Tie_Point_Grid.py 46.1 KB
Newer Older
1
2
3
4
5
6
# -*- coding: utf-8 -*-

import collections
import multiprocessing
import os
import warnings
7
import time
8
9

# custom
10
11
12
13
try:
    import gdal
except ImportError:
    from osgeo import gdal
14
import numpy as np
15
from matplotlib import pyplot as plt
16
17
18
from geopandas import GeoDataFrame, GeoSeries
from shapely.geometry import Point
from skimage.measure import points_in_poly, ransac
19
from skimage.transform import AffineTransform, PolynomialTransform
20
21

# internal modules
22
from .CoReg import COREG
23
from py_tools_ds.geo.projection import isProjectedOrGeographic, isLocal, get_UTMzone, dict_to_proj4, proj4_to_WKT
24
from py_tools_ds.io.pathgen import get_generic_outpath
25
from py_tools_ds.processing.progress_mon import ProgressBar
26
from py_tools_ds.geo.vector.conversion import points_to_raster
27
from py_tools_ds.io.vector.writer import write_shp
28
from geoarray import GeoArray
29

30
from .CoReg import GeoArray_CoReg  # noqa F401  # flake8 issue
31

32
__author__ = 'Daniel Scheffler'
33

34
global_shared_imref = None
35
36
37
global_shared_im2shift = None


38
39
40
41
42
43
44
45
46
47
48
def mp_initializer(imref, imtgt):
    """Declare global variables needed for self._get_spatial_shifts()

    :param imref:   reference image
    :param imtgt:   target image
    """
    global global_shared_imref, global_shared_im2shift
    global_shared_imref = imref
    global_shared_im2shift = imtgt


49
50
class Tie_Point_Grid(object):
    """See help(Tie_Point_Grid) for documentation!"""
51

52
    def __init__(self, COREG_obj, grid_res, max_points=None, outFillVal=-9999, resamp_alg_calc='cubic',
53
54
                 tieP_filter_level=3, outlDetect_settings=None, dir_out=None, CPUs=None, progress=True, v=False,
                 q=False):
55

56
57
58
        """Applies the algorithm to detect spatial shifts to the whole overlap area of the input images. Spatial shifts
        are calculated for each point in grid of which the parameters can be adjusted using keyword arguments. Shift
        correction performs a polynomial transformation using te calculated shifts of each point in the grid as GCPs.
59
        Thus 'Tie_Point_Grid' can be used to correct for locally varying geometric distortions of the target image.
60

61
        :param COREG_obj(object):       an instance of COREG class
62
        :param grid_res:                grid resolution in pixels of the target image (x-direction)
63
        :param max_points(int):         maximum number of points used to find coregistration tie points
64
65
66
                                        NOTE: Points are selected randomly from the given point grid (specified by
                                        'grid_res'). If the point does not provide enough points, all available points
                                        are chosen.
Daniel Scheffler's avatar
Daniel Scheffler committed
67
        :param outFillVal(int):         if given the generated tie points grid is filled with this value in case
68
                                        no match could be found during co-registration (default: -9999)
69
70
        :param resamp_alg_calc(str)     the resampling algorithm to be used for all warping processes during calculation
                                        of spatial shifts
71
72
                                        (valid algorithms: nearest, bilinear, cubic, cubic_spline, lanczos, average,
                                                           mode, max, min, med, q1, q3)
73
                                        default: cubic (highly recommended)
74
        :param tieP_filter_level(int):  filter tie points used for shift correction in different levels (default: 3).
75
                                        NOTE: lower levels are also included if a higher level is chosen
76
                                            - Level 0: no tie point filtering
77
78
79
                                            - Level 1: Reliablity filtering - filter all tie points out that have a low
                                                reliability according to internal tests
                                            - Level 2: SSIM filtering - filters all tie points out where shift
80
81
                                                correction does not increase image similarity within matching window
                                                (measured by mean structural similarity index)
82
                                            - Level 3: RANSAC outlier detection
83
84
85
86
        :param outlDetect_settings      a dictionary with the settings to be passed to
                                        arosics.TiePointGrid.Tie_Point_Refiner. Available keys: min_reliability,
                                        rs_max_outlier, rs_tolerance, rs_max_iter, rs_exclude_previous_outliers,
                                        rs_timeout, q. See documentation there.
87
88
        :param dir_out(str):            output directory to be used for all outputs if nothing else is given
                                        to the individual methods
Daniel Scheffler's avatar
Daniel Scheffler committed
89
        :param CPUs(int):               number of CPUs to use during calculation of tie points grid
90
                                        (default: None, which means 'all CPUs available')
91
        :param progress(bool):          show progress bars (default: True)
92
93
        :param v(bool):                 verbose mode (default: False)
        :param q(bool):                 quiet mode (default: False)
94
        """
95

96
97
        if not isinstance(COREG_obj, COREG):
            raise ValueError("'COREG_obj' must be an instance of COREG class.")
98

99
        self.COREG_obj = COREG_obj  # type: COREG
100
101
102
103
        self.grid_res = grid_res
        self.max_points = max_points
        self.outFillVal = outFillVal
        self.rspAlg_calc = resamp_alg_calc
104
        self.tieP_filter_level = tieP_filter_level
105
        self.outlDetect_settings = outlDetect_settings if outlDetect_settings else dict(q=q)
106
107
108
109
110
        self.dir_out = dir_out
        self.CPUs = CPUs
        self.v = v
        self.q = q if not v else False  # overridden by v
        self.progress = progress if not q else False  # overridden by q
111

112
113
        self.ref = self.COREG_obj.ref  # type: GeoArray_CoReg
        self.shift = self.COREG_obj.shift  # type: GeoArray_CoReg
114

115
        self.XY_points, self.XY_mapPoints = self._get_imXY__mapXY_points(self.grid_res)
116
117
118
119
        self._CoRegPoints_table = None  # set by self.CoRegPoints_table
        self._GCPList = None  # set by self.to_GCPList()
        self.kriged = None  # set by Raster_using_Kriging()

120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
    @property
    def mean_x_shift_px(self):
        return self.CoRegPoints_table['X_SHIFT_PX'][self.CoRegPoints_table['X_SHIFT_PX'] != self.outFillVal].mean()

    @property
    def mean_y_shift_px(self):
        return self.CoRegPoints_table['Y_SHIFT_PX'][self.CoRegPoints_table['Y_SHIFT_PX'] != self.outFillVal].mean()

