Wildfire risk detection using three-dimensional model of building structure and automated assessment of building-structure features

Abstract

A dimensionally calibrated three-dimensional model of a building structure is generated using overhead and ground-view images of a building structure on a property. The overhead image is segmented to define a polygon that represents a perimeter of the building structure and that includes vertices having respective geospatial positions. Ground-view images are used to determine actual dimensions and positions of structural features on external walls of the building structure. The dimensionally calibrated three-dimensional model is used to determine a wildfire risk using detected fuel sources and their distance from the nearest external wall of the building structure.

Description

Patent Application Attorney Docket No. FWIG-001W001WILDFIRE RISK DETECTION USING THREE-DIMENSIONAL MODEL OF BUILDING STRUCTURE AND AUTOMATED ASSESSMENT OF BUILDING-STRUCTURE FEATURESCross-Reference to Related Applications

[0001] This application claims priority to U.S. Provisional Application No. 63 / 677,202, titled “Dimensional Geometric Modeling Using Aerial And Ground Imagery,” filed on July 30, 2024, which is hereby incorporated by reference.Technical Field

[0002] This disclosure relates generally to systems and methods for assessing, mitigating and monitoring wildfire damage risk for building structures.Background

[0003] Wildfires are an increasingly important factor for homeowners to consider in various parts of the world, including in North America. In high-risk areas, wildfires present a real danger of total loss of a building structure, such as a residential home.

[0004] Assessing a building structure and corresponding land for wildfire requires that a skilled technician travel to the property to take measurements of the building structure including those of structural features on the building structure, such as windows, doors, vents, soffits, and decks. The technician also takes measurements of potential fuel sources on the property and their respective distances from the building structure. All of this data can then be used to assess the wildfire risk of the building structure. This is a time-intensive and expensive process for the technical to travel to the property and to perform the measurement work onsite.Summary

[0005] Example embodiments described herein have innovative features, no single one of which is indispensable or solely responsible for their desirable attributes. The following description and drawings set forth certain illustrative implementations of the disclosure in detail, which are indicative of several exemplary ways in which the variousPatent Application Attorney Docket No. FWIG-001W001 principles of the disclosure may be carried out. The illustrative examples, however, are not exhaustive of the many possible embodiments of the disclosure. Without limiting the scope of the claims, some of the advantageous features will now be summarized. Other objects, advantages, and novel features of the disclosure will be set forth in the following detailed description of the disclosure when considered in conjunction with the drawings, which are intended to illustrate, not limit, the invention.

[0006] An aspect of the invention is directed to a computer-implemented method for determining a wildfire risk for a building structure, comprising: receiving an overhead image from an overhead image database, the overhead image database in network communication with the one or more processors, the overhead image including geospatial data; determining, using the overhead image of a property including the building structure, first and second dimensions of the building structure, the first and second dimensions measured with respect to first and second axes, respectively, the overhead image having a known spatial scale; determining, using one or more ground-view images of the building structure, a third dimension of the building structure, the third dimension measured with respect to a third axis, the first, second, and third axes mutually orthogonal to each other, each ground-view image and the overhead image sharing a common dimension of the building structure; detecting, using one or more trained machine-learning (ML) models and the one or more ground-view images, structural features on one or more exterior walls of the building structure; generating a dimensionally calibrated three-dimensional model of an exterior of the building structure using the first, second, and third dimensions and the structural features; detecting, using the one or more trained ML models and at least the overhead image, fuel sources on the property; calculating a respective distance between each fuel source and a respective nearest exterior wall of the building structure using the dimensionally calibrated three-dimensional model; calculating a respective fuel load in terms of thermal energy generation potential for each fuel source; and determining a wildfire risk metric for the building structure based, at least in part, on the respective distance between each fuel source and a respective nearest exterior wall of the building structure and the respective fuel load for each fuel source.

[0007] In one or more embodiments, the method further comprises automatically determining respective dimensions of each structural feature, the respective dimensions determined relative to a respective common dimension of the building structure in aPatent Application Attorney Docket No. FWIG-001W001 respective ground-view image, wherein the wildfire risk metric for the building structure is further determined based, at least in part, on the respective dimensions of each structural feature. In one or more embodiments, the method further comprises determining a total number of pixels in the respective ground-view image corresponding to the respective common dimension of the building structure; calculating a respective pixel length as a ratio of the respective common dimension to the total number of pixels; determining a respective first number of pixels in the respective ground-view image corresponding to a respective first dimension of each structural feature; and determining the respective first dimension of each structural feature as a product of the respective first number of pixels in the respective ground-view image and the respective pixel length in the respective ground-view image.

[0008] In one or more embodiments, the method further comprises determining a respective second number of pixels in the respective ground-view image corresponding to a respective second dimension of each structural feature; and determining the respective second dimension of each structural feature as a product of the respective second number of pixels in the respective ground-view image and the respective pixel length in the respective groundview image.

[0009] In one or more embodiments, the one or more structural features includes one or more windows, one or more doors, one or more decks, one or more balconies, one or more vents, and / or one or more garage doors. In one or more embodiments, the method further comprises determining a respective position of each structural feature on the one or more exterior walls.

[0010] In one or more embodiments, the fuel sources include major vegetation, one or more non-vegetative fuels, and / or one or more secondary structures, the one or more non- vegetative fuels including one or more shipping pallets, one or more fuel containers, one or more industrial drums, one or more vehicles, one or more wood piles, and / or one or more pieces of furniture.

[0011] Another aspect of the invention is directed to a system for determining a wildfire risk of a building structure, comprising: one or more processors; non-transitory memory operably coupled to the one or more processors, the non-transitory memory storing computer-readable instructions that, when executed by the one or more processors, cause the one or more processors to: receive a geospatially calibrated overhead image from anPatent Application Attorney Docket No. FWIG-001W001 overhead image database, the overhead image database in network communication with the one or more processors, the geospatially calibrated overhead image including an overhead view of a property including the building structure; segment, using one or more trained machine-learning (ML) models, the geospatially calibrated overhead image to define a polygon that represents a perimeter of the building structure, the polygon including a plurality of vertices having respective geospatial vertex locations, each vertex corresponding to a respective corner of the building structure; receive a ground-view image of the building structure, the ground-view image showing one or more viewable external wall segments, each viewable external wall segment extending between a respective viewable pair of the vertices; determine, using input data, a geospatial location of a camera where the ground-view image was captured, a field-of-view (FOV) angle of the camera when the ground-view image was captured, and a compass bearing of the camera when the ground-view image was captured; segment, using the one or more trained ML models, the ground-view image to determine a boundary of the one or more viewable external wall segments and to detect at least a structural feature on a first viewable external wall segment, the structural feature having a pixel length in the ground-view image; define a plurality of lines, each line extending between the geospatial camera location and a respective vertex; define a plurality of subFOV angles, each subFOV angle defined between a respective pair of the lines; determine, using the geospatial camera location, the respective geospatial vertex locations, the subFOV angles, the FOV angle, and the compass bearing, a respective location of each viewable external wall segment, shown in the ground view image, on the polygon; calculate a respective length of each viewable external wall segment shown in the ground view image, the respective length determined using the respective geospatial vertex locations for the respective viewable pair of the vertices; determine a first subFOV angle defined between a first pair of the lines that extend to a first viewable pair of the vertices that define the first viewable external wall segment; determine a length of the structural feature using a first length of the first viewable external wall segment, the first subFOV angle, the FOV angle, and the pixel length of the structural feature; generate a dimensionally calibrated three-dimensional model of an exterior of the building structure using the polygon and the length of the structural feature on the first viewable external wall segment; detect, using the one or more trained ML models and at least the overhead image, one or more fuel sources on the property; calculate a respective distance between each fuel source and a respective nearest exterior wall of the building structure usingPatent Application Attorney Docket No. FWIG-001W001 the dimensionally calibrated three-dimensional model; calculate a respective fuel load in terms of thermal energy generation potential for each fuel source; and determine a wildfire risk metric for the building structure based, at least in part, on the respective distance between each fuel source and a respective nearest exterior wall of the building structure and the respective fuel load for each fuel source.

[0012] In one or more embodiments, the structural feature has an image position in the ground-view image defined, in part, by a pixel start location and the pixel length, and the computer-readable instructions, when executed by the one or more processors, further cause the one or more processors to determine a lateral physical location of the structural feature on the first viewable external wall segment using the first length of the first viewable external wall segment, the first subFOV angle, the FOV angle, the pixel start location, the pixel length, and the pixel resolution of the ground-view image.

[0013] In one or more embodiments, the pixel start location is a lateral pixel start location, the image position is further defined by a vertical pixel start location and a pixel height, and the computer-readable instructions, when executed by the one or more processors, further cause the one or more processors to determine a vertical physical location of the structural feature on the first viewable external wall segment using the length of the structural feature, the vertical pixel start location, the pixel height, the pixel length, and the pixel resolution of the ground-view image.

[0014] In one or more embodiments, the computer-readable instructions, when executed by the one or more processors, further cause the one or more processors to: calculate, using the geospatial camera location and the respective geospatial vertex locations, a respective distance between the geospatial camera location and each vertex; determine a closest vertex to the geospatial camera location within the FOV angle; and determine the respective location of each viewable external wall segment on the polygon based, at least in part, on the closest vertex to the geospatial camera location.

[0015] In one or more embodiments, the computer-readable instructions, when executed by the one or more processors, further cause the one or more processors to: compare(a) a first distance between the geospatial camera location and a first neighboring vertex to(b) a second distance between the geospatial camera location and a second neighboring vertex, each of the first and second neighboring vertices comprising an adjacent vertex to thePatent Application Attorney Docket No. FWIG-001W001 closest vertex on the polygon; define the first neighboring vertex as a second-closest vertex when the first distance is smaller than the second distance and the first neighboring vertex is within the FOV angle; and define the first viewable external wall segment as extending between the closest vertex and the second-closest vertex, the first viewable external wall segment having a first position on the polygon defined by the closest vertex and the second- closest vertex. In one or more embodiments, the structural feature has an image position defined by a pixel start location and the pixel length, and the computer-readable instructions, when executed by the one or more processors, further cause the one or more processors to: define an effective subFOV angle for the image position of the structural feature, the effective subFOV angle determined using the pixel start location, the pixel length, and the FOV angle; project the image position of the structural feature onto the first viewable wall segment using the effective subFOV angle and the FOV angle; and determine a lateral physical location of the structural feature on the first viewable external wall segment using the first length of the first viewable external wall segment, the first subFOV angle, the effective subFOV angle, the first distance between the geospatial camera location and the first neighboring vertex, and a third distance between the geospatial camera location and the closest vertex.

[0016] In one or more embodiments, the computer-readable instructions, when executed by the one or more processors, further cause the one or more processors to: compare the first subFOV angle with each of the other subFOV angles; determine that the first subFOV angle at least partially overlaps angularly with one or more of the other subFOV angles; and define each external wall segment defined by each of the one or more of the other subFOV angles that overlaps angularly with the first subFOV angle as a respective non- viewable external wall segment on the polygon. In one or more embodiments, the computer- readable instructions, when executed by the one or more processors, further cause the one or more processors to: determine that the first subFOV angle does not overlap angularly with one or more remaining subFOV angles, the one or more remaining subFOV angles excluding the one or more of the other subFOV angles that at least partially overlaps angularly with the first subFOV angle; compare the respective distances between the geospatial camera location and the vertices that defines the one or more remaining subFOV angles; define a first remaining vertex as a closest remaining vertex that partially defines a first remaining subFOV angle; compare (a) a first distance between a first neighboring remaining vertex to (b) aPatent Application Attorney Docket No. FWIG-001W001 second distance between the geospatial camera location and a second neighboring remaining vertex, each of the first and second neighboring remaining vertices comprising an adjacent vertex to the first remaining vertex on the polygon; define the first neighboring remaining vertex as a second-closest neighboring remaining vertex when the first distance is smaller than the second distance; and define a second viewable external wall segment that extends between the first remaining vertex and the second-closest neighboring remaining vertex, the second viewable external wall segment having a second position on the polygon defined by the first remaining vertex and the second-closest neighboring remaining vertex.

[0017] In one or more embodiments, the computer-readable instructions, when executed by the one or more processors, further cause the one or more processors to: compare the first remaining subFOV angle with each of the other remaining subFOV angle(s); determine that the first subFOV angle at least partially overlaps angularly with one or more of the other remaining subFOV angle(s); and define each external wall segment defined by each of the one or more of the remaining subFOV angle(s) that overlaps angularly with the first remaining subFOV angle as a respective additional non-viewable external wall segment on the polygon.

[0018] Another aspect of the invention is directed to a system for determining a wildfire risk of a building structure, comprising: one or more processors; non-transitory memory operably coupled to the one or more processors, the non-transitory memory storing computer-readable instructions that, when executed by the one or more processors, cause the one or more processors to: receive a geospatially calibrated overhead image from an overhead image database, the overhead image database in network communication with the one or more processors, the geospatially calibrated overhead image including an overhead view of a property including the building structure; segment, using one or more trained machine-learning (ML) models, the geospatially calibrated overhead image to define a polygon that represents a perimeter of the building structure, the polygon including a plurality of vertices having respective geospatial vertex locations, each vertex corresponding to a respective corner of the building structure; receive a ground-view image of the building structure, the ground-view image showing a viewable portion of an external wall segment of the building structure, the external wall segment extending between a viewable vertex that is shown in the ground-view image and a non-viewable vertex that is not shown in the ground-Patent Application Attorney Docket No. FWIG-001W001 view image; determine, using input data, a geospatial location of a camera where the groundview image was captured, a field-of-view (FOV) angle of the camera when the ground-view image was captured, and a compass bearing of the camera when the ground-view image was captured; segment, using the one or more trained ML models, the ground-view image to determine a boundary of the viewable portion of the external wall segment and to detect at least a structural feature on viewable portion of the external wall segment, the structural feature having a pixel length; calculate, using the geospatial camera location and the respective geospatial vertex locations, a respective distance between the geospatial camera location and each vertex within the FOV angle; determine a closest vertex to the geospatial camera location within the FOV angle at the compass bearing of the camera when the ground-view image was taken, the respective distance between the geospatial camera location and the closest vertex smaller than the respective distance between the geospatial camera location and other vertices within the FOV angle; define the closest vertex as the viewable vertex; compare (a) a first distance between the geospatial camera location and a first neighboring vertex to (b) a second distance between the geospatial camera location and a second neighboring vertex, each of the first and second neighboring vertices comprising an adjacent vertex to the closest vertex on the polygon; define the first neighboring vertex as a second-closest vertex when the first distance is smaller than the second distance; define the second-closest vertex as the non-viewable vertex; determine a location of the external wall segment on the polygon using the viewable and non-viewable vertices; define the viewable portion of the external wall segment, with respect to the polygon, using the location of the external wall segment on the polygon, the geospatial camera location, and the FOV angle, the FOV angle intersecting the external wall segment at an intersection point; define a plurality of vertex lines, each vertex line extending between the geospatial camera location and a respective vertex; define an intersection line that extends between the geospatial camera location and the intersection point; determine a first vertex angle between (a) a first vertex line that extends between the geospatial camera location and the viewable vertex and (b) and a second vertex line that extends between the geospatial camera location and the non- viewable vertex; determine a length of the external wall segment using the respective geospatial vertex locations of the viewable and non-viewable vertices; determine a length of the viewable portion of the external wall segment using the length of the external wall segment, the first vertex angle, the intersection line, the respective distance between thePatent Application Attorney Docket No. FWIG-001W001 geospatial camera location and the viewable vertex, and the first distance between the geospatial camera location and the non-viewable vertex; determine a length of the structural feature using the length of the viewable portion of the external wall segment, the intersection line, the FOV angle, the pixel length of the structural feature, and a pixel resolution of the ground-view image; generate a dimensionally calibrated three-dimensional model of an exterior of the building structure using the polygon and the length of the structural feature on the viewable portion of the external wall segment; detect, using the one or more trained ML models and at least the overhead image, fuel sources on the property; calculate a respective distance between each fuel source and a respective nearest exterior wall of the building structure using the dimensionally calibrated three-dimensional model; calculate a respective fuel load in terms of thermal energy generation potential for each fuel source; and determine a wildfire risk metric for the building structure based, at least in part, on the respective distance between each fuel source and a respective nearest exterior wall of the building structure and the respective fuel load for each fuel source.

[0019] In one or more embodiments, the structural feature has an image position defined, in part, by a pixel start location and the pixel length, and the computer-readable instructions, when executed by the one or more processors, further cause the one or more processors to: define an effective subFOV angle for the image position of the structural feature, the effective subFOV angle determined using the pixel start location, the pixel length, and the FOV angle; project the image position of the structural feature onto the viewable portion of the external wall segment using the effective subFOV angle and the FOV angle; and determine a lateral physical location of the structural feature on the viewable portion of the external wall segment using the length of the viewable portion of the external wall segment, the effective subFOV angle, the first distance between the geospatial camera location and the first neighboring vertex, and a third distance between the geospatial camera location and the closest vertex.

