In complex mining environments, traditional drone technology aerial surveys rely heavily on a large number of ground control points, leading to low operational efficiency and significant safety risks. This is especially problematic in high-relief, multi-slope terrains where deploying control points is difficult, costly, and often yields low accuracy. To address these challenges, our study proposes and implements an innovative image-free UAV aerial survey technology. Using the DJI M300 RTK unmanned aerial vehicle platform, integrated with high-precision PPK (Post-Processed Kinematic) and RTK (Real-Time Kinematic) modules, we directly obtain precise instantaneous spatial coordinates (POS data) of each aerial photograph through the collaborative work of the airborne GNSS system and a ground reference station. In the data processing stage, these high-precision POS data serve as strong constraints, combined with multi-view oblique photography technology, to perform block adjustment and aerial triangulation calculations. This enables high-precision application of image-free drone technology in modeling complex mine landforms.
Methodology
Fundamental Principles of Image-free UAV Surveying
Traditional drone technology requires ground control points to link images to the ground. The image-free approach eliminates this dependency by leveraging high-precision onboard GNSS/IMU systems. The core principle is based on the collinearity equation, expressed as:
$$
x – x_0 = -f \frac{a_1(X_A – X_S) + b_1(Y_A – Y_S) + c_1(Z_A – Z_S)}{a_3(X_A – X_S) + b_3(Y_A – Y_S) + c_3(Z_A – Z_S)}
$$
$$
y – y_0 = -f \frac{a_2(X_A – X_S) + b_2(Y_A – Y_S) + c_2(Z_A – Z_S)}{a_3(X_A – X_S) + b_3(Y_A – Y_S) + c_3(Z_A – Z_S)}
$$
Where (x, y) are image coordinates, (x0, y0) are principal point coordinates, f is focal length, (X, Y, Z) are ground coordinates, (Xs, Ys, Zs) are camera station coordinates, and (a1, b1, c1, a2, b2, c2, a3, b3, c3) are elements of the rotation matrix derived from three attitude angles. By using high-precision POS data from the drone’s RTK/PPK module, the need for ground control points is eliminated.
Technical Workflow
The image-free drone technology workflow consists of the following steps:
- Set up a GNSS base station near the survey area and record continuous observations.
- The drone flies on a pre-designed route; the onboard RTK module receives satellite and base station differential signals in real time, recording precise POS information (longitude, latitude, altitude, and attitude angles) at the moment of each exposure.
- After the flight, import the acquired images and high-precision POS data into processing software (e.g., DJI Terra, ContextCapture).
- Use the high-precision POS data as strong constraints to perform block adjustment and aerial triangulation, generating accurate 3D models and digital products.
This paradigm shift from “human running” to “data running” significantly improves field efficiency and avoids safety risks in hazardous areas such as steep cliffs or deep pits.
Application Case Study
Study Area Description
The study area is located in Hebei Province, characterized by hilly terrain with elevations ranging from 549 m to 1326.2 m, a maximum relative height difference of 777.2 m, and complex landforms. Large-scale open-pit mining has created deep pits, high steep slopes, waste rock piles, and tailings ponds, making traditional survey methods challenging and costly.
Data Acquisition
We used the DJI M300 RTK drone equipped with a D2-PSDK five-lens oblique camera. Key flight parameters are listed in the table below.
| Parameter | Value |
|---|---|
| Flight height (relative to takeoff point) | 450 m |
| Ground resolution | 7.7 cm |
| Forward overlap | 80% |
| Side overlap | 70% |
| Number of sorties | 9 |
| Camera focal length (nadir) | 22 mm |
| Camera focal length (oblique) | 35 mm |
| Total pixel count | ≥120 million |
Data Processing and Products
After flight, images and POS data were imported into DJI Terra software. Aerial triangulation was performed to generate dense point clouds, meshes, and textures. The final products included a digital orthophoto map (DOM), digital elevation model (DEM), digital surface model (DSM), and a 3D reality model.
