def computeForegroundMask(self, visualize=True):
"""
Computes the foreground mask. The camera location of the foreground view should
already be set correctly. Steps of the pipeline are
1. Set the background camera to same location as foreground camera
2. Render depth images from foreground and background scenes
3. Compute change detection based on pair of depth images
"""
# assume that foreground view is already in the correct place
view_f = self.views['foreground']
view_b = self.views['background']
view_f.forceRender()
cameraToWorld = director_utils.getCameraTransform(view_f.camera())
director_utils.setCameraTransform(view_f.camera(), cameraToWorld)
view_f.forceRender()
director_utils.setCameraTransform(view_b.camera(), cameraToWorld)
view_b.forceRender()
self.updateDepthScanners()
# get the depth images
# make sure to convert them to int32 since we want to avoid wrap-around issues when using
# uint16 types
depth_img_foreground_raw = self.depthScanners[
'foreground'].getDepthImageAsNumpyArray()
depth_img_f = np.array(depth_img_foreground_raw, dtype=np.int32)
depth_img_b = np.array(
self.depthScanners['background'].getDepthImageAsNumpyArray(),
dtype=np.int32)
# zero values are interpreted as max-range measurements
depth_img_b[depth_img_b == 0] = np.iinfo(np.int32).max
depth_img_f[depth_img_f == 0] = np.iinfo(np.int32).max
idx, mask = ChangeDetection.computeForegroundMaskFromDepthImagePair(
depth_img_f, depth_img_b, self.threshold)
depth_img_change = mask * depth_img_f
returnData = dict()
returnData['idx'] = idx
returnData['mask'] = mask
returnData['depth_img_foreground'] = depth_img_f
returnData['depth_img_background'] = depth_img_b
returnData['depth_img_foreground_raw'] = depth_img_foreground_raw
returnData['depth_img_change'] = depth_img_change
if visualize:
# make sure to remap the mask to [0,255], otherwise it will be all black
ChangeDetection.drawNumpyImage('Change Detection Mask', mask * 255)
return returnData
def testGetCameraTransforms(self):
vis.updateFrame(vtk.vtkTransform(), "origin frame", view=self.views['background'])
for viewType, view in self.views.iteritems():
print "\n\nviewType: ", viewType
cameraTransform = director_utils.getCameraTransform(view.camera())
pos, quat = transformUtils.poseFromTransform(cameraTransform)
print "pos: ", pos
print "quat: ", quat
vis.updateFrame(cameraTransform, viewType + "_frame")
def computeForegroundMask(self, visualize=True):
"""
Computes the foreground mask. The camera location of the foreground view should
already be set correctly. Steps of the pipeline are
1. Set the background camera to same location as foreground camera
2. Render depth images from foreground and background scenes
3. Compute change detection based on pair of depth images
"""
# assume that foreground view is already in the correct place
view_f = self.views['foreground']
view_b = self.views['background']
view_f.forceRender()
cameraToWorld = director_utils.getCameraTransform(view_f.camera())
director_utils.setCameraTransform(view_f.camera(), cameraToWorld)
view_f.forceRender()
director_utils.setCameraTransform(view_b.camera(), cameraToWorld)
view_b.forceRender()
self.updateDepthScanners()
# get the depth images
# make sure to convert them to int32 since we want to avoid wrap-around issues when using
# uint16 types
depth_img_foreground_raw = self.depthScanners['foreground'].getDepthImageAsNumpyArray()
depth_img_f = np.array(depth_img_foreground_raw, dtype=np.int32)
depth_img_b = np.array(self.depthScanners['background'].getDepthImageAsNumpyArray(), dtype=np.int32)
# zero values are interpreted as max-range measurements
depth_img_b[depth_img_b == 0] = np.iinfo(np.int32).max
depth_img_f[depth_img_f == 0] = np.iinfo(np.int32).max
idx, mask = ChangeDetection.computeForegroundMaskFromDepthImagePair(depth_img_f, depth_img_b, self.threshold)
depth_img_change = mask * depth_img_f
returnData = dict()
returnData['idx'] = idx
returnData['mask'] = mask
returnData['depth_img_foreground'] = depth_img_f
