Abstract: Insurance companies traditionally use aerial imagery to guide post-catastrophe response when other data sources are limited or nonexistent. Meanwhile recent advances in computer vision and deep learning have created new exciting opportunities in this space. We design novel ways to use aerial imagery to simplify insurance underwriting and quoting, improve claim processing, target inspections, and proactively identify hazards to people and property.
Success of this effort hinges upon how we retrieve, process, augment, merge, and store aerial images, and upon models that we build to extract key features for underwriting and pricing. We use advanced geodata processing and workflow and data pipeline management to convert thousands of postal addresses to high resolution and true distance aerial images of corresponding properties. By automating and version controlling individual tasks within the luigi batch workflow, we can efficiently run those tasks on big data only once and then share results across teams for downstream processing avoiding unnecessary duplication of work.
Computer vision models that use aerial imagery can predict key pricing such as roof complexity or condition, presence of additional structures, trees, pools, and other potential hazards. As a result, we can simultaneously avoid premium leakage and significantly improve our customers’ experience.
Bio: Oleg Poliannikov is Head of Data Science at Solaria Labs, an incubation arm within Liberty Mutual Innovation focused on exploring emerging technologies and non-traditional business opportunities. Prior to joining Solaria Labs, he had been on MIT research staff working on applications of statistics and machine learning to earthquake seismology and imaging.