Low altitude drones have become indispensable tools for efficient geospatial data acquisition. However, conventional route planning methods often neglect the critical influence of terrain morphology and surface features, leading to incomplete coverage, repeated surveys, and operational inefficiencies. To address this, we propose an intelligent algorithm that dynamically adapts flight paths based on integrated terrain and feature analysis. This approach minimizes resurvey needs and significantly enhances data collection efficiency for complex landscapes.
Our methodology comprises three core components: intelligent terrain-feature classification, holistic route planning, and supplementary planning for specific targets. Initially, we partition the survey area into grid cells and analyze digital elevation models (DEMs) and orthoimagery. Terrain classification uses slope thresholds:
$$ \text{Terrain Type} =
\begin{cases}
\text{Flat:} & \text{slope} < 3^\circ \\
\text{Hillock:} & \Delta H < 100\ \text{m} \\
\text{Hill:} & H < 400\ \text{m}, \Delta H > 100–200\ \text{m} \\
\text{Low Mountain:} & H = 400–800\ \text{m}, \Delta H > 200\ \text{m} \\
\text{High Mountain:} & H > 800\ \text{m}, \Delta H > 200\ \text{m}
\end{cases} $$
Surface features like forests and settlements are identified using Faster-RCNN deep learning models trained on annotated datasets. Table 1 summarizes classification parameters:
| Feature Type | Analysis Metric | Operational Impact |
|---|---|---|
| Forest (Dense/Sparse) | Canopy coverage %, Tree height | Requires LiDAR; influences flight altitude |
| Residential Area | Building density, Height variation | Needs oblique imaging; increases overlap |
| Water Body | Surface reflectivity | Triggers reflectance calibration |
For holistic planning, flat terrain employs standard grid patterns with fixed overlap. Mountainous regions use terrain-following paths with crosshatch patterns to maintain consistent ground sampling distance. The flight altitude $h$ adjusts dynamically to elevation $z$ and required resolution $\delta$:
$$ h = \frac{f \cdot GSD}{\delta} + z $$
where $f$ is focal length and $GSD$ is ground sampling distance. Overlap parameters are intensified for complex features:
$$ \text{Along-track overlap} = 85\%\ (\text{vs. 65\% baseline}), \quad \text{Cross-track overlap} = 75\% $$
Specific targets receive supplementary planning. High-rise structures use multi-altitude layered paths with electronic fences. Linear features like roads follow adaptive snake patterns. For residential zones, dense grid routes combine nadir and oblique angles to capture facade details. The energy consumption $E$ for added paths is optimized by:
$$ E = \sum_{i=1}^{n} \left( k_v v_i^2 \cdot d_i + k_h \Delta h_i \right) $$
where $v_i$ is velocity, $d_i$ is segment distance, $\Delta h_i$ is elevation change, and $k_v$, $k_h$ are drone-specific constants.
We validated the approach on a 1.68 km² test site (elevation range: 100 m) using a 35mm camera-equipped low altitude UAV. Key parameters included 10 m/s speed and 1s capture interval. Table 2 contrasts performance:
| Metric | Traditional Method | Proposed Algorithm | Improvement |
|---|---|---|---|
| Images Collected | 1,052 | 1,591 | 51.3% ↑ |
| Resurveys Required | 325 images (60 min) | 0 | 100% ↓ |
| Point Cloud Density (pts/m²) | 32.9M | 50.9M | 54.7% ↑ |
| Geometric Error (pixels) | 0.665 | 0.322 | 51.6% ↓ |
| Total Time (min) | 190 | 175 | 7.9% ↓ |
For residential areas, increasing overlap to 85% doubled image count and reduced modeling error by 52%. In mountainous zones, terrain-following paths enhanced point density by 48% compared to fixed-altitude flights. Target-specific planning, such as multi-layer flights around high-rise buildings, eliminated model gaps without additional surveys. The low altitude UAV’s adaptability proved critical for capturing nuanced features across varying topographies.
This intelligence-driven framework fundamentally transforms low altitude drone operations. By synthesizing terrain morphology with feature-specific planning, it eliminates redundant surveys and enhances data richness across diverse landscapes. Future work will integrate real-time weather adaptation and multi-sensor orchestration for further optimization.