mocap/mocap_tools.py

   1 # ##### BEGIN GPL LICENSE BLOCK #####
   2 #
   3 #  This program is free software; you can redistribute it and/or
   4 #  modify it under the terms of the GNU General Public License
   5 #  as published by the Free Software Foundation; either version 2
   6 #  of the License, or (at your option) any later version.
   7 #
   8 #  This program is distributed in the hope that it will be useful,
   9 #  but WITHOUT ANY WARRANTY; without even the implied warranty of
  10 #  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11 #  GNU General Public License for more details.
  12 #
  13 #  You should have received a copy of the GNU General Public License
  14 #  along with this program; if not, write to the Free Software Foundation,
  15 #  Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
  16 #
  17 # ##### END GPL LICENSE BLOCK #####
  18
  19 from math import sqrt, radians, floor, ceil
  20 import bpy
  21 import time
  22 from mathutils import Vector, Matrix
  23
  24
  25 # A Python implementation of n sized Vectors.
  26 # Mathutils has a max size of 4, and we need at least 5 for Simplify Curves and even more for Cross Correlation.
  27 # Vector utility functions
  28 class NdVector:
  29     vec = []
  30
  31     def __init__(self, vec):
  32         self.vec = vec[:]
  33
  34     def __len__(self):
  35         return len(self.vec)
  36
  37     def __mul__(self, otherMember):
  38         # assume anything with list access is a vector
  39         if isinstance(otherMember, NdVector):
  40             a = self.vec
  41             b = otherMember.vec
  42             n = len(self)
  43             return sum([a[i] * b[i] for i in range(n)])
  44         else:
  45             # int/float
  46             return NdVector([otherMember * x for x in self.vec])
  47
  48     def __sub__(self, otherVec):
  49         a = self.vec
  50         b = otherVec.vec
  51         n = len(self)
  52         return NdVector([a[i] - b[i] for i in range(n)])
  53
  54     def __add__(self, otherVec):
  55         a = self.vec
  56         b = otherVec.vec
  57         n = len(self)
  58         return NdVector([a[i] + b[i] for i in range(n)])
  59
  60     def __div__(self, scalar):
  61         return NdVector([x / scalar for x in self.vec])
  62
  63     @property
  64     def length(self):
  65         return sqrt(self * self)
  66
  67     @property
  68     def lengthSq(self):
  69         return (self * self)
  70
  71     def normalize(self):
  72         len = self.length
  73         if len > 0:
  74             self.vec = [x / len for x in self.vec]
  75
  76     def copy(self):
  77         return NdVector(self.vec)
  78
  79     def __getitem__(self, i):
  80         return self.vec[i]
  81
  82     @property
  83     def x(self):
  84         return self.vec[0]
  85
  86     @property
  87     def y(self):
  88         return self.vec[1]
  89
  90     def resize_2d(self):
  91         return Vector((self.x, self.y))
  92
  93
  94 #Sampled Data Point class for Simplify Curves
  95 class dataPoint:
  96     index = 0
  97     # x,y1,y2,y3 coordinate of original point
  98     co = NdVector((0, 0, 0, 0, 0))
  99     #position according to parametric view of original data, [0,1] range
 100     u = 0
 101     #use this for anything
 102     temp = 0
 103
 104     def __init__(self, index, co, u=0):
 105         self.index = index
 106         self.co = co
 107         self.u = u
 108
 109
 110 # Helper to convert from a sampled fcurve back to editable keyframes one.
 111 def make_editable_fcurves(fcurves):
 112     for fc in fcurves:
 113         if fc.sampled_points:
 114             fc.convert_to_keyframes(floor(fc.sampled_points[0].co[0]), ceil(fc.sampled_points[-1].co[0]) + 1)
 115
 116
 117 #Cross Correlation Function
 118 #http://en.wikipedia.org/wiki/Cross_correlation
 119 #IN:   curvesA, curvesB - bpy_collection/list of fcurves to analyze. Auto-Correlation is when they are the same.
 120 #        margin - When searching for the best "start" frame, how large a neighborhood of frames should we inspect (similar to epsilon in Calculus)
 121 #OUT:   startFrame, length of new anim, and curvesA
 122 def crossCorrelationMatch(curvesA, curvesB, margin):
 123     dataA = []
 124     dataB = []
 125     start = int(max(curvesA[0].range()[0], curvesB[0].range()[0]))
 126     end = int(min(curvesA[0].range()[1], curvesB[0].range()[1]))
 127
 128     #transfer all fcurves data on each frame to a single NdVector.
 129     for i in range(1, end):
 130         vec = []
 131         for fcurve in curvesA:
 132             if fcurve.data_path in [otherFcurve.data_path for otherFcurve in curvesB]:
 133                 vec.append(fcurve.evaluate(i))
 134         dataA.append(NdVector(vec))
 135         vec = []
 136         for fcurve in curvesB:
 137             if fcurve.data_path in [otherFcurve.data_path for otherFcurve in curvesA]:
 138                 vec.append(fcurve.evaluate(i))
 139         dataB.append(NdVector(vec))
 140
 141     #Comparator for Cross Correlation. "Classic" implementation uses dot product, as do we.
 142     def comp(a, b):
 143         return a * b
 144
 145     #Create Rxy, which holds the Cross Correlation data.
