def __init__(self, params, sensor_type):
config_file = params["SLAM.settings_path"]
vocab_file = params["SLAM.vocab_path"]
# check the existence of the configuration file
if not os.path.exists(config_file):
raise FileNotFoundError(config_file + " not found")
if not os.path.exists(vocab_file):
raise FileNotFoundError(vocab_file + " not found")
# ORB_Slam2 don't have the imu interfaces so the STEREO_IMU and
# MONOCULAR_IMU are mapped in the STEREO and MONOCULAR sensor
if sensor_type == Sensor.MONOCULAR or sensor_type == Sensor.MONOCULAR_IMU:
print("the input sensor select is MONOCULAR")
self.slam = orbslam2.System(vocab_file, config_file,
orbslam2.Sensor.MONOCULAR)
self.sensor_type = Sensor.MONOCULAR
if sensor_type == Sensor.STEREO or sensor_type == Sensor.STEREO_IMU:
print("the input sensor select is STEREO")
self.slam = orbslam2.System(vocab_file, config_file,
orbslam2.Sensor.STEREO)
self.sensor_type = Sensor.STEREO
if sensor_type == Sensor.RGBD:
print("the input sensor select is RGBD")
self.slam = orbslam2.System(vocab_file, config_file,
orbslam2.Sensor.RGBD)
self.sensor_type = sensor_type
self.slam.set_use_viewer(False)
self.slam.initialize()
def main(args):
sequence = []
with open(os.path.join(args.seq, 'rgb_left.txt'),
'r') as f_left, open(os.path.join(args.seq, 'rgb_right.txt'),
'r') as f_right:
for line_left, line_right in zip(f_left, f_right):
line_left, line_right = line_left.strip(), line_right.strip()
if line_left.startswith('#'):
continue
ts, img_left_path = line_left.split()
_, img_right_path = line_right.split()
sequence.append((float(ts), os.path.join(args.seq, img_left_path),
os.path.join(args.seq, img_right_path)))
slam = orbslam2.System(args.vocab, args.config, orbslam2.Sensor.STEREO)
slam.set_use_viewer(True)
slam.initialize()
for ts, path_left, path_right in sequence:
img_left = cv2.imread(path_left)
img_right = cv2.imread(path_right)
slam.process_image_stereo(img_left, img_right, ts)
sleep(0.1)
save_trajectory(slam.get_trajectory_points(), 'trajectory.txt')
slam.shutdown()
def main(vocab_path, settings_path):
slam = orbslam2.System(vocab_path, settings_path,
orbslam2.Sensor.MONOCULAR)
slam.set_use_viewer(True)
slam.initialize()
camera = Camera.instance(width=960,
height=540,
capture_width=1280,
capture_height=720)
print('-----')
print('Start processing sequence ...')
ts = 0
#cv2.namedWindow("test")
while (True):
img = camera.value.copy()
t1 = time.time()
slam.process_image_mono(img, float(ts))
t2 = time.time()
ttrack = t2 - t1
print("frame id:" + str(ts), ttrack)
time.sleep(0.01)
ts += 1
save_trajectory(slam.get_trajectory_points(), 'trajectory.txt')
slam.shutdown()
return 0
def __init__(
self,
vocab_path="./configs/ORBvoc.txt",
slam_settings_path="./configs/terrasentia_orb.yaml",
visualize=False,
):
super().__init__("orbslam2")
self.slam = orbslam2.System(vocab_path, slam_settings_path,
orbslam2.Sensor.MONOCULAR)
self.slam.set_use_viewer(visualize)
self.slam.initialize()
self.slam.reset()
self.start_time = self.get_clock().now().to_msg()