    @property
    def mean_x_shift_map(self):
        return self.CoRegPoints_table['X_SHIFT_M'][self.CoRegPoints_table['X_SHIFT_M'] != self.outFillVal].mean()

    @property
    def mean_y_shift_map(self):
        return self.CoRegPoints_table['Y_SHIFT_M'][self.CoRegPoints_table['Y_SHIFT_M'] != self.outFillVal].mean()
135

136
137
    @property
    def CoRegPoints_table(self):
138
139
        """Returns a GeoDataFrame with the columns 'geometry','POINT_ID','X_IM','Y_IM','X_UTM','Y_UTM','X_WIN_SIZE',
        'Y_WIN_SIZE','X_SHIFT_PX','Y_SHIFT_PX', 'X_SHIFT_M', 'Y_SHIFT_M', 'ABS_SHIFT' and 'ANGLE' containing all
140
        information containing all the results from coregistration for all points in the tie points grid.
141
        """
142
143
144
145
146
147
148
149
150
151
152
153
        if self._CoRegPoints_table is not None:
            return self._CoRegPoints_table
        else:
            self._CoRegPoints_table = self.get_CoRegPoints_table()
            return self._CoRegPoints_table

    @CoRegPoints_table.setter
    def CoRegPoints_table(self, CoRegPoints_table):
        self._CoRegPoints_table = CoRegPoints_table

    @property
    def GCPList(self):
154
155
        """Returns a list of GDAL compatible GCP objects.
        """
Daniel Scheffler's avatar
Daniel Scheffler committed
156

157
158
159
160
        if self._GCPList:
            return self._GCPList
        else:
            self._GCPList = self.to_GCPList()
161
            return self._GCPList
162
163
164
165
166
167

    @GCPList.setter
    def GCPList(self, GCPList):
        self._GCPList = GCPList

    def _get_imXY__mapXY_points(self, grid_res):
168
169
170
171
172
173
        """Returns a numpy array containing possible positions for coregistration tie points according to the given
        grid resolution.

        :param grid_res:
        :return:
        """
Daniel Scheffler's avatar
Daniel Scheffler committed
174

175
        if not self.q:
Daniel Scheffler's avatar
Daniel Scheffler committed
176
            print('Initializing tie points grid...')
177

178
179
        Xarr, Yarr = np.meshgrid(np.arange(0, self.shift.shape[1], grid_res),
                                 np.arange(0, self.shift.shape[0], grid_res))
180

181
182
        mapXarr = np.full_like(Xarr, self.shift.gt[0], dtype=np.float64) + Xarr * self.shift.gt[1]
        mapYarr = np.full_like(Yarr, self.shift.gt[3], dtype=np.float64) - Yarr * abs(self.shift.gt[5])
183

184
185
186
        XY_points = np.empty((Xarr.size, 2), Xarr.dtype)
        XY_points[:, 0] = Xarr.flat
        XY_points[:, 1] = Yarr.flat
187

188
189
190
        XY_mapPoints = np.empty((mapXarr.size, 2), mapXarr.dtype)
        XY_mapPoints[:, 0] = mapXarr.flat
        XY_mapPoints[:, 1] = mapYarr.flat
191

Daniel Scheffler's avatar
Daniel Scheffler committed
192
193
        assert XY_points.shape == XY_mapPoints.shape

194
        return XY_points, XY_mapPoints
195

196
197
198
199
200
201
202
203
204
205
206
    def _exclude_bad_XYpos(self, GDF):
        """Excludes all points outside of the image overlap area and all points where the bad data mask is True (if given).

        :param GDF:     <geopandas.GeoDataFrame> must include the columns 'X_UTM' and 'Y_UTM'
        :return:
        """

        # exclude all points outside of overlap area
        inliers = points_in_poly(self.XY_mapPoints,
                                 np.swapaxes(np.array(self.COREG_obj.overlap_poly.exterior.coords.xy), 0, 1))
        GDF = GDF[inliers].copy()
207
        # GDF = GDF[GDF['geometry'].within(self.COREG_obj.overlap_poly.simplify(tolerance=15))] # works but much slower
208

209
210
        # FIXME track that
        assert not GDF.empty, 'No coregistration point could be placed within the overlap area. Check your input data!'
211
212

        # exclude all point where bad data mask is True (e.g. points on clouds etc.)
213
214
215
216
217
218
        orig_len_GDF = len(GDF)  # length of GDF after dropping all points outside the overlap polygon
        mapXY = np.array(GDF.loc[:, ['X_UTM', 'Y_UTM']])
        GDF['REF_BADDATA'] = self.COREG_obj.ref.mask_baddata.read_pointData(mapXY) \
            if self.COREG_obj.ref.mask_baddata is not None else False
        GDF['TGT_BADDATA'] = self.COREG_obj.shift.mask_baddata.read_pointData(mapXY) \
            if self.COREG_obj.shift.mask_baddata is not None else False
Daniel Scheffler's avatar
Daniel Scheffler committed
219
        GDF = GDF[(~GDF['REF_BADDATA']) & (~GDF['TGT_BADDATA'])]
220
        if self.COREG_obj.ref.mask_baddata is not None or self.COREG_obj.shift.mask_baddata is not None:
Daniel Scheffler's avatar
Daniel Scheffler committed
221
222
            if not self.q:
                print('According to the provided bad data mask(s) %s points of initially %s have been excluded.'
223
                      % (orig_len_GDF - len(GDF), orig_len_GDF))
224
225
226

        return GDF

227
228
    @staticmethod
    def _get_spatial_shifts(coreg_kwargs):
Daniel Scheffler's avatar
Daniel Scheffler committed
229
        # unpack
230
        pointID = coreg_kwargs['pointID']
231
232
        fftw_works = coreg_kwargs['fftw_works']
        del coreg_kwargs['pointID'], coreg_kwargs['fftw_works']
233

Daniel Scheffler's avatar
Daniel Scheffler committed
234
        # assertions
235
        assert global_shared_imref is not None
236
        assert global_shared_im2shift is not None
Daniel Scheffler's avatar
Daniel Scheffler committed
237
238

        # run CoReg
239
        CR = COREG(global_shared_imref, global_shared_im2shift, CPUs=1, **coreg_kwargs)
240
        CR.fftw_works = fftw_works
241
        CR.calculate_spatial_shifts()
Daniel Scheffler's avatar
Daniel Scheffler committed
242
243

