Markerless 3D spatio-temporal reconstruction of microscopic swimmers from video

3D object reconstruction of deformable objects is a long standing challenge for computer vision. Here we develop a system for the 3D reconstruction of a single marker-less object – a freely moving biological swimmer in 3D space – using a passive, fixed-camera set-up. We focus on microscopic, long and thin (1 mm long, 80 µm thick) roundworms. Our set-up provides the resolution required both to track the animal’s coordinates across a large volume and to reconstruct its 3D posture at every frame. A data-pipeline is presented which combines model calibration, 2D image analysis and 3D reconstruction of the body midline, representing the complete posture up to orientation and internal twist. We present results, validation and open challenges, including instances of occlusion due to insufficient projected information, and experimental limitations of resolution and focus.