[0020] In one or more embodiments, the computer-readable instructions, when executed by the one or more processors, further cause the one or more processors to: define a plurality of vertex angles, each vertex angle defined between a respective pair of the lines, the vertex angles including the first vertex angle; compare the first vertex angle with each of the other vertex angles, the other vertex angles excluding the first vertex angle; determine thatPatent Application Attorney Docket No. FWIG-001W001 the first vertex angle at least partially overlaps angularly with one or more of the other subFOV angles; and define each external wall segment defined by each of the one or more of the other vertex angles that overlaps angularly with the first vertex angle as a respective non- viewable external wall segment on the polygon. In one or more embodiments, the camera comprises a stereoscopic camera.Brief Description of the Drawings

[0021] For a fuller understanding of the nature and advantages of the concepts disclosed herein, reference is made to the detailed description of preferred embodiments and the accompanying drawings.

[0022] Fig. 1 illustrates a structure with fuel sources according to one or more embodiments.

[0023] Fig. 2 illustrates a system for wildfire risk assessment and mitigation according to one or more embodiments.

[0024] Fig. 3 illustrates a risk engine according to one or more embodiments.

[0025] Fig. 4 illustrates a networked computing system according to one or more embodiments.

[0026] Fig. 5 is a flow chart of a method for wildfire risk assessment for a building structure according to one or more embodiments.

[0027] Fig. 6 is a flow chart of a method for analyzing a building structure according to one or more embodiments.

[0028] Fig. 7 is a flow chart of a method for analyzing fuel sources according to one or more embodiments.

[0029] Fig. 8 is a flow chart of a method for analyzing wildfire threat vectors according to one or more embodiments.

[0030] Fig. 9 illustrates an aerial view with exemplary grid according to one or more embodiments.

[0031] Fig. 10A illustrates an example of an dimensionally calibrated overhead image of a building structure according to one or more embodiments.Patent Application Attorney Docket No. FWIG-001W001

[0032] Fig. 10B illustrates an example of a ground-view image of the building structure shown in Fig. 10A according to one or more embodiments.

[0033] Fig. 11 illustrates a dimensionally calibrated three-dimensional model of a building structure according to one or more embodiments.

[0034] Fig. 12 is a flow chart of a method for determining a wildfire risk of a building structure according to one or more embodiments.

[0035] Fig. 13 illustrates an example polygon that defines a perimeter of a building structure according to one or more embodiments.

[0036] Fig. 14 is a flow chart of a method for determining one or more dimensions of a detected exterior building-structure feature on an exterior wall segment that extends between two vertices that are viewable in the ground-view image, according to one or more embodiments.

[0037] Fig. 15A illustrates an example polygon used in the flow chart shown in Fig. 14 according to one or more embodiments.

[0038] Fig. 15B illustrates an example boundary of an exterior wall segment in a ground-view image according to one or more embodiments.

[0039] Fig. 16 is a flow chart of a method for determining one or more dimensions of a detected exterior building-structure feature on an exterior wall segment that extends between two vertices that are viewable in the ground-view image, according to one or more embodiments.

[0040] Fig. 17 illustrates an example polygon used in the flow chart shown in Fig. 16.

[0041] Fig. 18 is a flow chart of a method for determining one or more dimensions of a detected exterior building-structure feature on a viewable portion of an exterior wall segment that extends to only one vertex that is viewable in the ground-view image, according to one or more embodiments.

[0042] Fig. 19 illustrates an example polygon used in the flow chart shown in Fig. 18.

[0043] Fig. 20 is a flow chart of a method for determining one or more dimensions of one or more detected exterior building-structure features on an exterior wall segment according to one or more embodiments.Patent Application Attorney Docket No. FWIG-001W001

[0044] Fig. 21 illustrates an example polygon where none of the lines are within the field-of-view of the camera.

[0045] Fig. 22 illustrates an example diagram that can be used in one or more steps of the flow chart shown in Fig. 18.

[0046] Fig. 23 illustrates an example diagram that can be used in one or more of the methods described herein.Detailed Description

[0047] A three-dimensional model of a building structure is generated using, as inputs, a geospatial overhead image of the building structure and one or more ground-view (e.g., side-view) images of the building structure. Each ground-view image includes metadata that include the geospatial location of the camera that captured the image and the Field of View (FOV) angle of the camera when the image was captured. The position of the external wall(s) of the building structure shown in each ground-view image are determined using the geospatial location of the camera that captured the image, the FOV angle, and geospatial data extracted from the overhead image that represents corners (e.g., vertices) of the building structure.

[0048] Structural features, such as windows, doors, vents, and / or soffits, are detected in one or more the ground-view images. Dimensions and / or positions of the structural features are determined and included in the three-dimensional model of the building structure. Different approaches to determining the dimensions and / or positions of the structural features can be used, in one or more embodiments, based on the relative position of the camera / FOV with respect to the external wall(s) captured in the image and the compass bearing of the camera when the image was taken.

[0049] Fuel sources on the property can be detected in the overhead image and / or in one or more of the ground-view image(s). The fuel load of each fuel source and the distance between each fuel source and the nearest external wall of the building structure can be determined. A wildfire risk metric can be determined using the three-dimensional model of the building structure, the fuel load of each fuel source, and the distance between each fuel source and the nearest external wall of the building structure.Patent Application Attorney Docket No. FWIG-001W001

[0050] The three-dimensional model can be generated without having a trained individual, such as a technician, travel to the property for purposes of wildfire risk assessment. For example, a lay individual (e.g., the property owner) can take ground-level digital photographs and / or ground-level digital video images of the external walls the building structure. The system uses these digital images, in combination with a dimensionally calibrated overhead image of the building structure, to identify the external wall(s) shown in the ground-level images, identify structural features on the external walls that are shown in the ground-level images, and determine the position and / or dimensions of each structural feature, which are used as inputs for determining the wildfire risk for the building structure.

[0051] Fig. 1 illustrates an example building structure (e.g., a residential building such as a house) 10 having horizontal and vertical construction elements and other three- dimensional components that are susceptible to varying degrees of wildfire damage risk and contribute to the risk in combination with environmental factors such as weather factors and the presence, nature, and quantity of fire fuels in the structure’s environment (e.g., trees and vegetation). The figure illustrates a ground-level view of the building structure 10 with fuel sources (e.g., trees and vegetation, generally fuel sources) in its environment in accordance with one or more embodiments. Fuel sources 104 may form into heat vectors affecting structure 10 in several different ways including direct flame exposure 110 (convection), radiant heat 112, ember accumulation 114, and / or ember penetration 116. Any of these risk or heat vectors may cause the building structure, such as a portion 102 of the building structure 10, to catch on fire. Fuel sources 104 surrounding the building structure 10 may generate flames 106 that may spread upon contact with flammable materials, and may further generate ember projectiles 108, which may be spread from the fuel sources 104 on air currents.Movement of flames 106 may be influenced by air currents and elevation, as well as the presence or absence of flammable fuel. Flames 106 may cause direct flame exposure 110 and thus ignition of materials comprising the building structure 10, as well as generation of radiant heat 112. Radiant heat 112 may cause combustion of materials by raising the material temperature to above the material's ignition temperature. Ember projectiles 108 may fall or be blown onto and gather upon surfaces of the building structure 10, resulting in ember accumulation 114, which may ignite those surfaces. Openings in a building structure 10, such as windows, gaps in roofing materials, and soffits, may allow ember penetration at one orPatent Application Attorney Docket No. FWIG-001W001 more locations, e.g., 116, and these embers may ignite materials interior to the building structure 10.

[0052] Fig. 2 is a block diagram of a system for wildfire risk assessment and mitigation 200 according to one or more embodiments. The system for wildfire risk assessment and mitigation 200 can include a risk engine and property ignition model system (RPIM) 202, mitigation management system (MMS) 216, and / or asset monitoring and protection system (AMP) 232.

[0053] The system for wildfire risk assessment and mitigation 200 may be implemented in or with one or more computers.

[0054] The RPIM 202 can include a risk engine 300 that interacts with a property ignition model 204 to analyze and assess aspects of a property, such as a building structure on the property, for wildfire risk. The risk engine 300 can request data from and provide data to application programming interfaces (APIs) such as external APIs 206 and / or third-party data providers 208 as part of this analysis and assessment. The risk engine 300 can interact with a property assessment application 210 in order to obtain information regarding the features of a property. The risk engine 300 can provide data to inform a property protection plan & mitigation report 212 and / or an insurance quote 214, and may also take information from these entities in as input to its analysis and assessment, and in order to refine its algorithms to provide improved analysis in future. The risk engine 300 can take as input and update as output client confirmation 220, subscription management 222, additional services 224, customer approvals 226, scheduling and billing 228, and / or service certifications 230 associated with the property under assessment. These entities may be further used in mitigating risk using MMS 216.

[0055] The MMS 216 can include a mitigation management hub 218 that may take as input and update as output subscription management 222, additional services 224, customer approvals 226, scheduling and billing 228, and / or service certifications 230 associated with the property under assessment. These entities may be further used to determine the binding of a policy (see policy bound 238), as well as inform the operations management property monitoring command center 234 of AMP 232.Patent Application Attorney Docket No. FWIG-001W001

[0056] The AMP 232 can include an operations management property monitoring command center 234 that can take as input and update as output subscription management 222, additional services 224, customer approvals 226, scheduling and billing 228, and / or service certifications 230 associated with the property under assessment. The AMP 232 can be informed by whether or not the policy is bound (policy bound 238). The operations management property monitoring command center 234 may interact with asset protection services and wildfire event communications 236 in order to remain up to date on changes to and status of the asset being monitored. The AMP 232 can comprise a monitoring platform in one or more embodiments. The monitoring platform may utilize predictive models to trigger pre-fire and during fire services entitled under additional services 224.

[0057] Fig. 3 is a block diagram of the risk engine 300 according to one or more embodiments. The risk engine 300 includes a machine learning stage 304 and a wildfire risk algorithm 312. The machine learning stage 304 may take in an address 302 and use machine learning 310 to identify fuel sources 306 and at least one protected building structure 308, which it may provide to the wildfire risk algorithm 312. Based on the fuel sources 306 and the protected building structure 308, informational databases may be used to gather information on and / or calculate maximum windspeed 314, an ember model 316, a maximum heat output (e.g., in British Thermal Units (BTUs)) 318, a maximum heat flux and flame decay 320, material failure thresholds 322, and / or heat flux and flame front models 324. These may each be applied to wall segments of the protected building structure 308 to determine risk accumulation for structure squares 326. This may inform a combination of risk with structure square material failure threshold 328. The result may be provided as a Failure Mode Effect Analysis (FMEA) framework output 330.

[0058] Risk accumulation for structure squares 326 may take into account radiant flux projection, flame front ellipse, and ember dispersion to intersect all structure squares encountered. Combination of risk with structure square material failure threshold 328 may combine each structure square accumulated input against material failure thresholds 322 for each failure mode (flux, flame, ember).

[0059] FMEA framework output 330 may include a detailed structure failure analysis as follows: each section of the structure that would experience an ignition during a wildfire may be inventoried, and the inventory may be ordered by severity and surplus. FMEAPatent Application Attorney Docket No. FWIG-001W001 framework output 330 may also include a detailed fuel contributors by zone (1-3) as follows: contributors may be inventoried by zone and ordered by contribution to failures related to each fuel source in order of contribution, this may facilitate identification and prioritization of actionable mitigation work.

[0060] Fig. 4 illustrates a networked computing system 400 according to one or more embodiments. The networked computing system 400 may be used to implement the system for wildfire risk assessment and mitigation 200 illustrated in Fig. 2. The networked computing system 400 may comprise a network 402, at least one user 404, at least one stationary computing device 406, at least one mobile computing device 408, and / or a server 410, in any of various combinations as may be readily apprehended by one of ordinary skill in the art.

[0061] Network 402 may be any one or more networks that allow elements of the present system to communicate with each other, as may be known in the art, for example wide area networks (WANs), LANs, and the like, that may be wired or wireless, and may include Bluetooth™, WiFi, and other approaches to local or remote communication.

[0062] There may be one or more stationary computing devices 406 that may be one or more computing systems that may be used various users of the system for wildfire risk assessment and mitigation 200, such as homeowners (owners of building structures), appraisers, mitigation consultants or companies, monitoring consultants or companies, insurance companies or brokers and adjusters, structure data providers (such as images of structures, weather data, elevation data, material data, and the like).

[0063] Mobile computing devices 408 may provide access to various functionality of system for wildfire risk assessment and mitigation 200, similar to stationary computing devices 406. In addition, mobile computing devices 408 may allow one or more users of system for wildfire risk assessment and mitigation 200 to input structure data that may need the device to be proximate to the structure (such as to take images of the structure and fuel sources that may not be obtained via other means such as satellite or Google™ imagery).

[0064] Either stationary computing devices 406 or mobile computing devices 408 may be used to perform various functionality of system for wildfire risk assessment and mitigation 200, as described herein - for example to initiate and perform risk assessment,Patent Application Attorney Docket No. FWIG-001W001 obtain, calculate or refine structure datasets, determining mitigation strategies and documenting completion of such mitigation, initiating and performing monitoring of one or more structures, and advising of monitoring or mitigation actions needed based on the monitoring. Together stationary computing devices 406 and mobile computing devices 408 may be referred to simply as computing devices.

[0065] In one or more embodiments, wildfire risk assessment of a building structure (e.g., protected building structure 308) can be performed using a Property Ignition Model (PIM).

[0066] The PIM is premised upon the empirically determined tenant that ignition is a function of a home’s structural features and their spatial relationship to the immediately surrounding fuel sources. Furthermore, research has determined that ignition is caused by some combination of the three principal ways in which fire spreads along the WUI (1) radiant heat aided by piloted ignition from embers, (2) direct flame impingement (convective heat), and (3). Firebrands entering the structure and accumulation of firebrands.

[0067] These three modes of wildfire fire transmission form the physical basis of a thermodynamic threat vector analysis. Not all structural elements are vulnerable to all threat vectors, and given that each threat vector represents a different physical phenomenon, the impact of each threat vector may be modeled separately: (a) heat flux for radiant impact, (b) flame front contact for direct flame impingement, (c) ember mass accumulation and size population for firebrand accumulation, and (d) ember penetration probability computation.

[0068] The likelihood of ignition by any of these threat vectors is a function of the variables encoded in a matrix, the spatial relationships between them, and the specific attributes of each individual element. To exemplify this, consider fire spread from a large 1- ton tree to a home with stucco siding. The PIM considers the total heat released by the tree under combustion, the heat capacity of stucco, the ignition temperature of treated wood, the distribution of ember size by distance from the tree, and several other attributes that, when taken together, fully characterize fire transmission between the tree and the home.

[0069] Finally, the assumption that if any feature of the home fails (ignites) then the entire home may ignite, allows for linear pair-wise (heat source - structure segment) modeling of the overall home ignition question. In practice, the PIM evaluates whether eachPatent Application Attorney Docket No. FWIG-001W001 exterior structural element of a given home ignites under the influence of each of the four threat vectors summed over all fuel sources with direct access to each exterior structural element. The impact of each of the four threat vectors may be summed over all “line of sight” fuel sources (e.g., a tree “on the other side of the house” may have no impact) at each structure tile. Then, for each structure feature at that tile location, an ignition (failure) determination may be computed. For radiant heat, the total incident heat flux may be used in combination with tables of structural material heat capacities to determine if ignition temperatures are reached. Breaking temperatures are used in the case of window glass. If the flame front occupies the same coordinates as a structural tile, that tile may be assumed to ignite / fail. For roof tiles, accumulated ember mass may be used to determine whether or not burn-through occurs. For wall tiles, sub 0.01g ember populations may be used to probabilistically determine if embers penetrate vents.

[0070] Wind may be an additional variable to be considered when modeling fire transmission. Since wind direction may affect the impact of any given fuel source on each roof or wall tile, threat-vector outcomes may be computed nine separate times: once for wind from each of eight compass headings and a “no wind” case to cover worst-case conditions for each structure tile. The model considers, for example, that a wind out of the north-east may yield maximum heat and ember transfer to structural elements on the north side of the house, but a south-west wind may yield little impact on those same elements. In that situation, the north-east results may be used for evaluation.

[0071] The PIM in one or more embodiments may utilize a simplified set of standard fuel sources, structural elements, and fire susceptibility parameters for modeling. Accuracy of the PIM may continue to improve as feature detection algorithms are trained and refined and more data is gathered on the fire susceptibility of individual materials.

