Comments

12 comments

  • Avatar
    Ondrej Trhan CR

    Hi RWiley-cid, there are options for using video for model creation. But not sure if someone tried that specific camera. Do you have a footage from a car or did you hold the camera in your hands?

    What were your import settings?

    What I saw, that camera have kind f a big distortion, so there could be issues also from that.

    1
    Comment actions Permalink
  • Avatar
    RWiley-cid

    I believe the camera was mounted to the windshield, which also captured part of the hood for the entire duration of the video. I initially imported close to every single frame available. It does have distortion. I can't seem to find any specific info about the sensor or focal length for the camera itself (Garmin 56) and the video doesn't store any of that info. It records GPS data every second on the video frames (I OCR extracted the numbers), so I'm going to try and insert those values into the extracted frames (though it's 60 images all w/ the same "coordinates" even thought they really don't).

    0
    Comment actions Permalink
  • Avatar
    Ondrej Trhan CR

    If the part of hood is on each capture it could be confusing for the application during alignment. You can consider using masks in that case.

    1
    Comment actions Permalink
  • Avatar
    RWiley-cid

    I also tried masking both the hood and sky using image transparency, but it still couldn't resolve the alignment. I'm going to try and spoof the gps into the images and see if that helps.

    0
    Comment actions Permalink
  • Avatar
    Ondrej Trhan CR

    Is it possible to share your video here?

    1
    Comment actions Permalink
  • Avatar
    RWiley-cid

    Sorry it took me so long to get back to you. I've been really busy with other projects. Here is the video: https://drive.google.com/file/d/1hpw-kKLOk4TPxXwLqXiarN2yh9wkB-K_/view it is still processing when I post this, so you may have to download it to view it.

    0
    Comment actions Permalink
  • Avatar
    Ondrej Trhan CR

    Hi, I chceked the video and the issue here could be, that there is quite a lot of movable objects during recording. This is crating still new environment and it is hard to find the same features over images for alignment.
    For alignment you will need empty roads, masked images and clean front car's window. Also, the text information are not ideal for this.

    I tried import the video with resolution 0.2 sec and I got some results (not ideal). So basically the masked images could work in this case, but you need also follow other rules. 

    1
    Comment actions Permalink
  • Avatar
    RWiley-cid

    I generated some image masks as well and just tried a draft alignment. I also took the gps coords written on the images and linearly interpolated the values since they only update every second, then embedded those in the exported frames. Without accurate elevation values, I adjusted the Z accuracy to 1000 meters in the camera prior settings hoping it would basically allow it to ignore the gps z value. It still gave me odd results:

    0
    Comment actions Permalink
  • Avatar
    Ondrej Trhan CR

    As I wrote, this is not an ideal dataset. To get better results you need to have empty road and cross that way in more lines, forward and backward. If you have the GPS information you can use them, but as it is video, it is possible that those info won't be written into the frames. Also, you can try increase the weight of those coordinates.

     

    1
    Comment actions Permalink
  • Avatar
    RWiley-cid

    That's what I was assuming about this video, but I had to at least try to get it to align. I'm not actually the person that's going to capture these drive-through videos, so I need to try and convince the people who are how to do it and make it as simple as possible while also still capturing usable footage for photogrammetry. We likely will never be able to do drive-throughs without other vehicles on the road (that just doesn't exist in daytime California).

    I'm thinking that if we hood mount a slightly down-angled gopro 10 (with gps enabled) that can see as much of the road as possible and having another camera looks slightly down on the rear. We could also shoot at 120fps4k to lower the amount of motion-blur. Add in multiple drive-throughs in different lanes. Could all of that give us better results?

    0
    Comment actions Permalink
  • Avatar
    RWiley-cid

    Would using a Stereo Camera setup help?

    0
    Comment actions Permalink
  • Avatar
    Ondrej Trhan CR

    Yes, the mentioned setups could bring better results. 

    1
    Comment actions Permalink

Please sign in to leave a comment.