Evaluation of Surveying Drone Mapping Accuracy Based on Ground Control Point Distribution

In civil engineering, Digital Elevation Models (DEMs) and orthophoto images serve as fundamental materials for project design and execution. Traditional acquisition methods often prove inefficient, whereas surveying UAV technology offers a transformative solution. This study evaluates how Ground Control Point (GCP) quantity and spatial distribution influence surveying drone mapping accuracy.

Methodology

Our 13.74-hectare study area in southeastern China features diverse topography (mean slope: 5.75%, elevation range: 210–266 m MSL). We deployed an octocopter surveying drone equipped with a Nikon D3100 DSLR (16 mm focal length, 14.2 MP resolution) capturing 180 images at 90% forward and 80% lateral overlap. Ground Sampling Distance (GSD) was 3.7 cm. Seventy-two A3-sized targets served as GCPs/Checkpoints (CPs), surveyed via GNSS for ground truth.

Five GCP distribution patterns were tested (Figure 2):

  1. Edge: Perimeter placement
  2. Center: Concentrated in central zone
  3. Corner: Positioned at vertices
  4. Stratified: Uniform layered distribution
  5. Random: Scattered arbitrarily

For each pattern, 12 GCP quantities (4, 5, 6, 7, 8, 9, 12, 16, 20, 24, 30, 36) underwent five repetitions, generating 300 photogrammetric projects processed via Agisoft PhotoScan Professional using Structure-from-Motion (SfM).

Accuracy Metrics

Horizontal (XY) and vertical (Z) errors were quantified using Root Mean Square Error (RMSE) between drone-derived coordinates and GNSS-measured CPs:

$$ RMSE_X = \sqrt{\frac{\sum_{i=1}^{n}(X_{O_i} – X_{{GNSS}_i})^2}{n}} $$
$$ RMSE_Y = \sqrt{\frac{\sum_{i=1}^{n}(Y_{O_i} – Y_{{GNSS}_i})^2}{n}} $$
$$ RMSE_{XY} = \sqrt{\frac{\sum_{i=1}^{n}\left[(X_{O_i} – X_{{GNSS}_i})^2 + (Y_{O_i} – Y_{{GNSS}_i})^2\right]}{n}} $$
$$ RMSE_Z = \sqrt{\frac{\sum_{i=1}^{n}(Z_{O_i} – Z_{{GNSS}_i})^2}{n}} $$

where \(n\) = number of CPs, \(X_{O_i}\), \(Y_{O_i}\), \(Z_{O_i}\) = drone-measured coordinates, and \(X_{{GNSS}_i}\), \(Y_{{GNSS}_i}\), \(Z_{{GNSS}_i}\) = GNSS-surveyed coordinates.

Point cloud accuracy was assessed using CloudCompare’s M3C2 algorithm, comparing drone-generated clouds against reference data.

Results

Reference Project (RP) using all 72 GCPs achieved \(RMSE_{XY}\) = 0.023 m and \(RMSE_Z\) = 0.038 m. Varying GCP distributions yielded distinct accuracy profiles:

Table 1: Mean \(RMSE_{XY}\) (m) for GCP distribution patterns
GCP Quantity Edge Center Corner Stratified Random
4 0.100 0.120 0.150 0.140 0.160
8 0.060 0.090 0.110 0.080 0.100
16 0.042 0.070 0.080 0.048 0.065
24 0.038 0.062 0.070 0.044 0.048
36 0.035 0.060 0.060 0.042 0.055
Table 2: Mean \(RMSE_Z\) (m) for GCP distribution patterns
GCP Quantity Edge Center Corner Stratified Random
4 0.085 0.065 0.090 0.055 0.080
8 0.070 0.055 0.075 0.045 0.065
16 0.055 0.048 0.060 0.040 0.050
24 0.048 0.042 0.055 0.038 0.045
36 0.042 0.040 0.048 0.036 0.042

Edge distribution minimized horizontal error (\(RMSE_{XY}\) = 0.035 m with 36 GCPs), while stratified distribution optimized vertical accuracy (\(RMSE_Z\) = 0.036 m). Random GCP placement consistently underperformed, requiring ≥24 GCPs to match the precision of structured distributions with fewer points. M3C2 analysis confirmed these trends, showing edge-placed GCPs reduced planimetric drift by 41% compared to corner distributions.

Discussion

Surveying drone accuracy is highly sensitive to GCP geometry. Perimeter placement counters systemic error accumulation in bundle adjustment, explaining its \(RMSE_{XY}\) superiority. For elevation, the surveying UAV benefits from internal GCPs in stratified arrangements to model terrain variability. The relationship between GCP density and accuracy follows a negative logarithmic curve:

$$ RMSE = a \cdot \ln(N_{GCP}) + b $$

where \(a\) and \(b\) are distribution-specific coefficients. Beyond 20 GCPs (≈1.5 GCPs/ha), marginal gains diminished by >60%.

Conclusion

Optimizing surveying UAV mapping requires strategic GCP deployment: edge distributions for planimetric precision and stratified patterns for elevation accuracy. A density threshold exists where additional GCPs yield negligible improvements. Project planners should prioritize perimeter control points supplemented by internal targets for 3D robustness. This methodology enhances efficiency in surveying drone operations without compromising accuracy.

Scroll to Top