"""Camera configuration for pyorc."""
import copy
import json
import os
from typing import Any, Dict, List, Literal, Optional, Tuple, Union
import cv2
import matplotlib.pyplot as plt
import numpy as np
import shapely.geometry
from pyproj import CRS, Transformer
from pyproj.exceptions import CRSError
from shapely import ops, wkt
from shapely.geometry import LineString, Point, Polygon
from pyorc import cv, helpers, plot_helpers
MODES = Literal["camera", "geographical", "3d"]
[docs]class CameraConfig:
"""Camera configuration containing information about the perspective of the camera.
The camera configuration contains information and functionalities to reconstruct perspective information
relating 2D image coordinates to 3D real world coordinates.
"""
def __str__(self):
return str(self.to_json())
def __repr__(self):
return self.to_json()
[docs] def __init__(
self,
height: int,
width: int,
crs: Optional[Any] = None,
window_size: int = 10,
resolution: float = 0.05,
bbox: Optional[Union[shapely.geometry.Polygon, str]] = None,
camera_matrix: Optional[List[List[float]]] = None,
dist_coeffs: Optional[List[List[float]]] = None,
lens_position: Optional[List[float]] = None,
corners: Optional[List[List[float]]] = None,
gcps: Optional[Dict[str, Union[List, float]]] = None,
lens_pars: Optional[Dict[str, float]] = None,
calibration_video: Optional[str] = None,
is_nadir: Optional[bool] = False,
stabilize: Optional[List[List]] = None,
rotation: Optional[int] = None,
rvec: Optional[List[float]] = None,
tvec: Optional[List[float]] = None,
):
"""Initialize a camera configuration instance.
Parameters
----------
height : int
height of frame in pixels
width : int
width of frame in pixels
crs : int, dict or str, optional
Coordinate Reference System. Accepts EPSG codes (int or str) proj (str or dict) or wkt (str). Only used if
the data has no native CRS.
window_size : int
pixel size of interrogation window (default: 15)
resolution : float, optional
resolution in m. of projected pixels (default: 0.01)
bbox : shapely.geometry.Polygon, optional
bounding box in geographical coordinates
camera_matrix : List[List[float]], optional
pre-defined camera matrix (if e.g. known from a separate calibration)
dist_coeffs : List[List[float]], optional
pre-defined distortion parameters (if e.g. known from a separate calibration)
lens_position : list of floats (3),
x, y, z coordinate of lens position in CRS
corners : list of lists of floats (2)
[x, y] coordinates defining corners of area of interest in camera cols/rows, bbox will be computed from this
gcps : dict
Can contain "src": list of lists, with column, row locations in objective of control points,
"dst": list of lists, with x, y or x, y, z locations (local or global coordinate reference system) of
control points,
"h_ref": float, measured water level [m] in local reference system (e.g. from staff gauge or pressure gauge)
during gcp survey,
"z_0": float, water level [m] in global reference system (e.g. from used GPS system CRS). This must be in
the same vertical reference as the measured bathymetry and other survey points,
"crs": int, str or CRS object, CRS in which "dst" points are measured. If None, a local coordinate system is
assumed (e.g. from spirit level).
lens_pars : dict, optional
Lens parameters, containing: "k1": float, barrel lens distortion parameter (default: 0.),
"c": float, optical center (default: 2.),
"focal_length": float, focal length (default: width of image frame)
calibration_video : str, optional
local path to video file containing a checkerboard pattern. Must be 9x6 if called directly, otherwise use
``.calibrate_camera`` explicitly and provide ``chessboard_size`` explicitly. When used, an automated camera
calibration on the video file will be attempted.
is_nadir : bool, optional
If set, the video is assumed to be taken at sub-drone position and only two control points are needed
for camera configuration
stabilize : list (of lists), optional
contains [x, y] pixels defining a polygon enclosing moving (water) areas. Areas outside of the polygon
are used for stabilization of the video, if this polygon is defined.
rotation : int [90, 180, 270]
enforces a rotation of the video of 90, 180 or 2780 degrees clock-wise.
rvec : list of floats (3), optional
OpenCV compatible rotation vector, if known. If None, the rotation will be computed from pnp solving if
gcps are available.
tvec : list of floats (3), optional
OpenCV compatible translation vector, if known. If None, the rotation will be computed from pnp solving if
gcps are available.
"""
assert isinstance(height, int), 'height must be provided as type "int"'
assert isinstance(width, int), 'width must be provided as type "int"'
assert isinstance(window_size, int), 'window_size must be of type "int"'
self.height = height
self.width = width
self.is_nadir = is_nadir
self.rvec = rvec
self.tvec = tvec
if crs is not None:
try:
crs = CRS.from_user_input(crs)
except CRSError:
raise CRSError(f'crs "{crs}" is not a valid Coordinate Reference System')
assert crs.is_geographic == 0, "Prodstvided crs must be projected with units like [m]"
self.crs = crs.to_wkt()
if resolution is not None:
self.resolution = resolution
if lens_position is not None:
self.set_lens_position(*lens_position)
else:
self.lens_position = None
if gcps is not None:
self.set_gcps(**gcps)
if camera_matrix is None or dist_coeffs is None:
if self.is_nadir:
# with nadir, no perspective can be constructed, hence, camera matrix and dist coeffs will be set
# to default values
self.camera_matrix = cv._get_cam_mtx(self.height, self.width)
self.dist_coeffs = cv.DIST_COEFFS
# camera pars are incomplete and need to be derived
else:
self.set_intrinsic(camera_matrix=camera_matrix, lens_pars=lens_pars)
else:
# camera matrix and dist coeffs can also be set hard, this overrules the lens_pars option
self.camera_matrix = camera_matrix
self.dist_coeffs = dist_coeffs
if calibration_video is not None:
self.set_lens_calibration(calibration_video, plot=False)
if bbox is not None:
self.bbox = bbox
if window_size is not None:
self.window_size = window_size
# override the transform and bbox with the set corners
if corners is not None:
self.set_bbox_from_corners(corners)
if stabilize is not None:
self.stabilize = stabilize
if rotation is not None:
self.rotation = rotation
@property
def bbox(self):
"""Give geographical bbox fitting around corners points of area of interest in camera perspective.
