FLEX: Extrinsic Parameters-free Multi-view 3D Human Motion Reconstruction

The increasing availability of video recordings made by multiple cameras has offered new means for mitigating occlusion and depth ambiguities in pose motion reconstruction methods. Yet, multi-view algorithms strongly depend on camera parameters, particularly relative transformations between the cameras. Such a dependency becomes hurdle once shifting to dynamic capture uncontrolled settings. We introduce FLEX (Free muLti-view rEconstruXion), an end-to-end extrinsic parameter-free model. is (dubbed ep-free) sense that it does not require parameters. Our key idea 3D angles skeletal parts, as well bone lengths, are invariant position. Hence, learning rotations lengths rather than locations allows predicting common values all views. network takes streams, learns fused deep features through novel fusion layer, reconstructs single consistent skeleton with temporally coherent joint rotations. demonstrate quantitative qualitative results three public data sets, multi-person synthetic streams captured compare our model state-of-the-art methods ep-free show absence we outperform them large margin while obtaining comparable when parameters available. Code, trained models, other materials available https://briang13.github.io/FLEX .