How does the classification tool classify each point cloud class individually?

[Leo_Deleon]

Hello,

I would like some detail in regards to how the Pix4D classification process identifies and sorts features into their respective point cloud class. For example, does the classification process use only color to identify brush and grass for the High vegetation class? Does it only use height to separate perceived building features from the rest of the point cloud for the building class? Finally, does it use flat, asphalt colored features at ground level in order to identify road surface features? I read that color and geometry does help the classification process figure out what class point features will be classified as, but is there something else being done behind the scenes to help create the point features classes?

Thanks,

[Blaz]

Hi Leo,

To give more insights, both color and geometry are used when assigning points to the classes. Depending on the class, the color might have a higher weight. Keep in mind that the point classification is done based on the machine learning algorithms that were trained with real case examples and that these values change once more training datasets are added. 

To get more information on this topic I recommend checking:

• Machine learning meets photogrammetry
• Classification of Aerial Photogrammetric 3D Point Clouds.
• How to generate the Point Cloud Classification.

Best,

[Leo_Deleon]

Blaz,

Thanks for the links, they were very helpful.

-Leo

[sbhgurun.g]

Hi Blaz,
Is there any setting in adobe lightroom so that I can enhance specific color of photo for achieving better point cloud classification?

[Rhea_Garratt]

Hi,

As far as I could tell, Adobe Lightroom is an image editor. We do not recommend editing your images with an image editor before processing them with Pix4Dmapper because:

• If you change the contrast, keypoint detection and matching might not be successful during step 1. Initial Processing.
• You might lose important metadata information in the process. As a consequence, information might be missing for processing step 1 successfully.

You can find more information here.

I recommend you process your original images and activate the point cloud classification in the processing options of step 2 (more). Then, as it is needed in most projects, edit the point cloud classification with the point cloud editing tool. More here.

Let me know if you have any questions.