Digital Architectural Photogrammetry and CAAD

Go back to list of examples

DIGITAL ARCHITECTURAL PHOTOGRAMMETRY AND CAAD
EXAMPLES FOR DIGITAL THREE-DIMENSIONAL RESTITUTION

HIP BUILDING, CAMPUS OF ETH ZURICH (SWITZERLAND)

OBJECT

The object used in this test is a building housing drawing rooms for architectural students on the campus of ETH Zurich. The architects Huber, Bolli and Gerber designed this building. It is composed of pre-manufactured parts which can be dismounted and rebuilt. The construction of the building follows the rules of a brickbox. It consists of wooden panels, screw connections, four different window types and a roof. These elements are placed on a concrete platform and are divided in the middle by a wall. The dimensions of the building are 55 by 11 m in width and depth and 8 m in height.

DATA ACQUISITION

During this campaign the image acquisition was performed with two representative solid-state camera systems, an inexpensive S-VHS camcorder and a "high-resolution" CCD-camera with digital read-out.

JVC GR-S77E

The imaging system JVC Camcorder GR-S77E is a medium price range consumer product. The free-hand portable camera allows on-site control for the acquired imagery via an internal monitor. It incorporates a 1/2" colour sensor (6.4x4.8 mm2) with nearly 420'000 sensor elements. The analog images are stored on a S-VHS video tape and have to be digitized by a framegrabber. The digitized images have a size of 728x568 pixel.
Using such an imaging system in architectural photogrammetry requires that most functions which are implemented for the convenience of a common user (e.g. autofocus, zooming functions or image stabilizer) are disabled. In this test the autofocus was disabled and the zoom lens was fixed at its shortest focal length. This system does, however, offer the ability to store a huge amount of data on inexpensive video-tapes and allows on-site quality control for the acquired imagery.

Kontron ProgRes 3000

The Kontron ProgRes 3000 is a high resolution camera with digital read-out. It uses the microscanning principle which is based on the idea of sensor displacement to improve the number of pixels. The whole equipment includes the camera head, a camera control unit, a host computer and an additional TV monitor to visualize the current viewpoint of the camera. The camera employs a standard 2/3" (8.8x6.6mm2) CCD-sensor with 580x512 sensor elements, which will be displaced by very small increments to get several different images. These single images are later interleaved to produce one high resolution image. The basic resolution, generated from two partial images of the TV frame, is increased 6-times in x-direction and 4-times in y-direction to achieve the high resolution. The displacement of the sensor in the camera is achieved by piezo controlled aperture displacement (PAD). Totally 96 subimages from 48 PAD positions are taken. The size of the final image is 3072x2320 pixel.

22 images were acquired with the JVC camcorder and 20 images with the ProgRes 3000.

RESULTS

The photogrammetric analysis of the digital image data was performed with DIPAD. In order to analyse the photogrammetric processing a reference system was defined by 19 signalised points which were fixed on the facades and used as control points. The reference data was determined geodetically, involving the precise measurement of the control points with a theodolite and the estimation of the three-dimensional coordinates with a geodetic triangulation program. The control points were determined with an accuracy of 1.5mm in plane and 1.0mm in height.
The coordinate system is defined with x- and y-axis in plane (x-axis parallel to the front facade) and z-axis in vertical direction. Referring to the dimensions of the object (55 m in x- and 11 m in y-direction) and to the camera configuration, the precision in the x- and z-directions is comparable to the precision within the plane of each facade and the precision in y-direction is comparable to the precision in depth.
The image analysis of the test object was performed with a bundle adjustment program which uses the image coordinates of all points as input data. Five different versions were calculated, three with the data set acquired by the JVC camcorder and two with the data set acquired by the Kontron ProgRes. All versions are performed as a bundle adjustment. The Version x1 uses the results of the testfield calibration without additional parameters. The Versions x2 are using the whole information of the testfield calibration, whereas the Versions x3 are performed by a bundle adjustment with self-calibration.

Numerical results of bundle adjustment for the test object

Version	# of stations	# of object points	redundancy	standard deviation of unit weight a posteriori	sX [mm]	sY [mm]	sZ [mm]
11	22	789	1072	7.48	19.53	34.62	15.33
12	22	789	1072	5.69	15.53	27.04	12.14
13	22	789	1062	5.79	15.93	27.63	12.31
22	20	571	869	2.06	6.74	12.35	4.99
23	20	571	860	2.17	6.99	13.19	5.22

All versions: Bundle adjustment.
Versions 1x: Image acquisition with video camera.
Versions 2x: Image acquisition with high-resolution CCD camera.
Version 11: no additional parameters.
Version x2: pre-calibrated camera.
Version x3: self-calibration.

It is remarkable that the high-resolution camera delivers a precision in object space improved by a factor 2 as compared to the precision delivered by a low-resolution camera, despite the fact that the ProgRes delivers an image with a twenty times higher number of pixels than the JVC. This because the precision in object space is not exclusively characterized by the number of pixels delivered by a camera. It is a result of the whole photogrammetric setup, including the camera configuration which provides poor intersection angles and only 2.3 image rays per object point. Especially the degraded precision in depth is caused by this particular camera configuration and perceptible in all calculated versions.
The comparison of Version 11 and 12 shows that the precision improvement attained by using this set of ten additional parameters for the JVC is in the order of 25%. This is typical for cameras employing inexpensive lenses. The comparison of the Versions x2 and x3 indicates that for both imaging systems, the precision derived by a bundle adjustment with self-calibration is comparable to a bundle adjustment with a calibrated camera. This confirms the well known fact that self-calibration is a tool to avoid additional work of a test field calibration, if the camera configuration is suitable for a self-calibration.
In general both imaging systems provide sufficient accuracies for architectural tasks. But it is remarkable that the number of details measurable in the ProgRes images is much higher than in the JVC images.

Photogrammetrically generated CAD model

The photogrammetric analysis with DIPAD delivers a three-dimensional geometric and semantic object description. This allows the automatic creation of a Digital Surface Model in a CAAD environment, which represents the object by surfaces instead of simply lines.
The CAAD system is able to preprocess the data and store it in data structures adapted to architectural purposes. The system is capable to find efficiently special data in a big data base. It allows easily data transformations into other representations. The task of the architect is just the creative finding of new solutions or to judge the current solution. The CAAD system is suitable for documentation and visualization, and for complex simulations, manipulations and analysis of the object.

Photogrammetrically generated Digital Surface Model

Photogrammetrically generated Digital Surface Model (zoomed view)

Generated 3D Model with different data types
(from left to right: Surface Model, Wireframe Model, Point Model).

Last change: 26-May-95 (André Streilein)
Problems and/or queries, send e-mail: andre@geod.ethz.ch