Tag Archives: 3D

Assessing Disaster Damage: How Close Do You Need to Be?

“What is the optimal spatial resolution for the analysis of disaster damage?”

I posed this question during an hour-long presentation I gave to the World Bank in May 2015. The talk was on the use of remote sensing aerial robotics (UAVs) for disaster damage assessments; I used Cyclone Pam in Vanuatu and the double Nepal Earthquakes as case studies.

One advantage of aerial robotics over space robotics (satellites) is that the former can capture imagery at far higher spatial resolutions—sub 1-centimeter if need be. Without hesitation, a World Bank analyst in the conference room replied to my question: “Fifty centimeters.” I was taken aback by the rapid reply. Had I per chance missed something? Was it so obvious that 50 cm resolution was optimal? So I asked, “Why 50 centimeters?” Again, without hesitation: “Because that’s the resolution we’ve been using.” Ah ha! “But how do you know this is the optimal resolution if you haven’t tried 30 cm or even 10 cm?”

Lets go back to the fundamentals. We know that “rapid damage assessment is essential after disaster events, especially in densely built up urban areas where the assessment results provide guidance for rescue forces and other immediate relief efforts, as well as subsequent rehabilitation and reconstruction. Ground-based mapping is too slow, and typically hindered by disaster-related site access difficulties” (Gerke & Kerle 2011). Indeed, studies have shown that the inability of physically access damaged areas results in field teams underestimating the damage (Lemoine et al 2013). Hence one reason for the use of remote sensing.

We know that remote sensing can be used for two purposes following disasters. The first, “Rapid Mapping”, aims at providing impact assessments as quickly as possible after a disaster. The second, “Economic Mapping” assists in quantifying the economic impact of the damage. “Major distinctions between the two categories are timeliness, completeness and accuracies” (2013).  In addition, Rapid Mapping aims to identify the relative severity of the damage (low, medium, high) rather than absolute damage figures. Results from Economic Mapping are combined with other geospatial data (building size, building type, etc) and economic parameters (e.g., cost per unit area, relocation costs, etc) to compute a total cost estimate (2013). The Post-Disaster Needs Assessment (PDNA) is an example of Economic Mapping.

It is worth noting that a partially destroyed building may be seen as a complete economic loss, identical to a totally destroyed structure (2011). “From a casualty / fatality assessment perspective, however, a structure that is still partly standing […] offers a different survival potential” (2011).

Screen Shot 2016-02-07 at 10.31.27 AM

We also know that “damage detectability is primarily a function of image resolution” (2011). The Haiti Earthquake was “one of the first major disasters in which very high resolution satellite and airborne imagery was embraced to delineate the event impact (2013). It was also “the first time that the PDNA was based on damage assessment produced with remotely sensed data” (2013). Imagery analysis of the impact confirmed that a “moderate resolution increase from 41 cm to 15 cm has profound effects on damage mapping accuracy” (2011). Indeed, a number of validation studies carried out since 2010 have confirmed that “the higher detail airborne imagery performs much better [than lower resolution] satellite imagery” (2013). More specifically, the detection of very heavy damage and destruction in Haiti aerial imagery “is approximately a factor 8 greater than in the satellite imagery.”

Comparing the aerial imagery analysis with field surveys, Lemoine et al. find that “the number of heavily affected and damaged buildings in the aerial point set is slightly higher than that obtained from the field survey” (2013). The correlation between the results of the aerial imagery analysis and field surveys is sensitive to land-use (e.g., commercial, downtown, industrial, residential high density, shanty, etc). In highly populated areas such as shanty zones, “the over-estimation of building damage from aerial imagery could simply be a result of an incomplete field survey while, in downtown, where field surveys seem to have been conducted in a more systematic way, the damage assessment from aerial imagery matches very well the one obtained from the field” (2013).


In sum, the results from Haiti suggests that the “damage assessment from aerial imagery currently represents the best possible compromise between timeliness and accuracy” (2013). The Haiti case study also “showed that the damage derived from satellite imagery was underestimated by a factor of eight, compared to the damage derived from aerial imagery. These results suggest that despite the fast availability of very high resolution satellite imagery […], the spatial resolution of 50 cm is not sufficient for an accurate interpretation of building damage.”

In other words, even though “damage assessments depend very much on the timeliness of the aerial images acquisitions and requires a considerable effort of visual interpretation as an element of compromise; [aerial imagery] remains the best trade-of in terms of required quality and timeliness for producing detailed damage assessments over large affected areas compared to satellite based assessments (insufficient quality) and exhaustive field inventories (too slow).” But there’s a rub with respect to aerial imagery. While the above results do “show that the identification of building damage from aerial imagery […] provides a realistic estimate of the spatial pattern and intensity of damage,” the aerial imagery analysis still “suffers from several limitations due to the nadir imagery” (2013).

