Example #1
0
def extract_and_fuse_single_scene(log_full_path, downsample=False):

    print "extracting and fusing scene:", log_full_path
    log = os.path.split(log_full_path)[-1]

    mem()

    if not already_extracted_rosbag(log_full_path):
        print "extracting", log_full_path
        bag_filepath = os.path.join(log_full_path, 'raw',
                                    'fusion_' + log + '.bag')

        # run this memory intensive step in a process, to allow
        # it to release memory back to the operating system
        # when finished
        proc = mp.Process(target=extract_data_from_rosbag,
                          args=(bag_filepath, ))
        proc.start()
        proc.join()

        print "finished extracting", log_full_path
        # counter_new_extracted += 1
    else:
        print "already extracted", log_full_path

    processed_dir = os.path.join(log_full_path, 'processed')
    images_dir = os.path.join(processed_dir, 'images')

    # we need to free some memory
    gc.collect()
    mem()

    if not already_ran_tsdf_fusion(log_full_path):
        print "preparing for tsdf_fusion for", log_full_path
        tsdf_fusion.format_data_for_tsdf(images_dir)
        gc.collect()

        print "running tsdf fusion"
        tsdf_fusion.run_tsdf_fusion_cuda(images_dir)
        gc.collect()

        print "converting tsdf to ply"
        tsdf_bin_filename = os.path.join(processed_dir, 'tsdf.bin')
        tsdf_mesh_filename = os.path.join(processed_dir, 'fusion_mesh.ply')
        tsdf_fusion.convert_tsdf_to_ply(tsdf_bin_filename, tsdf_mesh_filename)
        # counter_new_fused += 1
    else:
        print "already ran tsdf_fusion for", log_full_path

    if not already_downsampled(log_full_path) and downsample:
        fs = FusionServer()
        print "downsampling image folder"
        linear_distance_threshold = 0.03
        angle_distance_threshold = 10  # in degrees
        fs.downsample_by_pose_difference_threshold(images_dir,
                                                   linear_distance_threshold,
                                                   angle_distance_threshold)
        # counter_new_downsampled += 1
    else:
        print "already downsampled for", log_full_path
Example #2
0
def main():
    # extract_images_from_rosbag()

    # tsdf_fusion.format_data_for_tsdf(image_folder)
    tsdf_fusion.run_tsdf_fusion_cuda(image_folder)

    tsdf_bin_file = os.path.join(data_folder, 'tsdf.bin')
    tsdf_mesh_file = os.path.join(data_folder, 'fusion_mesh.ply')

    tsdf_fusion.convert_tsdf_to_ply(tsdf_bin_file, tsdf_mesh_file)
Example #3
0
def main():
    print "preparing for tsdf_fusion for"
    tsdf_fusion.format_data_for_tsdf(images_dir)

    print "running tsdf fusion"
    tsdf_fusion.run_tsdf_fusion_cuda(images_dir)

    print "converting tsdf to ply"
    tsdf_bin_filename = os.path.join(processed_dir, 'tsdf.bin')
    tsdf_mesh_filename = os.path.join(processed_dir, 'fusion_mesh.ply')
    tsdf_fusion.convert_tsdf_to_ply(tsdf_bin_filename, tsdf_mesh_filename)
Example #4
0
def extract_and_fuse_single_scene(log_full_path, downsample=False):

    print "extracting and fusing scene:", log_full_path

    if not already_extracted_rosbag(log_full_path):
        print "extracting", log_full_path
        bag_filepath = os.path.join(log_full_path, 'raw',
                                    'fusion_' + log + '.bag')
        processed_dir, images_dir = fs.extract_data_from_rosbag(bag_filepath)
        # counter_new_extracted += 1
    else:
        print "already extracted", log_full_path
        processed_dir = os.path.join(log_full_path, 'processed')
        images_dir = os.path.join(processed_dir, 'images')

    if not already_ran_tsdf_fusion(log_full_path):
        print "preparing for tsdf_fusion for", log_full_path
        tsdf_fusion.format_data_for_tsdf(images_dir)

        print "running tsdf fusion"
        tsdf_fusion.run_tsdf_fusion_cuda(images_dir)

        print "converting tsdf to ply"
        tsdf_bin_filename = os.path.join(processed_dir, 'tsdf.bin')
        tsdf_mesh_filename = os.path.join(processed_dir, 'fusion_mesh.ply')
        tsdf_fusion.convert_tsdf_to_ply(tsdf_bin_filename, tsdf_mesh_filename)
        # counter_new_fused += 1
    else:
        print "already ran tsdf_fusion for", log_full_path

