Patent Issued for Automated property inspections (USPTO 11302034): Tensorflight Inc.

2022 MAY 04 (NewsRx) -- By a News Reporter-Staff News Editor at Insurance Daily News -- From Alexandria, Virginia, NewsRx journalists report that a patent by the inventors Jarosz, Piotr (Warsaw, PL), Jocz, Lukasz (Warsaw, PL), Kozikowski, Robert (New York, NY, US), Paluba, Robert (Warsaw, PL), Surala, Wladyslaw (Warsaw, PL), Tybor, Rafal (Cracow, PL), Wojna, Zbigniew (London, GB), filed on July 9, 2020, was published online on April 12, 2022.

The patent’s assignee for patent number 11302034 is Tensorflight Inc. (New York, New York, United States).

News editors obtained the following quote from the background information supplied by the inventors: “Insurance or reinsurance for a commercial property is based on the value of the property including building structures and other relevant features, and the risks associated with the various attributes or characteristics of the building(s) and the property features. In general, there are four basic characteristics that an underwriter considers in performing a risk evaluation, namely, construction, occupancy, protection and exposure (COPE). Other attributes and/or characteristics of the commercial property can also be identified and evaluated as part of the risk analysis.

“The most basic element of a building is its construction, including the materials from which a building structure is made. Many insurers classify commercial buildings into categories based on their construction type. Each classification reflects both the building materials used (such as wood or concrete) and the combustibility and damageability of the materials used for major structural features. For example, a system developed by the Insurance Services Office (ISO) includes six building construction classifications: frame; joisted masonry; noncombustible; masonry noncombustible; modified fire resistive; and fire resistive.

“Occupancy describes the purpose for which the property is used, such as retail food market, furniture manufacturing, and apartments, to name a few, as well as how the insured manages the hazards associated with what they do. Protection means the methods used to safeguard a building from fire and other perils, including both public and private protection. Exposure refers to external hazards that exist largely due to the building’s location. Some hazards are natural. A building located in a breezy area may be subject to damage by high winds. Other natural hazards include sinkholes, hail, lightning, and heavy snow. Natural hazards can vary widely from one location to another.

“One conventional method for collecting relevant data is to conduct “in place” inspections, where a representative of the insurer (or the underwriter) visits the property and makes a visual assessment of the relevant property attributes. Another approach is to manually review photographs of the property and its buildings on publicly available websites. A more recent approach is to obtain satellite data and construct algorithms to evaluate the satellite images and extract relevant data.

“However, it is important when valuing a policy that the data relied upon is valid and current. Mistakes arise from the use of unreliable or outdated data sources, with errors appearing either in the manual input of data or from inaccurate information obtained from brokers, agents, or property owners. As a result, errors in the classification of properties can lead to a lower valuation thereby generating less in premium payments, i.e., premium leakage, a lost opportunity for insurers. An analysis by Verisk Analytics, Inc. estimated the premiums leakage due to misclassification at $4.5 billion annually. Further, the same Verisk analysis concluded that commercial real estate was wrongly classified by construction type in 32.5% of the cases studied, and the building area was incorrectly estimated in 80% of the cases studied. These problems can lead to significant undervaluing or overvaluing of insurance premiums for commercial properties.

“Therefore, there is a need for improved techniques for collecting reliable, precise and current attribute data for commercial properties in order to accurate value the properties for the purpose of insurance or reinsurance coverage on the properties.”

As a supplement to the background information on this patent, NewsRx correspondents also obtained the inventors’ summary information for this patent: “The following disclosure is directed to systems and methods for providing fully automated real property inspections in near-real time. In an embodiment, upon receiving a list of addresses from a user, multiple images are obtained of the real property parcels that correspond to each of the listed addresses, including images having at least two different perspective views of each parcel. For example, an overhead view of real property is analyzed with a first computer-based model that is configured and has been trained to identify buildings or other significant structures or objects from an overhead perspective. A different perspective view of the real property, such as a ground-level (street) view or an oblique angle view (e.g., 45°), is analyzed with a second computer-based model that is configured and has been trained to identify the building structures and/or other objects identified with the first model from the alternative perspective. A third computer-based model evaluates the different views together using an integrated approach that predicts the attributes associated with the identified objects, such as number of stories in the building. By using image recognition technology to evaluate each parcel from multiple view perspectives, better valuation estimates result leading to more competitive and accurate premium assessments.”

The claims supplied by the inventors are:

“1. A method implemented in a server, comprising: obtaining a plurality of images of a selected parcel of real property including a first set of images and a second set of images, the first set of images taken from a first view perspective relative to the first parcel and the second set of the images image taken from a second view perspective relative to the first parcel and different than the first view perspective; analyzing selected ones of the first set of images individually using optical recognition with a first model trained to identify major objects from the first view perspective; analyzing selected ones of the second set of images individually using optical recognition with a second model trained to identify the major objects from the second view perspective; analyzing the selected ones of the first and second set of images concurrently using optical recognition with a third model trained to identify attributes associated with building structures from an integrated analysis of the first and second view perspectives; identifying at least a first building structure and at least a first attribute associated with the first building structure; and generating a digital report on the selected parcel as a graphical user interface, the report including a selectable display of the selected ones of the first and second sets of images and a listing of the first attribute associated with the first building structure.

