def __init__(self, transf_file_path, sname, snum, objname, objnum,
ap_defaults=None, icp_defaults=None,
umbral_score=1e-3, **kwargs):
(super(SDTWithPostAlignment, self)
.__init__(transf_file_path, sname, snum, objname, objnum))
# alignment prerejective e ICP
self.umbral_score = umbral_score
if ap_defaults is None:
# alignment prerejective parameters
ap_defaults = APDefaults()
ap_defaults.leaf = 0.005
ap_defaults.max_ransac_iters = 100
ap_defaults.points_to_sample = 3
ap_defaults.nearest_features_used = 4
ap_defaults.simil_threshold = 0.1
ap_defaults.inlier_threshold = 3
ap_defaults.inlier_fraction = 0.8
# ap_defaults.show_values = True
self._ap_defaults = ap_defaults
if icp_defaults is None:
# icp parameters
icp_defaults = ICPDefaults()
icp_defaults.euc_fit = 1e-5
icp_defaults.max_corr_dist = 3
icp_defaults.max_iter = 50
icp_defaults.transf_epsilon = 1e-5
# icp_defaults.show_values = True
self._icp_defaults = icp_defaults
# Seteo el tamaño de las esferas usadas para filtrar de la escena
# los puntos del objeto encontrado
first_leaf_size = kwargs.get('first_leaf_size', 0.005)
self.adapt_leaf = kwargs.get(
'adapt_leaf',
AdaptLeafRatio(first_leaf=first_leaf_size)
)
# Seteo el porcentaje de puntos que permito conservar del modelo del
# objeto antes de considerar que lo que se encontró no es el objeto
self.perc_obj_model_points = kwargs.get('perc_obj_model_points', 0.5)
# Seteo el tamaño del frame de busqueda. Este valor se va a multiplicar
# por la altura y el ancho del objeto. Ej: si se multiplica por 2, el
# frame de busqueda tiene un area 4 (2*2) veces mayor que la del objeto
self.obj_mult = kwargs.get('obj_mult', 2)
def __init__(self,
transf_file_path,
sname,
snum,
objname,
objnum,
ap_defaults=None,
icp_defaults=None,
umbral_score=1e-3,
**kwargs):
(super(SDTWithPostAlignment, self).__init__(transf_file_path, sname,
snum, objname, objnum))
# alignment prerejective e ICP
self.umbral_score = umbral_score
if ap_defaults is None:
# alignment prerejective parameters
ap_defaults = APDefaults()
ap_defaults.leaf = 0.005
ap_defaults.max_ransac_iters = 100
ap_defaults.points_to_sample = 3
ap_defaults.nearest_features_used = 4
ap_defaults.simil_threshold = 0.1
ap_defaults.inlier_threshold = 3
ap_defaults.inlier_fraction = 0.8
# ap_defaults.show_values = True
self._ap_defaults = ap_defaults
if icp_defaults is None:
# icp parameters
icp_defaults = ICPDefaults()
icp_defaults.euc_fit = 1e-5
icp_defaults.max_corr_dist = 3
icp_defaults.max_iter = 50
icp_defaults.transf_epsilon = 1e-5
# icp_defaults.show_values = True
self._icp_defaults = icp_defaults
# Seteo el tamaño de las esferas usadas para filtrar de la escena
# los puntos del objeto encontrado
first_leaf_size = kwargs.get('first_leaf_size', 0.005)
self.adapt_leaf = kwargs.get(
'adapt_leaf', AdaptLeafRatio(first_leaf=first_leaf_size))
# Seteo el porcentaje de puntos que permito conservar del modelo del
# objeto antes de considerar que lo que se encontró no es el objeto
self.perc_obj_model_points = kwargs.get('perc_obj_model_points', 0.5)
# Seteo el tamaño del frame de busqueda. Este valor se va a multiplicar
# por la altura y el ancho del objeto. Ej: si se multiplica por 2, el
# frame de busqueda tiene un area 4 (2*2) veces mayor que la del objeto
self.obj_mult = kwargs.get('obj_mult', 2)
def __init__(self, icp_defaults=None, umbral_score=1e-4):
super(ICPFinder, self).__init__()
if icp_defaults is None:
icp_defaults = ICPDefaults()
icp_defaults.euc_fit = 1e-15
icp_defaults.max_corr_dist = 0.3
icp_defaults.max_iter = 50
icp_defaults.transf_epsilon = 1e-15
self._icp_defaults = icp_defaults
self.umbral_score = umbral_score
def prueba_solo_deteccion_automatica(img_provider):
# Set detection parameters values
det_template_threshold = 0.2
ap_defaults = APDefaults()