    if not already_downsampled(log_full_path) and downsample:
        print "downsampling image folder"
        linear_distance_threshold = 0.03
        angle_distance_threshold = 10  # in degrees
        fs.downsample_by_pose_difference_threshold(images_dir,
                                                   linear_distance_threshold,
                                                   angle_distance_threshold)
        # counter_new_downsampled += 1
    else:
        print "already downsampled for", log_full_path
def extract_and_fuse_single_scene(
        log_full_path,
        downsample=True,
        linear_distance_threshold=LINEAR_DISTANCE_THRESHOLD,
        angle_distance_threshold=ANGLE_DISTANCE_THRESHOLD):
    """
    Extracts all the images from the rosbag.
    :param log_full_path:
    :param downsample: whether or not to downsample
    :param linear_distance_threshold: threshold used in downsampling
    :param angle_distance_threshold: threshold used in downsampling
    :return:
    """
    global fs
    if fs is None:
        fs = FusionServer()

    print "extracting and fusing scene:", log_full_path
    log = os.path.split(log_full_path)[-1]

    mem()

    processed_dir = os.path.join(log_full_path, 'processed')
    images_dir = os.path.join(processed_dir, 'images')
    raw_dir = os.path.join(log_full_path, 'raw')
    raw_close_up_dir = os.path.join(log_full_path, 'raw_close_up')
    bag_filepath = os.path.join(log_full_path, 'raw', 'fusion_' + log + '.bag')
    bag_close_up_filepath = os.path.join(raw_close_up_dir,
                                         'fusion_' + log + '.bag')
    pose_data_filename = os.path.join(images_dir, 'pose_data.yaml')
    metadata_filename = os.path.join(images_dir, 'metadata.yaml')

    if not already_extracted_rosbag(log_full_path):
        print "extracting", bag_filepath

        # run this memory intensive step in a process, to allow
        # it to release memory back to the operating system
        # when finished
        proc = mp.Process(target=extract_data_from_rosbag,
                          args=(bag_filepath, images_dir))
        proc.start()
        proc.join()

        print "finished extracting", bag_filepath
        # counter_new_extracted += 1
    else:
        print "already extracted", bag_filepath

    # we need to free some memory
    gc.collect()
    mem()

    if not already_ran_tsdf_fusion(log_full_path):
        print "preparing for tsdf_fusion for", log_full_path
        tsdf_fusion.format_data_for_tsdf(images_dir)
        gc.collect()

        print "running tsdf fusion"
        fusion_params = FUSION_CONFIG[FUSION_LOCATION]
        tsdf_fusion.run_tsdf_fusion_cuda(
            images_dir,
            bbox_min=fusion_params['bbox_min'],
            bbox_max=fusion_params['bbox_max'],
            voxel_size=fusion_params['voxel_size'])
        gc.collect()

        print "converting tsdf to ply"
        tsdf_bin_filename = os.path.join(processed_dir, 'tsdf.bin')
        tsdf_mesh_filename = os.path.join(processed_dir, 'fusion_mesh.ply')
        tsdf_fusion.convert_tsdf_to_ply(tsdf_bin_filename, tsdf_mesh_filename)
        # counter_new_fused += 1
    else:
        print "already ran tsdf_fusion for", log_full_path

    if not already_downsampled(log_full_path) and downsample:
        print "downsampling image folder"

        # fs = FusionServer()
        fs.downsample_by_pose_difference_threshold(images_dir,
                                                   linear_distance_threshold,
                                                   angle_distance_threshold)
        # counter_new_downsampled += 1

        # create metadata.yaml file if it doesn't yet exist
        if not os.path.exists(metadata_filename):
            pose_data = spartanUtils.getDictFromYamlFilename(
                pose_data_filename)
            original_image_indices = pose_data.keys()

            metadata = dict()
            metadata['original_image_indices'] = original_image_indices
            metadata['close_up_image_indices'] = []
            spartanUtils.saveToYaml(metadata, metadata_filename)

    else:
        print "already downsampled for", log_full_path

    # extract the up close up log if it exists
    if os.path.exists(
            bag_close_up_filepath
    ) and not already_extracted_close_up_log(metadata_filename):
        print "\n\n-----------extracting close up images----------\n\n"
        print "extracting", bag_close_up_filepath
        close_up_temp_dir = os.path.join(processed_dir, 'images_close_up')