        # fetch results
244
        last_err = CR.tracked_errors[-1] if CR.tracked_errors else None
245
        win_sz_y, win_sz_x = CR.matchBox.imDimsYX if CR.matchBox else (None, None)
246
247
248
        CR_res = [win_sz_x, win_sz_y, CR.x_shift_px, CR.y_shift_px, CR.x_shift_map, CR.y_shift_map,
                  CR.vec_length_map, CR.vec_angle_deg, CR.ssim_orig, CR.ssim_deshifted, CR.ssim_improved,
                  CR.shift_reliability, last_err]
249

250
        return [pointID] + CR_res
251

252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
    def _get_coreg_kwargs(self, pID, wp):
        return dict(
            pointID=pID,
            fftw_works=self.COREG_obj.fftw_works,
            wp=wp,
            ws=self.COREG_obj.win_size_XY,
            resamp_alg_calc=self.rspAlg_calc,
            footprint_poly_ref=self.COREG_obj.ref.poly,
            footprint_poly_tgt=self.COREG_obj.shift.poly,
            r_b4match=self.ref.band4match + 1,  # band4match is internally saved as index, starting from 0
            s_b4match=self.shift.band4match + 1,  # band4match is internally saved as index, starting from 0
            max_iter=self.COREG_obj.max_iter,
            max_shift=self.COREG_obj.max_shift,
            nodata=(self.COREG_obj.ref.nodata, self.COREG_obj.shift.nodata),
            force_quadratic_win=self.COREG_obj.force_quadratic_win,
            binary_ws=self.COREG_obj.bin_ws,
            v=False,  # otherwise this would lead to massive console output
            q=True,  # otherwise this would lead to massive console output
            ignore_errors=True
        )

273
    def get_CoRegPoints_table(self):
274
275
        assert self.XY_points is not None and self.XY_mapPoints is not None

276
277
278
279
        # create a dataframe containing 'geometry','POINT_ID','X_IM','Y_IM','X_UTM','Y_UTM'
        # (convert imCoords to mapCoords
        XYarr2PointGeom = np.vectorize(lambda X, Y: Point(X, Y), otypes=[Point])
        geomPoints = np.array(XYarr2PointGeom(self.XY_mapPoints[:, 0], self.XY_mapPoints[:, 1]))
280

281
282
283
        if isLocal(self.COREG_obj.shift.prj):
            crs = None
        elif isProjectedOrGeographic(self.COREG_obj.shift.prj) == 'geographic':
284
            crs = dict(ellps='WGS84', datum='WGS84', proj='longlat')
285
        elif isProjectedOrGeographic(self.COREG_obj.shift.prj) == 'projected':
286
            UTMzone = abs(get_UTMzone(prj=self.COREG_obj.shift.prj))
287
288
289
290
            south = get_UTMzone(prj=self.COREG_obj.shift.prj) < 0
            crs = dict(ellps='WGS84', datum='WGS84', proj='utm', zone=UTMzone, south=south, units='m', no_defs=True)
            if not south:
                del crs['south']
291
292
293
        else:
            crs = None

294
295
296
        GDF = GeoDataFrame(index=range(len(geomPoints)), crs=crs,
                           columns=['geometry', 'POINT_ID', 'X_IM', 'Y_IM', 'X_UTM', 'Y_UTM'])
        GDF['geometry'] = geomPoints
297
        GDF['POINT_ID'] = np.array(range(len(geomPoints))).astype(np.object)
298
299
        GDF.loc[:, ['X_IM', 'Y_IM']] = self.XY_points
        GDF.loc[:, ['X_UTM', 'Y_UTM']] = self.XY_mapPoints
300

301
302
        # exclude offsite points and points on bad data mask
        GDF = self._exclude_bad_XYpos(GDF)
303
304
305
        if GDF.empty:
            self.CoRegPoints_table = GDF
            return self.CoRegPoints_table
306

307
        # choose a random subset of points if a maximum number has been given
308
        if self.max_points and len(GDF) > self.max_points:
309
            GDF = GDF.sample(self.max_points).copy()
310

311
        # equalize pixel grids in order to save warping time
312
313
314
315
        if len(GDF) > 100:
            # NOTE: actually grid res should be also changed here because self.shift.xgsd changes and grid res is
            # connected to that
            self.COREG_obj.equalize_pixGrids()
316

317
        # validate reference and target image inputs
318
        assert self.ref.footprint_poly  # this also checks for mask_nodata and nodata value
319
        assert self.shift.footprint_poly
320
321
322

        # ensure the input arrays for CoReg are in memory -> otherwise the code will get stuck in multiprocessing if
        # neighboured matching windows overlap during reading from disk!!
323
324
        self.ref.cache_array_subset(
            [self.COREG_obj.ref.band4match])  # only sets geoArr._arr_cache; does not change number of bands
Daniel Scheffler's avatar
Daniel Scheffler committed
325
326
        self.shift.cache_array_subset([self.COREG_obj.shift.band4match])

327
        # get all variations of kwargs for coregistration
328
        list_coreg_kwargs = (self._get_coreg_kwargs(i, self.XY_mapPoints[i]) for i in GDF.index)  # generator
329
330

        # run co-registration for whole grid
331
        if self.CPUs is None or self.CPUs > 1:
Daniel Scheffler's avatar
CoReg:    
Daniel Scheffler committed
332
            if not self.q:
333
                cpus = self.CPUs if self.CPUs is not None else multiprocessing.cpu_count()
334
                print("Calculating tie point grid (%s points) using %s CPU cores..." % (len(GDF), cpus))
335

336
            with multiprocessing.Pool(self.CPUs, initializer=mp_initializer, initargs=(self.ref, self.shift)) as pool:
337
338
339
340
                if self.q or not self.progress:
                    results = pool.map(self._get_spatial_shifts, list_coreg_kwargs)
                else:
                    results = pool.map_async(self._get_spatial_shifts, list_coreg_kwargs, chunksize=1)
341
                    bar = ProgressBar(prefix='\tprogress:')
342
343
                    while True:
                        time.sleep(.1)
344
345
                        # this does not really represent the remaining tasks but the remaining chunks
                        # -> thus chunksize=1
Daniel Scheffler's avatar
Fix.    
Daniel Scheffler committed
346
                        # noinspection PyProtectedMember
347
                        numberDone = len(GDF) - results._number_left
348
                        if self.progress:
349
                            bar.print_progress(percent=numberDone / len(GDF) * 100)
350
                        if results.ready():
351
352
353
                            # <= this is the line where multiprocessing can freeze if an exception appears within
                            # COREG ans is not raised
                            results = results.get()
354
                            break
Daniel Scheffler's avatar
Daniel Scheffler committed
355

