def __init__(self, **kwargs):
pose_cell_shape = kwargs.pop("pose_cell_dimensions", (100,100,60))
shift_match = kwargs.pop("odo_shift_match", 20)
# self.odometer = SimpleVisualOdometer()
self.odometer = SimpleVisualOdometer(fov_deg = kwargs.pop("fov_deg",50),
im_trans_y_range = kwargs.pop("im_trans_y_range"),
im_rot_y_range = kwargs.pop("im_rot_y_range"),
im_odo_x_range = kwargs.pop("im_odo_x_range"),
shift_match = shift_match,
trans_scale = kwargs.pop("odo_trans_scale",100.))
# vt_match_threshold = kwargs.pop('vt_match_threshold', None)
# self.view_templates = VisualTemplateCollection()
self.view_templates = VisualTemplateCollection( template_decay = kwargs.pop('template_decay', 1.0),
match_y_range = kwargs.pop('im_vt_y_range', None),
match_x_range = kwargs.pop('im_vt_x_range', None),
shift_match = shift_match,
match_threshold = kwargs.pop('vt_match_threshold', None))
self.pose_cell_network = PoseCellNetwork(pose_cell_shape)#,
# self.view_templates)
self.experience_map = ExperienceMap(self.pose_cell_network,
self.view_templates,
delta_pc_threshold = kwargs.pop("exp_delta_pc_threshold",1.0),
correction_factor = kwargs.pop("exp_correction_factor",0.5),
exp_loops = kwargs.pop("exp_loops",100))
def __init__( self ):
self.im_count = 0
# Initialize the Pose Cell Network
self.pcn = PoseCellNetwork(shape=POSE_SIZE)
self.pc_count = 0
self.twist_data = deque()
self.odom_freq = ODOM_FREQ
midpoint = (math.floor(POSE_SIZE[0]/2),math.floor(POSE_SIZE[1]/2),math.floor(POSE_SIZE[2]/2))
self.pcn.inject( 1, midpoint )
# Set up layout for ROS message
dim = [ MultiArrayDimension( label="x", size=POSE_SIZE[0], stride=POSE_SIZE[0] * POSE_SIZE[1] * POSE_SIZE[2] ),
MultiArrayDimension( label="y", size=POSE_SIZE[1], stride=POSE_SIZE[1] * POSE_SIZE[2] ),
MultiArrayDimension( label="th", size=POSE_SIZE[2], stride=POSE_SIZE[2] ) ]
self.pc_layout = MultiArrayLayout(dim=dim)
# Initialize the View Templates
# TODO: put reasonable values here
self.vts = ViewTemplates( x_range=X_RANGE, y_range=Y_RANGE,
x_step=X_STEP, y_step=Y_STEP,
im_x=IM_SIZE[0], im_y=IM_SIZE[1],
match_threshold=MATCH_THRESHOLD)
self.vt_count = 0
self.bridge = CvBridge()
# Initialize the Experience Map
self.em = ExperienceMap()
self.em_count = 0
self.vis_odom_data = deque()
rospy.init_node( 'posecells', anonymous=True )
if not USE_VISUAL_ODOMETRY:
self.sub_odom = rospy.Subscriber( 'navbot/odom',Odometry,self.odom_callback )
self.sub_vis = rospy.Subscriber( 'navbot/camera/image',Image,self.vis_callback )
self.pub_pc = rospy.Publisher( 'navbot/posecells', numpy_msg(Float64MultiArray) )
self.pub_em = rospy.Publisher( 'navbot/experiencemap', Pose2D )
# Set up publisher for the template matches (sends just the index value of a match)
self.pub_tm = rospy.Publisher( 'navbot/templatematches', Int32 )
# Set up a visual odometer
self.vis_odom = SimpleVisualOdometer()
if RECORD_ROSBAG:
date = datetime.datetime.now()
self.bag = rosbag.Bag( '../testdata/Output-{0}-{1}-{2}-{3}.bag'.format( date.month, date.day,
date.hour, date.minute ),'w' )
class RatslamRos():
def __init__( self ):
self.im_count = 0
# Initialize the Pose Cell Network
self.pcn = PoseCellNetwork(shape=POSE_SIZE)
self.pc_count = 0
self.twist_data = deque()
self.odom_freq = ODOM_FREQ
midpoint = (math.floor(POSE_SIZE[0]/2),math.floor(POSE_SIZE[1]/2),math.floor(POSE_SIZE[2]/2))
self.pcn.inject( 1, midpoint )
# Set up layout for ROS message
dim = [ MultiArrayDimension( label="x", size=POSE_SIZE[0], stride=POSE_SIZE[0] * POSE_SIZE[1] * POSE_SIZE[2] ),
MultiArrayDimension( label="y", size=POSE_SIZE[1], stride=POSE_SIZE[1] * POSE_SIZE[2] ),
MultiArrayDimension( label="th", size=POSE_SIZE[2], stride=POSE_SIZE[2] ) ]
self.pc_layout = MultiArrayLayout(dim=dim)
# Initialize the View Templates
# TODO: put reasonable values here
self.vts = ViewTemplates( x_range=X_RANGE, y_range=Y_RANGE,
x_step=X_STEP, y_step=Y_STEP,
im_x=IM_SIZE[0], im_y=IM_SIZE[1],
match_threshold=MATCH_THRESHOLD)
self.vt_count = 0
self.bridge = CvBridge()
# Initialize the Experience Map
self.em = ExperienceMap()
self.em_count = 0
self.vis_odom_data = deque()
rospy.init_node( 'posecells', anonymous=True )
if not USE_VISUAL_ODOMETRY:
self.sub_odom = rospy.Subscriber( 'navbot/odom',Odometry,self.odom_callback )
self.sub_vis = rospy.Subscriber( 'navbot/camera/image',Image,self.vis_callback )
self.pub_pc = rospy.Publisher( 'navbot/posecells', numpy_msg(Float64MultiArray) )
self.pub_em = rospy.Publisher( 'navbot/experiencemap', Pose2D )
# Set up publisher for the template matches (sends just the index value of a match)
self.pub_tm = rospy.Publisher( 'navbot/templatematches', Int32 )