        # run this memory intensive step in a process, to allow
        # it to release memory back to the operating system
        # when finished
        proc = mp.Process(target=extract_data_from_rosbag,
                          args=(bag_close_up_filepath, close_up_temp_dir))
        proc.start()
        proc.join()

        print "downsampling image folder %s" % (close_up_temp_dir)
        if downsample:
            fs.downsample_by_pose_difference_threshold(
                close_up_temp_dir, linear_distance_threshold,
                angle_distance_threshold)

        pose_data_close_up_filename = os.path.join(close_up_temp_dir,
                                                   'pose_data.yaml')
        pose_data_close_up = spartanUtils.getDictFromYamlFilename(
            pose_data_close_up_filename)

        pose_data = spartanUtils.getDictFromYamlFilename(pose_data_filename)
        original_image_indices = pose_data.keys()
        largest_original_image_idx = max(original_image_indices)

        close_up_image_indices = []

        print "largest_original_image_idx", largest_original_image_idx

        for img_idx, data in pose_data_close_up.iteritems():
            # just to give some buffer and avoid edge cases in some checks
            img_idx_new = img_idx + largest_original_image_idx + 4
            rgb_filename = "%06i_%s.png" % (img_idx_new, "rgb")
            depth_filename = "%06i_%s.png" % (img_idx_new, "depth")

            rgb_filename_prev = data['rgb_image_filename']
            depth_filename_prev = data['depth_image_filename']

            data['rgb_image_filename'] = rgb_filename
            data['depth_image_filename'] = depth_filename

            pose_data[img_idx_new] = data
            close_up_image_indices.append(img_idx_new)

            # move images around
            shutil.move(os.path.join(close_up_temp_dir, rgb_filename_prev),
                        os.path.join(images_dir, rgb_filename))

            shutil.move(os.path.join(close_up_temp_dir, depth_filename_prev),
                        os.path.join(images_dir, depth_filename))

        shutil.rmtree(close_up_temp_dir)

        original_image_indices.sort()
        close_up_image_indices.sort()
        spartanUtils.saveToYaml(pose_data, pose_data_filename)

        metadata = dict()
        metadata['normal_image_indices'] = original_image_indices
        metadata['close_up_image_indices'] = close_up_image_indices
        spartanUtils.saveToYaml(metadata, metadata_filename)

    else:
        if os.path.exists(bag_close_up_filepath):
            print "already extracted close up log, skipping"
        else:
            print "raw_close_up doesn't exist, skipping"
Example #6
0
            # publish the pointcloud to RVIZ
            elastic_fusion_output = resp3.fusion_output
            self.cache['fusion_output'] = elastic_fusion_output
            self.publish_pointcloud_to_rviz(
                elastic_fusion_output.point_cloud,
                self.cache['point_cloud_to_world_stamped'])

            response = CaptureSceneAndFuseResponse(elastic_fusion_output)

        elif self.config['fusion_type'] == FusionType.TSDF_FUSION:

            print "formatting data for tsdf fusion"
            tsdf_fusion.format_data_for_tsdf(images_dir)

            print "running tsdf fusion"
            tsdf_fusion.run_tsdf_fusion_cuda(images_dir)

            print "converting tsdf to ply"
            tsdf_bin_filename = os.path.join(processed_dir, 'tsdf.bin')
            tsdf_mesh_filename = os.path.join(processed_dir, 'fusion_mesh.ply')
            tsdf_fusion.convert_tsdf_to_ply(tsdf_bin_filename,
                                            tsdf_mesh_filename)

            # the response is not meaningful right now
            response = CaptureSceneAndFuseResponse()

        else:
            raise ValueError('unknown fusion type')

