A Data-driven 3D Reconstruction Pipeline for Flying Bats
Files
TR Number
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Bats are capable of highly dexterous flight maneuvers that rely heavily on highly articulated hand skeletons and malleable wing membranes. To understand the underlying mechanisms, large amounts of detailed data on bat flight kinematics are required. Conventional methods to obtain these data have been based on tracing landmarks and require substantial manual efforts. In this study, we present a markerless reconstruction pipeline that incorporates a physics simulator to estimate the full three-dimensional geometry of flying bats from synchronized high-speed video recordings. The pipeline employs an iterative optimization procedure that alternates between skeletal pose estimation and wing-membrane deformation optimization. The method was evaluated on 49 flight recordings comprising 73,451 targetpositive frame sets from five bat species recorded in a flight tunnel using 50 synchronized high-speed cameras. Successful reconstructions were obtained for 36,979 frame sets, yielding 66 continuous flight sequences spanning a variety of flight maneuvers and altitude behavior, including straight flight, gliding, turns, and S-curves. Reconstructions were evaluated using image-fitting accuracy, wing-surface curvature, reconstruction yield, and qualitative inspection.