Online Tree Reconstruction and Forest Inventory on a Mobile Robotic System

TL;DR

The paper tackles the challenge of producing timely, accurate forest inventories from mobile laser scanning by developing an online pipeline that incrementally builds submaps, segments trees, reconstructs trunks with a randomized Hough-based model, and maintains global consistency via a pose-graph SLAM framework. Key innovations include a Voronoi-inspired segmentation with multi-height clustering, a spatio-temporal aggregation strategy for a globally coherent DTMs, and real-time tree trait extraction (DBH, stem curve, height) during data acquisition. Experimental results across conifer, deciduous, and mixed plots show competitive DBH accuracy (average RMSE around 1.93 cm) and robust online performance, with significantly reduced data storage relative to TLS pipelines. The work enables immediate marteloscope generation and feedback during field missions, potentially accelerating forest management decisions and reducing post-processing burdens while facilitating mobile-robot-based deployments.

Abstract

Terrestrial laser scanning (TLS) is the standard technique used to create accurate point clouds for digital forest inventories. However, the measurement process is demanding, requiring up to two days per hectare for data collection, significant data storage, as well as resource-heavy post-processing of 3D data. In this work, we present a real-time mapping and analysis system that enables online generation of forest inventories using mobile laser scanners that can be mounted e.g. on mobile robots. Given incrementally created and locally accurate submaps-data payloads-our approach extracts tree candidates using a custom, Voronoi-inspired clustering algorithm. Tree candidates are reconstructed using an adapted Hough algorithm, which enables robust modeling of the tree stem. Further, we explicitly incorporate the incremental nature of the data collection by consistently updating the database using a pose graph LiDAR SLAM system. This enables us to refine our estimates of the tree traits if an area is revisited later during a mission. We demonstrate competitive accuracy to TLS or manual measurements using laser scanners that we mounted on backpacks or mobile robots operating in conifer, broad-leaf and mixed forests. Our results achieve RMSE of 1.93 cm, a bias of 0.65 cm and a standard deviation of 1.81 cm (averaged across these sequences)-with no post-processing required after the mission is complete.

Figures (11)

• Figure 1: Overview of our proposed online tree reconstruction pipeline. The central Tree Manager is fed with payload clouds from the local mapping module as well as the pose graph from the LiDAR SLAM system. It generates a real-time visualization and constructs a marteloscope.
• Figure 2: Overview of the Tree Manager module. Terrain models are produced by a cloth simulation filter (CSF) and clusters are computed by a Voronoi-inspired algorithm. We aggregate them over time in a globally consistent manner using the SLAM pose graph. After reconstruction of the trees, we can extract important tree traits.
• Figure 3: Point cloud $\mathcal{P}_\text{Hough}$ of triplet-circles represented in the Hough space. At a point $\mathcal{C}_\text{Hough}$ with a sphere $\mathcal{S}_{\mathcal{C}_\text{Hough}}$ around it containing many triplet-circles, a good circle fit can be found.
• Figure 4: Stem reconstruction as a stack of oblique cone frustums.
• Figure 5: Scatter plots of the DBH estimates of our reconstructions against reference measurements.