proximity-classifier

Viam vision service that reads a camera point cloud and returns one classification for every configured point that has cloud points nearby. Several classes can match the same frame.

Model: viam-labs:vision:proximity-classifier

CaptureAllFromCamera returns a grid image of the cloud, including when classes is empty. That picture is how you choose coordinates. The default side view (yz) draws +Y to the left and +Z up. Color is X, the axis that view collapses.

Configuration

{
  "camera": "cup-camera",
  "radius": 0.5,
  "classes": {
    "left-cup": { "x": 9.1, "y": 3, "z": -3 },
    "right-cup": { "x": 6.3, "y": -3, "z": -1.5 }
  }
}

Those two points are the short cup and the tall cup in the sample cloud testdata/cups.pcd. left-cup is the short one. A radius of 0.5 reaches that cup and misses the other. Coordinates are in the units of the camera's point cloud. GetPointCloud PCD data is in meters.

Attributes

Name	Type	Inclusion	Description
camera	string	Required	Name of a camera that returns a point cloud.
classes	object	Optional	Map of label to {x, y, z}. Leave this empty while you are reading the grid.
radius	float	Optional	How near a cloud point must be to count. Defaults to 0.5.
min_points	int	Optional	How many points must fall inside the radius. Defaults to 1.
view	string	Optional	Grid projection: yz (default), xz, or xy. Pass {"view": "xz"} in the capture extra to switch for one call.

Reading the grid

Call CaptureAllFromCamera with return_image set. The image has tick marks on the two drawn axes and a color bar for the collapsed axis. Once classes are configured, each one is drawn as a circle of radius.