Measurement of crown width by multi-scale template matching

  After the treetop solution is obtained, the optimum scale is found from the view that has the least oblique viewing angle. These views are best for the determination of crown width (Mäkinen et al. 2006, CJFR). Multi-scale template matching is used for that with the exception that the treetop image position is allowed to take values in the neighborhood (image window) of the solution. The crown width of the model tree is multiplied by the scale factor (e.g. 0.8) to give a crown diameter (Dcrm) estimate. Hence, there can only be as many values for Dcrm as there are scales (Fig 2).

Dcrm measurements
Fig 1. In this example five trees have estimates for Dcrm, which are the results of multi-scale TM in the image, which has the green circles. This image has the least nadir-angle. The model tree's Dcrm needs to measured manually [m]. If there are 11 scales, the Dcrm of the other trees can take only 11 values.

Back to index,   Next

Correlation over scale
Fig 2. Here the optimal scale is 3 (0.7 of the model tree's Dcrm). The correlation has a flat maximum - typical - and an indication of the unreliability of the method. Noise (varying backgroung, varying crown shapes, partial occlusion) in the images  is the main source of uncertainty.

Not so good results of Dcrm
Fig. 3. The method was tried here in a rather sparse pine stand (30 x 30 m) on barren CT (Calluna vulgaris) -site (sandy glacial delta). The yellow circles are set by an operator - and they give an idea that gross errors (underestimation) is possible and common. The method needs improvement.