Support Website Contact Support Blog

How to get best 3D point cloud reconstruction quality using multispectral RedEdge-M images?

Hi,

I am trying to reconstruct 3D point cloud from images captured via 2D multispectral RedEdge-M. It generates separate images per band, 5 bands in total.

Importing the images worked great. Pix4d seemed to process the images correctly.

The following 2 pictures show some results:

I configured the following settings regarding point cloud reconstruction. Basically, I wanted to have the best possible result. Processing time was not an issue.

I have a few questions:

1.) I am particularly interested in the best possible reconstruction of the electrical towers. On the first 2 images you can see two of them. Taking this into consideration, did I configure Pix4d correctly or is there some configuration to get a better result.

2.) Why is the resulting point cloud grayscale? I expected it to be colored, since all relevant color information is contained in the 5 bands. 

Just for comparison. The same UAV which captured the multispectral images was also equipped with a Sony UMC-R10C taking RGB images (much better resolution). I uploaded them to Pix4d cloud and got the following result:

Is a similar result in terms of texture possible with Pix4d Mapper too?

Thanks a lot.

Steffen

Hi Steffen,

Pix4Dmapper desktop and Pix4Dmapper cloud will produce the same results. I will now answer to your specific questions:

1.) I am particularly interested in the best possible reconstruction of the electrical towers. On the first 2 images, you can see two of them. Taking this into consideration, did I configure Pix4d correctly or is there some configuration to get a better result.

You can change your minimum number of matches to 3. That will generate less noise. https://support.pix4d.com/hc/en-us/articles/202557799

2.) Why is the resulting point cloud grayscale? I expected it to be colored since all relevant color information is contained in the 5 bands. 

The point cloud is generated separately for each band. You have generated the point cloud of the green band which is the default option (the rig is set to optimize relative rotation with a subset of secondaries with the reference camera as 1 which is the green band). If you go to image properties editor and change the processing mode in rig to optimize relative rotation, the point cloud of all bands will be generated but it will still be grayscale (but 5 point cloud models). You can read about rigs here: https://support.pix4d.com/hc/en-us/articles/115005852003-Processing-camera-rigs

 

Just for comparison. The same UAV which captured the multispectral images was also equipped with a Sony UMC-R10C taking RGB images (much better resolution). I uploaded them to Pix4d cloud and got the following result.

Since RGB images have better resolution, the point cloud generated will be better and will also have the color information (will be RGB) but for multispectral images, the pointcloud will be grayscale.

Thanks for the feedback.

> Pix4Dmapper desktop and Pix4Dmapper cloud will produce the same results.

Regarding Multispektral Images: Pix4Dmapper desktop results are much better than Pix4Dmapper cloud results. Probably Pix4Dmapper cloud processing in different mode (faster, less quaility) than my local Pix4Dmapper desktop run.

Regarding RGB Images: Pix4Dmapper cloud much better (last image i posted in initial post by Pix4Dmapper cloud).

1.) Changed minimum number of matches to 3. Yes, it reduces noise, but it doesn’t improve reconstruction quality of the electrical towers (gray objects you see on the field). Actually, noise is not a big issue.

2.) I changed processing mode in rig to optimize relative rotation. Resulting quality is more or less the same.

Any other ideas to get better results?

Best

Hi Matthias,

If you upload the project in cloud from Pix4D desktop, the project will be processed with same processing options that are selected (https://support.pix4d.com/hc/en-us/articles/202558589-How-to-upload-project-files-from-desktop-to-cloud). Otherwise, if only images are uploaded to cloud (https://support.pix4d.com/hc/en-us/articles/115004680703-How-to-upload-images-using-the-cloud-platform), cloud will process with the default 3D maps template. So for multispectral or RGB agriculture images, it is always better to start the project in desktop, use the processing options that will suit the best (ag multispectral or ag RGB for ag/homogeneous terrain) and then upload the project to cloud. In this case, the p4d file and the images get uploaded. Thus, the cloud will respect the processing options.

If you want a good reconstruction of the electrical tower, you will need to make a flight around the tower (orbital flight plan) and then merge it with this project. You can have a look at this for the flight acquisition plans: https://support.pix4d.com/hc/en-us/articles/202557459 and https://community.pix4d.com/t/4550-merge-nadir-image-with-45-image- for merging

2.) I changed processing mode in rig to optimize relative rotation. The resulting quality is more or less the same.

This will not change the quality of the output map so much. This option is mostly used for multispectral cameras not in our database… This is only when you want the point cloud of all bands.