“Essentially all conventional airborne and spacebar image data are taken from a quasi-vertical perspective” (2011). Vertical (or nadir) imagery is particularly useful for a wide range of applications, for sure. But when it comes to damage mapping, vertical data have great limitations, particularly when concerning structural building damage (2011). While complete collapse can be readily identified using vertical imagery (e.g., disintegrated roof structures and associated high texture values, or adjacent rubble piles, etc), “lower levels of damage are much harder to map. This is because such damage effects are largely expressed along the façades, which are not visible in such imagery” (2011). According to Gerke and Kerle, aerial oblique imagery is more useful for disaster damage assessment than aerial or satellite imagery taken with a vertical angle. I elaborated on this point vis-a-vis a more recent study in this blog post.


Clearly, “much of the challenge in detecting damage stems from the complex nature of damage” (2011). For some types and sizes of damage, using only vertical (nadir) imagery will result in missed damage (2013). The question, therefore, is not simply one of spatial resolution but the angle at which the aerial or space-based image is taken, land-use and the type of damage that needs to be quantified. Still, we do have a partial answer to the first question. Technically, the optimal spatial resolution for disaster damage assessments is certainly not 50 cm since 15 cm proved far more useful in Haiti.

Of course, if higher-resolution imagery is not available in time (or at all), than clearly 50 cm imagery is infinitely more optimal than no imagery. In fact, even 5 meter imagery that is available within 24-48 hours of a disaster can add value if this imagery comes with baseline imagery, i.e., imagery of the location of interest before the disaster. Baseline data enables the use of automated change-detection algorithms that can provide a first estimate of damage severity and the location or scope of that severity. What’s more, these change-detection algorithms could automatically plot a series of waypoints to chart the flight plan of an autonomous aerial robot (UAV) to carry out closer inspection. In other words, satellite and aerial data can be complementary, and the drawbacks of low resolution imagery can be offset if said imagery is available at a higher temporal frequency.


On the flip side, just because the aerial imagery used in the Haiti study was captured at 15 cm resolution does not imply that 15 cm resolution is the most optimal spatial resolution for disaster damage. It could very well be 10 cm. This depends entirely on the statistical distribution of the size of damaged features (e.g, the size of a crack, a window, a tile, etc,), the local architecture (e.g., type of building materials used and so on) and the type of hazard (e.g, hurricane, earthquake, etc). That said, “individual indicators of destruction, such as a roof or façade damage, do not linearly add to a given damage class [e.g., low, medium, high]” (2011). This limitation is alas “fundamental in remote sensing where no assessment of internal structural integrity is possible” (2011).

In any event, one point is certain: there was no need to capture aerial imagery (both nadir and obliques) at 5 centimeter resolution during the World Bank’s humanitarian UAV mission after Cyclone Pam—at least not for the purposes of 2D imagery analysis. This leads me to one final point. As recently noted during my opening Keynote at the 2016 International Drones and Robotics for Good Awards in Dubai, disaster damage is a 3D phenomenon, not a 2D experience. “Buildings are three-dimensional, and even the most detailed view at only one of those dimensions is ill-suited to describe the status of such features” (2011). In other words, we need “additional perspectives to provide a more comprehensive view” (2011).

There are very few peer-reviewed scientific papers that evaluate the use of high-resolution 3D models for the purposes of damage assessments. This one is likely the most up-to-date study. An earlier research effort by Booth et al. found that the overall correlation between the results from field surveys and 3D analysis of disaster damage was “an encouraging 74 percent.” But this visual analysis was carried out manually, which could have introduced non-random errors. After all, “the subjectivity inherent in visual structure damage mapping is considerable” (2011). Could semi-automated methods for the analysis of 3D models thus yield a higher correlation? This is the research question posed by Gerke and Kerle in their 2011 study.

Screen Shot 2016-02-07 at 5.39.06 AM

The authors tested this question using aerial imagery from the Haiti Earthquake. When 3D features were removed from their automated analysis, “classification performance declined, for example by some 10 percent for façades, the class that benefited most from the 3D derivates” (2011). The researchers also found that trained imagery analysts only agreed at most 76% of the time in their visual interpretation and assessments of aerial data. This is in part due to the lack of standards for damage categories (2011).

Screen Shot 2016-02-07 at 5.40.36 AM

I made this point more bluntly in this earlier blog post. Just imagine when you have hundreds of professionals and/or digital volunteers analyzing imagery (e.g, through crowdsourcing) and no standardized categories of disaster damage to inform the consistent interpretation of said imagery. This collective subjectivity introduces a non-random error into the overall analysis. And because it is non-random, this error cannot be accounted for. In contrast, a more semi-automated solution would render this error more random, which means the overall model could potentially be adjusted accordingly.

Gerke and Kerle conclude that high quality 3D models are “in principle well-suited for comprehensive semi-automated damage mapping. In particular façades, which are critical [to the assessment process], can be assessed [when] multiple views are provided.” That said, the methodology used by the authors was “still essentially based on projecting 3D data into 2D space, with conceptual and geometric limitations. [As such], one goal should be to perform the actual damage assessment and classification in 3D.” This explains why I’ve been advocating for Virtual Reality (VR) based solutions as described here.


