Photogrammetry 101
A free interactive course on drone and airborne photogrammetry: flight plan, GSD, overlap, ground control, point cloud, orthophoto and volumes, on a 3D model of a mine.
Every lesson is practised on a 3D model of a mining operation: move a control and the scene and the figures change. The same content, as text:
The course in words
-
Lesson 1
Measuring with photos
A drone flies over the ground in parallel lines and takes a photo every few metres.
Photogrammetry takes measurements of the ground from photographs. One photo on its own is flat. When the same point shows in several photos taken from different places, its position in three dimensions can be computed.
That is why the flight is planned: parallel lines, constant height and photos that overlap. The point cloud, the terrain model and the orthophoto all come out of that set of photos.
- The drone flies itself: it follows a plan loaded before take-off
- The camera points at the ground and shoots at fixed intervals
- Every point of the ground ends up in several photos
-
Lesson 2
The size of a pixel on the ground
The higher the camera flies, the more ground fits in each photo and the less detail each pixel keeps.
A photo is a grid of pixels. The GSD (ground sample distance) is the side of ground one pixel covers. With a 3 cm GSD, anything smaller than 3 cm cannot be told apart. It depends on the flight height, the focal length of the lens and the size of a pixel on the sensor.
The same formula holds for a drone and for a crewed aircraft. The drone flies low and gives the detail of a pit or a stockpile. The aircraft usually flies at 1,000 m or more with a medium or large-format metric camera: the pixel is larger, but each photo covers a kilometre and thousands of hectares are surveyed in a day. Global Mapping flies both, as an air operator certified by the DGAC since 2013.
- The GSD sets the finest detail that can be told apart
- Twice the height is twice the GSD and four times the area per photo
- A drone for one site and centimetres of detail; an aircraft for a concession, a corridor or a basin
-
Lesson 3
Photos that overlap
To rebuild the ground, every point has to show in several photos. Overlap is what makes that happen.
Forward overlap is how much one photo covers the next one on the same line. Side overlap is how much one line covers its neighbour. Both are given as a percentage of the photo.
With little overlap some areas are seen by one or two photos: one photo gives no height; two give a height, but with no check. With a lot of overlap the flight gets longer, and so does the processing. In mining the usual figures are 75 % forward and 65 % side, and more on high walls.
- A point needs at least two photos to have a height; with five or more the result is stable
- More overlap means more photos, more battery and more hours of processing
- High ground is closer to the camera and gets less overlap
-
Lesson 4
From two photos to a 3D point
One photo gives a direction. Two photos taken from different places give a point: where their rays cross.
Each pixel of a photo is a ray leaving the camera. With one photo you know which way the point lies, not how far. If the same point shows in another photo taken a few metres further on, the two rays cross, and that crossing is its position in 3D.
The elevation depends on the angle of the crossing: the base between photos against the height above the point (B/H). With a short base the rays arrive almost parallel, and half a pixel of error moves the elevation a lot. The position in plan barely depends on the base.
- A point is computed where the rays of two or more photos cross
- Parallax, the shift of the point between photos, gives the distance to the camera
- With a small B/H the elevation weakens; the position in plan barely changes
-
Lesson 5
Tying the model to the ground
The photos give the shape of the ground. Where that shape sits on the Earth depends on how the camera positions are known and on the ground control points.
The photo block builds a model with nearly the right shape, but places it with the coordinates of each photo. The drone's navigation GPS is off by metres: the model comes out shifted and tilted, and nothing restrains the doming left by the camera calibration. With RTK or PPK the camera position is known to a few centimetres and the model lands close.
Ground control points are targets measured with GNSS that go into the computation to tie the model down. Check points are targets the computation does not use: they measure the real error. Spread the control points over the whole site; bunched on one side, the error grows away from them, above all without RTK.
- Navigation GPS: metres of error. RTK or PPK: centimetres
- A ground control point goes into the computation; a check point judges it
- With RTK and no control, watch the elevation: an error in the focal length raises or lowers the whole model
-
Lesson 6
Millions of measured points
Once the cameras are oriented, the software measures almost every pixel of the photos and returns millions of points with a position and a colour.
First the software looks for the same detail in several photos: a corner, the crest of a bench, a tyre track on the road. These are the tie points. From them it works out where each camera was and where it pointed for every shot.
Then it matches almost every pixel between neighbouring photos and triangulates its position: this is the dense cloud. It holds millions of points, each with X, Y, Z and the colour from the photo. The terrain model, the orthophoto and the volumes all come from this cloud.
- One point every 2 to 4 photo pixels: with a GSD of 2.7 cm that is from tens to hundreds of points per m²
- Each point stores X, Y, Z and a colour; later it gets a class: ground, structure, vehicle or vegetation
- What the software cannot match is not measured: water, roofs with glare and thin objects leave gaps
-
Lesson 7
Surface model and terrain model
Two surfaces come out of the cloud: one with everything the camera saw, and one with the bare ground.
The point cloud becomes surfaces. The digital surface model (DSM) is everything the camera saw: ground, roofs, trucks, tree crowns. The digital terrain model (DTM) is the ground alone: the points are classified and whatever stands on top is removed.
Each one has its own use. Volumes, drainage and the design of roads and slopes are computed on the DTM. Sight lines, obstacles and the height of a structure are measured on the DSM. A stockpile is ground: it stays in both models.
- DSM: the ground and everything on it. DTM: the bare ground
- Contour lines are drawn on the DTM; on the DSM every building turns into a knot of contours
- Under a building the DTM is not measured, it is interpolated. Under dense vegetation we fly LiDAR
-
Lesson 8
A photo you can measure on
In an aerial photo, tall things lean towards the edges. The orthophoto puts every point in its place, seen from straight above at one scale.
An aerial photo is a perspective view: every ray passes through the lens. Anything with height leans away from the nadir, the point right under the camera, and the scale changes with the relief. You cannot measure on a photo like that as you would on a plan.
The orthophoto corrects every pixel with the digital surface model (DSM) and projects it vertically, at one scale. It is a mosaic: each area comes from the photo that sees it most vertically. On it you measure distances and areas as on a plan; heights come from the model.
- The top of a tall object lands away from its base: farther the taller it is and the farther it stands from the nadir
- At the nadir there is no displacement: high overlap lets you pick the most vertical photo for each area
- On the orthophoto you measure in plan; heights come from the DSM, volumes from the DTM
-
Lesson 9
From the model to the figure
The volume of a stockpile comes from the model: you add up, cell by cell, the height above a base.
The terrain model (DTM) gives the figures the mine uses every week: the volume of a stockpile, contour lines, a section. For the volume you set a base and a grid; each cell adds its height above the base times its area, and the sum is the volume.
The figure depends on the model, on the base you choose and, to get tonnes, on the density, which is measured separately and is usually the largest uncertainty. With this you can read a photogrammetry deliverable: where each number comes from and how far to trust it.
- Volume = sum of prisms: height above the base times the cell area
- Contour lines come from the same model: every 1 m, with an index contour every 5 m
- The base and the density move the figure more than the flight does