The STRS method applied on MARV4/2

Introduction - Manual photogrammetric 3D tree top positioning

  Photogrammetry can be applied to map 3D tree top positions. This requires that at least two images are available:
Treetop mapped in a pair
Treetop measured in a pair
Photogrammetric treetop positioning from an image pair. Upper figure gives the principle - lower figure displayes a real case. The operator (human) points the treetop in the left image and this image ray down towards the tree from the left camera is displayed over a range in Z in the right image with the blue line. The corresponding tree top can only reside along this ray - the epipolar line. The operator has to find the same object poit - treetop - by following the line.

Tree height requires a DTM
When the tree top is pointed in the right image, we have two image observations, i.e. two image-rays or 3D lines that nearly meet (intersect) in the object space. We find that 3D point by a method called least square ray-intersection. This is our estimate of the treetop position. Here that point is displayed in green. If we know the terrain elevation at that XY-location, we have an estiomate for tree height, 19.4 m here. The white line is a fictious trunk that connects the top with the base.

Note: Optionally it is possibly to use stereo measurements, where the operator sees a stereo model made of an image pair and places the 3D measurement mark on the tree top. This requires a good stereo vision.
Back to index,   Next

Multiple images improve accuracy
Treetop positioning using a set of six images. The estimation is more accurate and the solution of the correspondence problem is more reliable if several views are available (comparable to #DOP-values in satellite positioning). In the example, the tree top was pointed in the two leftmost images on the upper row. Their image-rays are dispayed over range of 10 meters in Z (± 5 m around "Z-good-guess") and the epipolar line segments intersect in the other 4 views over a tree top. The "Z-good guess" i.e. the expected elevation of the tree top has to be set by the operator in the beginning. Later it can be the elevation of the previously measured tree.
  • Large-scale images are required, > 1:12000
  • The very apex is often missed, which leads to height underestimation. The larger the scale the less underestimation
  • Crown shape affects the discernibility of the treetop. Trees can have round tops or multiple tops (old birch trees).
  • Clumped crowns of deciduous trees are difficult to discern. Trees fuse together and separation is impossible.
  • DTM by LiDAR helps, since in closed canopies the terrain cannot be seen
  • The probability for a treetop to be in direct sunlight is a function of the relative tree height. Trees shorter than 0.5 are usually not measurable at all.
Discernibility of trees
An example of the measurability of trees in six spruce stands. The likelyhood is a exponential function of the relative height of the tree. Stand density execises an effect as well. S3 is an unmanaged dense stand. S1 has been thinned most.

Introduction -  Manual image measurement of crown width
Manual measurement of Crown width
The operator can select a view in which the crown is best seen and make a measurement of the crown edges - here perpendicularly to the direction of the trunk. With the image scale known, the pixel size in meters is known and the measured image distance can be converted into an observation in meters, here 3.71 m.
  • Dcrm is usually underestimated, since the widest part of the crown is not seen in the images in direct light or it is occluded by neighboring trees. Underestimation is a function of tree height. Photo-based Dcrm observations are therefore rather unreliable.