(0 2
— dd eda pd
(D 0 2 (0
outer orientation is stable through the whole
sequence, then the bundle adjustment is only done
once in the beginning, but if it is unstable it will be
recalculated when needed.
The 3D measuring part of the Track Eye system for
high speed film consists of two main parts:
- Calibration
- 3D calculation
Both parts have similar interfaces regarding their
data-organisation and error handling.
The calibration is done for each camera separately on
a known test-field. Since some of the calibration
parameters, mainly the principal point, are unstable,
these can also be estimated in the 3D calculation part.
An eight bit flag controls the different camera status'
and the action to be taken by the program and/or
operator.
The 3D calculation is done in two steps. In the first
image set the orientation of the cameras (and possibly
the principal point) is established by bundle
adjustment. If the cameras have stable inner and
outer orientation the next frames will be calculated
directly from these parameters. If the camera is
unstable in some manner, which is often the case
with high speed film cameras, the bundle adjustment
will be redone when needed.
Point measurements 3D Calculations
Image coordinates
approximate orient. Calibration type
Image 0 Numbers of parameters Calibration
to be calibrated
o on known
X ner & outer. X test field
= orientation
&
=
Image coordinates @ Approx orientation
Image 1 = -
Ground truth 9 Approx obj. coord.
o Self calibration
& or only bundle adj.
® ©
& 5
= =
« ©
Image coordinates S 2
o =
= o
8 d
Image 2 = if ;
9 6 Stability >
— ©
Image n Control =
8
g
Continue with next frame OK Instab z
<< < > 9
Updated inner and
outer orientation
parameters
Image Sequence
fig4 3D module for TrackEye
4.2 Digitizing and Tracking
In the TrackEye Film Scanner the approach taken to
film digitizing is based on a CCD linear array camera,
which builds up an image by continuously moving
the film past the CCD array (Kállhammer, 1990). The
array covers the whole width of the 16 mm cine film
in order to be able to collect also timing and reference
mark, fig 5. The horizontal resolution is given by the
2592 pixels in the CCD array, giving 6.2 um for each
pixel. The vertical resolution is set up to be the same,
giving square pixels. A full frame is approximately
2592 by 1230 pixels, or about 3 Mega-bytes of data.
The digitizing rate is 0.83 seconds/frame on average if
a sequence of 100 frames including start-up time. The
data is stored on high speed, high capacity discs to
faciliate fast and easy access in the coordinate
extraction and analysis process. The disc system has a
normal storage capacity of 3.1 Giga-bytes, giving the
possibility to store at least 1000 full-frame images. The
access time for a 3 Mega-byte image is 0.3 seconds.
The tracking process is gray scale based and aided by
adaptive filters and path prediction. The tracking can
be performed in fully automated, semi-automated or
manual mode. A dedicated pipelined two-
dimensional processor is used for the compu-
tationally heavy tracking, image handling, zoom and
filtering operations.
Mechanical
continous feed
1 100 Points
Automatic extraction
: 7*7 of Timing Markers
fig5 Digitizing the film
4.3 Calibration
The calibration of the high speed cameras are done as
flexible as possible, since the type and construction of
the cameras may vary. The calibration is done either
as a single camera calibration with a known test field
or as additional parameters in the bundle adjustment
(self calibration).
One of the largest problems with most high speed
cameras is the lacking ability to produce some sort of
fiducial marks. The problem is fundamental since
most calculations are based on the collinearity
conditions. Instead of using fiducial marks in the
camera, stable object points may be used as reference
marks through one sequence of images. A
requirement for this to work is the stability of the
outer orientation during a sequence, a requirement