Re: Stereo Camera Optical Rangefinding >1000m - Solution Sought
From: CR Optiker (optiker.crb_at_pnl.gov)
Date: 08/23/04
- Next message: XQWU: "What is "detection / false positive rates" in fingerprint recognition?"
- Previous message: Cas Hill: "Radar Systems/Signal Processing Position"
- In reply to: TomO: "Stereo Camera Optical Rangefinding >1000m - Solution Sought"
- Messages sorted by: [ date ] [ thread ]
Date: Mon, 23 Aug 2004 13:32:08 -0700
On 16 Aug 2004 20:29:52 -0700, TomO wrote:
> reflective target. Anticipated targets are 1500m max.
>
> Is it a sensible quest to attempt stereo video digitising to generate depth
> information at this range? Am I stuck with the 80cm baseline? Autofocus
> on a digital SLR has been mentioned but that does seem optimistic. 10% accuracy
> is good enough - more a bonus.
Tom...
Just in case you're still watching this and in case you haven't resolved it
by doing some calculations as suggested by the other post, I'd suggest the
calculations. However, I've done this experiment but not with 80 cm
baseline, but with only 30 cm, and am not at all optimistic you can get 10%
at that distance. We were doing ranging inside buildings and if I recall
correctly (we did this maybe 10-15 years ago) found that digitizing to
512x480, with a 25 mm lens on a Cohu camera with a 1/2" CCD, 30 cm
baseline, and high contrast targets, we were getting about 1% precision at
about 12 meters. The parallactic separation decreases nonlinearly with
range as you can tell by just visual observation.
This was a relatively quick and dirty feasibility demo, so we probably
could have done somewhat better, but just my opinion, I'd guess 10% at 1500
meters is going to be tough.
Obvious improvements are greater baseline, higher res images, and assuring
that optics aren't the limiting feature by having good lenses. Assuming
outdoors, atmospheric conditions can cause problems. Focal length is an
important parameter, so knowing that accurately will help since nominal
focal length of off-the-shelf lenses can be accurate to only within 10%.
The math is pretty simple for a quick calc.
range = (baseline*focal length)/(shift of an image point between the two
images due to parallax)
Optiker
- Next message: XQWU: "What is "detection / false positive rates" in fingerprint recognition?"
- Previous message: Cas Hill: "Radar Systems/Signal Processing Position"
- In reply to: TomO: "Stereo Camera Optical Rangefinding >1000m - Solution Sought"
- Messages sorted by: [ date ] [ thread ]
Relevant Pages
|
|