 146     N = len(dataA)
 147     Rxy = [0.0] * N
 148     for i in range(N):
 149         for j in range(i, min(i + N, N)):
 150             Rxy[i] += comp(dataA[j], dataB[j - i])
 151         for j in range(i):
 152             Rxy[i] += comp(dataA[j], dataB[j - i + N])
 153         Rxy[i] /= float(N)
 154
 155     #Find the Local maximums in the Cross Correlation data via numerical derivative.
 156     def LocalMaximums(Rxy):
 157         Rxyd = [Rxy[i] - Rxy[i - 1] for i in range(1, len(Rxy))]
 158         maxs = []
 159         for i in range(1, len(Rxyd) - 1):
 160             a = Rxyd[i - 1]
 161             b = Rxyd[i]
 162             #sign change (zerocrossing) at point i, denoting max point (only)
 163             if (a >= 0 and b < 0) or (a < 0 and b >= 0):
 164                 maxs.append((i, max(Rxy[i], Rxy[i - 1])))
 165         return [x[0] for x in maxs]
 166         #~ return max(maxs, key=lambda x: x[1])[0]
 167
 168     #flms - the possible offsets of the first part of the animation. In Auto-Corr, this is the length of the loop.
 169     flms = LocalMaximums(Rxy[0:int(len(Rxy))])
 170     ss = []
 171
 172     #for every local maximum, find the best one - i.e. also has the best start frame.
 173     for flm in flms:
 174         diff = []
 175
 176         for i in range(len(dataA) - flm):
 177             diff.append((dataA[i] - dataB[i + flm]).lengthSq)
 178
 179         def lowerErrorSlice(diff, e):
 180             #index, error at index
 181             bestSlice = (0, 100000)
 182             for i in range(e, len(diff) - e):
 183                 errorSlice = sum(diff[i - e:i + e + 1])
 184                 if errorSlice < bestSlice[1]:
 185                     bestSlice = (i, errorSlice, flm)
 186             return bestSlice
 187
 188         s = lowerErrorSlice(diff, margin)
 189         ss.append(s)
 190
 191     #Find the best result and return it.
 192     ss.sort(key=lambda x: x[1])
 193     return ss[0][2], ss[0][0], dataA
 194
 195
 196 #Uses auto correlation (cross correlation of the same set of curves) and trims the active_object's fcurves
 197 #Except for location curves (which in mocap tend to be not cyclic, e.g. a walk cycle forward)
 198 #Transfers the fcurve data to a list of NdVector (length of list is number of fcurves), and calls the cross correlation function.
 199 #Then trims the fcurve accordingly.
 200 #IN: Nothing, set the object you want as active and call. Assumes object has animation_data.action!
 201 #OUT: Trims the object's fcurves (except location curves).
 202 def autoloop_anim():
 203     context = bpy.context
 204     obj = context.active_object
 205
 206     def locCurve(x):
 207         x.data_path == "location"
 208
 209     fcurves = [x for x in obj.animation_data.action.fcurves if not locCurve(x)]
 210
 211     margin = 10
 212
 213     flm, s, data = crossCorrelationMatch(fcurves, fcurves, margin)
 214     loop = data[s:s + flm]
 215
 216     #performs blending with a root falloff on the seam's neighborhood to ensure good tiling.
 217     for i in range(1, margin + 1):
 218         w1 = sqrt(float(i) / margin)
 219         loop[-i] = (loop[-i] * w1) + (loop[0] * (1 - w1))
 220
 221     for curve in fcurves:
 222         pts = curve.keyframe_points
 223         for i in range(len(pts) - 1, -1, -1):
 224             pts.remove(pts[i])
 225
 226     for c, curve in enumerate(fcurves):
 227         pts = curve.keyframe_points
 228         for i in range(len(loop)):
 229             pts.insert(i + 2, loop[i][c])
 230
 231     context.scene.frame_end = flm
 232
 233
 234 #simplifyCurves: performs the bulk of the samples to bezier conversion.
 235 #IN:    curveGroup - which can be a collection of singleFcurves, or grouped (via nested lists) .
 236 #         error - threshold of permittable error (max distance) of the new beziers to the original data
 237 #         reparaError - threshold of error where we should try to fix the parameterization rather than split the existing curve. > error, usually by a small constant factor for best performance.
 238 #         maxIterations - maximum number of iterations of reparameterizations we should attempt. (Newton-Rahpson is not guaranteed to converge, so this is needed).
 239 #         group_mode - boolean, indicating whether we should place bezier keyframes on the same x (frame), or optimize each individual curve.
 240 #OUT: None. Deletes the existing curves and creates the new beziers.
 241 def simplifyCurves(curveGroup, error, reparaError, maxIterations, group_mode):
 242     #Calculates the unit tangent of point v
 243     def unitTangent(v, data_pts):
 244         tang = NdVector((0, 0, 0, 0, 0))
 245         if v != 0:
 246             #If it's not the first point, we can calculate a leftside tangent
 247             tang += data_pts[v].co - data_pts[v - 1].co
 248         if v != len(data_pts) - 1:
 249             #If it's not the last point, we can calculate a rightside tangent
 250             tang += data_pts[v + 1].co - data_pts[v].co
 251         tang.normalize()
 252         return tang
 253
 254     #assign parametric u value for each point in original data, via relative arc length
 255     #http://en.wikipedia.org/wiki/Arc_length
 256     def chordLength(data_pts, s, e):
 257         totalLength = 0
 258         for pt in data_pts[s:e + 1]:
 259             i = pt.index
 260             if i == s:
 261                 chordLength = 0
 262             else:
 263                 chordLength = (data_pts[i].co - data_pts[i - 1].co).length
 264             totalLength += chordLength
 265             pt.temp = totalLength
 266         for pt in data_pts[s:e + 1]:
 267             if totalLength == 0:
 268                 print(s, e)
 269             pt.u = (pt.temp / totalLength)
 270
 271     # get binomial coefficient lookup table, this function/table is only called with args
 272     # (3,0),(3,1),(3,2),(3,3),(2,0),(2,1),(2,2)!