Geek and Kerle also “required extensive training data and substantial subjective evidence integration in the final damage class assessment. This raises the question to what extent rules could be formulated to create a damage ontology as the basis for per-building damage scoring.” This explains why I invited the Harvard Humanitarian Initiative (HHI) to create a damage ontology based on the aerial imagery from Cyclone Pam in Vanuatu. This ontology is based on 2D imagery, however. Ironically, very high spatial resolution 2D imagery can be more difficult to interpret than lower-res imagery since the high resolution imagery inevitably adds more “noise” to the data.

Ultimately, we’ll need to move on to 3D damage ontologies that can be visualized using VR headsets. 3D analysis is naturally more intuitive to us since we live in a mega-high resolution 3D world rather than a 2D one. As a result, I suspect there would be more agreement between different analysts studying dynamic, very high-resolution 3D models versus 2D static images at the same spatial res.

Screen Shot 2016-02-07 at 12.53.27 PM

Taiwan’s National Cheng Kung University created this 3D model from aerial imagery captured by UAV. This model was created and uploaded to Sketchfab on the same day the earthquake struck. Note that Sketchfab recently added a VR feature to their platform, which I tried out on this model. Simply open this page on your mobile device to view the disaster damage in Taiwan in VR. I must say it works rather well, and even seems to be higher resolution in VR mode compared to the 2D projections of the 3D model above. More on the Taiwan model here.

Screen Shot 2016-02-07 at 12.52.06 PM

But to be useful for disaster damage assessments, the VR headset would need to be combined with wearable technology that enables the end-user to digitally annotate (or draw on) the 3D models directly within the same VR environment. This would render the analysis process more intuitive while also producing 3D training data for the purposes of machine learning—and thus automated feature detection.

I’m still actively looking for a VR platform that will enable this, so please do get in touch if you know of any group, company, research institute, etc., that would be interested in piloting the 3D analysis of disaster damage from the Nepal Earthquake entirely within a VR solution. Thank you!

Using Aerial Robotics and Virtual Reality to Inspect Earthquake Damage in Taiwan (Updated)

The tragic 6.4 magnitude earthquake struck southern Taiwan shortly before 4 in the morning on Saturday, February 6th. Later in the day, aerial robots were used to capture areal videos and images of the disaster damage, like below.

Screen Shot 2016-02-07 at 1.25.26 PM

Within 10 hours of the earthquake, Dean Hosp at Taiwan’s National Cheng Kung University used screenshots of aerial videos posted on YouTube by various media outlets to create the 3D model below. As such, Dean used “second hand” data to create the model, which is why it is low resolution. Having the original imagery first hand would enable a far higher-res 3D model. Says Dean: “If I can fly myself, results can produce more fine and faster.”

Click the images below to enlarge.

Screen Shot 2016-02-07 at 1.32.46 PM

Screen Shot 2016-02-07 at 1.17.12 PM

Update: About 48 hours after the earthquake, Dean and team used their own UAV to create this much higher resolution version (see below), which they also annotated (click to enlarge).

Screen Shot 2016-02-10 at 6.45.04 AMScreen Shot 2016-02-10 at 6.47.33 AM Screen Shot 2016-02-10 at 6.48.19 AMScreen Shot 2016-02-10 at 6.51.23 AM

Here’s the embedded 3D model:

These 3D models were processed using AgiSoft PhotoScan and then uploaded to Sketchfab on the same day the earthquake struck. I’ve blogged about Sketchfab in the past—see this first-ever 3D model of a refugee camp, for example. A few weeks ago, Sketchfab added a Virtual Reality feature to their platform, so I just tried this out on the above model.

Screen Shot 2016-02-10 at 6.51.32 AM

The model appears equally crisp when viewed in VR mode on a mobile device (using Google Cardboard in my case). Simply open this page on your mobile device to view the disaster damage in VR. This works rather well; the model does seem to be of high resolution in Virtual Reality as well.


This is a good first step vis-a-vis VR applications. As a second step, we need to develop 3D disaster ontologies to ensure that imagery analysts actually interpret 3D models in the same way. As a third step, we need to combine VR headsets with wearable technology that enables the end-user to annotate (or draw on) the 3D models directly within the same VR environment. This would make the damage assessment process more intuitive while also producing 3D training data for the purposes of machine learning—and thus automated feature detection.

I’m still actively looking for a VR platform that will enable this, so please do get in touch if you know of any group, company, research institute, etc., that would be interested in piloting the 3D analysis of disaster damage from the Taiwan or Nepal Earthquakes entirely within a VR solution. Thank you.

Click here to view 360 aerial visual panoramas of the disaster damage.

Many thanks to Sebastien Hodapp for pointing me to the Taiwan model.

Video: Crisis Mapping Nepal with Aerial Robotics


I had the honor of spearheading this disaster recovery UAV mission in Nepal a few weeks ago as part of Kathmandu Flying Labs. I’ve been working on this new initiative (in my own time) with Kathmandu Living Labs (KLL), Kathmandu University (KU), DJI and Pix4D. This Flying Lab is the first of several local UAV innovation labs that I am setting up (in my personal capacity and during my holiday time) with friends and colleagues in disaster-prone countries around the world. The short film documentary above was launched just minutes ago by DJI and describes how we teamed up with local partners in Kathmandu to make use of aerial robotics (UAVs) to map Nepal’s recovery efforts.

