Comparing photogrammetry to laser scanning

Discuss all Autodesk ReCap Studio related issues here.
Post Reply
Jamesrye
V.I.P Member
V.I.P Member
Posts: 369
Joined: Mon Aug 11, 2008 4:13 pm
Full Name: James Rye
Company Details: Merrett Survey Partnership
Company Position Title: Spatial Analyst

Comparing photogrammetry to laser scanning

Post by Jamesrye » Thu Mar 08, 2018 11:50 am

I have carried out some testing where I look at the point cloud produced by RECAP using photogrammetry and compare it to the laser scanned point cloud. I returned to a sea wall that I had surveyed using a laser scanner and took photographs that I then processed in RECAP and also PIX4D. I thought that I would share here my initial results.

I used a SONY A6000 25 megapixel camera and a 19mm lens, I also experimented with a 30mm lens. The A6000 is a crop sensor and RECAP and PIX4D gave best results with the wider angle lens.

I took pictures in different ways, as a fan of images every 5m - looking at the sea wall (which was less than 10m from the camera), and also just a 'square on' photo every meter. Pix4D preferred the fan method and RECAP processed better with the second approach, both programs gave different outputs if the pictures weren't numbered sequentially. The PIX4D program stitched more of the images automatically and allowed for more user intervention. PIX4D costs significantly more, but produced a more detailed mesh with many more vertices and also less 'rounding' on steps etc...

What follows is a comparison of laser scanning to the mesh produced by RECAP Photo.

RESOLUTION

The first thing to note is that the point cloud density is much lower in the photogrammetry than in the laser scan data. We took a patch of wall approx 1m square, 10527 points in the scan data and 468 points in the photogrammetry point cloud. Just 4% of the points

DETAIL

It can be seen that surface detail in the laser scanned mesh is better represented than in the photogrammetric mesh, though image textures are far superior in the photogrammetry. Mesh edges are also sharper in the laser scanned mesh.
image1.png
The image below shows the photogrammetric mesh with and without textures. The high resolution imagery adds a lot to the model - especially for visual inspection surveys.
image2.png
NOISE

In some ways the photogrammetry is better than the laser scanning here because passing people would need to be present in four photographs, whereas they are likely to be moving through the scene and so wont appear in the photogrammetric data.

Although scanning has 'noise' where there is a high angle of incidence, the photogrammetry has artefacts where images do not capture the entire structure. By contrast, these areas would simply have no data in the laser scanned data set. The photogrammetry creates 'false information' - larger triangles on the periphery of the data.

ACCURACY TEST - accuracy measures the closeness of a measured value to a known value - in this case the photogrammetric point cloud is compared to the mesh from the laser scan.

The image shows that the points on the wall are within 2cm of the laser scanned surface. You can see some areas where the photogrammetric point cloud is not very accurate - on the steps. This is where the photos haven't been taken to capture the steps surface (due to RECAP preferring the photos taken square on to the surface).

It is also possible to see that there is an issue with the top of the wall and also with the ground surface that is quite oblique to the camera.
image3.png
I am interested to know if other users are combining scan data and photogrammetry using RECAP. I will be carrying out some testing with the Reality Capture application shortly.
You do not have the required permissions to view the files attached to this post.

User avatar
stevenramsey
Global Moderator
Global Moderator
Posts: 1602
Joined: Sun Aug 12, 2007 9:22 pm
Full Name: Steven Ramsey
Company Details: Leica Geosystems
Company Position Title: Technical Specialist Scanning EMEA
Country: UK
Skype Name: steven.ramsey
Linkedin Profile: Yes
Location: London
Contact:

Re: Comparing photogrammetry to laser scanning

Post by stevenramsey » Thu Mar 08, 2018 3:42 pm

James kindly sent the images over so here is a few pointers. Not about the software but the capture mainly.

Not enough images. for the square on ones there would need to be 4-5 as many images with approximately a 70-80% overlap, they were about 50%.
Next you need overlap in multiple directions, so at least three runs along the wall low, mid height and tall all looking at the whole object.
Fill the scene with the object no need for sky.
Never do a fan as there is not enough parallax to solve a good resection, doing a separate run at an angle to the object no more than 45deg essentially a fan with no overlap (that will start some discussions I'm sure).
Think of every pixel in the image needs to be seen by 3 images to create a good point.
Shoot in Raw and post process (light room/DXO for example).
Get a Color chart and color balance the images.
A CPL helps reduce any shine on the object (not an issue here) it looses one stop so may need to adjust the exposure.
shoot in manual or AV never auto and keep the DOF reasonably deep F8 min
Use a tripod and remote trigger, stops any movement and means you don't get a cricked neck.
Steven Ramsey

Home steven.ramsey@laserscanningforum.com
Work steven.ramsey@leica-geosystems.com
Mobile +44 7766 310 915

Jamesrye
V.I.P Member
V.I.P Member
Posts: 369
Joined: Mon Aug 11, 2008 4:13 pm
Full Name: James Rye
Company Details: Merrett Survey Partnership
Company Position Title: Spatial Analyst

Re: Comparing photogrammetry to laser scanning

Post by Jamesrye » Thu Mar 08, 2018 5:12 pm

Thank you for the tips Steven :)

When I have some free time :lol: I will see about returning to site and repeating the test.
shoot in manual or AV never auto and keep the DOF reasonably deep F8 min
I shot the images using various parameters:
2. One set in aperture priority set to F8
3. One set in aperture priority set to F16.

I didn't send you all of the images that we shot, but reading your comments, I think we were getting too much sky and even with the extra ones, we wont get 70% overlap.

The camera has an HDR option, so it might be interesting to try with that as well.

Thanks again,
James

Post Reply

Return to “Autodesk ReCap Studio”