 273     binomDict = {(3, 0): 1,
 274                  (3, 1): 3,
 275                  (3, 2): 3,
 276                  (3, 3): 1,
 277                  (2, 0): 1,
 278                  (2, 1): 2,
 279                  (2, 2): 1,
 280                  }
 281
 282     #value at pt t of a single bernstein Polynomial
 283     def bernsteinPoly(n, i, t):
 284         binomCoeff = binomDict[(n, i)]
 285         return binomCoeff * pow(t, i) * pow(1 - t, n - i)
 286
 287     # fit a single cubic to data points in range [s(tart),e(nd)].
 288     def fitSingleCubic(data_pts, s, e):
 289
 290         # A - matrix used for calculating C matrices for fitting
 291         def A(i, j, s, e, t1, t2):
 292             if j == 1:
 293                 t = t1
 294             if j == 2:
 295                 t = t2
 296             u = data_pts[i].u
 297             return t * bernsteinPoly(3, j, u)
 298
 299         # X component, used for calculating X matrices for fitting
 300         def xComponent(i, s, e):
 301             di = data_pts[i].co
 302             u = data_pts[i].u
 303             v0 = data_pts[s].co
 304             v3 = data_pts[e].co
 305             a = v0 * bernsteinPoly(3, 0, u)
 306             b = v0 * bernsteinPoly(3, 1, u)
 307             c = v3 * bernsteinPoly(3, 2, u)
 308             d = v3 * bernsteinPoly(3, 3, u)
 309             return (di - (a + b + c + d))
 310
 311         t1 = unitTangent(s, data_pts)
 312         t2 = unitTangent(e, data_pts)
 313         c11 = sum([A(i, 1, s, e, t1, t2) * A(i, 1, s, e, t1, t2) for i in range(s, e + 1)])
 314         c12 = sum([A(i, 1, s, e, t1, t2) * A(i, 2, s, e, t1, t2) for i in range(s, e + 1)])
 315         c21 = c12
 316         c22 = sum([A(i, 2, s, e, t1, t2) * A(i, 2, s, e, t1, t2) for i in range(s, e + 1)])
 317
 318         x1 = sum([xComponent(i, s, e) * A(i, 1, s, e, t1, t2) for i in range(s, e + 1)])
 319         x2 = sum([xComponent(i, s, e) * A(i, 2, s, e, t1, t2) for i in range(s, e + 1)])
 320
 321         # calculate Determinate of the 3 matrices
 322         det_cc = c11 * c22 - c21 * c12
 323         det_cx = c11 * x2 - c12 * x1
 324         det_xc = x1 * c22 - x2 * c12
 325
 326         # if matrix is not homogenous, fudge the data a bit
 327         if det_cc == 0:
 328             det_cc = 0.01
 329
 330         # alpha's are the correct offset for bezier handles
 331         alpha0 = det_xc / det_cc   # offset from right (first) point
 332         alpha1 = det_cx / det_cc   # offset from left (last) point
 333
 334         sRightHandle = data_pts[s].co.copy()
 335         sTangent = t1 * abs(alpha0)
 336         sRightHandle += sTangent  # position of first pt's handle
 337         eLeftHandle = data_pts[e].co.copy()
 338         eTangent = t2 * abs(alpha1)
 339         eLeftHandle += eTangent  # position of last pt's handle.
 340
 341         # return a 4 member tuple representing the bezier
 342         return (data_pts[s].co,
 343               sRightHandle,
 344               eLeftHandle,
 345               data_pts[e].co)
 346
 347     # convert 2 given data points into a cubic bezier.
 348     # handles are offset along the tangent at
 349     # a 3rd of the length between the points.
 350     def fitSingleCubic2Pts(data_pts, s, e):
 351         alpha0 = alpha1 = (data_pts[s].co - data_pts[e].co).length / 3
 352
 353         sRightHandle = data_pts[s].co.copy()
 354         sTangent = unitTangent(s, data_pts) * abs(alpha0)
 355         sRightHandle += sTangent  # position of first pt's handle
 356         eLeftHandle = data_pts[e].co.copy()
 357         eTangent = unitTangent(e, data_pts) * abs(alpha1)
 358         eLeftHandle += eTangent  # position of last pt's handle.
 359
 360         #return a 4 member tuple representing the bezier
 361         return (data_pts[s].co,
 362           sRightHandle,
 363           eLeftHandle,
 364           data_pts[e].co)
 365
 366     #evaluate bezier, represented by a 4 member tuple (pts) at point t.
 367     def bezierEval(pts, t):
 368         sumVec = NdVector((0, 0, 0, 0, 0))
 369         for i in range(4):
 370             sumVec += pts[i] * bernsteinPoly(3, i, t)
 371         return sumVec
 372
 373     #calculate the highest error between bezier and original data
 374     #returns the distance and the index of the point where max error occurs.
