Data derived from drone photogrammetry is great. It is feature rich and often contains more information than you would ever need. Pointclouds are at times so dense that they can exceed 50 points per square meter. But is all this high definition data really worth much to the end client and is it comparable to traditional outputs from terrestrial survey or LiDAR?
Some photogrammetry software allows for moderate filtering of tie-points that makes up the final pointcloud. While this removes undulation and inaccuracies in the model, it still doesn't extract vegetation, buildings, structures, etc. The pointcloud generated is based of multiple intersections of all objects visible in the image. Thus there is no distinction between these objects. This type of model is called a Digital Surface Model (DSM) and can not be used for contour generation
Some photogrammetry software allows for moderate filtering of tie-points that makes up the final pointcloud. While this removes undulation and inaccuracies in the model, it still doesn't extract vegetation, buildings, structures, etc. The pointcloud generated is based of multiple intersections of all objects visible in the image. Thus there is no distinction between these objects. This type of model is called a Digital Surface Model (DSM) and can not be used for contour generation
Many amateur mappers believe that to create ground level contours is to simply remove the concentric rings that form around tress, buildings etc. This is not recommended as the integrity of the model is still unedited and the terrain may vary due to classification around buildings or larger clumps of vegetation. This practice may work for small single trees or buildings, but what for large plantations? Contours do not simply disappear (unless they terminate into a body of water) and have to have a beginning and end.
The classification of pointcloud data is very time consuming. There are semi-automated tools available that can do limited DTM editing. Most of these tools do not allow for terrain editing and interpolation. Large fields of crops such as maize can not simply be classified as "vegetation" and removed from the DTM as it would create holes in the terrain model. Here the ground needs to be interpolated over to retain the profile.
Since pointcloud data is so dense, the client would need very powerful machines and expensive software to interrogate a dataset. This creates a processing bottleneck and should be thinned to make the model more meaningful and user-friendly. Spatial thinning of pointclouds does exactly this where we can define model keypoints that are crucial in representing the terrain while removing the superfluous ones. This dramatically cuts down file sizes while maintaining model integrity. Flatter areas will have less points but steeper or changes in slope angle will have a denser group. Typical data sets of 10 000 000's of point can be re-sampled to just 50 000 for a typical 1000ha area.
Comentários