def process_sequence_properties(import_path,
cutoff_distance=600.0,
cutoff_time=60.0,
interpolate_directions=False,
keep_duplicates=False,
duplicate_distance=0.1,
duplicate_angle=5,
offset_angle=0.0,
verbose=False,
rerun=False,
skip_subfolders=False,
video_import_path=None):
# sanity check if video file is passed
if video_import_path and not os.path.isdir(video_import_path) and not os.path.isfile(video_import_path):
print("Error, video path " + video_import_path +
" does not exist, exiting...")
sys.exit(1)
# in case of video processing, adjust the import path
if video_import_path:
# set sampling path
video_sampling_path = "mapillary_sampled_video_frames"
video_dirname = video_import_path if os.path.isdir(
video_import_path) else os.path.dirname(video_import_path)
import_path = os.path.join(os.path.abspath(import_path), video_sampling_path) if import_path else os.path.join(
os.path.abspath(video_dirname), video_sampling_path)
# basic check for all
if not import_path or not os.path.isdir(import_path):
print_error("Error, import directory " + import_path +
" does not exist, exiting...")
sys.exit(1)
sequences = []
if skip_subfolders:
process_file_list = processing.get_process_file_list(import_path,
"sequence_process",
rerun,
verbose,
True,
import_path)
if not len(process_file_list):
if verbose:
print("No images to run sequence process in root " + import_path)
print(
"If the images have already been processed and not yet uploaded, they can be processed again, by passing the argument --rerun")
else:
# LOAD TIME AND GPS POINTS ------------------------------------
file_list, capture_times, lats, lons, directions = processing.load_geotag_points(
process_file_list, verbose)
# ---------------------------------------
# SPLIT SEQUENCES --------------------------------------
if len(capture_times) and len(lats) and len(lons):
sequences.extend(processing.split_sequences(
capture_times, lats, lons, file_list, directions, cutoff_time, cutoff_distance, verbose))
# ---------------------------------------
else:
# sequence limited to the root of the files
for root, dirs, files in os.walk(import_path):
if os.path.join(".mapillary", "logs") in root:
continue
if len(files):
process_file_list = processing.get_process_file_list(import_path,
"sequence_process",
rerun,
verbose,
True,
root)
if not len(process_file_list):
if verbose:
print("No images to run sequence process in root " + root)
print(
"If the images have already been processed and not yet uploaded, they can be processed again, by passing the argument --rerun")
continue
# LOAD TIME AND GPS POINTS ------------------------------------
file_list, capture_times, lats, lons, directions = processing.load_geotag_points(
process_file_list, verbose)
# ---------------------------------------
# SPLIT SEQUENCES --------------------------------------
if len(capture_times) and len(lats) and len(lons):
sequences.extend(processing.split_sequences(
capture_times, lats, lons, file_list, directions, cutoff_time, cutoff_distance, verbose))
# ---------------------------------------
if not keep_duplicates:
if verbose:
print("Flagging images as duplicates if consecutive distance difference less than {} and angle difference less than {}".format(
duplicate_distance, duplicate_angle))
# process for each sequence
for sequence in sequences:
file_list = sequence["file_list"]
directions = sequence["directions"]
latlons = sequence["latlons"]
capture_times = sequence["capture_times"]
# COMPUTE DIRECTIONS --------------------------------------
interpolated_directions = [compute_bearing(ll1[0], ll1[1], ll2[0], ll2[1])
for ll1, ll2 in zip(latlons[:-1], latlons[1:])]
if len(interpolated_directions):
interpolated_directions.append(interpolated_directions[-1])
else:
interpolated_directions.append(directions[-1])
# use interpolated directions if direction not available or if flag for
# interpolate_directions
for i, d in enumerate(directions):
directions[i] = d if (
d is not None and not interpolate_directions) else (interpolated_directions[i] + offset_angle) % 360.0
# ---------------------------------------
# COMPUTE SPEED -------------------------------------------
computed_delta_ts = [(t1 - t0).total_seconds()
for t0, t1 in zip(capture_times[:-1], capture_times[1:])]
computed_distances = [gps_distance(l1, l0)
for l0, l1 in zip(latlons[:-1], latlons[1:])]
computed_speed = gps_speed(
computed_distances, computed_delta_ts) # in meters/second
if len([x for x in computed_speed if x > MAX_CAPTURE_SPEED]) > 0:
print("Warning: The distance in sequence including images\n{}\nto\n{}\nis too large for the time difference (very high apparent capture speed). Are you sure timestamps and locations are correct?".format(
file_list[0], file_list[-1]))
# INTERPOLATE TIMESTAMPS, in case of identical timestamps
capture_times = processing.interpolate_timestamp(capture_times)
final_file_list = file_list[:]
final_directions = directions[:]
final_capture_times = capture_times[:]
# FLAG DUPLICATES --------------------------------------
if not keep_duplicates:
final_file_list = [file_list[0]]
final_directions = [directions[0]]
final_capture_times = [capture_times[0]]
prev_latlon = latlons[0]
prev_direction = directions[0]
for i, filename in enumerate(file_list[1:]):
log_root = uploader.log_rootpath(filename)
duplicate_flag_path = os.path.join(log_root,
"duplicate")
sequence_process_success_path = os.path.join(log_root,
"sequence_process_success")
k = i + 1
distance = gps_distance(latlons[k],
prev_latlon)
if directions[k] is not None and prev_direction is not None:
direction_diff = diff_bearing(directions[k],
prev_direction)
else:
# dont use bearing difference if no bearings are
# available
direction_diff = 360
if distance < duplicate_distance and direction_diff < duplicate_angle:
open(duplicate_flag_path, "w").close()
open(sequence_process_success_path, "w").close()
open(sequence_process_success_path + "_" +
str(time.strftime("%Y_%m_%d_%H_%M_%S", time.gmtime())), "w").close()
else:
prev_latlon = latlons[k]
prev_direction = directions[k]
final_file_list.append(filename)
final_directions.append(directions[k])
final_capture_times.append(capture_times[k])
# ---------------------------------------
# FINALIZE ------------------------------------
for i in range(0, len(final_file_list), MAX_SEQUENCE_LENGTH):
finalize_sequence_processing(str(uuid.uuid4()),
final_file_list[i:i +
MAX_SEQUENCE_LENGTH],
final_directions[i:i +
MAX_SEQUENCE_LENGTH],
final_capture_times[i:i +
MAX_SEQUENCE_LENGTH],
import_path,
verbose)
print("Sub process ended")
def process_sequence_properties(import_path,
cutoff_distance=600.0,
cutoff_time=60.0,
interpolate_directions=False,
flag_duplicates=False,
duplicate_distance=0.1,
duplicate_angle=5,
offset_angle=0.0,
verbose=False,
rerun=False,
skip_subfolders=False,
video_import_path=None):
# sanity check if video file is passed
if video_import_path and not os.path.isdir(video_import_path):
print("Error, video path " + video_import_path +
" does not exist, exiting...")
sys.exit(1)
# in case of video processing, adjust the import path
if video_import_path:
# set sampling path
video_sampling_path = "mapillary_sampled_video_frames"
import_path = os.path.join(
os.path.abspath(import_path),
video_sampling_path) if import_path else os.path.join(
os.path.abspath(video_import_path), video_sampling_path)
# basic check for all
if not import_path or not os.path.isdir(import_path):
print("Error, import directory " + import_path +
" does not exist, exiting...")