 375     def maxErrorAmount(data_pts, bez, s, e):
 376         maxError = 0
 377         maxErrorPt = s
 378         if e - s < 3:
 379             return 0, None
 380         for pt in data_pts[s:e + 1]:
 381             bezVal = bezierEval(bez, pt.u)
 382             normalize_error = pt.co.length
 383             if normalize_error == 0:
 384                 normalize_error = 1
 385             tmpError = (pt.co - bezVal).length / normalize_error
 386             if tmpError >= maxError:
 387                 maxError = tmpError
 388                 maxErrorPt = pt.index
 389         return maxError, maxErrorPt
 390
 391     #calculated bezier derivative at point t.
 392     #That is, tangent of point t.
 393     def getBezDerivative(bez, t):
 394         n = len(bez) - 1
 395         sumVec = NdVector((0, 0, 0, 0, 0))
 396         for i in range(n - 1):
 397             sumVec += (bez[i + 1] - bez[i]) * bernsteinPoly(n - 1, i, t)
 398         return sumVec
 399
 400     #use Newton-Raphson to find a better parameterization of datapoints,
 401     #one that minimizes the distance (or error)
 402     # between bezier and original data.
 403     def newtonRaphson(data_pts, s, e, bez):
 404         for pt in data_pts[s:e + 1]:
 405             if pt.index == s:
 406                 pt.u = 0
 407             elif pt.index == e:
 408                 pt.u = 1
 409             else:
 410                 u = pt.u
 411                 qu = bezierEval(bez, pt.u)
 412                 qud = getBezDerivative(bez, u)
 413                 #we wish to minimize f(u),
 414                 #the squared distance between curve and data
 415                 fu = (qu - pt.co).length ** 2
 416                 fud = (2 * (qu.x - pt.co.x) * (qud.x)) - (2 * (qu.y - pt.co.y) * (qud.y))
 417                 if fud == 0:
 418                     fu = 0
 419                     fud = 1
 420                 pt.u = pt.u - (fu / fud)
 421
 422     #Create data_pts, a list of dataPoint type, each is assigned index i, and an NdVector
 423     def createDataPts(curveGroup, group_mode):
 424         make_editable_fcurves(curveGroup if group_mode else (curveGroup,))
 425
 426         if group_mode:
 427             print([x.data_path for x in curveGroup])
 428             comp_cos = (0,) * (4 - len(curveGroup))  # We need to add that number of null cos to get our 5D vector.
 429             kframes = sorted(set(kf.co.x for fc in curveGroup for kf in fc.keyframe_points))
 430             data_pts = [dataPoint(i, NdVector((fra,) + tuple(fc.evaluate(fra) for fc in curveGroup) + comp_cos))
 431                         for i, fra in enumerate(kframes)]
 432         else:
 433             data_pts = [dataPoint(i, NdVector((kf.co.x, kf.co.y, 0, 0, 0)))
 434                         for i, kf in enumerate(curveGroup.keyframe_points)]
 435         return data_pts
 436
 437     #Recursively fit cubic beziers to the data_pts between s and e
 438     def fitCubic(data_pts, s, e):
 439         # if there are less than 3 points, fit a single basic bezier
 440         if e - s < 3:
 441             bez = fitSingleCubic2Pts(data_pts, s, e)
 442         else:
 443             #if there are more, parameterize the points
 444             # and fit a single cubic bezier
 445             chordLength(data_pts, s, e)
 446             bez = fitSingleCubic(data_pts, s, e)
 447
 448         #calculate max error and point where it occurs
 449         maxError, maxErrorPt = maxErrorAmount(data_pts, bez, s, e)
 450         #if error is small enough, reparameterization might be enough
 451         if maxError < reparaError and maxError > error:
 452             for i in range(maxIterations):
 453                 newtonRaphson(data_pts, s, e, bez)
 454                 if e - s < 3:
 455                     bez = fitSingleCubic2Pts(data_pts, s, e)
 456                 else:
 457                     bez = fitSingleCubic(data_pts, s, e)
 458
 459             #recalculate max error and point where it occurs
 460             maxError, maxErrorPt = maxErrorAmount(data_pts, bez, s, e)
 461
 462         #repara wasn't enough, we need 2 beziers for this range.
 463         #Split the bezier at point of maximum error
 464         if maxError > error:
 465             fitCubic(data_pts, s, maxErrorPt)
 466             fitCubic(data_pts, maxErrorPt, e)
 467         else:
 468             #error is small enough, return the beziers.
 469             beziers.append(bez)
 470             return
 471
 472     # deletes the sampled points and creates beziers.
 473     def createNewCurves(curveGroup, beziers, group_mode):
 474         #remove all existing data points
 475         if group_mode:
 476             for fcurve in curveGroup:
 477                 for i in range(len(fcurve.keyframe_points) - 1, 0, -1):
 478                     fcurve.keyframe_points.remove(fcurve.keyframe_points[i], fast=True)
 479         else:
 480             fcurve = curveGroup
 481             for i in range(len(fcurve.keyframe_points) - 1, 0, -1):
 482                 fcurve.keyframe_points.remove(fcurve.keyframe_points[i], fast=True)
 483
 484         #insert the calculated beziers to blender data.
