odm-dji

Python 3 scripts for processing DJI drone imagery with OpenDroneMap, including multispectral processing and vegetation index calculation.

Features

• Lens calibration injection - Auto-detects DJI camera model from EXIF and injects correct distortion coefficients
• Rolling shutter correction - Applies model-specific readout times
• Multispectral processing - Process DJI Mavic 3M and other multispectral drones
• Vegetation indices - Calculate NDVI, NDRE, GNDVI, SAVI, EVI, and more
• 3D visualization - Generate interactive 3D HTML maps and animations
• Publication plots - Create print-ready maps with scale bars and north arrows

Measured improvement on DJI Phantom 3 (FC330), ultra quality:

Setup	Reprojection error
No correction (ODM default)	1.321 px
Rolling shutter + lens calibration	0.731 px

---

Requirements

Tool	Install
Python 3.8+	Pre-installed on Ubuntu
exiftool	sudo apt install libimage-exiftool-perl
Docker	docs.docker.com/engine/install
opendronemap/odm:fixed	See Build the Docker image

Python Dependencies

pip install numpy rasterio matplotlib scipy plotly

Build the Docker image

The :fixed tag patches a gdal_array incompatibility in ODM v3.5.6 that prevents PDF report generation:

docker build -t opendronemap/odm:fixed \
  -f docker/Dockerfile.fix \
  docker/

---

Usage

Publication Plots

Generate publication-quality maps from ODM outputs:

python3 publication_plots.py \
  --dem /path/to/dsm.tif \
  --rgb /path/to/ortho.tif \
  --output /path/to/plots/ \
  --indices /path/to/indices/ \
  --offset 0 \
  --title "Your Site Name"

Options:

• --dem - Path to DEM/DSM GeoTIFF
• --rgb - Path to RGB orthophoto GeoTIFF
• --output - Output directory for plots
• --indices - Directory containing vegetation index GeoTIFFs (optional)
• --offset - Elevation offset in meters (default: 0)
• --title - Title for plots (optional)

Outputs: dem_contours.png, rgb_orthophoto.png, 3d_perspective.png, ndvi.png, ndre.png, gndvi.png, ndwi.png

3D Interactive Visualization

Create an interactive 3D HTML map with RGB texture:

python3 dem_3d_rgb_mesh.py \
  --dem /path/to/dsm.tif \
  --rgb /path/to/ortho.tif \
  --output /path/to/3d_map.html \
  --downsample 4 \
  --exaggeration 1.0

Options:

• --dem - Path to DEM/DSM GeoTIFF
• --rgb - Path to RGB orthophoto GeoTIFF
• --output - Output HTML file path
• --downsample - Downsample factor (default: 4, lower = higher quality)
• --exaggeration - Vertical exaggeration (default: 1.0 = no exaggeration)

Vegetation Indices

Calculate vegetation indices from multispectral bands:

from vegetation_indices import calculate_all_indices, save_index_geotiff

bands = {
    'red': red_array,
    'green': green_array,
    'nir': nir_array,
    'rededge': rededge_array
}

indices = calculate_all_indices(bands)
# Returns: ndvi, ndre, gndvi, savi, ndwi, evi, evi2, mcari, ndvire, ccci

---

Supported Cameras

Camera model is auto-detected from EXIF. Known models receive pre-loaded lens calibration (Brown radial distortion k1/k2) and the correct rolling shutter readout time.

EXIF Model	Drone	k1	k2	Readout
FC330	DJI Phantom 3	-0.270	0.090	16 ms
FC6310	DJI Phantom 4 Pro	-0.098	0.010	29 ms
FC220	DJI Mavic Pro	-0.154	0.025	32 ms
FC7203	DJI Mini 2	-0.143	0.028	33 ms
FC3582	DJI Mini 3 Pro	-0.120	0.021	30 ms
FC3411	DJI Mini 4 Pro	-0.110	0.018	28 ms
FC3170	DJI Air 2S	-0.095	0.015	25 ms
FC2103	DJI Mavic Air 2	-0.131	0.022	30 ms
FC8282	DJI Neo	-0.158	0.030	30 ms
FC2105	DJI Mavic 3 Multispectral	-0.105	0.020	28 ms

Unknown cameras fall back to rolling shutter only (k1=k2=0, readout=30 ms), which is always safer than ODM's default of no rolling shutter correction at all.

See CONTRIBUTING.md to add a new model.

---

Vegetation Indices

The following indices are calculated from multispectral bands:

Index	Name	Bands Used
NDVI	Normalized Difference Vegetation Index	NIR, Red
NDRE	Normalized Difference Red Edge	NIR, RedEdge
GNDVI	Green NDVI	NIR, Green
SAVI	Soil Adjusted Vegetation Index	NIR, Red
NDWI	Normalized Difference Water Index	Green, NIR
EVI	Enhanced Vegetation Index	NIR, Red
EVI2	Two-band EVI	NIR, Red
MCARI	Modified Chlorophyll Absorption Index	RedEdge, Red, Green
NDVIre	Red Edge NDVI	NIR, RedEdge
CCCI	Canopy Chlorophyll Content Index	NDRE, NDVI

---

Outputs

After processing, typical output structure:

<project_name>/
├── odm_orthophoto/
│   └── odm_orthophoto.tif    # Georeferenced orthophoto
├── odm_dem/
│   ├── dsm.tif               # Digital Surface Model
│   └── dtm.tif               # Digital Terrain Model
├── indices/                   # Vegetation indices (if processed)
│   ├── ndvi.tif
│   ├── ndre.tif
│   └── ...
└── plots/                     # Publication plots (if generated)
    ├── dem_contours.png
    ├── rgb_orthophoto.png
    └── ...

---

How It Works

1. Lens calibration injection

ODM accepts an initial cameras.json via --cameras. The scripts write the known Brown distortion coefficients (k1, k2) for each DJI model and let ODM self-calibrate from that starting point rather than from zero.

2. Rolling shutter correction

All consumer DJI cameras use CMOS sensors that expose line-by-line rather than all at once. At flight speed this creates a "jello" distortion. The scripts pass --rolling-shutter --rolling-shutter-readout <N> with the correct readout time for each model.

3. Multispectral processing

For multispectral drones like the DJI Mavic 3M, the pipeline extracts individual bands (Red, Green, NIR, RedEdge) from the multispectral orthophoto and calculates vegetation indices for analysis.

---

License

MIT - see LICENSE.