Case Study: First Insight Damascus

Report Highlights:

• 1,500km² imaged around the Syrian capital Damascus
• Persistent imagery surveillance over a 1-month period
• 59 volunteers identified signs of military and protestor activity
• 2,400 tagged points of insight
• CrowdRankTM consensus locations disseminated via a real-time private Twitter feed

Overview
Syrian civilians came in to violent conflict with the Assad regime in late 2011 / early 2012 in the cities of Homs, Hamah and Latakia. In March 2012, the protests spread to the Syrian capital, Damascus. With an entire country in turmoil, the huge area to examine and regular imagery updates exceeded the capabilities of DigitalGlobe’s in-house analysis center. Tomnod’s Global InSight crowdsourcing platform was deployed to enlist volunteer assistance in monitoring the unfolding crisis. The result was First Insight – a combination of Tomnod’s crowd insight with DigitalGlobe’s First Look fast-delivery, subscription imagery service.

First Insight: Damascus
First Insight: Damascus went live on March 1, 2012, running in a web browser. Given the sensitivity of the Syrian crisis, First Insight: Damascus employed a “community sourcing” approach with a password-protected site sent to invite-only groups. The example below shows the user interface where a volunteer in the community tagged a landmark, a crowd and a possible military building.

Figure 1 — Screenshot of the FirstInsight: Damascus user interface for tagging satellite imagery. In this example, a volunteer in the community tagged a landmark, a crowd and a possible military building.

Crowdsourcing Imagery Insight
The imagery in First Insight was updated immediately after new DigitalGlobe collects. New imagery was added 3 times throughout the month of March and displayed on First Insight within 24 hours of arriving on DigitalGlobe servers. A trusted community of 59 volunteers visited the First Insight website where they viewed small segments of the most current Damascus imagery (~1km2 per view, Figure 1). Imagery segments were presented to volunteers based on areas of interest prioritized by DigitalGlobe analysts to emphasize downtown Damascus and the airfield (Figure 2(a)). In total, multiple independent volunteers viewed 1,829 segments, densely covering the priority areas (Figure 2(b)). Volunteers were asked to examine each segment and identify 4 features of interest by dropping pins or “tags” on the image. The 4 features of interest were:

• Military buildings
• Military vehicles
• Crowds
• Landmark

The First Insight website included examples and instructions for how to identify each of these 4 feature types.

Figure 2(a) — Priority locations in the 1,500km² of regularly updated satellite imagery are input by First Insight administrators.

Figure 2(b) — Imagery segments from priority regions are viewed by multiple, independent volunteers in the crowd. Each yellow box corresponds to a single view by a crowd volunteer.

Figure 2(c) — Volunteers tag locations of interest, including possible military buildings (yellow) and vehicles (green), crowd or protestors (red) and landmarks (blue).

CrowdRank Consensus
First Insight volunteers identified 2,400 tags, with the following breakdown across the 4 features of interest:

•    773 Military Buildings
• 1,169 Military Vehicles
•      52 Groups of People
•    446 Landmarks

Figure 2(c) illustrates locations tagged by independent members of the community. To quantify consensus between multiple volunteers, Tomnod’s CrowdRank algorithm consumes all the tagged locations and computes the points with maximum agreement. CrowdRank iteratively estimates consensus between tags, as well as the reliability of each individual, resulting in probabilistic estimates that rank the most salient locations in the image. CrowdRank estimates update every hour and, whenever the consensus exceeds a confidence threshold, a confirmed location is disseminated to First Insight’s administrators and analysts.

See the gallery below for examples of top image locations identified by CrowdRank’s consensus.

Real-Time Reporting
As CrowdRank identifies consensus locations, image “chips” (small highlights) are extracted and disseminated via a real-time, protected Twitter feed @crowdrank_out (Figure 3). Authorized administrators were given access to this stream of intelligence tips, providing rapid, reliable updates about areas of interest in Damascus.

Figure 3 — Real-time dissemination of CrowdRank insight via a protected Twitter feed.
Move your mouse over the image to see an example Tweet.

Figure 3(b) — Individual tweet highlighting the image and location of a group of people identified by CrowdRank.

Conclusions
Modern imaging sensors collect vast amounts of data every day, capturing high-resolution detail of every inch of the planet. DigitalGlobe’s FirstLook offers real-time imagery of the most interesting events and objects across the world. However, with such huge amounts of information available, imagery analysts face the challenge of how to examine every pixel.

In response to the humanitarian crisis in Syria, Tomnod and DigitalGlobe partnered to launch First Insight – a crowdsourcing platform that engaged communities of trusted volunteers to transform pixels into points of interest. Tomnod’s CrowdRank algorithm extracted consensus from thousands of human inputs. The result was a stream of intelligence tips about hotspots in Damascus, delivered in real-time to administrators and analysts via web browsers and mobile devices.