Page 29 - Surveyor 54.3 and 4
P. 29
The Potential of Panoramic Images for 3D Indoor Model Reconstruction
Vol.54 No.3 & 4 2019
(a) images able to produce 3D indoor
modelling. As mentioned above, the The result shows
surfaces of the interior building are
distorted, most probably because that researchers
of the residual error of marked point
on each image are larger than it could make use
supposed to be.
(b) of cheap camera
4. Conclusion and for surface
Future Work measurement
The research has presented an using the
(c) approach in which photogrammetric photogrammetric
method could be used for 3D
indoor modelling. This research has software.
achieved the aim and objectives for
3D indoor modelling. The consumer
camera proved that DSLR SLT-A35
can generate a 3D indoor model to
Figure 9: 3D indoor model millimetres accuracy at the images.
(a) Front view, (b) right view, Resolution remains important and
(c) left view time can be considered. However,
it is suggested that a 16 megapixels
camera can provide a good the 3D surface needs to be improved
resolution for surface measurement. effectively to preserve the features
The result shows that researchers on the surface of the buildings.
There are 30 samples of distance could make use of the cheap camera There are several recommendations
have been compared based on for surface measurement using the that would be useful in the future
actual measurement and 3D interior photogrammetric software. The research such as ensure that the
model results. The accuracy of comparison of actual measurement camera used for the research is
the 3D interior model is 0.017m with the measurement that the well-calibrated. The information of
(Root Mean Square Error) and the PhotoModeler used is interesting, camera calibration gives changes
standard deviation is +0.018m. which it is reliable to evaluate the to minimize accuracy of the
This result explained the errors of results. The PhotoModeler Scanner research, the number of images
the measurement in the oriented software uses their ability in stitching for each point is marked must
images that are affected by the more than 10 images to complete be maximized to have a good
marking point’s process. From the the orientation and manage to overlapping coverage, ensure that
results, analysis can be made by generate a 3D indoor modelling. The all points appear on three or more
taking precautions that the errors successful to complete the research images, the number of points that
of the model will be minimized, in PhotoModeler proves that the appear at two photographs have to
and measurement accuracy has a generation of 3D model can be used maximize otherwise the research
possibility to maximize. in any applications related to indoor will not successful to carry out
space. Performing the process in this tie points measurement on next
All four models are captured in four research with a simple measurement images, ensure that the angle
different views: front, right, left and such as tape distances, between the camera positions close
back view. The surfaces are not as inexpensiveness of instrumentation to 90 degrees as possible. For the
perfect as the original images due like digital camera fits the procedure research, the camera captured
to the overlapping of marking points in orientation processes and easy 360° view, at least, 25 points in the
during the orientation process. The to capture the sequential photos photographs have good coverage
3D model in Figure 9 is produced are several advantages that can be with 60% of overlapping coverage
using 15 images from digital camera described. and the points need to be marked
DLSR SLT-A35. The procedure is on each image precisely to minimize
applied, and the results indicate For future developments on 3D the residual errors in referencing
that the accuracy of panoramic modelling, reduction of processing process.
29