Returns
-------
bbox : shapely.geometry.Polygon
bbox of area of interest
"""
return self._bbox
@bbox.setter
def bbox(self, pol):
if isinstance(pol, str):
self._bbox = wkt.loads(pol)
else:
self._bbox = pol
@property
def camera_matrix(self):
"""Get camera matrix."""
return self._camera_matrix
@camera_matrix.setter
def camera_matrix(self, camera_matrix):
self._camera_matrix = camera_matrix.tolist() if isinstance(camera_matrix, np.ndarray) else camera_matrix
@property
def dist_coeffs(self):
"""Get distortion coefficients."""
return self._dist_coeffs
@dist_coeffs.setter
def dist_coeffs(self, dist_coeffs):
self._dist_coeffs = dist_coeffs.tolist() if isinstance(dist_coeffs, np.ndarray) else dist_coeffs
@property
def gcps_dest(self):
"""Get destination coordinates of GCPs.
Returns
-------
dst : np.ndarray
destination coordinates of ground control point. z-coordinates are parsed from z_0 if necessary
"""
if hasattr(self, "gcps"):
if "dst" in self.gcps:
return np.array(
self.gcps["dst"]
if len(self.gcps["dst"][0]) == 3
else np.c_[self.gcps["dst"], np.ones(4) * self.gcps["z_0"]],
dtype=np.float64,
)
# if conditions are not yet met, then return None
return None
@property
def gcps_dest_bbox(self):
"""Give destination coordinates as row, col in intended bounding box.
Returns
-------
dst : np.ndarray
Destination coordinates measured as column, row in the intended bounding box with the intended resolution
"""
return np.array(cv.transform_to_bbox(self.gcps_dest, self.bbox, self.resolution))
@property
def gcps_bbox_reduced(self):
"""Give col, row coordinates of gcps within intended bounding box, reduced by mean coordinate.
Returns
-------
dst : np.ndarray
Destination coordinates in col, row in the intended bounding box, reduced with their mean coordinate
"""
return self.gcps_dest_bbox - self.gcps_dest_bbox.mean(axis=0)
@property
def gcps_reduced(self):
"""Get location of gcp destination points, reduced with their mean for better local projection.
Returns
-------
dst : np.ndarray
Reduced coordinate (x, y) or (x, y, z) of gcp destination points
"""
return np.array(self.gcps_dest - self.gcps_mean)
@property
def gcps_mean(self):
"""Get mean location of gcp destination points.
Returns
-------
dst_mean : np.ndarray
mean coordinate (x, y) or (x, y, z) of gcp destination points
"""
return np.array([0.0, 0.0, 0.0]) if self.gcps_dest is None else np.array(self.gcps_dest).mean(axis=0)
@property
def gcps_dims(self):
"""Return amount of dimensions of GCPs provided (2 or 3).
Returns
-------
dims : int
amount of dimensions of gcps (can be 2 or 3)
"""
return len(self.gcps["dst"][0]) if hasattr(self, "gcps") else None
@property
def is_nadir(self):
"""Return if the camera configuration belongs to nadir video.
Returns
-------
is_nadir : bool
False if not nadir, True if nadir
"""
return self._is_nadir
@is_nadir.setter
def is_nadir(self, nadir_prop: bool):
self._is_nadir = nadir_prop
@property
def pnp(self):
"""Return Precise N point solution from ground control points, intrinsics and distortion."""
# solve rvec and tvec with reduced coordinates, this ensure that the solvepnp solution is stable.
_, rvec, tvec = cv.solvepnp(self.gcps_reduced, self.gcps["src"], self.camera_matrix, self.dist_coeffs)
# ensure that rvec and tvec are corrected for the fact that mean gcp location was subtracted
rvec_cam, tvec_cam = cv.pose_world_to_camera(rvec, tvec)
tvec_cam += self.gcps_mean
# transform back to world
rvec, tvec = cv.pose_world_to_camera(rvec_cam, tvec_cam)
return _, rvec, tvec
@property
def rvec(self):
"""Return rvec from precise N point solution."""
return self.pnp[1].tolist() if self._rvec is None else self._rvec
@rvec.setter
def rvec(self, _rvec):
self._rvec = _rvec.tolist() if isinstance(_rvec, np.ndarray) else _rvec
@property
def shape(self):
"""Return rows and columns in projected frames from `Frames.project`.
Returns
-------
rows : int
Amount of rows in projected frame
cols : int
Amount of columns in projected frame
"""
cols, rows = cv._get_shape(self.bbox, resolution=self.resolution, round=1)
return rows, cols
@property
def stabilize(self):
"""Return stabilization polygon (anything outside is used for stabilization.
Returns
-------
coords : list of lists
coordinates in original image frame comprising the polygon for use in stabilization
"""
return self._stabilize
@stabilize.setter
def stabilize(self, coords: List[List[float]]):
self._stabilize = coords
@property
def rotation(self):
"""Return rotation OpenCV code.
Returns
-------
code : int
integer code belonging to rotation, 0 for 90 deg, 1 for 180 deg and 2 for 270 deg
"""
if hasattr(self, "_rotation"):
return self._rotation
else:
return None
@rotation.setter
def rotation(self, rotation_code: int):
self._rotation = rotation_code
@property
def transform(self):
"""Returns Affine transform of projected frames from `Frames.project`.