# This makes it so NO data is dropped, a large buffer will occur
# This is necessary because the map building doesn't have to be
# meeting any safety critical times.
q = QoSProfile(history=2)
self.subscription = self.create_subscription(
Image,
"/terrasentia/usb_cam_node/image_raw",
self.update_internal_state,
qos_profile=q,
)
self.publisher = self.create_publisher(PoseStamped,
"pose",
qos_profile=q)
self.bridge = CvBridge()
self.get_logger().info("ORBSLAM2 succesfully started")
def main(args):
sequence = []
with open(os.path.join(args.seq, 'rgb_left.txt'),
'r') as f_left, open(os.path.join(args.seq, 'depth.txt'),
'r') as f_depth:
for line_left, line_depth in zip(f_left, f_depth):
line_left, line_depth = line_left.strip(), line_depth.strip()
if line_left.startswith('#'):
continue
ts, img_left_path = line_left.split()
_, depth_path = line_depth.split()
sequence.append((float(ts), os.path.join(args.seq, img_left_path),
os.path.join(args.seq, depth_path)))
slam = orbslam2.System(args.vocab, args.config, orbslam2.Sensor.RGBD)
slam.set_use_viewer(True)
slam.initialize()
for ts, path_left, path_depth in sequence:
img_left = cv2.imread(path_left)
depth = cv2.imread(path_depth, cv2.IMREAD_UNCHANGED)
slam.process_image_rgbd(img_left, depth, ts)
sleep(0.1)
save_trajectory(slam.get_trajectory_points(), 'trajectory.txt')
slam.shutdown()
def orbslam(imgs, timestamps, vocab, settings):
num_images = len(imgs)
slam = orbslam2.System(vocab, settings, orbslam2.Sensor.MONOCULAR)
slam.set_use_viewer(False)
slam.initialize()
times_track = [0 for _ in range(num_images)]
print('-----')
print('Start processing sequence ...')
print('Images in the sequence: {0}'.format(num_images))
for idx in tnrange(num_images):
image = cv2.imread(imgs[idx], cv2.IMREAD_UNCHANGED)
tframe = timestamps[idx]
if image is None:
print("failed to load image at {0}".format(imgs[idx]))
break
t1 = time.time()
slam.process_image_mono(image, tframe)
t2 = time.time()
ttrack = t2 - t1
times_track[idx] = ttrack
t = 0
if idx < num_images - 1:
t = timestamps[idx + 1] - tframe
elif idx > 0:
t = tframe - timestamps[idx - 1]
if ttrack < t:
time.sleep(t - ttrack)
times_track = sorted(times_track)
total_time = sum(times_track)
print('-----')
print('median tracking time: {0}'.format(times_track[num_images // 2]))
print('mean tracking time: {0}'.format(total_time / num_images))
tmp = np.expand_dims([0, 0, 0, 1], axis=0)
#convert pose and inverse pose into 4x4 matrices
pose = np.array(slam.get_keyframe_poses())
tframe = [t[0] for t in pose]
pose = [np.concatenate((f[1:].reshape(3, 4), tmp), axis=0) for f in pose]
inverse_pose = np.array(slam.get_inverse_keyframe_poses())
inverse_pose = [
np.concatenate((f[1:].reshape(3, 4), tmp), axis=0)
for f in inverse_pose
]
points = [np.array(frame) for frame in slam.get_keyframe_points()]
slam.shutdown()
return points, pose, inverse_pose, tframe
def __init__(
self,
config,
device=torch.device("cuda:0"),
monocheckpoint="habitat_baselines/slambased/data/mp3d_resnet50.pth",
):
super(ORBSLAM2MonodepthAgent, self).__init__(config)
self.num_actions = config.NUM_ACTIONS
self.dist_threshold_to_stop = config.DIST_TO_STOP
self.slam_vocab_path = config.SLAM_VOCAB_PATH
assert os.path.isfile(self.slam_vocab_path)
self.slam_settings_path = config.SLAM_SETTINGS_PATH
assert os.path.isfile(self.slam_settings_path)
self.slam = orbslam2.System(
self.slam_vocab_path, self.slam_settings_path, orbslam2.Sensor.RGBD
)
self.slam.set_use_viewer(False)
self.slam.initialize()
self.device = device
self.map_size_meters = config.MAP_SIZE
self.map_cell_size = config.MAP_CELL_SIZE
