Near Real-Time UAV Imagery Processing

Contents 1 Near-Real Time UAV Imagery Processing 1.1 Summary 1.2 Use Case 1.3 Description 1.4 Resources 1.5 Additional Notes

[edit] Near-Real Time UAV Imagery Processing

(testing available in conjunction with STAR-Tides/Camp Roberts Field Simulation)

• Discussion thread about this problem

• Technical notes (includes notes from CrisisCommons notes page on this task)

[edit] Summary

There is a strong need for near real-time, geo-referenced imagery and geospatial information in the aftermath of natural disasters and humanitarian crises. One means of collection is to use Unmanned Aerial Systems (UAS). In recent years, the development of next generation UAV platforms has become a vibrant research field, particularly mini- and micro-UAVs. It is also expected that civilian and commercial UAS will become an indispensable data acquisition platform for a multitude of geomatic applications as soon as the technology matures and a legal framework for its deployment has been established.

A major hurdle preventing the widespread deployment of mini- and micro-UAVs for mapping purposes is their inability to provide geo-referenced high resolution imagery in near real time.

[edit] Use Case

Deployment of UAV’s over/around disaster or crisis areas for data capture, including but not limited to aerial imagery would lead to collection of unprocessed imagery/data – this app would live either on a field system or in the cloud. Data would be processed and made available in various system-friendly formats (kml, geotiff, other).

End users are likely to be:

• OpenStreetMappers before and during crises in areas which don't have high-resolution satellite imagery (e.g. they would like to map crisis-prone areas but are unlikely to get good satellite imagery until a crisis actually happens)
• Sahana Agasti and Sahana Eden users that wish to use the imagery as underlying layers for our Emergency Management Information Systems
• CrisisCommoners etc during crises, for building damage assessment, camp finding (in Haiti, we found camps by looking for areas covered with tarpaulins) etc

[edit] Description

An open source system for the efficient registration of numerous frames, likely through feature detection, bundle-block adjustment, or some combination there-of, will vastly improve the utility of UAV acquired image sets and make a significant step toward the operational deployment of mini- and micro-UAV systems for mapping purposes.

• Micro UAVs typically operate at low altitudes (i.e., <1000 meters) and carry small-medium format sensors
• This results in a very large number (i.e. 100s-1000s) of frames at very high resolution.
• The limited resolution of the on-board inertial monitoring units (IMU), combined with the relatively stringent registration requirements of imagery with spatial resolutions in the cm range, make direct projection insufficient for the production of quality mosaics.
• The large number of frames makes any human interpretation all but impossible.

[edit] Resources

• Turquilt Google Group - email list for ongoing coordination

[edit] Additional Notes

• ftp location for sample data from Camp Roberts experimentation to be provided (Can we get this soon? --Gav in Sydney)
• This application MUST be under fully open-source license and not encumbered by IP of the large corporations participating in the RHOK. Proprietary solutions will not be considered for actual fielding.
• Consider mapwarper (source on github)