Returns
-------
transform : rasterio.transform.Affine object
"""
return cv._get_transform(self.bbox, resolution=self.resolution)
@property
def tvec(self):
"""Return tvec from precise N point solution."""
return self.pnp[2].tolist() if self._tvec is None else self._tvec
@tvec.setter
def tvec(self, _tvec):
self._tvec = _tvec.tolist if isinstance(_tvec, np.ndarray) else _tvec
[docs] def set_lens_calibration(
self,
fn: str,
chessboard_size: Optional[Tuple] = (9, 6),
max_imgs: Optional[int] = 30,
plot: Optional[bool] = True,
progress_bar: Optional[bool] = True,
**kwargs,
):
"""Calibrate and set the properties `camera_matrix` and `dist_coeffs` using a video of a chessboard pattern.
Follows methods described on https://docs.opencv.org/4.x/dc/dbb/tutorial_py_calibration.html
Parameters
----------
fn : str
path to video file
df : int, optional
amount of frames to skip after a valid frame with corner points is found, defaults to fps of video.
chessboard_size : tuple, optional
amount of internal corner points expected on chessboard pattern, default is (9, 6).
max_imgs : int, optional
maximum amount of images to use for calibration (default: 30).
tolerance : float, optional
error tolerance alowed for reprojection of corner points (default: 0.1, if set to None, no filtering will
be done). images that exceed the tolerance are excluded from calibration. This is to remove images with
spuriously defined points, or blurry images.
plot : bool, optional
if set, chosen frames will be plotted for the user to inspect on-=the-fly with a one-second delay
(default: True).
progress_bar : bool, optional
if set, a progress bar going through the frames is plotted (default: True).
**kwargs : dict
keyword arguments to pass to underlying methods
"""
assert os.path.isfile(fn), f"Video calibration file {fn} not found"
camera_matrix, dist_coeffs = cv.calibrate_camera(fn, chessboard_size, max_imgs, plot, progress_bar, **kwargs)
self.camera_matrix = camera_matrix
self.dist_coeffs = dist_coeffs
[docs] def estimate_lens_position(self):
"""Estimate lens position from distortion and intrinsec/extrinsic matrix."""
rvec, tvec = np.array(self.rvec), np.array(self.tvec)
rmat = cv2.Rodrigues(rvec)[0]
# determine lens position related to center of objective
lens_pos = (np.array(-rmat).T @ tvec).flatten()
return lens_pos
[docs] def get_bbox(
self,
camera: Optional[bool] = False,
mode: Optional[MODES] = "geographical",
h_a: Optional[float] = None,
z_a: Optional[float] = None,
within_image: Optional[bool] = False,
expand_exterior=True,
) -> Polygon:
"""Get bounding box.
Can be retrieved in geographical or camera perpective.
Parameters
----------
camera : bool, optional
If set, the bounding box will be returned as row and column coordinates in the camera perspective.
In this case ``h_a`` may be set to provide the right water level, to estimate the bounding box for.
This option is deprecated, instead use mode="camera".
mode : Literal["geographical", "camera", "3d"], optional
Determines the type of bounding box to return. If set to "geographical" (default), the bounding box
is returned in the geographical coordinates. If set to "camera", the bounding box is returned in the
camera perspective. If set to "3d", the bounding box is returned as a 3D polygon in the CRS
h_a : float, optional
If set with `mode="camera"`, then the bbox coordinates will be transformed to the camera perspective,
using h_a as a present water level (in local datum). In case a video with higher (lower) water levels is
used, this will result in a different perspective plane than the control video.
z_a : float, optional
similar to setting h_a, but z_a represent the vertical coordinate in the coordinate reference system of
the camera configuration instead of the local datum.
within_image : bool, optional (default False)
Set to True to make an attempt to remove parts of the polygon that lie outside of the image field of view
expand_exterior : bool, optional
Set to True to expand the corner points to more points. This is particularly useful for plotting purposes.
Returns
-------
A bounding box, that in the used CRS is perfectly rectangular, and aligned in the up/downstream direction.
It can (and certainly will) be rotated with respect to a typical bbox with xlim, ylim coordinates.
If user sets ``mode="camera"`` then the geographical bounding box will be converted into a camera perspective,
using the homography belonging to the available ground control points and current water level.
This can then be used to reconstruct the grid for velocimetry calculations.
"""
if camera:
import warnings
warnings.warn(
"The camera=True option is deprecated, use mode='camera' instead. This option will be removed in "
"a future release.",
DeprecationWarning,
stacklevel=2,
)
mode = "camera"
bbox = self.bbox
coords = np.array(bbox.exterior.coords)
if within_image:
# in this case, always more points than just corners are needed, so expand_exterior is forced to True
expand_exterior = True
if expand_exterior:
# make a new set of bbox coordinates with a higher density. This is meant to enable plotting of
# distortion on image frame, and to plot partial coverage in the real-world coordinates
coords_expand = np.zeros((0, 2))
for n in range(0, len(coords) - 1):
new_coords = np.linspace(coords[n], coords[n + 1], 100)
coords_expand = np.r_[coords_expand, new_coords]
coords = coords_expand
if not z_a:
z_a = self.get_z_a(h_a)
# add vertical coordinates to the set
coords = np.c_[coords, np.ones(len(coords)) * z_a]
# project points to pixel image coordinates
corners = self.project_points(coords, within_image=within_image)
corners = corners[np.isfinite(corners[:, 0])]
if not mode == "camera":
# project back to real-world coordinates after possibly cutting at edges of visibility
corners = self.unproject_points(np.array(np.array(list(zip(*corners, strict=False))).T), z_a)
if mode == "3d":
return Polygon(corners[np.isfinite(corners[:, 0])])
return Polygon(corners[np.isfinite(corners[:, 0])][:, 0:2])
[docs] def get_depth(self, z: List[float], h_a: Optional[float] = None) -> List[float]:
"""Retrieve depth for measured bathymetry points.
