Forum

Author Topic: Quality measurement of point cloud  (Read 5205 times)

Ingsayyad

  • Jr. Member
  • **
  • Posts: 89
    • View Profile
Quality measurement of point cloud
« on: January 06, 2015, 02:21:12 PM »
I just wonder if there is some quality measures of the dense point cloud generated by images in photoscan.
How can I know the quality of single point in dense cloud.

If my images are max at 10 m and I 'm using middle, high, ultra high setting for dense cloud, what approximately pixel accuracy should I have.
« Last Edit: January 07, 2015, 12:03:25 PM by Ingsayyad »

nadar

  • Full Member
  • ***
  • Posts: 130
    • View Profile
Re: Quality measurement of point cloud
« Reply #1 on: January 08, 2015, 01:01:35 PM »
a practical solution is to visualize a "slice" crossing a planar surface such as a concrete wall.
On a perfect cloud, all points should be co-planar, and hence generate a "1-dot thick" cloud.
As the model is not perfect, the plane will be represented by a more or less thick cloud. You can measure the width of the cloud. I usually divide this measure by the distance between the camera and the wall, assuming the noise will be proportional to the distance.
My first findings are that noise is related to
1) the quality of images (type and quality of the camera, but also sharpness of the images, exposure settinge, etc...). Simply said: don't expect good clouds from poor images.
2) the overlap between images. Apparently, the more images I use, the better results I got

Of course, a more rigourous approach to assess point cloud quality should be to measure the 3D coordinates of a large number of individual points, to compare these to actual positions (measured with a d-GPS) and express it as a RMSE. So far, I didn't find the time (and patience) to do this.
Please keep us informed about your results.
It's easy to get spectacular results. Having accurate models is another story...
 

Ingsayyad

  • Jr. Member
  • **
  • Posts: 89
    • View Profile
Re: Quality measurement of point cloud
« Reply #2 on: January 08, 2015, 04:03:40 PM »
@ nadar
Thank you for your detailed reply.
About your suggestions:
1-It was also my original plan to do so.   I don't have perfect plane in my object. So I  have not applied it.
2- It is standard solution in classical photogrammetry. I have taken 20 markers in my project. 10 are fixed (active) and the rest
are used to see RMS. The results are about 5mm.
 

nadar

  • Full Member
  • ***
  • Posts: 130
    • View Profile
Re: Quality measurement of point cloud
« Reply #3 on: January 08, 2015, 05:29:57 PM »
5 mm for photos acquired from 10m (1/2000 of the distance) is roughly in par of what I get on my aerial photos acquired at 200-500m

Ingsayyad

  • Jr. Member
  • **
  • Posts: 89
    • View Profile
Re: Quality measurement of point cloud
« Reply #4 on: January 08, 2015, 05:59:36 PM »
@ nadar
What is your accuracy??????? A litle bit detail please?
What should be expected, 1/5000 of the distance?
« Last Edit: January 08, 2015, 06:40:06 PM by Ingsayyad »

nadar

  • Full Member
  • ***
  • Posts: 130
    • View Profile
Re: Quality measurement of point cloud
« Reply #5 on: January 08, 2015, 08:12:51 PM »
On photos aquired on an urban site from an helicopter flying about 500 m above ground, on vertical photos, I have a "point cloud fuziness" of about 25 cm on flat surfaces. I have more or less the same ratio (0.25m / 500 = 1/2000) on other projects, but as I mentionned earlier, it seems to be very dependent to the image quality and overlap.