Research

Topics of Research

Pinhole-like Cameras

Self-calibration of Projective Cameras with Parallel Screw Axis Motion

Self-calibration of Projective Cameras with Planar Motion

Generic Cameras

Self-calibration of Generic Cameras with Ground Plane Motion

Self-calibration of Generic Central Cameras using Two Rotational Flows

Computation of Generic Rotational Flows

Self-calibration of Projective Cameras with Parallel Screw Axis Motion

	In [SCIA'11] we presented a closed-form method for the self-calibration of a camera (intrinsic and extrinsic parameters) from at least three images acquired with parallel screw axis motion, i.e. the camera rotates about parallel axes while performing general translations. The considered camera motion is more general than pure rotation and planar motion, which are not always easy to produce. The proposed solution is nearly as simple as the existing for those motions, and it has been evaluated by using both synthetic and real data from acquired images.

In [SCIA'11] we presented a closed-form method for the self-calibration of a camera (intrinsic and extrinsic parameters) from at least three images acquired with parallel screw axis motion, i.e. the camera rotates about parallel axes while performing general translations. The considered camera motion is more general than pure rotation and planar motion, which are not always easy to produce. The proposed solution is nearly as simple as the existing for those motions, and it has been evaluated by using both synthetic and real data from acquired images.

Self-calibration of Projective Cameras with Planar Motion

	We consider the self-calibration problem (affine/metric reconstruction and camera motion estimation) from images acquired with a camera with unchanging internal parameters undergoing planar motion. The general self-calibration methods (e.g. modulus constraint, Kruppa equations) are known to fail with this camera motion. For two images, we present a novel linear constraint on the coordinates of the plane at infinity in a projective reconstruction (linear version of the quartic modulus constraint). The constraint allows us to linearly reconstruct the horizon (line at infinity of the plane of motion) from at least three images, which knowledge is sufficient to determine the camera motion. The self-calibration can be completed by using the apex or another planar motion. See [JMIV'07] and [AGGM'06] for details.

We consider the self-calibration problem (affine/metric reconstruction and camera motion estimation) from images acquired with a camera with unchanging internal parameters undergoing planar motion. The general self-calibration methods (e.g. modulus constraint, Kruppa equations) are known to fail with this camera motion. For two images, we present a novel linear constraint on the coordinates of the plane at infinity in a projective reconstruction (linear version of the quartic modulus constraint). The constraint allows us to linearly reconstruct the horizon (line at infinity of the plane of motion) from at least three images, which knowledge is sufficient to determine the camera motion. The self-calibration can be completed by using the apex or another planar motion. See [JMIV'07] and [AGGM'06] for details.

Self-calibration of Generic Cameras with Ground Plane Motion

	The generic camera model can describe any imaging system. V. Caglioti and P. Taddei showed at OMNIVIS'08 that the planar motion of a generic camera can be estimated by using (a sparse set of) detected and matched features in at least three images of a ground plane. In addition, they gave a preliminar algorithm to perform the rectification of the image of the ground plane. In collaboration with these researchers, in [IJCV'12] we extended and improved the previous work, showing that the rectification can be performed linearly from three views and with quantitetively good accuracy.

The generic camera model can describe any imaging system. V. Caglioti and P. Taddei showed at OMNIVIS'08 that the planar motion of a generic camera can be estimated by using (a sparse set of) detected and matched features in at least three images of a ground plane. In addition, they gave a preliminar algorithm to perform the rectification of the image of the ground plane. In collaboration with these researchers, in [IJCV'12] we extended and improved the previous work, showing that the rectification can be performed linearly from three views and with quantitetively good accuracy.

Self-calibration of Generic Central Cameras using Two Rotational Flows

	The generic central camera model can describe any imaging system with a single effective viewpoint. We address the self-calibration of a smooth generic central camera from only two dense rotational flows produced by rotations of the camera about two unknown linearly independent axes passing through the camera centre. Here we consider the optical flow as the velocity field of the image sequence produced by each camera rotation (i.e. not the image displacement, but its tangent field). In [VISAPP'07] we give a closed-form theoretical solution to this problem, and we prove that it can be solved exactly up to a linear orthogonal transformation ambiguity. In [OMNIVIS'08] we proved the applicability of this method with flows computed from image sequences. See [IJCV'11] for a new geometric formulation of the closed-form self-calibration solution and promising results using three flows.

The generic central camera model can describe any imaging system with a single effective viewpoint. We address the self-calibration of a smooth generic central camera from only two dense rotational flows produced by rotations of the camera about two unknown linearly independent axes passing through the camera centre. Here we consider the optical flow as the velocity field of the image sequence produced by each camera rotation (i.e. not the image displacement, but its tangent field). In [VISAPP'07] we give a closed-form theoretical solution to this problem, and we prove that it can be solved exactly up to a linear orthogonal transformation ambiguity. In [OMNIVIS'08] we proved the applicability of this method with flows computed from image sequences. See [IJCV'11] for a new geometric formulation of the closed-form self-calibration solution and promising results using three flows.

Computation of Generic Rotational Flows

	The generic central camera model can describe any imaging system with a single effective viewpoint. Our self-calibration formulae in [VISAPP'07] involve second order derivatives of the optical flow produced by the rotation of a generic central camera (generic rotational flow). Here we consider the optical flow as the velocity field of the image sequence produced by each camera rotation (i.e. not the image displacement, but its tangent field). The rotational motion of the camera implies some rigidity on the optical flow. In particular, we do not need to worry about occlusions or discontinuities, but the smoothness of the flow is a critical concern. Accordingly, in [OMNIVIS'08] we developed a method for the computation of smooth generic rotational flows using bicubic splines. In [SSBA'10] we explored the use of warping for such computation.

The generic central camera model can describe any imaging system with a single effective viewpoint. Our self-calibration formulae in [VISAPP'07] involve second order derivatives of the optical flow produced by the rotation of a generic central camera (generic rotational flow). Here we consider the optical flow as the velocity field of the image sequence produced by each camera rotation (i.e. not the image displacement, but its tangent field). The rotational motion of the camera implies some rigidity on the optical flow. In particular, we do not need to worry about occlusions or discontinuities, but the smoothness of the flow is a critical concern. Accordingly, in [OMNIVIS'08] we developed a method for the computation of smooth generic rotational flows using bicubic splines. In [SSBA'10] we explored the use of warping for such computation.