 485         if group_mode:
 486             for fullbez in beziers:
 487                 for i, fcurve in enumerate(curveGroup):
 488                     bez = [Vector((vec[0], vec[i + 1])) for vec in fullbez]
 489                     newKey = fcurve.keyframe_points.insert(frame=bez[0].x, value=bez[0].y, options={'FAST'})
 490                     newKey.handle_right = (bez[1].x, bez[1].y)
 491
 492                     newKey = fcurve.keyframe_points.insert(frame=bez[3].x, value=bez[3].y, options={'FAST'})
 493                     newKey.handle_left = (bez[2].x, bez[2].y)
 494         else:
 495             for bez in beziers:
 496                 for vec in bez:
 497                     vec.resize_2d()
 498                 newKey = fcurve.keyframe_points.insert(frame=bez[0].x, value=bez[0].y, options={'FAST'})
 499                 newKey.handle_right = (bez[1].x, bez[1].y)
 500
 501                 newKey = fcurve.keyframe_points.insert(frame=bez[3].x, value=bez[3].y, options={'FAST'})
 502                 newKey.handle_left = (bez[2].x, bez[2].y)
 503
 504         # We used fast remove/insert, time to update the curves!
 505         for fcurve in (curveGroup if group_mode else (curveGroup,)):
 506             fcurve.update()
 507
 508     # indices are detached from data point's frame (x) value and
 509     # stored in the dataPoint object, represent a range
 510
 511     data_pts = createDataPts(curveGroup, group_mode)
 512
 513     if not data_pts:
 514         return
 515
 516     s = 0  # start
 517     e = len(data_pts) - 1  # end
 518
 519     beziers = []
 520
 521     #begin the recursive fitting algorithm.
 522     fitCubic(data_pts, s, e)
 523
 524     #remove old Fcurves and insert the new ones
 525     createNewCurves(curveGroup, beziers, group_mode)
 526
 527
 528 #Main function of simplification, which called by Operator
 529 #IN:
 530 #       sel_opt- either "sel" (selected) or "all" for which curves to effect
 531 #       error- maximum error allowed, in fraction (20% = 0.0020, which is the default),
 532 #       i.e. divide by 10000 from percentage wanted.
 533 #       group_mode- boolean, to analyze each curve separately or in groups,
 534 #       where a group is all curves that effect the same property/RNA path
 535 def fcurves_simplify(context, obj, sel_opt="all", error=0.002, group_mode=True):
 536     # main vars
 537     fcurves = obj.animation_data.action.fcurves
 538
 539     if sel_opt == "sel":
 540         sel_fcurves = [fcurve for fcurve in fcurves if fcurve.select]
 541     else:
 542         sel_fcurves = fcurves[:]
 543
 544     #Error threshold for Newton Raphson reparamatizing
 545     reparaError = error * 32
 546     maxIterations = 16
 547
 548     if group_mode:
 549         fcurveDict = {}
 550         #this loop sorts all the fcurves into groups of 3 or 4,
 551         #based on their RNA Data path, which corresponds to
 552         #which property they effect
 553         for curve in sel_fcurves:
 554             if curve.data_path in fcurveDict:  # if this bone has been added, append the curve to its list
 555                 fcurveDict[curve.data_path].append(curve)
 556             else:
 557                 fcurveDict[curve.data_path] = [curve]  # new bone, add a new dict value with this first curve
 558         fcurveGroups = fcurveDict.values()
 559     else:
 560         fcurveGroups = sel_fcurves
 561
 562     if error > 0.00000:
 563         #simplify every selected curve.
 564         totalt = 0
 565         for i, fcurveGroup in enumerate(fcurveGroups):
 566             print("Processing curve " + str(i + 1) + "/" + str(len(fcurveGroups)))
 567             t = time.clock()
 568             simplifyCurves(fcurveGroup, error, reparaError, maxIterations, group_mode)
 569             t = time.clock() - t
 570             print(str(t)[:5] + " seconds to process last curve")
 571             totalt += t
 572             print(str(totalt)[:5] + " seconds, total time elapsed")
 573
 574     return
 575
 576
 577 def detect_min_max(v):
 578     """
 579     Converted from MATLAB script at http://billauer.co.il/peakdet.html
 580
 581     Yields indices of peaks, i.e. local minima/maxima.
 582
 583     % Eli Billauer, 3.4.05 (Explicitly not copyrighted).
 584     % This function is released to the public domain; Any use is allowed.
 585     """
 586
 587     min_val, max_val = float('inf'), -float('inf')
 588
 589     check_max = True
 590
 591     for i, val in enumerate(v):
 592         if val > max_val:
 593             max_val = val
 594         if val < min_val:
 595             min_val = val
 596
 597         if check_max:
 598             if val < max_val:
 599                 yield i
 600                 min_val = val
 601                 check_max = False
 602         else:
 603             if val > min_val:
 604                 yield i
 605                 max_val = val
 606                 check_max = True
 607
 608
 609 def denoise(obj, fcurves):
 610     """
 611     Implementation of non-linear blur filter.
 612     Finds spikes in the fcurve, and replaces spikes that are too big with the average of the surrounding keyframes.
 613     """
 614     make_editable_fcurves(fcurves)
 615
 616     for fcurve in fcurves:
 617         org_pts = fcurve.keyframe_points[:]
 618
 619         for idx in detect_min_max(pt.co.y for pt in fcurve.keyframe_points[1:-1]):
 620             # Find the neighbours
 621             prev_pt = org_pts[idx - 1].co.y
 622             next_pt = org_pts[idx + 1].co.y
 623             this_pt = org_pts[idx]
 624
 625             # Check the distance from the min/max to the average of the surrounding points.
 626             avg = (prev_pt + next_pt) / 2
 627             is_peak = abs(this_pt.co.y - avg) > avg * 0.02
 628
 629             if is_peak:
 630                 diff = avg - fcurve.keyframe_points[idx].co.y
 631                 fcurve.keyframe_points[idx].co.y = avg
 632                 fcurve.keyframe_points[idx].handle_left.y += diff
 633                 fcurve.keyframe_points[idx].handle_right.y += diff
 634
 635         # Important to update the curve after modifying it!
