File size: 9,214 Bytes
e9fe176
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
require 'csvigo'
require 'image'
local pl = require 'pl.import_into' ()
cv = require 'cv'
require 'cv.imgproc'

local utils = {}

function utils.transform(transformation, points)
   if points:dim() == 1 then
      points = points:repeatTensor(1, 1)
   end
   points = torch.cat(points, torch.ones(points:size(1)), 2)
   local newPoints = (transformation * points:transpose(1, 2)):transpose(1, 2)
   newPoints = torch.cdiv(newPoints[{{}, {1, 3}}], newPoints[{{}, {4}}]:expand(newPoints:size(1), 3))
   return newPoints
end

function utils.project(transformation, points)
   if points:dim() == 1 then
      points = points:repeatTensor(1, 1)
   end
   points = torch.cat(points, torch.ones(points:size(1)), 2)
   local points2D = (transformation * points:transpose(1, 2)):transpose(1, 2)
   points2D = torch.cdiv(points2D[{{}, {1, 2}}], points2D[{{}, {3}}]:expand(points2D:size(1), 2))
   return points2D:squeeze()
end

function utils.unproject(transformation, points2D, additionalConstraint)
   if points2D:dim() == 1 then
      points2D = points2D:repeatTensor(1, 1)
   end
   points2D = torch.cat(points2D, torch.ones(points2D:size(1)), 2)
   local A = torch.cat(transformation, additionalConstraint, 1)
   local b = torch.cat(points2D, torch.zeros(points2D:size(1), additionalConstraint:size(1)), 2)

   local points = (torch.inverse(A) * b:transpose(1, 2)):transpose(1, 2)
   points = torch.cdiv(points[{{}, {1, 3}}], points[{{}, {4}}]:expand(points:size(1), 3))
   return points:squeeze()
end

-- function utils.fitPlanes(points, numPlanes = 100, numIterations = 20)
--    local numPoints = points:size(1)
--    local numSampledPoints = 10000
--    local sampledPoints = points:index(1, perm:narrow(1, 1, numSampledPoints):long())

--    local perm = torch.randperm(numSampledPoints)
--    local indices = perm:narrow(1, 1, numIterations * 3):long()
--    local selectedPoints = sampledPoints:index(1, indices):reshape(numIterations, 3, 3)
--    local vectors_1 = selectedPoints[{{}, 1, {}}] - selectedPoints[{{}, 2, {}}]
--    local vectors_2 = selectedPoints[{{}, 1, {}}] - selectedPoints[{{}, 3, {}}]
--    local normals = torch.zeros(vectors_1:size())
--    normals[{{}, 1}] = vectors_1[{{}, 2}] * vectors_2[{{}, 3}] - vectors_1[{{}, 3}] * vectors_2[{{}, 2}]
--    normals[{{}, 2}] = vectors_1[{{}, 3}] * vectors_2[{{}, 1}] - vectors_1[{{}, 1}] * vectors_2[{{}, 3}]
--    normals[{{}, 3}] = vectors_1[{{}, 1}] * vectors_2[{{}, 2}] - vectors_1[{{}, 2}] * vectors_2[{{}, 1}]
--    local mean = torch.mean(selectedPoints, 2):squeeze()
--    local planeD = torch.sum(torch.cmul(normals, mean), 2)
--    local planes = normals * planeD:repeatTensor(1, 3)
-- end

function utils.projectLines(transformation, lines)
   local lines2D
   for lineIndex = 1, lines:size(1) do
      local line = lines[lineIndex]
      local points = torch.cat(line, torch.ones(2), 2)
      local points2D = (transformation * points:transpose(1, 2)):transpose(1, 2)
      points2D[{{}, 1}] = torch.cdiv(points2D[{{}, 1}], points2D[{{}, 3}])
      points2D[{{}, 2}] = torch.cdiv(points2D[{{}, 2}], points2D[{{}, 3}])
      local lineExists = false
      for pointIndex = 1, 2 do
         local point2D = points2D[pointIndex]
         --if point2D[3] > 0 and point2D[1] >= 1 and point2D[1] <= width and point2D[2] >= 1 and point2D[2] <= height then
         if point2D[3] > 0 then
            lineExists = true
         end
      end
      lineExists = true
      local line2D = torch.zeros(1, 2, 2)
      if lineExists then
         line2D = points2D[{{}, {1, 2}}]:repeatTensor(1, 1, 1)
      end
      if not lines2D then
         lines2D = line2D
      else
         lines2D = torch.cat(lines2D, line2D, 1)
      end
   end
   return lines2D
end

function utils.loadPointCloud(filename)
   local representationExists, representationInfo = pcall(function()
         return csvigo.load({path=filename, mode="large", header=false, separator=' ', verbose=false})
   end)
   local points = {}
   if representationExists and representationInfo ~= nil then
      local numPoints = tonumber(representationInfo[1][3])
      for pointIndex, point in pairs(representationInfo) do
	 if pointIndex >= 3 then
	    table.insert(points, {point[2], point[3], point[4]})
	 end
	 if pointIndex - 2 == numPoints then
	    break
	 end
      end
   end
   return torch.Tensor(points)
end

function utils.samplePoints(points, numSampledPoints)
   local indices = torch.randperm(points:size(1)):narrow(1, 1, numSampledPoints):long()
   local sampledPoints = points:index(1, indices)
   return sampledPoints
end

function utils.drawTopDownView(width, height, points, angle, transformation)
   local X = points[{{}, 1}]
   local Y = points[{{}, 2}]
   local points2D = torch.cat(X, Y, 2)

