TD,
Just a quick note for now ... more to follow.
1) Sorry Herschel jumped on you ... you wrote 'images', not 'image'.
2) Those 'industry standard' 1280x1024 @ 500 fps cameras (most based
on the venerable Micron MV-13 sensor) are now selling for under U$D
15k. Beware that many are progressive scan, eg, the camera reads-out
one line at a time @ 2 us per line, so 2 ms elapses per frame. For
some targets, that'll s-t-r-e-t-c-h the image a bit, as if you pulled
on two opposite corners.
3) You'll notice a minor discrepancy in target speed in prior posts -
like three zeroes worth. I get 100,000 m/hr / 3600 sec/hr => 28
meters (not cm) / second. That's 56 cm in 2 ms, which can be a fair
amount of blur. Hold the camera the 'normal' way, and the tires will
move a good bit more than the roof of the truck during the one
frame. The 'solution' in your case, since you are looking for
vertical movement in the target during its horizontal path, is turn
the camera on its side; the image of the vehicle will be a nice,
satisfying rectangle, either stretched or compressed, but the
distortion won't work too much against you.
4) The lateral (across the FoV) movement of the truck is the base of
a triangle - if you do not know the exact height of the triangle (eg
exactly how *far* to the side of the road) you cannot know how many
degrees of lateral movement you have across the FoV and, therefor,
how many pixels that movement will cover with a given sensor. 'By
the side of the road' in America can mean 50 yards (the easements are
often quite wide) while the same phrase in a rural environment in
other countries might mean 2 meters. You may need to work this
backwards, eg, figure-out what's possible in terms of degrees of
movement across the sensor, and then move the camera far enough away
- the further away, the smaller an angle the same movement will make.
5) The isoceles triangle (assumed above) offers the advantage that
the further away you place the sensor, the less depth-of-field you
need for the target to be in focus during its movement. That's the
good news. The bad news is that the target will move very quickly
across the FoV (Field of View). If you make a shallow right-angle
triangle - eg, place the camera closer to the road, and shoot the
truck as it 'approaches' or 'recedes' - you get less lateral movement
across the sensor and, therefor, a longer period of time to
shoot. The problem then becomes maintaining focus, as the truck
moves towards you.
5a) One way to maintain focus is to do the math (in a spreadsheet),
buy a motorized lens, and set the software that controls the
translator/actuator/motor on the lens to read the math (a smooth
curve, but a curve).
5b) Another way to maintain focus is to do real-time image
processing, which you intend to do anyway, adjusting the focus
frame-by-frame. This is not as hard to do as it sounds. Put a white
paper plate on the front of the truck, measure it, and you're most of
the way to a solution.
5c) A freebie you get when you motorize & control focus is that you
can also motorize & control zoom; you've already done the math, which
is the hard part, so you might as well do the zoom.
6) Speaking of white paper plates, people who do crash testing put
stickers all over their targets. Typically we're talking about a 10
cm diameter circle, with two white quadrants and two black quadrants
... no, the white quadrants are *not* adjacent to each other ... that
would be one white hemisphere and one black hemisphere ... I hope you
have a sense of humor. Shooters buy blaze orange (or yellow) circles
to put on their paper targets (no fair putting them on the deer) to
help when zeroing their sights (etc); cheap shooters use yellow
Post-It notes, which are only cheaper because they steal them from
their place of employment. Simply put, you need a visible (good
contrast against its background) object of known size placed on the
target; this will help both the focus control (above) and the
bump'b'bounce measurements you're making.
7) And all that, of course, assumes you control the target, and are
not just filming whatever truck happens to drive by (my, lots of
assumptions today). If you are filming targets of opportunity, the
shallow angle approach will kill two birds with one stone, as you can
use the front or rear license plate (often nicely visible, and of
known dimensions) for both the focus/zoom control and for the
bump'n'bounce measurements.
8) In *all* of these scenarios, you will need to do some extra math
if you want to know the *speed* of the bumps'n'bounces, rather than
just the maximum travel up'n'down.
Okay, so much for "short'.
More to follow on camera choices,
/s/ Michael Storch
Ask not what your laptop can do for you,
Ask what you can do with your necktop.