 636         fcurve.update()
 637
 638
 639 # Receives armature, and rotations all bones by 90 degrees along the X axis
 640 # This fixes the common axis issue BVH files have when importing.
 641 # IN: Armature (bpy.types.Armature)
 642 def rotate_fix_armature(arm_data):
 643     global_matrix = Matrix.Rotation(radians(90), 4, "X")
 644     bpy.ops.object.mode_set(mode='EDIT', toggle=False)
 645     #disconnect all bones for ease of global rotation
 646     connectedBones = []
 647     for bone in arm_data.edit_bones:
 648         if bone.use_connect:
 649             connectedBones.append(bone.name)
 650             bone.use_connect = False
 651
 652     #rotate all the bones around their center
 653     for bone in arm_data.edit_bones:
 654         bone.transform(global_matrix)
 655
 656     #reconnect the bones
 657     for bone in connectedBones:
 658         arm_data.edit_bones[bone].use_connect = True
 659     bpy.ops.object.mode_set(mode='OBJECT', toggle=False)
 660
 661
 662 #Roughly scales the performer armature to match the enduser armature
 663 #IN: perfromer_obj, enduser_obj, Blender objects whose .data is an armature.
 664 def scale_fix_armature(performer_obj, enduser_obj):
 665     perf_bones = performer_obj.data.bones
 666     end_bones = enduser_obj.data.bones
 667
 668     def calculateBoundingRadius(bones):
 669         # Calculate the average position of each bone
 670         center = sum((bone.head_local for bone in bones), Vector())
 671         center /= len(bones)
 672
 673         # The radius is defined as the max distance from the center.
 674         radius = max((bone.head_local - center).length for bone in bones)
 675         return radius
 676
 677     perf_rad = calculateBoundingRadius(performer_obj.data.bones)
 678     end_rad = calculateBoundingRadius(enduser_obj.data.bones)
 679
 680     factor = end_rad / perf_rad
 681     performer_obj.scale = factor * Vector((1, 1, 1))
 682
 683
 684 #Guess Mapping
 685 #Given a performer and enduser armature, attempts to guess the hierarchy mapping
 686 def guessMapping(performer_obj, enduser_obj):
 687     perf_bones = performer_obj.data.bones
 688     end_bones = enduser_obj.data.bones
 689
 690     root = perf_bones[0]
 691
 692     def findBoneSide(bone):
 693         if "Left" in bone:
 694             return "Left", bone.replace("Left", "").lower().replace(".", "")
 695         if "Right" in bone:
 696             return "Right", bone.replace("Right", "").lower().replace(".", "")
 697         if "L" in bone:
 698             return "Left", bone.replace("Left", "").lower().replace(".", "")
 699         if "R" in bone:
 700             return "Right", bone.replace("Right", "").lower().replace(".", "")
 701         return "", bone
 702
 703     def nameMatch(bone_a, bone_b):
 704         # nameMatch - receives two strings, returns 2 if they are relatively the same, 1 if they are the same but R and L and 0 if no match at all
 705         side_a, noside_a = findBoneSide(bone_a)
 706         side_b, noside_b = findBoneSide(bone_b)
 707         if side_a == side_b:
 708             if noside_a in noside_b or noside_b in noside_a:
 709                 return 2
 710         else:
 711             if noside_a in noside_b or noside_b in noside_a:
 712                 return 1
 713         return 0
 714
 715     def guessSingleMapping(perf_bone):
 716         possible_bones = [end_bones[0]]
 717
 718         while possible_bones:
 719             for end_bone in possible_bones:
 720                 match = nameMatch(perf_bone.name, end_bone.name)
 721                 if match == 2 and not perf_bone.map:
 722                     perf_bone.map = end_bone.name
 723                 #~ elif match == 1 and not perf_bone.map:
 724                     #~ oppo = perf_bones[oppositeBone(perf_bone)].map
 725                     # if oppo:
 726                     #   perf_bone = oppo
 727             newPossibleBones = []
 728             for end_bone in possible_bones:
 729                 newPossibleBones += list(end_bone.children)
 730             possible_bones = newPossibleBones
 731
 732         for child in perf_bone.children:
 733             guessSingleMapping(child)
 734
 735     guessSingleMapping(root)
 736
 737
 738 # Creates limit rotation constraints on the enduser armature based on range of motion (max min of fcurves) of the performer.
 739 # IN: context (bpy.context, etc.), and 2 blender objects which are armatures
 740 # OUT: creates the limit constraints.
