Using the example of the meshes of a photogrammetry project the dimensional accuracy of the drone mapping is evaluated. Concerning to the Pix4D technical documentation the accuracy of the coordinates of a point is 1 – 3 GSD (Ground Sampling Distance) under good conditions. The maximal inaccuracy of a distance should be ±6 GSD, because in the worst case the inaccuracy of the coordinates of the endpoints is its maxiumum (3 GSD), is in the direction of the segment and in opposite oriantation. In the best case the inaccuracies of the endpoints are similar in value, direction and orientation. In that case the inaccuracies compensate each other and distance is exact. The influence of that component of the inaccuracy, which is orthogonal to the segment, decreases on increasing distance. So the mean absolute inaccuracy decreases on increasing distance. Because the maximum absolute inaccuracy of ±6 GSD is constant, the relative inaccuracy of the distance decreases on increasing distance anyhow.
The computational printout of the Pix4D software reports a average Ground Sampling Distance of 0.56 cm. So the maximum inaccuracy should be 3.36 cm. But additionally the resolution of the in-place-measurements as well as of the point cloud is only 1 cm, which corresponds with an accuracy of ±0.5 cm. This results in an additional uncertainty of ±1.35 cm. So all things considered there is a possible difference between in-place-measurements and point cloud of up to ±4.71 cm. But this does not represent the (smaller) inaccuracy of drone-mapping itself. And you have to keep in mind, that even though we used an official verified measuring tape there is a tolerance of ±0.23 % in the measuring tape itself.
| Mesh | Segment | Measured Length | Length Pix4D | Deviation Pix4D | Length Photoscan | Deviation Photoscan |
|---|---|---|---|---|---|---|
| 1 | 1 | 6.30 m | 6.28 m | -2.1 cm / -0.33 % | 6.29 m | -1.0 cm / -0.16 % |
| 2 | 2.95 m | 2.99 m | +3.8 cm / +1.29 % | 2.96 m | +0.5 cm / +0.18 % | |
| 3 | 6.68 m | 6.65 m | -3.3 cm / -0.50 % | 6.72 m | +4.0 cm / 0.18 % | |
| 4 | 2.79 m | 2.78 m | -0.5 cm / -0.18 % | 2.78 m | -1.0 cm / -0.35 % | |
| 5 | 7.16 m | 7.13 m | -2.7 cm / -0.37 % | 7.18 m | +2.3 cm / +0.32 % | |
| 6 (not drawn) | 7.01 m | 7.02 m | +0.7 cm / +0.10 % | 7.03 m | +1.8 cm / +0.32 % | |
| 2 | 1 | 5.64 m | 5.61 m | -3.4 cm / -0.61 % | 5.63 m | -0.7 cm / -0.12 % |
| 2 | 2.79 m | 2.78 m | -0.5 cm / -0.18 % | 2.78 m | -1.0 cm / -0.35 % | |
| 3 | 5.30 m | 5.27 m | -3.2 cm / -0.61 % | 5.29 m | -1.0 cm / -0.20 % | |
| 4 | 2.86 m | 2.84 m | -1.7 cm / -0.59 % | 2.88 m | +2.0 cm / +0.71 % | |
| 5 | 6.13 m | 6.07 m | -5.7 cm / -0.93 % | 6.10 m | -3.0 cm / -0.49 % | |
| 6 (not drawn) | 6.17 m | 6.17 m | -0.4 cm / -0.07 % | 6.20 m | +2.7 cm / +0.43 % | |