self.pos_th = config.DIST_REACHED_TH
self.next_wp_th = config.NEXT_WAYPOINT_TH
self.angle_th = config.ANGLE_TH
self.obstacle_th = config.MIN_PTS_IN_OBSTACLE
self.depth_denorm = config.DEPTH_DENORM
self.planned_waypoints = []
self.mapper = DirectDepthMapper(
camera_height=config.CAMERA_HEIGHT,
near_th=config.D_OBSTACLE_MIN,
far_th=config.D_OBSTACLE_MAX,
h_min=config.H_OBSTACLE_MIN,
h_max=config.H_OBSTACLE_MAX,
map_size=config.MAP_SIZE,
map_cell_size=config.MAP_CELL_SIZE,
device=device,
)
self.planner = DifferentiableStarPlanner(
max_steps=config.PLANNER_MAX_STEPS,
preprocess=config.PREPROCESS_MAP,
beta=config.BETA,
device=device,
)
self.slam_to_world = 1.0
self.timestep = 0.1
self.timing = False
self.checkpoint = monocheckpoint
if not os.path.isfile(self.checkpoint):
mp3d_url = "http://cmp.felk.cvut.cz/~mishkdmy/navigation/mp3d_ft_monodepth_resnet50.pth"
# suncg_me_url = "http://cmp.felk.cvut.cz/~mishkdmy/navigation/suncg_me_resnet.pth"
# suncg_mf_url = "http://cmp.felk.cvut.cz/~mishkdmy/navigation/suncg_mf_resnet.pth"
url = mp3d_url
print("No monodepth checkpoint found. Downloading...", url)
download(url, self.checkpoint)
self.monodepth = MonoDepthEstimator(self.checkpoint)
self.reset()
return
def Initialize_mapping_calibration(self,
disparity_bool=True,
slam_bool=False,
file_capture=False,
fisheye=True):
# Initialize the mapping and the disparity matcher, to be called once and outside the loop
#The mapping for the correction, careful always give the left image to the left parameters
# and the right to the right parameters
if fisheye:
self.map1l, self.map2l = cv.fisheye.initUndistortRectifyMap(
self.K_l, self.D_l, self.R_l, self.P_l, (self.w, self.h),
cv.CV_32FC1)
self.map1r, self.map2r = cv.fisheye.initUndistortRectifyMap(
self.K_r, self.D_r, self.R_r, self.P_r, (self.w, self.h),
cv.CV_32FC1)
else:
self.map1l, self.map2l = cv.initUndistortRectifyMap(
self.K_l, self.D_l, self.R_l, self.P_l, (self.w, self.h),
cv.CV_32FC1)
self.map1r, self.map2r = cv.initUndistortRectifyMap(
self.K_r, self.D_r, self.R_r, self.P_r, (self.w, self.h),
cv.CV_32FC1)
if disparity_bool:
# The accuracy and the range of the disparity depends on these parameters
self.left_matcher = cv.StereoSGBM_create(
minDisparity=-self._minDisparity,
numDisparities=16 *
self.a, # max_disp has to be dividable by 16 f. E. HH 192, 256
blockSize=self._blockSize,
P1=8 * 3 * self.window_size**
2, # wsize default 3; 5; 7 for SGBM reduced size image; 15 for SGBM full size image (1300px and above); 5 Works nicely
P2=32 * 3 * self.window_size**2,
disp12MaxDiff=self._disp12MaxDiff,
uniquenessRatio=self._uniquenessRatio,
speckleWindowSize=self._speckleWindowSize,
speckleRange=self._speckleRange,
preFilterCap=self._preFilterCap,
mode=cv.STEREO_SGBM_MODE_SGBM_3WAY)
#Though this seems unecessary, it lowers the computation of the disparity map
wls_filter = cv.ximgproc.createDisparityWLSFilter(
matcher_left=self.left_matcher)
if slam_bool:
vocab_path = "Parameters/ORBvoc.txt"
settings_path = "Parameters/EuRoC.yaml"
self.slam = orbslam2.System(vocab_path, settings_path,
orbslam2.Sensor.STEREO)
self.slam.set_use_viewer(True)
self.slam.initialize()
if file_capture:
self.f = open("Data.txt", "w+")
def main(vocab_path, settings_path, sequence_path, association_path):
rgb_filenames, depth_filenames, timestamps = load_images(association_path)
num_images = len(timestamps)
slam = orbslam2.System(vocab_path, settings_path, orbslam2.Sensor.RGBD)
slam.set_use_viewer(True)
slam.initialize()
times_track = [0 for _ in range(num_images)]
print('-----')
print('Start processing sequence ...')