        # downsample data (this should be specifiable by an arg)
        print "downsampling image folder"
Example #7
0
class FusionServer(object):
    def __init__(self, camera_serial_number="carmine_1"):
        self.camera_serial_number = camera_serial_number
        self.bagging = False
        self.rosbag_proc = None
        self.tfBuffer = None
        storedPosesFile = os.path.join(spartanUtils.getSpartanSourceDir(),
                                       'src', 'catkin_projects',
                                       'station_config', 'RLG_iiwa_1',
                                       'stored_poses.yaml')
        self.storedPoses = spartanUtils.getDictFromYamlFilename(
            storedPosesFile)
        self.robotService = rosUtils.RobotService(
            self.storedPoses['header']['joint_names'])
        self.setupConfig()
        self.setupSubscribers()
        self.setupPublishers()
        self.setupTF()
        self.setupCache()

    def setupConfig(self):
        self.config = dict()
        self.config['scan'] = dict()
        #self.config['scan']['pose_list'] = ['scan_back', 'scan_left', 'scan_top', 'scan_right', 'scan_back']
        # with the Grasping group
        #self.config['scan']['pose_group'] = 'Grasping'
        #self.config['scan']['pose_list'] = ['scan_left_close', 'scan_left', 'scan_left_center', 'scan_above_table_far', 'scan_right_center', 'scan_right', 'scan_right_close']

        # regular far out scanning poses
        pose_list = []
        pose_list.append(["Elastic Fusion", 'home'])
        pose_list.append(["Elastic Fusion", 'home_closer'])
        pose_list.append(["Elastic Fusion", 'center_right'])
        pose_list.append(["Elastic Fusion", 'right'])
        pose_list.append(["Elastic Fusion", 'right_low'])
        pose_list.append(["Elastic Fusion", 'right_low_closer'])
        pose_list.append(["Elastic Fusion", 'center_right'])
        pose_list.append(["Elastic Fusion", 'home_closer'])
        pose_list.append(["Elastic Fusion", 'center_left_closer'])
        pose_list.append(["Elastic Fusion", 'center_left_low_closer'])
        pose_list.append(["Elastic Fusion", 'left_low'])
        pose_list.append(["Elastic Fusion", 'left_mid'])
        pose_list.append(["Elastic Fusion", 'center_left_low'])
        pose_list.append(["Elastic Fusion", 'center_left_low_closer'])
        pose_list.append(["Elastic Fusion", 'center_left_closer'])
        pose_list.append(["Elastic Fusion", 'home_closer'])
        pose_list.append(["Elastic Fusion", 'top_down'])
        pose_list.append(["Elastic Fusion", 'top_down_right'])
        pose_list.append(["Elastic Fusion", 'top_down_left'])

        self.config['scan']['pose_list'] = pose_list

        # quick scan for testing purposes
        pose_list = []
        pose_list.append(["Elastic Fusion", 'home'])
        pose_list.append(["Elastic Fusion", 'home_closer'])
        pose_list.append(["Elastic Fusion", 'top_down'])
        pose_list.append(["Elastic Fusion", 'top_down_right'])
        pose_list.append(["Elastic Fusion", 'top_down_left'])
        pose_list.append(["Elastic Fusion", 'home'])
        self.config['scan']['pose_list_quick'] = pose_list

        # close up scanning
        pose_list = []
        pose_list.append(["close_up_scan", "center"])
        pose_list.append(["close_up_scan", "center_2"])
        pose_list.append(["close_up_scan", "center_3"])
        pose_list.append(["close_up_scan", "center_back_1"])
        pose_list.append(["close_up_scan", "center_back_2"])
        pose_list.append(["close_up_scan", "center_back_3"])
        pose_list.append(["close_up_scan", "center"])
        pose_list.append(["close_up_scan", "right_1"])
        pose_list.append(["close_up_scan", "right_2"])
        pose_list.append(["close_up_scan", "right_3"])
        pose_list.append(["close_up_scan", "right_4"])
        pose_list.append(["close_up_scan", "far_right_1"])
        # pose_list.append(["close_up_scan", "far_right_2"])
        # pose_list.append(["close_up_scan", "far_right_1"])
        pose_list.append(["close_up_scan", "right_1"])
        pose_list.append(["close_up_scan", "center"])
        pose_list.append(["close_up_scan", "left_1"])
        pose_list.append(["close_up_scan", "left_2"])
        pose_list.append(["close_up_scan", "left_3"])
        pose_list.append(["close_up_scan", "far_left_1"])
        pose_list.append(["close_up_scan", "far_left_2"])
        pose_list.append(["close_up_scan", "extreme_left_1"])
        pose_list.append(["close_up_scan", "extreme_left_2"])
        pose_list.append(["close_up_scan", "extreme_left_3"])
        pose_list.append(["close_up_scan", "left_1"])
        pose_list.append(["close_up_scan", "center"])

