ROBUST APPROACH FOR URBAN ROAD SURFACE EXTRACTION USING MOBILE LASER SCANNING 3D POINT CLOUDS

Abdul Nurunnabi, Felix Norman Teferle, Roderik Lindenbergh, Jonathan Li, S. Zlatanova2022 · DOI

TL;DR

This paper presents a robust, five-step algorithm designed to extract urban road features—like pavement, curbs, and dividers—from noisy Mobile Laser Scanning (MLS) point clouds. By leveraging robust locally weighted regression (RLWR), the approach effectively ignores outliers (like cars or vegetation) and handles complex geometry (like steep slopes) where traditional methods often fail.

What Problem Does This Paper Address?

Mobile Laser Scanning (MLS) is the gold standard for mapping urban environments, but the raw 3D point clouds it produces are "messy." They are frequently riddled with noise, occlusions from parked cars, and outliers like low-hanging branches or road debris.

Most existing road extraction algorithms rely on "classical" geometry or standard regression. These methods are fragile: even a small amount of noise can lead to inaccurate curb detection or misclassification of the road surface. Furthermore, many struggle with "sharp" features like steep curbs or the varying slopes found in complex urban intersections. The authors aim to create a "robust" system—one that remains accurate even when the data is far from perfect.

Methods

The researchers developed a five-step "split-and-merge" workflow that processes the point cloud in manageable chunks:

  1. Slicing: The raw point cloud is sliced into thin "stripes" perpendicular to the direction of the road.
  2. Ground Filtering: For each stripe, the algorithm separates ground points from non-ground points (buildings, trees, etc.) using Robust Locally Weighted Regression (RLWR). This iterative process uses Tukey’s bi-square weight function to "down-weight" points that sit significantly above the estimated ground level, effectively ignoring obstacles.
  3. Patching: The ground-filtered stripes are further subdivided into tiny patches along the road's width.
  4. Elevation Range ($R_z$) Calculation: The algorithm calculates the vertical range ($R_z$) of points within each patch.
  5. Robust Classification: Using the Median Absolute Deviation (MAD)—a robust statistical measure of variability—the algorithm identifies patches with "abnormal" elevation ranges. These outliers are classified as curbs or road dividers. Once the curbs are found, the remaining points are classified as pavement (between the curbs) or footpaths (outside the curbs).

Datasets

The algorithm was evaluated on two real-world MLS datasets:

  • Experiment 1: A 53-meter urban road corridor containing over 1.1 million points, featuring typical urban furniture, trees, and steep slopes.
  • Experiment 2: A 31-meter section (approx. 900k points) that included more challenging elements like curved road sections, road islands, and vehicles that partially occluded the curbs.

Key Insights and Ideas

  • Local beats Global: Global parametric models often smooth over the very details needed to find a curb. By using local regression within small neighborhoods, the algorithm preserves sharp edges and corners.
  • The Power of Robust Statistics: Traditional Least Squares regression is easily skewed by a single outlier. By using RLWR and MAD, the authors ensure that a stray point on a bush or a car bumper doesn't trick the system into thinking the road has a 2-meter high curb.
  • Curbs as Statistical Outliers: One of the "aha" moments is treating the elevation range of a patch as a data point. In a flat road, most patches have a very small $R_z$. Curbs naturally stand out as statistical outliers in this distribution, making them detectable without complex geometric templates.

Key Results

The proposed method significantly outperformed existing benchmarks, particularly in its ability to ignore non-ground objects that other models mistook for curbs.

  • Curb Detection (Exp 1): Achieved a 97.28% Precision and 100% Recall, with a Matthews Correlation Coefficient (MCC) of 0.986. For comparison, a popular existing method (Zhao et al., 2021) achieved only 72.61% precision on the same data.
  • Pavement Accuracy: In both experiments, pavement extraction was nearly perfect, with precision and recall scores consistently above 99.5%.
  • Versatility: The model proved successful for both straight and curved road sections, as well as roads with complex features like central dividers/islands.

Limitations and Future Work

While robust, the system is not yet perfect:

  • Occlusions: In Experiment 2, the algorithm missed some curb points because they were physically blocked by parked vehicles. The current version cannot "guess" where a curb is if the laser can't see it.
  • Parameter Sensitivity: Some settings, such as the size of the stripes and patches, are data-dependent. This means a human might need to "tune" the algorithm when moving from one city's data to another.
  • Future Directions: The authors plan to incorporate robust curb-line fitting. This would allow the system to mathematically "fill in" the gaps in curbs caused by obstructions like cars or thick vegetation.

Summary updated

Abstract

Road surface extraction is crucial for 3D city analysis. Mobile laser scanning (MLS) is the most appropriate data acquisition system for the road environment because of its efficient vehicle-based on-road scanning opportunity. Many methods are available for road pavement, curb and roadside way extraction. Most of them use classical approaches that do not mitigate problems caused by the presence of noise and outliers. In practice, however, laser scanning point clouds are not free from noise and outliers, and it is apparent that the presence of a very small portion of outliers and noise can produce unreliable and non-robust results. A road surface usually consists of three key parts: road pavement, curb and roadside way. This paper investigates the problem of road surface extraction in the presence of noise and outliers, and proposes a robust algorithm for road pavement, curb, road divider/islands, and roadside way extraction using MLS point clouds. The proposed algorithm employs robust statistical approaches to remove the consequences of the presence of noise and outliers. It consists of five sequential steps for road ground and non-ground surface separation, and road related components determination. Demonstration on two different MLS data sets shows that the new algorithm is efficient for road surface extraction and for classifying road pavement, curb, road divider/island and roadside way. The success can be rated in one experiment in this paper, where we extract curb points; the results achieve 97.28%, 100% and 0.986 of precision, recall and Matthews correlation coefficient, respectively.
BibTeX
@inproceedings{nurunnabi2022robust,
  author = {Abdul Nurunnabi and Felix Norman Teferle and Roderik Lindenbergh and Jonathan Li and S. Zlatanova},
  title = {{ROBUST APPROACH FOR URBAN ROAD SURFACE EXTRACTION USING MOBILE LASER SCANNING 3D POINT CLOUDS}},
  booktitle = {˜The},
  year = {2022},
  doi = {10.5194/isprs-archives-xliii-b1-2022-59-2022},
}
Figure