Intensity Scan Context: Coding Intensity and Geometry Relations for Loop Closure Detection

Han Wang, Chen Wang, Lihua Xie

Abstract

Loop closure detection is an essential and challenging problem in simultaneous localization and mapping (SLAM). It is often tackled with light detection and ranging (LiDAR) sensor due to its view-point and illumination invariant properties. Existing works on 3D loop closure detection often leverage the matching of local or global geometrical-only descriptors, but without considering the intensity reading. In this paper we explore the intensity property from LiDAR scan and show that it can be effective for place recognition. Concretely, we propose a novel global descriptor, intensity scan context (ISC), that explores both geometry and intensity characteristics. To improve the efficiency for loop closure detection, an efficient two-stage hierarchical re-identification process is proposed, including a binary-operation based fast geometric relation retrieval and an intensity structure re-identification. Thorough experiments including both local experiment and public datasets test have been conducted to evaluate the performance of the proposed method. Our method achieves higher recall rate and recall precision than existing geometric-only methods.

Introduction

In Simultaneous Localization and Mapping (SLAM), loop closure detection is the critical ability of a robot to recognize a previously visited location. This process is essential for correcting the cumulative drift in a robot's estimated trajectory. While visual sensors often struggle with varying illumination and weather conditions, LiDAR sensors offer a more robust alternative. However, most existing LiDAR-based methods rely exclusively on the geometric shape of the environment. This paper argues that geometry alone can be ambiguous—for instance, different hallways or forest paths may appear identical. By incorporating the intensity channel, which records the reflectance of surfaces (e.g., distinguishing a metal sign from a concrete wall), the proposed Intensity Scan Context (ISC) provides a more unique "fingerprint" for place recognition.

Intensity Scan Context (ISC)

The core of the proposed method is the ISC descriptor, which integrates both spatial structure and material properties. The 3D point cloud is projected onto a 2D polar grid consisting of rings and sectors. For each cell in this grid, the descriptor stores the maximum intensity value of the points contained within it. This results in a 2D matrix that captures the reflectance structure of the scene.

Intensity Scan Context (ISC) visualization alongside corresponding camera views, illustrating how environmental features are encoded into the descriptor for matching.

Hierarchical Retrieval and Verification

To ensure real-time performance, the authors implement a two-stage hierarchical matching process:

  1. Fast Geometry Re-identification: The intensity matrix is first simplified into a binary matrix representing the presence or absence of points. Fast bitwise XOR operations are then used to quickly filter the database for potential candidates and identify the relative rotation (yaw) between scans.
  2. Intensity Structure Matching: For the top candidates, a more granular comparison is performed using cosine distance on the original intensity values.

Finally, the system employs consistency verification to prevent false positives. This includes checking for temporal consistency across sequential scans and geometrical consistency via Iterative Closest Point (ICP) alignment.

Experimental Evaluation

The ISC method was evaluated using the standard KITTI Dataset (specifically sequences 00, 05, and 02) and data from a local warehouse robot equipped with a Velodyne VLP-16.

A typical street scene from the KITTI dataset, representing the urban environment used for benchmarking loop closure detection.

Key results include:

  • Superior Accuracy: ISC achieved a precision of 98% and a recall of 91% on the challenging KITTI sequence 02, which features "reverse visits" where the robot returns from the opposite direction. It outperformed existing geometric-only methods like Scan Context and vision-based methods like DBoW2.
  • High Efficiency: The entire query process takes approximately 1.2 ms on a standard desktop, making it highly suitable for real-time applications.
  • Robustness: The system successfully corrected trajectory drifts in warehouse environments where vision-based SLAM failed due to feature scarcity and viewpoint changes.

Conclusion

Intensity Scan Context demonstrates that LiDAR intensity is a valuable signature for place recognition, particularly in handling the "reverse visit" problem. By utilizing a polar representation, changes in heading are reduced to simple column shifts in the descriptor matrix. While the method shows some limitations in feature-poor rural areas, it offers a significant advancement in robust, real-time loop closure detection for autonomous systems.

Summary updated

Abstract

Loop closure detection is an essential and challenging problem in simultaneous localization and mapping (SLAM). It is often tackled with light detection and ranging (LiDAR) sensor due to its view-point and illumination invariant properties. Existing works on 3D loop closure detection often leverage the matching of local or global geometrical-only descriptors, but without considering the intensity reading. In this paper we explore the intensity property from LiDAR scan and show that it can be effective for place recognition. Concretely, we propose a novel global descriptor, intensity scan context (ISC), that explores both geometry and intensity characteristics. To improve the efficiency for loop closure detection, an efficient two-stage hierarchical re-identification process is proposed, including a binary-operation based fast geometric relation retrieval and an intensity structure re-identification. Thorough experiments including both local experiment and public datasets test have been conducted to evaluate the performance of the proposed method. Our method achieves higher recall rate and recall precision than existing geometric-only methods.
BibTeX
@inproceedings{wang2020intensity,
  author = {Han Wang and Chen Wang and Lihua Xie},
  title = {{Intensity Scan Context: Coding Intensity and Geometry Relations for Loop Closure Detection}},
  booktitle = {IEEE International Conference on Robotics and Automation},
  year = {2020},
  url = {https://openreview.net/forum?id=CenQaGEsDB},
}
Figure