Tree Motion in Three Dimensions: Video-Based Reconstruction using a Calibrated Multi-Camera System

Sammanfattning

Wind-induced tree damage poses a significant risk in both managed forests andurban environments. While computational models could improve how these risksare assessed, their development and validation require reliable experimental data,including spatially resolved measurements of tree motion. A low-cost, non-invasivevideo-based method is presented for reconstructing time-resolved three-dimensionaltree motion from a calibrated three-camera system. Physically attached colourmarkers are detected through image segmentation, initialised using triangulationand tracked using an Extended Kalman Filter with a Singer acceleration model,requiring no training data. The method was tested and validated in a controlledmotion-tracking experiment against an industrial robot arm and a terrestrial Li-DAR scanner. It achieves sub-centimetre accuracy with point-wise variation closeto the limit set by the camera resolution. The method produced reproducible threedimensionaltrajectories and relatively consistent frequency estimates wherever themarkers remained sufficiently visible when it was applied to pull-and-release experimentson a spruce, pine and birch tree. The reconstructed motion revealed locationdependentdynamics, with stem and branch features showing different apparentfrequencies and decay behaviour. Branch features generally sustained oscillationslonger than stem features, consistent with the role branches are thought to play indamping tree motion. Tracking reliability depended strongly on marker visibility,with spruce performing best and pine proving most challenging due to dense foliageand marker occlusion. The thesis establishes that a small, low-cost multi-camera systemcan deliver accurate, physically interpretable three-dimensional measurementsat multiple points simultaneously, providing a foundation for extending the methodto mature trees under natural wind loading outdoors.