# Set up a visual odometer
self.vis_odom = SimpleVisualOdometer()
if RECORD_ROSBAG:
date = datetime.datetime.now()
self.bag = rosbag.Bag( '../testdata/Output-{0}-{1}-{2}-{3}.bag'.format( date.month, date.day,
date.hour, date.minute ),'w' )
# This is called whenever new visual information is received
def vis_callback( self, data ):
cv_im = self.bridge.imgmsg_to_cv( data, "mono8" )
im = asarray( cv_im )
pc_max = self.pcn.get_pc_max()
template_match = self.vts.match( input=im,pc_x=pc_max[0],pc_y=pc_max[1],pc_th=pc_max[2] )
index = template_match.get_index()
#TEMP - just inject with this template for now
#self.pcn.inject( .02, template_match.location() )
#self.pcn.inject( .2, template_match.location() )
# Send the template match index to the viewer
index_msg = Int32()
index_msg.data = index
self.pub_tm.publish( index_msg )
if RECORD_ROSBAG:
self.bag.write( 'navbot/camera/image', data )
self.bag.write( 'navbot/templatematches', index_msg )
# If using visual odometry, update the odometry information using this new image
if USE_VISUAL_ODOMETRY:
delta = self.vis_odom.update( im )
self.vis_odom_data.append( delta )
# This is called whenever new odometry information is received
def odom_callback( self, data ):
twist = data.twist.twist
# If there is barely any motion, don't bother flooding the queue with it
if abs(twist.linear.x) > 0.001 or abs(twist.angular.z) > 0.001:
self.twist_data.append(twist)
if RECORD_ROSBAG:
self.bag.write( 'navbot/odom', data )
def update_posecells( self, vtrans, vrot ):
self.pcn.update( ( vtrans, vrot ) )
pc_max = self.pcn.get_pc_max()
self.em.update( vtrans, vrot, pc_max )
#pc = sparse.csr_matrix( self.pcn.posecells ) # Create scipy sparse matrix
pc = self.pcn.posecells
em_msg = Pose2D()
em_msg.x, em_msg.y = self.em.get_current_point() # TODO: add theta as well
self.pub_pc.publish( self.pc_layout, pc.ravel() )
self.pub_em.publish( em_msg )
if RECORD_ROSBAG:
self.bag.write( 'navbot/experiencemap', em_msg )
self.bag.write( 'navbot/posecells', numpy_msg(self.pc_layout, pc.ravel()) )
def run( self ):
count = 0
start_time = time.time()
while not rospy.is_shutdown():
if self.twist_data:
twist = self.twist_data.popleft()
vtrans = twist.linear.x / self.odom_freq
vrot = twist.angular.z / self.odom_freq
self.update_posecells( vtrans, vrot )
count += 1
if self.vis_odom_data:
vtrans, vrot = self.vis_odom_data.popleft()
self.update_posecells( vtrans, vrot )
count += 1
class RatSLAM:
""" A complete rat slam instance
(Matlab version: equivalent to rs_main)
"""
def __init__(self, **kwargs):
pose_cell_shape = kwargs.pop("pose_cell_dimensions", (100,100,60))
shift_match = kwargs.pop("odo_shift_match", 20)
# self.odometer = SimpleVisualOdometer()
self.odometer = SimpleVisualOdometer(fov_deg = kwargs.pop("fov_deg",50),
im_trans_y_range = kwargs.pop("im_trans_y_range"),
im_rot_y_range = kwargs.pop("im_rot_y_range"),
im_odo_x_range = kwargs.pop("im_odo_x_range"),
shift_match = shift_match,
trans_scale = kwargs.pop("odo_trans_scale",100.))
# vt_match_threshold = kwargs.pop('vt_match_threshold', None)
# self.view_templates = VisualTemplateCollection()
self.view_templates = VisualTemplateCollection( template_decay = kwargs.pop('template_decay', 1.0),
match_y_range = kwargs.pop('im_vt_y_range', None),
match_x_range = kwargs.pop('im_vt_x_range', None),
shift_match = shift_match,
match_threshold = kwargs.pop('vt_match_threshold', None))
self.pose_cell_network = PoseCellNetwork(pose_cell_shape)#,
# self.view_templates)
self.experience_map = ExperienceMap(self.pose_cell_network,
self.view_templates,
delta_pc_threshold = kwargs.pop("exp_delta_pc_threshold",1.0),
correction_factor = kwargs.pop("exp_correction_factor",0.5),
exp_loops = kwargs.pop("exp_loops",100))
def update(self, video_frame):
# TODO save experience map
# convert video_frame to grayscale # done outside
# plt.imshow(video_frame)
# plt.show()
# get odometry from video
speed, rotation, odo = self.odometer.estimate_odometry(video_frame)
# get most active visual template (bassed on current video_frame)
vt_id = self.view_templates.match(video_frame, odo[0], odo[1], odo[2])
# update pose cells
self.pose_cell_network.update(self.view_templates.templates[vt_id], speed, rotation)
# get 'best' center of pose cell activity
px, py, pr = self.pose_cell_network.activty_packet_center()
# run an iteration fo the experience map
experience_map.update(vt_id, speed, rotation, (px, py, pr))
# store current odometry for next update
# pass
print self.experience_map.get_current_experience()