This is done using the camera configuration and an actual water level,
measured in local reference (e.g. staff gauge).
Parameters
----------
z : list of floats
measured bathymetry point depths
h_a : float, optional
actual water level measured [m], if not set, assumption is that a single video
is processed and thus changes in water level are not relevant. (default: None)
Returns
-------
depths : list of floats
"""
if h_a is None:
h_a = self.gcps["h_ref"]
z_pressure = np.maximum(self.gcps["z_0"] - self.gcps["h_ref"] + h_a, z)
return z_pressure - z
[docs] def get_dist_shore(
self, x: List[float], y: List[float], z: List[float], h_a: Optional[float] = None
) -> List[float]:
"""Retrieve depth for measured bathymetry points.
This is done using the camera configuration and an actual water level, measured
in local reference (e.g. staff gauge).
Parameters
----------
x : list of floats
measured bathymetry point x-coordinates
y : list of floats
measured bathymetry point y-coordinates
z : list of floats
measured bathymetry point depths
h_a : float, optional
actual water level measured [m], if not set, assumption is that a single video
is processed and thus changes in water level are not relevant. (default: None)
Returns
-------
depths : list of floats
"""
# retrieve depth
depth = self.get_depth(z, h_a=h_a)
if h_a is None:
assert self.gcps["h_ref"] is None, (
"No actual water level is provided, but a reference water level is " "provided "
)
# h_a = 0.
# h_ref = 0.
# else:
# h_ref = self.gcps["h_ref"]
z_dry = depth <= 0
z_dry[[0, -1]] = True
# compute distance to nearest dry points with Pythagoras
dist_shore = np.array(
[(((x[z_dry] - _x) ** 2 + (y[z_dry] - _y) ** 2) ** 0.5).min() for _x, _y in zip(x, y, strict=False)]
)
return dist_shore
[docs] def get_dist_wall(self, x: List[float], y: List[float], z: List[float], h_a: Optional[float] = None) -> List[float]:
"""Retrieve distance to wall for measured bathymetry points.
Done by using the camera configuration and an actual water level, measured in local reference (e.g. staff
gauge).
Parameters
----------
x : list of floats
measured bathymetry point x-coordinates
y : list of floats
measured bathymetry point y-coordinates
z : list of floats
measured bathymetry point depths
h_a : float, optional
actual water level measured [m], if not set, assumption is that a single video
is processed and thus changes in water level are not relevant. (default: None)
Returns
-------
distance : list of floats
"""
depth = self.get_depth(z, h_a=h_a)
dist_shore = self.get_dist_shore(x, y, z, h_a=h_a)
dist_wall = (dist_shore**2 + depth**2) ** 0.5
return dist_wall
[docs] def z_to_h(self, z: float) -> float:
"""Convert z coordinates of bathymetry to height coordinates in local reference (e.g. staff gauge).
Parameters
----------
z : float
measured bathymetry point
Returns
-------
h : float
"""
h_ref = 0 if self.gcps["h_ref"] is None else self.gcps["h_ref"]
h = z + h_ref - self.gcps["z_0"]
return h
[docs] def h_to_z(self, h_a: float) -> float:
"""Convert z coordinates of bathymetry to height coordinates in local reference (e.g. staff gauge).
Parameters
----------
h_a : float
measured level in local datum
Returns
-------
z : float
level in global datum
"""
h_ref = 0 if self.gcps["h_ref"] is None else self.gcps["h_ref"]
return h_a - h_ref + self.gcps["z_0"]
[docs] def get_M(
self, h_a: Optional[float] = None, to_bbox_grid: Optional[bool] = False, reverse: Optional[bool] = False
) -> np.ndarray:
"""Establish a transformation matrix for a certain actual water level `h_a`.
This is done by mapping where the ground control points, measured at `h_ref` will end up with new water level
`h_a`, given the lens position.
Parameters
----------
h_a : float, optional
actual water level [m] (Default: None)
to_bbox_grid : bool, optional
if set, the M will be computed in row, column values of the target bbox, with set resolution
reverse : bool, optional
if True, the reverse matrix is prepared, which can be used to transform projected
coordinates back to the original camera perspective. (Default: False)
Returns
-------
M : np.ndarray
2x3 transformation matrix
"""
src = cv.undistort_points(self.gcps["src"], self.camera_matrix, self.dist_coeffs)
if to_bbox_grid:
dst_a = self.gcps_bbox_reduced
else:
dst_a = self.gcps_reduced
# compute the water level in the coordinate system reduced with the mean gcp coordinate
z_a = self.get_z_a(h_a)
z_a -= self.gcps_mean[-1]
# treating 3D homography
return cv.get_M_3D(
src=src,
dst=dst_a,
camera_matrix=self.camera_matrix,
dist_coeffs=cv.DIST_COEFFS, # self.dist_coeffs,
z=z_a,
reverse=reverse,
)
[docs] def get_z_a(self, h_a: Optional[float] = None) -> float:
"""Get actual water level measured in global vertical datum (+z_0) from water level in local datum (+h_ref).
Parameters
----------
h_a : float, optional
actual water level measured [m], if not set, assumption is that a single video
is processed and thus changes in water level are not relevant. (default: None)
Returns
-------
Actual locations of control points (in case these are only x, y) given the current set water level and
the camera location
"""
if h_a is None:
return self.gcps["z_0"]
else:
return self.gcps["z_0"] + (h_a - self.gcps["h_ref"])
[docs] def set_bbox_from_corners(self, corners: List[List[float]]):
"""Establish bbox based on a set of camera perspective corner points.