print('Images in the sequence: {0}'.format(num_images))
for idx in range(num_images):
rgb_image = cv2.imread(os.path.join(sequence_path, rgb_filenames[idx]),
cv2.IMREAD_UNCHANGED)
depth_image = cv2.imread(
os.path.join(sequence_path, depth_filenames[idx]),
cv2.IMREAD_UNCHANGED)
tframe = timestamps[idx]
if rgb_image is None:
print("failed to load image at {0}".format(rgb_filenames[idx]))
return 1
if depth_image is None:
print("failed to depth at {0}".format(depth_filenames[idx]))
return 1
t1 = time.time()
slam.process_image_rgbd(rgb_image, depth_image, tframe)
t2 = time.time()
ttrack = t2 - t1
times_track[idx] = ttrack
t = 0
if idx < num_images - 1:
t = timestamps[idx + 1] - tframe
elif idx > 0:
t = tframe - timestamps[idx - 1]
if ttrack < t:
time.sleep(t - ttrack)
save_trajectory(slam.get_trajectory_points(), 'trajectory.txt')
slam.shutdown()
times_track = sorted(times_track)
total_time = sum(times_track)
print('-----')
print('median tracking time: {0}'.format(times_track[num_images // 2]))
print('mean tracking time: {0}'.format(total_time / num_images))
return 0
def main(vocab_path, settings_path, path_to_image_folder, path_to_times_file,
save_file='trajectory.txt', with_time=False):
image_filenames, timestamps = load_images(path_to_image_folder, path_to_times_file)
num_images = len(image_filenames)
slam = orbslam2.System(vocab_path, settings_path, orbslam2.Sensor.MONOCULAR)
slam.set_use_viewer(True)
slam.initialize()
times_track = [0 for _ in range(num_images)]
print('-----')
print('Start processing sequence ...')
print('Images in the sequence: {0}'.format(num_images))
for idx in range(num_images):
image = cv2.imread(image_filenames[idx], cv2.IMREAD_UNCHANGED)
tframe = timestamps[idx]
if image is None:
print("failed to load image at {0}".format(image_filenames[idx]))
return 1
t1 = time.time()
slam.process_image_mono(image, tframe)
t2 = time.time()
ttrack = t2 - t1
times_track[idx] = ttrack
t = 0
if idx < num_images - 1:
t = timestamps[idx + 1] - tframe
elif idx > 0:
t = tframe - timestamps[idx - 1]
if ttrack < t:
time.sleep(t - ttrack)
# save_trajectory(slam.get_trajectory_points(), 'trajectory.txt')
save_trajectory(slam.get_trajectory_points(), save_file, with_time=with_time)
slam.shutdown()
times_track = sorted(times_track)
total_time = sum(times_track)
print('-----')
print('median tracking time: {0}'.format(times_track[num_images // 2]))
print('mean tracking time: {0}'.format(total_time / num_images))
return 0
def __init__(self, config, device=torch.device("cuda:0")):
self.num_actions = config.NUM_ACTIONS
self.dist_threshold_to_stop = config.DIST_TO_STOP
self.slam_vocab_path = config.SLAM_VOCAB_PATH
assert os.path.isfile(self.slam_vocab_path)
self.slam_settings_path = config.SLAM_SETTINGS_PATH
assert os.path.isfile(self.slam_settings_path)
self.slam = orbslam2.System(self.slam_vocab_path,
self.slam_settings_path,
orbslam2.Sensor.RGBD)
self.slam.set_use_viewer(False)
self.slam.initialize()
self.device = device
self.map_size_meters = config.MAP_SIZE
self.map_cell_size = config.MAP_CELL_SIZE
self.pos_th = config.DIST_REACHED_TH
self.next_wp_th = config.NEXT_WAYPOINT_TH
self.angle_th = config.ANGLE_TH
self.obstacle_th = config.MIN_PTS_IN_OBSTACLE
self.depth_denorm = config.DEPTH_DENORM
self.planned_waypoints = []
self.mapper = DirectDepthMapper(
camera_height=config.CAMERA_HEIGHT,
near_th=config.D_OBSTACLE_MIN,
far_th=config.D_OBSTACLE_MAX,
h_min=config.H_OBSTACLE_MIN,
h_max=config.H_OBSTACLE_MAX,
map_size=config.MAP_SIZE,
map_cell_size=config.MAP_CELL_SIZE,
device=device,
)
self.planner = DifferentiableStarPlanner(
max_steps=config.PLANNER_MAX_STEPS,