356
        else:
357
358
359
360
361
            # declare global variables needed for self._get_spatial_shifts()
            global global_shared_imref, global_shared_im2shift
            global_shared_imref = self.ref
            global_shared_im2shift = self.shift

Daniel Scheffler's avatar
CoReg:    
Daniel Scheffler committed
362
            if not self.q:
363
                print("Calculating tie point grid (%s points) 1 CPU core..." % len(GDF))
364
365
366
            results = np.empty((len(geomPoints), 14), np.object)
            bar = ProgressBar(prefix='\tprogress:')
            for i, coreg_kwargs in enumerate(list_coreg_kwargs):
367
                if self.progress:
368
369
                    bar.print_progress((i + 1) / len(GDF) * 100)
                results[i, :] = self._get_spatial_shifts(coreg_kwargs)
370

371
372
        # merge results with GDF
        records = GeoDataFrame(results,
373
                               columns=['POINT_ID', 'X_WIN_SIZE', 'Y_WIN_SIZE', 'X_SHIFT_PX', 'Y_SHIFT_PX', 'X_SHIFT_M',
374
                                        'Y_SHIFT_M', 'ABS_SHIFT', 'ANGLE', 'SSIM_BEFORE', 'SSIM_AFTER',
375
376
                                        'SSIM_IMPROVED', 'RELIABILITY', 'LAST_ERR'])

377
378
379
        GDF = GDF.merge(records, on='POINT_ID', how="inner")
        GDF = GDF.fillna(int(self.outFillVal))

380
381
382
        if not self.q:
            print("Found %s matches." % len(GDF[GDF.LAST_ERR == int(self.outFillVal)]))

383
        # filter tie points according to given filter level
384
        if self.tieP_filter_level > 0:
385
386
            if not self.q:
                print('Performing validity checks...')
387
            TPR = Tie_Point_Refiner(GDF[GDF.ABS_SHIFT != self.outFillVal], **self.outlDetect_settings)
388
            GDF_filt, new_columns = TPR.run_filtering(level=self.tieP_filter_level)
389
            GDF = GDF.merge(GDF_filt[['POINT_ID'] + new_columns], on='POINT_ID', how="outer")
390
        GDF = GDF.fillna(int(self.outFillVal))
391

392
        self.CoRegPoints_table = GDF
393
394
395

        return self.CoRegPoints_table

396
397
398
399
    def calc_rmse(self, include_outliers=False):
        # type: (bool) -> float
        """Calculates root mean square error of absolute shifts from the tie point grid.

Daniel Scheffler's avatar
Daniel Scheffler committed
400
        :param include_outliers:    whether to include tie points that have been marked as false-positives (if present)
401
402
403
        """

        tbl = self.CoRegPoints_table
404
        tbl = tbl if include_outliers else tbl[tbl['OUTLIER'] == 0].copy() if 'OUTLIER' in tbl.columns else tbl
405
406
407
408
409
410

        shifts = np.array(tbl['ABS_SHIFT'])
        shifts_sq = [i * i for i in shifts if i != self.outFillVal]

        return np.sqrt(sum(shifts_sq) / len(shifts_sq))

411
412
413
    def calc_overall_ssim(self, include_outliers=False, after_correction=True):
        # type: (bool, bool) -> float
        """Calculates the median value of all SSIM values contained in tie point grid.
414
415

        :param include_outliers:    whether to include tie points that have been marked as false-positives
416
        :param after_correction:    whether to compute median SSIM before correction or after
417
418
419
        """

        tbl = self.CoRegPoints_table
420
        tbl = tbl if include_outliers else tbl[tbl['OUTLIER'] == 0].copy()
421

422
423
        ssim_col = np.array(tbl['SSIM_AFTER' if after_correction else 'SSIM_BEFORE'])
        ssim_col = [i * i for i in ssim_col if i != self.outFillVal]
424

425
        return float(np.median(ssim_col))
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441

    def plot_shift_distribution(self, include_outliers=True, unit='m', interactive=False, figsize=None, xlim=None,
                                ylim=None, fontsize=12, title='shift distribution'):
        # type: (bool, str, bool, tuple, list, list, int) -> tuple
        """Creates a 2D scatterplot containing the distribution of calculated X/Y-shifts.

        :param include_outliers:    whether to include tie points that have been marked as false-positives
        :param unit:                'm' for meters or 'px' for pixels (default: 'm')
        :param interactive:         interactive mode uses plotly for visualization
        :param figsize:             (xdim, ydim)
        :param xlim:                [xmin, xmax]
        :param ylim:                [ymin, ymax]
        :param fontsize:            size of all used fonts
        :param title:               the title to be plotted above the figure
        """

442
443
        if unit not in ['m', 'px']:
            raise ValueError("Parameter 'unit' must have the value 'm' (meters) or 'px' (pixels)! Got %s." % unit)
444
445

        tbl = self.CoRegPoints_table
Daniel Scheffler's avatar
Daniel Scheffler committed
446
        tbl = tbl[tbl['ABS_SHIFT'] != self.outFillVal]
447
        tbl_il = tbl[tbl['OUTLIER'] == 0].copy() if 'OUTLIER' in tbl.columns else tbl
Daniel Scheffler's avatar
Daniel Scheffler committed
448
        tbl_ol = tbl[tbl['OUTLIER']].copy() if 'OUTLIER' in tbl.columns else None
449
450
        x_attr = 'X_SHIFT_M' if unit == 'm' else 'X_SHIFT_PX'
        y_attr = 'Y_SHIFT_M' if unit == 'm' else 'Y_SHIFT_PX'
451
452
        rmse = self.calc_rmse(include_outliers=False)  # always exclude outliers when calculating RMSE
        figsize = figsize if figsize else (10, 10)
453
454
455
456

        if interactive:
            from plotly.offline import iplot, init_notebook_mode
            import plotly.graph_objs as go
Daniel Scheffler's avatar
Daniel Scheffler committed
457
            # FIXME outliers are not plotted
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478

            init_notebook_mode(connected=True)

            # Create a trace
            trace = go.Scatter(
                x=tbl_il[x_attr],
                y=tbl_il[y_attr],
                mode='markers'
            )

            data = [trace]

