Acquisition of digital elevation models at high spatial and temporal resolutions using automated digital photogrammetry

Tests carried out on the Flood Channel Facility (FCF) and the Tilting Flume at Wallingford, have revealed that automated digital photogrammetry can generate digital elevation models (DEMs) to represent the surfaces of dry sand and gravel bed flumes. The technique makes use of the latest generation of digital cameras combined with modern digital photogrammetric software, which offers several unique advantages. All DEMs are generated 'off line' using photographs taken of the flume. Consequently, experimental work only has to pause for the image acquisition phase, which is potentially less than a few minutes. The scale of the photographs is flexible and so various size objects can be measured, with a commensurate and known precision. Much of the data processing is carried out automatically using sophisticated image correlation methods, running on a workstation. These allow DEMs to be generated at rates exceeding 100 points per second, allowing complex and detailed surfaces to be represented accurately. The intention of this review is to clarify basic operating requirements and present results achieved during funded research work at Wallingford.

Digital image acquisition

Although conventional film based cameras are traditionally used for photogrammetric measurement (e.g. Wolf, 1983) the latest digital cameras are more efficient and potentially more accurate. A digital camera not only obviates the need to scan photographic negatives but systematic error sources due to film unflatness and distortion are eradicated. The imagery can also be downloaded and processed within minutes of acquisition and this allows exposure settings to be checked and modified if necessary. High-resolution digital cameras are expensive (e.g. Kodak DCS460 camera- costs £17,000) and are not designed for photogrammetric usage. Despite this, recent literature (Fraser, 1997) and our own tests (Chandler et al., 2001) demonstrate that the internal geometry is sufficiently stable for accurate data to be generated.

Images acquired from a vertical perspective (Figure 1) allow efficient coverage of an area and the gantries at Wallingford provide a versatile and movable platform for the camera. A major advantage with photogrammetric methods is the positive relationship between photo-scale and precision. Typically high precision is required for studies involving small areas (i.e. large-scale photography) and a lower precision is acceptable for larger areas (i.e. smaller scale imagery is most efficient).

Control requirements

It is necessary to relate measurements derived from the photographic images to a 3D-site coordinate system (typically rectangular Cartesians). The most effective means of achieving this involves placing a limited number of "photo-control" points throughout the area of interest, (Note: black/white targets visible in Figure 1). The design of these is flexible but points need to be both visible on the imagery and at a known location within the desired coordinate system. Once established, it is most efficient if such a control framework can remain in place during the experimental work, although this is not essential.

Data processing

The stages involved in transforming the photographs and control points into digital elevation models follows established photogrammetric procedures and are well documented elsewhere (e.g. Brunsden & Chandler, 1996; Pyle et al., 1997; Greve, 1996; Walker, 1996). The usage of a digital camera requires an additional stage that involves calibration of the camera itself, which may not be considered routine. Further details of this approach can be found in Chandler & Cooper, 1989 and more recently in Stojic et al., 1998 and Chandler, 1999.

The Flood Channel Case Study

The Engineering and Physical Sciences Research Council (EPSRC) funded project (GR/L69213), awarded to Dr. K. Shiono and Dr J. Chandler at Loughborough University involves studying the influence of turbulence and secondary flow within the Flood Channel Facility. Part of the study requires generating digital elevation models to represent the complex channel bed morphology over one full wavelength (15m) by digital photogrammetry. A DCS460 digital camera (loaned from EPSRC project GR/L58118- Prof. Lane/Richards and Dr J. Chandler) was used to acquire the vertical imagery, the camera located on a simple board attached to the gantry using 'G' clamps. This provided a stable but movable camera platform that was 4.5m above the flood plain of the FCF. Thirty-three control targets were placed on the flood plain and the channel banks and their positions determined using a total station and a digital level (survey work, approx. 5hrs). The channel was illuminated using two standard arc lamps and photography was acquired in less than one hour, (Figure 1). Images were downloaded to a portable PC to verify exposure settings and coverage, prior to returning to Loughborough for processing and DEM generation (Processing time, approx. 10 hrs). Figure 2 portrays the final DEM extracted at a resolution of 10mm over one full wavelength of the channel. Note how the morphological structures apparent in the photograph are reflected in this high resolution DEM.

Figure 2, Slope shaded digital elevation model of bed surface

The Flood Channel Case Study- Accuracy assessments

The touch sensitive 2-D profiling system employed on the FCF was used to measure cross sectional profiles at various critical locations along the flume. The accuracy of elevations determined by digital photogrammetry could therefore be assessed, by extracting the same cross sectional profiles through the DEM and comparing them with those directly measured. Figure 3 represents a typical profile and illustrates that accuracies being achieved at this photo-scale (1: 160) are between 2-3mm. It is also apparent that the density of data generation is far higher (one point per 10mm), compared with the touch sensitive probe (one point per 25mm). It should be remembered also that this higher density is achieved across the whole flume in three dimensions and is not restricted to just the arbitrary plane selected for comparative purposes and accuracy assessment.

Figure 3, Accuracy of bed data- comparing photogrammetry with manual profiler

Conclusion

This short report has demonstrated the value of automated digital photogrammetry for extracting the bedforms created during flume experiments. The methods have also been applied successfully in monitoring bed changes in a real braided river in the Canadian Rockies using oblique, ground-based digital photographs (Chandler & Ashmore, 2000).

Wolf, P.R., 1983. Elements of photogrammetry. McGraw Hill: Singapore.