Discussions

Modelling buildings

Posted by Dave on July 27, 2016 at 10:12am

Hi all

Was wondering if anyone knows how to autonomously capture images for photogrammetry of buildings. The idea would be to set up a spiral flight plan around for example a tower with a common point of interest.

regards

You need to be a member of diydrones to add comments!

Join diydrones

Email me when people reply –
Follow

Replies

  • Dave August 1, 2016 at 8:31am

    Stephen thanks

    This info has helped alot. I am busy processing dense cloud now after 2h of marker matching :)



    Stephen Zidek said:

    Dave,

    In photoscan you must have the photos window up, and double click a photo to bring it up on the main viewer.  From there you can zoom in to add markers.  I usually use roof corners because they are visible from the ground and air.  Each individual marker needs to be tagged from multiple different photos in both chunks.

  • Dave August 1, 2016 at 8:30am

    Thanks Mario

    Very cool model

    MarioSpeedwagon said:

    As stated before, Tower can do this.

    This was done with Tower, 3dr solo and a stock GP4.

    https://skfb.ly/LErs
  • Dave August 1, 2016 at 8:29am

    Thanks Will

    Im onto it :)

    Will Fellmeth said:

    Dave,

    Yea the ordering of these messages is getting pretty messy rather quickly.

    The MAP2 from skysight/seegul does work with the A5000 as well as the samsung NX1100. Its all on their website (http://skysight.eu/). We have had great success with produces from skysight/seegul. When you go to place the order for the MAP2, it asks you for the manufacturer and model of the camera your using and automatically sets you up with the other cables and such that you'll need.

    Cheers,

    Will F 

  • T3
    Stephen Zidek August 1, 2016 at 6:26am

    Dave,

    In photoscan you must have the photos window up, and double click a photo to bring it up on the main viewer.  From there you can zoom in to add markers.  I usually use roof corners because they are visible from the ground and air.  Each individual marker needs to be tagged from multiple different photos in both chunks.

  • Moderator
    MarioSpeedwagon July 31, 2016 at 10:21am
    As stated before, Tower can do this.

    This was done with Tower, 3dr solo and a stock GP4.

    https://skfb.ly/LErs
  • Will Fellmeth July 30, 2016 at 10:11pm

    Dave,

    Yea the ordering of these messages is getting pretty messy rather quickly.

    The MAP2 from skysight/seegul does work with the A5000 as well as the samsung NX1100. Its all on their website (http://skysight.eu/). We have had great success with produces from skysight/seegul. When you go to place the order for the MAP2, it asks you for the manufacturer and model of the camera your using and automatically sets you up with the other cables and such that you'll need.

    Cheers,

    Will F 

    "skysight.eu"
  • Dave July 29, 2016 at 12:57pm

    Hi Will

    The ordering of these messages gets confusing.


    Thanks once again for your feedback. I have a correction to make though. Our camera is a Sony A5000 not A7.

    Our other camera is a Samsung NX1100 which triggers perfectly from a APM 2.6 but continuously triggers from a Pixhawk.

    I need to get a cable or two which can trigger these cameras.

    Do you know whether the previous advice you gave me on the trigger cable would still be relevant?

     

    Best regards

    Will Fellmeth said:

    Dave,

    So, for the camera trigger cable we use a SkySight Mono (shown in the picture i attached), it goes from the AUX output of the Pixhawk to the micro USB of the Sony. Apparently Skysight has gone out of business because when I go to their website I'm directed to here (http://skysight.eu/) which is a company called 'Seggul' (why their website is skysight, I'm sure I do not know). They have a similar product they call the MAP2, you just need to get the servo cables and USB cable to use it, and they have helpful tutorials on their website that shows you everything you need to know. And here's a wiki page on the Mono (http://ardupilot.org/copter/docs/common-camera-shutter-triggering-f...).

    Now for your model. We don't use photoscan for our 3D models, we use Pix4D mostly (sometimes agisoft). We only use photoscan for stitching together 2D panoramic images for things like power line inspection. So point of the story, I'm not too familiar with photoscan's process to make 3D models. But I can tell you that when their are blurred spots in our 3D models, that means there was poor overlap between images or the stitching algorithm couldn't find good tie points. So to fix this I recommend increasing your overlap between images. Our standard is 80% forward and 60% side.

    Hope this helps.

    Cheers,

    Will F


    Dave said:

    Hi Will

    Many thanks for your constructive feedback. We also have a Sony A7 and a machine which uses a Pixhawk FC. Unfortunately we do not have a trigger cable which works for this. Do you have any recommendations on where we could find this? 

    Currently we are using another setup with a Samsung NX1100 which works really nicely and I will be testing using this setup. 

    I have done a few models of houses and buildings so far but in all cases I have found certain areas to be blurred or twisted. Stephen mentions manual alignment in Photoscan?  

    I guess that Photoscan gets confused if you enter top down pictures with aerial view from around the building and ground data all in one data set?