returnData['depth_img_background'] = depth_img_b
returnData['depth_img_foreground_raw'] = depth_img_foreground_raw
returnData['depth_img_change'] = depth_img_change
if visualize:
# make sure to remap the mask to [0,255], otherwise it will be all black
ChangeDetection.drawNumpyImage('Change Detection Mask', mask*255)
return returnData
def testGetCameraTransforms(self):
vis.updateFrame(vtk.vtkTransform(), "origin frame", view=self.views['background'])
for viewType, view in self.views.iteritems():
print "\n\nviewType: ", viewType
cameraTransform = director_utils.getCameraTransform(view.camera())
pos, quat = transformUtils.poseFromTransform(cameraTransform)
print "pos: ", pos
print "quat: ", quat
vis.updateFrame(cameraTransform, viewType + "_frame")
def computeForegroundMaskUsingCropStrategy(self, visualize=True):
"""
Computes the foreground mask. The camera location of the foreground view should
already be set correctly. Steps of the pipeline are
1. Render depth image from cropped foreground
2. Everything in the "crop" should be foreground for now
"""
# assume that foreground view is already in the correct place
view_f = self.views['foreground']
view_b = self.views['background']
view_f.forceRender()
cameraToWorld = director_utils.getCameraTransform(view_f.camera())
self.updateDepthScanners()
# get the depth images
# make sure to convert them to int32 since we want to avoid wrap-around issues when using
# uint16 types
depth_img_foreground_raw = self.depthScanners[
'foreground'].getDepthImageAsNumpyArray()
depth_img_f = np.array(depth_img_foreground_raw, dtype=np.int32)
idx = depth_img_f > 0
mask = np.zeros(np.shape(depth_img_f))
mask[idx] = 1
returnData = dict()
returnData['idx'] = idx
returnData['mask'] = mask
returnData['depth_img_foreground_raw'] = depth_img_foreground_raw
returnData['depth_img_foreground'] = depth_img_f
if visualize:
# make sure to remap the mask to [0,255], otherwise it will be all black
ChangeDetection.drawNumpyImage('Change Detection Mask', mask * 255)
return returnData
def computeForegroundMaskUsingCropStrategy(self, visualize=True):
"""
Computes the foreground mask. The camera location of the foreground view should
already be set correctly. Steps of the pipeline are
1. Render depth image from cropped foreground
2. Everything in the "crop" should be foreground for now
"""
# assume that foreground view is already in the correct place
view_f = self.views['foreground']
view_b = self.views['background']
view_f.forceRender()
cameraToWorld = director_utils.getCameraTransform(view_f.camera())
self.updateDepthScanners()
# get the depth images
# make sure to convert them to int32 since we want to avoid wrap-around issues when using
# uint16 types
depth_img_foreground_raw = self.depthScanners['foreground'].getDepthImageAsNumpyArray()
depth_img_f = np.array(depth_img_foreground_raw, dtype=np.int32)
idx = depth_img_f > 0
mask = np.zeros(np.shape(depth_img_f))
mask[idx] = 1
returnData = dict()
returnData['idx'] = idx
returnData['mask'] = mask
returnData['depth_img_foreground_raw'] = depth_img_foreground_raw
returnData['depth_img_foreground'] = depth_img_f
if visualize:
# make sure to remap the mask to [0,255], otherwise it will be all black
ChangeDetection.drawNumpyImage('Change Detection Mask', mask*255)
return returnData
def testIdentity(self):
view = self.views['foreground']
cameraToWorld = director_utils.getCameraTransform(view.camera())
self.setCameraTransform(cameraToWorld)
self.updateDepthScanners()
def showCameraTransform(self):
t = director_utils.getCameraTransform(
self.views['foreground'].camera())
vis.updateFrame(t, 'camera transform', scale=0.15)
def testIdentity(self):
view = self.views['foreground']
cameraToWorld = director_utils.getCameraTransform(view.camera())
self.setCameraTransform(cameraToWorld)
self.updateDepthScanners()
def showCameraTransform(self):
t = director_utils.getCameraTransform(self.views['foreground'].camera())
vis.updateFrame(t, 'camera transform', scale=0.15)
