Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

COIN-LIO: Complementary Intensity-Augmented LiDAR Inertial Odometry

Patrick Pfreundschuh, Helen Oleynikova, Cesar Cadena, Roland Siegwart, Olov Andersson

TL;DR

COIN-LIO tackles robustness gaps in LiDAR-inertial odometry under geometrically degenerate conditions by tightly fusing intensity-based photometric residuals with geometry-based registration in an iterative EKF framework. It contributes a brightness-normalized intensity image filter, a geometry-aware, patch-based feature selection strategy, and a multi-patch management scheme to extract complementary information along non-degenerate directions. The approach demonstrates strong robustness on the newly introduced ENWIDE dataset and competitive accuracy on geometry-rich sequences such as Newer College, while maintaining real-time performance. The ENWIDE dataset and open-source implementation are intended to spur further research in robust LIO under challenging geometry.

Abstract

We present COIN-LIO, a LiDAR Inertial Odometry pipeline that tightly couples information from LiDAR intensity with geometry-based point cloud registration. The focus of our work is to improve the robustness of LiDAR-inertial odometry in geometrically degenerate scenarios, like tunnels or flat fields. We project LiDAR intensity returns into an intensity image, and propose an image processing pipeline that produces filtered images with improved brightness consistency within the image as well as across different scenes. To effectively leverage intensity as an additional modality, we present a novel feature selection scheme that detects uninformative directions in the point cloud registration and explicitly selects patches with complementary image information. Photometric error minimization in the image patches is then fused with inertial measurements and point-to-plane registration in an iterated Extended Kalman Filter. The proposed approach improves accuracy and robustness on a public dataset. We additionally publish a new dataset, that captures five real-world environments in challenging, geometrically degenerate scenes. By using the additional photometric information, our approach shows drastically improved robustness against geometric degeneracy in environments where all compared baseline approaches fail.

COIN-LIO: Complementary Intensity-Augmented LiDAR Inertial Odometry

TL;DR

COIN-LIO tackles robustness gaps in LiDAR-inertial odometry under geometrically degenerate conditions by tightly fusing intensity-based photometric residuals with geometry-based registration in an iterative EKF framework. It contributes a brightness-normalized intensity image filter, a geometry-aware, patch-based feature selection strategy, and a multi-patch management scheme to extract complementary information along non-degenerate directions. The approach demonstrates strong robustness on the newly introduced ENWIDE dataset and competitive accuracy on geometry-rich sequences such as Newer College, while maintaining real-time performance. The ENWIDE dataset and open-source implementation are intended to spur further research in robust LIO under challenging geometry.

Abstract

We present COIN-LIO, a LiDAR Inertial Odometry pipeline that tightly couples information from LiDAR intensity with geometry-based point cloud registration. The focus of our work is to improve the robustness of LiDAR-inertial odometry in geometrically degenerate scenarios, like tunnels or flat fields. We project LiDAR intensity returns into an intensity image, and propose an image processing pipeline that produces filtered images with improved brightness consistency within the image as well as across different scenes. To effectively leverage intensity as an additional modality, we present a novel feature selection scheme that detects uninformative directions in the point cloud registration and explicitly selects patches with complementary image information. Photometric error minimization in the image patches is then fused with inertial measurements and point-to-plane registration in an iterated Extended Kalman Filter. The proposed approach improves accuracy and robustness on a public dataset. We additionally publish a new dataset, that captures five real-world environments in challenging, geometrically degenerate scenes. By using the additional photometric information, our approach shows drastically improved robustness against geometric degeneracy in environments where all compared baseline approaches fail.
Paper Structure (18 sections, 11 equations, 6 figures, 3 tables)

This paper contains 18 sections, 11 equations, 6 figures, 3 tables.

Table of Contents

  1. INTRODUCTION
  2. Related Work
  3. LiDAR (Intertial) Odometry
  4. Intensity Assisted Odometry
  5. Method
  6. Definitions
  7. IMU Prediction and point cloud undistortion
  8. Image Projection Model
  9. Image Processing
  10. Geometrically Complementary Patch Selection
  11. Feature Management
  12. Photometric Residual & Kalman Update
  13. Experimental Results
  14. Newer College Results
  15. ENWIDE Dataset
  16. ...and 3 more sections

Figures (6)

  • Figure 1: Top: Accumulated point cloud colorized by intensity and trajectory (orange) resulting from COIN-LIO. Our approach achieves accurate odometry despite geometric degeneracy along the tunnel, resulting in clearly visible correct ground and wall markings. Mid: Filtered intensity with tracked features (orange). Bottom: Top view of the resulting point cloud (gray) and trajectory (orange) from the tunnel.
  • Figure 2: System Overview: The input point cloud is used geometrically (green) for map registration and as a projected image (blue) for photometric error minimization. Both residuals are combined in an iterated update (orange). We use the registration Jacobian to find uninformative directions in the geometry and select features with complementary image information (right bottom). Lines indicate information flow before (- - -) and after (---) the update step.
  • Figure 3: Projection model. The offset between LiDAR origin and laser emitter is denoted as $r$. A measured point is depicted on the top right ($p$).
  • Figure 4: (1): The intensity image is over- (center) and under-exposed (sides). (2): Our filtered image has consistent brightness across the image. (3) & (4): Detail views from a grass field (3) and tunnel (4). The reflectivity image is under-exposed and does not show the ground markings (4). The intensity suffers from strong line artefacts that dominate the texture (3). Our filter removes the line artefacts (Intensity w/o). Our brightness compensation produces consistent exposure and shows details at larger range (ground markings in (4), grass texture in (3)).
  • Figure 5: Top: Example frame Bottom: Features are colored by contribution strength in the uninformative axis along the tunnel (increases from red to green). Uninformative features along the tunnel edges are correctly marked in red, while features with strong gradients along the tunnel show up green.
  • ...and 1 more figures