Oblique Imagery to Make 3D Models

Introduction:

UAS imagery flown in a spiral like order around an object with the camera adjusting its orientation to capture images can be turned into a 3D model using Pix4D software.  This lab goes over the process of creating a 3D model using Pix4D and how to annotate it.

Methodology:

The first step to creating a 3D model is to have the proper orientation of the images.  The following image is an example of how the images were taken around a bulldozer at the Litchfield mine.  This flight pattern is necessary to get all the angles needed to create a 3D image.  These images are to be brought into Pix4D and the 3D model template is to be used.

Once the images are loaded the user can create a 3D model by running processing steps one and two.  This will create a 3D Model.  The image below is of a 3D model that was created without doing any additional editing.  The program does not know which pixels are suppose to be the model and which ones are suppose to be the back ground.

To get a 3D model with no distortion the user needs to edit the images using annotation.  This is the process of removing pixels from images that the user does not want included in the background.  This could be the sky, or ground, or a pole obstructing the whatever is the object to be modeled.  Below is an image showing what the annotation process looks like.  The user uses the annotation tool to highlight the area or objects to be annotated then the annotations are applied to the images.

When the annotation process is complete the user re-optimizes the initial processing and runs part two of the processing.  This creates a 3D model using the images that were annotated and removes pixels that are not wanted.  This was done with three different sets of imagery.


Discussion:

In theory annotation will take away all pixels in the image that create distortion.  Unfortunately it is not a perfect process and some of the models did not turn out the way they should have.  The first data set of the bulldozer turned out very distorted with lots of holes in the model.  This could be because of pixels being wrongly removed from the bulldozer and the program then wrongly takes those pixels out of all the images.  This could also have to do with only three images being annotated.  Annotating an image takes a long time to get all of the pixels selected.  The difference between the non-annotated and annotated model of the bulldozer can be seen below.

Pre-Annotations

Post-Annotations

The next model was of a truck parked in a parking lot.  This model turned out better but still not great. 5 images were annotated for this model.  The annotated image is crisper and has less distortion but the bottom of the truck is still very distorted.  This is because no images were taken from that low of an angle and the program did not know what to do with that area.

Pre-Annotation

Post-Annotation

Conclusion:

Annotation is a very useful tool but can be applied in a more useful way than it was in this lab.  Annotation can be used to remove an object like a pole or stick that is interfering with the model.  The application it was used in for this lab resulted in okay results but not perfect.  The bulldozer was left as it was because it seemed good to show that mistakes could be make and the resulting model can turn out bad.  If this were being done again more images would be annotated and they would be done more carefully and the model would turn out much nicer.