        self.config['scan']['close_up'] = pose_list

        self.config['speed'] = dict()
        self.config['speed']['scan'] = 25
        self.config['speed']['fast'] = 30
        self.config['speed']['wrist_rotation'] = 45

        self.config['spin_rate'] = 1

        self.config['home_pose_name'] = 'home'
        self.config['sleep_time_before_bagging'] = 3.0
        self.config['world_frame'] = 'base'
        self.config[
            'camera_frame'] = "camera_" + self.camera_serial_number + "_rgb_optical_frame"

        self.config['sleep_time_at_each_pose'] = 0.01

        self.config["reconstruction_frame_id"] = "fusion_reconstruction"

        self.config['fusion_type'] = FusionType.TSDF_FUSION

        self.topics_to_bag = [
            "/tf", "/tf_static", "/camera_" + self.camera_serial_number +
            "/depth_registered/sw_registered/image_rect",
            "/camera_" + self.camera_serial_number +
            "/depth_registered/sw_registered/camera_info",
            "/camera_" + self.camera_serial_number + "/rgb/camera_info",
            "/camera_" + self.camera_serial_number + "/rgb/image_rect_color"
        ]

        self.topics_dict = dict()
        self.topics_dict[
            'rgb'] = "/camera_" + self.camera_serial_number + "/rgb/image_rect_color"
        self.topics_dict[
            'depth'] = "/camera_" + self.camera_serial_number + "/depth_registered/sw_registered/image_rect"
        self.topics_dict[
            'camera_info'] = "/camera_" + self.camera_serial_number + "/rgb/camera_info"

    def setupTF(self):
        tfWrapper = TFWrapper()
        tfBuffer = tfWrapper.getBuffer()
        if self.tfBuffer is None:
            self.tfBuffer = tfBuffer

        self.tf_broadcaster = tf2_ros.TransformBroadcaster()

    def setupSubscribers(self):
        self.subscribers = dict()

        self.subscribers['rgb'] = rosUtils.SimpleSubscriber(
            self.topics_dict['rgb'], sensor_msgs.msg.Image)

        self.subscribers['depth'] = rosUtils.SimpleSubscriber(
            self.topics_dict['depth'], sensor_msgs.msg.Image)

    def setupPublishers(self):
        topic_name = "fusion_reconstruction"
        self.pointcloud_publisher = rospy.Publisher(
            topic_name, sensor_msgs.msg.PointCloud2, queue_size=1)

    def startImageSubscribers(self):
        for key, sub in self.subscribers.iteritems():
            sub.start()

    def stopImageSubscribers(self):
        for key, sub in self.subscribers.iteritems():
            sub.stop()

    def setupCache(self):
        self.cache = dict()

    def flushCache(self):
        self.setupCache()

    """
    Get transform from rgb_optical_frame to world
    """

    def getRGBOpticalFrameToWorldTransform(self):
        rgbOpticalFrameToWorld = self.tfBuffer.lookup_transform(
            self.config['world_frame'],
            rosUtils.getRGBOpticalFrameName(self.camera_serial_number),
            rospy.Time(0))
        return rgbOpticalFrameToWorld

    def start_bagging(self,
                      bag_folder='logs_proto',
                      full_path_to_bag_file=None):
        """

        :param bag_folder:
        :type bag_folder:
        :param full_path_to_bag_file: (optional) full path to where we save bag
        :type full_path_to_bag_file:
        :return:
        :rtype:
        """
        self.flushCache()

        if full_path_to_bag_file is None:
            base_path = os.path.join(spartanUtils.getSpartanSourceDir(),
                                     'data_volume', 'pdc', bag_folder)
            log_id_name = spartanUtils.get_current_YYYY_MM_DD_hh_mm_ss()
            log_subdir = "raw"
            bagfile_directory = os.path.join(base_path, log_id_name,
                                             log_subdir)
            bagfile_name = "fusion_" + log_id_name

            full_path_to_bag_file = os.path.join(bagfile_directory,
                                                 bagfile_name)
            # make bagfile directory with name
            os.system("mkdir -p " + bagfile_directory)

