LiDAR vs photogrammetry: which is better for a survey?

A LiDAR measures distance. It fires a pulse of laser light, times its return, and multiplies by the speed of light: one point, exactly, for every pulse that comes back, hundreds of thousands of times a second. A camera measures nothing; it records what the light looked like, and the measuring is done afterwards by comparing pictures. That one difference explains almost everything that follows: what each sees, what each costs, and where each belongs.

What each one sees

The animation at the top is the whole argument for LiDAR in one picture. Where a laser pulse finds a gap in a canopy, it reaches the soil and comes back; the sensor records several returns from one pulse, the first off the leaves, the last off the ground. Classify the last returns as terrain and a bare-earth model of a wooded site falls out. A camera sees the surface of the leaves and nothing beneath, and no processing can recover what no photograph contains.

The same holds for anything thin: a power line, a fence wire, a handrail. A laser pulse that hits it returns a point; a camera’s matcher needs texture on both sides of an edge and a wire has none, so it vanishes from the model. And a laser carries its own light, so it works at night and in the shade of a pit wall where a photograph would be mud.

The camera’s side of the argument is colour and density. Every point in a photogrammetric cloud is the colour of the ground it came from, and there are as many points as there are pixels, which at a 2 cm pixel is far more than any drone LiDAR returns. Edges, markings, cracks and the texture of a stockpile are in the model because they are in the pictures. The orthomosaic, which is the product most people actually look at, exists only because there were photographs.

LiDAR sees

Ground under canopy, wires, dark and shaded ground, the same at night

LiDAR misses

Colour; glass and water (the pulse goes through or away); fine texture between its points

Camera sees

Everything with texture, in colour, at pixel density; edges and markings

Camera misses

Anything under leaves; blank, shiny and transparent surfaces; thin wires; the dark

Accuracy

On open ground the two are closer than the marketing suggests. A current drone LiDAR with RTK positioning is quoted at around 5 cm vertical and 10 cm horizontal; a 45 MP camera on the same drone at 3 cm horizontal and 5 cm vertical without targets, and a centimetre or two with them. The ASPRS standard, which both are tested against, no longer ties accuracy to the sensor at all: it asks for the error against independent check points, and either method can be made to meet a class.

2–5 cm

Drone LiDAR vertical, with RTK and control

3–10 cm

Drone photogrammetry with GCPs, as commonly quoted

1–2 cm

Drone photogrammetry with RTK and GCPs, in published checks

0 points

Ground returns from a camera under dense canopy

The important difference is in the shape of the error. LiDAR’s error is mostly in positioning the sensor (GNSS, IMU and the calibration between them) and is roughly the same everywhere; it does not depend on what the ground looks like. Photogrammetry’s error depends on texture and geometry: excellent on a gravel pad with targets, poor on a bare concrete slab, and systematically bent across a long nadir-only corridor. The accuracy page goes through what decides it.

Cost and effort

What each produces

Which to use

For how the sensors themselves work, from tripod scanners to drone LiDAR, the laser scanners page; for photogrammetry against close-range scanners, structured light and handheld devices, the 3D scanning page. For a site’s photographs turned into the surface and the orthomosaic without a workstation, Clarity’s photogrammetry processing.