Hierarchical registration of laser point clouds between airborne andvehicle-borne data considering building eave attributes
2021
Laser point cloud registration is a key step in multisource laser
scanning data fusion and application. Aimed at the problems of fewer
overlapping regional features and the influence of building eaves on
registration accuracy, a hierarchical registration algorithm of laser
point clouds that considers building eave attributes is proposed in this
paper. After extracting the building feature points of airborne and
vehicle-borne light detection and ranging data, the similarity measurement
model is constructed to carry out coarse registration based on
pseudo-conjugate points. To obtain the feature points of the potential
eaves (FPPE), the building contour lines of the vehicle-borne data are
extended using the direction prediction algorithm. The FPPE data are
regarded as the search set, in which the iterative closest point (ICP)
algorithm is employed to match the true conjugate points between the
airborne laser scanning data and vehicle-borne laser scanning data. The
ICP algorithm is used again to complete the fine registration. To evaluate
the registration performance, the developed method was applied to the data
processing near Shandong University of Science and Technology, Qingdao,
China. The experimental results showed that the FPPE dataset can
effectively address the coarse registration accuracy effects on the
convergence of the iterative ICP. Before considering eave attributes, the
mean registration errors (MREs) of the proposed method in the
xoz plane, yoz plane, and
xoy plane are 0.318, 0.96, and 0.786 m, respectively.
After considering eave attributes, the MREs decrease to 0.129, 0.187, and
0.169 m, respectively. The developed method can effectively improve the
registration accuracy of the laser point clouds, which not only solves the
problem of matching true conjugate points under the effects of the eaves
but also avoids converging to a local minimum due to ICP’s poor coarse
registration.
Keywords:
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
36
References
0
Citations
NaN
KQI