| |
 |
|
|
Science Forum Index » Image Processing Forum » Visual Odometry to Estimate Position without GPS??
Page 1 of 1
|
| Author |
Message |
| Guest |
 Posted: Tue Oct 31, 2006 5:13 pm |
|
|
|
|
Hi all,
Just wondering if anyone knows of alternatives to getting position
information without GPS? I have an indoor helicopter which I have
autonomously hovering, but it starts to drift because of drafts or the
IMU is not perfectly positioned (i.e. it's not exactly parallel to the
ground). Double integrating accelerometer data is prone to error
accumulation so that's not an option.
The only thing I can think of doing (excluding indoor GPS) is to point
a camera down towards the ground and use some sort of visual odometry
algorithm. If anyone else has any other ideas, I'd appreciate it.
If someone thinks visual odometry will work, I'd appreciate any advice
on the fastest way to successful implementation.
Thanks in advance!
-weg22 |
|
|
| Back to top |
|
| Dave Martindale |
| Posted: Wed Nov 01, 2006 8:49 am |
|
|
|
Guest
|
weg22@drexel.edu writes:
Quote: Just wondering if anyone knows of alternatives to getting position
information without GPS? I have an indoor helicopter which I have
autonomously hovering, but it starts to drift because of drafts or the
IMU is not perfectly positioned (i.e. it's not exactly parallel to the
ground). Double integrating accelerometer data is prone to error
accumulation so that's not an option.
I once read of an engineering student project that used GPS-like
algorithms at audio frequencies (I think ultrasonic). You place a
bunch of synchronized sound emitters around the measurement area, then
place a microphone on the object whose position you want to measure.
Time differences in signal arrival times get resolved into position.
The speed of sound in air is so much less than the speed of light that
you can get quite good positioning precision with audio frequency. The
system is also simplified because you can place the emitters in
stationary locations around the measuring area, rather than having them
moving constantly.
Dave |
|
|
| Back to top |
|
| ImageAnalyst |
| Posted: Wed Nov 01, 2006 2:09 pm |
|
|
|
Guest
|
Weg22:
There is a field of study in image processing called Dynamic Scene
Analysis (do a web search) that looks for changes from frame to frame
in a video sequence. If you could get a camera over or under your
helicopter, you could detect its centroid and the movement of the
centroid from some reference (say, the center of field of view or the
frame before), then you could send some sort of correction signal to
the helicopter to fly it back to the reference position. Interesting
project! Is it just some sort of fun or hobbyist or classroom project
kind of thing or is there a more serious real world application of this
that you can share?
ImageAnalyst
weg22@drexel.edu wrote:
Quote: Hi all,
Just wondering if anyone knows of alternatives to getting position
information without GPS? I have an indoor helicopter which I have
autonomously hovering, but it starts to drift because of drafts or the
IMU is not perfectly positioned (i.e. it's not exactly parallel to the
ground). Double integrating accelerometer data is prone to error
accumulation so that's not an option.
The only thing I can think of doing (excluding indoor GPS) is to point
a camera down towards the ground and use some sort of visual odometry
algorithm. If anyone else has any other ideas, I'd appreciate it.
If someone thinks visual odometry will work, I'd appreciate any advice
on the fastest way to successful implementation.
Thanks in advance!
-weg22 |
|
|
| Back to top |
|
| bh |
| Posted: Fri Dec 01, 2006 9:20 pm |
|
|
|
Guest
|
<weg22@drexel.edu> wrote in message
news:1162329222.503847.13640@h48g2000cwc.googlegroups.com...
Quote: Hi all,
Just wondering if anyone knows of alternatives to getting position
information without GPS? I have an indoor helicopter which I have
autonomously hovering, but it starts to drift because of drafts or the
IMU is not perfectly positioned (i.e. it's not exactly parallel to the
ground). Double integrating accelerometer data is prone to error
accumulation so that's not an option.
The only thing I can think of doing (excluding indoor GPS) is to point
a camera down towards the ground and use some sort of visual odometry
algorithm. If anyone else has any other ideas, I'd appreciate it.
If someone thinks visual odometry will work, I'd appreciate any advice
on the fastest way to successful implementation.
Thanks in advance!
-weg22
try this:
http://www.techreview.com/InfoTech/17841/ |
|
|
| Back to top |
|
| serg271 |
| Posted: Sat Dec 02, 2006 2:11 am |
|
|
|
Guest
|
weg22@drexel.edu wrote:
Quote: Hi all,
Just wondering if anyone knows of alternatives to getting position
information without GPS? I have an indoor helicopter which I have
autonomously hovering, but it starts to drift because of drafts or the
IMU is not perfectly positioned (i.e. it's not exactly parallel to the
ground). Double integrating accelerometer data is prone to error
accumulation so that's not an option.
The only thing I can think of doing (excluding indoor GPS) is to point
a camera down towards the ground and use some sort of visual odometry
algorithm. If anyone else has any other ideas, I'd appreciate it.
If someone thinks visual odometry will work, I'd appreciate any advice
on the fastest way to successful implementation.
It all depend on the amount computing power you can allocate ot the
task. If you are not restricted to weak on-board CPU and have for
example WiFi link to desktop PC you can make complete 3d reconstruction
of the scene from camera sequence and get very precise 3d position and
orientation of your helo relatively to ground. Make feature extraction
and matching for each frame and solve epipolar matrix for example. If
you have landscape with distinctive features in each frame (_not_ a
clean white floor) and frame overlaps more than 50% (that is camera is
not extremely slow) that will work for sure with enough computing
power. Precise algorithms for feature extraction and 3d reconstruction
should be tailored to your task - depending on the type of the camera,
landscape, movement and amount of computing power. You can start with
something ready-made for start, check OpenCV for example. |
|
|
| Back to top |
|
| |
|
Page 1 of 1
All times are GMT - 5 Hours
The time now is Tue Oct 07, 2008 4:50 am
|
|