            # Plot and embed in ipython notebook!
            iplot(data, filename='basic-scatter')

            return None, None

        else:
            fig = plt.figure(figsize=figsize)
            ax = fig.add_subplot(111)

Daniel Scheffler's avatar
Daniel Scheffler committed
479
            if include_outliers and 'OUTLIER' in tbl.columns:
Daniel Scheffler's avatar
Daniel Scheffler committed
480
                ax.scatter(tbl_ol[x_attr], tbl_ol[y_attr], marker='+', c='r', label='false-positives')
481
482
483
484
485
486
487
488
489
490
491
492
            ax.scatter(tbl_il[x_attr], tbl_il[y_attr], marker='+', c='g', label='valid tie points')

            # set axis limits
            if not xlim:
                xmax = np.abs(tbl_il[x_attr]).max()
                xlim = [-xmax, xmax]
            if not ylim:
                ymax = np.abs(tbl_il[y_attr]).max()
                ylim = [-ymax, ymax]
            ax.set_xlim(xlim)
            ax.set_ylim(ylim)

Daniel Scheffler's avatar
Daniel Scheffler committed
493
            # add text box containing RMSE of plotted shifts
494
            xlim, ylim = ax.get_xlim(), ax.get_ylim()
495
496
            plt.text(xlim[1] - (xlim[1] / 20), -ylim[1] + (ylim[1] / 20),
                     'RMSE:  %s m / %s px' % (np.round(rmse, 2), np.round(rmse / self.shift.xgsd, 2)),
497
                     ha='right', va='bottom', fontsize=fontsize, bbox=dict(facecolor='w', pad=None, alpha=0.8))
498

Daniel Scheffler's avatar
Daniel Scheffler committed
499
            # add grid and increase linewidth of middle line
500
501
502
            plt.grid()
            xgl = ax.get_xgridlines()
            middle_xgl = xgl[int(np.median(np.array(range(len(xgl)))))]
Daniel Scheffler's avatar
Daniel Scheffler committed
503
            middle_xgl.set_linewidth(2)
504
505
506
            middle_xgl.set_linestyle('-')
            ygl = ax.get_ygridlines()
            middle_ygl = ygl[int(np.median(np.array(range(len(ygl)))))]
Daniel Scheffler's avatar
Daniel Scheffler committed
507
            middle_ygl.set_linewidth(2)
508
509
            middle_ygl.set_linestyle('-')

Daniel Scheffler's avatar
Daniel Scheffler committed
510
511
            # set title and adjust tick labels
            ax.set_title(title, fontsize=fontsize)
512
513
            [tick.label.set_fontsize(fontsize) for tick in ax.xaxis.get_major_ticks()]
            [tick.label.set_fontsize(fontsize) for tick in ax.yaxis.get_major_ticks()]
Daniel Scheffler's avatar
Daniel Scheffler committed
514
515
            plt.xlabel('x-shift [%s]' % 'meters' if unit == 'm' else 'pixels', fontsize=fontsize)
            plt.ylabel('y-shift [%s]' % 'meters' if unit == 'm' else 'pixels', fontsize=fontsize)
516

517
518
            # add legend with labels in the right order
            handles, labels = ax.get_legend_handles_labels()
Daniel Scheffler's avatar
Daniel Scheffler committed
519
520
            leg = plt.legend(reversed(handles), reversed(labels), fontsize=fontsize, loc='upper right', scatterpoints=3)
            leg.get_frame().set_edgecolor('black')
521

522
523
524
525
            plt.show()

            return fig, ax

526
    def dump_CoRegPoints_table(self, path_out=None):
527
528
529
530
531
        path_out = path_out if path_out else \
            get_generic_outpath(dir_out=self.dir_out, fName_out="CoRegPoints_table_grid%s_ws(%s_%s)__T_%s__R_%s.pkl"
                                                                % (self.grid_res, self.COREG_obj.win_size_XY[0],
                                                                   self.COREG_obj.win_size_XY[1], self.shift.basename,
                                                                   self.ref.basename))
532
533
534
        if not self.q:
            print('Writing %s ...' % path_out)
        self.CoRegPoints_table.to_pickle(path_out)
535

536
    def to_GCPList(self):
Daniel Scheffler's avatar
Daniel Scheffler committed
537
        # get copy of tie points grid without no data
Daniel Scheffler's avatar
Daniel Scheffler committed
538
539
540
541
542
        try:
            GDF = self.CoRegPoints_table.loc[self.CoRegPoints_table.ABS_SHIFT != self.outFillVal, :].copy()
        except AttributeError:
            # self.CoRegPoints_table has no attribute 'ABS_SHIFT' because all points have been excluded
            return []
543

544
        if getattr(GDF, 'empty'):  # GDF.empty returns AttributeError
545
546
            return []
        else:
547
            # exclude all points flagged as outliers
548
            if 'OUTLIER' in GDF.columns:
549
                GDF = GDF[GDF.OUTLIER.__eq__(False)].copy()
550
551
            avail_TP = len(GDF)

552
553
554
555
            if not avail_TP:
                # no point passed all validity checks
                return []

556
            if avail_TP > 7000:
557
558
559
                GDF = GDF.sample(7000)
                warnings.warn('By far not more than 7000 tie points can be used for warping within a limited '
                              'computation time (due to a GDAL bottleneck). Thus these 7000 points are randomly chosen '
560
                              'out of the %s available tie points.' % avail_TP)
561

562
563
564
            # calculate GCPs
            GDF['X_UTM_new'] = GDF.X_UTM + GDF.X_SHIFT_M
            GDF['Y_UTM_new'] = GDF.Y_UTM + GDF.Y_SHIFT_M
565
566
            GDF['GCP'] = GDF.apply(lambda GDF_row: gdal.GCP(GDF_row.X_UTM_new, GDF_row.Y_UTM_new, 0,
                                                            GDF_row.X_IM, GDF_row.Y_IM), axis=1)
567
568
569
            self.GCPList = GDF.GCP.tolist()

            if not self.q:
570
                print('Found %s valid tie points.' % len(self.GCPList))
571
572

            return self.GCPList
573

574
    def test_if_singleprocessing_equals_multiprocessing_result(self):
575
576
        # RANSAC filtering always produces different results because it includes random sampling
        self.tieP_filter_level = 1
577

