Figure 1.
(a) Intertidal rocky shore at Greenfields Beach (study site shown in red box) at Jervis Bay, New South Wales, Australia, (b) ground-based photography at the study site taken at low-tide. Google maps imagery of the site is available at: http://maps.google.com.au/?ll=-35.085668,150.693294&spn=0.005759,0.0109&t=h&z=17.
Figure 2.
(a) 2.7 m wingspan conynes-delta kite using to lift (b) a Picavet suspension rig that was used to attach each of the two downwards-facing Sony NEX-5 digital cameras (one colour and one near-infrared). The operator walks the kite across the intertidal zone collecting multiple, overlapping photographs. Examples of aerial images collected from an altitude of approximately 15 m are shown in colour (c) and near-infrared (d).
Figure 3.
(a) Spectral response functions for the colour and near-infrared converted cameras: the red, green and blue channels correspond to the three channels of the colour camera whereas the near-infrared curve corresponds to the red channel of the near-infrared converted camera, which was found to have the highest response of each of the channels for this camera. (b) Reflectance spectra for key surface coverage types in the intertidal zone measured using a handheld spectroradiometer. The reflectance spectra were used in conjunction with the camera spectral response functions to validate the measured colour of objects in the kite-based imagery.
Table 1.
Peak response values of the colour and near-infrared converted cameras.
Figure 4.
Overview of the kite-based mapping process.
During data acquisition, ground control points are placed in the environment and images collected over the the terrain using the kite and camera. After data acquisition, images are processed to extract and match features across multiple overlapping images. These features are used to reconstruct the poses from which images were captured and a 3D pointcloud of the terrain using a structure-from-motion algorithm. The pointcloud is geo-referenced using the ground control points and a photo-texturing process is used to create 3D topographic maps and high-resolution geo-mosaics.
Table 2.
Overview of software implementations used within the processing procedure.
Table 3.
Comparison of the number of matching image pairs and average feature matches per image for colour-to-colour, colour-to-near-infrared and near-infrared-to-near-infrared image pairs.
Figure 5.
Example of extracted and matched SIFT image feature data.
Shown are two overlapping images (image 1 and image 2) with annotated positions of SIFT features that have been matched between the two images (blue) and lines displaying the computed correspondence between points (green) (only every hundredth correspondence shown for clarity). Also shown are detailed sections of each image illustrating the feature points.
Figure 6.
Colour and near-infrared photomosaic layers of the intertidal rocky shore reconstruction at Greenfields Beach.
(a) Colour mosaic and (b) near-infrared mosaic for the whole shoreline. (c) and (d) show a detailed section of the colour and near-infrared mosaics illustrating the different scales achieved across the entire map.
Figure 7.
Elevation, slope and aspect data derived from the 3D topographic reconstruction.
(a) Elevation above maximum low tide (b) slope of the terrain and (c) aspect of the terrain.
Figure 8.
Detailed views of elevation and slope data derived from the 3D topographic reconstruction.
(a) Elevation above maximum low tide and (b) slope of the terrain.
Figure 9.
3D photo-textured rocky shore reconstruction.
(a) 3D oblique view of the shoreline, (b) and (c) Detailed oblique views of rock platform section from different viewing angles. The level-of-detail visualisation system allows for different model scales to be visualised in a single, continuous terrain model.
Table 4.
Measures of both global and local-scale (fine-scale) spatial map errors from Ground Control Point (GCP) residual errors.
Figure 10.
Normalised Difference Vegetation Index (NDVI) maps derived from raw colour and near-infrared imagery.
(a) NDVI map for entire shoreline, (b) detailed view of NDVI and (c) corresponding colour imagery of detailed section.
Figure 11.
Example mosaic imagery highlighting some of the dominant surface types compared in this study.
Figure 12.
Comparison of predicted and measured Normalised Difference Vegetation Index (NDVI) values for different intertidal zone coverage types.
Predicted NDVI values were derived using knowledge of coverage type reflectance spectra and camera spectral response functions (see Figure 3). Measured NDVI values correspond to manually extracted image patches taken from the reconstructed NDVI mosaic maps (error bars show the standard deviation from multiple sample points).