ap_defaults.leaf = 0.005
ap_defaults.max_ransac_iters = 120
ap_defaults.points_to_sample = 3
ap_defaults.nearest_features_used = 4
ap_defaults.simil_threshold = 0.6
ap_defaults.inlier_threshold = 4
ap_defaults.inlier_fraction = 0.3
icp_detection_defaults = ICPDefaults()
icp_detection_defaults.euc_fit = 1e-10
icp_detection_defaults.max_corr_dist = 3
icp_detection_defaults.max_iter = 50
icp_detection_defaults.transf_epsilon = 1e-6
det_umbral_score = 1e-3
det_obj_scene_leaf = 0.005
det_perc_obj_model_points = 0.5
det_depth_area_extra = 0.333
detector = RGBDDetector(
template_threshold=det_template_threshold,
ap_defaults=ap_defaults,
icp_defaults=icp_detection_defaults,
umbral_score=det_umbral_score,
perc_obj_model_points=det_perc_obj_model_points,
first_leaf_size=det_obj_scene_leaf,
depth_area_extra=det_depth_area_extra,
)
finder = Finder()
follower = DetectionWithCombinedFollowers(
img_provider,
detector,
finder,
finder
)
for i in range(3):
FollowingSchemeExploringParameterRGBD(
img_provider,
follower,
'pruebas_guardadas',
'probando_deteccion_automatica_RGBD',
'UNICO',
).run()
img_provider.restart()
def __init__(self,
matfile_path,
obj_rgbd_name,
obj_rgbd_num,
ap_defaults=APDefaults(),
icp_defaults=ICPDefaults(),
leaf_size=0.002,
icp_threshold=1e-3,
perc_obj_model_pts=0.5):
(super(StaticDetectorWithModelAlignment,
self).__init__(matfile_path, obj_rgbd_name, obj_rgbd_num))
self._ap_defaults = ap_defaults
self._icp_defaults = icp_defaults
# El tamaño del radio de la esfera para buscar puntos en la escena
self.leaf_size = leaf_size
# Umbral de aceptacion para ICP
self.icp_threshold = icp_threshold
# Porcentaje minimo de puntos del modelo que deben detectarse
self.perc_obj_model_pts = perc_obj_model_pts
def sistema_de_seguimiento_automatico_completo_rgb_rgbd(img_provider):
# Set detection parameters values
det_template_threshold = 0.2
ap_defaults = APDefaults()
ap_defaults.leaf = 0.005
ap_defaults.max_ransac_iters = 120
ap_defaults.points_to_sample = 3
ap_defaults.nearest_features_used = 4
ap_defaults.simil_threshold = 0.6
ap_defaults.inlier_threshold = 4
ap_defaults.inlier_fraction = 0.3
icp_detection_defaults = ICPDefaults()
icp_detection_defaults.euc_fit = 1e-10
icp_detection_defaults.max_corr_dist = 3
icp_detection_defaults.max_iter = 50
icp_detection_defaults.transf_epsilon = 1e-6
det_umbral_score = 1e-3
det_obj_scene_leaf = 0.005
det_perc_obj_model_points = 0.5
det_depth_area_extra = 0.333
# Set depth following parameters values
icp_finder_defaults = ICPDefaults()
icp_finder_defaults.euc_fit = 1e-10
icp_finder_defaults.max_corr_dist = 0.1
icp_finder_defaults.max_iter = 50
icp_finder_defaults.transf_epsilon = 1e-6
find_umbral_score = 1e-4
find_adapt_area = FixedSearchArea(3)
find_adapt_leaf = AdaptLeafRatio()
find_obj_scene_leaf = 0.002
find_perc_obj_model_points = 0.6
# Set RGB following parameters values
find_template_comp_method = cv2.cv.CV_COMP_BHATTACHARYYA
template_perc = 0.6
template_worst = 1
find_template_reverse = False
find_frame_comp_method = cv2.cv.CV_COMP_BHATTACHARYYA
frame_perc = 0.3
frame_worst = 1
find_frame_reverse = False
metodo_de_busqueda = BusquedaCambiandoSizePeroMismoCentro()
# Repetir 3 veces para evitar detecciones fallidas por RANSAC
for i in range(3):
detector = RGBDDetector(
template_threshold=det_template_threshold,
ap_defaults=ap_defaults,
icp_defaults=icp_detection_defaults,
umbral_score=det_umbral_score,
perc_obj_model_points=det_perc_obj_model_points,
first_leaf_size=det_obj_scene_leaf,
depth_area_extra=det_depth_area_extra,
)
template_comparator = HistogramComparator(
method=find_template_comp_method,
perc=template_perc,
worst_case=template_worst,
reverse=find_template_reverse,
)
frame_comparator = HistogramComparator(
method=find_frame_comp_method,
perc=frame_perc,
worst_case=frame_worst,