[0072] For each structure tile for which the model determines an ignition (failure), the PIM catalogues the following (a) tile location, (b) feature (or features) that failed, (c) threat Vector (or vectors) that caused each of the feature failure(s), (d) individual fuel sources contributing to each of the specific vector - feature failure(s), (e) wind direction(s) during failure, and / or (f) failure surplus (a measure of “how close” the feature came to non- failure / non -ignition). For example, a feature might fail at 2kJ / m2of accumulated heat flux over the combustion period. The model determines that the feature fails because the incidentPatent Application Attorney Docket No. FWIG-001W001 heat at that tile is 2.5kJ / m2. Thus the “failure surplus” is 0.5kJ / m2. The “failure surplus” may be useful for determining the remediation classification. Large surpluses indicate more extensive remediation may be needed (or may even not be possible), whereas small surpluses may need less extensive remediation.

[0073] With the modeling approach described in the previous section, each fuel source and structural element in scope is backed by specific calculations for thermal energy output, probability of ignition, and remediation opportunity. The rich level of detail obtained through data acquisition methods may support calculations that may vary considerably from item to item. For example, a short tree with a small crown may have a much smaller thermal energy output potential than a tall tree with a large crown. All failure items may be compiled into a modified FMEA framework, and the relative risk of each item may be quantified in a Risk Priority Number (RPN).

[0074] While results of the overall risk assessment may be presented in the context of standards set by the Institute for Business and Home Safety (IBHS) and include four discrete risk zones, listed below, the relative risk may be computed for each of five threat vectors (convective, radiant, ember accumulation, ember penetration, and nearby non-primary structures).

[0075] Fig. 9 illustrates an aerial view with exemplary grid 900 in accordance with one or more embodiments. A multi-dimensional 300 x 300 grid 908 with tiles 918 at, e.g., 2ft x 2ft resolution for a total of 90,000 analysis points per property, noting that higher precision may be chosen based on computing resources and multiple calculations may be performed for each grid, as described).

[0076] The model disclosed herein may consider four discrete risk zones: (1) Home Ignition Zone (HIZ): The building structure 10 itself and a boundary of 5 feet minimum from the building structure 10, inside the line 902; (2) Zone One: 5 feet to 30 feet from the building structure 10, inside the line 904 and outside the line 902; (3) Zone Two: 30 to 100 feet from the building structure 10, inside the line 906 and outside the line 904; and (4) Zone Three: 100 to 300 feet from the building structure 10, outside the line 906.

[0077] The centroid of the building structure 10 is positioned at the center of the structure grid 900, and every structure tile within the grid may be encoded with fuel andPatent Application Attorney Docket No. FWIG-001W001 structure elements or structure data. For example, the structure outline or footprint may determine which tiles surrounding the center tile are roof tiles or wall tiles and each of those may be encoded with the structural materials at that part of the exterior wall (window, siding, roof material, etc.).

[0078] Once each threat vector score is computed, a cumulative risk score may be calculated, and then for the entire property based on threat vector contributions to ignition failure points. The PIM incorporates the compounding effect of risk from multiple threat vectors. This also helps in understanding contributors to ignition, as most items may fail due to a cumulative effect of heat flux from multiple sources.

[0079] It is also important to understand that each failure item may be attached to multiple fuel sources and there may be overlap between them. The FMEA framework analysis sits on top of each individual failure item so each failure and the point at which it fails may be properly represented. This gives us the ability to show multiple failures in the same location of a home. For example, a wall section may comprise a window and siding that both may fail, but at different points. Reducing one risk may or may not eliminate the adjacent risk. The PIM's risk prioritization approach may augment the FMEA framework in two ways. First, traditional detectability scoring may be replaced with remediation scoring. This reflects the ability to address remediation in a failure event. Second, risk prioritization considers failure contributions and surplus amount to subsequent related structure failures.

[0080] Results may be presented to the owner and / or their insurer in a property protection plan, where all supporting calculations, images, and features evaluated throughout the analysis may be attached with each component for review. This approach is unique in that a component-level view of risk for a property may be compiled, and a component-level remediation plan to mitigate risks. From a system perspective, at the end of this process each Risk Component (structure feature or fuel source) may be a separate structured data object that includes a Risk Score, a “Treated” Risk Score, and a Mitigation Treatment. These data objects and the granularity they represent serve as the foundation for the Mitigation Management system, informing the downstream integrated service delivery and logistics around approvals, scheduling, pricing, billing, and other services.

[0081] In one or more embodiments, scoring may be based on the FMEA framework model, where each failure may be identified, and scoring may be computed based on scope,Patent Application Attorney Docket No. FWIG-001W001 probability, and detectability. Other embodiments may expand on this to map scope to energy, or may score by failure vector, for which more accurate and deterministic data regarding point of ignition may be available.

[0082] Each failure vector (convective, radiant, ember accumulation, ember entry, and other structures) involves distinct calculations to determine ignition points. These calculations may be performed using a model for fuel load on the structure and information about when a structure may exhibit a thermal failure (ignition) based on materials and ignition points of those materials. Depending on the failure mode, each calculation may represent a relative overage (excess energy) to ignition based on both energy and amount of the structure that may fail. As these calculations are different, the range of scoring may be fit (based on a large sample of properties >10000) to a 1-100 range, representing a thermal excess of energy needed to cause ignition based on the assumed fire conditions. This may be varied in the disclosed system, but is based on standard peak ignition models of a Peak Burn: all trees burn in a 1-minute heat pulse, with 20 mph winds, and 30% humidity. The scale of output may indicate how much excess energy may be present at the structure during a fire. Zero is no ignition at peak, while the further away from Zero, the more excess energy may be present to drive ignition. This informs both the structure risk as well as amount of energy reduction needed to solve for ignition. Ignition may in some cases be solved by removing fuels or by hardening a structure to be more resilient to thermal exposure. An example of this would be to replace wood siding with cement board siding.

[0083] Fig. 5 is a flow chart of a method 500 for wildfire risk assessment for a building structure according to one or more embodiments.

[0084] Method 500 begins at 502 where an application may be received to perform one or more steps of a wildfire risk assessment. An application may be received at RPIM 202 and may be received from a computing device, having been initiated by a user 404. T5he application may include some limited data about one or more building structures, such as one or more building structure addresses. Method 500 may also start at 502 with one or more other triggers to perform the method. For example, system for wildfire risk assessment and mitigation 200 may receive new, or further, information that may affect a prior risk assessment or mitigation assessment, such as a user 404 providing more information about a building structure (e.g., building structure 10, 308), mitigation step information beingPatent Application Attorney Docket No. FWIG-001W001 received by system for wildfire risk assessment and mitigation 200 (such as confirmation that a tree or other fuel source 104 affecting a particular building structure has been cut down), and / or that monitoring has been initiated for a given building structure that is already in system for wildfire risk assessment and mitigation 200 but has not previously been part of a monitoring service. In some of such cases method 500 may not have to fully re-perform all of 504-512 but may update the results based on the new or additional information.

[0085] Method 500 continues to then perform various steps to calculate a quantitative risk assessment for each of the structures received in the application. Broadly speaking the ingestion and creation of structure data may be performed at 506-510. Then risk assessment, with mitigation strategy determination, if indicated, may be performed at 510. Method 500 may then display results of the risk assessment and / or mitigation assessment at 514.Returning to risk assessment at 506-510, method 500 begins to ingest and create structure data used to calculate the risk for a given building structure, for each building structure in a group or set of building structures.

[0086] At 506, a building structure may be analyzed and ingested, such as by receiving an address, and creation of a structure dataset may be initiated. Ingesting or analyzing the structure may comprise the following, as may be described in method 600 of Fig. 6:

[0087] Step 602: Get an address from an application or via a data source available through a computing device (e.g., 406, 408, 410). Using the address, a latitude / longitude point may be obtained. Global positioning system (GPS) coordinates, geographic information system (GIS) coordinates, or other forms of locating and orienting may be used for the various features described herein.

[0088] Step 604: Determining a structure outline. This may be the top-down outline view of the structure. This may be determined from a data source and using GPS coordinates of each comer of the structure to create an outline in GPS coordinates. Machine learning may perform this determination in one or more embodiments.

[0089] Step 606: Determining the structure centroid. This may be determined with reference to data from a data source and the structure outline. The structure centroid may be used to create a grid.Patent Application Attorney Docket No. FWIG-001W001

[0090] Step 608: Further structure classification, such as from the exterior features of the structure(s) on the property, where the features (such as windows, decks and the like) are located along the perimeter of the respective structure, their dimensions, and material composition. This may be arrived at from oblique satellite imagery, aerial imagery, ground imagery, real estate and other property data, and input via an app accessible on computing devices. Trained machine-learning (ML) model(s) can be used on these images to detect windows, doors, garage doors, vents (e.g., air vents, laundry vents, and / or other vents), soffits, and other exterior building-structure feature to model fire susceptibility.

[0091] Method 600 may then return to method 500 where, at 508 a grid fuel source dataset may be created by determining fuel sources 104 within a grid, as may be described in method 700 of Fig. 7. This may involve determining or detecting major vegetation (trees, shrubs), structures (sheds, fences), and neighboring rooftops within some distance of the building structure that could act as a fuel and contribute to the advancement of wildfire. Other characteristics may also be extracted, such as local topological features (slope, roads, hydrants, arroyos). For modeling purposes, specific attributes may then be determined for each feature or fuel source 104. In the case of a large tree, for example, algorithms estimate tree height and measure crown size. In the case of a shed or pool house, algorithms determine height and surface area. Feature-specific attributes may then be used to calculate fuel load in terms of thermal energy (e.g., BTU) generation potential. This may be done for every feature or fuel source, effectively translating the sources of data (such as satellite imagery and other sources of data about fuel sources 104) into a full inventory of fuel sources surrounding the structure. This may preferably all be done in an automated fashion, though some may be manual, such as by operators visiting building structures and assessing one or more elements of building structure and fuel sources 104. In automated functioning, images may be provided to an artificial intelligence (Al) and / or machine learning (ML) image processing system (e.g., an Al tree detection system and / or an ML tree detection system) that identifies fuel sources. The AI / ML tree detection system may have previously been provided training data, from existing data sources 210 and augmented with manual data as may be desired, for example that identified fuel sources 104, such as trees.

[0092] Ingesting or analyzing the structure may comprise the following, as may be described in method 700 of Fig. 7:Patent Application Attorney Docket No. FWIG-001W001

[0093] Step 702: Obtaining one or more images, such as satellite images, of structure102.

[0094] Step 704: Images may be provided to a machine learning system to detect and extract fuel sources 104.

[0095] Output encoding may include GIS layer(s) of vector polygons with an appropriate coordinate reference system (global projection) and / or may include GPS coordinates. A projection may provide correct cardinal orientation of the extracted features when projected onto the grid, computing correct lengths and angles of incidence, and correct overlaying of other data such as spectral band imagery and slope data.

[0096] Given a raster based geoTIFF or other geo-encoded vector or raster file, features may be raster mapped to a 300 feet x 300 feet grid at 2-foot resolution (conforming to industry defined 300-f00t home defensible space radius) with a defined geo-projection (World Geodetic System (WGS) 84 or other standard). Note that radiant heat flux threat computations use wall segment normals, which may be better computed pre-rasterization from vector polygons. For rasterization, map centroids to grid intersections-thus, e.g., with 400x400 50cm resolution (or better, such as 20cm resolution), grid points would be pixel centers in the original image.

[0097] Step 706: Analysis of the detected fuel sources may be completed and a fuel source dataset results.

[0098] Analysis may include computing fuel source / structure-segment (tile) distance vectors (using vector computation from points snapped (raster mapped) to a grid for example on structure tiles), and filtering fuel source / structure distance vectors to exclude vectors for wall segments shaded by other wall segments.

[0099] Fuel source centroids, wall and roof segments / tiles may be raster mapped (“snapped”) to grid points. However, distance vectors may be computed using Euclidean distance (and include direction), not taxicab geometry. The number of distance vectors xy is O(z * j) where i is the number of wall and roof segments and j is the number of fuel sources. Step 708: Manual processes may augment the automated dataset that may result from 702-706, such as be users 404 visiting structure 102 and doing an audit, such as via anPatent Application Attorney Docket No. FWIG-001W001 app accessible on computing devices. That may allow types of trees to be determined, and the like.

[0100] Method 700 may then return to method 500 where, at 510 a structure grid may be created, and structure data may be added to each applicable tile. Using the structure center (GPS coordinates for example), a grid of tiles may be created around the structure center. The structure grid may be set to extend a certain distance to ensure that relevant fuel sources 104 may be included in the risk assessment. For example, a structure grid that extends 300 feet radially from the structure center may be chosen, which may mean the grid extends beyond property boundaries. Tiles may be various sizes, for example 2 foot by 2 foot square tiles. Tiles may be any size and shape, provided that the number and shape of the tiles may be chosen to allow for accurate calculations and risk, mitigation and monitoring functionality, while balancing the computational challenges as the number of tiles increases. Tiles may be classified in many ways, for example as being a tile that forms part of the structure (structure tile, which may be further classified as wall structure tile, roof structure tile or wall / roof structure tile) and fuel source tile. This may allow more efficient computation at 512. Note that any tile may contain part of a wall and part of a roof, for example. Tiles may be “top down” tiles (i.e., from above) such that a depth of a tile (the vertical height) may be determined from the height of the house in that location.

[0101] At 510 data gathered at 506 and 508 may be placed in the grid.

[0102] Having ingested and created relevant data regarding the structure 102 (structure dataset) and the fuel sources 104 (fuel source data source), method 500 continues to perform the structure or property ignition risk assessment, also referred to as a property ignition model 204 (PIM), at 512.

[0103] PIM is premised upon the empirically determined tenet that ignition is a function of a structure’s structural features and spatial relationship to the immediately surrounding fuel sources. Furthermore, research has determined that ignition is caused by some combination of these principal ways in which fire spreads along the wildland-urban interface: (1) radiant heat aided by piloted ignition from embers, (2) direct flame impingement (convective heat), and (3) firebrands / embers entering the structure or accumulating on surfaces of the building structure.Patent Application Attorney Docket No. FWIG-001W001

[0104] These three modes of wildfire fire transmission form the physical basis of the thermodynamic threat vector analysis. Not all structural elements are vulnerable to all threat vectors, and given that each threat vector represents a different physical phenomenon, the impact of each threat vector may be modeled separately, starting at 802 of Fig. 8.

[0105] Flame front contact for direct flame impingement (convection, see 804 in method 800), determined by considering - True / False: Is there flame touch? (per wall segment in the, e.g., 2’x2' tiles in a grid). All of wind direction and magnitude, crown heat, canopy bulk density, flame length and flame angle may be computed using known approaches, making various assumptions, which may vary over time to increase accuracy. Such all may have an impact on threat vectors caused by convective heat, as may be seen by the impact of changing wind vectors.

[0106] Heat flux for radiant impact (see 806 in method 800), determined by considering Joules under / over to ignition (per wall segment in the, e.g., 2'x2' grid tiles).

[0107] The rate at which radiant heat (or any heat) is transmitted to a substance determines whether or not the temperature of the substance rises (the substance may absorb, transmit, reflect, and re-emit heat). If the heat per unit time (heat flux) is sufficiently large, the temperature of the receiving material may rise. The PIM determines ignition of a material at a location by whether or not total heat flux incident on that location causes the temperature of that material to exceed its ignition temperature.

[0108] From a fuel of given shape, emissivity, and temperature, the incident radiant heat flux at a structural element is a function of view factor (i.e., the share of total radiation from a source that is incident on a receiver / structure tile) which is in turn a function of source shape, structure shape, the relative orientations of the two, and the distance between them.

[0109] Heat flux from tree combustion decays with the inverse square of the distance to a structure feature.

[0110] Ember mass accumulation and size population for firebrand accumulation (see 808 in method 800), determined by considering kg of ember under / over to bum through (per roof tile segment in the, e.g., 2'x2' structure tiles in grid).Patent Application Attorney Docket No. FWIG-001W001[OHl] The failure calculation comprises both a load calculation and an ignition calculation. Fuel sources 104 may either be within ember throw distance or not. Those outside of ember throw distance may be ignored for these calculations.

[0112] Ember downrange mass projection in a given direction follows a LN distribution as may be known, based on variables such as wind speed. For PIM, embers at each distance may be considered to fall within a perpendicular range equal to the crown width along a path in the direction of the effective wind. Ember mass accumulation from tree z on surface element j falling within 12 crown width of the downwind line from the tree and less than 22.6m from the tree is metj = Q.QQ SMtideij where M is the mass of the tree or fuel source 104, and d is the ember mass share from fuel source deposited on a given tile.

[0113] Ember mass accumulation may occur on flat roof tops, gutters, gable junctures, re-entrant corners on wall sections, or anywhere with abrupt changes in wind velocity. Embers falling out of the air stream at wall sections may fall to the nearest horizontal surface creating piles on sills against doors and windows, decks against siding, and ground abutting footings. Accumulated embers transmit heat through conduction and radiation, both measured as a heat flux density. For modeling purposes, wall and roof segments may assume the area of the grid square. Tests on wood demonstrate reliable smoldering combustion at ember densities of 0.247g / cm2. For Mej / Aj > 0.247g / cm2bum- through occurs on roof segments made of wood or thin coatings over wood where Aj is the effective area of the roof or wall segment j (the grid area).

