A Riemannian Framework for the Elastic Analysis of the Spatiotemporal Variability in the Shape and Structure of Tree-like 4D Objects.

This paper introduces a novel computational framework for modeling and analyzing the spatiotemporal shape variability of tree-like 4D objects, i.e., three-dimensional structures whose shapes deform and evolve over time. Tree-like 3D objects, such as botanical trees and plants, deform and grow at different rates. In this process, they bend and stretch their branches and change their branching structure, making their spatiotemporal registration challenging. We address this problem within a Riemannian framework that represents tree-like 3D objects as points in a tree-shape space endowed with a proper elastic metric that quantifies branch bending, stretching, and topological changes. With this setting, a 4D tree-like object becomes a trajectory in the tree-shape space. Thus, the problem of modeling and analyzing the spatiotemporal variability in tree-like 4D objects reduces to the analysis of trajectories within this tree-shape space. However, performing spatiotemporal registration and subsequently computing geodesics and statistics in the nonlinear tree-shape space is inherently challenging, as these tasks rely on complex nonlinear optimizations. Our core contribution is the mapping of the tree-like 3D objects to the space of the Extended Square Root Velocity Function (ESRVF) [1] where the complex elastic metric is reduced to the L2 metric. By solving spatial registration in the ESRVF space, analyzing tree-like 4D objects can be reformulated as the problem of analyzing elastic trajectories in the ESRVF space. Based on this formulation, we develop a comprehensive framework for analyzing the spatiotemporal dynamics of tree-like objects, including registration and geodesic computation under large deformations and topological differences, statistical summarization through mean trajectories and modes of variation, and the synthesis of new, random tree-like 4D shapes. We demonstrate the effectiveness of the proposed framework on both synthetic 4D botanical tree models and real plant data, illustrating its robustness and practical applicability in capturing complex spatiotemporal shape evolution.