        # create parent folder if it doesn't exist
        parent_folder = os.path.dirname(full_path_to_bag_file)
        if not os.path.exists(parent_folder):
            os.makedirs(parent_folder)

        topics_to_bag = [
            "/tf", "/tf_static", "/camera_" + self.camera_serial_number +
            "/depth_registered/sw_registered/image_rect",
            "/camera_" + self.camera_serial_number +
            "/depth_registered/sw_registered/camera_info",
            "/camera_" + self.camera_serial_number + "/rgb/camera_info",
            "/camera_" + self.camera_serial_number + "/rgb/image_rect_color"
        ]

        # add simple subscribers to fix xtion driver issues
        self.startImageSubscribers()

        # sleep to allow for xtion driver compression issues to be resolved
        rospy.sleep(self.config['sleep_time_before_bagging'])

        # get camera to world transform
        self.cache[
            'point_cloud_to_world_stamped'] = self.getRGBOpticalFrameToWorldTransform(
            )
        transform_stamped = self.cache['point_cloud_to_world_stamped']

        # build up command string
        rosbag_cmd = "rosbag record __name:=" + ROS_BAGGING_NODE_NAME
        rosbag_cmd += " -O " + full_path_to_bag_file
        for i in topics_to_bag:
            rosbag_cmd += " " + i

        # add some visibility
        print rosbag_cmd

        # start bagging
        self.bagging = True
        rosbag_proc = subprocess.Popen(rosbag_cmd,
                                       stdin=subprocess.PIPE,
                                       shell=True,
                                       cwd=parent_folder)

        rospy.loginfo("started image subscribers, sleeping for %d seconds",
                      self.config['sleep_time_before_bagging'])

        return os.path.join(full_path_to_bag_file), rosbag_proc

    def _stop_bagging(self):
        """
        Stops ROS bagging
        :return:
        :rtype:
        """
        cmd = "rosnode kill /" + ROS_BAGGING_NODE_NAME
        print "cmd", cmd
        os.system(cmd)

        self.bagging = False
        self.stopImageSubscribers()

    def handle_start_bagging_fusion_data(self, req):
        ## check if bagging already
        if self.bagging:
            return StartBaggingFusionDataResponse("ERROR: Already bagging!")

        ## start bagging
        filepath, self.rosbag_proc = self.start_bagging()
        self.bagging = True

        ## return the full path string to the data
        print "Returning filepath"
        return StartBaggingFusionDataResponse(filepath)

    def handle_stop_bagging_fusion_data(self, req):
        ## check if bagging already
        if not self.bagging:
            return StopBaggingFusionDataResponse(
                "ERROR: Not currently bagging! Nothing to stop...")

        self._stop_bagging()
        return StopBaggingFusionDataResponse("success")

    def handle_perform_elastic_fusion(self, req):
        ## call executable for filename
        fusion_folder = os.path.join(os.path.dirname(req.bag_filepath),
                                     "images")

        cmd = ". /opt/ros/kinetic/setup.sh && $SPARTAN_SOURCE_DIR/src/ElasticFusion/GUI/build/ElasticFusion -f -q -t 900 -d 2 -l " + fusion_folder

        rospy.loginfo("elastic fusion cmd = %s", cmd)
        os.system(cmd)

        ply_filename = fusion_folder + ".ply"

        # of type sensor_msgs.msg.PointCloud2
        point_cloud = self.convert_ply_to_pointcloud2(
            PlyData.read(ply_filename))

        res = PerformElasticFusionResponse()
        res.fusion_output.pointcloud_filepath = ply_filename
        res.fusion_output.point_cloud = point_cloud
        res.fusion_output.point_cloud_to_world_stamped = self.cache[
            'point_cloud_to_world_stamped']

        return res

    def convert_ply_to_pointcloud2(self, plydata):

        cloud_arr = np.zeros((len(plydata.elements[0].data), 3))
        for i in xrange(len(plydata.elements[0].data)):
            cloud_arr[i] = list(plydata.elements[0].data[i])[:3]

        return array_to_xyz_pointcloud2f(cloud_arr)

    @staticmethod
    def get_numpy_position_from_pose(pose):
        x = pose["camera_to_world"]["translation"]["x"]
        y = pose["camera_to_world"]["translation"]["y"]
        z = pose["camera_to_world"]["translation"]["z"]
        return np.asarray([x, y, z])

