Use of a convolutional neural network to auto-determine a floor height and floor height elevation of a building
A method using a convolutional neural network to auto-determine a first floor height (FFH) and a FFH elevation (FFE) of a building. The FFH, and FFE of the building are determined with respect to the terrain or surface of the parcel of land on which the building is located. In turn, by knowing the FFH and/or FFE of the building on the parcel, it is possible to use that information while performing a flood risk assessment to a property without requiring a personal inspection of the parcel by a human.
1 . A computer-implemented method of detecting a first floor height (FFH) of a first floor of a subject building, the method comprising:
extracting, with a computer, digital surface model (DSM) information from a database, the DSM information includes surface elevation information of the subject building on a parcel of land on which the subject building is located;
applying an image of the subject building to an AI engine implemented on the computer, and trained to identify a first floor of the subject building from the image, the AI engine having previously been trained with other images of a plurality of other buildings;
analyzing the image with the AI engine to estimate the FFH of the subject building, the analyzing including
applying by the image to a first feature extracting input layer to generate a features,
applying the features from the input layer to a hidden layer, the features including detected building components of the subject building detected in the image,
identifying a roof elevation of the roof of the subject building from the DSM information, and
determining an interior height differential between the roof elevation and the first floor of the subject building; and
detecting the FFH of the subject building as a difference between the roof elevation and the height differential between the roof elevation and first floor.
2 . The method of claim 1 , wherein the other images previously applied to the AI engine include interior images of interior spaces of the plurality of other buildings.
3 . The method of claim 2 , wherein the analyzing includes the AI engine identifying regions of interest (ROIs) in the interior images of the interior spaces of the plurality of other buildings.
4 . The method of claim 3 , wherein the ROIs include at least one of an appliance, a countertop, or a piece of furniture.
5 . The method of claim 3 , wherein the ROIs includes an interior structure including at least one of an interior staircase, an entrance, a door, or a doorstep.
6 . The method of claim 1 , further comprising:
extracting previously stored data from another database that includes interior property information for the subject building.
7 . The method of claim 6 , wherein the another database is a multiple listing service (MLS) database.
8 . The method of claim 6 , wherein the interior property information includes at least one of room dimensional measurements, number of floors, foundation type, property type, basement information, structure design drawing, or a 3D design model.
9 . A method of detecting a first floor height (FFH) of a first floor of a subject building relative to a terrain or surface of a parcel of land on which the subject building is located, the method comprising:
obtaining information on a building footprint of the subject building on the parcel of land;
applying an image of the subject building to an AI engine trained to identify a first floor of the subject building from the image, the AI engine having previously been trained with other images of a plurality of other buildings, the other images including at least one of a front, a side, a top, or a back-side view of individual buildings of the plurality of other buildings;
analyzing the image with the AI engine and detecting the FFH of the subject building, the analyzing including
applying the image to an input layer to extract features of the building,
in a hidden layer, non-linearly transforming the features extracted in the input layer by weighting the features and applying the features to an activation function,
adapting the weighting of the features by comparing estimated features from the hidden layer to features contained in a ground truth image, and
outputting an estimate of the FFH from an output layer;
extracting digital elevation map information of the terrain and/or surface from a dataset for the parcel of land; and
converting the FFH of the subject building to a first floor elevation (FFE) from the FFH and the digital elevation map information.
10 . The method of claim 9 , further comprising:
retrieving base flood elevation (BFE) information for the subject building; and
determining a floor elevation of the building that is at least higher than the BFE by a predetermined amount so as to determine whether the subject building would experience flood damage during an occurrence of a flood that corresponds in size with a base flood.
11 . The method of claim 9 , further comprising:
training the AI engine by inputting other images as training images and ground truth images to the AI engine, and backpropagating losses so as to establish data extraction parameters for a data extraction network portion of the AI engine.
12 . The method of claim 11 , wherein the training comprises
training the AI engine to detect a whether the subject building has a floor below a main floor.
13 . The method of claim 12 , wherein the method further comprises
adjusting the FFH to a lower height in response to the AI engine having detected a presence of the floor below the main floor.
14 . The method of claim 13 , wherein the training of the AI engine to detect whether the subject building has the floor below the main floor includes detecting from the other images at least one of a basement window, a crawlspace, and a back entrance at a level that is below the main floor.
15 . The method of claim 9 , wherein the training comprises
training the AI engine by inputting and processing other images as training images to detect a main entrance of the subject building.
16 . The method of claim 15 , wherein the training to detect the main entrance includes training the AI engine with images that include buildings having front steps above ground that lead to the main entrance.
17 . The method of claim 15 , wherein the training of the AI engine to detect the main entrance includes training the AI engine to detect the main entrance from street-view images of buildings having at least one of
entrance lights adjacent to the main entrance,
a doorbell or a door knocker,
a package adjacent to the main entrance;
a door handle, and
a lock.
18 . A computer-implemented method of detecting a first floor height (FFH) of a first floor of a subject building, the method comprising:
extracting, with a computer, digital surface model (DSM) information from a database, the DSM information includes surface elevation information of the subject building on a parcel of land on which the subject building is located;
analyzing an image of the subject building with a computer programmed to identify a first floor of the subject building from the image;
estimating with the computer the FFH of the subject building from the first floor identified in the image and from the DSM information;
retrieving base flood elevation (BFE) information for the subject building; and
determining a floor elevation of the subject building that is at least higher than the BFE by a predetermined amount so as to determine whether the subject building would experience flood damage during an occurrence of a flood that corresponds in size with a base flood.
19 . The method of claim 18 , wherein
the estimating includes analyzing the image with an AI engine trained to detect the FFH of the subject building, the analyzing the image with an AI engine including
applying the image to an input layer to extract features of the subject building,
in a hidden layer, non-linearly transforming the features extracted in the input layer by weighting the features and applying the features to an activation function,
adapting the weighting of the features by comparing estimated features from the hidden layer to features contained in a ground truth image, and
outputting an estimate of the FFH from an output layer.
20 . The method of claim 19 , further comprising:
training the AI engine by inputting other images as training images and ground truth images to the AI engine, and backpropagating losses so as to establish data extraction parameters for a data extraction network portion of the AI engine.