Method for property feature segmentation

Patent metadata

Field	Value
Publication number	US-11640667-B2
Application number	US-202117529836-A
Country	US
Kind code	B2
Filing date	Nov 18, 2021
Priority date	Jun 2, 2020
Publication date	May 2, 2023
Grant date	May 2, 2023

Abstract

The method for determining property feature segmentation includes: receiving a region image for a region; determining parcel data for the region; determining a final segmentation output based on the region image and parcel data using a trained segmentation module; optionally generating training data; and training a segmentation module using the training data S 500.

First claim

We claim:

1. A method, comprising:
receiving a region image depicting a property feature;
determining an instance-aware mask for the property feature based on the region image;
determining a semantic segmentation mask for the property feature based on the region image;
computing a distance transform from the instance-aware mask; and
generating a pixel-accurate mask by associating instance identifiers for different property feature instances with pixels of the semantic segmentation mask based on the distance transform.

2. The method of claim 1 , wherein generating the pixel-accurate mask comprises determining a nearest property feature instance for pixels lacking an instance identifier within the semantic segmentation mask, based on the distance transform.

3. The method of claim 2 , wherein the nearest property feature instance is determined using a watershed transform.

4. The method of claim 1 , wherein the instance-aware mask is further determined based on parcel data for a geographic region depicted in the region image.

5. The method of claim 4 , wherein the parcel data comprises a parcel mask for the geographic region.

6. The method of claim 1 , wherein the instance identifiers are further assigned to pixels of the semantic segmentation mask based on parcel data for a region depicted in the region image.

7. The method of claim 1 , wherein the instance-aware mask comprises an under-segmented mask of the property feature.

8. The method of claim 1 , wherein the instance-aware mask is determined by an instance-aware segmentation module trained on training data, wherein the training data is determined by:
determining a set of polygons for a training image depicting a plurality of property features in a geographic region;
determining a set of parcels for the geographic region;
generating a set of instance polygons from the polygon set by combining adjacent polygons sharing a common parcel; and
determining labels for each instance polygon in the set, wherein the training image and the labels are used to train the instance-aware segmentation module.

9. The method of claim 1 , wherein the region image comprises a remote image.

10. The method of claim 1 , wherein the property feature comprises at least one of: a roof, driveway, paved surface, vegetation, or waterfront.

11. A method, comprising:
receiving a region image depicting a set of instances of a property feature;
retrieving parcel data representative of parcel extents for parcels associated with the region image;
determining a segmentation mask for the property feature based on the region image;
determining an instance-aware mask for the property feature based on the region image and the parcel data, using an instance-aware segmentation module trained on training data determined by:
determining a set of polygons for a training image depicting a plurality of property features in a geographic region;
determining a set of parcels for the geographic region;
generating a set of instance polygons from the polygon set by combining adjacent polygons sharing a common parcel; and
labelling each instance polygon in the set, wherein the training image and the labels are used to train the instance-aware segmentation module; and

determining property feature image segments corresponding to property feature instances based on the segmentation mask, the instance-aware mask, and the parcel data.

12. The method of claim 11 , wherein the property feature image segments are pixel-accurate representations of the respective property feature instance depicted in the region image.

13. The method of claim 11 , wherein the instance-aware mask comprises an under-segmented mask of the property feature.

14. The method of claim 11 , wherein determining the property feature image segments comprises:
dilating each instance within the instance-aware mask to generate a dilated mask; and
masking the dilated mask using the segmentation mask.

15. The method of claim 11 , wherein determining the property feature image segments comprises segmenting the segmentation mask using the parcel data.

16. The method of claim 15 , wherein segmenting the segmentation mask using the parcel data comprises:
identifying property feature pixels, from the semantic segmentation mask, that share a common parcel; and
treating the identified property feature pixels as part of a shared property feature image segment.

17. The method of claim 11 , wherein the region image is comprises a remote image.

18. The method of claim 11 , wherein the property feature comprises at least one of: a roof, driveway, paved surface, vegetation, or waterfront.

Assignees

• Cape Analytics Inc

Inventors

• Richter Fabian
• Portail Matthieu
• Erickson Jason

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• G06T7/11 Primary

  Region-based segmentation · CPC title

• G06F16/29

  Geographical information databases · CPC title