Results and Accuracy Assessment
Accuracy of Image-free UAV Surveying
To validate accuracy, 18 checkpoints were distributed across the study area and measured using RTK GNSS. The errors between UAV measurements and ground truth were calculated. The aerial triangulation quality report is summarized in the following table.
| Metric | Value |
|---|---|
| Number of tie points | 4,113,929 |
| Reprojection error (mean) | 0.054 m |
| Reprojection RMSE | 0.077 m |
| Distance RMSE | 0.089 m |
The root mean square error (RMSE) for checkpoints was calculated using:
$$
RMSE_{planar} = \sqrt{\frac{\sum(\Delta X^2 + \Delta Y^2)}{n}}
$$
$$
RMSE_{vertical} = \sqrt{\frac{\sum(\Delta Z^2)}{n}}
$$
Results showed a planar RMSE of ±6.5 cm and a vertical RMSE of ±16.5 cm. These values meet the 1:1000 scale topographic mapping standard (GB 50026-2020). The elevation errors are shown in the following table.
| Checkpoint ID | Elevation from DEM (m) | Elevation from RTK (m) | Error (m) |
|---|---|---|---|
| 1 | 765.12 | 765.25 | -0.13 |
| 2 | 810.05 | 810.18 | -0.13 |
| 3 | 802.00 | 802.20 | -0.20 |
| 4 | 779.20 | 779.35 | -0.15 |
| 5 | 757.25 | 757.40 | -0.15 |
| 6 | 761.40 | 761.55 | -0.15 |
| 7 | 760.70 | 760.85 | -0.15 |
| 8 | 688.35 | 688.50 | -0.15 |
| 9 | 730.65 | 730.80 | -0.15 |
| 10 | 710.05 | 710.20 | -0.15 |
| 11 | 752.60 | 752.75 | -0.15 |
| 12 | 749.75 | 749.90 | -0.15 |
| 13 | 709.55 | 709.70 | -0.15 |
| 14 | 764.40 | 764.55 | -0.15 |
| 15 | 809.00 | 809.15 | -0.15 |
| 16 | 801.95 | 801.90 | 0.05 |
| 17 | 779.15 | 779.05 | 0.10 |
| 18 | 757.20 | 757.05 | 0.15 |
Analysis of Mine Landforms
Using the generated DOM and 3D model, we identified and quantified three types of landform features: slopes, stockpiles, and pits. The following sections detail the analysis.
Slope Analysis
Thirteen slope faces were identified. Slope geometric parameters were extracted from the 3D model and are summarized below.
| Slope ID | Toe Elevation (m) | Crest Elevation (m) | Height Difference (m) | Slope Length (m) |
|---|---|---|---|---|
| 1 | 764.12 | 878.58 | 114.46 | 643.74 |
| 2 | 809.95 | 882.32 | 72.37 | 268.86 |
| 3 | 801.93 | 867.35 | 65.42 | 188.54 |
| 4 | 779.10 | 848.18 | 69.08 | 146.74 |
| 5 | 757.15 | 830.54 | 73.39 | 302.95 |
| 6 | 761.33 | 801.47 | 40.14 | 112.81 |
| 7 | 760.63 | 870.70 | 110.07 | 294.82 |
| 8 | 688.29 | 780.47 | 92.18 | 366.50 |
| 9 | 730.58 | 831.24 | 100.66 | 182.17 |
| 10 | 709.98 | 881.24 | 171.26 | 410.34 |
| 11 | 752.57 | 865.02 | 112.45 | 323.89 |
| 12 | 749.69 | 874.17 | 124.48 | 284.83 |
| 13 | 709.51 | 806.45 | 96.94 | 173.29 |
Slope 10 had the greatest height difference (171.26 m), and Slope 1 had the longest length (643.74 m). Slope aspect and steepness were also classified, as shown in the next table.