    Looking forward to your response

    Dave



    Will Fellmeth said:

    Dave,

    There are a whole bunch of ways you could go about doing this. I personally like one of two options: using a pixhawk controlled Turbo Ace Matrix with a Sony A7 camera or using a DJI Inspire with their X5 camera. The former requires a more tech savvy and open to do some research individual but provides better images with smaller Ground Sample Distances (GSD), while the later is more lay-man friendly but does not have the image quality of the former. As for modeling software I prefer Pix4D or Agisoft. And as for the actual flights to capture the images, I would do one nadir flight over the building to capture its top and surrounding area, and then follow up with a number of oblique flights (depending on how tall this structure is and what GSD your looking for) around the structure.

    This is a (very) quick explanation of the processes my colleagues and I follow to do our photogrammetry. If you have more specific questions please don't hesitate to ask.

    But to directly answer your question, yes many people (including myself) have experience doing this kind of work.

    Cheers,

    Will F

  • Dave July 29, 2016 at 12:53pm

    Hi Stephan

    Thanks again for the advice. I am currently working on this. I dont know how you find common points from a sparse point cloud though. It seems impossible to me.

    Hope to figure it out soon

    Regards


    Stephen Zidek said:

    Dave,

    I start photoscan with two chunks.  One for the aerial photos, and one for the ground photos.  Sparse cloud on both.  From there I manually add markers to features which are visible both from the air and the ground.  I then align chunks by marker, then merge chunks.  Then I do further processing like dense cloud.

    I made this using this method.

  • Will Fellmeth July 29, 2016 at 7:49am

    Dave,

    So, for the camera trigger cable we use a SkySight Mono (shown in the picture i attached), it goes from the AUX output of the Pixhawk to the micro USB of the Sony. Apparently Skysight has gone out of business because when I go to their website I'm directed to here (http://skysight.eu/) which is a company called 'Seggul' (why their website is skysight, I'm sure I do not know). They have a similar product they call the MAP2, you just need to get the servo cables and USB cable to use it, and they have helpful tutorials on their website that shows you everything you need to know. And here's a wiki page on the Mono (http://ardupilot.org/copter/docs/common-camera-shutter-triggering-f...).

    Now for your model. We don't use photoscan for our 3D models, we use Pix4D mostly (sometimes agisoft). We only use photoscan for stitching together 2D panoramic images for things like power line inspection. So point of the story, I'm not too familiar with photoscan's process to make 3D models. But I can tell you that when their are blurred spots in our 3D models, that means there was poor overlap between images or the stitching algorithm couldn't find good tie points. So to fix this I recommend increasing your overlap between images. Our standard is 80% forward and 60% side.

    Hope this helps.

    Cheers,

    Will F


    Dave said:

    Hi Will

    Many thanks for your constructive feedback. We also have a Sony A7 and a machine which uses a Pixhawk FC. Unfortunately we do not have a trigger cable which works for this. Do you have any recommendations on where we could find this? 

    Currently we are using another setup with a Samsung NX1100 which works really nicely and I will be testing using this setup. 

    I have done a few models of houses and buildings so far but in all cases I have found certain areas to be blurred or twisted. Stephen mentions manual alignment in Photoscan?  

    I guess that Photoscan gets confused if you enter top down pictures with aerial view from around the building and ground data all in one data set?

    Looking forward to your response

    Dave



    Will Fellmeth said:

    Dave,

    There are a whole bunch of ways you could go about doing this. I personally like one of two options: using a pixhawk controlled Turbo Ace Matrix with a Sony A7 camera or using a DJI Inspire with their X5 camera. The former requires a more tech savvy and open to do some research individual but provides better images with smaller Ground Sample Distances (GSD), while the later is more lay-man friendly but does not have the image quality of the former. As for modeling software I prefer Pix4D or Agisoft. And as for the actual flights to capture the images, I would do one nadir flight over the building to capture its top and surrounding area, and then follow up with a number of oblique flights (depending on how tall this structure is and what GSD your looking for) around the structure.

    This is a (very) quick explanation of the processes my colleagues and I follow to do our photogrammetry. If you have more specific questions please don't hesitate to ask.

    But to directly answer your question, yes many people (including myself) have experience doing this kind of work.

    Cheers,

    Will F

    camera trigger cable.PNG

    "skysight.eu"
  • T3
    Stephen Zidek July 28, 2016 at 8:43am

    Dave,

    I start photoscan with two chunks.  One for the aerial photos, and one for the ground photos.  Sparse cloud on both.  From there I manually add markers to features which are visible both from the air and the ground.  I then align chunks by marker, then merge chunks.  Then I do further processing like dense cloud.

    I made this using this method.

This reply was deleted.

Activity

Neville Rodrigues liked Neville Rodrigues's profile
Jun 30
Santiago Perez liked Santiago Perez's profile
Jun 21
More…