Computer vision makes geospatial imagery come alive: 3 key insurance use cases
The other night, I looked up and saw a long line of blinking stars move across the sky. Even though I’d never seen anything like it, I was pretty sure it wasn’t a UFO. A quick Google search turned up the answer. The line of moving “stars” are part of Elon Musk’s Starlink venture, an initiative to bring satellite-based broadband internet to people across the globe. Starlink launches satellites in clusters of 60 and it can take several months for them to reach their orbit.
As both satellite size and launch costs shrink, the quality of the data they capture and transmit has increased dramatically. Starlink is just one of many companies engaging in a race-to-space to capitalize on better and cheaper satellite-based capabilities. Some geospatial satellites can now capture location images within one-meter (or less) accuracy. Additional image post-processing can improve the accuracy to within 0.5 meters even within poor conditions like cloud cover.
As of March 2020, approximately 2,700 satellites were in orbit and many more planned for launch this year. What do all of those “eyes in the sky” mean for the commercial insurance industry?
Computer vision, an artificial intelligence-enabled capability, allows computers to assess, interpret and act on vast amounts of video and imagery data. It encompasses three areas: reconstruction (the ability to estimate a 3D scene from a 2D image), recognition (assessing and tagging objects in an image), and reorganization (the segmentation of tags within an image to determine the relationships of objects within an image). Computer vision helps us turn the vast amount of geospatial imagery generated by satellite, reconnaissance planes, and drones into geospatial intelligence to be used in risk assessment, loss prevention, risk management, and product innovation.
Let’s look at a few examples:
Property Risk Assessment
A multinational corporation may have hundreds of locations across the world. A commercial property insurer needs to understand which locations are most at risk of loss so that they can prioritize which locations need review. Even with a good set of geocoded location data, it’s not feasible for risk engineers to conduct site surveys for all of a large company’s high-value, high-risk locations.
By combining geospatial imagery with computer vision and machine learning techniques, aerial site information can be used to determine not only information about the building (size, construction, age), but proximity to hazards (flood, wildfire) or other structures as well as predict the likelihood and magnitude of occurrence (how likely is the building to flood or catch on fire and how severe the damage might be). This information gives underwriters and risk engineers a more accurate picture of the risk, providing details about the building conditions and location that often the insured doesn’t even have. The insight also helps insureds enhance their risk mitigation strategies.
A hurricane has struck an area where you have significant exposure: How can you quickly get help to your insureds while more accurately estimating losses? Post-event analysis is another computer vision application. For smaller, targeted areas, loss adjusters use drones to survey rooftop damage after a storm. The drone transmits data and imagery to a device like an iPad, using photogrammetry to develop models of the site. Sophisticated computer vision algorithms can detect anomalies in the damage patterns (e.g. wind versus hail) and analysis results can be transmitted directly into claims management systems. Not only does this process reduce risk of injury to the adjuster, but it’s faster and more accurate, resulting in quicker claims resolution.
Aerial and satellite imagery provides a more comprehensive view of pre- and post-catastrophe events such as hurricanes, floods or wildfires. This allows insurers to more proactively develop response strategies. However, post-event analysis is most powerful when combined with the upfront risk assessment so that insurers can develop strategies in advance of events.
The same can be applied to marine cargo by pinpointing ship location or even marine liability for environmental monitoring.
Employee Safety and Loss Prevention
Companies with disciplined loss prevention practices can not only reduce risk but lower their cost of insurance coverage. Computer visions systems “see” and react to an environment in real-time. In commercial fleets, cameras placed on vehicles surveille the surroundings, alerting drivers to hazards or engaging automated collision avoidance systems. Applied at the geospatial level, computer vision enables more robust mapping and routing capabilities to get employees to their destinations efficiently and safely.
Computer vision assists with worker safety on construction sites. While geospatial imagery can be used to analyze the site at a macro-level, cameras placed around sites can leverage computer vision to analyze and tag photos and video to determine adherence to safety practices. For example, did the crew remember to put their hard hats on? Large-scale agribusiness and livestock operations are also adopting computer vision (both geospatial and location-based imagery) to optimize yield and monitor crop and animal conditions.
The number of applications for geospatial computer vision are limitless. While the capabilities are nascent, the accuracy of the image analysis and prediction methods are improving rapidly, resulting in a wider evaluation of application across industries. Innovation will drive new use cases, and in the insurance industry, expect to see computer vision as a vehicle for new product development and solution ideas.
However, due to regulatory challenges with data privacy and surveillance laws within different countries, insurers may need to tread carefully. As the overall market for geospatial intelligence grows and technology outpaces regulations, governments and regulators need to evolve Spatial Law policy frameworks to ensure that citizen rights and corporate interests are carefully balanced.