auto-calibration in single camera view

Simultaneous camera calibration and foot-head homology estimation from frontal human detections. We recognized that parallel shifted homographies with properly chosen coordinate system and parametrization can be written in such a form that it allows to factorize the homography and monomial relations of rotation and focal length in an elegant way to solve for the unknowns in a closed form for minimal case and in the rectangular Quadratic Eigenvalue Problem (QEP) for the overconstrained case.

• B. Micusik and T. Pajdla: Simultaneous surveillance camera calibration and foot-head homology estimation from human detections. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), San Francisco, USA, 2010. accepted

A robust self-calibration method for surveillance cameras, based on a combination of RANSAC (random sample consensus) and EM (expectation-maximization).

• G. Nebehay and R. Pflugfelder: A self-calibration method for smart video cameras. IEEE Workshop on Embedded Computer Vision at Int. Conference on Computer Vision (ICCV), Kyoto, Japan, 2009.

localization and trajectory reconstruction for non-overlapping camera views

Augmentation of Carsten Rother's Direct Reference Plane Method to the case where the cameras' fields of view are non-overlapping. The idea is to see the points as dynamic forming smooth trajectories which allows regularization to overcome the otherwise ill-posed problem.

• R. Pflugfelder and H. Bischof: Localization and Trajectory Reconstruction in Surveillance Cameras with Nonoverlapping Views, IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), 32(4), 2010. [paper link]

demo video

Localizing non-overlapping surveillance cameras under the L-Infinity Norm. We formulate the problem of reconstructing camera centers and 3D points for non-overlapping cameras with known internal camera parameters and known rotations as a Second Order Cone Program (SOCP). We enlarge the class of geometric problems solvable by convex program delivering a global optimum.

• B. Micusik and R. Pflugfelder: Localizing Non-Overlapping Surveillance Cameras under the L-Infinity Norm. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), San Francisco, USA, 2010. accepted