Here are some of the 3D results, courtesy of Pix4D (click to enlarge):


Screen Shot 2015-11-02 at 5.17.58 PM

Why work in 3D? Because disaster damage is a 3D phenomenon. This newfound ability to work in 3D has important implications for Digital Humanitarians. To be sure, the analysis of these 3D models could potentially be crowdsourced and eventually analyzed entirely within a Virtual Reality environment.

Since most of our local partners in Nepal don’t have easy access to computers or VR headsets, I found another way to unlock and liberate this digital data by printing our high-resolution maps on large, rollable banners.



We brought these banner maps back to the local community and invited them to hack the map. How? Directly, by physically adding their local knowledge to the map; knowledge about the location of debris, temporary shelters, drinking water and lots more. We brought tape and color-coded paper with us to code this knowledge so that the community could annotate the map themselves.


In other words, we crowdsourced a crisis map of Panga, which was highly participatory. The result was a rich, contextual social layer on top of the base map, which further inform community discussions on strategies and priorities guiding their recovery efforts. For the first time ever, the community of Panga was working off the one and same dataset to inform their rebuilding. In short, our humanitarian mission combined aerial robotics, computer vision, water-proof banners, local knowledge, tape, paper and crowdsourcing to engage local communities on the reconstruction process.


I’m now spending my evenings & weekends working with friends and colleagues to plan a follow-up mission in early 2016. We’ll be returning to Kathmandu Flying Labs with new technology partners to train our local partners on how to use fixed-wing UAVs for large scale mapping efforts. In the meantime, we’re also exploring the possibility of co-creating Jakarta Flying Labs, Monrovia Flying Labs and Santiago Flying Labs in 2016.

I’m quitting my day job next week to devote myself full time to these efforts. Fact is, I’ve been using all of my free time (meaning evenings, weekends and many, many weeks of holiday time) to pursue my passion in aid robotics and to carry out volunteer-based UAV missions like the one in Nepal. I’ve also used holiday time (and my own savings) to travel across the globe to present this volunteer-work at high-profile events, such as the 2015 Web Summit here in Dublin where the DJI film documentary was just publicly launched.


My Nepali friends & I need your help to make sure that Kathmandu Flying Labs take-off and become a thriving and sustainable center of social entrepreneur-ship. To this end, we’re actively looking for both partners and sponsors to make all this happen, so please do get in touch if you share our vision. And if you’d like to learn more about how UAVs other emerging technologies are changing the face of humanitarian action, then check out my new book Digital Humanitarians.

In the meantime, big, big thanks to our Nepali partners and technology partners for making our good work in Kathmandu possible!

The First Ever 3D Model of a Refugee Camp Made with UAV Imagery

A colleague of mine just returned from overseas where he flew a UAV as part of an independent exploratory project. He did so with permission and also engaged directly with local communities in the process—as per the guidelines listed in the Humanitarian UAV Code of Conduct. He subsequently sent me this aerial video footage of a camp, which he recorded using a DJI Phantom 2 Vision+:

The analysis of aerial imagery for humanitarian & development purposes is an active area of research at UAViators. He thus kindly gave me permission to share this footage with colleague Matt Shroyer so that we could explore the possibility of creating a mosaic and 3D model from the video.

Screen Shot 2015-08-16 at 5.21.38 PM

Incidentally, the image below is the highest resolution and most recent satellite image available of the camp on Google Maps. As you can tell, the satellite image is very much out of date.

Screen Shot 2015-08-14 at 10.56.08 AM

And here is the mosaic, which Matt kindly produced by taking hundreds of screenshots of the aerial video footage (click to enlarge):

Screen Shot 2015-08-14 at 11.00.01 AM

A close up:

Screen Shot 2015-08-14 at 11.02.13 AM

We then explored the possibility of creating a 3D model of the camp using the screenshots and SketchFab. The results are displayed below (click to enlarge). The numbers are annotations we added to provide relevant information on the camp. Perhaps in the future we’ll be able to add photographs & videos (captured from hand-held cameras) and other types of data to the 3D model.

Screen Shot 2015-08-14 at 11.09.00 AM

Screen Shot 2015-08-14 at 11.11.18 AM

Screen Shot 2015-08-14 at 11.11.39 AM

It’s worth noting that this 3D model would be far higher resolution if the UAV had been flown with the expressed purpose of creating a 3D model. Either way, you’ll note that no individuals appear either in the mosaic or in the 3D model, which is important for data privacy and security.

Here are two short video fly-throughs of the 3D model:

You can also fly through the model yourself here.

The purpose of this visual exploration is to solicit feedback from humanitarian organizations vis-a-vis the potential added value that this imagery could provide for camp management and related humanitarian efforts. So please feel free to get in touch via email and/or to post comments below with your feedback. In the meantime, a big thanks to my colleague for sharing the aerial videos and equally big thanks to Matt for all his time on the imagery processing. UAViators will be carrying out additional projects like this one over the coming months. So if you’d like to get involved, please do get in touch.

