Rolling Shutter Relative Pose Estimation Made Practical

Researchers have made significant advancements in rolling shutter relative pose estimation and strategic classification, introducing new methods that improve accuracy and efficiency.

A team of researchers has introduced affine correspondences to make rolling shutter relative pose estimation practical, reducing the required number of point correspondences from 20 to 7^[1]. The new method exploits the physical smallness of RS parameters to linearize the constraints, achieving the best pose and RS parameter accuracy on the TUM RS benchmark. Additionally, the solver achieves comparable accuracy to the standard 5-point algorithm on the global-shutter EuRoC MAV dataset^[1]. In a separate development, researchers have presented a novel method for approximating the best response in Strategic Classification, allowing for non-linear classifiers in settings with computational intractability^[2]. The method exploits Lagrangian duality to reformulate the strategic response as a constrained optimisation problem and uses the Implicit Function Theorem to compute the total gradient of the loss during classifier learning, resulting in improved strategic accuracy on common machine learning datasets.