   local mean = torch.mean(points2D, 1)
   points2D = points2D - mean:expandAs(points2D)
   points2D:div(math.sqrt(points2D:size(1) - 1))

   if not transformation then
      if not angle then
      --local w, _, _ = torch.svd(points2D:t())
      --angle = torch.atan2(w[1][2], w[1][1])
      angle = 0
      end

      local newX = X * torch.cos(angle) + Y * torch.sin(angle)
      local newY = X * torch.sin(angle) - Y * torch.cos(angle)
      local newPoints2D = torch.cat(newX, newY, 2)

      local mins = torch.min(newPoints2D, 1)[1]
      local maxs = torch.max(newPoints2D, 1)[1]
      local paddingRatio = 0.05
      local padding = (maxs - mins) * paddingRatio
      mins = mins - padding
      maxs = maxs + padding

      local scaleFactor = math.min(width / (maxs[1] - mins[1]), height / (maxs[2] - mins[2]))
      transformation = torch.zeros(3, 4)
      transformation[1][1] = torch.cos(angle)
      transformation[1][2] = torch.sin(angle)
      transformation[1][4] = -mins[1]
      transformation[2][1] = torch.sin(angle)
      transformation[2][2] = -torch.cos(angle)
      transformation[2][4] = -mins[2]
      transformation[3][4] = 1
      transformation[1] = transformation[1] * scaleFactor
      transformation[2] = transformation[2] * scaleFactor
      --transformation[1], transformation[2] = transformation[2], transformation[1]\
      
   end

   local uv = utils.project(transformation, points)
   local topDownView = torch.zeros(height, width)
   for i = 1, uv:size(1) do
      local point = uv[i]
      topDownView[math.min(math.max(point[2], 1), height)][math.min(math.max(point[1], 1), width)] = topDownView[math.min(math.max(point[2], 1), height)][math.min(math.max(point[1], 1), width)] + 1
   end
   --image.save('test/pointcloud.png', topDownView)
   local pointDensity = 1
   topDownView = topDownView / topDownView:max()
--   topDownView[topDownView:gt(1)] = 1

   topDownView = torch.repeatTensor(topDownView, 3, 1, 1)
   return topDownView, transformation
end

function utils.getRotationMatrix(quaternion)
   qi = quaternion[1]
   qj = quaternion[2]
   qk = quaternion[3]
   qr = quaternion[4]
   local rotation = torch.zeros(4, 4)
   rotation[1][1] = 1 - 2 * (qj^2 + qk^2)
   rotation[1][2] = 2 * (qi * qj - qk * qr)
   rotation[1][3] = 2 * (qi * qk + qj * qr)
   rotation[2][1] = 2 * (qi * qj + qk * qr)
   rotation[2][2] = 1 - 2 * (qi^2 + qk^2)
   rotation[2][3] = 2 * (qj * qk - qi * qr)
   rotation[3][1] = 2 * (qi * qk - qj * qr)
   rotation[3][2] = 2 * (qj * qk + qi * qr)
   rotation[3][3] = 1 - 2 * (qi^2 + qj^2)
   rotation[4][4] = 1
   return rotation
end

function utils.getTransformation(camera, pose, orientation)
   local K = torch.zeros(3, 3)
   K[1][2] = 1
   K[2][1] = -1
   K[3][3] = 1
   local intrinsics = camera.intrinsics:clone()

   if not orientation or orientation == 1 then
      intrinsics[{{}, 1}] = -intrinsics[{{}, 1}]
      intrinsics[{{}, 3}] = -intrinsics[{{}, 3}]
   elseif orientation == 2 then
      intrinsics[{{}, 3}] = -intrinsics[{{}, 3}]
      local temp = intrinsics[{{}, 1}]:clone()
      intrinsics[{{}, 1}] = -intrinsics[{{}, 2}]
      intrinsics[{{}, 2}] = -temp
      local temp = intrinsics[1][3]
      intrinsics[1][3] = intrinsics[2][3]
      intrinsics[2][3] = temp
   end

   --intrinsics[2][2] = -intrinsics[2][2]
   intrinsics = torch.cat(intrinsics, torch.zeros(3), 2)

   local inverseQuaternion = pose[{{4, 7}}]:clone()
   inverseQuaternion[{{1, 3}}] = -inverseQuaternion[{{1, 3}}]
   local rotation = utils.getRotationMatrix(inverseQuaternion)
   local translation = torch.zeros(4, 4)
   translation[1][1] = 1
   translation[2][2] = 1
   translation[3][3] = 1
   translation[1][4] = -pose[1]
   translation[2][4] = -pose[2]
   translation[3][4] = -pose[3]
   translation[4][4] = 1
   local transformation = intrinsics * rotation * translation
   print('frame')
   print(pose)
   print(intrinsics)
   -- print(K)
   print(rotation)
   print(translation)
   print(transformation)

   local point = torch.Tensor({-0.0527, 2.0441, 1.000, 1})
   -- print(point)
   -- print(rotation * point)
   -- print(translation * rotation * point)
   -- print(transformation * point)
   return transformation
end

-- function utils.getRotationFromAngle(angle)
--    rotation[1][1] = torch.cos(angle)
--    rotation[1][2] = torch.sin(angle)
--    rotation[2][1] = torch.sin(angle)
--    rotation[2][2] = -torch.cos(angle)
--    rotation[3][3] = 1
--    return rotation
-- end

return utils