3D Digital Humanitarians: The Irony

In 2009 I wrote this blog post entitled “The Biggest Problem with Crisis Maps.” The gist of the post: crises are dynamic over time and space but our crisis maps are 2D and static. More than half-a-decade later, Digital Humanitarians have still not escaped from Plato’s Cave. Instead, they continue tracing 2D shadows cast by crisis data projected on their 2D crisis maps. Is there value in breaking free from our 2D data chains? Yes. And the time will soon come when Digital Humanitarians will have to make a 3D run for it.

Screen Shot 2015-07-21 at 3.31.27 PM

Aerial imagery captured by UAVs (Unmanned Aerial Vehicles) can be used to create very high-resolution 3D point clouds like the one below. It only took a 4-minute UAV flight to capture the imagery for this point cloud. Of course, the processing time to convert the 2D imagery to 3D took longer. But solutions already exist to create 3D point clouds on the fly, and these solutions will only get more sophisticated over time.

Stitching 2D aerial imagery into larger “mosaics” is already standard practice in the UAV space. But that’s so 2014. What we need is the ability to stitch together 3D point clouds. In other words, I should be able to mesh my 3D point cloud of a given area with other point clouds that overlap spatially with mine. This would enable us to generate high-resolution 3D point clouds for larger areas. Lets call these accumulated point clouds Cumulus Clouds. We could then create baseline data in the form of Cumulus Clouds. And when a disaster happens, we could create updated Cumulus Clouds for the affected area and compare them with our baseline Cumulus Cloud for changes. In other words, instead of solely generating 2D mapping data for the Missing Maps Project, we could add Cumulus Clouds.

Meanwhile, breakthroughs in Virtual Reality will enable Digital Humanitarians to swarm through these Cumulus Clouds. Innovations such as Oculus Rift, the first consumer-targeted virtual reality headsets, may become the pièce de résistance of future Digital Humanitarians. This shift to 3D doesn’t mean that our methods for analyzing 2D crisis maps are obsolete when we leave Plato’s Cave. We simply need to extend our microtasking and crowdsourcing solutions to the 3D space. As such, a 3D “tasking manager” would just assign specific areas of a Cumulus Cloud to individual Digital Jedis. This is no different to how field-based disaster assessment surveys get carried out in the “Solid World” (Real Word). Our Oculus headsets would “simply” need to allow Digital Jedis to “annotate” or “trace various” sections of the Cumulus Clouds just like they already do with 2D maps; otherwise we’ll be nothing more than disaster tourists.

wp oculus

The shift to 3D is not without challenges. This shift necessarily increases visual complexity. Indeed, 2D images are a radical (and often welcome) simplification of the Solid World. This simplification comes with a number of advantages like reducing the signal to noise ratio. But 2D imagery, like satellite imagery, “hides” information, which is one reason why imagery-interpretation and analysis is difficult, often requiring expert training. But 3D is more intuitive; 3D is the world we live in. Interpreting signs of damage in 3D may thus be easier than doing so with a lot less information in 2D. Of course, this also depends on the level of detail required for the 3D damage assessments. Regardless, appropriate tutorials will need to be developed to guide the analysis of 3D point clouds and Cumulus Clouds. Wait a minute—shouldn’t existing assessment methodologies used for field-based surveys in the Solid World do the trick? After all, the “Real World” is in 3D last time I checked.

Ah, there’s the rub. Some of the existing methodologies developed by the UN and World Bank to assess disaster damage are largely dysfunctional. Take for example the formal definition of “partial damage” used by the Bank to carry out their post-disaster damage and needs assessments: “the classification used is to say that if a building is 40% damaged, it needs to be repaired. In my view this is too vague a description and not much help. When we say 40%, is it the volume of the building we are talking about or the structural components?” The question is posed by a World Bank colleague with 15+ years of experience. Since high-resolution 3D data enables more of us to more easily see more details, our assessment methodologies will necessarily need to become more detailed both for manual and automated analysis solutions. This does add more complexity but such is the price if we actually want reliable damage assessments regardless.

Isn’t it ironic that our shift to Virtual Reality may ultimately improve the methodologies (and thus data quality) of field-based surveys carried out in the Solid World? In any event, I can already “hear” the usual critics complaining; the usual theatrics of cave-bound humanitarians who eagerly dismiss any technology that appears after the radio (and maybe SMS). Such is life. Moving along. I’m exploring practical ways to annotate 3D point clouds here but if anyone has additional ideas, do please get in touch. I’m also looking for any solutions out there (imperfect ones are fine too) that can can help us build Cumulus Clouds—i.e., stitch overlapping 3D point clouds. Lastly, I’d love to know what it would take to annotate Cumulus Clouds via Virtual Reality. Thanks!

Acknowledgements: Thanks to colleagues from OpenAerialMap, Cadasta and MapBox for helping me think through some of the ideas above.