 741 def limit_dof(context, performer_obj, enduser_obj):
 742     limitDict = {}
 743     perf_bones = [bone for bone in performer_obj.pose.bones if bone.bone.map]
 744     c_frame = context.scene.frame_current
 745     for bone in perf_bones:
 746         limitDict[bone.bone.map] = [1000, 1000, 1000, -1000, -1000, -1000]
 747     for t in range(context.scene.frame_start, context.scene.frame_end):
 748         context.scene.frame_set(t)
 749         for bone in perf_bones:
 750             end_bone = enduser_obj.pose.bones[bone.bone.map]
 751             bake_matrix = bone.matrix
 752             rest_matrix = end_bone.bone.matrix_local
 753
 754             if end_bone.parent and end_bone.bone.use_inherit_rotation:
 755                 srcParent = bone.parent
 756                 parent_mat = srcParent.matrix
 757                 parent_rest = end_bone.parent.bone.matrix_local
 758                 parent_rest_inv = parent_rest.inverted()
 759                 parent_mat_inv = parent_mat.inverted()
 760                 bake_matrix = parent_mat_inv @ bake_matrix
 761                 rest_matrix = parent_rest_inv @ rest_matrix
 762
 763             rest_matrix_inv = rest_matrix.inverted()
 764             bake_matrix = rest_matrix_inv @ bake_matrix
 765
 766             mat = bake_matrix
 767             euler = mat.to_euler()
 768             limitDict[bone.bone.map][0] = min(limitDict[bone.bone.map][0], euler.x)
 769             limitDict[bone.bone.map][1] = min(limitDict[bone.bone.map][1], euler.y)
 770             limitDict[bone.bone.map][2] = min(limitDict[bone.bone.map][2], euler.z)
 771             limitDict[bone.bone.map][3] = max(limitDict[bone.bone.map][3], euler.x)
 772             limitDict[bone.bone.map][4] = max(limitDict[bone.bone.map][4], euler.y)
 773             limitDict[bone.bone.map][5] = max(limitDict[bone.bone.map][5], euler.z)
 774     for bone in enduser_obj.pose.bones:
 775         existingConstraint = [constraint for constraint in bone.constraints if constraint.name == "DOF Limitation"]
 776         if existingConstraint:
 777             bone.constraints.remove(existingConstraint[0])
 778     end_bones = [bone for bone in enduser_obj.pose.bones if bone.name in limitDict.keys()]
 779     for bone in end_bones:
 780         #~ if not bone.is_in_ik_chain:
 781         newCons = bone.constraints.new("LIMIT_ROTATION")
 782         newCons.name = "DOF Limitation"
 783         newCons.owner_space = "LOCAL"
 784         newCons.min_x, newCons.min_y, newCons.min_z, newCons.max_x, newCons.max_y, newCons.max_z = limitDict[bone.name]
 785         newCons.use_limit_x = True
 786         newCons.use_limit_y = True
 787         newCons.use_limit_z = True
 788     context.scene.frame_set(c_frame)
 789
 790
 791 # Removes the constraints that were added by limit_dof on the enduser_obj
 792 def limit_dof_toggle_off(context, enduser_obj):
 793     for bone in enduser_obj.pose.bones:
 794         existingConstraint = [constraint for constraint in bone.constraints if constraint.name == "DOF Limitation"]
 795         if existingConstraint:
 796             bone.constraints.remove(existingConstraint[0])
 797
 798
 799 # Reparameterizes a blender path via keyframing it's eval_time to match a stride_object's forward velocity.
 800 # IN: Context, stride object (blender object with location keyframes), path object.
 801 def path_editing(context, stride_obj, path):
 802     y_fcurve = [fcurve for fcurve in stride_obj.animation_data.action.fcurves if fcurve.data_path == "location"][1]
 803     s, e = context.scene.frame_start, context.scene.frame_end  # y_fcurve.range()
 804     s = int(s)
 805     e = int(e)
 806     y_s = y_fcurve.evaluate(s)
 807     y_e = y_fcurve.evaluate(e)
 808     direction = (y_e - y_s) / abs(y_e - y_s)
 809     existing_cons = [constraint for constraint in stride_obj.constraints if constraint.type == "FOLLOW_PATH"]
 810     for cons in existing_cons:
 811         stride_obj.constraints.remove(cons)
 812     path_cons = stride_obj.constraints.new("FOLLOW_PATH")
 813     if direction < 0:
 814         path_cons.forward_axis = "TRACK_NEGATIVE_Y"
 815     else:
 816         path_cons.forward_axis = "FORWARD_Y"
 817     path_cons.target = path
 818     path_cons.use_curve_follow = True
 819     path.data.path_duration = e - s
 820     try:
 821         path.data.animation_data.action.fcurves
 822     except AttributeError:
 823         path.data.keyframe_insert("eval_time", frame=0)
 824     eval_time_fcurve = [fcurve for fcurve in path.data.animation_data.action.fcurves if fcurve.data_path == "eval_time"]
 825     eval_time_fcurve = eval_time_fcurve[0]
 826     totalLength = 0
 827     parameterization = {}
 828     print("evaluating curve")
 829     for t in range(s, e - 1):
 830         if s == t:
 831             chordLength = 0
 832         else:
 833             chordLength = (y_fcurve.evaluate(t) - y_fcurve.evaluate(t + 1))
 834         totalLength += chordLength
 835         parameterization[t] = totalLength
 836     for t in range(s + 1, e - 1):
 837         if totalLength == 0:
 838             print("no forward motion")
 839         parameterization[t] /= totalLength
 840         parameterization[t] *= e - s
 841     parameterization[e] = e - s
 842     for t in parameterization.keys():
 843         eval_time_fcurve.keyframe_points.insert(frame=t, value=parameterization[t])
 844     y_fcurve.mute = True
 845     print("finished path editing")
 846
 847
 848 #Animation Stitching
 849 #Stitches two retargeted animations together via NLA settings.
 850 #IN: enduser_obj, a blender armature that has had two retargets applied.