| 3 | 1 | 5.77 m | 5.77 m | ±0.0 cm / ±0.00 % | 5.76 m | -0.9 cm / -0.16 % |
| 2 | 2.86 m | 2.84 m | -1.7 cm / -0.59 % | 2.88 m | +2.0 cm / +0.71 % | |
| 3 | 6.24 m | 6.21 m | -2.7 cm / -0.43 % | 6.23 m | -1.3 cm / -0.20 % | |
| 4 | 3.02 m | 3.02 m | -0.1 cm / -0.03 % | 3.02 m | -0.4 cm / -0.14 % | |
| 5 | 6.60 m | 6.58 m | -0.2 cm / -0.31 % | 6.59 m | -0.5 cm / -0.08 % | |
| 6 (not drawn) | 6.76 m | 6.75 m | -0.9 cm / -0.13 % | 6.76 m | -0.4 cm / -0.06 % | |
| 4 | 1 | 5.79 m | 5.74 m | -4.7 cm / -0.81 % | 5.77 m | -1.8 cm / -0.31 % |
| 2 | 3.02 m | 3.02 m | -0.1 cm / -0.03 % | 3.02 m | -0.4 cm / -0.14 % | |
| 3 | 5.45 m | 5.41 m | -3.9 cm / -0.71 % | 5.41 m | -4.0 cm / -0.74 % | |
| 4 | 2.93 m | 2.91 m | -1.6 cm / -0.55 % | 2.91 m | -1.6 cm / -0.55 % | |
| 5 | 6.27 m | 6.24 m | -3.4 cm / -0.55 % | 6.25 m | -2.4 cm / -0.38 % | |
| 6 (not drawn) | 6.43 m | 6.39 m | -3.7 cm / -0.58 % | 6.40 m | -2.6 cm / -0.40 % | |
| 5 | 1 | 6.24 m | 6.21 m | -2.7 cm / -0.43 % | 6.23 m | -1.3 cm / -0.20 % |
| 2 | 3.73 m | 3.71 m | -1.5 cm / -0.41 % | 3.69 m | -3.9 cm / -1.04 % | |
| 3 | 1.85 m | 1.82 m | -3.0 cm / -1.60 % | 1.85 m | +0.3 cm / +0.15 % | |
| 4 | 6.04 m | 6.02 m | -2.2 cm / -0.36 % | 6.03 m | -0.6 cm / -0.10 % | |
| 5 | 7.26 m | 7.22 m | -4.1 cm / -0.57 % | 7.26 m | -0.3 cm / -0.04 % | |
| 6 (not drawn) | 4.49 m | 4.47 m | -1.5 cm / -0.34 % | 4.46 m | -2.8 cm / -0.62 % | |
| 6 | 1 | 5.30 m | 5.27 m | -3.2 cm / -0.61 % | 5.29 m | -1.0 cm / -0.20 % |
| 2 | 5.04 m | 5.01 m | -3.1 cm / -0.62 % | 5.03 m | -1.1 cm / -0.21 % | |
| 3 | 3.08 m | 3.10 m | +1.8 cm / +0.57 % | 3.07 m | -1.1 cm / -0.37 % | |
| 4 | 3.73 m | 3.71 m | -1.5 cm / -0.41 % | 3.69 m | -3.9 cm / -1.04 % | |
| 5 | 5.38 m | 5.37 m | -0.6 cm / -0.11 % | 5.35 m | -3.1 cm / -0.58 % | |
| 6 (not drawn) | 6.47 m | 6.45 m | -1.5 cm / -0.24 % | 6.46 m | -0.9 cm / -0.13 % |
The average inaccuracy of the Pix4D point cloud is -0.35 % without any manual rescaling. Using Photoscan a manual rescaling is essential, because Photoscan is not able to do a proper scaling of the point cloud without geo reference points. With manual rescaling related to the longest measured segment (segment 5 of mesh 5), which principally leads to the smallest effect of inaccuracies due to the reading, the average inaccuracy of the Photoscan point cloud is -0.16 %. An additonal – but unnecessary – manual rescaling shifts the average inaccuracy of the Pix4D point cloud to +0.22 %. This does not lead to a change in quality at all.
So as a conclusion the accuracy of drone mapping point clouds is regularly on the level of careful measurements with a measuring tape and better than measurements with a metering wheel.
(the underlying measurements have been made by J. and D. Bärwald during their students internship)