Assessing Disaster Damage from 3D Point Clouds

Humanitarian and development organizations like the United Nations and the World Bank typically carry out disaster damage and needs assessments following major disasters. The ultimate goal of these assessments is to measure the impact of disasters on the society, economy and environment of the affected country or region. This includes assessing the damage caused to building infrastructure, for example. These assessment surveys are generally carried out in person—that is, on foot and/or by driving around an affected area. This is a very time-consuming process with very variable results in terms of data quality. Can 3D (Point Clouds) derived from very high resolution aerial imagery captured by UAVs accelerate and improve the post-disaster damage assessment process? Yes, but a number of challenges related to methods, data & software need to be overcome first. Solving these challenges will require pro-active cross-disciplinary collaboration.

The following three-tiered scale is often used to classify infrastructure damage: “1) Completely destroyed buildings or those beyond repair; 2) Partially destroyed buildings with a possibility of repair; and 3) Unaffected buildings or those with only minor damage . By locating on a map all dwellings and buildings affected in accordance with the categories noted above, it is easy to visualize the areas hardest hit and thus requiring priority attention from authorities in producing more detailed studies and defining demolition and debris removal requirements” (UN Handbook). As one World Bank colleague confirmed in a recent email, “From the engineering standpoint, there are many definitions of the damage scales, but from years of working with structural engineers, I think the consensus is now to use a three-tier scale – destroyed, heavily damaged, and others (non-visible damage).”

That said, field-based surveys of disaster damage typically overlook damage caused to roofs since on-the-ground surveyors are bound by the laws of gravity. Hence the importance of satellite imagery. At the same time, however, “The primary problem is the vertical perspective of [satellite imagery, which] largely limits the building information to the roofs. This roof information is well suited for the identification of extreme damage states, that is completely destroyed structures or, to a lesser extent, undamaged buildings. However, damage is a complex 3-dimensional phenomenon,” which means that “important damage indicators expressed on building façades, such as cracks or inclined walls, are largely missed, preventing an effective assessment of intermediate damage states” (Fernandez Galaretta et al. 2014).

Screen Shot 2015-04-06 at 10.58.31 AM

This explains why “Oblique imagery [captured from UAVs] has been identified as more useful, though the multi-angle imagery also adds a new dimension of complexity” as we experienced first-hand during the World Bank’s UAV response to Cyclone Pam in Vanuatu (Ibid, 2014). Obtaining photogrammetric data for oblique images is particularly challenging. That is, identifying GPS coordinates for a given house pictured in an oblique photograph is virtually impossible to do automatically with the vast majority of UAV cameras. (Only specialist cameras using gimbal mounted systems can reportedly infer photogrammetric data in oblique aerial imagery, but even then it is unclear how accurate this inferred GPS data is). In any event, oblique data also “lead to challenges resulting from the multi-perspective nature of the data, such as how to create single damage scores when multiple façades are imaged” (Ibid, 2014).

To this end, my colleague Jorge Fernandez Galarreta and I are exploring the use of 3D (point clouds) to assess disaster damage. Multiple software solutions like Pix4D and PhotoScan can already be used to construct detailed point clouds from high-resolution 2D aerial imagery (nadir and oblique). “These exceed standard LiDAR point clouds in terms of detail, especially at façades, and provide a rich geometric environment that favors the identification of more subtle damage features, such as inclined walls, that otherwise would not be visible, and that in combination with detailed façade and roof imagery have not been studied yet” (Ibid, 2014).

Unlike oblique images, point clouds give surveyors a full 3D view of an urban area, allowing them to “fly through” and inspect each building up close and from all angles. One need no longer be physically onsite, nor limited to simply one façade or a strictly field-based view to determine whether a given building is partially damaged. But what does partially damaged even mean when this kind of high resolution 3D data becomes available? Take this recent note from a Bank colleague with 15+ years of experience in disaster damage assessments: “In the [Bank’s] official Post-Disaster Needs Assessment, the classification used is to say that if a building is 40% damaged, it needs to be repaired. In my view this is too vague a description and not much help. When we say 40%, is it the volume of the building we are talking about or the structural components?”

Screen Shot 2015-05-17 at 1.45.50 PM

In their recent study, Fernandez Galaretta et al. used point clouds to generate per-building damage scores based on a 5-tiered classification scale (D1-D5). They chose to compute these damage scores based on the following features: “cracks, holes, intersection of cracks with load-carrying elements and dislocated tiles.” They also selected non-damage related features: “façade, window, column and intact roof.” Their results suggest that the visual assessment of point clouds is very useful to identify the following disaster damage features: total collapse, collapsed roof, rubble piles, inclined façades and more subtle damage signatures that are difficult to recognize in more traditional BDA [Building Damage Assessment] approaches. The authors were thus able to compute a per building damage score, taking into account both “the overall structure of the building,” and the “aggregated information collected from each of the façades and roofs of the building to provide an individual per-building damage score.”

