A Deformable Shape Model for Automatic and Real-Time Dendrometry

We present a stereo image-based algorithm for tree stem diameter measurement and form analysis. The algorithm uses planar parametric curves to represent two-dimensional projections of tree stems in stereo images. The curves evolve according to an energy formulation based on the gradients of the images and inductive priors related to biomechanics and morphology of tree stems. After energy minimization, the curves are reconstructed to three dimensions, allowing for diameter measurements at any point along the height of the stem. We describe the algorithm and report the validation test results comparing predicted diameter measurements to external observations. Our findings demonstrate that the algorithm can automatically estimate diameters for trees within 20 m of the camera with an error of 5.52%. We highlight how this method can aid product value optimization through taper analysis and sweep or crook detection. A run-time analysis shows that the algorithm can estimate dendrometric variables for ten trees simultaneously at 15 frames per second on a consumer-grade computer. Furthermore, we discuss the opportunity to produce training data for machine learning algorithms that generalize across domains and eliminate the need to manually tune parameters.