StemReg: A Marker-Free Automated Method for Registering Multiscan Terrestrial Laser Scanning Data in Forests Using Stem Mapping

Registering multiscan terrestrial laser scanning (TLS) data is a prerequisite for applying TLS in precision forestry. Marker-free automatic registration methods based on tree attributes or stem positions have been considered promising solutions. However, tree attribute–based methods rely heavily on accurate data preprocessing, while stem-based methods are often sensitive to parameter settings. We propose an automatic, robust, and efficient marker-free registration method (StemReg). Our approach first extracts stem candidates using a density-adaptive RANSAC algorithm tailored to local point densities, followed by a visibility check to eliminate occluded stems. Stem correspondences between scans are then established based on spatial configuration, and the registration transformation is estimated accordingly. Experiments conducted on multiple plots with diverse vegetation structures demonstrate that StemReg achieves registration accuracy comparable to that of manual methods, with an average mean deviation (MD) difference of only 0.8 cm across all test plots. Compared to state-of-the-art stem-based methods, the proposed approach retains comparable effectiveness while enhancing robustness and extending applicability to more complex forest environments. Our method is expected to provide reliable support for the application of TLS in large-scale forestry research.