Parameters
----------
corners : list of lists (4)
[columns, row] coordinates in original camera perspective without any undistortion applied
"""
assert np.array(corners).shape == (4, 2), (
f"a list of lists of 4 coordinates must be given, resulting in (4, "
f"2) shape. Current shape is {corners.shape} "
)
# get homography
corners_xyz = self.unproject_points(corners, np.ones(4) * self.gcps["z_0"])
bbox = cv.get_aoi(corners_xyz, resolution=self.resolution)
self.bbox = bbox
[docs] def set_intrinsic(
self,
camera_matrix: Optional[List[List]] = None,
dist_coeffs: Optional[List[List]] = None,
lens_pars: Optional[Dict[str, float]] = None,
):
"""Set lens and distortion parameters.
If not provided, they are derived by optimizing pnp fitting together with optimizing the focal length.
Parameters
----------
camera_matrix : Optional[List[List]]
A defined camera matrix to set as intrinsic parameters. If not provided, it will use default values or
those derived from ground control points (GCPs) if available.
dist_coeffs : Optional[List[List]]
Distortion coefficients to be used for the camera. If not provided, it will use default values or those
derived from GCPs if available.
lens_pars : Optional[Dict[str, float]]
Lens parameters to be set. These will override any default settings or those derived from GCPs if provided.
"""
# first set a default estimate from pose if 3D gcps are available
self.set_lens_pars() # default parameters use width of frame
if hasattr(self, "gcps"):
if len(self.gcps["src"]) >= 4:
self.camera_matrix, self.dist_coeffs, err = cv.optimize_intrinsic(
self.gcps["src"],
self.gcps_dest,
# self.gcps["dst"],
self.height,
self.width,
lens_position=self.lens_position,
)
if lens_pars is not None:
# override with lens parameter set by user
self.set_lens_pars(**lens_pars)
if camera_matrix is not None and dist_coeffs is not None:
# override with
self.camera_matrix = camera_matrix
self.dist_coeffs = dist_coeffs
[docs] def set_lens_pars(self, k1: Optional[float] = 0.0, c: Optional[float] = 2.0, focal_length: Optional[float] = None):
"""Set the lens parameters of the given CameraConfig.
Parameters
----------
k1 : float, optional
lens curvature [-], zero (default) means no curvature
c : float, optional
optical centre [1/n], where n is the fraction of the lens diameter, 2.0 (default) means in the
centre.
focal_length : float, optional
focal length [mm], typical values could be 2.8, or 4 (default).
"""
assert isinstance(k1, (int, float)), "k1 must be a float"
assert isinstance(c, (int, float)), "c must be a float"
if focal_length is not None:
assert isinstance(focal_length, (int, float, None)), "f must be a float"
self.dist_coeffs = cv._get_dist_coefs(k1)
self.camera_matrix = cv._get_cam_mtx(self.height, self.width, c=c, focal_length=focal_length)
[docs] def set_gcps(
self, src: List[List], dst: List[List], z_0: float, h_ref: Optional[float] = None, crs: Optional[Any] = None
):
"""Set ground control points for the given CameraConfig.
Parameters
----------
src : list of lists (2, 4 or 6+)
[x, y] pairs of columns and rows in the frames of the original video
dst : list of lists (2, 4 or 6+)
[x, y] or [x, y, z] pairs of real world coordinates in the given coordinate reference system.
z_0 : float
Water level measured in global reference system such as a geoid or ellipsoid used
by a GPS device. All other surveyed points (lens position and cross section) must have the same vertical
reference.
h_ref : float, optional
Water level, belonging to the 4 control points in `dst`. This is the water level
as measured by a local reference (e.g. gauge plate) during the surveying of the control points. Control
points must be taken on the water surface. If a single movie is processed, h_ref can be left out.
(Default: None)
crs : int, dict or str, optional
Coordinate Reference System. Accepts EPSG codes (int or str) proj (str or dict) or wkt (str). CRS used to
measure the control points (e.g. 4326 for WGS84 lat-lon). Destination control points will automatically be
reprojected to the local crs of the CameraConfig. (Default: None)
"""
assert isinstance(src, list), "src must be a list of (x, y) or (x, y, z) coordinates"
assert isinstance(dst, list), "dst must be a list of (x, y) or (x, y, z) coordinates"
if np.array(dst).shape[1] == 2:
assert len(src) in [2, 4], f"2 or 4 source points are expected in src, but {len(src)} were found"
if len(src) == 4:
assert len(dst) == 4, f"4 destination points are expected in dst, but {len(dst)} were found"
else:
assert len(dst) == 2, f"2 destination points are expected in dst, but {len(dst)} were found"
else:
assert len(src) == len(
dst
), f"Amount of (x, y, z) coordinates in src ({len(src)}) and dst ({len(dst)} must be equal"
assert len(src) >= 6, f"for (x, y, z) points, at least 6 pairs must be available, only {len(src)} provided"
if h_ref is not None:
assert isinstance(h_ref, (float, int)), "h_ref must contain a float number"
if z_0 is not None:
assert isinstance(z_0, (float, int)), "z_0 must be provided as type float"
assert all(isinstance(x, (float, int)) for p in src for x in p), "src contains non-int parts"
assert all(isinstance(x, (float, int)) for p in dst for x in p), "dst contains non-float parts"
if crs is not None:
if not (hasattr(self, "crs")):
raise ValueError(
"CameraConfig does not contain a crs, so gcps also cannot contain a crs. Ensure that the provided "
"destination coordinates are in a locally defined coordinate reference system, e.g. established "
"with a spirit level."
)
dst = helpers.xyz_transform(dst, crs, CRS.from_wkt(self.crs))
# if there is no h_ref, then no local gauge system, so set h_ref to zero
# check if 2 points are available
if len(src) == 2:
self.is_nadir = True
src, dst = cv._get_gcps_2_4(src, dst, self.width, self.height)
if h_ref is None:
h_ref = 0.0
self.gcps = {
"src": src,
"dst": dst,
"h_ref": h_ref,
"z_0": z_0,
}
[docs] def set_lens_position(self, x: float, y: float, z: float, crs: Optional[Any] = None):
"""Set the geographical position of the lens of current CameraConfig.