sys.exit(1)
sequences = []
if skip_subfolders:
process_file_list = processing.get_process_file_list(
import_path, "sequence_process", rerun, verbose, True, import_path)
if not len(process_file_list):
if verbose:
print("No images to run sequence process in root " +
import_path)
print(
"If the images have already been processed and not yet uploaded, they can be processed again, by passing the argument --rerun"
)
else:
# LOAD TIME AND GPS POINTS ------------------------------------
file_list, capture_times, lats, lons, directions = processing.load_geotag_points(
process_file_list, verbose)
# ---------------------------------------
# SPLIT SEQUENCES --------------------------------------
if len(capture_times) and len(lats) and len(lons):
sequences.extend(
processing.split_sequences(capture_times, lats, lons,
file_list, directions,
cutoff_time, cutoff_distance,
verbose))
# ---------------------------------------
else:
# sequence limited to the root of the files
for root, dirs, files in os.walk(import_path):
if os.path.join(".mapillary", "logs") in root:
continue
if len(files):
process_file_list = processing.get_process_file_list(
import_path, "sequence_process", rerun, verbose, True,
root)
if not len(process_file_list):
if verbose:
print("No images to run sequence process in root " +
root)
print(
"If the images have already been processed and not yet uploaded, they can be processed again, by passing the argument --rerun"
)
continue
# LOAD TIME AND GPS POINTS ------------------------------------
file_list, capture_times, lats, lons, directions = processing.load_geotag_points(
process_file_list, verbose)
# ---------------------------------------
# SPLIT SEQUENCES --------------------------------------
if len(capture_times) and len(lats) and len(lons):
sequences.extend(
processing.split_sequences(capture_times, lats, lons,
file_list, directions,
cutoff_time,
cutoff_distance, verbose))
# ---------------------------------------
if flag_duplicates:
if verbose:
print(
"Flagging images as duplicates if consecutive distance difference less than {} and angle difference less than {}"
.format(duplicate_distance, duplicate_angle))
# process for each sequence
for sequence in sequences:
file_list = sequence["file_list"]
directions = sequence["directions"]
latlons = sequence["latlons"]
capture_times = sequence["capture_times"]
# COMPUTE DIRECTIONS --------------------------------------
interpolated_directions = [
compute_bearing(ll1[0], ll1[1], ll2[0], ll2[1])
for ll1, ll2 in zip(latlons[:-1], latlons[1:])
]
if len(interpolated_directions):
interpolated_directions.append(interpolated_directions[-1])
else:
interpolated_directions.append(directions[-1])
# use interpolated directions if direction not available or if flag for
# interpolate_directions
for i, d in enumerate(directions):
directions[i] = d if (
d is not None and not interpolate_directions
) else (interpolated_directions[i] + offset_angle) % 360.0
# ---------------------------------------
# INTERPOLATE TIMESTAMPS, in case of identical timestamps
capture_times = processing.interpolate_timestamp(capture_times)
final_file_list = file_list[:]
final_directions = directions[:]
final_capture_times = capture_times[:]
# FLAG DUPLICATES --------------------------------------
if flag_duplicates:
final_file_list = [file_list[0]]
final_directions = [directions[0]]
final_capture_times = [capture_times[0]]
prev_latlon = latlons[0]
prev_direction = directions[0]
for i, filename in enumerate(file_list[1:]):
log_root = uploader.log_rootpath(filename)
duplicate_flag_path = os.path.join(log_root, "duplicate")
sequence_process_success_path = os.path.join(
log_root, "sequence_process_success")
k = i + 1
distance = gps_distance(latlons[k], prev_latlon)
if directions[k] is not None and prev_direction is not None:
direction_diff = diff_bearing(directions[k],
prev_direction)
else:
# dont use bearing difference if no bearings are
# available
direction_diff = 360
if distance < duplicate_distance and direction_diff < duplicate_angle:
open(duplicate_flag_path, "w").close()
open(sequence_process_success_path, "w").close()
open(
sequence_process_success_path + "_" +
str(time.strftime("%Y_%m_%d_%H_%M_%S", time.gmtime())),
"w").close()
else:
prev_latlon = latlons[k]
prev_direction = directions[k]
final_file_list.append(filename)
final_directions.append(directions[k])
final_capture_times.append(capture_times[k])
# ---------------------------------------
# FINALIZE ------------------------------------
for i in range(0, len(final_file_list), MAX_SEQUENCE_LENGTH):
finalize_sequence_processing(
str(uuid.uuid4()), final_file_list[i:i + MAX_SEQUENCE_LENGTH],
final_directions[i:i + MAX_SEQUENCE_LENGTH],
final_capture_times[i:i + MAX_SEQUENCE_LENGTH], import_path,
verbose)
print("Sub process ended")
def space_distance(a, b):
return geo.gps_distance(a[1:3], b[1:3])
def split_sequences(capture_times, lats, lons, file_list, directions, cutoff_time, cutoff_distance, verbose=False):
sequences = []
# sort based on time
sort_by_time = zip(capture_times,
file_list,
lats,
lons,
directions)
sort_by_time.sort()
capture_times, file_list, lats, lons, directions = [
list(x) for x in zip(*sort_by_time)]
latlons = zip(lats,
lons)
# initialize first sequence
sequence_index = 0
sequences.append({"file_list": [
file_list[0]], "directions": [directions[0]], "latlons": [latlons[0]], "capture_times": [capture_times[0]]})
if len(file_list) >= 1:
# diff in capture time
capture_deltas = [
t2 - t1 for t1, t2 in zip(capture_times, capture_times[1:])]
# distance between consecutive images
distances = [gps_distance(ll1, ll2)
for ll1, ll2 in zip(latlons, latlons[1:])]
# if cutoff time is given use that, else assume cutoff is
# 1.5x median time delta
if cutoff_time is None:
if verbose:
print(
"Warning, sequence cut-off time is None and will therefore be derived based on the median time delta between the consecutive images.")
median = sorted(capture_deltas)[
len(capture_deltas) // 2]
if type(median) is not int:
median = median.total_seconds()
cutoff_time = 1.5 * median
else:
cutoff_time = float(cutoff_time)
cut = 0
for i, filepath in enumerate(file_list[1:]):
cut_time = capture_deltas[i].total_seconds(
) > cutoff_time
cut_distance = distances[i] > cutoff_distance
if cut_time or cut_distance:
cut += 1
# delta too big, start new sequence
sequence_index += 1
sequences.append({"file_list": [
filepath], "directions": [directions[1:][i]], "latlons": [latlons[1:][i]], "capture_times": [capture_times[1:][i]]})
if verbose:
if cut_distance:
print('Cut {}: Delta in distance {} meters is bigger than cutoff_distance {} meters at {}'.format(
cut, distances[i], cutoff_distance, file_list[i + 1]))
elif cut_time:
print('Cut {}: Delta in time {} seconds is bigger then cutoff_time {} seconds at {}'.format(
cut, capture_deltas[i].total_seconds(), cutoff_time, file_list[i + 1]))
else:
# delta not too big, continue with current
# group
sequences[sequence_index]["file_list"].append(
filepath)
sequences[sequence_index]["directions"].append(
directions[1:][i])
sequences[sequence_index]["latlons"].append(
latlons[1:][i])
sequences[sequence_index]["capture_times"].append(
capture_times[1:][i])
return sequences
def process_sequence_properties(import_path,
cutoff_distance=600.0,
cutoff_time=60.0,
interpolate_directions=False,
flag_duplicates=False,
duplicate_distance=0.1,
duplicate_angle=5,
offset_angle=0.0,
verbose=False,
rerun=False,
skip_subfolders=False):
# basic check for all
import_path = os.path.abspath(import_path)
if not os.path.isdir(import_path):
print("Error, import directory " + import_path +
" doesnt not exist, exiting...")