Fernandez Galaretta et al. also explore the possibility of automating this damage assessment process based on point clouds. Their conclusion: “More research is needed to extract automatically damage features from point clouds, combine those with spectral and pattern indicators of damage, and to couple this with engineering understanding of the significance of connected or occluded damage indictors for the overall structural integrity of a building.” That said, the authors note that this approach would “still suffer from the subjectivity that characterizes expert-based image analysis.”

Hence my interest in using crowdsourcing to analyze point clouds for disaster damage. Naturally, crowdsourcing alone will not eliminate subjectivity. In fact, having more people analyze point clouds may yield all kinds of disparate results. This is explains why a detailed and customized imagery interpretation guide is necessary; like this one, which was just released by my colleagues at the Harvard Humanitarian Initiative (HHI). This also explains why crowdsourcing platforms require quality-control mechanisms. One easy technique is triangulation: have ten different volunteers look at each point cloud and tag features in said cloud that show cracks, holes, intersection of cracks with load-carrying elements and dislocated tiles. Surely more eyes are better than two for tasks that require a good eye for detail.

Screen Shot 2015-05-17 at 1.49.59 PM

Next, identify which features have the most tags—this is the triangulation process. For example, if one area of a point cloud is tagged as a “crack” by 8 or more volunteers, chances are there really is a crack there. One can then count the total number of distinct areas tagged as cracks by 8 or more volunteers across the point cloud to calculate the total number of cracks per façade. Do the same with the other metrics (holes, dislocated titles, etc.), and you can compute a per building damage score based on overall consensus derived from hundreds of crowdsourced tags. Note that “tags’ can also be lines or polygons; meaning that individual cracks could be traced by volunteers, thus providing information on the approximate lengths/size of a crack. This variable could also be factored in the overall per-building damage score.

In sum, crowdsourcing could potentially overcome some of the data quality issues that have already marked field-based damage assessment surveys. In addition, crowdsourcing could potentially speed up the data analysis since professional imagery and GIS analysts tend to already be hugely busy in the aftermath of major disasters. Adding more data to their plate won’t help anyone. Crowdsourcing the analysis of 3D point clouds may thus be our best bet.

So why hasn’t this all been done yet? For several reasons. For one, creating very high-resolution point clouds requires more pictures and thus more UAV flights, which can be time consuming. Second, processing aerial imagery to construct point clouds can also take some time. Third, handling, sharing and hosting point clouds can be challenging given how large those files quickly get. Fourth, no software platform currently exists to crowdsource the annotation of point clouds as described above (particularly when it comes to the automated quality control mechanisms that are necessary to ensure data quality). Fifth, we need more robust imagery interpretation guides. Sixth, groups like the UN and the World Bank are still largely thinking in 2D rather than 3D. And those few who are considering 3D tend to approach this from a data visualization angle rather than using human and machine computing to analyze 3D data. Seventh, this area, point cloud analysis for 3D feature detection, is still a very new area of research. Many of the methodology questions that need answers have yet to be answered, which is why my team and I at QCRI are starting to explore this area from the perspective of computer vision and machine learning.

The holy grail? Combining crowdsourcing with machine learning for real-time feature detection of disaster damage in 3D point clouds rendered in real-time via airborne UAVs surveying a disaster site. So what is it going to take to get there? Well, first of all, UAVs are becoming more sophisticated; they’re flying faster and for longer and will increasingly be working in swarms. (In addition, many of the new micro-UAVs come with a “follow me” function, which could enable the easy and rapid collection of aerial imagery during field assessments). So the first challenge described above is temporary as are the second and third challenges since computer processing power is increasing, not decreasing, over time.

This leaves us with the software challenge and imagery guides. I’m already collaborate with HHI on the latter. As for the former, I’ve spoken with a number of colleagues to explore possible software solutions to crowdsource the tagging of point clouds. One idea is simply to extend MicroMappers. Another is to add simple annotation features to PLAS.io and PointCloudViz since these platforms are already designed to visualize and interact with point clouds. A third option is to use a 3D model platform like SketchFab, which already enables annotations. (Many thanks to colleague Matthew Schroyer for pointing me to SketchFab last week). I’ve since had a long call with SketchFab and am excited by the prospects of using this platform for simple point cloud annotation.

In fact, Matthew already used SketcFab to annotate a 3D model of Durbar Square neighborhood in downtown Kathmandu post-earthquake. He found an aerial video of the area, took multiple screenshots of this video, created a point cloud from these and then generated a 3D model which he annotated within SketchFab. This model, pictured below, would have been much higher resolution if he had the original footage or 2D images. Click pictures to enlarge.

3D Model 1 Nepal

3D Model 2 Nepal

3D Model 3 Nepal

3D Model 4 Nepal

Here’s a short video with all the annotations in the 3D model:

And here’s the link to the “live” 3D model. And to drive home the point that this 3D model could be far higher resolution if the underlying imagery had been directly accessible to Matthew, check out this other SketchFab model below, which you can also access in full here.

