Abstract
Roadside trees in the city play a crucial role in addressing the issues of air pollution, urban heat island effects, road noise, and so on. This paper proposes an efficient and robust method to automatically extract individual roadside trees with morphological parameters from mobile laser scanning (MLS) point clouds for ecological benefits estimation. The proposed method consists of four steps: MLS data pre-processing, pole-like objects classification, individual tree extraction, and morphological parameters calculation for ecological benefits estimation. The proposed method is verified using three complex datasets in Shanghai, China. Comprehensive experiments demonstrate that the proposed method achieves good performance in extracting individual trees in terms of average precision and recall (91.83%, 92.60%), and provides detailed information for ecological benefits estimation.
Introduction
Urban roadside trees are essential "green infrastructure" that mitigate heat islands, reduce noise, and scrub CO2 from the air. However, effective management of these assets requires precise data on their location and physical dimensions (morphology).
Existing remote sensing methods have significant trade-offs. Airborne Laser Scanning (ALS) is effective for mapping tree canopies but often misses trunks and bases due to its nadir perspective. Terrestrial Laser Scanning (TLS) provides high detail but is too slow for city-wide surveys. Mobile Laser Scanning (MLS), mounted on vehicles, offers a high-speed, high-density alternative. However, raw MLS point clouds are often cluttered with buildings, power lines, and overlapping tree crowns, making automated extraction difficult. This paper presents a robust, four-step automated workflow to transform noisy MLS data into actionable ecological metrics.
Methodology
The proposed pipeline follows a logical progression from raw point clouds to refined ecological estimates:
- Pre-processing: To optimize computation, the system filters ground points using a grid-based height difference method and removes large, flat man-made surfaces like building facades using curvature-based region growing.
- Pole-like Object Extraction: The system identifies vertical structures by slicing the data horizontally and applying Euclidean distance clustering. A "quasi-circularity" metric is used to distinguish potential trunks from other urban objects.
- Individual Tree Extraction: This core stage utilizes a hierarchical min-cut approach. A Random Forest classifier calculates the "confidence" of a pole being a tree trunk. This score guides the graph-based min-cut algorithm, ensuring that the most "tree-like" objects are segmented first to prevent accidental data loss.
- Morphological & Ecological Calculation: Isolated trees are modeled with cylinders and spheres to calculate height, crown width, and Diameter at Breast Height (DBH). These parameters are used to calculate the Living Vegetation Volume (LVV), which is then converted into environmental benefits (e.g., kg of SO2 absorbed) using local standards.
Experimental Results
The method was validated using three datasets from Shanghai, China, covering 1.6 km of roadway and 1.8 billion points. These areas ranged from old urban districts with high occlusion (Huangpu) to newer districts with dense, overlapping canopies (Zhangjiang).
The system demonstrated high efficiency and accuracy:
- Accuracy: The pipeline achieved an average precision of 91.83% and a recall of 92.60%.
- Speed: Processing a ~500m road segment took approximately 9 to 10 minutes, proving its viability for large-scale urban inventory.
- Ecological Impact: In the test areas, the system calculated a total LVV of over 11,000 m³, translating to an annual CO2 absorption of roughly 30.87 tons.
Discussion and Key Insights
A primary innovation of this research is the prioritization of "tree-ness" through soft classification. By using trunk probability to guide segmentation, the algorithm becomes resilient to the clutter of urban streets. Furthermore, the paper emphasizes the importance of 3D Living Vegetation Volume (LVV) over traditional 2D green cover indices. By treating trees as 3D volumes, ecological estimates like oxygen production become significantly more accurate.
While the results are strong, challenges remain regarding extremely dense, interlocking canopies and highly irregular or slanted trees. Future work will focus on integrating multi-spectral data to identify specific tree species, which would further refine the ecological benefit constants used in the calculations.
Conclusion
This research provides a scalable, automated solution for quantifying the environmental contributions of urban roadside trees. By bridging the gap between raw laser data and ecological modeling, the proposed method offers city planners a powerful tool for managing green infrastructure and meeting sustainability goals.
Summary updated