Parameters
----------
x : float
x-coordinate
y : float
y-coordinate
z : float
z-coordinate
crs : int, dict or str, optional
Coordinate Reference System. Accepts EPSG codes (int or str) proj (str or dict) or wkt (str). CRS used to
measure the lens position (e.g. 4326 for WGS84 lat-lon). The position's x and y coordinates will
automatically be reprojected to the local crs of the CameraConfig.
"""
if crs is not None:
if self.crs is None:
raise ValueError("CameraConfig does not contain a crs, ")
x, y = helpers.xyz_transform([[x, y]], crs, self.crs)[0]
self.lens_position = [x, y, z]
[docs] def project_points(self, points: List[List], within_image=False, swap_y_coords=False) -> np.ndarray:
"""Project real world x, y, z coordinates into col, row coordinates on image.
If col, row coordinates are not allowed to go outside of the image frame, then set `within_image = True`.
Method uses the intrinsics and extrinsics and distortion parameters to perform the projection.
Parameters
----------
points : list of lists or array-like
list of points [x, y, z] in real world coordinates
within_image : bool, optional
Set coordinates to NaN if these fall outside of the image.
swap_y_coords : bool, optional
If set to True (default: False), y-coordinates will be swapped, in order to match plotting defaults
which return row counting from top to bottom instead of bottom to top.
Returns
-------
points_project : list or array-like
list of points (equal in length as points) with [col, row] coordinates
"""
rvec, tvec = np.array(self.rvec), np.array(self.tvec)
# normalize points wrt mean of gcps
points = np.array(points, dtype=np.float64)
points_proj, jacobian = cv2.projectPoints(
points, rvec, tvec, np.array(self.camera_matrix), np.array(self.dist_coeffs)
)
points_proj = np.array([list(point[0]) for point in points_proj])
# points_back = cv.unproject_points(src=points_proj, z=points[:, -1], )
if within_image:
# also filter points outside edges of image
points_proj[points_proj[:, 0] < 0, 0] = -1.0
points_proj[points_proj[:, 0] > self.width - 1, 0] = self.width
points_proj[points_proj[:, 1] < 0, 1] = -1.0
points_proj[points_proj[:, 1] > self.height - 1, 1] = self.height
# check which points lie behind the camera
R, _ = cv2.Rodrigues(rvec)
points_camera = cv.world_to_camera(points, rvec, tvec)
behind_camera = points_camera[:, 2] <= 0.0
# set points behind camera to nan
points_proj[behind_camera, :] = np.nan
# swap y coords if set
if swap_y_coords:
points_proj[:, 1] = self.height - points_proj[:, 1]
return points_proj
[docs] def project_grid(self, xs, ys, zs, swap_y_coords=False):
"""Project gridded coordinates to col, row coordinates on image.
Method uses the intrinsics and extrinsics and distortion parameters to perform the projection.
Parameters
----------
xs : np.ndarray
2d array of real-world x-coordinates
ys : np.ndarray
2d array of real-world y-coordinates
zs : np.ndarray
2d array of real-world z-coordinates
swap_y_coords : bool, optional
If set to True (default: False), y-coordinates will be swapped, in order to match plotting defaults
which return row counting from top to bottom instead of bottom to top.
Returns
-------
xp : np.ndarray
list of col coordinates of image objective
yp : np.ndarray
list of row coordinates of image objective
"""
points = list(zip(xs.flatten(), ys.flatten(), zs.flatten(), strict=False))
points_proj = np.array(self.project_points(points, swap_y_coords=swap_y_coords))
xp, yp = points_proj[:, 0], points_proj[:, 1]
# reshape back
xp = np.reshape(xp, (len(xs), -1))
yp = np.reshape(yp, (len(xs), -1))
return xp, yp
[docs] def unproject_points(self, points: List[List], zs: Union[float, List[float]]) -> np.ndarray:
"""Reverse projects points in [column, row] space to [x, y, z] real world.
Parameters
----------
points : List of lists or array-like
Points in [col, row] to unproject
zs : float or list of floats
z-coordinates on which to unproject points
Returns
-------
points_unproject : List of lists or array-like
unprojected points as list of [x, y, z] coordinates
"""
rvec, tvec = self.rvec, self.tvec
# reduce zs by the mean of the gcps
dst = cv.unproject_points(
np.array(points, dtype=np.float64),
zs,
rvec=rvec,
tvec=tvec,
camera_matrix=self.camera_matrix,
dist_coeffs=self.dist_coeffs,
)
dst = np.array(dst, dtype=np.float64)
return dst
[docs] def plot(
self,
figsize: Optional[Tuple] = (13, 8),
ax: Optional[plt.Axes] = None,
tiles: Optional[Any] = None,
buffer: Optional[float] = 0.0005,
zoom_level: Optional[int] = 19,
camera: Optional[bool] = False,
mode: Optional[MODES] = "geographical",
pose_length: float = 1.0,
tiles_kwargs: Optional[Dict] = None,
) -> plt.Axes:
"""Plot geographical situation of the CameraConfig.
This is very useful to check if the CameraConfig seems to be in the right location. Requires `cartopy`
to be installed.
Parameters
----------
figsize : tuple, optional
width and height of figure (Default value = (13)
ax : plt.axes, optional
if not provided, axes is setup (Default: None)
tiles : str, optional
name of tiler service to use (called as attribute from cartopy.io.img_tiles) (Default: None)
buffer : float, optional
buffer in lat-lon around points, used to set extent (default: 0.0005)
zoom_level : int, optional
zoom level of image tiler service (default: 18)
camera : bool, optional
If set to True, all camera config information will be back projected to the original camera objective.