Daniel Scheffler's avatar
Daniel Scheffler committed
578
        self.CPUs = None
579
        dataframe = self.get_CoRegPoints_table()
580
        mp_out = np.empty_like(dataframe.values)
581
        mp_out[:] = dataframe.values
Daniel Scheffler's avatar
Daniel Scheffler committed
582
        self.CPUs = 1
583
        dataframe = self.get_CoRegPoints_table()
584
        sp_out = np.empty_like(dataframe.values)
585
586
        sp_out[:] = dataframe.values

587
        return np.array_equal(sp_out, mp_out)
588

589
590
    def _get_line_by_PID(self, PID):
        return self.CoRegPoints_table.loc[PID, :]
591

592
    def _get_lines_by_PIDs(self, PIDs):
593
594
595
596
        assert isinstance(PIDs, list)
        lines = np.zeros((len(PIDs), self.CoRegPoints_table.shape[1]))
        for i, PID in enumerate(PIDs):
            lines[i, :] = self.CoRegPoints_table[self.CoRegPoints_table['POINT_ID'] == PID]
597
598
        return lines

599
    def to_PointShapefile(self, path_out=None, skip_nodata=True, skip_nodata_col='ABS_SHIFT'):
600
        # type: (str, bool, str) -> None
Daniel Scheffler's avatar
Daniel Scheffler committed
601
        """Writes the calculated tie points grid to a point shapefile containing
602
        Tie_Point_Grid.CoRegPoints_table as attribute table. This shapefile can easily be displayed using GIS software.
603
604
605

        :param path_out:        <str> the output path. If not given, it is automatically defined.
        :param skip_nodata:     <bool> whether to skip all points where no valid match could be found
606
        :param skip_nodata_col: <str> determines which column of Tie_Point_Grid.CoRegPoints_table is used to
607
608
                                identify points where no valid match could be found
        """
Daniel Scheffler's avatar
Daniel Scheffler committed
609

610
611
        GDF = self.CoRegPoints_table
        GDF2pass = GDF if not skip_nodata else GDF[GDF[skip_nodata_col] != self.outFillVal]
Daniel Scheffler's avatar
Daniel Scheffler committed
612
613

        # replace boolean values (cannot be written)
614
615
        GDF2pass = GDF2pass.replace(False, 0)  # replace all booleans where column dtype is not np.bool but np.object
        GDF2pass = GDF2pass.replace(True, 1)
Daniel Scheffler's avatar
Daniel Scheffler committed
616
617
618
        for col in GDF2pass.columns:
            if GDF2pass[col].dtype == np.bool:
                GDF2pass[col] = GDF2pass[col].astype(int)
619
620
621
622
623
624

        path_out = path_out if path_out else \
            get_generic_outpath(dir_out=os.path.join(self.dir_out, 'CoRegPoints'),
                                fName_out="CoRegPoints_grid%s_ws(%s_%s)__T_%s__R_%s.shp"
                                          % (self.grid_res, self.COREG_obj.win_size_XY[0],
                                             self.COREG_obj.win_size_XY[1], self.shift.basename, self.ref.basename))
Daniel Scheffler's avatar
CoReg:    
Daniel Scheffler committed
625
        if not self.q:
626
            print('Writing %s ...' % path_out)
627
628
        GDF2pass.to_file(path_out)

629
    def _to_PointShapefile(self, skip_nodata=True, skip_nodata_col='ABS_SHIFT'):  # pragma: no cover
630
631
632
633
634
        warnings.warn(DeprecationWarning(
            "'_tiepoints_grid_to_PointShapefile' is deprecated."  # TODO delete if other method validated
            " 'tiepoints_grid_to_PointShapefile' is much faster."))
        GDF = self.CoRegPoints_table
        GDF2pass = GDF if not skip_nodata else GDF[GDF[skip_nodata_col] != self.outFillVal]
635
        shapely_points = GDF2pass['geometry'].values.tolist()
636
        attr_dicts = [collections.OrderedDict(zip(GDF2pass.columns, GDF2pass.loc[i].values)) for i in GDF2pass.index]
637

638
        fName_out = "CoRegPoints_grid%s_ws%s.shp" % (self.grid_res, self.COREG_obj.win_size_XY)
639
        path_out = os.path.join(self.dir_out, fName_out)
640
        write_shp(path_out, shapely_points, prj=self.COREG_obj.shift.prj, attrDict=attr_dicts)
641

642
    def to_vectorfield(self, path_out=None, fmt=None, mode='md'):
643
        # type: (str) -> GeoArray
644
645
646
647
648
        """Saves the calculated X-/Y-shifts to a 2-band raster file that can be used to visualize a vectorfield
        (e.g. using ArcGIS)

        :param path_out:    <str> the output path. If not given, it is automatically defined.
        :param fmt:         <str> output raster format string
649
650
        :param mode:        <str> The mode how the output is written ('uv' or 'md'; default: 'md')
                                    'uv': outputs X-/Y shifts
651
652
653
                                    'md': outputs magnitude and direction
        """

654
655
        assert mode in ['uv', 'md'], "'mode' must be either 'uv' (outputs X-/Y shifts) or 'md' " \
                                     "(outputs magnitude and direction)'. Got %s." % mode
656
657
        attr_b1 = 'X_SHIFT_M' if mode == 'uv' else 'ABS_SHIFT'
        attr_b2 = 'Y_SHIFT_M' if mode == 'uv' else 'ANGLE'
658

659
660
661
662
663
        xshift_arr, gt, prj = points_to_raster(points=self.CoRegPoints_table['geometry'],
                                               values=self.CoRegPoints_table[attr_b1],
                                               tgt_res=self.shift.xgsd * self.grid_res,
                                               prj=proj4_to_WKT(dict_to_proj4(self.CoRegPoints_table.crs)),
                                               fillVal=self.outFillVal)
664

665
666
667
668
669
        yshift_arr, gt, prj = points_to_raster(points=self.CoRegPoints_table['geometry'],
                                               values=self.CoRegPoints_table[attr_b2],
                                               tgt_res=self.shift.xgsd * self.grid_res,
                                               prj=proj4_to_WKT(dict_to_proj4(self.CoRegPoints_table.crs)),
                                               fillVal=self.outFillVal)
670
671
672

        out_GA = GeoArray(np.dstack([xshift_arr, yshift_arr]), gt, prj, nodata=self.outFillVal)

