"Joel and I both serve on the “Core Team” of the Humanitarian UAV Network (UAViators). It is in this context that we’ve been exploring ways to render aerial imagery more actionable for rapid disaster damage assessments and tactical decision making. To overcome some of the challenges around the consistent analysis of aerial imagery, Joel suggested we take a rank-order approach. His proposal is quite simple: display two geo-tagged aerial images side by side with the following question: “Which of the two images shows more disaster damage?” Each combination of images could be shown to multiple individuals. Images that are voted as depicting more damage would “graduate” to the next display stage and in turn be compared to each other, and so on and so forth along with those images voted as showing less damage.

In short, a dedicated algorithm would intelligently select the right combination of images to display side by side. The number and type of votes could be tabulated to compute reliability and confidence scores for the rankings. Each image would have a unique damage score which could potentially be used to identify thresholds for fully destroyed versus partially damaged versus largely intact infrastructure. Much of this could be done on MicroMappers or similar microtasking solutions. Such an approach would do away with the need for detailed imagery interpretation guides. As noted above, consistent analysis is difficult even when such guides are available. The rank-order approach could help quickly identify and map the most severely affected areas to prioritize tactical response efforts.  Note that this approach could be used with both crowd-sourced analysis and professional analysis. Note also that the GPS coordinates for each image would not be made publicly available for data privacy reasons.

Is this strategy worth pursuing? What are we missing? Joel and I would be keen to get some feedback. So please feel free to use the comments section below to share your thoughts or to send an email here."

Full Post Here.

E-mail me when people leave their comments –

You need to be a member of diydrones to add comments!

Join diydrones


  • @Patrick: No problem - anytime! 

  • And thanks Rosalie for cross-posting!

    Original blog post:


  • Thanks Grant and Chris

    @Grant -

    1) Yes, I do mean algorithm that can intelligently select which follow up images to display.
    2) Images are of the same disaster. The immediate need is to assess disaster damage, rather than identify what type of infrastructure is damaged.
    3) Yes, we need more people power, hence inviting the 3,000+ digital volunteers over at MicroMappers who have participated in past disasters. See also the Digital Humanitarian Network, which I co-founded with the UN back in 2012.
    4) Yes, UAVs come into this because they capture the imagery. See also:


    @Chris -

    1) Both phases
    2) Incorrect, we have already crowdsourced the analysis of aerial imagery to provide rapid disaster damage assessment for the World Bank following Cyclone Pam. UK and Swedish Search and Rescue teams have also expressed strong interest in using MicroMappers and crowdsourcing for S&R efforts.
    3) We are returning to Nepal in March to do large-scale mapping via UAV as part of the recovery and reconstruction efforts.

    For context, my background: http://iRevolutions.org/bio

    Thanks again for your feedback, Grant and Chris, very much appreciated.

  • Sorry second question was about using the untrained persons.  I think on the whole it will work, but it will depend on the number of people setting eyes on the image.  How do you plan to improve data interpretation?  I can see it working if a review system is used, you'd have to give the consumers of the information a way to say the data was accurate or not.  How would you integrate that feedback into the system?


  • Two questions, at which part of the operation are you planning this for?

    The response phase or the recovery phase?

    On the response side I see little value as you wont be able to get the data to first responders quickly enough, plus the trust level of the information will be low.

    On the later response phase (property damage and access issues - ie: not life threatening) and the recovery phase, the issue I see is who is analysing those images, I've been to many disaster sites in Australia and to the the untrained observer something that looks okay can be far more severe that first meets the eye.  Having said that from a rapid damage assessment perspective the data would be useful.  For example, For a cyclone I had to use 6 teams to check a large section of a town, this cost us almost a whole day to gather the information and the night operations team then spend the night sorting and prioritizing for my teams to action the next morning.  So yes it might has saved us a day if the drones were deployed before we arrived (if it was done at the same time, we'd have to analyse the data still and it would be of little gain).  Also in the later response phase and then the recovery phase we do not typically deal with the most damaged places first, we will restore access and then we may try for the light to moderate damage areas to restore as many people back to normal - less people in shelters is a good thing.  Having said all that if we had rapid availability of the damage from the air it could have speed up our operation (ie: no analysis work already done), particularly if we could have been reviewing it as we were heading to the location (but then again we didn't know our exact destination till we hit the ground).


  • Developer

    I have no expertise in this area however I have a couple of "layperson" comments/questions.

    - I don't think you mean algorithm but process right?  You want to create a new process for analysing images by  humans - not with computers?

    - The images your analysing side by side - are they off the same area and your trying to decide which image is better or are you analysing images of different areas and comparing damage in each.  If its the latter I'm going to guess that the amount of damage done to an area is only one factor to be considered?  For example if you have a barn that has been completely destroyed but a hospital that only took partial damage the hospital still rates higher on the "important image" scale?  I guess if the images are standardised in some way i.e. your comparing the damage in images of whole villages of similar sizes it could work in the sense you can figure out which village appears to be damaged more.

    - Your going to increase the amount of person power required to analsye images if you need multiple people to look at the same image.

    Drones come into this as they are taking the images?

    Thanks, Grant.

  • Moderator

    I don't know how applicable this is, but I have been developing a low cost ($500) thermal camera solution for drones based on the SEEK Thermal camera.   The value in a low cost camera is that many can be deployed for Disaster Relief / Assessment.   We're currently beta testing and looking for new testers.  

    Here's a short video describing the camera in more detail. 

This reply was deleted.