The Millimeter-Wave Radar SLAM Assisted by the RCS Feature of the Target and IMU

Yang Li, Yutong Liu, Yanping Wang, Yun Lin, Wenjie ShenSensors 2020 · DOI

TL;DR

This paper introduces a novel millimeter-wave (MMW) radar SLAM system that leverages the Radar Cross Section (RCS) of targets and IMU data to overcome the inherent challenges of MMW radar, such as low data volume and precision. By combining multi-scan point clouds and using RCS-enhanced DBSCAN for outlier filtering, the system achieves more accurate and robust localization and mapping, particularly in challenging environments.

What Problem Does This Paper Address?

Simultaneous Localization and Mapping (SLAM) is crucial for autonomous systems, but traditional sensors like LiDAR and cameras have limitations (e.g., weather dependency, lighting conditions). Millimeter-wave (MMW) radar offers all-weather, all-day capabilities and can measure target speed and RCS, making it an attractive alternative. However, MMW radar data suffers from small point cloud volumes, a high number of outliers, and low precision, which significantly degrade SLAM performance. This paper aims to address these limitations by proposing an MMW radar SLAM system that effectively utilizes RCS information and IMU data to improve accuracy and robustness.

Methods

The proposed system involves several key steps:

  • Data Preprocessing:
    • Region of Interest (ROI) Filtering: Low-precision points are filtered out based on their position within the radar's detection range, as precision varies with distance and angle.
  • Multi-scan Creation: Odometry and IMU data are used to combine consecutive radar scans into "Multi-scans." This increases point cloud density and allows for the estimation of radar velocity, aiding in the filtering of moving targets.
  • Outlier Filtering with DBSCAN and RCS: The Density-based Spatial Clustering of Applications with Noise (DBSCAN) algorithm is employed to filter outliers. Crucially, this paper extends DBSCAN by incorporating the Radar Cross Section (RCS) feature of targets alongside their 2D position. Mahalanobis distance is used to measure similarity in this multi-dimensional feature space, enabling adaptive parameter determination for DBSCAN.
  • Localization and Mapping:
    • Improved Correlative Scan Matching (CSM): The authors propose an improved "Scan to Map" CSM method, which matches the current radar scan against a local submap of the global grid map, rather than just the previous scan.
  • RCS-Occupancy Grid Map (R-OGM): A novel map representation, R-OGM, is introduced. This map utilizes RCS information to assign different probabilities to grid cells, enhancing the representation of strong reflectors and improving landmark extraction. Three variations of the improved CSM method are presented, with the third method, which uses RCS as a weight in an occupancy grid map, demonstrating the best performance.

Datasets

The paper uses three groups of experimental data collected from different scenarios:

  1. Scene 1: A relatively empty ground parking lot with few parked cars, representing a degraded environment.
  2. Scene 2: A suburban street with stationary objects (trees, guardrails, vehicles) and some moving targets (pedestrians, bicycles, cars), presenting a challenging scenario with both weak and strong reflectors.
  3. Scene 3: A relatively ideal scene with stationary vehicles on both sides of the road, providing stable geometric information.

The paper also references the Multimodal range dataset (Mulran) and Oxford radar robotcar dataset as existing benchmarks, noting that their radar data refresh rates are lower than the one used in this study.

Key Insights and Ideas

  • RCS as a Discriminative Feature: The paper highlights that RCS is a valuable feature for MMW radar SLAM, going beyond simple positional data. It helps to distinguish between different types of targets and improve outlier rejection.
  • Adaptive DBSCAN for MMW Radar: Standard DBSCAN struggles with the non-uniform density and noise in MMW radar data. Incorporating RCS and using Mahalanobis distance allows for more effective outlier identification and adaptive parameter tuning.
  • "Scan to Map" for Robustness: Shifting from "Scan to Scan" to "Scan to Map" matching in CSM significantly improves robustness, especially in environments with sparse data.
  • R-OGM for Enhanced Mapping: The R-OGM representation, which leverages RCS, provides a more detailed and geometrically accurate map, particularly beneficial for landmark extraction.
  • IMU for Multi-scan and Motion Compensation: IMU data is crucial for creating denser point clouds ("Multi-scans") and estimating radar motion, which is essential for filtering moving targets and improving overall SLAM performance.

Key Results

  • Outlier Filtering: The proposed DBSCAN method with RCS features demonstrated better outlier recognition compared to methods without RCS, especially in areas with uneven outlier distribution.
  • Localization Accuracy: The proposed SLAM system achieved better trajectory accuracy and robustness across various scenarios, including degraded environments, compared to Odometry, original CSM, and even the improved CSM without full preprocessing.
  • Map Quality: The R-OGM map representation effectively highlighted the geometric contours of targets, especially strong reflectors, outperforming standard Occupancy Grid Maps (OGM) in this regard.
  • Data Volume Reduction: The data preprocessing steps (ROI filtering and outlier removal) significantly reduced the point cloud data volume, leading to reduced computational load and storage requirements. For instance, in Scene 1, the screening rate reached 65.91%.

Limitations and Future Work

  • Odometry Fusion: The authors acknowledge that the current Odometry fusion method might introduce errors due to ground slippage and hardware inaccuracies. They suggest exploring better fusion methods or graph optimization for future work.
  • Back-end Optimization: The proposed SLAM system does not include back-end optimization. The authors plan to incorporate graph optimization in future work to further refine the pose estimates and improve accuracy.
  • RCS Assumption: The assumption that RCS is constant for static targets within a short time interval might be too simplistic for a full 360° observation, as RCS can be irregular. This was a limitation identified in one of the CSM variants.
  • Path Planning: While R-OGM is beneficial for landmark extraction, it might be less suitable for path planning as it can suppress information from small targets with low probability.

Summary updated

Abstract

Compared with the commonly used lidar and visual sensors, the millimeter-wave radar has all-day and all-weather performance advantages and more stable performance in the face of different scenarios. However, using the millimeter-wave radar as the Simultaneous Localization and Mapping (SLAM) sensor is also associated with other problems, such as small data volume, more outliers, and low precision, which reduce the accuracy of SLAM localization and mapping. This paper proposes a millimeter-wave radar SLAM assisted by the Radar Cross Section (RCS) feature of the target and Inertial Measurement Unit (IMU). Using IMU to combine continuous radar scanning point clouds into "Multi-scan," the problem of small data volume is solved. The Density-based Spatial Clustering of Applications with Noise (DBSCAN) clustering algorithm is used to filter outliers from radar data. In the clustering, the RCS feature of the target is considered, and the Mahalanobis distance is used to measure the similarity of the radar data. At the same time, in order to alleviate the problem of the lower accuracy of SLAM positioning due to the low precision of millimeter-wave radar data, an improved Correlative Scan Matching (CSM) method is proposed in this paper, which matches the radar point cloud with the local submap of the global grid map. It is a "Scan to Map" point cloud matching method, which achieves the tight coupling of localization and mapping. In this paper, three groups of actual data are collected to verify the proposed method in part and in general. Based on the comparison of the experimental results, it is proved that the proposed millimeter-wave radar SLAM assisted by the RCS feature of the target and IMU has better accuracy and robustness in the face of different scenarios.
BibTeX
@inproceedings{li2020millimeter,
  author = {Yang Li and Yutong Liu and Yanping Wang and Yun Lin and Wenjie Shen},
  title = {{The Millimeter-Wave Radar SLAM Assisted by the RCS Feature of the Target and IMU}},
  booktitle = {Sensors},
  year = {2020},
  doi = {10.3390/s20185421},
}
Figure