preprocess=config.PREPROCESS_MAP,
beta=config.BETA,
device=device,
)
self.slam_to_world = 1.0
self.timestep = 0.1
self.timing = False
self.reset()
# print('self', self, dir(self))
for attr_name in dir(self):
if '__' not in attr_name:
print(attr_name, getattr(self, attr_name),
type(getattr(self, attr_name)))
type_name = str(type(getattr(self, attr_name)))
if 'numpy' in type_name or 'ensor' in type_name:
print(getattr(getattr(self, attr_name), 'shape'))
return
def __init__(self, fps=10., scale=20., height=40.,
online=False, settings_path='carla.yaml',
vocab_path='~/ORB_SLAM2/Vocabulary/ORBvoc.txt'):
self.slam = orbslam2.System(os.path.expanduser(vocab_path),
settings_path, orbslam2.Sensor.MONOCULAR)
self.slam.set_use_viewer(True)
self.slam.initialize()
self.fps = fps
self.scale = scale
self.height = height
self.prev_line = None
self.online = online
self.reset_between = False
self.num_train_images = 0
self.blank_image = np.zeros(0)
self.tframe = 0
self.coord_transform = lambda x: x
def main(args):
sequence = []
with open(os.path.join(args.seq, 'rgb_left.txt'), 'r') as f:
for line in f:
line = line.strip()
if line.startswith('#'):
continue
ts, img_path = line.split()
sequence.append((float(ts), os.path.join(args.seq, img_path)))
slam = orbslam2.System(args.vocab, args.config, orbslam2.Sensor.MONOCULAR)
slam.set_use_viewer(True)
slam.initialize()
for ts, path in sequence:
img = cv2.imread(path)
slam.process_image_mono(img, ts)
sleep(0.1)
save_trajectory(slam.get_trajectory_points(), 'trajectory.txt')
slam.shutdown()
def Initialize_Realsense(self, slam_bool=False, file_capture=False):
# Configure depth and color streams
self.pipeline = rs.pipeline()
config = rs.config()
config.enable_stream(rs.stream.depth, self.w, self.h, rs.format.z16, 60)
config.enable_stream(rs.stream.color, self.w, self.h, rs.format.bgr8, 60)
# Start streaming
self.pipeline.start(config)
align_to = rs.stream.color
self.align = rs.align(align_to)
if slam_bool:
vocab_path = "Parameters/ORBvoc.txt"
settings_path = "Parameters/Realsense.yaml"
self.slam = orbslam2.System(vocab_path, settings_path, orbslam2.Sensor.RGBD)
self.slam.set_use_viewer(True)
self.slam.initialize()
if file_capture:
self.f = open("Data.txt", "w+")
def __init__(self, config, follower, device=torch.device("cuda:0")):
self.slam_vocab_path = config.SLAM_VOCAB_PATH
assert os.path.isfile(self.slam_vocab_path)
self.slam_settings_path = config.SLAM_SETTINGS_PATH
assert os.path.isfile(self.slam_settings_path)
self.slam = orbslam2.System(self.slam_vocab_path,
self.slam_settings_path,
orbslam2.Sensor.RGBD)
self.slam.set_use_viewer(False)
self.slam.initialize()
self.device = device
self.map_size_meters = config.MAP_SIZE
self.map_cell_size = config.MAP_CELL_SIZE
self.obstacle_th = config.MIN_PTS_IN_OBSTACLE
self.depth_denorm = config.DEPTH_DENORM
self.mapper = DirectDepthMapper(
camera_height=config.CAMERA_HEIGHT,
near_th=config.D_OBSTACLE_MIN,
far_th=config.D_OBSTACLE_MAX,
h_min=config.H_OBSTACLE_MIN,
h_max=config.H_OBSTACLE_MAX,
map_size=config.MAP_SIZE,
map_cell_size=config.MAP_CELL_SIZE,
device=device,
)
self.slam_to_world = 1.0
self.timestep = 0.1
self.timing = False
self.follower = follower
self.reset()
return
def main(vocab_path, settings_path, sequence_path, map_file):
rgb_filenames, timestamps = load_images(sequence_path)
num_images = len(timestamps)
slam = orbslam2.System(vocab_path, settings_path, orbslam2.Sensor.MONOCULAR)
slam.set_use_viewer(True)
slam.set_map_file(map_file) # set mapfile
slam.initialize()
times_track = [0 for _ in range(num_images)]
print('-----')
print('Start processing sequence ...')