| Slope ID | Slope Angle (°) | Steepness Class | Aspect (°) | Aspect Category |
|---|---|---|---|---|
| 1 | 19 | Low steep | 127 | Semi-shady |
| 2 | 21 | Low steep | 84 | Semi-shady |
| 3 | 26 | Medium steep | 50 | Semi-shady |
| 4 | 23 | Medium steep | 91 | Semi-shady |
| 5 | 19 | Low steep | 28 | Shady |
| 6 | 33 | Medium steep | 123 | Semi-sunny |
| 7 | 26 | Medium steep | 13 | Shady |
| 8 | 25 | Medium steep | 140 | Semi-sunny |
| 9 | 32 | Medium steep | 53 | Semi-shady |
| 10 | 31 | Medium steep | 79 | Semi-shady |
| 11 | 31 | Medium steep | 36 | Shady |
| 12 | 31 | Medium steep | 95 | Semi-shady |
| 13 | 34 | Steep | 31 | Shady |
By examining the 3D model, we identified a potential landslide feature on Slope 12. The slope body exhibited fragmentation, loose material, and a debris fan at the toe, indicating instability. This drone technology enabled early warning of such hazards.
Stockpile Analysis
Nine stockpiles were identified, all artificial. Using filtered point clouds and Global Mapper software, we calculated geometric parameters including volume, area, and height.
| Stockpile ID | Minimum Elevation (m) | Maximum Elevation (m) | Height (m) | Base Area (m²) | Perimeter (m) | Volume (m³) |
|---|---|---|---|---|---|---|
| 1 | 764.36 | 883.49 | 119.13 | 250,055.62 | 2,206.95 | 22,440,045.72 |
| 2 | 692.48 | 785.92 | 93.44 | 113,145.23 | 1,697.29 | 7,723,492.10 |
| 3 | 807.58 | 886.86 | 79.28 | 123,991.12 | 1,546.31 | 6,996,485.54 |
| 4 | 802.33 | 876.25 | 73.92 | 80,456.20 | 1,435.43 | 4,223,608.17 |
| 5 | 782.45 | 849.15 | 66.70 | 49,255.61 | 1,197.07 | 2,221,125.88 |
| 6 | 756.94 | 837.06 | 80.12 | 239,206.44 | 3,021.48 | 15,887,280.89 |
| 7 | 735.96 | 813.10 | 77.14 | 37,539.16 | 1,014.43 | 2,066,551.99 |
| 8 | 781.57 | 692.76 | 88.81 | 112,074.96 | 1,726.31 | 7,639,721.26 |
| 9 | 803.78 | 762.62 | 41.16 | 40,553.11 | 911.46 | 1,391,465.21 |
Mine Pit Analysis
Three mine pits were identified. For each pit, we extracted the boundary contour from the DEM and calculated area, perimeter, and fill volume (volume needed to fill the pit to original topography).
| Pit ID | Perimeter (m) | Area (m²) | Fill Volume (m³) |
|---|---|---|---|
| I | 2,187.68 | 258,686.48 | 8,549,693.54 |
| II | 3,567.76 | 237,186.70 | 1,216,343.46 |
| III | 1,576.05 | 74,276.31 | 440,736.98 |
Discussion
The results demonstrate that image-free drone technology using high-precision RTK/PPK modules can achieve centimeter-level accuracy in complex mine terrain, meeting 1:1000 mapping standards. Compared to traditional manned aerial surveys or ground-based methods, this approach reduces field time by 60-70% and eliminates the need for dangerous ground control point deployment on steep slopes.
The combination of multi-view oblique photography with drone technology provides comprehensive 3D information, enabling detailed analysis of slope geometry, stockpile volumes, and pit dimensions. The successful identification of a potential landslide on Slope 12 illustrates the value of this technique for geohazard monitoring.
Future work will focus on improving image matching algorithms to further enhance modeling accuracy, especially for deformation monitoring of stockpiles and slopes. Integration of drone technology with LiDAR and InSAR could provide multi-scale monitoring capabilities for mining areas.
Conclusion
Our study validates that image-free drone technology, based on a DJI M300 RTK platform and five-lens oblique camera, is a reliable, efficient, and safe method for 3D reconstruction of complex mine landforms. The achieved planar RMSE of ±6.5 cm and vertical RMSE of ±16.5 cm satisfy rigorous engineering requirements. The technology enables systematic extraction of geometric parameters of slopes, stockpiles, and pits, and facilitates early identification of geohazards such as landslides. This paradigm shift from ground control point dependency to direct POS-based georeferencing substantially reduces cost and risk, making drone technology a cornerstone for modern mine surveying, resource management, and ecological restoration.