673
674
675
676
677
        path_out = path_out if path_out else \
            get_generic_outpath(dir_out=os.path.join(self.dir_out, 'CoRegPoints'),
                                fName_out="CoRegVectorfield%s_ws(%s_%s)__T_%s__R_%s.tif"
                                          % (self.grid_res, self.COREG_obj.win_size_XY[0],
                                             self.COREG_obj.win_size_XY[1], self.shift.basename, self.ref.basename))
678
679
680
681
682

        out_GA.save(path_out, fmt=fmt if fmt else 'Gtiff')

        return out_GA

683
    def to_Raster_using_Kriging(self, attrName, skip_nodata=1, skip_nodata_col='ABS_SHIFT', outGridRes=None,
684
                                fName_out=None, tilepos=None, tilesize=500, mp=None):
685

686
        mp = False if self.CPUs == 1 else True
687
688
        self._Kriging_sp(attrName, skip_nodata=skip_nodata, skip_nodata_col=skip_nodata_col,
                         outGridRes=outGridRes, fName_out=fName_out, tilepos=tilepos)
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710

        # if mp:
        #     tilepositions = UTL.get_image_tileborders([tilesize,tilesize],self.tgt_shape)
        #     args_kwargs_dicts=[]
        #     for tp in tilepositions:
        #         kwargs_dict = {'skip_nodata':skip_nodata,'skip_nodata_col':skip_nodata_col,'outGridRes':outGridRes,
        #                        'fName_out':fName_out,'tilepos':tp}
        #         args_kwargs_dicts.append({'args':[attrName],'kwargs':kwargs_dict})
        #     # self.kriged=[]
        #     # for i in args_kwargs_dicts:
        #     #     res = self.Kriging_mp(i)
        #     #     self.kriged.append(res)
        #     #     print(res)
        #
        #     with multiprocessing.Pool() as pool:
        #        self.kriged = pool.map(self.Kriging_mp,args_kwargs_dicts)
        # else:
        #     self.Kriging_sp(attrName,skip_nodata=skip_nodata,skip_nodata_col=skip_nodata_col,
        #                     outGridRes=outGridRes,fName_out=fName_out,tilepos=tilepos)
        res = self.kriged if mp else None
        return res

711
712
    def _Kriging_sp(self, attrName, skip_nodata=1, skip_nodata_col='ABS_SHIFT', outGridRes=None,
                    fName_out=None, tilepos=None):
713
714
        GDF = self.CoRegPoints_table
        GDF2pass = GDF if not skip_nodata else GDF[GDF[skip_nodata_col] != self.outFillVal]
715

716
        X_coords, Y_coords, ABS_SHIFT = GDF2pass['X_UTM'], GDF2pass['Y_UTM'], GDF2pass[attrName]
717

718
        xmin, ymin, xmax, ymax = GDF2pass.total_bounds
719

720
721
        grid_res = outGridRes if outGridRes else int(min(xmax - xmin, ymax - ymin) / 250)
        grid_x, grid_y = np.arange(xmin, xmax + grid_res, grid_res), np.arange(ymax, ymin - grid_res, -grid_res)
722
723
724

        # Reference: P.K. Kitanidis, Introduction to Geostatistcs: Applications in Hydrogeology,
        #            (Cambridge University Press, 1997) 272 p.
725
        from pykrige.ok import OrdinaryKriging
726
727
        OK = OrdinaryKriging(X_coords, Y_coords, ABS_SHIFT, variogram_model='spherical', verbose=False)
        zvalues, sigmasq = OK.execute('grid', grid_x, grid_y, backend='C', n_closest_points=12)
728

729
        if self.CPUs is None or self.CPUs > 1:
730
            fName_out = fName_out if fName_out else \
731
                "Kriging__%s__grid%s_ws%s_%s.tif" % (attrName, self.grid_res, self.COREG_obj.win_size_XY, tilepos)
732
733
        else:
            fName_out = fName_out if fName_out else \
734
735
                "Kriging__%s__grid%s_ws%s.tif" % (attrName, self.grid_res, self.COREG_obj.win_size_XY)
        path_out = get_generic_outpath(dir_out=self.dir_out, fName_out=fName_out)
736
        # add a half pixel grid points are centered on the output pixels
737
        xmin, ymin, xmax, ymax = xmin - grid_res / 2, ymin - grid_res / 2, xmax + grid_res / 2, ymax + grid_res / 2
738
739
740
741

        GeoArray(zvalues,
                 geotransform=(xmin, grid_res, 0, ymax, 0, -grid_res),
                 projection=self.COREG_obj.shift.prj).save(path_out)
742
743
744

        return zvalues

745
    def _Kriging_mp(self, args_kwargs_dict):
746
747
        args = args_kwargs_dict.get('args', [])
        kwargs = args_kwargs_dict.get('kwargs', [])
748

749
        return self._Kriging_sp(*args, **kwargs)
750
751


752
class Tie_Point_Refiner(object):
Daniel Scheffler's avatar
Daniel Scheffler committed
753
    def __init__(self, GDF, min_reliability=60, rs_max_outlier=10, rs_tolerance=2.5, rs_max_iter=15,
754
                 rs_exclude_previous_outliers=True, rs_timeout=20, q=False):
755
        """A class for performing outlier detection.
Daniel Scheffler's avatar
Daniel Scheffler committed
756

757
758
        :param GDF:                             GeoDataFrame like TiePointGrid.CoRegPoints_table containing all tie
                                                points to be filtered and the corresponding metadata
Daniel Scheffler's avatar
Daniel Scheffler committed
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
        :param min_reliability:                 <float, int> minimum threshold for previously computed tie X/Y shift
                                                reliability (default: 60%)
        :param rs_max_outlier:                  <float, int> RANSAC: maximum percentage of outliers to be detected
                                                (default: 10%)
        :param rs_tolerance:                    <float, int> RANSAC: percentage tolerance for max_outlier_percentage
                                                (default: 2.5%)
        :param rs_max_iter:                     <int> RANSAC: maximum iterations for finding the best RANSAC threshold
                                                (default: 15)
        :param rs_exclude_previous_outliers:    <bool> RANSAC: whether to exclude points that have been flagged as
                                                outlier by earlier filtering (default:True)
        :param rs_timeout:                      <float, int> RANSAC: timeout for iteration loop in seconds (default: 20)

        :param q:
        """

        self.GDF = GDF.copy()
        self.min_reliability = min_reliability
        self.rs_max_outlier_percentage = rs_max_outlier
        self.rs_tolerance = rs_tolerance
        self.rs_max_iter = rs_max_iter
        self.rs_exclude_previous_outliers = rs_exclude_previous_outliers
        self.rs_timeout = rs_timeout
        self.q = q
        self.new_cols = []
783
784
        self.ransac_model_robust = None

785
    def run_filtering(self, level=3):
786
787
        """Filter tie points used for shift correction.