This option is deprecated, instead use mode="camera".
mode : Literal["geographical", "camera", "3d"], optional
Determines the type of bounding box to return. If set to "geographical" (default), the bounding box
is returned in the geographical coordinates. If set to "camera", the bounding box is returned in the
camera perspective. If set to "3d", the bounding box is returned as a 3D polygon in the CRS
pose_length: float, optional
length of pose axes to draw (only used in mode="3d").
tiles_kwargs : dict
additional keyword arguments to pass to ax.add_image when tiles are added
Returns
-------
ax : plt.axes
"""
if camera:
import warnings
warnings.warn(
"The camera=True option is deprecated, use mode='camera' instead. This option will be removed in "
"a future release.",
DeprecationWarning,
stacklevel=2,
)
mode = "camera"
if not tiles_kwargs:
tiles_kwargs = {}
# initiate transform
transformer = None
# if there is an axes, get the extent
xlim = ax.get_xlim() if ax is not None else None
ylim = ax.get_ylim() if ax is not None else None
# prepare points for plotting
if mode == "camera":
points = [Point(x, y) for x, y in self.gcps["src"]]
elif mode == "geographical":
points = [Point(p[0], p[1]) for p in self.gcps["dst"]]
else:
# 3d points are needed
if len(self.gcps["dst"]) == 3:
points = [Point(*p) for p in self.gcps["dst"]]
else:
points = [Point(p[0], p[1], self.gcps["z_0"]) for p in self.gcps["dst"]]
if mode != "camera":
if self.lens_position is not None:
lens_position = self.lens_position
else:
lens_position = self.estimate_lens_position()
if mode == "3d":
points.append(Point(*lens_position))
else:
points.append(Point(lens_position[0], lens_position[1]))
# transform points in case a crs is provided and we want a geographical plot
if mode == "geographical" and hasattr(self, "crs"):
# make a transformer to lat lon
transformer = Transformer.from_crs(
CRS.from_user_input(self.crs), CRS.from_epsg(4326), always_xy=True
).transform
points = [ops.transform(transformer, p) for p in points]
if mode == "geographical":
xmin, ymin, xmax, ymax = list(np.array(LineString(points).bounds))
extent = [xmin - buffer, xmax + buffer, ymin - buffer, ymax + buffer]
x = [p.x for p in points]
y = [p.y for p in points]
if mode == "3d":
z = [p.z for p in points]
if ax is None:
plt.figure(figsize=figsize)
if hasattr(self, "crs") and mode == "geographical":
ax = helpers.get_geo_axes(tiles=tiles, extent=extent, zoom_level=zoom_level, **tiles_kwargs)
else:
if mode == "3d":
ax = plt.axes(projection="3d")
else:
ax = plt.axes()
if hasattr(ax, "add_geometries"):
import cartopy.crs as ccrs
plot_kwargs = dict(transform=ccrs.PlateCarree())
else:
plot_kwargs = {}
if mode == "3d":
ax.plot(
x[0 : len(self.gcps["dst"])],
y[0 : len(self.gcps["dst"])],
z[0 : len(self.gcps["dst"])],
"o",
label="Control points",
markersize=12,
markeredgecolor="w",
zorder=2,
**plot_kwargs,
)
else:
ax.plot(
x[0 : len(self.gcps["dst"])],
y[0 : len(self.gcps["dst"])],
".",
label="Control points",
markersize=12,
markeredgecolor="w",
zorder=2,
**plot_kwargs,
)
if len(x) > len(self.gcps["dst"]):
if mode == "3d":
ax.plot(
x[-1],
y[-1],
z[-1],
"o",
label="Lens position",
markersize=12,
zorder=2,
markeredgecolor="w",
**plot_kwargs,
)
# add pose
_ = self.plot_3d_pose(ax=ax, length=pose_length)
if hasattr(self, "bbox"):
# also plot dashed lines from cam to bbox
for xy in self.bbox.exterior.coords:
ax.plot([x[-1], xy[0]], [y[-1], xy[1]], [z[-1], self.gcps["z_0"]], linestyle="--", color="gray")
# plot bbox exterior
ax.plot(*self.bbox.exterior.xy, [self.gcps["z_0"]] * 5, color="k", label="bbox exterior")
else:
ax.plot(
x[-1],
y[-1],
".",
label="Lens position",
markersize=12,
zorder=2,
markeredgecolor="w",
**plot_kwargs,
)
patch_kwargs = {
**plot_kwargs,
"alpha": 0.5,
"zorder": 2,
"edgecolor": "w",
"label": "bbox visible",
**plot_kwargs,
}
if hasattr(self, "bbox"):
self.plot_bbox(ax=ax, mode=mode, transformer=transformer, within_image=True, **patch_kwargs)
if mode == "camera":
# make sure that zero is on the top
ax.set_aspect("equal")
if xlim is not None:
ax.set_xlim(xlim)
ax.set_ylim(ylim)
ax.set_xlabel("column [-]")
ax.set_ylabel("row [-]")
elif mode == "3d":
ax.set_xlabel("x [m]")
ax.set_ylabel("y [m]")
ax.set_zlabel("z [m]")
ax.legend()
return ax
[docs] def plot_bbox(
self,
ax: Optional[plt.Axes] = None,
camera: Optional[bool] = False,
mode: Optional[MODES] = "geographical",
transformer: Optional[Any] = None,
h_a: Optional[float] = None,
within_image: Optional[bool] = True,
**kwargs,
):
"""Plot bounding box.
This can be done for orthorectification in a geographical projection (`camera=False`) or the camera
Field Of View (`mode="camera"`).
Parameters
----------
ax : plt.axes, optional
if not provided, axes is setup (Default: None)
camera : bool, optional
If set to True, all camera config information will be back projected to the original camera objective.
