Stereo photogrammetry is a technique whereby two or more photographs are used to create a 3D model by triangulating common points using parallax. Parallax is the phenomenon by which we are able to percieve depth: the right eye and left eye both see different viewpoints of the same scene, where foreground objects are in different positions relative to background objects. By turning one's head left and right, you can see this effect in action. Photogrammetric modelling is based on common points relative to one another within two separate images to triangulate the positions of the camera used to capture the images. The ADAMtech Mine Mapping Suite is the preferred software for creating photogrammetric models, though more cost-effective software suites are available (1, 2, 3D Catch, Photomodeler, and Agisoft). The strength of ADAMtech is its relative-only point generation, which automates the process of generating matching points between images to triangulate the camera positions and generate an accurate camera calibration and finally point-cloud.
It is important to understand the principles behind photogrammetry prior to attempting a project. First and foremost is overlap, or how much of the subject is seen in two separate images. In a convergent pair project, as close to 100% overlap as possible is ideal, whereas a strip project requires overlap of around 66% (see below) between images. While keeping overlap in mind, base to distance ratio is just as important when creating a 3D model. Where base is the distance between camera stations (where the camera was when the photo was taken) and distance is the distance from the camera to the subject. The ideal range of base to distance ratios is between 1:1 and 1:7 depending on the type of project. There is one cavaeat to base to distance ratios: the closer to 1:1, the more depth accuracy, however it becomes more difficult to match between images at such ratios.
Another important aspect of accuracy in a photogrammetric project is camera calibration. The camera calibration solves for the distortion of any given camera and lens combination. Barrel distortion can be easily modeled, while complex distortion cannot. For this reason the use of prime lenses rather than zoom lenses is ideal. For convergent pairs and strip models, a lens between 24mm and 60mm is ideal, while fan projects are better suited to telephoto lenses. To calibrate a camera, one must take at least six overlapping images, 2 horizontal, 2 vertical (90º) and 2 vertical (270º).
Once these images are captured, a camera calibration may be modeled using 3DM Calibcam. A visulization of the calibration will appear, which can be used to check for proper calibration. Ideally, it should concentric, like this one where blue is undistorted and red is badly distorted. After calibration, a model can be created using the captured images.
Three types of projects can be used to create photogrammetric models:
A convergent pair project consists of two (or more) photographs with close to 100% overlap. It is very fast and easy to capture the images required to build a convergent pair, but resolution is limited because of the distance to the object: i.e., ground pixel size (the real-world size of each pixel on the subject) will be larger than with a close range strip project. However, with the advent of ultra high resolution DSLR cameras, it will become increasingly easier to create a high resolution photogrammetric model using convergent pair geometry.
Figure 1: Two convergent pairs (Birch, J. 2007) Figure 2: 3D view in 3DM Calibcam of convergent pair, purple
squares are cameras, yellow are matching bundle points.
Strip projects are more time consuming to capture in the field than convergent pairs, however they can create incredibly high resolution models when captured with a macro lens. A strip project is captured as overlapping strips of images with 60% horizontal overlap and 20% vertical overlap. A strip project can comprise 2, 10, 50 or hundreds of images, although building large strip models is time consuming. A secondary outcome when capturing strip models is that the constituent images can be combined in another program such as Adobe Photoshop or Microsoft Ice to create a high resolution photo mosaic of subjects that can be a useful overview when linking individual images in ADAMtech, or as a standalone image for publication.
Figure 1: A strip model with ideal overlap (Matthews, N.A. 2008) Figure 2: 3D view in 3DM Calibcam of strip model
A fan project consists of two or more camera stations of multiple images. These multiple images can be either a single strip or multiple strips of overlapping images (see also: gigapixel images). These camera stations should be located with a base-distance ratio between 1:1 and 1:4. While two camera stations should suffice, a third station in the center can be useful in case problems arise with look-angle or occlusion (this occurs when part of the subject is blocked by another part of itself and cannot be viewed from both camera stations). This photogrammetric model is useful for long-range subjects such as cliff faces, or large boulders that are inaccesable to strip photogrammetry or would not be sufficiently modeled by a convergent pair.
Figure 1: Fan model with two camera stations (Birch, J. 2007) Figure 2: 3D view in 3DM Calibcam of fan model
1 - 7
Photo of signed drawing.
Depth map of same
Detail of depth map. Notice the figure riding the left camel, added at a greater depth and carved through the body of the camel.
Detail of signature, notice the erasure of the third character, which is more legible here than in the original photograph.
Photo of WHPS 348.
Depth map of WHPS 348.
Depth map of WHPS 07-0013. Notice the figure on the right: his left arm is carved into his body.
Once a model is built, an OBJ file can be exported from 3DM Analyst. This is a mesh that can be opened in MeshLab. A depth map is an effective way of discerning details too difficult to see in a normal photograph. This technique works well on relatively flat surfaces. Details such as varied depths between overlaid carvings become clearly visible.
1 - 1
Photo of WHPS 454
3D model of WHPS 454: half shown with ambient occlusion filter applied, half without
Once a model is built, a PTS file can be exported from 3DM Analyst. This is a point that can be opened in Cloud Compare. A point cloud can be processed in a number of ways, as it is the raw data that comes from building a photogrammetric model. A particularly useful tool is the PCV Ambient Occlusion filter in cloud compare. This filter helps to clarify inscriptions that have been erased or heavily weathered.