Collecting Your Own Aerial Data

An exciting possibility: increasingly affordable hardware and sophisticated software should make it available for anyone to collect high-resolution georegistered aerial imagery.  Beyond just viewing the result, with the use of photogrammetry and ground control points, the imagery could provide a base for collecting all the 2D and 3D features that are the last-mile problem in modeling and visualizing the whole world.

UAVs

• There are a wide range of UAVs that can capture imagery, from toys at the low-end to military drone surveying at the high end.
• French company Pict'Earth gave a good presentation in 2008, A NeoGeographical Approach to Aerial Image Acquisition and Processing, which explains how UAVs opens a way to do something new and important: Site Specific Base Maps.
• Popular platforms include RC helicopters (gas and electric), Quadrotors, and regular fixed-wing (and flying wing) planes, generally of a modest size.  There are degrees of control from manual to complete autonomy. Even the cheapest models now often have a high degree of self-stabilization (attitude control).
• A frequently used term is "Ready To Fly" (RTF).
• DIY Drones has a substantial community around the open "ArduCopter" platform.
  • The 3DR drones (e.g. Iris) start at around $600; with battery and camera it is >$1000.
  • ArduCopter: what do I need? nicely explains the popular components.
• The Parrot AR Drone ($360 and up) seems to be the popular low-end quadcopter
  • It sends a 720p live video stream, but is really meant for remote viewing rather than surveying, because the camera is wide-angle and points forward, not down.  It also uses Wifi for control, so you can only go ~150' above the operator. Hobbyists have reportedly made replacement boards for the Parrot which lets you use RC instead of Wifi (for farther range), add GPS, etc.
• SteadiDrone/a> (South Africa) sells RTF systems from around $2,000 to $20,000 including large copters that can lift heavy DSLR cameras.
• DJI (China) makes RTF UAVs from the  quad Phantom ($680)
  •  sold in the US through dealers like quadrocopter.com and uavproducts.com
  
• Falcon UAV sells fixed-wing UAV systems for $13,000 to $20,000
• Sensefly swinglet CAM (Switzerland) is a another fixed-wing
  • The eBee (~$10,000) came out in 2013.
• At the high end, the GateWing X-100 is a supposedly turnkey solution, but it costs around $100,000.

Kites and Balloons

• Some of what a UAV does can also be done with Balloon and Kite Mapping, which has a long history; some of the very first aerials were taken from kites over a hundred years ago.
• A balloon can be very stable and much less  expensive than an RC aircraft.  A basic Balloon Mapping Kit is only $95.
• Paper: Unmanned Small-Format Aerial Photography from Kites for Acquiring Large-Scale, High-Resolution, Multiview-Angle Imagery (2002)

Cameras

• There are very few consumer-priced cameras that can be set to take a high-quality still picture at regular intervals.
• As of 2013,  there are basically only 4 high-megapixel fixed-lens cameras which can take pictures continuously (commonly called time-lapse, or more obscurely as "intervalometer"):
  1. Pentax Optio WG-2 (and WG-3), 16 MP, 192 g, $205 on amazon.
  2. Panasonic Lumix DMC-TS5, 16.1 MP, 214 g, $360 on amazon.
  3. Nikon Coolpix P510, 16 MP, (41.7× zoom!), 555 g, $370 on amazon.
  4. Ricoh GR, 16.2 megapixels, 245 g, $800 on amazon.
• As of 2014, another reasonably affordable high-quality option is the Sony A6000 (24.3 MP, ~$650) which doesn't have an intervalometer, but can be remotely triggered.
  • an example of this usage with a UAV is QuestUAV's Visible Sensor (a Sony A6000)
• An older option: certain Canon cameras can be hacked with CHDK (see Unleash your camera’s inner Hulk with a free hack to the firmware)
• Smartphone camera
  • In theory, any smartphone has a programmable camera which can be set to take regular full-res photos with an app like Tina Time-lapse.
  • In practice, a phone dangling from a moving aerial platform has a very low chance of focusing properly.
  • June 2013: Aerial photos with UAV and an Android phone: first results