 851 def anim_stitch(context, enduser_obj):
 852     stitch_settings = enduser_obj.data.stitch_settings
 853     action_1 = stitch_settings.first_action
 854     action_2 = stitch_settings.second_action
 855     if stitch_settings.stick_bone != "":
 856         selected_bone = enduser_obj.pose.bones[stitch_settings.stick_bone]
 857     else:
 858         selected_bone = enduser_obj.pose.bones[0]
 859     scene = context.scene
 860     TrackNamesA = enduser_obj.data.mocapNLATracks[action_1]
 861     TrackNamesB = enduser_obj.data.mocapNLATracks[action_2]
 862     enduser_obj.data.active_mocap = action_1
 863     anim_data = enduser_obj.animation_data
 864     # add tracks for action 2
 865     mocapAction = bpy.data.actions[TrackNamesB.base_track]
 866     mocapTrack = anim_data.nla_tracks.new()
 867     mocapTrack.name = TrackNamesB.base_track
 868     mocapStrip = mocapTrack.strips.new(TrackNamesB.base_track, stitch_settings.blend_frame, mocapAction)
 869     mocapStrip.extrapolation = "HOLD_FORWARD"
 870     mocapStrip.blend_in = stitch_settings.blend_amount
 871     mocapStrip.action_frame_start += stitch_settings.second_offset
 872     mocapStrip.action_frame_end += stitch_settings.second_offset
 873     constraintTrack = anim_data.nla_tracks.new()
 874     constraintTrack.name = TrackNamesB.auto_fix_track
 875     constraintAction = bpy.data.actions[TrackNamesB.auto_fix_track]
 876     constraintStrip = constraintTrack.strips.new(TrackNamesB.auto_fix_track, stitch_settings.blend_frame, constraintAction)
 877     constraintStrip.extrapolation = "HOLD_FORWARD"
 878     constraintStrip.blend_in = stitch_settings.blend_amount
 879     userTrack = anim_data.nla_tracks.new()
 880     userTrack.name = TrackNamesB.manual_fix_track
 881     userAction = bpy.data.actions[TrackNamesB.manual_fix_track]
 882     userStrip = userTrack.strips.new(TrackNamesB.manual_fix_track, stitch_settings.blend_frame, userAction)
 883     userStrip.extrapolation = "HOLD_FORWARD"
 884     userStrip.blend_in = stitch_settings.blend_amount
 885     #stride bone
 886     if enduser_obj.parent:
 887         if enduser_obj.parent.name == "stride_bone":
 888             stride_bone = enduser_obj.parent
 889             stride_anim_data = stride_bone.animation_data
 890             stride_anim_data.use_nla = True
 891             stride_anim_data.action = None
 892             for track in stride_anim_data.nla_tracks:
 893                 stride_anim_data.nla_tracks.remove(track)
 894             actionATrack = stride_anim_data.nla_tracks.new()
 895             actionATrack.name = TrackNamesA.stride_action
 896             actionAStrip = actionATrack.strips.new(TrackNamesA.stride_action, 0, bpy.data.actions[TrackNamesA.stride_action])
 897             actionAStrip.extrapolation = "NOTHING"
 898             actionBTrack = stride_anim_data.nla_tracks.new()
 899             actionBTrack.name = TrackNamesB.stride_action
 900             actionBStrip = actionBTrack.strips.new(TrackNamesB.stride_action, stitch_settings.blend_frame, bpy.data.actions[TrackNamesB.stride_action])
 901             actionBStrip.action_frame_start += stitch_settings.second_offset
 902             actionBStrip.action_frame_end += stitch_settings.second_offset
 903             actionBStrip.extrapolation = "NOTHING"
 904             #we need to change the stride_bone's action to add the offset
 905             aStrideCurves = [fcurve for fcurve in bpy.data.actions[TrackNamesA.stride_action].fcurves if fcurve.data_path == "location"]
 906             bStrideCurves = [fcurve for fcurve in bpy.data.actions[TrackNamesB.stride_action].fcurves if fcurve.data_path == "location"]
 907             scene.frame_set(stitch_settings.blend_frame - 1)
 908             desired_pos = (enduser_obj.matrix_world * selected_bone.matrix.to_translation())
 909             scene.frame_set(stitch_settings.blend_frame)
 910             actual_pos = (enduser_obj.matrix_world * selected_bone.matrix.to_translation())
 911             print(desired_pos, actual_pos)
 912             offset = Vector(actual_pos) - Vector(desired_pos)
 913
 914             for i, fcurve in enumerate(bStrideCurves):
 915                 print(offset[i], i, fcurve.array_index)
 916                 for pt in fcurve.keyframe_points:
 917                     pt.co.y -= offset[i]
 918                     pt.handle_left.y -= offset[i]
 919                     pt.handle_right.y -= offset[i]
 920
 921             #actionBStrip.blend_in = stitch_settings.blend_amount
 922
 923
 924 #Guesses setting for animation stitching via Cross Correlation
 925 def guess_anim_stitch(context, enduser_obj):
 926     stitch_settings = enduser_obj.data.stitch_settings
 927     action_1 = stitch_settings.first_action
 928     action_2 = stitch_settings.second_action
 929     TrackNamesA = enduser_obj.data.mocapNLATracks[action_1]
 930     TrackNamesB = enduser_obj.data.mocapNLATracks[action_2]
 931     mocapA = bpy.data.actions[TrackNamesA.base_track]
 932     mocapB = bpy.data.actions[TrackNamesB.base_track]
 933     curvesA = mocapA.fcurves
 934     curvesB = mocapB.fcurves
 935     flm, s, data = crossCorrelationMatch(curvesA, curvesB, 10)
 936     print("Guessed the following for start and offset: ", s, flm)
 937     enduser_obj.data.stitch_settings.blend_frame = flm
 938     enduser_obj.data.stitch_settings.second_offset = s