[0114] Fuel sources 104 whose ember throw distances reach the building structure may remain part of ember mass accumulation calculations, resulting in the impact on each tile of the building structure, with impacts ranging from innocuous, as in unaffected tiles, to failure, as in failed tiles.

[0115] Ember penetration probability computation (see 810 in method 800), determined by considering the number of embers under / over to bum through (per roof tile segment in the 2x2' structure tiles in grid). At any given downwind distance from a tree, ember particle mass has a distribution that can be modeled. Given the mass and mass distribution at a structure segment j, the population of a given ember size, the mass share at a distance for embers with mass less than 0.01g and populations of embers < 4mm at structure segment j may be computed. Various fuel sources 104 (fuel sources within ember throwPatent Application Attorney Docket No. FWIG-001W001 distance) impact ember penetration for a given structure tile (being a roof tile or soffit tile). The collective possibility of ember penetration is thus the sum of each fuel source 104 (fuel source within ember throw distance) releasing suitable embers.

[0116] Ignition from ember penetration through soffits may be modeled as a probability based on interior attic (rafterjoist, sheathing shaded area) space and can include: (1) temperature and / or relative humidity; (2) corresponding 1 hour dead fuel moisture content; (3) the probability of ignition (per ember) as determined by (1) and (2), and / or (4) the number of ember particles penetrating into the space.

[0117] Soffit panels are typically 12’ x 16". Seams per structure segment then are, on average, Ns = Sg / 12 where Sgis the grid size (side length). Assuming a 4mm soffit seam gap, the exposure area per segment area ratio may be calculated. The number of embers penetrating the soffit gaps then is NP= Nei,<4mm5x. The total ignition probability from ember penetration is then Pep= NPP(ignition|lh-MC, T) = 0.2NP. This value may be greater than 1 indicating a surplus of embers penetrating over and above what may cause ignition. Surplus may then be calculated as well.

[0118] Accuracy may improve as the feature detection algorithms (e.g., 506-508) improve and more data sources provide additional, and more accurate, data on the fire susceptibility of individual materials of structure 10, 308 characteristics of fuel sources 104, and knowledge about wildfire.

[0119] The likelihood of ignition of a tile by any of these vectors is a function of the data and variables encoded in a grid, as part of the structure dataset, the spatial relationships between them, and the specific attributes of each individual element. To exemplify this, consider fire spread from a large 1-ton tree to a structure 10, 308 with stucco siding. The PIM considers the total heat released by the tree under combustion, the heat capacity of stucco, the ignition temperature of treated wood, the distribution of ember size by distance from the tree, and several other attributes that, when taken together, fully characterize fire transmission between the tree and the structure (see 812 in method 800).

[0120] In practice, the PIM evaluates whether each exterior structural element of a given home ignites under the influence of each of the four threat vectors summed over all fuel sources with direct access to each exterior structural element (see 812 in method 800). ThePatent Application Attorney Docket No. FWIG-001W001 impact of each of the four threat vectors may be summed over all “line of sight” fuel sources (e.g., a tree “on the other side of the house” may have no impact) at each structure tile. Then, for each structure feature at that tile location, an ignition (failure) determination may be computed (see 814 in method 800). For radiant heat, the total incident heat flux may be used in combination with tables of structural material heat capacities to determine if ignition temperatures are reached. Breaking temperatures may be used in the case of window glass. If the flame front occupies the same coordinates as a structural tile, that tile may be assumed to ignite / fail. For roof tiles, accumulated ember mass may be used to determine whether or not burn-through occurs. For wall tiles, sub 0.01g ember populations may be used to probabilistically determine if embers penetrate vents. Each of the four vectors may be summed, for each structure tile, to arrive at an visual overall failure view.

[0121] Ignition failure determinations may evolve. For example, specific determinations for new materials or material interfaces (e.g., where brick and windows meet) may be added or updated to reflect advances in specific research in the area, without changing the overall PIM and approaches described herein.

[0122] Wind is an additional variable for the PIM. Since wind direction may affect the impact of any given fuel source on each roof or wall tile, threat vector outcomes may be computed nine separate times: once for wind from each of eight compass headings and a “no wind” case to cover worst case conditions for each structure tile. The model considers, for example, that a wind out of the north-east may yield maximum heat and ember transfer to structural elements on the north side of the house, but a south-west wind may yield little impact on those same elements. In that situation, the north-east results may be used for evaluation.

[0123] For each structure tile for which the model determines an ignition (failure), the PIM catalogues one or more of the following (see 816 in method 800): tile location; feature (or features) that failed; threat vector (or vectors) that caused each of the feature failure(s); individual fuel sources contributing, or summing, to each of the specific vector - feature failure(s); wind directi on(s) during failure; and / or failure Surplus - a measure of “how close” the feature came to non-failure / non-ignition. For example, a feature might fail at 2kJ / m2of accumulated heat flux over the combustion period. The model determines that the feature fails because the incident heat at that tile is 2.5kJ / m2. Thus the “failure surplus” is 0.5kJ / m2.Patent Application Attorney Docket No. FWIG-001W001The “failure surplus” may be useful for determining the remediation classification. Large surpluses indicate more extensive remediation may be needed (or may not be possible), whereas small surpluses may need less extensive remediation.

[0124] As part of 512, wildfire risk scoring may occur - taking the output of PIM and determining a risk of ignition, in a quantitative way (for example to allow quick comprehension, comparisons between properties and portfolios and mitigation or hardening impacts). Risk scoring may be based on one or more known risk scoring approaches, such as failure mode effects analysis, tailored to the assessment of wildfire risk for buildings such as structure 10, 308. Risk scoring may take the output of the PIM and calculate or compute the risk that a particular property, or amount of a portfolio, may ignite. In a risk assessment, surplus heat may be the characteristic that is primarily used.

[0125] Risk scoring may take into account, or highlight, differences in heat surplus between properties / portfolios. For example, a building that bums at Im joules and would expect to experience 4m joules of surplus heat in a wildfire (Building X) versus a building that burns at 500k joules of heat and would expect to experience 10k joules of surplus heat in a wildfire (Building Y). A given user 404 may view Building X as higher risk than Building Y, but if the absolute or percentage surplus heat was the same, Building X may be viewed as equally risky as Building Y, by a given user 404. A given user 404 may also set thresholds for acceptable risk, or for how buildings are scored or categorized based on surplus heat or surplus heat percentages, and the like. One user 404 may decide “high risk” is more than 500k joules of surplus or more than 20% surplus for a particular tile to fail, while another may set those values at 100k joules or 10%. System for wildfire risk assessment and mitigation 200 may allow for various parties and thresholds to be established and used - for example for mitigation and hardening, and monitoring.

[0126] With the modeling approach described above, each fuel source 104 and structural element or tile in scope may be backed by specific calculations for thermal energy output, probability of ignition, and remediation opportunity.

[0127] The rich level of data included in a structure dataset and assembled at 510 may be helpful because these calculations may vary considerably from item to item. For example, a short tree with a small crown may have a much smaller thermal energy output potential than a tall tree with a large crown.Patent Application Attorney Docket No. FWIG-001W001

[0128] All failure items may be compiled into a modified FMEA framework, and the relative risk of each item may be quantified in a risk priority number (see 818 in method 800).

[0129] The model may consider four discrete risk zones: (1) Home Ignition Zone (HIZ): The structure 10 itself and a boundary of 5 feet minimum from the structure 102, inside the line 902 (see Fig. 9); (2) Zone One: 5 feet to 30 feet from the structure 10, inside the line 904 and outside the line 902; (3) Zone Two: 30 to 100 feet from the structure 10, inside the line 906 and outside the line 904; (4) Zone Three: 100 to 500 feet from the structure 10, outside the line 906.

[0130] Once each threat vector risk score is computed, a cumulative risk score may be calculated for the entire property based on threat vector contributions to ignition failure points. The risk score may incorporate the compounding effect of risk from multiple threat vectors on a structure. This also helps in understanding contributors to ignition, as most items may fail due to a cumulative effect of heat flux from multiple sources. A risk score may be a single score, multiple scores, or a single score made up of multiple underlying scores. Scores may be classified as low / medium / high, or any other system. Multiple users 404 may set their own scores and classifications, based on the PIM outputs.

[0131] It is also important to understand that each failure item may be attached to multiple fuel sources and there may be overlap between them. The risk scoring may be in addition each individual failure item, to properly represent each failure and the point at which it fails. This allows showing multiple failures in the same location of a home. For example, a wall section (particular tile) where a window and siding both may fail, but at different points. Reducing one risk may or may not eliminate the adjacent risk. Importantly, the risk prioritization approach may augment a typical FMEA framework in two ways. First, traditional detectability scoring may be replaced with remediation scoring. This reflects the ability to address remediation in a failure event. Second, risk prioritization considers failure contributions and surplus amount to subsequent related structure failures. Because a wildfire advances inward from Zone 3 to the Home Ignition Zone, each zone has a quantifiable risk impact on its interior (i.e., zones 2, 1 and HIZ).

[0132] Method 800 may then return to method 500 where, at 514 results may be presented to a user 404, digitally on one or more screens of computing devices or physicallyPatent Application Attorney Docket No. FWIG-001W001 via written reports, where all supporting calculations, images, and features evaluated throughout the analysis may be retrieved and examined with each component for review. This allows a component-level view of risk for a property, and a component-level remediation plan to mitigate risks.

[0133] From a system and data perspective, at the end of this process each risk component (structure feature or tile or fuel source) may be a separate structured data object that may include a risk score, a “treated” risk score, and a mitigation treatment. These data objects and the granularity they represent serve as the foundation for MMS 216, informing the downstream integrated service delivery and logistics around approvals, scheduling, pricing, billing, and other services. This allows viewing of results in a more meaningful way than schemes that classify structures as High / Medium / Low risk, for example.

[0134] Additional details regarding wildfire risk assessment, PIM, and / or other details described herein may be disclosed in U.S. Patent Application Publication 2023 / 0023808, titled “System And Method For Wildfire Risk Assessment, Mitigation And Monitoring For Building Structures,” which is hereby incorporated by reference.

[0135] In one or more embodiments, the dimensions of a building structure can be determined using a combination of a dimensionally calibrated aerial image of the building structure and at least one ground-view image of the building structure. An example of a dimensionally calibrated overhead image 1000 of a building structure 1010 is shown in Fig. 10A. An example of a ground-view image 1050 of the building structure 1010 is shown in Fig. 10B

[0136] The dimensionally calibrated overhead image 1000 can provide a vertical or aerial view of the roof 1012 of the building structure 1010 and can show the perimeter of the building structure 1010. The dimensionally calibrated overhead image 1000 can also show fuel sources 1040 such as major vegetation (e.g., trees and / or shrubs) and / or secondary / ancillary structures such as a detached garage, a shed, and / or a fence. The dimensionally calibrated overhead image 1000 can be captured by a camera (or other image sensor) on a satellite, on a drone, on an airplane, or on a balloon.

[0137] The dimensionally calibrated aerial image 1000 has a known scale 1030 such that dimensions of an object in the dimensionally calibrated aerial image 1000 can bePatent Application Attorney Docket No. FWIG-001W001 determined. For example, first and second dimensions 1021, 1022 (e.g., length and width, respectively) of the building structure 1010, as measured with respect to first and second axes 1001, 1002, respectively, can be determined and / or calculated using the scale 1030.

[0138] The ground-view image 1050 can provide a horizontal view of the building structure 1010 including at least one exterior wall 1060 and the roof 1012 of the building structure 1010. The ground-view image 1050 can be captured by a camera, a smartphone, or another image-acquiring device (in general, a camera 1070), and can be handheld, mounted on a tripod, disposed on a robot, or disposed on a drone that flies at a low level. The camera can be a stereoscopic camera in one or more embodiments. When capturing a ground-view image 1050, the camera 1070 can be disposed in a camera plane 1072 that is parallel to a plane defined by the closest exterior wall 1060 to the camera 1070. The camera plane 1072 can also be parallel to a plane defined by the first and third axes 1001, 1003. The camera plane 1072 is oriented according to the compass bearing of the camera 1070 when the ground-view image was taken. When capturing a ground-view image 1050, the camera 1070 can also be disposed at a height or elevation, as measured with respect to a third axis 1003, that is equal to or below the roof 1012 (e.g., between the roof 1012 and ground 1080), for example to minimize skew of the exterior wall 1060 in the ground-view image 1050. In one or more embodiments, when capturing a ground-view image 1050, the camera 1070 can also be disposed at a height or elevation, as measured with respect to the third axis 1003, that is equal to or below 10 feet higher than the roof 1012 (e.g., between 10 feet higher than the roof 1012 and ground 1080).

[0139] In one or more embodiments, the ground-view image 1050 is not dimensionally calibrated. Even though the ground-view image 1050 may not be dimensionally calibrated, the ground- view image 1050 can be used to determine a third dimension 1023 (e.g., a height) of the building structure 1010. The third dimension 1023 is measured with respect to the third axis 1003. The axes 1001-1003 are mutually orthogonal.

[0140] As can be seen in Figs. 10A and 10B, the dimensionally calibrated overhead image 1000 and the ground-view image 1050 show a common dimension 1024 of the building structure 1010, in this case the first dimension 1021. In one or more other embodiments, the common dimension 1024 can be the second dimension 1022. Using the common dimension 1024, having a known value in the dimensionally calibrated overheadPatent Application Attorney Docket No. FWIG-001W001 image 1000 (e.g., using the scale 1030), the system can automatically determine a scale of the ground-view image 1050. For example, the scale can be determined as a ratio of the known value of the first dimension 1021 (e.g., the common dimension 1024) and a measurement (e.g., in inches or number of pixels) of the first dimension 1021 (e.g., the common dimension 1024), along or with respect to the first axis 1001. The value of the third dimension 1023 can be determined by taking an equivalent measurement of the third dimension 1023 (e.g., in inches or number of pixels, respectively), along or with respect to the third axis 1003 and multiplying that measurement by the calculated ratio / scale for the ground-view image 1050. Alternatively, the value of the third dimension 1023 can be determined by Equation 1.Value of Third Dimension = - x Known Value of First Dimension (1)Meas.of First Dimension where Meas, of Third Dimension is the measurement of the third dimension 1023 (e.g., in inches or number of pixels) in the ground-view image 1050, along or with respect to the third axis 1003, Meas, of First Dimension is the measurement of the first dimension 1021 (e.g., in inches or number of pixels, respectively) in the ground-view image 1050, along or with respect to the first axis 1001, and the Known Value of First Dimension is the known value of the first dimension 1021 (e.g., the common dimension 1024) determined from the dimensionally calibrated overhead image 1000. The Meas, of First Dimension and the Meas, of Third Dimension are in the same units (e.g., inches, number of pixels, or another unit). For example, when the measurement of the third dimension 1023 is 60% of the known value of the first dimension 1021 (e.g., the common dimension 1024) determined from the dimensionally calibrated overhead image 1000 is 100 feet, the value of the third dimension 1023 is 0.6*100 feet or 60 feet.

[0141] In one or more embodiments, one or more trained machine-learning (ML) models can be used to detect objects in the ground-view image 1050 and / or in the dimensionally calibrated overhead image 1000. For example, the trained ML model(s) can be used to detect fuel source(s) 1040 in the ground-view image 1050 and / or in the dimensionally calibrated overhead image 1000. The dimensions and position / location of a fuel source 1040, such as a tree, can be determined using a calculated scale of the ground-view image 1050 and / or the known scale 1030 of the dimensionally calibrated aerial image 1000.Patent Application Attorney Docket No. FWIG-001W001

[0142] In another example, the trained ML model(s) can be used to detect exterior building-structure features 1081, such as window(s) 1082, door(s) 1084, garage door(s) 1086, vent(s) 1088, deck(s) 1089 (Fig. 11), and / or balcony(ies). The dimensions and position / location of each exterior building-structure feature 1081 can be determined using a calculated scale of the ground-view image 1050 and / or the known scale 1030 of the dimensionally calibrated aerial image 1000.

[0143] In one or more embodiments, the fuel source(s) 1040 can be detected with a first trained ML model and the building-structure features 1081 can be detected with a second trained ML model that is different than the first trained ML mode. In one or more embodiments, the fuel source(s) 1040 can be detected with the same trained ML model (e.g., the first and second trained ML models are the same).

[0144] The dimensions 1021-1023 can be used to generate a dimensionally calibrated three-dimensional model 1100 of the building structure 1010, for example as shown in Fig.11. The dimensionally calibrated three-dimensional model 1100 can have a scale 1130 to determine the physical dimensions of the building structure 1010 and / or the physical dimensions and physical locations of the exterior building-structure features 1081 on the building structure 1010.

[0145] Fig. 12 is a flow chart of a method 1200 for determining a wildfire risk of a building structure according to one or more embodiments. The method 1200 can be performed by a computer.