This option is deprecated, instead use mode="camera".
mode : Literal["geographical", "camera", "3d"], optional
Determines the type of bounding box to return. If set to "geographical" (default), the bounding box
is returned in the geographical coordinates. If set to "camera", the bounding box is returned in the
camera perspective. If set to "3d", the bounding box is returned as a 3D polygon in the CRS
transformer : pyproj transformer transformation function, optional
used to reproject bbox to axes object projection (e.g. lat lon)
h_a : float, optional
If set with `mode="camera"`, then the bbox coordinates will be transformed to the camera perspective,
using h_a as a present water level. In case a video with higher (lower) water levels is used, this
will result in a different perspective plane than the control video.
within_image : bool, optional
If set (default), points outside the camera objective are removed.
**kwargs : dict
additional keyword arguments used for plotting the bbox polygon with `matplotlib.patches.Polygon`
Returns
-------
p : matplotlib.patch mappable
"""
# collect information to plot
if camera:
import warnings
warnings.warn(
"The camera=True option is deprecated, use mode='camera' instead. This option will be removed in "
"a future release.",
DeprecationWarning,
stacklevel=2,
)
mode = "camera"
bbox = self.get_bbox(mode=mode, h_a=h_a, within_image=within_image)
if mode == "geographical" and transformer is not None:
# geographical projection is needed
bbox = ops.transform(transformer, bbox)
if mode == "3d":
return plot_helpers.plot_3d_polygon(bbox, ax=ax, **kwargs)
return plot_helpers.plot_polygon(bbox, ax=ax, **kwargs)
#
# bbox_x, bbox_y = bbox.exterior.xy
# bbox_coords = list(zip(bbox_x, bbox_y, strict=False))
# patch = patches.Polygon(bbox_coords, **kwargs)
# p = ax.add_patch(patch)
# return p
def plot_3d_pose(self, ax=None, length=1):
"""Plot 3D pose of a camera using its rotation and translation vectors.
Parameters
----------
ax : axes, optional
3d axes to plot on, if not set, a new axes will be established.
length : float
length of the axes drawn in meters
Returns
-------
list[handles]
list of handles to the plotted pose axes
"""
rvec = np.array(self.rvec)
tvec = np.array(self.tvec)
# rvec, tvec = cv.pose_world_to_camera(rvec, tvec)
# Convert the rotation vector to a 3x3 rotation matrix
R, _ = cv2.Rodrigues(rvec.flatten())
# Define the camera's axis directions in its local coordinate system
camera_axes = (
np.array(
[
[0, 0, 0], # lens center
[1, 0, 0], # X-axis (red - right looking)
[0, 1, 0], # Y-axis (green - down looking)
[0, 0, 1], # Z-axis (blue - forward looking)
]
)
* length
)
pts_trans = camera_axes - tvec
world_axes_translated = (R.T @ pts_trans.T).T
ax = plt.axes(projection="3d") if ax is None else ax
# Plot the origin of the camera
ps = []
# Plot the camera axes
for i, (color, label) in enumerate(
zip(["r", "g", "b"], ["right-pose", "down-pose", "forward-pose"], strict=False)
):
# if i == 2:
xx = [world_axes_translated[0, 0], world_axes_translated[i + 1, 0]]
yy = [world_axes_translated[0, 1], world_axes_translated[i + 1, 1]]
zz = [world_axes_translated[0, 2], world_axes_translated[i + 1, 2]]
ps.append(ax.plot(xx, yy, zz, color=color, label=label, linewidth=3))
return ps
[docs] def to_dict(self) -> Dict:
"""Return the CameraConfig object as dictionary.
Returns
-------
camera_config_dict : dict
serialized CameraConfig
"""
d = copy.deepcopy(self.__dict__)
# replace underscore keys for keys without underscore
for k in list(d.keys()):
if k[0] == "_":
d[k[1:]] = d.pop(k)
return d
[docs] def to_dict_str(self) -> Dict:
"""Convert the current instance to a dictionary with all values converted to strings.
Returns
-------
dict
A dictionary representation of the instance where all values are strings. If an attribute is an instance
of `Polygon`, it is converted to its string representation.
"""
d = self.to_dict()
# convert anything that is not string in string
dict_str = {k: v if not (isinstance(v, Polygon)) else v.__str__() for k, v in d.items()}
return dict_str
[docs] def to_file(self, fn: str):
"""Write the CameraConfig object to json structure.
Parameters
----------
fn : str
Path to file to write camera config to
"""
with open(fn, "w") as f:
f.write(self.to_json())
[docs] def to_json(self) -> str:
"""Convert CameraConfig object to string.
Returns
-------
json_str : str
json string with CameraConfig components
"""
return json.dumps(self, default=lambda o: o.to_dict_str(), indent=4)
depr_warning_height_width = """
Your camera configuration does not have a property "height" and/or "width", probably because your configuration file is
from an older < 0.3.0 version. Please rectify this by editing your .json config file. The top of your file should e.g.
look as follows for a HD video:
{
"height": 1080,
"width": 1920,
"crs": ....
...
}
"""
def get_camera_config(s: str) -> CameraConfig:
"""Read camera config from string.
Parameters
----------
s : str
json string containing camera config
Returns
-------
cam_config : CameraConfig
"""
d = json.loads(s)
if "height" not in d or "width" not in d:
raise IOError(depr_warning_height_width)
# ensure the bbox is a Polygon object
if "bbox" in d:
if isinstance(d["bbox"], str):
d["bbox"] = wkt.loads(d["bbox"])
return CameraConfig(**d)
def load_camera_config(fn: str) -> CameraConfig:
"""Load a CameraConfig from a geojson file.
Parameters
----------
fn : str
path to file with camera config in json format.
Returns
-------
cam_config : CameraConfig
"""
with open(fn, "r") as f:
camera_config = get_camera_config(f.read())
return camera_config
