ml/driveway_detection

Detects driveways in front of houses. The workflow downloads road geometry from Open Street Maps and segments the front of houses in the input image using a machine learning model. It then uses the input image, segmentation map, road geometry, and input property boundaries to detect the presence of driveways in the front of each house.

graph TD inp1>input_raster] inp2>property_boundaries] out1>properties] out2>driveways] tsk1{{segment}} tsk2{{osm}} tsk3{{detect}} tsk1{{segment}} -- segmentation_raster --> tsk3{{detect}} tsk2{{osm}} -- roads --> tsk3{{detect}} inp1>input_raster] -- input_raster --> tsk1{{segment}} inp1>input_raster] -- input_raster --> tsk3{{detect}} inp1>input_raster] -- user_input --> tsk2{{osm}} inp2>property_boundaries] -- property_boundaries --> tsk3{{detect}} tsk3{{detect}} -- properties_with_driveways --> out1>properties] tsk3{{detect}} -- driveways --> out2>driveways]

Sources

• input_raster: Aerial imagery of the region of interest with RBG + NIR bands.

• property_boundaries: Property boundary information for the region of interest.

Sinks

• properties: Boundaries of properties that contain a driveway.

• driveways: Regions of each property boundary where a driveway was detected.

Parameters

• min_region_area: Minimum contiguous region that will be considered as a potential driveway, in meters.

• ndvi_thr: Only areas under this NDVI threshold will be considered for driveways.

• car_size: Expected size of a car, in pixels, defined as [height, width].

• num_kernels: Number of rotated kernels to try to fit a car inside a potential driveway region.

• car_thr: Ratio of pixels of a kernel that have to be inside a region in order to consider it a parkable spot.

Tasks

• segment: Segments the front of houses in the input raster using a machine learning model.

• osm: Downloads road geometry for input region from Open Street Maps.

• detect: Detects driveways in the front of each house, using the input image, segmentation map, road geometry, and input property boundaries.

Workflow Yaml

name: driveway_detection
sources:
  input_raster:
  - segment.input_raster
  - detect.input_raster
  - osm.user_input
  property_boundaries:
  - detect.property_boundaries
sinks:
  properties: detect.properties_with_driveways
  driveways: detect.driveways
parameters:
  min_region_area: null
  ndvi_thr: null
  car_size: null
  num_kernels: null
  car_thr: null
tasks:
  segment:
    op: segment_driveway
  osm:
    workflow: data_ingestion/osm_road_geometries
    parameters:
      network_type: drive_service
      buffer_size: 100
  detect:
    op: detect_driveway
    parameters:
      min_region_area: '@from(min_region_area)'
      ndvi_thr: '@from(ndvi_thr)'
      car_size: '@from(car_size)'
      num_kernels: '@from(num_kernels)'
      car_thr: '@from(car_thr)'
edges:
- origin: segment.segmentation_raster
  destination:
  - detect.segmentation_raster
- origin: osm.roads
  destination:
  - detect.roads
description:
  short_description: Detects driveways in front of houses.
  long_description: The workflow downloads road geometry from Open Street Maps and
    segments the front of houses in the input image using a machine learning model.
    It then uses the input image, segmentation map, road geometry, and input property
    boundaries to detect the presence of driveways in the front of each house.
  sources:
    input_raster: Aerial imagery of the region of interest with RBG + NIR bands.
    property_boundaries: Property boundary information for the region of interest.
  sinks:
    properties: Boundaries of properties that contain a driveway.
    driveways: Regions of each property boundary where a driveway was detected.
  parameters:
    min_region_area: Minimum contiguous region that will be considered as a potential
      driveway, in meters.
    ndvi_thr: Only areas under this NDVI threshold will be considered for driveways.
    car_size: Expected size of a car, in pixels, defined as [height, width].
    num_kernels: Number of rotated kernels to try to fit a car inside a potential
      driveway region.
    car_thr: Ratio of pixels of a kernel that have to be inside a region in order
      to consider it a parkable spot.