    @staticmethod
    def get_quaternion_from_pose(pose):
        quat = spartanUtils.getQuaternionFromDict(pose["camera_to_world"])
        x = quat["x"]
        y = quat["y"]
        z = quat["z"]
        w = quat["w"]
        return np.asarray([w, x, y, z])

    def get_joint_positions_for_pose(self, pose_data):
        """
        Looks up the joint positions for a given pose
        :param pose_data: list of strings [group_name, pose_name]. e.g.
        ["General", "q_nom"]
        :return: list[double] of joint angles
        """
        return copy.copy(self.storedPoses[pose_data[0]][pose_data[1]])

    def _move_robot_through_pose_list(self,
                                      pose_list,
                                      randomize_wrist=False,
                                      hit_original_poses=True):
        """
        Moves robot through the given list of poses
        :param pose_list: list of list of strings of form [pose_group, pose_name]
        :type pose_list:
        :param with_randomize_wrist: boolean flag on whether to randomizde wrist or not
        :type with_randomize_write:
        :return:
        :rtype:
        """

        joint_limit_safety_factor = 0.9
        wrist_joint_limit_degrees = 175.0
        safe_wrist_joint_limit_radians = (wrist_joint_limit_degrees * np.pi /
                                          180.0) * joint_limit_safety_factor

        for pose_data in pose_list:
            print "moving to", pose_data[1]
            joint_positions = self.get_joint_positions_for_pose(pose_data)

            if randomize_wrist:
                print "before randomize wrist", joint_positions
                joint_positions_random_wrist = copy.copy(joint_positions)
                joint_positions_random_wrist[-1] = np.random.uniform(
                    -safe_wrist_joint_limit_radians,
                    safe_wrist_joint_limit_radians)
                print "after randomize wrist", joint_positions
                self.robotService.moveToJointPosition(
                    joint_positions_random_wrist,
                    maxJointDegreesPerSecond=self.config['speed']['scan'])

                if hit_original_poses:
                    self.robotService.moveToJointPosition(
                        joint_positions,
                        maxJointDegreesPerSecond=self.config['speed']
                        ['wrist_rotation'])

            else:
                self.robotService.moveToJointPosition(
                    joint_positions,
                    maxJointDegreesPerSecond=self.config['speed']['scan'])

            rospy.sleep(self.config['sleep_time_at_each_pose'])

    def capture_scene(self):
        """
        This "moves around and captures all of the data needed for fusion". I.e., it:

        1. Moves the robot "home"
        2. Starts bagging all data needed for fusion
        3. Moves the robot around
        4. Stops bagging data
        5. Moves the robot back home

        This is not a service handler itself, but intended to be modularly called by service handlers.

        :return: bag_filepath, the full path to where (one) of the rosbag (fusion-*.bag) was saved
        :rtype: string

        """

        # first move home
        home_pose_joint_positions = self.storedPoses["Elastic Fusion"][
            self.config['home_pose_name']]
        print home_pose_joint_positions
        self.robotService.moveToJointPosition(
            home_pose_joint_positions,
            maxJointDegreesPerSecond=self.config['speed']['fast'])

        print "moved to home"

        # Start bagging for far out data collection
        # base_path = os.path.join(spartanUtils.getSpartanSourceDir(), 'data_volume', 'pdc', 'logs_proto')
        base_path = os.path.join(spartanUtils.getSpartanSourceDir(),
                                 'data_volume', 'pdc', 'logs_shoes')
        log_id_name = spartanUtils.get_current_YYYY_MM_DD_hh_mm_ss()
        log_subdir = "raw"
        bagfile_directory = os.path.join(base_path, log_id_name, log_subdir)
        bagfile_name = "fusion_" + log_id_name + ".bag"
        full_path_to_bagfile = os.path.join(bagfile_directory, bagfile_name)

        print "moving robot through regular scan poses"
        self.start_bagging(full_path_to_bag_file=full_path_to_bagfile)
        pose_list = self.config['scan']['pose_list']
        joint_positions = self.get_joint_positions_for_pose(pose_list[0])
        self.robotService.moveToJointPosition(
            joint_positions,
            maxJointDegreesPerSecond=self.config['speed']['scan'])
        # rospy.sleep(3.0)
        self._move_robot_through_pose_list(pose_list,
                                           randomize_wrist=True,
                                           hit_original_poses=True)

        self._stop_bagging()

        # # move robot through close up scan poses
        log_subdir = "raw_close_up"
        bagfile_directory = os.path.join(base_path, log_id_name, log_subdir)
        bagfile_name = "fusion_" + log_id_name + ".bag"
        full_path_to_bagfile = os.path.join(bagfile_directory, bagfile_name)
        #
        print "moving robot through close up scan poses"
        pose_list = self.config['scan']['close_up']

