Drone that generates a point cloud from photography

Question

I read this in a trade magazine today,

​“Esri has just released the Drone2Map for ArcGIS app, which transforms high-resolution drone imagery into ready-to-use aerial data in the GIS platform. Users can bring point-clouds, mosaic datasets, 3D meshes and orthomosaics directly into the platform in near real time without any third-party application. … Simple photographic data obtained from a miniature camera mounted on a drone can be used for point-cloud collection, rather than rely on airborne lidar.”

How does it generate a point-cloud from the photographic data without some sort of laser or other active sensor? The part in bold above.

Is it merely stereo imagery? https://doc.arcgis.com/en/drone2map/process-and-share/processing-options.htm

Answer 1

Probably they have made a structure from motion function. With the mosaic of several pictures around an object you can create its 3d model and get also the point cloud. Thats nothing too spectacular. Probably the calculations are made in the cloud and sent back to the device for the "near real time".

For example VirtualSFM is a extremly strong opensource tool where you also only input the picutres and you get the object out if you want to play around with such stuff.

The part where it becomes interesting is the accuracy. Without high precision positioning the output is not very nice. The orthomosaics and point clouds are looking just really rough. Professional companies combine 60-100 Megapixel cameras on medium sized UAV with high precision ground control points and get mosaics, 3d models or point clouds to the accuracy of 1-2cm in about 1-2 days for large buildings for example. But therefore quiete some processing time and some field work is necessary. Have seen those on the same conference as some projects that tried the same with normal <1000 € UAV where the outcome was really bad for the input in time and money. The point is not the software but the recording system.