Support Website Contact Support Blog

Masking prior Step 1

Dear Pix4D team,
Could you help me with the following:
Source data: FullHD footage (about 1500 images) taken from onboard of aircraft.
Goal: identify the trajectory of the aircraft by using Pix4D.
Issue: source images contains areas without any useful data for analysis: the clouds and parts of aircraft with sunlight reflection.
Problem: After Step 1 I saw that some calibrated Cameras are in wrong positions, but Pix4D consider them as calibrated. I want to redo Step 1 with masks for each Image. I created grayscale PNG mask images for each frame (Rotoscope mask in Blender). How I can apply them as the mask for Step 1? Or should I add alpha channel in the source images?

Additional data: I read some articles about Pix4D image annotation feature, but the problem is that “The Image Annotation feature is available only if step 1. Initial Processing has been completed.”, but I not satisfied with Step 1 result and I want to apply mask on Step 1 in order to exclude inapplicable images area from analysis.

Additional data: real aircraft tracjectory avalable from GPS for results comparison.

Kind regards,

Hi @Alexandr_Dyachenko,

Is the image you shared one from your acquisition?
Because if so the orientation is really not optimal.
We recommend this type of acquisitions:

As a general comment, we recommend to NOT modify anyhow the images before processing.

If your images do not calibrate, it is because either your acquisition is not good, either poor overlap, either wrong processing options.
Which camera have you been using?
If you share with us your quality report and *.log file we can have a look and advice you on the processing options to use.
Then if we manage to properly calibrate during the first step, we will be able to use the mask and annotations.

If I understand correctly, you want to get the position from your camera using Pix4Dmapper and hence get the trajectory of your flight.
In order to get correct computed positions, it is important to have cameras properly calibrated. To achieve this, it is mandatory to have a robust data set.

Waiting for your files,

Hi Marco,
Thanks for your answer!
Our agency investigating civil air accidents. Due to that, we use Pix4D in two ways:

  1. Creating orthomosaic and sometimes 3D model of accident site.
  2. Evaluating possibility to use Pix4D in order to extract flight path from video that was taken on board of aircraft.
    First way:
    I can say that we have very good results for the orthomosaic and 3D models. Because we control the process of image acquisition by Pix4DCapture and by using drones (nothing interferes with the drone camera view). So we are OK with the first way of how we use Pix4D with the drones.
    Second way:
    In few previous investigations, involving analysis of video taken onboard of aircraft I manage to obtain flight trajectory by Pix4D with good quality (comparing to the GPS trajectory). Therefore, I know for sure that Pix4D can be used to determine the aircraft flight trajectory.
    About your question regarding not optimal camera orientation: as we are talking about accidents – we don’t have any opportunity to adjust existing video and camera location.
    The images showed in my post – related to the test video (not accident) taken by GoPro Hero 6 camera (a good example for the development of skills that can be used in a real practice).
    Regarding first step: my initial idea was to help Pix4D to calibrate cameras property as some images areas doesn’t contains data for calibration – parts of aircraft with sunlight reflection, propeller and moving clouds. As I understand – you do not recommend modifying the images before processing but such opportunity exists? If so then how can I check it?
    Attached quality report.
    Kind regards,

test flight_report.pdf (2.6 MB)

Hi Alexandr,

Thank you for your detailed explanation.
Well if you already have a model from the area that you flew over, I would recommend to you to merge it with this flight acquisition that you did with the aircraft.

Then by adding some MTP, it will help to reconstruction the two models.

Then from a processing point of view, I would recommend you the following to improve the calibration:
Set the image scale on 1/2
(I would try with use Geomatrically Verified Matching)
Internal Parameters Optimization: All Prior
Rematch: Yes

This should help to calibrate but I do not guarantee that it will work as your project seems that lack of matching:

Good luck and best,