[0146] In step 1201, an overhead image of a property including a building structure is received. The overhead image can be received from an overhead image database that is in communication (e.g., network communication) with the computer. The overhead image can be a dimensionally calibrated overhead image 1000. For example, the overhead image can have a known spatial scale such as a scale 1030.

[0147] In step 1202, the perimeter of the building structure in the overhead image is detected and / or segmented using one or more trained ML models. The perimeter of a building structure 1010 can be defined by vertices 1310A-D (in general, vertex / vertices 1310) that define a polygon 1300, for example as shown in Fig. 13. The polygon 1300 can include additional vertices 1310 in one or more embodiments, for example when the buildingPatent Application Attorney Docket No. FWIG-001W001 structure 1010 includes a bump-out, a wing, an addition, an attached garage, and / or other architectural feature(s). Each vertex 1310 has corresponding geospatial data (e.g., latitude and longitude data) that represents a comer 1090 (Figs. 10A, 10B) of the building structure 1010 and has a physical / actual geolocation on the property.

[0148] In step 1203, at least first and second dimensions of the building structure are determined using the overhead image and the known spatial scale. The first and second dimensions can be the same as the first and second dimensions 1021, 1022 (e.g., determined with respect to first and second axes 1001, 1002, respectively).

[0149] In step 1204, one or more ground-view images is / are received. Each groundview image can be the same as a ground-view image 1050. Each ground-view image shows at least one exterior wall of the building structure. Each ground-view image can also show a roof of the building structure and / or one or more building-structure features on the external wall(s).

[0150] In optional step 1205, a third dimension of the building structure can be determined using at least one of the ground-view image(s) and a common dimension (e.g., a common dimension 1024) that is shown in both the overhead image the ground-view image. The third dimension can be the same as the third dimension 1023 (e.g., determine with respect to a third axis 1003). For example, the third dimension can be determined using Equation 1 or a calculated scale of the ground-view image.

[0151] In step 1206, exterior building-structure features are detected in the groundview image(s) on one or more of the exterior walls. The exterior building-structure features can be the same as exterior building-structure features 1081. The exterior building-structure features can be detected using a trained ML model, which can be the same as or different than the trained ML model used in step 1202 to detect a building-structure perimeter in an overhead image.

[0152] In step 1207 (via placeholder A), one or more dimensions of each detected exterior building-structure feature is / are determined. The physical dimensions and physical position of each detected exterior building-structure feature can be determined using a calculated spatial scale of the respective ground-view image. Alternatively, each detected exterior building-structure feature can be represented as a percentage range of pixels in thePatent Application Attorney Docket No. FWIG-001W001 image, which can be used to calculate the physical dimensions and physical position of a respected detected exterior building-structure feature.

[0153] In step 1208, a dimensionally calibrated three-dimensional model of the exterior of the building structure is generated. The dimensionally calibrated three- dimensional model can be generated using the first, second, and third dimensions of the building structure and the detected exterior building-structure features (e.g., including their respective dimensions and positions).

[0154] In step 1209. fuel sources on the property are detected in the overhead image and / or in the ground-view image(s). The fuel sources can be the same as fuel sources 1040. The fuel sources can be detected with a trained ML model. The trained ML model used to detect the fuel sources can be the same as or different than the trained ML model that can be used in step 1202 to detect a building structure in an overhead image and / or than the trained ML model that can be used in step 1206 to detect exterior building-structure features in ground-view images. The fuel sources can be detected in the overhead image and / or in the ground view images. The fuel sources can include major vegetation, one or more non- vegetative fuels such as shipping pallets, fuel containers, industrial drums, vehicles, wood piles, and / or furniture, and / or one or more secondary structures.

[0155] In step 1210, a distance between each detected fuel source and the nearest exterior wall of the building structure is calculated, for example using the known spatial scale of the overhead image and / or a calculated spatial scale of a respective ground-view image.

[0156] In step 1211, the fuel load for each detected fuel source is calculated or estimated. The fuel load can be determined / calculated according to step 508 and / or method 700 by determining feature-specific attributes of each fuel source and the thermal energy (e.g., BTU) generation potential.

[0157] In step 1212, a fire-risk metric for the building structure is determined. The fire-risk metric can be determined structure based, at least in part, on the respective distance between each fuel source and a respective nearest exterior wall of the building structure and the respective fuel load for each fuel source. The fire-risk metric can be determined according to step 512 (e.g., risk scoring) and / or method 800.Patent Application Attorney Docket No. FWIG-001W001

[0158] Fig. 14 is a flow chart of a method 1400 for determining one or more dimensions of a detected exterior building-structure feature on an exterior wall segment that extends between two vertices that are viewable in a ground-view image, according to one or more embodiments. The building-structure feature is detected in a ground-view image. In one or more embodiments, step 1207 can be performed according to method 1400. Method 1400 can be performed when the camera is approximately centered and orthogonal to an exterior wall of a building structure.

[0159] In step 1401, the geospatial camera location of the camera when the groundview image was captured, the FOV angle of the camera when the ground-view image was captured, and the compass bearing of the camera when the ground-view image was taken are determined using input data. Additionally or alternatively, the input data can include an identification of the exterior wall(s) that is / are captured in the ground-view image. The input data can include user-input data, metadata of the ground-view image, and / or camera data associated with the ground-view image.

[0160] In step 1402, the location of the exterior wall captured in the ground-view image is located on the polygon using input data that includes the FOV of the camera when the ground-view image was taken, the geospatial data of the vertices, and an identification of the exterior wall(s) that is / are captured in the ground-view image. The input data can include user-input data, metadata of the ground-view image, and / or camera data associated with the ground-view image. In one or more embodiments, the input data can further include geospatial data representing the physical location of the camera when the ground-view image was taken and / or compass bearing data representing the compass bearing (or approximate compass bearing) of the camera when the ground-view image was taken.

[0161] The input data can include user-input data, metadata of the ground-view image, and / or camera data associated with the ground-view image. The terms FOV and FOV angle can be used interchangeably.

[0162] In one or more embodiments, the exterior wall represented in the ground-view image can be determined by locating the exterior wall that extends between the two vertices that are closest spatially to the geospatial position of the camera. For example, in Fig. 15A vertices 1310C and 1310D are closest spatially to the geospatial position of the camera 1320 that captured the image, and thus it can be assumed that the exterior wall segment 1330Patent Application Attorney Docket No. FWIG-001W001 between vertices 13 IOC and 1310D is the exterior wall represented in a ground-view image taken by the camera 1320. The camera 1320 can comprise a standalone digital camera or can comprise a digital camera that is incorporated into another electronic device, such as a smartphone, a laptop, a tablet, or other electronic device.

[0163] The angle of the FOV can be determined relative to a line or plane 1520 of the camera 1320 according to the compass bearing of the camera when the ground-view image was taken, for example with 0 degrees and 180 degrees defined as shown in Figs. 15A and 15B. The line / plane 1520 can have another orientation relative to the polygon 1300 in one or more other embodiments.

[0164] It can also be determined that the camera location is centered and orthogonal to the exterior wall segment 1330 between vertices 1310C and 1310D because the distance along a first line / ray 1500 A from the camera location to the vertex 1310C is equal to or approximately equal to (e.g., within 10% of the relevant value) the distance along a second line / ray 1500B from the camera location to the vertex 1310D. Lines / rays 1500A, 1500B can be referred to generally as lines / rays 1500.

[0165] In step 1403, the boundary of the exterior wall in the ground-view image is segmented in the ground-view image. An example boundary 1530 of an exterior wall segment 1330 is shown in Fig. 15B. The boundary 1530 is shown as slightly outside of the lines for the exterior wall segment 1300 so as to not obscure the boundary 1530. The boundary of the exterior wall in the ground-view image can be segmented using a trained ML model.

[0166] In step 1404, the length of the exterior wall represented in the ground-view image is determined. The length can be determined using the geospatial data of the vertices that define the exterior wall represented in the ground-view image. In Fig. 13, the length 1350 of the exterior wall segment 1330 between vertices 1310A and 1310B is measured with respect to the first axis 1001. An example FOV 1322 of the camera 1320 is shown in Figs. 15A and 15B, which can represent a pinhole model of the camera 1320.

[0167] In step 1405, a scale is determined or calculated for the ground-view image. The scale can represent the length of an individual pixel in the ground-view image and can be calculated using the length of the exterior wall segment (calculated in step 1404) and numberPatent Application Attorney Docket No. FWIG-001W001 of pixels in the ground-view image, along or parallel to the first axis, that represent the length 1350 of the exterior wall segment 1330 in the ground-view image. The length per pixel can be calculated according to Equation 2.where pLis the length of an individual pixel in the same units as WL, the length of the exterior wall segment 1330, and npis the number of pixels the object detection detected as the width of the exterior wall segment 1330.

[0168] In step 1406, the number of pixels corresponding to a dimension, such as a length, of one or more building-structure features, detected on the exterior wall segment 1330 viewable in the ground-view image, is determined. The building-structure(s) can be detected using a trained ML model, for example in step 1206.

[0169] For example, the number of pixels corresponding to a length 1552 of a garage door 1086 is determined in the ground-view image. The position of the garage door 1086 (or other a building-structure feature 1081) can be represented as with a first (x,y) coordinate representing a relative percentage of pixels in the (x,y) image space of the ground-view image. The width / length and height of the garage door 1086 (or other a building-structure feature 1081) can also be represented as with a second (x,y) coordinate representing a relative percentage of pixels in the (x,y) image space of the ground-view image. For example, a first coordinate of (0.1, 0.5) indicates that the garage door 1086 (or other a building-structure feature 1081) begins at 10% of the pixels in the X direction (e.g., relative to an origin in the bottom-hand side of the ground-view image) and at 50% of the pixels in the Y direction. A second coordinate of (0.2, 0.2) indicates that the garage door 1086 (or other a buildingstructure feature 1081) extends for 20% of the pixels, from its beginning at the first coordinate of 10%, in the X direction (e.g., with respect to the first axis 1001) and for 20% of the pixels, from its beginning at the first coordinate of 50% of the pixels, in the Y direction (e.g., with respect to the third dimension 1003).

[0170] In this example, the length / width of the garage door 1086 is 20% of the pixels in the X direction. If the image resolution is 1,000 pixels, then the length / width of the garage door 1086 is 200 pixels.Patent Application Attorney Docket No. FWIG-001W001

[0171] In step 1407, the dimension(s) and / or the location of the building-structure feature is determined. The length of the garage door 1086 (e.g., in the x direction along / parallel to the first axis 1001) can be calculated by Equation 3.where Pfeaturelengtllisthe percentage length in pixels of the garage door 1086 (or other a building-structure feature 1081) and presoiution is the image resolution (total number of pixels) in the X dimension. The height of the garage door 1086 can be calculated using Equation 3 but substituting the y pixel percentage (Pfeatureheight) for the x pixel percentage (PfeatUrelength) and the y pixel resolution (preSoiutionY) for the x pixel resolution (presoiution_x) as shown in Equation 4.F eature _height — pLxpfeatureheightxpresoiution_Y (4)

[0172] The position of the garage door 1086 is determined in the same manner as the dimensions but using the respective first coordinate (x or y value depending on which component of the position is being determined) in Equations 5 and 6.F eatUre _pOSitiOTl_X p^ XPfeatureXpositionPresolution_X (5)Feature _position_Y = pLxpfeatureYpositionxpresolution Y(6) where PfeatureXposi tlonposition of the garage door 1086 (or other a buildingstructure feature 1081) as represented as relative percentage of pixels in the x image space of the ground-view image (e.g., relative to the first dimension 1001) and PfeatureYpositionP°siti°nof the garage door 1086 (or other a building-structure feature 1081) as represented as relative percentage of pixels in the y image space of the ground-view image (e.g., relative to the third dimension 1002). The first coordinate (x, y) is the same as PfeatureXposition, PfeatureYposition)- The structural feature length and height can be used to determine the surface area of the garage door 1086 (or other a building-structure feature 1081).

[0173] Fig. 16 is a flow chart of a method 1600 for determining one or more dimensions of a detected exterior building-structure feature on one or more exterior wall segment that are viewable in the ground-view image, according to one or more embodiments.Patent Application Attorney Docket No. FWIG-001W001The building-structure feature is detected in a ground-view image that shows two exterior wall segments that extend to respective vertices. In one or more embodiments, step 1207 can be performed according to method 1600.

[0174] In step 1601, the geospatial camera location of the camera when the groundview image was captured, the FOV angle of the camera when the ground-view image was captured, and the compass bearing of the camera when the ground-view image was taken are determined using input data. Additionally or alternatively, the input data can include an identification of the exterior wall(s) that is / are captured in the ground-view image. The input data can include user-input data, metadata of the ground-view image, and / or camera data associated with the ground-view image.

[0175] In step 1602, the location of the exterior walls on the polygon that are visible in the ground-view image is determined using the geospatial camera location of the camera when the ground-view image was captured, the FOV angle of the camera when the groundview image was captured, and the compass bearing of the camera when the ground-view image was taken.

[0176] In one or more embodiments, the location of the exterior walls on the polygon visible in the ground image can be determined by defining lines or rays 1700 that extend from the geospatial location of the camera 1320 that took the ground-view image to each vertex 1310 of the polygon 1300, for example as shown in Fig. 17. Since all lines / rays 1700 are within the FOV 1322 at the compass bearing of the camera when the ground-view image was taken, the vertex 1310 that is closest to the start (left-most edge 1324 of the FOV 1322 from the perspective of the camera when viewing the building structure) of the FOV 1322 defines one end of the first viewable exterior wall segment. The closest point 1310 to the left-most edge 1324 of the FOV 1322 can be determined by scanning each ray from the start of the FOV until an intersection with the polygon 1300 is found. Vertex 1310A is the closest vertex 1310 to the left-most edge 1324 of the FOV 1322. In the case two vertices lie along the same ray, the closer to the camera of the two is chosen.

[0177] Moving counter-clockwise from vertex 1310A along the polygon 1300, the next vertex encountered is 1310C. The pair of vertices 1310A and 1310C define the wall segment 1730D. The subFOV angle 1710D for segment 1730D is recorded, as is the start and end angles of subFOV 1710D with respect to the line through the camera perpendicular to thePatent Application Attorney Docket No. FWIG-001W001 camera bearing. By convention, all angles are measured in a clockwise direction. For example, the start of subFOV 1710D is less than the end of subFOV 1710D, but the start of subFOV 1710B is greater than the end of subFOV 1710B. Continuing counterclockwise along the polygon 1300 to the next vertex (1310C to 1310D), the next segment in the polygon is identified and in the same way the subFOV is recorded for that segment. This process continues until all segments 1730A-1730D and subFOV angles 1710A-1710D are recorded.

[0178] The wall segment beginning at the closest vertex to the left-most edge 1324 of the FOV 1322 is visible by necessity. Moving counterclockwise around the polygon compare the end (relative to the line through the camera perpendicular to the camera bearing) of the subFOV of the next wall segment B to the end subFOV angle of the previous wall segment A. If the end subFOV angle B is greater than the end subFOV A, then wall segment B is visible; if it is less than or equal to subFOV angle A, wall segment B is occluded.

[0179] The process is repeated until either all of the segments in the polygon have been checked or the end angle is greater than or equal to the end (right-most edge 1326 of the FOV 1322 from the perspective of the camera when viewing the building structure) of the FOV angle 1322. The viewable wall segments in Fig. 17 would therefore be 1730D and 1730A. Vertices 1310A, 1310C, and 1310D are defined as viewable vertices 1710A, 1710C, and 1710D, respectively. Vertex 1310B is defined as a non-viewable vertex 1710B.

[0180] It is noted that the process of determining viewable wall segments can be performed by starting at the right-most edge 1326 of the FOV 1322 and determining the closest vertex to the right-most edge 1324 of the FOV 1322 by scanning each line / ray 1700 from the right-most edge 1324 of the FOV 1322 until an intersection with the polygon 1300 is found. Vertex 1310D is the closest vertex 1310 to the right-most edge 1326 of the FOV 1322. The process can continue as described above moving clockwise from vertex 1310D along the polygon 1300. In one or more other embodiments, the process of determining viewable wall segments can be performed by starting at the vertex 1310C that is closest to the geospatial location of the camera 1320 and moving counter-clockwise (or clockwise) from vertex 1310C along the polygon 1300 in the same manner as described above.

[0181] In step 1603, the boundary of the exterior walls in the ground-view image are segmented in the ground-view image. The boundary of the exterior walls in the ground-view image can be segmented using a trained ML model.Patent Application Attorney Docket No. FWIG-001W001

[0182] In step 1604, the subFOVs corresponding to the exterior wall segments 1730 that are viewable in the ground-view image. In this example, it is determined that subFOVs 1710A and 1710D represent and / or define the exterior wall segments 1730A, 1730D, respectively that are viewable in the ground-view image. Vertices 1310A, 1310C, and 1310D can be defined as viewable vertices 1710A, 1710C, and 1710D, respectively.

