Motion capture usually calls for specialized cameras, tracking markers, and a space built around the job. Researchers at the Max Planck Institute for Intelligent Systems and Carnegie Mellon University have demonstrated a very different setup using just four iPhones.
Called MAMMA, the system recovers human movement from several camera views without asking performers to wear tracking markers. It can also deal with close interactions between two people, including moments when one person partly disappears behind another.
How does MAMMA pull this off
MAMMA analyzes footage from several viewpoints and tracks 512 landmarks across each person’s body. That gives the system enough information to estimate pose and shape even when someone turns away from a camera or becomes partly obscured.
It also works out where people are touching each other and where their bodies meet the floor. Those extra clues help with movements such as hugging and dancing, where overlapping limbs can make markerless motion capture much harder.
From there, MAMMA builds a 3D body model using ordinary RGB video without requiring physical tracking markers attached to the performer.
How close does MAMMA get to Vicon
The researchers separately evaluated MAMMA against a professional Vicon marker-based motion capture system. In that comparison, the difference in reconstruction error came in at under 1 millimeter.
That figure shouldn’t be read as the measured accuracy of the four-iPhone setup. The researchers demonstrated four-iPhone capture separately, including indoor and outdoor environments. Together, the results show both how accurate the underlying system can be and how little camera hardware it can work with.
Consumer hardware has already become surprisingly capable for spatial capture and tracking, but MAMMA pushes that idea into full-body movement without relying on a conventional mocap rig.
Where could this actually be useful
MAMMA remains research software rather than a finished commercial product. Its code is publicly available, but the current release is intended for non-commercial scientific research, which limits how directly studios or game developers can put it into production.
Still, the four-iPhone demonstrations show why the approach is interesting. Smaller teams could eventually have another route to motion capture without building their workflow around a traditional mocap stage.
The next hurdle is less about proving that smartphone cameras can capture the movement and more about turning the research into something creators can use reliably, legally, and without a research-grade setup.
