Page 49 - FDMAsia JulAug 2024

TECHNOLOGY 47
www.fdmasia.com | FDM ASIA JUL/AUG 2024

inaccuracy and partly due to the differences between internal to manual measurements. Difficulties arose when the scars
knot and external branch characteristics. However, some quality were not evident, resulting in errors greater than 5 cm. In all
indexes, such as the maximum knot diameter, showed no other cases, the error was less than 1 cm.
statistically significant difference from the maximum branch Finally, machine learning techniques were effectively applied

diameter at tree level. for the classification of stem surface defects, including branch
Similarly, the number, diameter, and insertion angle of scars, epicormic shoots, burls, and small defects identified
branches were detected both automatically and by an operator in laser point clouds.
from the TLS point cloud. The results indicated good agreement A step further was performed to quantify the dimensions of
between the two methods for branch angle; however, automatic these singularities with promising results. However, difficulties
calculation tended to underestimate diameter, and precision for some flat defects were observed in accurately defining
in branch detection decreased with tree height due to the their borders when compared to assessments made by
presence of the leaving crown. human operators.
Noteworthy is the work of Winberg et al., who utilised a Photogrammetry uses images (photographs) captured from

mobile handheld laser scanner (HHLS) to assess branching different angles of an object to construct a 3D model and
properties in standing Norway spruce trees. determine its dimensions. Either a multiple-image or single-
Two different algorithms for data extraction from point clouds image approach can be utilised in the analysis.
were applied and compared in terms of branch detection, A recent advancement in photogrammetric applications,
determination of branch diameter, and insertion angle. The known as Structure from Motion (SfM), has significantly
collected data were also compared with sawmill data (log contributed to the advancement and widespread adoption
and knot volume, and whorl characteristics) obtained with of this technique. SfM utilises overlapping images captured

X-rays on the same material (at sawlog level). from moving cameras to reconstruct a 3D model through
Overall, the laser scanner successfully predicted branch computer vision techniques.
structures in standing trees, although these attributes showed In a more recent study, Kędra et al. compared single-
only a moderate correlation with sawlog features. image photogrammetry with TLS results to characterise the
TLSs have also been effectively applied to detect and tree architecture of sessile oak. While the primary focus was
measure stem damage from various sources, including on crown characteristics, branch length, thickness, and angle
mechanical actions, insects, animals, weathering, and physical were also obtained through image analysis.
defects such as fork or butt swell. Damage was visually Again, the already-cited Morgan et al. utilised both HHLS
detected and measured both in the field and on the point and photogrammetry techniques to count and measure the

clouds by an operator; the comparison revealed no significant length of stem damages. Their results showed that the extent
difference in the size of damage between the field and desk
surveys. However, the defect count was underestimated when
assessed using point clouds, likely due to the limited tree Corvallis Gazette-Times
height captured instrumentally.
Kretschmer et al. developed two approaches to detect and
measure surface defects, such as bark scars, in TLS point

clouds: one based on data intensity and the other using bark
surface models. These automatic techniques were compared