788
        :param level:   tie point filter level (default: 3).
789
790
791
792
793
794
795
796
                        NOTE: lower levels are also included if a higher level is chosen
                            - Level 0: no tie point filtering
                            - Level 1: Reliablity filtering - filter all tie points out that have a low
                                reliability according to internal tests
                            - Level 2: SSIM filtering - filters all tie points out where shift
                                correction does not increase image similarity within matching window
                                (measured by mean structural similarity index)
                            - Level 3: RANSAC outlier detection
Daniel Scheffler's avatar
Daniel Scheffler committed
797
798
799
800

        :return:
        """

801
802
        # TODO catch empty GDF

803
        # RELIABILITY filtering
804
        if level > 0:
805
            marked_recs = GeoSeries(self._reliability_thresholding())
806
807
            self.GDF['L1_OUTLIER'] = marked_recs
            self.new_cols.append('L1_OUTLIER')
Daniel Scheffler's avatar
Daniel Scheffler committed
808

809
            if not self.q:
810
                print('%s tie points flagged by level 1 filtering (reliability).'
Daniel Scheffler's avatar
Daniel Scheffler committed
811
                      % (len(marked_recs[marked_recs])))
Daniel Scheffler's avatar
Daniel Scheffler committed
812

813
        # SSIM filtering
814
        if level > 1:
815
            marked_recs = GeoSeries(self._SSIM_filtering())
816
817
            self.GDF['L2_OUTLIER'] = marked_recs
            self.new_cols.append('L2_OUTLIER')
Daniel Scheffler's avatar
Daniel Scheffler committed
818

819
            if not self.q:
Daniel Scheffler's avatar
Daniel Scheffler committed
820
                print('%s tie points flagged by level 2 filtering (SSIM).' % (len(marked_recs[marked_recs])))
Daniel Scheffler's avatar
Daniel Scheffler committed
821

822
        # RANSAC filtering
823
        if level > 2:
Daniel Scheffler's avatar
Daniel Scheffler committed
824
            # exclude previous outliers
Daniel Scheffler's avatar
Daniel Scheffler committed
825
            ransacInGDF = self.GDF[~self.GDF[self.new_cols].any(axis=1)].copy() \
826
                    if self.rs_exclude_previous_outliers else self.GDF
Daniel Scheffler's avatar
Daniel Scheffler committed
827

828
            if len(ransacInGDF) > 4:
829
                # running RANSAC with less than four tie points makes no sense
Daniel Scheffler's avatar
Daniel Scheffler committed
830
831

                marked_recs = GeoSeries(self._RANSAC_outlier_detection(ransacInGDF))
832
833
                # we need to join a list here because otherwise it's merged by the 'index' column
                self.GDF['L3_OUTLIER'] = marked_recs.tolist()
Daniel Scheffler's avatar
Daniel Scheffler committed
834

835
                if not self.q:
Daniel Scheffler's avatar
Daniel Scheffler committed
836
837
                    print('%s tie points flagged by level 3 filtering (RANSAC)'
                          % (len(marked_recs[marked_recs])))
838
839
840
841
            else:
                print('RANSAC skipped because too less valid tie points have been found.')
                self.GDF['L3_OUTLIER'] = False

842
            self.new_cols.append('L3_OUTLIER')
Daniel Scheffler's avatar
Daniel Scheffler committed
843

844
845
846
847
848
        self.GDF['OUTLIER'] = self.GDF[self.new_cols].any(axis=1)
        self.new_cols.append('OUTLIER')

        return self.GDF, self.new_cols

Daniel Scheffler's avatar
Daniel Scheffler committed
849
    def _reliability_thresholding(self):
850
        """Exclude all records where estimated reliability of the calculated shifts is below the given threshold."""
851

Daniel Scheffler's avatar
Daniel Scheffler committed
852
        return self.GDF.RELIABILITY < self.min_reliability
853
854

    def _SSIM_filtering(self):
Daniel Scheffler's avatar
Daniel Scheffler committed
855
        """Exclude all records where SSIM decreased."""
856

857
        # ssim_diff  = np.median(self.GDF['SSIM_AFTER']) - np.median(self.GDF['SSIM_BEFORE'])
858

859
860
        # self.GDF.SSIM_IMPROVED = \
        #     self.GDF.apply(lambda GDF_row: GDF_row['SSIM_AFTER']>GDF_row['SSIM_BEFORE'] + ssim_diff, axis=1)
861

862
        return ~self.GDF.SSIM_IMPROVED
863

Daniel Scheffler's avatar
Daniel Scheffler committed
864
865
    def _RANSAC_outlier_detection(self, inGDF):
        """Detect geometric outliers between point cloud of source and estimated coordinates using RANSAC algorithm."""
866

Daniel Scheffler's avatar
Daniel Scheffler committed
867
        src_coords = np.array(inGDF[['X_UTM', 'Y_UTM']])
868
        xyShift = np.array(inGDF[['X_SHIFT_M', 'Y_SHIFT_M']])
869
870
871
        est_coords = src_coords + xyShift

        for co, n in zip([src_coords, est_coords], ['src_coords', 'est_coords']):
872
            assert co.ndim == 2 and co.shape[1] == 2, "'%s' must have shape [Nx2]. Got shape %s." % (n, co.shape)
873

874
875
        if not 0 < self.rs_max_outlier_percentage < 100:
            raise ValueError
876
        min_inlier_percentage = 100 - self.rs_max_outlier_percentage
877

878
        class PolyTF_1(PolynomialTransform):  # pragma: no cover
879
880
881
882
            def estimate(*data):
                return PolynomialTransform.estimate(*data, order=1)

        # robustly estimate affine transform model with RANSAC
883
        # eliminates not more than the given maximum outlier percentage of the tie points
884
885

        model_robust, inliers = None, None
886
887
888
889
890
891
892
        count_inliers = None
        th = 5  # start RANSAC threshold
        th_checked = {}  # dict of thresholds that already have been tried + calculated inlier percentage
        th_substract = 2
        count_iter = 0
        time_start = time.time()
        ideal_count = min_inlier_percentage * src_coords.shape[0] / 100
893

Daniel Scheffler's avatar