print('Images in the sequence: {0}'.format(num_images))
for idx in range(num_images):
image = cv2.imread(os.path.join(sequence_path, rgb_filenames[idx]), cv2.IMREAD_UNCHANGED)
tframe = timestamps[idx]
if image is None:
print("failed to load image at {0}".format(os.path.join(sequence_path, rgb_filenames[idx])))
return 1
t1 = time.time()
slam.process_image_mono(image, tframe)
t2 = time.time()
ttrack = t2 - t1
times_track[idx] = ttrack
t = 0
if idx < num_images - 1:
t = timestamps[idx + 1] - tframe
elif idx > 0:
t = tframe - timestamps[idx - 1]
if ttrack < t:
time.sleep(t - ttrack)
slam.save_map("saved_maps/map_1.bin")
slam.reset_and_load_map("saved_maps/map_1.bin")
for idx in range(num_images):
image = cv2.imread(os.path.join(sequence_path, rgb_filenames[idx]), cv2.IMREAD_UNCHANGED)
tframe = timestamps[idx]
if image is None:
print("failed to load image at {0}".format(os.path.join(sequence_path, rgb_filenames[idx])))
return 1
t1 = time.time()
slam.process_image_mono(image, tframe)
t2 = time.time()
print(slam.get_tracking_state())
ttrack = t2 - t1
times_track[idx] = ttrack
t = 0
if idx < num_images - 1:
t = timestamps[idx + 1] - tframe
elif idx > 0:
t = tframe - timestamps[idx - 1]
if ttrack < t:
time.sleep(t - ttrack)
slam.save_map("saved_maps/map_2.bin")
save_trajectory(slam.get_trajectory_points(), 'trajectory.txt')
slam.shutdown()
times_track = sorted(times_track)
total_time = sum(times_track)
print('-----')
print('median tracking time: {0}'.format(times_track[num_images // 2]))
print('mean tracking time: {0}'.format(total_time / num_images))
return 0
def __init__(self, tonic: Tonic, vocab_path, settings_path):
self.tonic = tonic
self.slam = orbslam2.System(vocab_path, settings_path,
orbslam2.Sensor.MONOCULAR)
self.prevtime = None
self.to_world = Pose3DtoWorld()
def main(vocab_path, settings_path, sequence_path, masktp):
rgb_filenames, timestamps = load_images(sequence_path)
num_images = len(timestamps)
slam = orbslam2.System(vocab_path, settings_path,
orbslam2.Sensor.MONOCULAR)
# slam.set_use_viewer(True)
slam.set_use_viewer(False)
slam.initialize()
times_track = [0 for _ in range(num_images)]
print('-----')
print('Start processing sequence ...')