[0183] In step 1605, the length of the exterior walls represented in the ground-view image are determined. The length can be determined using the geospatial data of the vertices. In Fig. 17, the length of the exterior wall segment 1330 between vertices 1310A and 1310C can be measured with respect to the second axis 1002. The length of the exterior wall segment 1330 between vertices 1310C and 1310D can be measured with respect to the first axis 1001.

[0184] In step 1606, one or more physical dimensions and / or locations of the detected building-structure feature is / are determined.

[0185] The position of the detected building-structure feature can be determined by defining a plurality of vectors 2300 between the geospatial location of the camera 1320 and a viewable external wall segment 1330, as shown in Fig. 23.

[0186] A vector 2300C extends between the geospatial location of the camera 1320 and a first vertex 1310C that partially defines the viewable external wall segment 1330. A vector 2300D extends between the geospatial location of the camera 1320 and a second vertex 1310D that partially defines the viewable external wall segment 1330. In one or more other embodiments, the vector 2300C and / or the line 2300D can extend between the geospatial location of the camera 1320 and a respective intersection point on the viewable external wall segment 1330, the respective intersection point defining a respective bounds of a viewable portion of an external wall segment.

[0187] In one or more embodiments, the lateral pixels defining the pixel length start location and the pixel length of the detected building-structure feature 1081 are projected onto a corresponding viewable exterior wall segment using the FOV angle of the camera. For example, when a detected building-structure feature has a starting pixel location at 0.25 (25%) of the ground image edge and the detected building-structure feature has a ending pixel location of 0.30 (30%). If the FOV angle of the camera is 60 degrees, the detectedPatent Application Attorney Docket No. FWIG-001W001 building-structure feature can be defined by a pixel start angle 2301 of 15 degrees (60 degrees x 0.25) and a pixel end angle 2302 of 18 degrees (60 degrees x 0.30).

[0188] A pixel start angle 2301 is defined between line 2300C and a vector 2300A that extends between the geospatial location of the camera 1320 and a first edge 2311 of the detected building-structure feature 1081. A pixel end angle 2302 is defined between vector 2300A and a vector 2300B that extends between the geospatial location of the camera 1320 and a second edge 2312 of the detected building-structure feature 1081. An angle 2303 between vectors 2300 A, 2300B defines a length 2381 of the detected building-structure feature 1081. The angle 2303 can represent an effective subFOV angle of the detected building-structure feature 1081. An angle 2304 between vectors 2300B and 2300D defines a width of the portion 2331 of the viewable external wall segment 1330. Angle 2301 defines a width of the portion 2332 of the viewable external wall segment 1330.

[0189] The position of detected building-structure feature 1081 on the corresponding wall segment 1330 can be determined according to Equations 7-11. For simplicity of these equations, pO represents the geospatial location of the camera 1320, pl represents vertex 1310C, p2 represents vertex 1310D, p3 represents the first edge 2311 of the detected building-structure feature 1081, p4 represents the second edge 2311 of the detected buildingstructure feature 1081, y represents angle 2301, represents angle 2303, and <J represents angle 2304.

[0190] The position of the detected building-structure feature 1081 is defined by theVectors p0p3 and p0p4. The distance between points p3 and p4 represents the length of thePatent Application Attorney Docket No. FWIG-001W001 detected building-structure feature 1081. The distance between points p3 and p4 can be determined using angle and the lengths of vectors pOp3 and pOp4.

[0191] In one or more alternative embodiments, the length 2381 of the detected building-structure feature 1081 can be determined using Equations 2 and 3.

[0192] In one or more embodiments, the height and / or vertical position of the detected building-structure feature 1081 can be determined in the same manner using the vertical FOV, vertical camera bearing, and vertical wall segment representation (e.g., by rotating the axes shown in Fig. 23 such that p3 and p4 represents the first and second vertical edges of the detected building structure and pl and p2 represent the top and bottom edges of the viewable exterior wall segment).

[0193] Fig. 18 is a flow chart of a method 1800 for determining one or more dimensions of a detected exterior building-structure feature on a viewable portion of an exterior wall segment that extends to only one vertex that is viewable in the ground-view image, according to one or more embodiments. In one or more embodiments, step 1207 can be performed according to method 1800.

[0194] In step 1801, the geospatial camera location of the camera when the groundview image was captured, the FOV angle of the camera when the ground-view image was captured, and the compass bearing of the camera when the ground-view image was taken are determined using input data. Additionally or alternatively, the input data can include an identification of the exterior wall(s) that is / are captured in the ground-view image. The input data can include user-input data, metadata of the ground-view image, and / or camera data associated with the ground-view image.

[0195] In step 1802, the location of the exterior wall(s), shown in the ground-view image, on the polygon is / are determined using geospatial data that represents the physical location of the camera when the ground-view image was taken, the FOV of the camera when the ground-view image was taken, and the geospatial data of the vertices.

[0196] In one or more embodiments, the location of the exterior walls on the polygon can be determined by defining lines or rays 1900 that extend from the geospatial location of the camera 1320 that took the ground-view image to each vertex 1310 of the polygon 1300, for example as shown in Fig. 19. Each line / ray 1900 can be referred to as a vertex line / rayPatent Application Attorney Docket No. FWIG-001W0011900. Each pair of vertex lines / rays 1900 defines a respective vertex angle or a subFOV angle. A vertex angle 1950 can be defined with respect to each vertex line / ray 1900 that falls outside of the FOV 1322 at the compass bearing of the camera when the ground-view image was taken. SubFOV angles 1910A-1910C are defined within the FOV 1322 at the compass bearing of the camera when the ground-view image was taken.

[0197] Since the lines / rays 1900 to vertices 1310A, 1310C are not within the FOV 1322 at the compass bearing of the camera when the ground-view image was taken, it can be determined that those vertices 1310A, 1310C are not viewable in the ground-view image. Thus, only vertex 1310B and / or 1310D is viewable in the ground-view image. To determine which of the wall segments or partial wall segments are viewable, find the closest point of intersection 1934 on the ray 1935 corresponding to the edge 1924 of the FOV 1322 that passes through an external wall segment 1330. The point of intersection 1934 is on the external wall segment 1330 that extends between vertices 1310C, 1310D. Thus, the external wall segment 1330 that extends between vertices 1310C, 1310D is defined as the external wall segment 1330 shown, in part, in the ground-view image.

[0198] In step 1803, a new vertex 1936 is defined and / or created at the intersection point 1934.

[0199] Using this new vertex 1936 as the nearest vertex to the left edge 1924 of the FOV, the process is the same as described in 1603 through 1606 for identifying visible wall segments (in this case partial wall segment 1932 is the only viewable wall segment or partial wall segment in the ground-view image). Thus, steps 1804-1807 are the same as steps 1603- 1606, respectively. In yet another implementation, it should be appreciated that if the camera position, FOV and bearing were such that the vertex 1310D were not within the camera FOV (e.g., if the right edge of the FOV in Fig. 19 were more to the left), a virtual vertex defined / created in a manner similar to the vertex 1934 could be used as a new vertex.

[0200] It should also be appreciated that if at least one vertex does appear in the ground-view image, it is possible to perform some degree of error correction, if warranted, in connection with the camera location. For example, if the camera location (e.g., as provided by metadata) is incorrect (e.g., off by a few feet), the presence of an actual vertex in the ground-view image allows an operator to adjust the camera location so that the actual vertexPatent Application Attorney Docket No. FWIG-001W001 present in the ground-view image lines up with the corresponding vertex of the polygon (and thereby adjust for any error in the camera location provided by metadata).

[0201] The position of the detected building-structure feature can be determined by defining a plurality of vectors 2200 between the geospatial location of the camera 1320 and a viewable external wall segment 1330, as shown in Fig. 23.

[0202] The location p4 at intersection 1934 can be calculated using the following Equations 12-14. For simplicity of these equations, pO represents the geospatial location of the camera 1320, pl represents vertex 1310C, p2 represents vertex 1310D, p4 represents the intersection point 1934, represents vertex angle 1910A, < represents the angle between ray 2200C that extends from the geospatial location of the camera 1320 (pO) to vertex 1310C (pl), ray 2200D extends from the extends from the geospatial location of the camera 1320 (pO) to vertex 1310D, as shown in Fig. 22.

[0203] In Equation 13, Z represents the length 2230 of the viewable partial -wall segment 1932.

[0204] Fig. 20 is a flow chart of a method 2000 for determining one or more dimensions of one or more detected exterior building-structure features on an exterior wall segment according to one or more embodiments. In one or more embodiments, step 1207 can be performed according to method 2000.

[0205] In step 2001, the location of the camera that captured a ground-view image of a building structure and the location of the vertices 1310 of the polygon 1300 that defines the perimeter of a building structure are determined using respective geospatial data.

[0206] In step 2002, vertex lines / rays (e.g., lines / rays 1500, 1700, 1900) are defined between the camera location and each vertex 1310 of the polygon.Patent Application Attorney Docket No. FWIG-001W001

[0207] In step 2003, the FOV of the camera when it captured the ground-view image is determined, for example from metadata of the ground-view image.

[0208] In step 2004, it is determined whether all lines / rays are within the FOV of the camera. If at least one line / ray is not within the FOV of the camera (e.g., the line / ray 1900 to vertex 1310A in Fig. 19) (i.e., step 2004=no), then it is determined in step 2005 whether at least one line / ray is within the FOV of the camera. If at least one line / ray is within the FOV of the camera (e.g., the line 1900 to vertex 1310D in Fig. 19) (i.e., step 2005=yes), the dimensions of the detected exterior building-structure feature(s) are determined using method 1800. If none of the lines / rays are within the FOV the camera (i.e., step 2005=no), then it is determined that the building structure is not shown in the ground-view image and the groundview image is not processed to determine the dimensions of a detected exterior buildingstructure feature(s). An example where none of the lines / rays 1900 are within the FOV 1322 of the camera 1320 is shown in Fig. 21. Fig. 21 is the same as Fig. 19 except that the camera 1320 is oriented differently to change the position of the FOV 1322 with respect to the polygon 1300. Vertex angles 1950 and subFOV angles 1910A-C are not shown in Fig. 21 due to the change in orientation of the FOV 1322 and for simplicity.

[0209] If all lines are within the FOV of the camera (i.e., step 2004=yes), then in step 2006 it is determined whether the camera was orthogonal to and centered on an external wall captured in the ground-view image. Step 2005 can be performed by calculating the distance between the camera location and the two vertices closest to the camera location. The camera was orthogonal to and centered on the external wall when the distances between the camera location and each vertex are equal or approximately equal (e.g., within 10% of the relevant value) to one another. If the camera was orthogonal to and centered on the external wall (i.e., step 2005=yes), then the dimensions of the detected exterior building-structure feature(s) are determined using method 1400. If not (i.e., step 2005=no), then the dimensions of the detected exterior building-structure feature(s) are determined using method 1600.

[0210] Example Embodiments

[0211] Example 1. A computer-implemented method for determining a wildfire risk for a building structure, comprising: receiving an overhead image from an overhead image database, the overhead image database in network communication with the one or more processors, the overhead image including geospatial data; determining, using the overheadPatent Application Attorney Docket No. FWIG-001W001 image of a property including the building structure, first and second dimensions of the building structure, the first and second dimensions measured with respect to first and second axes, respectively, the overhead image having a known spatial scale; determining, using one or more ground-view images of the building structure, a third dimension of the building structure, the third dimension measured with respect to a third axis, the first, second, and third axes mutually orthogonal to each other, each ground-view image and the overhead image sharing a common dimension of the building structure; detecting, using one or more trained machine-learning (ML) models and the one or more ground-view images, structural features on one or more exterior walls of the building structure; generating a dimensionally calibrated three-dimensional model of an exterior of the building structure using the first, second, and third dimensions and the structural features; detecting, using the one or more trained ML models and at least the overhead image, fuel sources on the property; calculating a respective distance between each fuel source and a respective nearest exterior wall of the building structure using the dimensionally calibrated three-dimensional model; calculating a respective fuel load in terms of thermal energy generation potential for each fuel source; and determining a wildfire risk metric for the building structure based, at least in part, on the respective distance between each fuel source and a respective nearest exterior wall of the building structure and the respective fuel load for each fuel source.

[0212] Example 2. The method of Example 1, further comprising automatically determining respective dimensions of each structural feature, the respective dimensions determined relative to a respective common dimension of the building structure in a respective ground-view image, wherein the wildfire risk metric for the building structure is further determined based, at least in part, on the respective dimensions of each structural feature.

[0213] Example 3. The method of Example 1 or 2, further comprising determining a total number of pixels in the respective ground-view image corresponding to the respective common dimension of the building structure; calculating a respective pixel length as a ratio of the respective common dimension to the total number of pixels; determining a respective first number of pixels in the respective ground-view image corresponding to a respective first dimension of each structural feature; and determining the respective first dimension of eachPatent Application Attorney Docket No. FWIG-001W001 structural feature as a product of the respective first number of pixels in the respective ground-view image and the respective pixel length in the respective ground-view image.

[0214] Example 4. The method of any of the previous Examples, further comprising determining a respective second number of pixels in the respective ground-view image corresponding to a respective second dimension of each structural feature; and determining the respective second dimension of each structural feature as a product of the respective second number of pixels in the respective ground-view image and the respective pixel length in the respective ground-view image.

[0215] Example 5. The method of any of the previous Examples wherein the one or more structural features includes one or more windows, one or more doors, one or more decks, one or more balconies, one or more vents, and / or one or more garage doors.

[0216] Example 6. The method of any of the previous Examples, further comprising determining a respective position of each structural feature on the one or more exterior walls.

[0217] Example 7. The method of any of the previous Examples wherein the fuel sources include major vegetation, one or more non-vegetative fuels, and / or one or more secondary structures, the one or more non-vegetative fuels including one or more shipping pallets, one or more fuel containers, one or more industrial drums, one or more vehicles, one or more wood piles, and / or one or more pieces of furniture.

[0218] Example 8. A system for determining a wildfire risk of a building structure, comprising: one or more processors; non-transitory memory operably coupled to the one or more processors, the non-transitory memory storing computer-readable instructions that, when executed by the one or more processors, cause the one or more processors to: receive a geospatially calibrated overhead image from an overhead image database, the overhead image database in network communication with the one or more processors, the geospatially calibrated overhead image including an overhead view of a property including the building structure; segment, using one or more trained machine-learning (ML) models, the geospatially calibrated overhead image to define a polygon that represents a perimeter of the building structure, the polygon including a plurality of vertices having respective geospatial vertex locations, each vertex corresponding to a respective corner of the building structure; receive a ground-view image of the building structure, the ground-view image showing onePatent Application Attorney Docket No. FWIG-001W001 or more viewable external wall segments, each viewable external wall segment extending between a respective viewable pair of the vertices; determine, using input data, a geospatial location of a camera where the ground-view image was captured, a field-of-view (FOV) angle of the camera when the ground-view image was captured, and a compass bearing of the camera when the ground-view image was captured; segment, using the one or more trained ML models, the ground-view image to determine a boundary of the one or more viewable external wall segments and to detect at least a structural feature on a first viewable external wall segment, the structural feature having a pixel length in the ground-view image; define a plurality of lines, each line extending between the geospatial camera location and a respective vertex; define a plurality of subFOV angles, each subFOV angle defined between a respective pair of the lines; determine, using the geospatial camera location, the respective geospatial vertex locations, the subFOV angles, the FOV angle, and the compass bearing, a respective location of each viewable external wall segment, shown in the ground view image, on the polygon; calculate a respective length of each viewable external wall segment shown in the ground view image, the respective length determined using the respective geospatial vertex locations for the respective viewable pair of the vertices; determine a first subFOV angle defined between a first pair of the lines that extend to a first viewable pair of the vertices that define the first viewable external wall segment; determine a length of the structural feature using a first length of the first viewable external wall segment, the first subFOV angle, the FOV angle, and the pixel length of the structural feature; generate a dimensionally calibrated three-dimensional model of an exterior of the building structure using the polygon and the length of the structural feature on the first viewable external wall segment; detect, using the one or more trained ML models and at least the overhead image, one or more fuel sources on the property; calculate a respective distance between each fuel source and a respective nearest exterior wall of the building structure using the dimensionally calibrated three-dimensional model; calculate a respective fuel load in terms of thermal energy generation potential for each fuel source; and determine a wildfire risk metric for the building structure based, at least in part, on the respective distance between each fuel source and a respective nearest exterior wall of the building structure and the respective fuel load for each fuel source.

[0219] Example 9. The system of Example 8 wherein the structural feature has an image position in the ground-view image defined, in part, by a pixel start location and the pixel length, and the computer-readable instructions, when executed by the one or morePatent Application Attorney Docket No. FWIG-001W001 processors, further cause the one or more processors to determine a lateral physical location of the structural feature on the first viewable external wall segment using the first length of the first viewable external wall segment, the first subFOV angle, the FOV angle, the pixel start location, the pixel length, and the pixel resolution of the ground-view image.

