Stitching over Jungle Canopy

[t_harris]

Hello Forum:

We're working to stitch imagery acquired with a modified GoPro over jungle canopy, and are encountering lots of artifacts in orthomosaic.

Geotagging of photos looks good, and we have plenty of overlap between image frames.  See the attached image showing image locations overlaid in Google Earth.

Wondering if the issue was simply that the polygons needed more points to properly display the data, we tried running a dense point cloud, and then used that to generate the mesh.

Here's what it looks like with the lowest quality dense point cloud:
http://screencast.com/t/D6wH3NXiCGqd

We can stitch an orthomosaic, but the ortho has severe artefacts due to the severe artifacts in the dense point cloud, visible in the video.

Has anyone solved similar problems?

[Alexey Pasumansky]

Hello t_harris,

More informative will be a screenshot with the sparse cloud, bounding box and camera positions. Now it looks like there's almost no side overlap.

[t_harris]

Thanks for your help!

Here's some screen shots.

[t_harris]

We're flying at 100m.  200m.
This is a GoPro 3+Silver with a converted 4.14mm lens.  The focal length = 23mm in 35mm equivalent.
We don't have any flights using a camera with a mechanical shutter.
Over a variety of other surfaces, we haven't seen any issues stitching high quality orthomosaics with this setup.

Thanks!

[Alexey Pasumansky]

Hello t_harris,

It looks like the camera alignment is incorrect (assuming that the flight track should lay in the same plane), so the next processing steps are already based on the incorrect data.
Can you provide the original image set and the project file for our internal tests?

[t_harris]

We can provide the data for sure.  Can you point me to an upload area?

[Alexey Pasumansky]

Hello t_harris,

Sent you the details in PM.

[t_harris]

Thanks!  I uploaded the photo set to your FTP.

Regards and thank you for your help.

[Alexey Pasumansky]

Hello t_harris,

It looks like the problem is in the insufficient side overlap between the flight lines. The camera is not looking vertically down, so on the neighboring lines the trees in the overlap area are seen from different sides, and this fact makes the matching point detection very difficult.