Skip to main content

Cameras and synchronization

For 3D reconstruction to work, every camera's frames need to correspond to the same moment in time. Without that, a time lag between cameras shows up as inaccurate 3D data: the triangulation math assumes it's looking at the same instant from multiple angles, and a few milliseconds of drift breaks that assumption.

SkellyCam​

SkellyCam is FreeMoCap's synchronized video acquisition component, and its core directive is to make high-quality synchronized recording work with low-cost, consumer-grade hardware. The validation data behind Accuracy, validity, and limits was collected entirely on six USB webcams, about 20each,recordingat1280x720and30FPS,foratotalcameracostofroughly20 each, recording at 1280x720 and 30 FPS, for a total camera cost of roughly 120.

That matters beyond the sticker price. A comparably sized OpenCap setup needs six iOS devices, roughly 1,000to1,000 to 1,500 even with the cheapest available iPhones, and Theia3D's dedicated hardware costs more still. Since the same architecture that keeps SkellyCam's per-camera cost low also makes it cheap to add more cameras, and more cameras generally means more accurate 3D reconstruction (see triangulation and 3D reconstruction), accessibility and accuracy end up reinforcing each other here rather than trading off.

Beyond USB webcams​

$20 webcams aren't a requirement; they're what the validation study happened to use. You can also record with cameras that aren't connected to the computer, like GoPros or phones, and bring the videos in afterwards with Playback > Import Videos. FreeMoCap can line them up using their audio, or a light flash that every camera saw. See process a recording after the fact. That keeps the door open for higher frame rate cameras for athletic movement, or non-USB cameras outdoors.

Single camera versus multi-camera​

FreeMoCap can reconstruct 3D data from a single camera, but monocular reconstruction falls well short of what multiple cameras can provide. A second viewpoint adds real information about where a joint sits in space, rather than requiring the system to infer depth from one image, and it substantially reduces the impact of occlusion, a major failure point for single-camera systems. Multi-camera setups are recommended for anything beyond a quick test.

Next steps​