[0220] Example 10. The system of Example 8 or 9, wherein the pixel start location is a lateral pixel start location, the image position is further defined by a vertical pixel start location and a pixel height, and the computer-readable instructions, when executed by the one or more processors, further cause the one or more processors to determine a vertical physical location of the structural feature on the first viewable external wall segment using the length of the structural feature, the vertical pixel start location, the pixel height, the pixel length, and the pixel resolution of the ground-view image.

[0221] Example 11. The system of any of Examples 8-10 wherein the computer- readable instructions, when executed by the one or more processors, further cause the one or more processors to: calculate, using the geospatial camera location and the respective geospatial vertex locations, a respective distance between the geospatial camera location and each vertex; determine a closest vertex to the geospatial camera location within the FOV angle; and determine the respective location of each viewable external wall segment on the polygon based, at least in part, on the closest vertex to the geospatial camera location.

[0222] Example 12. The system of any of Examples 8-11 wherein the computer- readable instructions, when executed by the one or more processors, further cause the one or more processors to: compare (a) a first distance between the geospatial camera location and a first neighboring vertex to (b) a second distance between the geospatial camera location and a second neighboring vertex, each of the first and second neighboring vertices comprising an adjacent vertex to the closest vertex on the polygon; define the first neighboring vertex as a second-closest vertex when the first distance is smaller than the second distance and the first neighboring vertex is within the FOV angle; and define the first viewable external wall segment as extending between the closest vertex and the second-closest vertex, the first viewable external wall segment having a first position on the polygon defined by the closest vertex and the second-closest vertex.

[0223] Example 13. The system of any of Examples 8-12 wherein the structural feature has an image position defined by a pixel start location and the pixel length, and thePatent Application Attorney Docket No. FWIG-001W001 computer-readable instructions, when executed by the one or more processors, further cause the one or more processors to: define an effective subFOV angle for the image position of the structural feature, the effective subFOV angle determined using the pixel start location, the pixel length, and the FOV angle; project the image position of the structural feature onto the first viewable wall segment using the effective subFOV angle and the FOV angle; and determine a lateral physical location of the structural feature on the first viewable external wall segment using the first length of the first viewable external wall segment, the first subFOV angle, the effective subFOV angle, the first distance between the geospatial camera location and the first neighboring vertex, and a third distance between the geospatial camera location and the closest vertex.

[0224] Example 14. The system of any of Examples 8-13 wherein the computer- readable instructions, when executed by the one or more processors, further cause the one or more processors to: compare the first subFOV angle with each of the other subFOV angles; determine that the first subFOV angle at least partially overlaps angularly with one or more of the other subFOV angles; and define each external wall segment defined by each of the one or more of the other subFOV angles that overlaps angularly with the first subFOV angle as a respective non-viewable external wall segment on the polygon.

[0225] Example 15. The system of any of Examples 8-14 wherein the computer- readable instructions, when executed by the one or more processors, further cause the one or more processors to: determine that the first subFOV angle does not overlap angularly with one or more remaining subFOV angles, the one or more remaining subFOV angles excluding the one or more of the other subFOV angles that at least partially overlaps angularly with the first subFOV angle; compare the respective distances between the geospatial camera location and the vertices that defines the one or more remaining subFOV angles; define a first remaining vertex as a closest remaining vertex that partially defines a first remaining subFOV angle; compare (a) a first distance between a first neighboring remaining vertex to (b) a second distance between the geospatial camera location and a second neighboring remaining vertex, each of the first and second neighboring remaining vertices comprising an adjacent vertex to the first remaining vertex on the polygon; define the first neighboring remaining vertex as a second-closest neighboring remaining vertex when the first distance is smaller than the second distance; and define a second viewable external wall segment that extendsPatent Application Attorney Docket No. FWIG-001W001 between the first remaining vertex and the second-closest neighboring remaining vertex, the second viewable external wall segment having a second position on the polygon defined by the first remaining vertex and the second-closest neighboring remaining vertex.

[0226] Example 16. The system of any of Examples 8-15 wherein computer-readable instructions, when executed by the one or more processors, further cause the one or more processors to: compare the first remaining subFOV angle with each of the other remaining subFOV angle(s); determine that the first subFOV angle at least partially overlaps angularly with one or more of the other remaining subFOV angle(s); and define each external wall segment defined by each of the one or more of the remaining subFOV angle(s) that overlaps angularly with the first remaining subFOV angle as a respective additional non -viewable external wall segment on the polygon.

[0227] Example 17. A system for determining a wildfire risk of a building structure, comprising: one or more processors; non-transitory memory operably coupled to the one or more processors, the non-transitory memory storing computer-readable instructions that, when executed by the one or more processors, cause the one or more processors to: receive a geospatially calibrated overhead image from an overhead image database, the overhead image database in network communication with the one or more processors, the geospatially calibrated overhead image including an overhead view of a property including the building structure; segment, using one or more trained machine-learning (ML) models, the geospatially calibrated overhead image to define a polygon that represents a perimeter of the building structure, the polygon including a plurality of vertices having respective geospatial vertex locations, each vertex corresponding to a respective corner of the building structure; receive a ground-view image of the building structure, the ground-view image showing a viewable portion of an external wall segment of the building structure, the external wall segment extending between a viewable vertex that is shown in the ground-view image and a non-viewable vertex that is not shown in the ground-view image; determine, using input data, a geospatial location of a camera where the ground-view image was captured, a field-of-view (FOV) angle of the camera when the ground-view image was captured, and a compass bearing of the camera when the ground-view image was captured; segment, using the one or more trained ML models, the ground-view image to determine a boundary of the viewable portion of the external wall segment and to detect at least a structural feature on viewablePatent Application Attorney Docket No. FWIG-001W001 portion of the external wall segment, the structural feature having a pixel length; calculate, using the geospatial camera location and the respective geospatial vertex locations, a respective distance between the geospatial camera location and each vertex within the FOV angle; determine a closest vertex to the geospatial camera location within the FOV angle at the compass bearing of the camera when the ground-view image was taken, the respective distance between the geospatial camera location and the closest vertex smaller than the respective distance between the geospatial camera location and other vertices within the FOV angle; define the closest vertex as the viewable vertex; compare (a) a first distance between the geospatial camera location and a first neighboring vertex to (b) a second distance between the geospatial camera location and a second neighboring vertex, each of the first and second neighboring vertices comprising an adjacent vertex to the closest vertex on the polygon; define the first neighboring vertex as a second-closest vertex when the first distance is smaller than the second distance; define the second-closest vertex as the non-viewable vertex; determine a location of the external wall segment on the polygon using the viewable and non- viewable vertices; define the viewable portion of the external wall segment, with respect to the polygon, using the location of the external wall segment on the polygon, the geospatial camera location, and the FOV angle, the FOV angle intersecting the external wall segment at an intersection point; define a plurality of vertex lines, each vertex line extending between the geospatial camera location and a respective vertex; define an intersection line that extends between the geospatial camera location and the intersection point; determine a first vertex angle between (a) a first vertex line that extends between the geospatial camera location and the viewable vertex and (b) and a second vertex line that extends between the geospatial camera location and the non-viewable vertex; determine a length of the external wall segment using the respective geospatial vertex locations of the viewable and non-viewable vertices; determine a length of the viewable portion of the external wall segment using the length of the external wall segment, the first vertex angle, the intersection line, the respective distance between the geospatial camera location and the viewable vertex, and the first distance between the geospatial camera location and the non-viewable vertex; determine a length of the structural feature using the length of the viewable portion of the external wall segment, the intersection line, the FOV angle, the pixel length of the structural feature, and a pixel resolution of the ground-view image; generate a dimensionally calibrated three-dimensional model of an exterior of the building structure using the polygon and the length of thePatent Application Attorney Docket No. FWIG-001W001 structural feature on the viewable portion of the external wall segment; detect, using the one or more trained ML models and at least the overhead image, fuel sources on the property; calculate a respective distance between each fuel source and a respective nearest exterior wall of the building structure using the dimensionally calibrated three-dimensional model; calculate a respective fuel load in terms of thermal energy generation potential for each fuel source; and determine a wildfire risk metric for the building structure based, at least in part, on the respective distance between each fuel source and a respective nearest exterior wall of the building structure and the respective fuel load for each fuel source.

[0228] Example 18. The system of Example 17 wherein the structural feature has an image position defined, in part, by a pixel start location and the pixel length, and the computer-readable instructions, when executed by the one or more processors, further cause the one or more processors to: define an effective subFOV angle for the image position of the structural feature, the effective subFOV angle determined using the pixel start location, the pixel length, and the FOV angle; project the image position of the structural feature onto the viewable portion of the external wall segment using the effective subFOV angle and the FOV angle; and determine a lateral physical location of the structural feature on the viewable portion of the external wall segment using the length of the viewable portion of the external wall segment, the effective subFOV angle, the first distance between the geospatial camera location and the first neighboring vertex, and a third distance between the geospatial camera location and the closest vertex.

[0229] Example 19. The system of Example 17 or 18 wherein the computer-readable instructions, when executed by the one or more processors, further cause the one or more processors to: define a plurality of vertex angles, each vertex angle defined between a respective pair of the lines, the vertex angles including the first vertex angle; compare the first vertex angle with each of the other vertex angles, the other vertex angles excluding the first vertex angle; determine that the first vertex angle at least partially overlaps angularly with one or more of the other subFOV angles; and define each external wall segment defined by each of the one or more of the other vertex angles that overlaps angularly with the first vertex angle as a respective non-viewable external wall segment on the polygon.

[0230] Example 20. The system of any of claims 17-19 wherein the camera comprises a stereoscopic cameraPatent Application Attorney Docket No. FWIG-001W001

[0231] The invention should not be considered limited to the particular embodiments described above. Various modifications, equivalent processes, as well as numerous structures to which the invention may be applicable, will be readily apparent to those skilled in the art to which the invention is directed upon review of this disclosure. The above-described embodiments may be implemented in numerous ways. One or more aspects and embodiments involving the performance of processes or methods may utilize program instructions executable by a device (e.g., a computer, a processor, or other device) to perform, or control performance of, the processes or methods.

[0232] In this respect, various inventive concepts may be embodied as a non- transitory computer readable storage medium (or multiple non-transitory computer readable storage media) (e.g., a computer memory of any suitable type including transitory or non- transitory digital storage units, circuit configurations in Field Programmable Gate Arrays or other semiconductor devices, or other tangible computer storage medium) encoded with one or more programs that, when executed on one or more computers or other processors, perform methods that implement one or more of the various embodiments described above. When implemented in software (e.g., as an app), the software code may be executed on any suitable processor or collection of processors, whether provided in a single computer or distributed among multiple computers.

[0233] Further, it should be appreciated that a computer may be embodied in any of a number of forms, such as a rack-mounted computer, a desktop computer, a laptop computer, or a tablet computer, as non-limiting examples. Additionally, a computer may be embedded in a device not generally regarded as a computer but with suitable processing capabilities, including a Personal Digital Assistant (PDA), a smartphone or any other suitable portable or fixed electronic device.

[0234] Also, a computer may have one or more communication devices, which may be used to interconnect the computer to one or more other devices and / or systems, such as, for example, one or more networks in any suitable form, including a local area network or a wide area network, such as an enterprise network, and intelligent network (IN) or the Internet. Such networks may be based on any suitable technology and may operate according to any suitable protocol and may include wireless networks or wired networks.Patent Application Attorney Docket No. FWIG-001W001

[0235] Also, a computer may have one or more input devices and / or one or more output devices. These devices can be used, among other things, to present a user interface. Examples of output devices that may be used to provide a user interface include printers or display screens for visual presentation of output and speakers or other sound generating devices for audible presentation of output. Examples of input devices that may be used for a user interface include keyboards, and pointing devices, such as mice, touch pads, and digitizing tablets. As another example, a computer may receive input information through speech recognition or in other audible formats.

[0236] The non-transitory computer readable medium or media may be transportable, such that the program or programs stored thereon may be loaded onto one or more different computers or other processors to implement various one or more of the aspects described above. In some embodiments, computer readable media may be non- transitory media.

[0237] The terms "program," “app,” and "software" are used herein in a generic sense to refer to any type of computer code or set of computer-executable instructions that may be employed to program a computer or other processor to implement various aspects as described above. Additionally, it should be appreciated that, according to one aspect, one or more computer programs that when executed perform methods of this application need not reside on a single computer or processor but may be distributed in a modular fashion among a number of different computers or processors to implement various aspects of this application.

[0238] Computer-executable instructions may be in many forms, such as program modules, executed by one or more computers or other devices. Generally, program modules include routines, programs, objects, components, data structures, etc. that performs particular tasks or implement particular abstract data types. The functionality of the program modules may be combined or distributed as desired in various embodiments.

[0239] Also, data structures may be stored in computer-readable media in any suitable form. For simplicity of illustration, data structures may be shown to have fields that are related through location in the data structure. Such relationships may likewise be achieved by assigning storage for the fields with locations in a computer-readable medium that convey relationship between the fields. However, any suitable mechanism may be used to establish a relationship between information in fields of a data structure, including through the use of pointers, tags or other mechanisms that establish relationship between data elements.Patent Application Attorney Docket No. FWIG-001W001

[0240] Thus, the disclosure and claims include new and novel improvements to existing methods and technologies, which were not previously known nor implemented to achieve the useful results described above. Users of the method and system will reap tangible benefits from the functions now made possible on account of the specific modifications described herein causing the effects in the system and its outputs to its users. It is expected that significantly improved operations can be achieved upon implementation of the claimed invention, using the technical components recited herein.

[0241] Also, as described, some aspects may be embodied as one or more methods. The acts performed as part of the method may be ordered in any suitable way. Accordingly, embodiments may be constructed in which acts are performed in an order different than illustrated, which may include performing some acts simultaneously, even though shown as sequential acts in illustrative embodiments.

[0242] What is claimed is:

Claims

Patent Application Attorney Docket No. FWIG-001W001Claims1. A computer-implemented method for determining a wildfire risk for a building structure, comprising: receiving an overhead image from an overhead image database, the overhead image database in network communication with the one or more processors, the overhead image including geospatial data; determining, using the overhead image of a property including the building structure, first and second dimensions of the building structure, the first and second dimensions measured with respect to first and second axes, respectively, the overhead image having a known spatial scale; determining, using one or more ground-view images of the building structure, a third dimension of the building structure, the third dimension measured with respect to a third axis, the first, second, and third axes mutually orthogonal to each other, each ground-view image and the overhead image sharing a common dimension of the building structure; detecting, using one or more trained machine-learning (ML) models and the one or more ground-view images, structural features on one or more exterior walls of the building structure; generating a dimensionally calibrated three-dimensional model of an exterior of the building structure using the first, second, and third dimensions and the structural features; detecting, using the one or more trained ML models and at least the overhead image, fuel sources on the property; calculating a respective distance between each fuel source and a respective nearest exterior wall of the building structure using the dimensionally calibrated three- dimensional model; calculating a respective fuel load in terms of thermal energy generation potential for each fuel source; and determining a wildfire risk metric for the building structure based, at least in part, on the respective distance between each fuel source and a respective nearest exterior wall of the building structure and the respective fuel load for each fuel source.

2. The method of claim 1, further comprising automatically determining respective dimensions of each structural feature, the respective dimensions determined relative to aPatent Application Attorney Docket No. FWIG-001W001 respective common dimension of the building structure in a respective ground-view image, wherein the wildfire risk metric for the building structure is further determined based, at least in part, on the respective dimensions of each structural feature.

3. The method of claim 2, further comprising: determining a total number of pixels in the respective ground-view image corresponding to the respective common dimension of the building structure; calculating a respective pixel length as a ratio of the respective common dimension to the total number of pixels; determining a respective first number of pixels in the respective ground-view image corresponding to a respective first dimension of each structural feature; and determining the respective first dimension of each structural feature as a product of the respective first number of pixels in the respective ground-view image and the respective pixel length in the respective ground-view image.

4. The method of claim 3, further comprising: determining a respective second number of pixels in the respective ground-view image corresponding to a respective second dimension of each structural feature; and determining the respective second dimension of each structural feature as a product of the respective second number of pixels in the respective ground-view image and the respective pixel length in the respective ground-view image.

5. The method of claim 1, wherein the one or more structural features includes one or more windows, one or more doors, one or more decks, one or more balconies, one or more vents, and / or one or more garage doors.

6. The method of claim 5, further comprising determining a respective position of each structural feature on the one or more exterior walls.

7. The method of claim 1, wherein the fuel sources include major vegetation, one or more non-vegetative fuels, and / or one or more secondary structures, the one or more non- vegetative fuels including one or more shipping pallets, one or more fuel containers, one orPatent Application Attorney Docket No. FWIG-001W001 more industrial drums, one or more vehicles, one or more wood piles, and / or one or more pieces of furniture.

