In terms of viewing figures, live sports events are one of the most important contents for broadcasting stations. Therefore, these programs have top priority concerning innovations and the usage of new technologies. Not only sports events but also video productions are often faced with combining real and virtual graphics. Up to now, this process of combining real world objects with virtual reality was a consuming task in terms of time, money and personnel. Only a few tools (e.g. Adobe After Effects) do support the combining of virtual and real objects by using motion tracking.

The main goal of the project v-graph is the implementation of a user-friendly and semi-automatic tool, which is able to calculate a camera calibration model. Calibration is considered as the extraction of 3D information from 2D image data which is also known as the determination of intrinsic and extrinsic camera parameters. The interrelationship between image, camera and world coordinate system is shown in figure 1. An object point P from the world coordinate system is projected on the image plane (p) by a straight line going through the projection center O.

The camera projection model is then able to find a corresponding 3D point in the real world for each 2D point from the image plane and vice versa. This enables a considerably easy integration of virtual images and animations in real images. The project v-graph creates a completely new interface between the producers of real motion media and the creators of virtual worlds.

The project v-graph is expanding this model by an essential factor: A mapping between world and image points can not only be performed by pan and tilt of the camera but also by zooming.
The main calibration algorithm to recover intrinsic and extrinsic parameters of a PTZ camera is based on the methods proposed in [2] and [3].

In a first step, corresponding interest points are collected using the Random Sample Consensus ( RANSAC) algorithm. In general, the RANSAC method takes some points for calculating an initial guess of the resulting parameters and searches for all the samples which fit to the initial parameter estimation under the assumption of a certain threshold. This simply means, that outliers get excluded from further processing. In our case, RANSAC is used to find the corresponding points between multiple different views and to remove those points, where no corresponding point is found in the all views.

After having corresponding points, the inter-image homographies may be estimated. A homography mathematically describes the projection of a 3D ray onto two different images ( or views ).

When the homography is induced by the plane at infinity, the homography is called infinite homography. An infinite homography maps the points between an image and the plane at infinity. This homography can be calculated by using the given relative rotation of the camera.

The calculation of the infinite homography is necessary to recover the image of the absolute conic, which is also located on the plane at infinity. The image of the absolute conic is then used to recover the intrinsic parameters of the PTZ camera.


Throughout the project, the following PTZ camera was used.

Sony SNC-RZ50P