Can pix4d process historic imagery captured with a film camera?

PIX4Dmapper PIX4Dmapper Questions/Troubleshooting
1

I have several historic imagery datasets that were captured a few decades ago with a film camera. The photos all have fiducial marks and have been metrically scanned.

I want to know if its possible to process these photos using pix4d. Specifically, does Pix4d have an option to measure fiducial marks and perform interior orientation?

Has anyone else managed to process historic datasets?

Thanks,
Laura

2

Laura, I have not done any history data sets, but I would like to help you.
By having the targets or marks and known dimension you can set scale to your project. There is a new version coming up soon (from the latest pix4d webinar) that will make this process very easy.

3

Hi Laura,
Pix4Dmapper is not designed to process historical scanned images. However, we did manage to process a few projects with analogue images and they gave good results.

It could work if you:

  1. Scan the images by aligning the fiducials.
  2. Crop the images to remove the fiducials. Make sure that the resulting images have the same width and height or you will obtain a different camera model for each image in Pix4Dmapper.
  3. Create a new camera model in Pix4Dmapper and find the optimized parameters. You can find more information about the camera model options here: https://support.pix4d.com/hc/en-us/articles/202560169

Regards,

4

This is quite an old topic, however we had great success with it this year. The short answer is YES! It does require a fair bit of control though, and often you can use google earth for common locations and elevations that haven’t changed. We have produced orthomosaic outputs for thousands of square miles using Pix4D. Large film scans (9") scanned at high resolution can often be much more than 55 megapixels, so the large-frame camera add-on is necessary.

5

Hi Nathan, could you please tell me a little bit more about the parameters you used in Pix 4D? How many control points per image, on average? What was your RMSE? Also, did you need to crop every single image in advance of stitching? I am not sure if I will be able to do that since I only have digital scans (not the paper originals), and only some, not all, of the scans have the fiducials/black border.

6

Any luck with this?
I am also looking for these parameters.
Thanks,
Nicolas

7

Hi nicolas,
PIX4Dmapper can definitely process archival imagery. I recently created a project of Vail Ski Resort where I downloaded eight image frames from from 1955 at earthexplorer.usgs.gov.

vail
vail1862×1452 465 KB

As you can see the project worked fairly well. You will want to also download the camera parameters so that you can enter them into the camera properties editor. I used 10 GCPs all created in Google Earth. Three are the minimum but with such a large project I opted for 10. The GSD was 55" and the RMSE was 12’. All in all, I believe it processed very well.

2 Likes
8

Hey Laura can you share some film image for mirror stereoscope on sparshshekhar4077@gmail.com

9

Can you share the workflow you followed?

10

Hi Demetres,
I am happy to share the workflow. At least I can hit upon some of the more essential aspects. First, it is very important to pre-process the imagery. Your scanned imagery will likely have a dark border around all the images, like the screenshot below.

image
image1920×1872 402 KB

It is very important to clip out only the image itself. Otherwise, the border will leave artifacts and can cause calibration issues. I used ArcGIS Pro and was able to clip out many images using batch mode.

After all the images are clipped, you will need to create what I call ad hoc GCPs. Basically, you will need to use some modern imagery and identify features that exist in the old imagery that exists today. Using ArcGIS Pro, I created points on these features and extracted the coordinates. You will also need elevation data. For the sake of ease, I extracted surface information from SRTM elevation data. You can probably use better elevation data but SRTM worked pretty well.

After you clip the images and mark your GCPs, you are ready to process. These images tend to be extremely large (approximately 90mp in my case). For step 1, I used 1/2 image scale, with Geometrically Verified Matching, and All Prior for the internal parameters optimization. These three processing options worked best for me.

Let me know if you have any questions.

11

Thank you very much for sharing! I will try to make it work and get back to you!!!

12

We have seen your post regarding your workflow for processing historical imagery. We have some further questions as our project has some extra challenges. We also have some historical imagery, however these photos are from 1910, meaning they are:
Taken from a ground based camera
There are gaps between the sequence of images
Not intended to be used for stereoscopic analysis

Our objective is to create a 3D model using these images. The photos are masked already.
What would the workflow look like to manually create tie points so Pix4D can stitch the images together?

13

Hi claplante,

The biggest problem is going to be the gaps between the images. For this process to work, you will need good overlap between images. The minimum overlap should be 60% but it will work much better if you have between 70-80%.

Assuming you have good overlap, when you go to process these images the orientation of the model will likely be tilted, skewed, or even inverted. This is because there is no orientation data contained in the Exif. What you will need to do is add an orientation constraint to the Z axis. This will force Mapper to straighten the model. If this is a building, then it would be quite easy to add one along a corner, door, or window.