“2. The method of claim 1, further comprising: the first view perspective is an overhead view and the second view perspective is a ground level view or an oblique view.

“3. The method of claim 1, further comprising repeating the method for a plurality of parcels.

“4. A method implemented in a server, comprising: receiving a first physical address corresponding to a first parcel of real property; obtaining a plurality of images of the first parcel, at least a first image taken from a first view perspective relative to the first parcel and at least a second image taken from a second view perspective relative to the first parcel different than the first view perspective; analyzing the first image using optical recognition with a first model trained to identify major objects from the first view perspective; identifying at least a first building structure on the first image; analyzing the second image using optical recognition with a second model trained to identify the major objects from the second view perspective; identifying at least the first building structure on the second image; analyzing the first image and the second image together using optical recognition with a third model trained to identify attributes associated with building structures from the first view perspective and the second view perspective; and identifying at least one attribute associated with the first building structure; generating a digital report configured for display as a graphical user interface, the report including a selectable display of the first image and the second image and a listing of the at least one attribute associated with the first building structure.

“5. The method of claim 4, further comprising: the first view perspective is an overhead view and the second view perspective is a ground level view or an oblique view.

“6. The method of claim 4, further comprising repeating the method for a plurality of parcels.

“7. The method of claim 5, further comprising: obtaining a plurality of second images taken from the ground level view, each of the second images taken from a different lateral position relative to the first building structure; analyzing the plurality of second images using optical recognition with the second model; and analyzing the first image and the plurality of second images together using optical recognition with the third trained model.

“8. The method of claim 7, further comprising: the digital report configured to include a selectable display of the first image and the plurality of second images.

“9. The method of claim 4, further comprising: re-training the first model, the second model and the third model based on the results of the analyses steps.

“10. The method of claim 4, further comprising: receiving a plurality of physical addresses corresponding to a plurality of parcels of real property; and repeating the method for each parcel.

“11. The method of claim 4, further comprising the attributes associated with building structures include number of stories, construction type, occupancy type, estimated year built, floor area, roof material, and roof pitch.

“12. The method of claim 4, further comprising: adding a polygon to the first building structure, the polygon corresponds with a perimeter of the first building structure to represent the footprint of the first building structure.

“13. The method of claim 1, the step of analyzing the first and second images further comprising: adding an identifier to the attributes identified in the combined images.

“14. The method of claim 4, the step of analyzing the first and second images further comprising: assigning a classification to at least some of the attributes identified in the combined images using the third model, the classification describing the attribute.

“15. The method of claim 4, wherein the first model, the second model and the third model are different neural networks configured to apply computer vision in order to identify and characterize objects.

“16. The method of claim 15, wherein the neural networks are trained using human-annotated images and machine learning algorithms.

“17. A method implemented in a server, comprising: receiving a list of physical addresses corresponding to a plurality of parcels of real property; obtaining a plurality of images of a first parcel of the plurality of parcels, including a first set of the images taken from an overhead view of the first parcel and a second set of the images taken from a perspective view of the first parcel; selecting a first image from the first set of images; analyzing the first image using optical recognition with a first model trained to identify major objects from the first view perspective; identifying at least a first building structure on the first image; selecting a second image from the second set of images; analyzing the second image using optical recognition with a second model trained to identify the major objects from the second view perspective; identifying at least the first building structure on the second image; analyzing the first image and the second image together using optical recognition with a third model trained to identify attributes associated with building structures from the first view perspective and the second view perspective; and identifying at least one attribute associated with the first building structure; generating a digital report configured for display as a graphical user interface, the report including a selectable display of the first image and the second image and a listing of the at least one attribute associated with the first building structure.

“18. The method of claim 17, further comprising: selecting a third image from the second set of images, the third image having a different lateral position relative to the first building structure than the second image; analyzing the third image using optical recognition with the second model; identifying at least the first building structure on the third image.

“19. The method of claim 17, further comprising repeating the method for a plurality of parcels.

“20. An apparatus, comprising: a processor-based server system; a memory accessible to the server system and storing instructions that when executed by the server cause the server to: receive a first physical address corresponding to a first parcel of real property; obtain a plurality of images of the first parcel, at least a first image taken from a first view perspective relative to the first parcel and at least a second image taken from a second view perspective relative to the first parcel different than the first view perspective; analyze the first image using optical recognition with a first model trained to identify major objects from the first view perspective; identify at least a first building structure on the first image; analyze the second image using optical recognition with a second model trained to identify the major objects from the second view perspective; identify at least the first building structure on the second image; analyze the first image and the second image together using optical recognition with a third model trained to identify attributes associated with the major objects from the first view perspective and the second view perspective; and identify at least one attribute associated with the first building structure; and generate a digital report configured for display as a graphical user interface, the report including a selectable display of the first image and the second image and a listing of the at least one attribute associated with the first building structure.”

For additional information on this patent, see: Jarosz, Piotr. Automated property inspections. U.S. Patent Number 11302034, filed July 9, 2020, and published online on April 12, 2022. Patent URL: http://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO1&Sect2=HITOFF&d=PALL&p=1&u=%2Fnetahtml%2FPTO%2Fsrchnum.htm&r=1&f=G&l=50&s1=11302034.PN.&OS=PN/11302034RS=PN/11302034

(Our reports deliver fact-based news of research and discoveries from around the world.)