computer vision solution?

bridged with qdn.public.qnxrtp.applications
Post Reply
Julien Beaudry

computer vision solution?

Post by Julien Beaudry » Mon Mar 11, 2002 7:18 pm

Hi,

We are looking for a cost sensitive solution for computer vision under QNX
RTP. We were originally using a pc104+ bt878 based video frame grabber for
which we had a free driver under linux. Drivers for such a device under QNX
RTP are not easy to find (http://www.nexwarecorp.com/ has a commercial one,
and it's the only one we found). This type of grabber is probably the most
affordable solution that offers good quality video. Using a good quality usb
webcam would also be an affordable solution, but I didn't see any existing
drivers for such devices. So we are now considering 3 options:
1-buying the bt878 driver which is expensive
2-porting the bt878 linux driver to qnx ourselves
3-using an usb webcam and write the driver for it ourselves

We are currently building a multi robot laboratory and money and manpower
must be wisely used...

So I was wondering if somebody who is experienced with computer vision under
QNX would have good advices for us. Thanks a lot!

Julien

Srikanth Saripalli

Re: computer vision solution?

Post by Srikanth Saripalli » Tue Mar 12, 2002 8:12 am

Hi,
If you are using QNX 4.25 then Imagenation sells as PC-104+ frame grabber
for which they give the QNX driver for free. If you are using QNX 6.1 then
I am currently porting the driver for a PC104+ progressive scan frame
grabber from linux to QNX 6.1, and I will make it GPLed so that anybody can
use it.

Hope that helps.

Thanks
Srikanth


Julien Beaudry wrote:
Hi,

We are looking for a cost sensitive solution for computer vision under QNX
RTP. We were originally using a pc104+ bt878 based video frame grabber
for which we had a free driver under linux. Drivers for such a device
under QNX RTP are not easy to find (http://www.nexwarecorp.com/ has a
commercial one, and it's the only one we found). This type of grabber is
probably the most affordable solution that offers good quality video.
Using a good quality usb webcam would also be an affordable solution, but
I didn't see any existing drivers for such devices. So we are now
considering 3 options: 1-buying the bt878 driver which is expensive
2-porting the bt878 linux driver to qnx ourselves
3-using an usb webcam and write the driver for it ourselves

We are currently building a multi robot laboratory and money and manpower
must be wisely used...

So I was wondering if somebody who is experienced with computer vision
under QNX would have good advices for us. Thanks a lot!

Julien

QNX

Re: computer vision solution?

Post by QNX » Fri Mar 15, 2002 3:13 am

We are looking for a cost sensitive solution for computer vision under QNX
RTP. We were originally using a pc104+ bt878 based video frame grabber
for
which we had a free driver under linux. Drivers for such a device under
QNX
RTP are not easy to find (http://www.nexwarecorp.com/ has a commercial
one,
and it's the only one we found). This type of grabber is probably the most
affordable solution that offers good quality video. Using a good quality
usb
webcam would also be an affordable solution, but I didn't see any existing
drivers for such devices. So we are now considering 3 options:
1-buying the bt878 driver which is expensive
2-porting the bt878 linux driver to qnx ourselves
3-using an usb webcam and write the driver for it ourselves

We are currently building a multi robot laboratory and money and manpower
must be wisely used...

So I was wondering if somebody who is experienced with computer vision
under
QNX would have good advices for us. Thanks a lot!

Julien
This kind of question comes up a lot. Do the *real* math, and determine
which
is cheapest.

1) If you buy the driver, you don't worry about it anymore. You get either
a
binary that just works, and is supported by the manufacturer, or you get a
documented source set that you only need to learn if something isn't to your
liking. Let's say this driver costs $1000.

2) If you write your own driver, you spend manpower and time to completion.
If you are doing something commercial, decide what the cost of a delay is,
in
terms of lost revenue. You must also decide what the cost per hour of an
average developer is. If your project is non-commercial, just use the
manpower
number. Let's say that a developer costs $20/hour (I'm guessing that your
project is academic, not commercial). Add to this the cost of the rest of
the
lab denizens finding the bugs, and the cost of fixing them.

3) If you port a driver, the same argument as 2) is used, except that you
will
also have to pay your lawyers to determine whether the open-source license
is acceptable to your company. Let's say a lawyer costs $150/hour.

Now the math:

Buy: $1000
Build: $20/hour * 1 person * 160 hours to develop +
$20/hour * 3 people * 40 hours to find all of the bugs +
$20/hour * 1 person * 160 hours to debug =
Total 440 hours * $20 = $8800.
Port: $20/hour * 1 person * 20 hours to port +
$20/hour * 3 people * 40 hours to find all of the bugs +
$20/hour * 1 person * 160 hours to debug +
(4 lawyer hours, if commercial) =
Total 300 hours * $20 = $6000 ( possibly + $600 for lawyer).

These numbers are probably low. Unless you are completely broke, or
you are a masochist, or you are just plain insane, you should buy the
driver.

Just my $0.02.

Cheers,
Andrew

Post Reply

Return to “qdn.public.qnxrtp.applications”