How to improve the outputs of dense vegetation areas?

Trees, forest and dense vegetation areas have specific properties that require special care. When processing such datasets, 3 type of issues might occur:

Initial Processing: Low number of calibrated images
Point Cloud Densification: Areas without 3D points generated
Orthomosaic: Visual deformation and artifacts

Initial Processing: Low number of calibrated images

Pix4Dmapper relies on visual similarities between overlapping images to reconstruct the model. Trees and dense vegetation, due to their complex geometry (thousands of branches and leaves), often appear very different between overlapping images. Therefore, it is more difficult to find enough similarities between overlapping images and, if the overlap is not enough, it leads to a low number of calibrated images.

To ensure to have a high number of calibrated images it is recommended to:

  • Process with 1/2 Keypoints Image Scale: This processing option can lead to a higher number of calibrated images in case the default option (original keypoint image scale) does not calibrate all the images. For more information: Menu Process > Processing Options... > 1. Initial Processing > General.
  • Increase the overlap between the images to at least 85% frontal overlap and 70% side overlap.
  • Increase the flight altitude: When flying higher, the images suffer less from perspective distortions and it is easier to detect visual similarities between overlapping images in vegetation areas.

For more details about the flight plan: Step 1. Before Starting a Project > 1. Designing the Image Acquisition Plan > a. Selecting the Image Acquisition Plan Type.

Point Cloud Densification: Areas without 3D points generated

Even when all the images are calibrated, it is possible that no additional 3D points are computed at areas with trees and vegetation during step 2. Point Cloud and Mesh. In order to improve the Point Cloud in such areas it is recommended to:

  • Increase the overlap between the images to at least 85% frontal and side overlap.
  • Increase the flight altitude: When flying higher, the images suffer less from perspective distortions and it is easier to detect visual similarities between overlapping images in vegetation areas.
  • Set the Image Scale option for the Point Cloud to 1/4 or 1/8 if the dataset consists only of vegetation areas. If the dataset has roads or buildings as well, set the Image scale option to 1/2 and use the Multiscale option. For more information: Menu Process > Processing Options... > 2. Point Cloud and Mesh > Point Cloud.

Orthomosaic: Visual deformation and artifacts

The Orthomosaic is generated from the DSM and it suffers from artifacts in areas where the 3D model is not good. As trees have an infinite amount of details (trunk, branches, leaves) they can only be approximately modeled. This approximation creates visual distortions and artifacts when generating the Orthomosaic. In order to remove the distortions it is recommended to:

  • Improve the DSM by:
  • Improve directly the orthomosaic by:
    • Using the Mosaic Editor: Replace areas with distortions by images generated using a planar projection instead of an ortho projection. For more information about how to use the Mosaic Editor: Using the Mosaic Editor.
       
      Important: Planar projections may not preserve distances and should be used only to improve the visual aspect of the orthomosaic and not to generate orthomosaics that will be used for measuring purposes.
Was this article helpful?
33 out of 35 found this helpful

Article feedback (for troubleshooting, post here)

2 comments

  • Chandramouli Arunachalam

    I used DJI mavic mini drone to survey a Neem plantation site of 25 acres for a test case to estimate the Plant/ Tree height. The Plants and Trees are planted at equal spacing intervals of 4m between each tree and the site is distributed with small plants of height ranging from 0.5 m to big trees of height ranging upto 7.5 meters. Image acquisition was executed using dronelink software at 60m AGL, Nadir angle with Front overlap of 85% and Side overlap of 75%. The produced Orthomosaic GSD was 2.05 cm.

    The issue I faced was with point cloud results.
    1. At some places, few trees which are dense and have bigger crown area didn't have point cloud data/ very less point cloud/ holes in point cloud.
    2. But some trees which are small got high point clouds information at some places and at some places some small trees also dint get point cloud data.

    Due to this I was not able to estimate the height for Neem Trees Properly. I am not able to conclude whether the issue in point cloud generation is due to

    1. Lack of denseness of branches or leaves in a tree?
    2. Lack of bigger tree crown/ smaller tree height (about 1 m measured at fied) ?
    3. Similarity in soil color (beach Sand- White Color) and tree color (light Green)?
    4. Existing wind speed (5.5 m/s to 8 m/s) while UAV data acquisition?.

    Also attached the images of

    1. Pix4d mapper point cloud generation settings,

    2. Tree Distribution at site imagery and

    3. Some images of results (Point cloud data - top and cross section view with orthomosaic imagery)

    for your reference. Kindly suggest me the possible cause for this issue and also solutions to rectify this and achieve good results in point cloud data for estimating the plant height.

    Thanks in Advance..

     

     

    Edited by Chandramouli Arunachalam
  • Momtanu (Pix4D)

    Can we get your quality report (pdf)? The point cloud/3D model/ any other outputs depend on step 1 as the main input for the generation of all these outputs in the automatic tie points that are generated from step 1. The reconstruction will also depend on the flight plan. Along with a nadir flight, you can also try an oblique flight to capture the plants from other angles.

Please sign in to leave a comment.