ECTLO: Effective Continuous-time Odometry Using Range Image for LiDAR with Small FoV

Xin Zheng,Jianke Zhu

DOI: https://doi.org/10.48550/arXiv.2206.08517

2023-10-19

Abstract:Prism-based LiDARs are more compact and cheaper than the conventional mechanical multi-line spinning LiDARs, which have become increasingly popular in robotics, recently. However, there are several challenges for these new LiDAR sensors, including small field of view, severe motion distortions, and irregular patterns, which hinder them from being widely used in LiDAR odometry, practically. To tackle these problems, we present an effective continuous-time LiDAR odometry (ECTLO) method for the Risley-prism-based LiDARs with non-repetitive scanning patterns. A single range image covering historical points in LiDAR's small FoV is adopted for efficient map representation. To account for the noisy data from occlusions after map updating, a filter-based point-to-plane Gaussian Mixture Model is used for robust registration. Moreover, a LiDAR-only continuous-time motion model is employed to relieve the inevitable distortions. Extensive experiments have been conducted on various testbeds using the prism-based LiDARs with different scanning patterns, whose promising results demonstrate the efficacy of our proposed approach.

Robotics,Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the challenges faced by prism - based LiDAR (especially Risley - prism LiDAR) in robot navigation and localization. Specifically, these new - type LiDAR sensors have the following characteristics and problems:

1. **Small Field of View (FoV)**: Compared with the traditional multi - line mechanical - rotation LiDAR, the horizontal field of view of prism LiDAR is smaller, resulting in a sparse and irregular scanning pattern.
2. **Severe Motion Distortion**: Due to the non - repetitive scanning pattern, the motion distortion is more severe, which affects the quality of point - cloud data.
3. **Irregular Scanning Pattern**: The scanning patterns of these LiDARs are non - repetitive, making feature extraction and point - cloud registration more complicated.

To solve these problems, the author proposes an effective continuous - time LiDAR odometry method (ECTLO), and the main objectives include:

- **Utilizing Range - Image Representation**: Use a single range image to efficiently represent the map and only cover the historical points within the small field of view of LiDAR, thereby improving memory efficiency and adapting to parallel computing.
- **Handling Noisy Data**: In order to deal with the noisy data (such as noise caused by occlusion) that may occur after updating the map, use a filter - based point - to - plane Gaussian Mixture Model (GMM) for robust registration.
- **Compensating for Motion Distortion**: Introduce a continuous - time motion model to alleviate the inevitable motion distortion, especially in the non - repetitive scanning pattern.

Through the above methods, the author aims to improve the odometry accuracy and robustness of prism - based LiDAR, enabling it to be more widely used in various application scenarios.

### Formula Summary

1. **Motion Parameterization in Special Euclidean Space**:
\[
\text{right - }\oplus: Y = X \oplus \xi = X \exp(\xi) \in SE(3)
\]
\[
\text{right - }\ominus: \xi = Y \ominus X = \log(X^{-1}Y) \in \mathbb{R}^6
\]

2. **Spherical Projection Formula**:
\[
\begin{bmatrix}
u \\
v
\end{bmatrix}
= \Pi(x, y, z) =
\begin{bmatrix}
\frac{1}{2} + \arctan\left(\frac{y}{x}\right) \cdot f_h^{-1} \\
\frac{1}{2} - \arcsin\left(\frac{z}{r}\right) \cdot f_v^{-1}
\end{bmatrix}
\]
where \( r = \sqrt{x^2 + y^2 + z^2} \).

3. **E - step and M - step in EM Algorithm**:
\[
m_0^{p_i} = \sum_{q_j} N(p_i(s_{\text{old}}); q_j, \Sigma_{xyz})
\]
\[
m_1^{p_i} = \sum_{q_j} N(p_i(s_{\text{old}}); q_j, \Sigma_{xyz}) q_j
\]
\[
n_{p_i} = \left( \sum_{q_j} N(p_i(s_{\text{old}}); q_j, \Sigma_{xyz}) n_{q_j} \right) / m_0^{p_i}
\]
\[
E_{\text{reg}}(s) = \frac{1}{M} \sum_{i = 1}^M m_0^{p_i} \left( \frac{n_{p_i} \cdot (p_i(s) - m_1^{p_i})}{m_0^{p_i}} \right)^2
\]

4. **Continuous - time**