def connect_ros_point_cloud(self): self.depthTo3d = calib.DepthTo3d('Depth ~> 3D') self.erode = imgproc.Erode('Mask Erosion', kernel=3) # -> 7x7 # this is for SXGA mode scale handling. self.rescale_depth = RescaledRegisteredDepth('Depth scaling') self.point_cloud_converter = MatToPointCloudXYZRGB('To Point Cloud') self.to_ecto_pcl = ecto_pcl.PointCloudT2PointCloud( 'converter', format=ecto_pcl.XYZRGB) self.cloud2msg = ecto_pcl_ros.PointCloud2Message("cloud2msg") self.cloud_pub = ecto_sensor_msgs.Publisher_PointCloud2( "cloud_pub", topic_name='/ecto_pcl/sample_output') self.graph += [ self.observation_renderer['depth', 'image', 'mask'] >> self.rescale_depth['depth', 'image', 'mask'], self.observation_renderer['K'] >> self.rescale_depth['K'], self.rescale_depth['K'] >> self.depthTo3d['K'], self.rescale_depth['depth'] >> self.depthTo3d['depth'], self.depthTo3d['points3d'] >> self.point_cloud_converter['points'], self.observation_renderer['image'] >> self.point_cloud_converter['image'], self.rescale_depth['mask'] >> self.erode['image'], self.erode['image'] >> self.point_cloud_converter['mask'], self.point_cloud_converter['point_cloud'] >> self.to_ecto_pcl['input'], self.to_ecto_pcl['output'] >> self.cloud2msg['input'], self.cloud2msg['output'] >> self.cloud_pub['input'], ]
def generate_pointclouds_in_object_space(dbs, session, args): object_name = dbs[session.object_id]['object_name'] if not os.path.exists(object_name): os.mkdir(object_name) obs_ids = models.find_all_observations_for_session(dbs, session.id) if len(obs_ids) == 0: raise RuntimeError("There are no observations available.") db_reader = capture.ObservationReader( 'Database Source', db_params=dbtools.args_to_db_params(args)) #observation dealer will deal out each observation id. observation_dealer = ecto.Dealer( tendril=db_reader.inputs.at('observation'), iterable=obs_ids) depthTo3d = calib.DepthTo3d('Depth ~> 3D') rescale_depth = capture.RescaledRegisteredDepth( 'Depth scaling') #this is for SXGA mode scale handling. point_cloud_transform = reconstruction.PointCloudTransform( 'Object Space Transform', do_transform=False ) #keeps the points in camera coordinates, but populates the global sensor position and orientatino. point_cloud_converter = conversion.MatToPointCloudXYZRGB('To Point Cloud') to_ecto_pcl = ecto_pcl.PointCloudT2PointCloud('converter', format=ecto_pcl.XYZRGB) plasm = ecto.Plasm() plasm.connect(observation_dealer[:] >> db_reader['observation'], db_reader['K'] >> depthTo3d['K'], db_reader['image'] >> rescale_depth['image'], db_reader['depth'] >> rescale_depth['depth'], rescale_depth[:] >> depthTo3d['depth'], depthTo3d['points3d'] >> point_cloud_converter['points'], db_reader['image'] >> point_cloud_converter['image'], db_reader['mask'] >> point_cloud_converter['mask'], db_reader['R', 'T'] >> point_cloud_transform['R', 'T'], point_cloud_converter['point_cloud'] >> to_ecto_pcl[:], to_ecto_pcl[:] >> point_cloud_transform['cloud']) ply_writer = ecto_pcl.PLYWriter('PLY Saver', filename_format='%s/cloud_%%05d.ply' % (object_name)) pcd_writer = ecto_pcl.PCDWriter('PCD Saver', filename_format='%s/cloud_%%05d.pcd' % (object_name)) plasm.connect(point_cloud_transform['view'] >> (ply_writer['input'], pcd_writer['input'])) if args.visualize: global cloud_view plasm.connect( point_cloud_transform['view'] >> cloud_view, db_reader['image'] >> imshows['image'], db_reader['depth'] >> imshows['depth'], db_reader['mask'] >> imshows['mask'], ) from ecto.opts import run_plasm run_plasm(args, plasm, locals=vars())
