Intelligent system and method for assessing structural damage using aerial imagery

A method of improving the accuracy of a damage assessment based on aerial imagery, the method comprising:

• obtaining a first set of imagery of a first property captured at a first time;
• clipping images in the first set of imagery to generate first clipped images, the clipping comprising:
• selecting a first selected property, from properties queried for in the first set of imagery, having a point in a first footprint polygon that is found by querying a geographic information source;
• and
• clipping images in the first set of imagery, based on a location of the first selected property, at two zooms including a neighborhood level zoom and a structure level zoom;
• obtaining a second set of imagery of the first property captured at a second time subsequent to the first time, wherein a natural disaster has impacted the first property during the interval between the first time and the second time;
• clipping images in the second set of imagery to generate second clipped images, the clipping comprising:
• selecting a second selected property, from properties queried for in the second set of imagery, having a point in a second footprint polygon that is found by querying the geographic information source;
• and
• clipping images in the second set of imagery, based on a location of the second selected property, at two zooms including a neighborhood level zoom and a structure level zoom;
• feeding the first clipped images and the second clipped images to a deep learning damage classification model;
• determining, via the deep learning classification model, that a first portion of a first structure located on the first property has been damaged, and identifying varying damage magnitude levels associated with the first portion;
• and
• generating and presenting a first heat map that visually represents the damage magnitude levels associated with the first portion;
• wherein the first set of imagery and the second set of imagery each include oblique aerial images of the first structure, and the method further comprises employing a machine learning model to evaluate the oblique aerial images and detect structural damage to sides of the first structure and to interior portions of the first structure.

A system for improving the accuracy of a damage assessment based on aerial imagery, the system comprising a processor and machine-readable media including instructions which, when executed by the processor, cause the processor to:

• obtain a first set of imagery of a first property captured at a first time;
• automatically create a first three-dimensional model of the first property based on the first set of imagery, wherein the first set of imagery is clipped to generate first clipped images used to create the first three-dimensional model, the clipping of the first set of imagery comprising:
• identifying a first selected property, from properties queried for in the first set of imagery, having a point in a first footprint polygon that is found by querying a geographic information source;
• and
• clipping images in the first set of imagery, based on a location of the first selected property, at two zooms including a neighborhood level zoom and a structure level zoom;
• obtain a second set of imagery of the first property captured at a second time subsequent to the first time;
• automatically create a second three-dimensional model of the first property based on the second set of imagery, wherein the second set of imagery is clipped to generate second clipped images used to create the second three-dimensional model, the clipping of the second set of imagery comprising:
• identifying a second selected property, from properties queried for in the second set of imagery, having a point in a second footprint polygon that is found by querying the geographic information source;
• and
• clipping images in the second set of imagery, based on a location of the second selected property, at two zooms including a neighborhood level zoom and a structure level zoom;
• automatically compare the first three-dimensional model and the second three-dimensional model using a machine learning model to detect damage on the first property that has occurred in the interval between the first time and the second time;
• employ a deep learning damage classification model to determine that a first area of a yard located on the first property includes debris, and identify varying damage magnitude levels associated with the first area;
• generate a first heat map that visually represents the damage magnitude levels associated with the first area;
• and
• generate and present a damage report based on the comparison of the first three-dimensional model and the second three-dimensional model, the damage report comprising the first heat map.