Computer Vision and Pattern Recognition Technologies for Construction - Mobile and Pervasive Computing in Construction - page 202

Flexibility and portability: the system can be used in a wide range of applica-
tions. Moreover, the sensor system is low in weight and battery operated, which
make it highly portable.

*

Videogrammetric surveying suffers from a number of limitations as well, which
counteract its benefits for some mapping activities:

* Need for sufficient daylight: since videogrammetry uses the natural light in
the environment to capture visual data, its effectiveness is highly dependent on
the environmental conditions of the site and the presence of significant
daylight.
* Limited range distance: the range measurement error in videogrammetry is
proportional to the square of depth. Therefore, the accuracy of the measure-
ments over large distances can be unacceptable for construction site surveying.
* Sensitivity: severe illumination changes, abrupt motions of the sensor system
and noise in recorded frames can result in the failure of the system. Moreover,
accurate camera calibration is critical to achieve precise measurements.

A videogrammetric system does not have any limitation on the number of
cameras. However, the single and stereo camera systems are mostly used in practice.
Although there are some differences, the general process of videogrammetry for any
arrangement of cameras is almost same and mainly composed of seven steps:
(1) calibration of the camera system, (2) acquisition of video sequences, (3) pre-
processing to select appropriate frames, (4) feature detection and matching,
(5) structure and motion recovery, (6) stereo mapping, and (7) reconstruction
of the target points in 3D space. Acquired 3D data can then be represented as a
triangular mesh with a photo-realistic appearance. A common challenge for all
existing remote sensing technologies is that a lot of manual editing and correction
processes are still needed to convert 3D spatial data into a surface of a construction
site model.
A complete system to build geometric models from video sequences was
presented by Pollefeys et al . (2004). A 20-second video sequence captured from
a monumental fountain was used to generate a 3D model (Figure 10.7). The
approach resulted in a relative error of less than 1mm for the reconstructed points.
The same approach is applicable to construction site videogrammetric surveying.
However, the typical measurement range required in construction sites is much
more than the range value used for monumental surveying. Therefore, a higher
degree of error should be expected in this case, except when using very high
resolution video cameras.
In general, videogrammetry is a promising approach for construction site
surveying which allows the calculation of 3D spatial data through highly automated
methods. However, the possibility of fully automated videogrammetric surveying is
still at the early stage of development, as the accumulated errors do not allow the
system to run for long time. Errors in the calculation of the 3D spatial coordinates
of a target point limit the use of videogrammetry in practice. There is, therefore, a
need for further research to increase the quality and robustness of the process.