def create_capture_plasm(bag_name, angle_thresh, segmentation_cell, n_desired, sensor, res=SXGA_RES, fps=FPS_30, orb_matches=False, preview=False, use_turn_table=True): ''' Creates a plasm that will capture openni data into a bag, using a dot pattern to sparsify views. @param bag_name: A filename for the bag, will write to this file. @param angle_thresh: The angle threshhold in radians to sparsify the views with. ''' graph = [] # Create ros node, subscribing to openni topics argv = sys.argv[:] ecto_ros.strip_ros_args(argv) ecto_ros.init(argv, "object_capture_msd_pcl", False) cloud_sub = ecto_ros.ecto_sensor_msgs.Subscriber_PointCloud2("cloud_sub", topic_name='/camera/depth_registered/points') # openni&freenect requires '/camera/rgb/image_rect_color', openni2 just need '/camera/rgb/image_raw' if sensor in ['kinect']: rgb_image_sub = ecto_ros.ecto_sensor_msgs.Subscriber_Image("rgb_image_sub", topic_name='/camera/rgb/image_color') elif sensor in ['xtion']: rgb_image_sub = ecto_ros.ecto_sensor_msgs.Subscriber_Image("rgb_image_sub", topic_name='/camera/rgb/image_raw') depth_image_sub = ecto_ros.ecto_sensor_msgs.Subscriber_Image("depth_image_sub", topic_name='/camera/depth_registered/image_raw') rgb_camera_info_sub = ecto_ros.ecto_sensor_msgs.Subscriber_CameraInfo("rgb_camera_info_sub", topic_name='/camera/rgb/camera_info') depth_camera_info_sub = ecto_ros.ecto_sensor_msgs.Subscriber_CameraInfo("depth_camera_info_sub", topic_name='/camera/depth_registered/camera_info') # Converte ros topics to cv types msg2cloud = ecto_pcl_ros.Message2PointCloud("msg2cloud", format=XYZRGB) image2cv = ecto_ros.Image2Mat() depth2cv = ecto_ros.Image2Mat() rgb_info2cv = ecto_ros.CameraInfo2Cv() depth_info2cv = ecto_ros.CameraInfo2Cv() graph += [ cloud_sub['output'] >> msg2cloud[:], rgb_image_sub[:] >> image2cv[:], depth_image_sub[:] >> depth2cv[:], rgb_camera_info_sub[:] >> rgb_info2cv[:], depth_camera_info_sub[:] >> depth_info2cv[:] ] """ rgb_display = highgui.imshow(name='rgb') depth_display = highgui.imshow(name='depth') graph += [image2cv[:] >> rgb_display[:], depth2cv[:] >> depth_display[:], ] # Create pointcloud viewer viewer = CloudViewer("viewer", window_name="Clouds!") graph += [ msg2cloud[:] >> viewer[:] ] """ # Process pointcloud # Cut off some parts cut_x = PassThrough(filter_field_name="x", filter_limit_min=-0.2, filter_limit_max=0.2) cut_y = PassThrough(filter_field_name="y", filter_limit_min=-0.5, filter_limit_max=0.5) cut_z = PassThrough(filter_field_name="z", filter_limit_min=-0.0, filter_limit_max=1.0) graph += [ msg2cloud[:] >> cut_x[:], cut_x[:] >> cut_y[:], cut_y[:] >> cut_z[:] ] # Voxel filter voxel_grid = VoxelGrid("voxel_grid", leaf_size=0.002) graph += [ cut_z[:] >> voxel_grid[:] ] # Find plane and delete it normals = NormalEstimation("normals", k_search=0, radius_search=0.02) graph += [ voxel_grid[:] >> normals[:] ] plane_finder = SACSegmentationFromNormals("planar_segmentation", model_type=SACMODEL_NORMAL_PLANE, eps_angle=0.09, distance_threshold=0.1) plane_deleter = ExtractIndices(negative=True) graph += [ voxel_grid[:] >> plane_finder['input'], normals[:] >> plane_finder['normals'] ] graph += [ voxel_grid[:] >> plane_deleter['input'], plane_finder['inliers'] >> plane_deleter['indices'] ] # Create pointcloud viewer viewer = CloudViewer("viewer", window_name="Clouds!") graph += [ plane_deleter[:] >> viewer[:] ] # Compute vfh vfh_signature = VFHEstimation(radius_search=0.01) graph += [ plane_deleter[:] >> vfh_signature['input'], normals[:] >> vfh_signature['normals'] ] # convert the image to grayscale rgb2gray = imgproc.cvtColor('rgb -> gray', flag=imgproc.Conversion.RGB2GRAY) graph += [image2cv[:] >> rgb2gray[:] ] # get a pose use the dot pattern: there might be a scale ambiguity as this is 3d only poser = OpposingDotPoseEstimator(rows=5, cols=3, pattern_type=calib.ASYMMETRIC_CIRCLES_GRID, square_size=0.04, debug=True) graph += [ image2cv[:] >> poser['color_image'], rgb_info2cv['K'] >> poser['K_image'], rgb2gray[:] >> poser['image'] ] #poser gives us R&T from camera to dot pattern points3d = calib.DepthTo3d() graph += [ depth_info2cv['K'] >> points3d['K'], depth2cv['image'] >> points3d['depth'] ] plane_est = PlaneFinder(min_size=10000) compute_normals = ComputeNormals() # Convert K if the resolution is different (the camera should output that) graph += [ # find the normals depth_info2cv['K'] >> compute_normals['K'], points3d['points3d'] >> compute_normals['points3d'], # find the planes compute_normals['normals'] >> plane_est['normals'], depth_info2cv['K'] >> plane_est['K'], points3d['points3d'] >> plane_est['points3d'] ] pose_filter = object_recognition_capture.ecto_cells.capture.PlaneFilter(); # make sure the pose is centered at the origin of the plane graph += [ depth_info2cv['K'] >> pose_filter['K_depth'], poser['R', 'T'] >> pose_filter['R', 'T'], plane_est['planes', 'masks'] >> pose_filter['planes', 'masks'] ] delta_pose = ecto.If('delta R|T', cell=object_recognition_capture.DeltaRT(angle_thresh=angle_thresh, n_desired=n_desired)) poseMsg = RT2PoseStamped(frame_id='/camera_rgb_optical_frame') graph += [ pose_filter['R', 'T', 'found'] >> delta_pose['R', 'T', 'found'], pose_filter['R', 'T'] >> poseMsg['R', 'T'] ] trainer_ = Trainer() graph += [ pose_filter['R', 'T'] >> trainer_['R', 'T'], vfh_signature['output'] >> trainer_['features'], delta_pose['last'] >> trainer_['last'], delta_pose['novel'] >> trainer_['novel'], plane_deleter[:] >> trainer_['input'] ] novel = delta_pose['novel'] cloud2msg = ecto_pcl_ros.PointCloud2Message("cloud2msg") graph += [ plane_deleter[:] >> cloud2msg[:] ] baggers = dict(points=PointCloudBagger(topic_name='/camera/depth_registered/points'), pose=PoseBagger(topic_name='/camera/pose') ) bagwriter = ecto.If('Bag Writer if R|T', cell=ecto_ros.BagWriter(baggers=baggers, bag=bag_name)) pcd_writer = ecto.If('msd pcd writer', cell = PCDWriter()) graph += [ cloud2msg[:] >> bagwriter['points'], poseMsg['pose'] >> bagwriter['pose'] ] graph += [ plane_deleter['output'] >> pcd_writer['input'] ] delta_pose.inputs.__test__ = True pcd_writer.inputs.__test__ = True graph += [novel >> (bagwriter['__test__'])] graph += [novel >> (pcd_writer['__test__'])] rt2pose = RT2PoseStamped(frame_id = "camera_rgb_optical_frame") rt2pose_filtered = RT2PoseStamped(frame_id = "camera_rgb_optical_frame") posePub = PosePublisher("pose_pub", topic_name='pattern_pose') posePub_filtered = PosePublisher("pose_pub_filtered", topic_name='pattern_pose_filtered') graph += [ poser['R', 'T'] >> rt2pose['R', 'T'], rt2pose['pose'] >> posePub['input'], pose_filter['R', 'T'] >> rt2pose_filtered['R', 'T'], rt2pose_filtered['pose'] >> posePub_filtered['input'] ] plasm = ecto.Plasm() plasm.connect(graph) return (plasm, segmentation_cell) # return segmentation for tuning of parameters.