print('Images in the sequence: {0}'.format(num_images))
for idx in range(num_images):
image = cv2.imread(os.path.join(sequence_path, rgb_filenames[idx]),
cv2.IMREAD_UNCHANGED)
image2 = seg_process(image, masktp)
'''
x,y = process_img(image,[640,360],device,model.cuda(), masktp)
z = img_as_ubyte(y)
z = cv2.resize(z, (image.shape[1],image.shape[0]))
gz = cv2.cvtColor(z, cv2.COLOR_BGR2GRAY)
image2 = cv2.bitwise_and(image, image, mask=gz)
cv2.imshow('img2',z)
print("hi2")
k = cv2.waitKey(1)
if k == 27:
exit(0)
'''
tframe = timestamps[idx]
if image is None:
print("failed to load image at {0}".format(
os.path.join(sequence_path, rgb_filenames[idx])))
return 1
t1 = time.time()
slam.process_image_mono(image, tframe)
t2 = time.time()
ttrack = t2 - t1
times_track[idx] = ttrack
t = 0
if idx < num_images - 1:
t = timestamps[idx + 1] - tframe
elif idx > 0:
t = tframe - timestamps[idx - 1]
if ttrack < t:
time.sleep((t - ttrack) * 0.0004)
save_trajectory(slam.get_trajectory_points(), 'trajectory.txt')
slam.shutdown()
times_track = sorted(times_track)
total_time = sum(times_track)
print('-----')
print('median tracking time: {0}'.format(times_track[num_images // 2]))
print('mean tracking time: {0}'.format(total_time / num_images))
return 0
type=str,
default=None,
help="path of the map to be loaded")
parser.add_argument('--start',
type=int,
default=0,
help="Number of starting frame")
parser.add_argument('--end',
type=int,
default=None,
help="Number of ending frame")
args = parser.parse_args()
vocab_path = args.vocab_path
settings_path = args.settings_path
data_tuple = read_folder_as_tuples(args.images_path)
slam = orbslam2.System(vocab_path, settings_path,
orbslam2.Sensor.MONOCULAR)
slam.set_use_viewer(True)
slam.initialize()
slam.osmap_init()
filenames, timestamps = load_images(args.images_path)
#slam.activate_localisation_only()
first_datapoint = data_tuple[args.start]
slam.process_image_mono(first_datapoint[0], first_datapoint[1])
new_ids = []
if args.end is None:
end = len(data_tuple) - 1
elif abs(args.start - args.end) < 10:
print("WARGNING!!! the number of usable frames is less than 10!")
else:
end = args.end
if args.load_map is not None:
def __init__(self, vocab_path, settings_path):
self.slam = orbslam2.System(vocab_path, settings_path,
orbslam2.Sensor.MONOCULAR)
def run_orbslam(output_queue, input_queue, vocab_file, settings_file, mode):
"""
Actually run the orbslam system. This is done in a separate process to isolate memory leaks,
and in case it crashes.
:param output_queue:
:param input_queue:
:param vocab_file:
:param settings_file:
:param mode:
:return:
"""
import orbslam2
import logging
import trials.slam.tracking_state
logging.getLogger(__name__).info("Starting ORBSLAM2...")
sensor_mode = orbslam2.Sensor.RGBD
if mode == SensorMode.MONOCULAR:
sensor_mode = orbslam2.Sensor.MONOCULAR
elif mode == SensorMode.STEREO:
sensor_mode = orbslam2.Sensor.STEREO
tracking_stats = {}
orbslam_system = orbslam2.System(vocab_file, settings_file, sensor_mode)
orbslam_system.set_use_viewer(False)
orbslam_system.initialize()
output_queue.put(
True
) # Tell the parent process we've set-up correctly and are ready to go.
logging.getLogger(__name__).info("ORBSLAM2 Ready.")
running = True
while running:
in_data = input_queue.get(block=True)
if isinstance(in_data, tuple) and len(in_data) == 3:
img1, img2, timestamp = in_data
if mode == SensorMode.MONOCULAR:
orbslam_system.process_image_mono(img1, timestamp)
elif mode == SensorMode.STEREO:
orbslam_system.process_image_stereo(img1, img2, timestamp)
elif mode == SensorMode.RGBD:
orbslam_system.process_image_rgbd(img1, img2, timestamp)
tracking_state = orbslam_system.get_tracking_state()
if (tracking_state == orbslam2.TrackingState.SYSTEM_NOT_READY
or tracking_state == orbslam2.TrackingState.NO_IMAGES_YET
or tracking_state
== orbslam2.TrackingState.NOT_INITIALIZED):
tracking_stats[
timestamp] = trials.slam.tracking_state.TrackingState.NOT_INITIALIZED
elif tracking_state == orbslam2.TrackingState.OK:
tracking_stats[
timestamp] = trials.slam.tracking_state.TrackingState.OK
else:
tracking_stats[
timestamp] = trials.slam.tracking_state.TrackingState.LOST
else:
# Non-matching input indicates the end of processing, stop the main loop
logging.getLogger(__name__).info(
"Got terminate input, finishing up and sending results.")