Screen Shot 2015-05-16 at 9.35.20 AM

Screen Shot 2015-05-16 at 9.35.41 AM

Screen Shot 2015-05-16 at 9.37.33 AM

The SketchFab team has kindly given me a SketchFab account that allows up to 50 annotations per 3D model. So I’ll be uploading a number of point clouds from Vanuatu (post Cyclone Pam) and Nepal (post earthquakes) to explore the usability of SketchFab for crowdsourced disaster damage assessments. In the meantime, one could simply tag-and-number all major features in a point cloud, create a Google Form, and ask digital volunteers to rate the level of damage near each numbered tag. Not a perfect solution, but one that works. Ultimately, we’d need users to annotate point clouds by tracing 3D polygons if we wanted a more easy way to use the resulting data for automated machine learning purposes.

In any event, if readers do have any suggestions on other software platforms, methodologies, studies worth reading, etc., feel free to get in touch via the comments section below or by email, thank you. In the meantime, many thanks to colleagues Jorge, Matthew, Ferda & Ji (QCRI), Salvador (PointCloudViz), Howard (PLAS.io) and Corentin (SketchFab) for the time they’ve kindly spent brainstorming the above issues with me.

Crowdsourcing Point Clouds for Disaster Response

Point Clouds, or 3D models derived from high resolution aerial imagery, are in fact nothing new. Several software platforms already exist to reconstruct a series of 2D aerial images into fully fledged 3D-fly-through models. Check out these very neat examples from my colleagues at Pix4D and SenseFly:

What does a castle, Jesus and a mountain have to do with humanitarian action? As noted in my previous blog post, there’s only so much disaster damage one can glean from nadir (that is, vertical) imagery and oblique imagery. Lets suppose that the nadir image below was taken by an orbiting satellite or flying UAV right after an earthquake, for example. How can you possibly assess disaster damage from this one picture alone? Even if you had nadir imagery for these houses before the earthquake, your ability to assess structural damage would be limited.

Screen Shot 2015-04-09 at 5.48.23 AM

This explains why we also captured oblique imagery for the World Bank’s UAV response to Cyclone Pam in Vanuatu (more here on that humanitarian mission). But even with oblique photographs, you’re stuck with one fixed perspective. Who knows what these houses below look like from the other side; your UAV may have simply captured this side only. And even if you had pictures for all possible angles, you’d literally have 100’s of pictures to leaf through and make sense of.

Screen Shot 2015-04-09 at 5.54.34 AM

What’s that famous quote by Henry Ford again? “If I had asked people what they wanted, they would have said faster horses.” We don’t need faster UAVs, we simply need to turn what we already have into Point Clouds, which I’m indeed hoping to do with the aerial imagery from Vanuatu, by the way. The Point Cloud below was made only from single 2D aerial images.

It isn’t perfect, but we don’t need perfection in disaster response, we need good enough. So when we as humanitarian UAV teams go into the next post-disaster deployment and ask what humanitarians they need, they may say “faster horses” because they’re not (yet) familiar with what’s really possible with the imagery processing solutions available today. That obviously doesn’t mean that we should ignore their information needs. It simply means we should seek to expand their imaginations vis-a-vis the art of the possible with UAVs and aerial imagery. Here is a 3D model of a village in Vanuatu constructed using 2D aerial imagery:

Now, the title of my blog post does lead with the word crowdsourcing. Why? For several reasons. First, it takes some decent computing power (and time) to create these Point Clouds. But if the underlying 2D imagery is made available to hundreds of Digital Humanitarians, we could use this distributed computing power to rapidly crowdsource the creation of 3D models. Second, each model can then be pushed to MicroMappers for crowdsourced analysis. Why? Because having a dozen eyes scrutinizing one Point Cloud is better than 2. Note that for quality control purposes, each Point Cloud would be shown to 5 different Digital Humanitarian volunteers; we already do this with MicroMappers for tweets, pictures, videos, satellite images and of course aerial images as well. Each digital volunteer would then trace areas in the Point Cloud where they spot damage. If the traces from the different volunteers match, then bingo, there’s likely damage at those x, y and z coordinate. Here’s the idea:

We could easily use iPads to turn the process into a Virtual Reality experience for digital volunteers. In other words, you’d be able to move around and above the actual Point Cloud by simply changing the position of your iPad accordingly. This technology already exists and has for several years now. Tracing features in the 3D models that appear to be damaged would be as simple as using your finger to outline the damage on your iPad.

What about the inevitable challenge of Big Data? What if thousands of Point Clouds are generated during a disaster? Sure, we could try to scale our crowd-sourcing efforts by recruiting more Digital Humanitarian volunteers, but wouldn’t that just be asking for a “faster horse”? Just like we’ve already done with MicroMappers for tweets and text messages, we would seek to combine crowdsourcing and Artificial Intelligence to automatically detect features of interest in 3D models. This sounds to me like an excellent research project for a research institute engaged in advanced computing R&D.

I would love to see the results of this applied research integrated directly within MicroMappers. This would allow us to integrate the results of social media analysis via MicroMappers (e.g, tweets, Instagram pictures, YouTube videos) directly with the results of satellite imagery analysis as well as 2D and 3D aerial imagery analysis generated via MicroMappers.

Anyone interested in working on this?