reverse=find_frame_reverse,
)
rgb_finder = TemplateAndFrameHistogramFinder(
template_comparator,
frame_comparator,
metodo_de_busqueda,
)
depth_finder = ICPFinderWithModelToImproveRGB(
icp_defaults=icp_finder_defaults,
umbral_score=find_umbral_score,
adapt_area=find_adapt_area,
adapt_leaf=find_adapt_leaf,
first_leaf_size=find_obj_scene_leaf,
perc_obj_model_points=find_perc_obj_model_points,
)
follower = DetectionWithCombinedFollowers(
image_provider=img_provider,
detector=detector,
main_finder=depth_finder,
secondary_finder=rgb_finder,
)
# mostrar_seguimiento = MuestraSeguimientoEnVivo('Seguimiento')
# FollowingScheme(img_provider, follower, mostrar_seguimiento).run()
FollowingSchemeExploringParameterRGBD(
img_provider,
follower,
'pruebas_guardadas',
'definitivo_automatico_RGB_RGBD',
'UNICO',
).run()
img_provider.restart()
def correr_con_depth_como_principal(img_provider, scenename, scenenumber,
objname, objnum):
# Set alignment detection parameters values
ap_defaults = APDefaults()
ap_defaults.leaf = 0.005
ap_defaults.max_ransac_iters = 120
ap_defaults.points_to_sample = 3
ap_defaults.nearest_features_used = 4
ap_defaults.simil_threshold = 0.6
ap_defaults.inlier_threshold = 4
ap_defaults.inlier_fraction = 0.3
icp_detection_defaults = ICPDefaults()
icp_detection_defaults.euc_fit = 1e-10
icp_detection_defaults.max_corr_dist = 3
icp_detection_defaults.max_iter = 50
icp_detection_defaults.transf_epsilon = 1e-6
det_umbral_score = 1e-3
det_obj_scene_leaf = 0.005
det_perc_obj_model_points = 0.5
# Set depth following parameters values
icp_finder_defaults = ICPDefaults()
icp_finder_defaults.euc_fit = 1e-10
icp_finder_defaults.max_corr_dist = 0.1
icp_finder_defaults.max_iter = 50
icp_finder_defaults.transf_epsilon = 1e-6
find_umbral_score = 1e-4
find_adapt_area = FixedSearchArea(3)
find_adapt_leaf = AdaptLeafRatio()
find_obj_scene_leaf = 0.002
find_perc_obj_model_points = 0.6
# Set RGB following parameters values
find_template_comp_method = cv2.cv.CV_COMP_BHATTACHARYYA
template_perc = 0.6
template_worst = 1
find_template_reverse = False
find_frame_comp_method = cv2.cv.CV_COMP_BHATTACHARYYA
frame_perc = 0.3
frame_worst = 1
find_frame_reverse = False
metodo_de_busqueda = BusquedaCambiandoSizePeroMismoCentro()
# Repetir 3 veces para evitar detecciones fallidas por RANSAC
for i in range(3):
detector = StaticDetectorForRGBD(
matfile_path=('videos/rgbd/scenes/{sname}/{sname}_{snum}.mat'
.format(sname=scenename, snum=scenenumber)),
obj_rgbd_name=objname,
obj_rgbd_num=objnum,
ap_defaults=ap_defaults,
icp_defaults=icp_detection_defaults,
leaf_size=det_obj_scene_leaf,
icp_threshold=det_umbral_score,
perc_obj_model_pts=det_perc_obj_model_points
)
template_comparator = HistogramComparator(
method=find_template_comp_method,
perc=template_perc,
worst_case=template_worst,
reverse=find_template_reverse,
)
frame_comparator = HistogramComparator(
method=find_frame_comp_method,
perc=frame_perc,
worst_case=frame_worst,
reverse=find_frame_reverse,
)
rgb_finder = ImproveObjectFoundWithHistogramFinder(
template_comparator,
frame_comparator,
metodo_de_busqueda,
)
depth_finder = ICPFinderWithModel(
icp_defaults=icp_finder_defaults,
umbral_score=find_umbral_score,
adapt_area=find_adapt_area,
adapt_leaf=find_adapt_leaf,
first_leaf_size=find_obj_scene_leaf,
perc_obj_model_points=find_perc_obj_model_points,
)
follower = DetectionWithCombinedFollowers(
image_provider=img_provider,
detector=detector,
main_finder=depth_finder,
secondary_finder=rgb_finder,
)
# mostrar_seguimiento = MuestraSeguimientoEnVivo('Seguimiento')
# FollowingScheme(img_provider, follower, mostrar_seguimiento).run()
FollowingSchemeExploringParameterRGBD(
img_provider,
follower,
'pruebas_guardadas',
'definitivo_RGBD_preferD',
'DEFINITIVO',
).run()
img_provider.restart()