Page 309 -

288 6 Feature-based alignment

(a) (b) (c) (d)
Figure 6.6 The VideoMouse can sense six degrees of freedom relative to a specially printed mouse pad using
its embedded camera (Hinckley, Sinclair, Hanson et al. 1999) c 1999 ACM: (a) top view of the mouse; (b) view
of the mouse showing the curved base for rocking; (c) moving the mouse pad with the other hand extends the
interaction capabilities; (d) the resulting movement seen on the screen.

outdoor locations, such as film sets, it is more common to place special markers such as
brightly colored balls in the scene to make it easier to find and track them (Bogart 1991). In
older applications, surveying techniques were used to determine the locations of these balls
before filming. Today, it is more common to apply structure-from-motion directly to the film
footage itself (Section 7.4.2).
Rapid pose estimation is also central to tracking the position and orientation of the hand-
held remote controls used in Nintendo’s Wii game systems. A high-speed camera embedded
in the remote control is used to track the locations of the infrared (IR) LEDs in the bar that
is mounted on the TV monitor. Pose estimation is then used to infer the remote control’s
location and orientation at very high frame rates. The Wii system can be extended to a variety
of other user interaction applications by mounting the bar on a hand-held device, as described
by Johnny Lee. 11
Exercises 6.4 and 6.5 have you implement two different tracking and pose estimation sys-
tems for augmented-reality applications. The first system tracks the outline of a rectangular
object, such as a book cover or magazine page, and the second has you track the pose of a
hand-held Rubik’s cube.

6.3 Geometric intrinsic calibration

As described above in Equations (6.42–6.43), the computation of the internal (intrinsic) cam-
era calibration parameters can occur simultaneously with the estimation of the (extrinsic)
pose of the camera with respect to a known calibration target. This, indeed, is the “classic”
approach to camera calibration used in both the photogrammetry (Slama 1980) and the com-
puter vision (Tsai 1987) communities. In this section, we look at alternative formulations
(which may not involve the full solution of a non-linear regression problem), the use of alter-
native calibration targets, and the estimation of the non-linear part of camera optics such as
radial distortion. 12

11 http://johnnylee.net/projects/wii/.
12 In some applications, you can use the EXIF tags associated with a JPEG image to obtain a rough estimate of a
camera’s focal length but this technique should be used with caution as the results are often inaccurate.