12 replies to this topic

[pujari]

I am new here on the forum & looking for recommendation of software which can do aerial image stitching with geo referencing along with it, I tried Map Cruncher but found that it needs all (31000) images anchored & it does not stitch into 1 large file instead it tiles them correctly & display them in web pages whereas I need 1 large image from the 31000 images geo referenced.

These images are extracted jpeg frames (total 31000 frames) from video footage.

I have the following question if any experienced cartography expert can answer them.
1. Is there a software which can do image stitching with few anchored images lets say I geo reference(anchor)1000 images out of 31000 frames, will all the images stitch around the anchored images correctly?

2.When I stitch the images rows the straight flight path appears curved, I don't know why the conventional stitching program are making it curved, Any reason for that?  2011_08_24_175630.jpg   16.29K   41 downloads

I hope you can advice me the right software to do such large stitch & geo reference it correctly

[David Medeiros]

Not exactly sure of its functionality but have a look at Geographic Imager by Avenza. It's a GIS plugin for Adobe Photo Shop. http://www.avenza.co...ographic-imager

Are your images ortho rectified but without a reference file, or simply raw uncorrected images from the camera? Was the video camera used purpose built and set up to take ortho images, or a standard consumer camera?

The curve in your image path likely has to do with these pictures not being ortho rectified which means they retain all of the distortion that comes with taking a picture through a lens or at an angle to the ground. Stitching un-rectified images together can result in a straight line path curving or wiggling around in layout. Because of this I suspect you will need to rectify and reference all 31000 images if you want to use them together. But I'm not a remote sensing expert. Hopefully someone with more info will come along to help.

[CGIS]

Well - First, I'd take issue with your number of images to stitch. 31k is insane. If you're extracting those from 29.97fps video, your frame-to-frame overlap would be huge. Especially when shot from the air.

Please post two of the frames (sequential, ie. frames 5001 & 5002).

Second, you're after two things that should be handled seperately; the stiching, and the warp/georeferencing. In order to understand your situation better:

>Why do you need one large image?
>What do you expect the pixel dimensions of this image to be? If it's one continuous flight path in a straightline, you're looking at something like 50000x 480? If that's that's the case, what's wrong with using say, 4 images at 12500x 480?
>What is your frame overlap? If it's more than 40%, I'd drop frames until I reach the 40% marker (or whatever is best for your sticher). More isn't always better.

[pujari]

Thanks for the replies it helps to know some of the answers which I was seeking.

Please find more details as you asked for.
The company took this video footage with a custom build thermal camera whose technical parameters & lens details are not known to me, the pilot flew over the area making number of passes over the landscape & in the process I feel the camera must have changed its angle hence the stitched images make a swayed flight path although the GPS track shows straight flight path.


>Why do you need one large image?
The image needs to be printed at 200-300 ppi

>What is your frame overlap?
almost 80% -90% but I have a choice of dropping frames (every 5th or 10th frame) the trouble is at ground there are many areas with too less features so they wont stitch at all in such case, I have tried that. Most stitching apps seek 20-40% overlap else you manually have to use control points if the overlap is less, I cant use manual method since in many areas ground shows repeating pattern of fields/trees/plain ground.

>two things that should be handled seperately; the stiching, and the warp/georeferencing.
If I stitch each column of images & then georeference it, I get different wrapping of each column, that is indicating some issue may have created a problem while the camera was in flight.

>Are your images ortho rectified but without a reference file,
No they are not ortho rectified, I tried Map cruncher for georeferencing & it corrected the geometrical distortion but it does not give me output as single images back so I cant stitch them together.

I will post 2 images in just a while from now.

[pujari]

See attached zip file have sample images

1st set of 2 images have feature

2nd set of 2 images have repetitive features almost impossible to use control points & hard to geo reference too unless stitched with other images.
 carto_sample.zip   58.5K   24 downloads

Attached Files

•  carto_sample.zip   58.5K   6 downloads

[Jacques Gélinas]

Not sure, but in your case, I would consider doing a rubbing sheeting on the stitched output image using a good distribution of control points along the path of the image.

As David mentioned you could look into Avenza's Geographic Imager or Blue Marble Geographics Transformer software.

Regards,

[pujari]

Not sure, but in your case, I would consider doing a rubbing sheeting on the stitched output image using a good distribution of control points along the path of the image.

Thanks for the reply,
I tried rubber sheeting on the stitched images but each columns have been curved differently while stitching as stitch apps don't have any clue how to undistort these columns, some columns are straight some are bent outwards which is then making the rubber sheeting among each column not match with next column hence the final image remains.

Before I try other software mentioned I would like to know whats the best method to do such aerial stitch?
ortho rectify first all the 31000 images then stitch them
or
stitch them column by column then ortho rectify using the georeferencing

Second method is what I have used as of now but stitcher apps have no way or method to geo reference it while stitching. If I straighten the columns with a guided straight line then the columns wont match as seen in the attached image.

Attached Files

•  2011_08_25_181518.jpg   211.34K   24 downloads

[Jacques Gélinas]

Sorry but if you don't have camera parameters you won't be able to create any ortho photo or an ortho photo mosaic.

Anyway, sorry but I can't help you. I would suggest you look up photogrammetry software on Google and read up on the subject.

[pujari]

Sorry but if you don't have camera parameters you won't be able to create any ortho photo or an ortho photo mosaic.

That is exactly the problem I am facing due to camera parameters missing since this is custom built camera.

[CGIS]

Sorry but if you don't have camera parameters you won't be able to create any ortho photo or an ortho photo mosaic.

That is exactly the problem I am facing due to camera parameters missing since this is custom built camera.

Since we can't help you on this, here's a place to go next:

http://www.erdas.com...c_Pro.sflb.ashx

http://www.erdas.com...nc/Details.aspx

[pujari]

Since we can't help you on this, here's a place to go next:

http://www.erdas.com...c_Pro.sflb.ashx

http://www.erdas.com...nc/Details.aspx

Thanks
these links will be quite helpful.
I must say I am happy to find such a helpful community here at cartotalk.

[Kalai Selvan]

Pujari,

You might end up merging all your images remember; but loading time will be increased too much, so divide them to 8-10 nos..hope that will help..Erdas is a good joining tool..

[Lui]

Well six months ago I did an thermo orthophoto from several 1000's images. But I had a LIDAR data, trajectory, GCPs (Ground Control Points). From this data I've calibrated camera (both principal points, focus lenght and radial distorsion). To produce a quality orthophoto you'll need a good terrain representation especially using thermo camera. Thermal cameras have lower resolution with less detail present in image. That doesn't allow automatic terrain creation from image pairs. That also mean that tie point collection (identical points in overlapped images) is critical at best scenario. I didn't perform an automatic tie point collection, I've just used exterior parameters from trajectory. Well trajectory was quite good using professional IMU and GPS with adequate postprocessing.

Just my 0.02$