8. A system for determining a wildfire risk of a building structure, comprising: one or more processors; non-transitory memory operably coupled to the one or more processors, the non- transitory memory storing computer-readable instructions that, when executed by the one or more processors, cause the one or more processors to: receive a geospatially calibrated overhead image from an overhead image database, the overhead image database in network communication with the one or more processors, the geospatially calibrated overhead image including an overhead view of a property including the building structure; segment, using one or more trained machine-learning (ML) models, the geospatially calibrated overhead image to define a polygon that represents a perimeter of the building structure, the polygon including a plurality of vertices having respective geospatial vertex locations, each vertex corresponding to a respective corner of the building structure; receive a ground-view image of the building structure, the ground-view image showing one or more viewable external wall segments, each viewable external wall segment extending between a respective viewable pair of the vertices; determine, using input data, a geospatial location of a camera where the ground-view image was captured, a field-of-view (FOV) angle of the camera when the ground-view image was captured, and a compass bearing of the camera when the ground-view image was captured; segment, using the one or more trained ML models, the ground-view image to determine a boundary of the one or more viewable external wall segments and to detect at least a structural feature on a first viewable external wall segment, the structural feature having a pixel length in the ground-view image; define a plurality of lines, each line extending between the geospatial camera location and a respective vertex; define a plurality of subFOV angles, each subFOV angle defined between a respective pair of the lines;Patent Application Attorney Docket No. FWIG-001W001 determine, using the geospatial camera location, the respective geospatial vertex locations, the subFOV angles, the FOV angle, and the compass bearing, a respective location of each viewable external wall segment, shown in the ground view image, on the polygon; calculate a respective length of each viewable external wall segment shown in the ground view image, the respective length determined using the respective geospatial vertex locations for the respective viewable pair of the vertices; determine a first subFOV angle defined between a first pair of the lines that extend to a first viewable pair of the vertices that define the first viewable external wall segment; determine a length of the structural feature using a first length of the first viewable external wall segment, the first subFOV angle, the FOV angle, and the pixel length of the structural feature; generate a dimensionally calibrated three-dimensional model of an exterior of the building structure using the polygon and the length of the structural feature on the first viewable external wall segment; detect, using the one or more trained ML models and at least the overhead image, one or more fuel sources on the property; calculate a respective distance between each fuel source and a respective nearest exterior wall of the building structure using the dimensionally calibrated three-dimensional model; calculate a respective fuel load in terms of thermal energy generation potential for each fuel source; and determine a wildfire risk metric for the building structure based, at least in part, on the respective distance between each fuel source and a respective nearest exterior wall of the building structure and the respective fuel load for each fuel source.

9. The system of claim 8, wherein the structural feature has an image position in the ground-view image defined, in part, by a pixel start location and the pixel length, andPatent Application Attorney Docket No. FWIG-001W001 the computer-readable instructions, when executed by the one or more processors, further cause the one or more processors to determine a lateral physical location of the structural feature on the first viewable external wall segment using the first length of the first viewable external wall segment, the first subFOV angle, the FOV angle, the pixel start location, the pixel length, and the pixel resolution of the ground-view image.

10. The system of claim 9, wherein: the pixel start location is a lateral pixel start location, the image position is further defined by a vertical pixel start location and a pixel height, and the computer-readable instructions, when executed by the one or more processors, further cause the one or more processors to determine a vertical physical location of the structural feature on the first viewable external wall segment using the length of the structural feature, the vertical pixel start location, the pixel height, the pixel length, and the pixel resolution of the ground-view image.

11. The system of claim 8, wherein the computer-readable instructions, when executed by the one or more processors, further cause the one or more processors to: calculate, using the geospatial camera location and the respective geospatial vertex locations, a respective distance between the geospatial camera location and each vertex; determine a closest vertex to the geospatial camera location within the FOV angle; and determine the respective location of each viewable external wall segment on the polygon based, at least in part, on the closest vertex to the geospatial camera location.

12. The system of claim 11, wherein the computer-readable instructions, when executed by the one or more processors, further cause the one or more processors to: compare (a) a first distance between the geospatial camera location and a first neighboring vertex to (b) a second distance between the geospatial camera location and a second neighboring vertex, each of the first and second neighboring vertices comprising an adjacent vertex to the closest vertex on the polygon;Patent Application Attorney Docket No. FWIG-001W001 define the first neighboring vertex as a second-closest vertex when the first distance is smaller than the second distance and the first neighboring vertex is within the FOV angle; and define the first viewable external wall segment as extending between the closest vertex and the second-closest vertex, the first viewable external wall segment having a first position on the polygon defined by the closest vertex and the second-closest vertex.

13. The system of claim 12, wherein: the structural feature has an image position defined by a pixel start location and the pixel length, and the computer-readable instructions, when executed by the one or more processors, further cause the one or more processors to: define an effective subFOV angle for the image position of the structural feature, the effective subFOV angle determined using the pixel start location, the pixel length, and the FOV angle; project the image position of the structural feature onto the first viewable wall segment using the effective subFOV angle and the FOV angle; and determine a lateral physical location of the structural feature on the first viewable external wall segment using the first length of the first viewable external wall segment, the first subFOV angle, the effective subFOV angle, the first distance between the geospatial camera location and the first neighboring vertex, and a third distance between the geospatial camera location and the closest vertex.

14. The system of claim 12, wherein the computer-readable instructions, when executed by the one or more processors, further cause the one or more processors to: compare the first subFOV angle with each of the other subFOV angles; determine that the first subFOV angle at least partially overlaps angularly with one or more of the other subFOV angles; and define each external wall segment defined by each of the one or more of the other subFOV angles that overlaps angularly with the first subFOV angle as a respective non- viewable external wall segment on the polygon.Patent Application Attorney Docket No. FWIG-001W00115. The system of claim 14, wherein the computer-readable instructions, when executed by the one or more processors, further cause the one or more processors to: determine that the first subFOV angle does not overlap angularly with one or more remaining subFOV angles, the one or more remaining subFOV angles excluding the one or more of the other subFOV angles that at least partially overlaps angularly with the first subFOV angle; compare the respective distances between the geospatial camera location and the vertices that defines the one or more remaining subFOV angles; define a first remaining vertex as a closest remaining vertex that partially defines a first remaining subFOV angle; compare (a) a first distance between a first neighboring remaining vertex to (b) a second distance between the geospatial camera location and a second neighboring remaining vertex, each of the first and second neighboring remaining vertices comprising an adjacent vertex to the first remaining vertex on the polygon; define the first neighboring remaining vertex as a second-closest neighboring remaining vertex when the first distance is smaller than the second distance; and define a second viewable external wall segment that extends between the first remaining vertex and the second-closest neighboring remaining vertex, the second viewable external wall segment having a second position on the polygon defined by the first remaining vertex and the second-closest neighboring remaining vertex.

16. The system of claim 15, wherein the computer-readable instructions, when executed by the one or more processors, further cause the one or more processors to: compare the first remaining subFOV angle with each of the other remaining subFOV angle(s); determine that the first subFOV angle at least partially overlaps angularly with one or more of the other remaining subFOV angle(s); and define each external wall segment defined by each of the one or more of the remaining subFOV angle(s) that overlaps angularly with the first remaining subFOV angle as a respective additional non-viewable external wall segment on the polygon.

17. A system for determining a wildfire risk of a building structure, comprising:Patent Application Attorney Docket No. FWIG-001W001 one or more processors; non-transitory memory operably coupled to the one or more processors, the non- transitory memory storing computer-readable instructions that, when executed by the one or more processors, cause the one or more processors to: receive a geospatially calibrated overhead image from an overhead image database, the overhead image database in network communication with the one or more processors, the geospatially calibrated overhead image including an overhead view of a property including the building structure; segment, using one or more trained machine-learning (ML) models, the geospatially calibrated overhead image to define a polygon that represents a perimeter of the building structure, the polygon including a plurality of vertices having respective geospatial vertex locations, each vertex corresponding to a respective corner of the building structure; receive a ground-view image of the building structure, the ground-view image showing a viewable portion of an external wall segment of the building structure, the external wall segment extending between a viewable vertex that is shown in the ground-view image and a non-viewable vertex that is not shown in the ground-view image; determine, using input data, a geospatial location of a camera where the ground-view image was captured, a field-of-view (FOV) angle of the camera when the ground-view image was captured, and a compass bearing of the camera when the ground-view image was captured; segment, using the one or more trained ML models, the ground-view image to determine a boundary of the viewable portion of the external wall segment and to detect at least a structural feature on viewable portion of the external wall segment, the structural feature having a pixel length; calculate, using the geospatial camera location and the respective geospatial vertex locations, a respective distance between the geospatial camera location and each vertex within the FOV angle; determine a closest vertex to the geospatial camera location within the FOV angle at the compass bearing of the camera when the ground-view image was taken, the respective distance between the geospatial camera location and the closest vertexPatent Application Attorney Docket No. FWIG-001W001 smaller than the respective distance between the geospatial camera location and other vertices within the FOV angle; define the closest vertex as the viewable vertex; compare (a) a first distance between the geospatial camera location and a first neighboring vertex to (b) a second distance between the geospatial camera location and a second neighboring vertex, each of the first and second neighboring vertices comprising an adjacent vertex to the closest vertex on the polygon; define the first neighboring vertex as a second-closest vertex when the first distance is smaller than the second distance; define the second-closest vertex as the non-viewable vertex; determine a location of the external wall segment on the polygon using the viewable and non-viewable vertices; define the viewable portion of the external wall segment, with respect to the polygon, using the location of the external wall segment on the polygon, the geospatial camera location, and the FOV angle, the FOV angle intersecting the external wall segment at an intersection point; define a plurality of vertex lines, each vertex line extending between the geospatial camera location and a respective vertex; define an intersection line that extends between the geospatial camera location and the intersection point; determine a first vertex angle between (a) a first vertex line that extends between the geospatial camera location and the viewable vertex and (b) and a second vertex line that extends between the geospatial camera location and the non-viewable vertex; determine a length of the external wall segment using the respective geospatial vertex locations of the viewable and non-viewable vertices; determine a length of the viewable portion of the external wall segment using the length of the external wall segment, the first vertex angle, the intersection line, the respective distance between the geospatial camera location and the viewable vertex, and the first distance between the geospatial camera location and the non-viewable vertex;Patent Application Attorney Docket No. FWIG-001W001 determine a length of the structural feature using the length of the viewable portion of the external wall segment, the intersection line, the FOV angle, the pixel length of the structural feature, and a pixel resolution of the ground-view image; generate a dimensionally calibrated three-dimensional model of an exterior of the building structure using the polygon and the length of the structural feature on the viewable portion of the external wall segment; detect, using the one or more trained ML models and at least the overhead image, fuel sources on the property; calculate a respective distance between each fuel source and a respective nearest exterior wall of the building structure using the dimensionally calibrated three-dimensional model; calculate a respective fuel load in terms of thermal energy generation potential for each fuel source; and determine a wildfire risk metric for the building structure based, at least in part, on the respective distance between each fuel source and a respective nearest exterior wall of the building structure and the respective fuel load for each fuel source.

18. The system of claim 17, wherein the structural feature has an image position defined, in part, by a pixel start location and the pixel length, and the computer-readable instructions, when executed by the one or more processors, further cause the one or more processors to: define an effective subFOV angle for the image position of the structural feature, the effective subFOV angle determined using the pixel start location, the pixel length, and the FOV angle; project the image position of the structural feature onto the viewable portion of the external wall segment using the effective subFOV angle and the FOV angle; and determine a lateral physical location of the structural feature on the viewable portion of the external wall segment using the length of the viewable portion of the external wall segment, the effective subFOV angle, the first distance between thePatent Application Attorney Docket No. FWIG-001W001 geospatial camera location and the first neighboring vertex, and a third distance between the geospatial camera location and the closest vertex.

19. The system of claim 17, wherein the computer-readable instructions, when executed by the one or more processors, further cause the one or more processors to: define a plurality of vertex angles, each vertex angle defined between a respective pair of the lines, the vertex angles including the first vertex angle; compare the first vertex angle with each of the other vertex angles, the other vertex angles excluding the first vertex angle; determine that the first vertex angle at least partially overlaps angularly with one or more of the other subFOV angles; and define each external wall segment defined by each of the one or more of the other vertex angles that overlaps angularly with the first vertex angle as a respective non-viewable external wall segment on the polygon.

20. The system of claim 17, wherein the camera comprises a stereoscopic camera.

21. A method of generating a three-dimensional (3D) model of a structure to facilitate assessment of a wildfire risk for the structure, the 3D model including at least one structural feature susceptible to wildfire damage and at least one of a geospatial position or at least one geospatial dimension of the at least one structural feature susceptible to wildfire damage, the method comprising:A) receiving overhead image data associated with a dimensionally calibrated overhead image of at least a perimeter of the structure, the overhead image data including geospatial data for respective vertices of the perimeter of the structure;B) receiving ground-view image data associated with at least one ground-view image of at least one external wall segment of the structure, wherein the at least one external wall segment of the structure includes the at least one structural feature susceptible to wildfire damage;C) applying at least one trained machine-learning (ML) model to the received groundview image data to identify the at least one structural feature susceptible to wildfire damage on the at least one external wall segment of the structure; andPatent Application Attorney Docket No. FWIG-001W001D) determining, based at least in part on the geospatial data in the overhead image data in A), the at least one of the geospatial position or the at least one geospatial dimension of the at least one structural feature susceptible to wildfire damage identified in C) so as to facilitate the assessment of the wildfire risk for the structure.

22. The method of claim 21, wherein the at least one ground-view image includes at least a first vertex of the perimeter of the structure corresponding to one of the respective vertices of the perimeter of the structure in the overhead image data in A).

23. The method of claim 21, wherein the at least one structural feature susceptible to wildfire damage includes at least one of: a window; a door; a deck; a balcony; a vent; or a garage door.

24. The method of claim 21, wherein D) comprises: determining the at least one of the geospatial position or the at least one geospatial dimension of the at least one structural feature susceptible to wildfire damage identified in C) based at least in part on: the geospatial data in the overhead image data in A); a geospatial location of a camera upon acquiring the at least one ground-view image; a field-of-view (FOV) angle of the camera upon acquiring the at least one groundview image; and a compass bearing of the camera upon acquiring the at least one ground-view image.

25. The method of claim 24, wherein in B), the ground-view image data includes the geospatial location of the camera, the FOV angle of the camera, and the compass bearing of the camera as metadata.Patent Application Attorney Docket No. FWIG-001W00126. The method of claim 24, wherein: in A), the overhead image data includes a polygon representing the perimeter of the structure, the polygon defined by the geospatial data for the respective vertices of the perimeter of the structure; andD) further comprises:DI) determining a location of the at least one exterior wall segment of the structure on the polygon based at least in part on the ground-view image data in B).

27. The method of claim 26, wherein D) further comprises:D2) determining a length of the at least one exterior wall segment of the structure based at least in part on DI) and the geospatial data for at least one of the respective vertices of the perimeter of the structure.

28. The method of claim 27, wherein D) further comprises:D3) determining a pixel scale for the at least one ground-view image based at least in part on the ground-view image data in B) and the length of the at least one exterior wall segment of the structure determined in D2).

29. The method of claim 28, wherein:C) comprises applying the at least one trained machine-learning (ML) model to the received ground-view image data to identify pixel dimension for the at least one structural feature susceptible to wildfire damage on the at least one external wall segment of the structure; andD) comprises determining at least the at least one geospatial dimension of the at least one structural feature susceptible to wildfire damage identified in C), based at least in part on the pixel scale determined in D3) and the pixel dimensions identified in C).

30. The method of claim 29, wherein the at least one ground-view image includes at least a first vertex of the perimeter of the structure corresponding to one of the respective vertices of the perimeter of the structure in the overhead image data in A).Patent Application Attorney Docket No. FWIG-001W00131. The method of claim 30, wherein the at least one structural feature susceptible to wildfire damage includes at least one of: a window; a door; a deck; a balcony; a vent; or a garage door.

32. The method of claim 26, wherein DI) further comprises:DI a) defining a plurality of lines between the geospatial location of the camera and the respective vertices defining the polygon representing the perimeter of the structure;Dlb) defining a plurality of subFOV angles, each subFOV angle defined between a respective pair of the lines; andDie) determining the location of the at least one exterior wall segment of the structure on the polygon based at least in part on the ground-view image data in B) and the plurality of subFOV angles.

33. The method of claim 32, wherein D) further comprises:D2) determining a length of the at least one exterior wall segment of the structure based at least in part on Die) and the geospatial data for at least one of the respective vertices of the perimeter of the structure.

34. The method of claim 33, wherein D) comprises determining the at least one of the geospatial position or the at least one geospatial dimension of the at least one structural feature susceptible to wildfire damage identified in C) based at least in part on the location determined in Die) and the length determined in D2).

35. The method of claim 34, wherein the at least one ground-view image includes at least a first vertex of the perimeter of the structure corresponding to one of the respective vertices of the perimeter of the structure in the overhead image data in A).Patent Application Attorney Docket No. FWIG-001W00136. The method of claim 35, wherein the at least one structural feature susceptible to wildfire damage includes at least one of: a window; a door; a deck; a balcony; a vent; or a garage door.

Images