        # move to first pose before we start bagging
        joint_positions = self.get_joint_positions_for_pose(pose_list[0])
        self.robotService.moveToJointPosition(
            joint_positions,
            maxJointDegreesPerSecond=self.config['speed']['scan'])

        # now start bagging and move the robot through the poses
        self.start_bagging(full_path_to_bag_file=full_path_to_bagfile)
        self._move_robot_through_pose_list(pose_list,
                                           randomize_wrist=True,
                                           hit_original_poses=True)
        # rospy.sleep(3.0)
        self._stop_bagging()
        rospy.sleep(1.0)

        # move back home
        self.robotService.moveToJointPosition(
            home_pose_joint_positions,
            maxJointDegreesPerSecond=self.config['speed']['fast'])

        return full_path_to_bagfile

    def extract_data_from_rosbag(self,
                                 bag_filepath,
                                 images_dir=None,
                                 rgb_only=False):
        """
        This wraps the ImageCapture calls to load and process the raw rosbags, to prepare for fusion.

        :param: bag_filepath, the full path to where the rosbag (fusion-*.bag) was saved
        :ptype: string

        :return: data dir, images_dir the full path to the directory where all the extracted data is saved
                            and its images subdirectory
        :rtype: two strings, separated by commas
        """

        # extract RGB and Depth images from Rosbag
        rgb_topic = self.topics_dict['rgb']
        depth_topic = self.topics_dict['depth']
        camera_info_topic = self.topics_dict['camera_info']

        if images_dir is None:
            log_dir = os.path.dirname(os.path.dirname(bag_filepath))
            processed_dir = os.path.join(log_dir, 'processed')
            images_dir = os.path.join(processed_dir, 'images')

        print "Using images_dir %s" % (images_dir)
        image_capture = ImageCapture(rgb_topic,
                                     depth_topic,
                                     camera_info_topic,
                                     self.config['camera_frame'],
                                     self.config['world_frame'],
                                     rgb_encoding='bgr8')
        image_capture.load_ros_bag(bag_filepath)
        image_capture.process_ros_bag(image_capture.ros_bag,
                                      images_dir,
                                      rgb_only=rgb_only)

        rospy.loginfo("Finished writing images to disk")

        return images_dir

    def handle_capture_scene(self, req):
        print "handling capture_scene"

        # Capture scene
        bag_filepath = self.capture_scene()

        response = CaptureSceneResponse()
        response.bag_filepath = bag_filepath

        rospy.loginfo("handle_capture_scene finished!")
        return response

    def handle_capture_scene_and_fuse(self, req):
        print "handling capture_scene_and_fuse"

        # Capture scene
        bag_filepath = self.capture_scene()

        # Extract images from bag
        processed_dir, images_dir = self.extract_data_from_rosbag(bag_filepath)

        if self.config['fusion_type'] == FusionType.ELASTIC_FUSION:
            # Perform fusion
            try:
                perform_elastic_fusion = rospy.ServiceProxy(
                    'perform_elastic_fusion', PerformElasticFusion)
                resp3 = perform_elastic_fusion(bag_filepath)
            except rospy.ServiceException, e:
                print "Service call failed: %s" % e

            # publish the pointcloud to RVIZ
            elastic_fusion_output = resp3.fusion_output
            self.cache['fusion_output'] = elastic_fusion_output
            self.publish_pointcloud_to_rviz(
                elastic_fusion_output.point_cloud,
                self.cache['point_cloud_to_world_stamped'])

            response = CaptureSceneAndFuseResponse(elastic_fusion_output)

        elif self.config['fusion_type'] == FusionType.TSDF_FUSION:

            print "formatting data for tsdf fusion"
            tsdf_fusion.format_data_for_tsdf(images_dir)

            print "running tsdf fusion"
            tsdf_fusion.run_tsdf_fusion_cuda(images_dir)

            print "converting tsdf to ply"
            tsdf_bin_filename = os.path.join(processed_dir, 'tsdf.bin')
            tsdf_mesh_filename = os.path.join(processed_dir, 'fusion_mesh.ply')
            tsdf_fusion.convert_tsdf_to_ply(tsdf_bin_filename,
                                            tsdf_mesh_filename)

            # the response is not meaningful right now
            response = CaptureSceneAndFuseResponse()