running = False
# send the final trajectory to the parent
output_queue.put((orbslam_system.get_trajectory_points(), tracking_stats))
# shut down the system. This is going to crash it, but that's ok, because it's a subprocess
orbslam_system.shutdown()
logging.getLogger(__name__).info("Finished running ORBSLAM2")
def main(vocab_path, settings_path, path_to_left_folder, path_to_right_folder,
path_to_times_file):
left_filenames, right_filenames, timestamps = load_images(
path_to_left_folder, path_to_right_folder, path_to_times_file)
# read stereo rectification parameters from settings
m1l, m2l, m1r, m2r = load_stereo_rectification(settings_path)
num_images = len(left_filenames)
slam = orbslam2.System(vocab_path, settings_path, orbslam2.Sensor.STEREO)
slam.set_use_viewer(True)
slam.initialize()
times_track = [0 for _ in range(num_images)]
print('-----')
print('Start processing sequence ...')
print('Images in the sequence: {0}'.format(num_images))
for idx in range(num_images):
left_image = cv2.imread(left_filenames[idx], cv2.IMREAD_UNCHANGED)
right_image = cv2.imread(right_filenames[idx], cv2.IMREAD_UNCHANGED)
tframe = timestamps[idx]
if left_image is None:
print("failed to load image at {0}".format(left_filenames[idx]))
return 1
if right_image is None:
print("failed to load image at {0}".format(right_filenames[idx]))
return 1
left_image = cv2.remap(left_image, m1l, m2l, cv2.INTER_LINEAR)
right_image = cv2.remap(right_image, m1r, m2r, cv2.INTER_LINEAR)
t1 = time.time()
slam.process_image_stereo(left_image, right_image, tframe)
t2 = time.time()
ttrack = t2 - t1
times_track[idx] = ttrack
t = 0
if idx < num_images - 1:
t = timestamps[idx + 1] - tframe
elif idx > 0:
t = tframe - timestamps[idx - 1]
if ttrack < t:
time.sleep(t - ttrack)
save_trajectory(slam.get_trajectory_points(), 'trajectory.txt')
slam.shutdown()
times_track = sorted(times_track)
total_time = sum(times_track)
print('-----')
print('median tracking time: {0}'.format(times_track[num_images // 2]))
print('mean tracking time: {0}'.format(total_time / num_images))
return 0
def main(vocab_path, settings_path, sequence_path, map_file):
rgb_filenames, timestamps = load_images(sequence_path)
num_images = len(timestamps)
slam = orbslam2.System(vocab_path, settings_path,
orbslam2.Sensor.MONOCULAR)
slam.set_use_viewer(True)
slam.initialize()
times_track = [0 for _ in range(num_images)]
print('-----')
print('Start processing sequence ...')
print('Images in the sequence: {0}'.format(num_images))
iterations = 0
while iterations < 2:
slam.reset()
frame_count = 0
for idx in range(num_images):
image = cv2.imread(os.path.join(sequence_path, rgb_filenames[idx]),
cv2.IMREAD_UNCHANGED)
tframe = timestamps[idx]
if image is None:
print("failed to load image at {0}".format(
os.path.join(sequence_path, rgb_filenames[idx])))
return 1
t1 = time.time()
slam.process_image_mono(image, tframe)
t2 = time.time()
ttrack = t2 - t1
times_track[idx] = ttrack
t = 0
if idx < num_images - 1:
t = timestamps[idx + 1] - tframe
elif idx > 0:
t = tframe - timestamps[idx - 1]
if ttrack < t:
time.sleep(t - ttrack)
print(frame_count, slam.get_tracking_state())
if slam.get_tracking_state() == orbslam2.TrackingState.OK:
break
frame_count += 1
iterations += 1
save_trajectory(slam.get_trajectory_points(), 'trajectory.txt')
save_trajectory(slam.get_keyframe_points(),
'keyframe.txt') # actual keyframes
points = slam.get_map_points()
print(points)
slam.shutdown()
times_track = sorted(times_track)
total_time = sum(times_track)
print('-----')
print('median tracking time: {0}'.format(times_track[num_images // 2]))
print('mean tracking time: {0}'.format(total_time / num_images))
return 0