Processing the Imagery

• To make an orthophoto (for visualization purposes), the video or still frames from the UAV generally need to be stitched together (mosaicked) in a process similar to that used for panoramas.  You can also do photogrammetry to produce a DEM from the images.  Some approaches require knowing the location the camera took the images from to some degree of precision, other approaches don't require that information.
• Another (newer, more general) term for this process is "structure from motion" (SfM)
• Ground Control Points (GCPs)
  • Regardless of what hardware or software you use, to georeference / georectify, you will need GCPs, which are identifiable points on the ground with a known position.  The accuracy of your GCPs determines the accuracy of anything tied to them, so it's important to do them well.
  • Permanent GCP's are sometimes maintained by government or local authorities.
  • If you have any clearly identifiable objects in the image (such as buildings) for which you do know their location, then those can serve as GCPs.
  • You can't just use a consumer GPS; see Data Collection page for more about that.
• Producing an Orthophoto
  • You could, in theory, simply stich images yourself in any image editing software, then rectify to match GCPs with software such as Global Mapper ($400).  This is similar to the process for working with traditional aerial photos.
  • The PLOTS MapKnitter is a free opens-source online tool for semi-automatically stitching images, in part by overlaying them on an existing image layer (such as Google Maps).
  • Also in the open-source world, GDAL has recently incorporated a SOC project to do image correlation (one small step in the image processing chain)
• Producing a DEM
  • LISA Geo-Software (starting at €5000, €820 educational)
    • Appears to be powerful and complex photogrammetry software, needing a good amount of training to use. They list "UAV photogrammetry" as a field of application.
  • How to create a Digital Elevation Model from Photosynth point clouds (2010)
    • A detailed (and hacky) 25-step process moving the data through Photosynth, PLY, Meshlab, text editing, XYZ, ScanView, Excel, Graticule 3D (JAG3D), Excel again, then ArcGIS, Spatial Analyst, then writing a DEM.  Wow.
  • Accuracy and cost-effectiveness of Unmanned Aerial Vehicle derived DEM for localized areas (2012)
    • A very good presentation by Jurjen van der Sluijs, comparing the above (partly free) Photosynth approach and the (commercial) LISA software with a traditional DEM survey.
    • Conclusion: Photosynth is OK but the whole resulting DEM tends to be curved. LISA worked better.
• Producing both an Orthophoto and a DEM
  • DroneMapper (per project)
    • It is "Software as a Service", you upload your images to their servers and they process them.
    • Pricing is very reasonable "$20 USD per km sq with original imagery at 10-12 mega pixel generally", slightly higher for higher resolution.
    • Products: "Geo-referenced DEM, DSM, Ortho and Point Cloud (low/high res)"
  • PhotoModeler Scanner ($2600)
    • Seems mainly intended to extract colored point clouds from photos of single objects or small scenes, but some of the geology examples (like "Open Pit Mine via Micro-drone") show it working on small areas of terrain.
  • Agisoft PhotoScan ($3500 professional edition / $500 educational)
    • claims "fully automated workflow enables a non-specialist to process" large datasets
  • Pix4D Pix4UAV ($?)
    • Available either as a Service, or as desktop software.  The Service has a free trial available.
    • "2D processing starting at 60 euros for 0.5 km2 at 5 cm/pixel"
• DIYDrones thread: UAV for photogrammetry / surveys
  • Recommendations: use a high (>18) MPixel camera, with fixed focal length, fixed exposure, you must calibrate it, and beware barrel distortion.
  • Agisoft Lens is free lens calibration software.
• A book on this subject is Small-Format Aerial Photography (2010, hardcover only, $86 on Amazon)
• The Open Source Photogrammetry blog mostly deals with point clouds of objects, not of terrain.
• MeshLab is open-source software for working with points clouds, converting them to meshes.
  • See: Meshing Point Clouds for an example of working with terrain.