential for real-time GPU parallelization is clear. Future work involves integrating this robust detection into 3D urban reconstruction to automatically remove dynamic objects from static maps.

Summary updated

Abstract

Detecting moving objects in dynamic scenes from sequences of lidar scans is an important task in object tracking, mapping, localization, and navigation. Many works focus on changes detection in previously observed scenes, while a very limited amount of literature addresses moving objects detection. The state-of-the-art method exploits Dempster-Shafer Theory to evaluate the occupancy of a lidar scan and to discriminate points belonging to the static scene from moving ones. In this paper we improve both speed and accuracy of this method by discretizing the occupancy representation, and by removing false positives through visual cues. Many false positives lying on the ground plane are also removed thanks to a novel ground plane removal algorithm. Efficiency is improved through an octree indexing strategy. Experimental evaluation against the KITTI public dataset shows the effectiveness of our approach, both qualitatively and quantitatively with respect to the state- of-the-art.
BibTeX
@inproceedings{postica2016robust,
  author = {Gheorghii Postica and Andrea Romanoni and Matteo Matteucci},
  title = {{Robust Moving Objects Detection in Lidar Data Exploiting Visual Cues}},
  booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems},
  year = {2016},
  url = {https://openreview.net/forum?id=aJvuAlNJu6},
}
Figure