OpenDroneMap is an open source photogrammetry toolkit to process aerial imagery (usually from a drone) into maps and 3D models.^[3][4][5] The software is hosted and distributed freely on GitHub.^[6]

OpenDroneMap has been integrated within American Red Cross's in-field Portable OpenStreetMap system.^[7]

Overview

OpenDroneMap can be controlled either from a command-line interface or through a web interface (WebODM). It is recommended to run OpenDroneMap via Docker.^[8]

Under the hood, OpenDroneMap uses OpenSfM, CMVS and other libraries to perform the specific tasks in its workflow. Before processing the images, it can lower their resolution in order to save computational resources. OpenDroneMap uses the OpenSfM library to detect and match features, create tracks and determine their 3D positions along with the positions of the cameras. Then it uses either the OpenMVS or CMVS library to generate a dense point cloud from which it generates meshes. After that, the OpenDroneMap uses the Geospatial Data Abstraction Library and the Point Data Abstraction Library for orthomosaic generation and georeferencing.^[4]

OpenDroneMap can also process the aerial videos by cutting them into still images.^[8]

Performance

OpenDroneMap supports parallel computing and can utilize GPUs. It has a split-merge feature for computers with small amount of RAM in cases where the dataset is too large to fit in the memory, but it significantly reduces the performance. The official recommendation is to use 128 GB of memory to process 2500 images. Even though increasing the number of CPU cores generally reduces the processing time, it was determined that the optimal number of cores for large datasets is 20, and there is little to no performance gain beyond 20 cores.^[8]

References