obtained using a T-lidar Illris-3
D
instrument (Optech). Based on
the approach of Cote
et al
. (2009), the perspective of the
PBRT
camera was set to cover a field of view of 40° × 40°, and indi-
vidual rays were launched within the field of view from the
camera to the objective tree. The coordinates of each intersec-
tion point between rays and the surface of the virtual tree were
recorded. The virtual tree was “scanned” at azimuthal angles
of 0°, 90°, 180°, and 270°, with each simulated scan producing
a 3000 × 3000 raster image. The final point cloud, containing
2 023 709 points, was obtained after ray tracing (Plate 1). The
two trees and one virtual tree differed in tree height and tree
branch diameter, resulting in a gradient in processing difficulty.
Methodology
Geometric Method for Wood-Leaf Separation
Trees were horizontally sliced using the proposed geometric
method. Geometric primitives, including circles, circle-like
shapes such as arcs and incomplete circles, and line segments
were detected in each sliced bin. The raw point cloud can be
classified into wood or leaf components by thresholding the
sizes of the detected geometric primitives. A flow chart of this
method is shown in Figure 1.
Step 1 - Horizontal Tree Slicing
In the first step, the point cloud for each tree was sliced into
successive bins of height around 10 mm. Figure 2 shows the
top views of four typical sliced bins at different height lev-
els for the camphor tree. The circle in Figure 2a represents
the main trunk, and those in Figure 2b represent first-order
branches. In Figures 2c and 2d, lines, rather than circles, are
used to identify small twigs and leaf clusters, because of the
decrease in branch size. The changes in circle or line sizes
along the tree height constitute the cornerstone of the proposed
method; this will be fully described in the following sections.
Step 2 - Noise Filtering Using Triangular Irregular Network (TIN)
In the second step, 2
D
Delaunay triangles were built among
points in each sliced bin. The area of each constructed
Figure 1. Scheme of proposed geometric method for wood-leaf
separation.
Figure 2. Points in four sliced bins at different heights of camphor tree showing that different geometric primitives (circles, arcs or lines)
can be found at different tree heights.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
October 2015
769