sys.exit()
sequences = []
if skip_subfolders:
process_file_list = processing.get_process_file_list(import_path,
"sequence_process",
rerun,
verbose,
True,
import_path)
if not len(process_file_list):
if verbose:
print("No images to run sequence process in root " + import_path)
print(
"If the images have already been processed and not yet uploaded, they can be processed again, by passing the argument --rerun")
else:
# LOAD TIME AND GPS POINTS ------------------------------------
file_list, capture_times, lats, lons, directions = processing.load_geotag_points(
process_file_list, import_path, verbose)
# ---------------------------------------
# SPLIT SEQUENCES --------------------------------------
if len(capture_times) and len(lats) and len(lons):
sequences.extend(processing.split_sequences(
capture_times, lats, lons, file_list, directions, cutoff_time, cutoff_distance, verbose))
# ---------------------------------------
else:
# sequence limited to the root of the files
for root, dirs, files in os.walk(import_path):
if ".mapillary" in root:
continue
if len(files):
process_file_list = processing.get_process_file_list(import_path,
"sequence_process",
rerun,
verbose,
True,
root)
if not len(process_file_list):
if verbose:
print("No images to run sequence process in root " + root)
print(
"If the images have already been processed and not yet uploaded, they can be processed again, by passing the argument --rerun")
continue
# LOAD TIME AND GPS POINTS ------------------------------------
file_list, capture_times, lats, lons, directions = processing.load_geotag_points(
process_file_list, import_path, verbose)
# ---------------------------------------
# SPLIT SEQUENCES --------------------------------------
if len(capture_times) and len(lats) and len(lons):
sequences.extend(processing.split_sequences(
capture_times, lats, lons, file_list, directions, cutoff_time, cutoff_distance, verbose))
# ---------------------------------------
# process for each sequence
for sequence in sequences:
file_list = sequence["file_list"]
directions = sequence["directions"]
latlons = sequence["latlons"]
capture_times = sequence["capture_times"]
# COMPUTE DIRECTIONS --------------------------------------
interpolated_directions = [compute_bearing(ll1[0], ll1[1], ll2[0], ll2[1])
for ll1, ll2 in zip(latlons, latlons[1:])]
interpolated_directions.append(directions[-1])
# use interpolated directions if direction not available or if flag for
# interpolate_directions
for i, d in enumerate(directions):
directions[i] = d if (
d is not None and not interpolate_directions) else (interpolated_directions[i] + offset_angle) % 360.0
# ---------------------------------------
# INTERPOLATE TIMESTAMPS, incase of identical timestamps
capture_times, file_list = processing.interpolate_timestamp(capture_times,
file_list)
final_file_list = file_list[:]
final_directions = directions[:]
final_capture_times = capture_times[:]
# FLAG DUPLICATES --------------------------------------
if flag_duplicates:
final_file_list = [file_list[0]]
final_directions = [directions[0]]
prev_latlon = latlons[0]
prev_direction = directions[0]
for i, filename in enumerate(file_list[1:]):
log_root = uploader.log_rootpath(import_path,
filename)
duplicate_flag_path = os.path.join(log_root,
"duplicate")
sequence_process_success_path = os.path.join(log_root,
"sequence_process_success")
k = i + 1
distance = gps_distance(latlons[k],
prev_latlon)
if directions[k] is not None and prev_direction is not None:
direction_diff = diff_bearing(directions[k],
prev_direction)
else:
# dont use bearing difference if no bearings are
# available
direction_diff = 360
if distance < duplicate_distance and direction_diff < duplicate_angle:
open(duplicate_flag_path, "w").close()
open(sequence_process_success_path, "w").close()
open(sequence_process_success_path + "_" +
str(time.strftime("%Y_%m_%d_%H_%M_%S", time.gmtime())), "w").close()
else:
prev_latlon = latlons[k]
prev_direction = directions[k]
final_file_list.append(filename)
final_directions.append(directions[k])
# ---------------------------------------
# FINALIZE ------------------------------------
for i in range(0, len(final_file_list), MAX_SEQUENCE_LENGTH):
finalize_sequence_processing(str(uuid.uuid4()),
final_file_list[i:i +
MAX_SEQUENCE_LENGTH],
final_directions[i:i +
MAX_SEQUENCE_LENGTH],
final_capture_times[i:i +
MAX_SEQUENCE_LENGTH],
import_path,
verbose)
def split_sequences(capture_times, lats, lons, file_list, directions, cutoff_time, cutoff_distance, verbose=False):
sequences = []
# sort based on time
sort_by_time = zip(capture_times,
file_list,
lats,
lons,
directions)
sort_by_time.sort()
capture_times, file_list, lats, lons, directions = [
list(x) for x in zip(*sort_by_time)]
latlons = zip(lats,
lons)
# initialize first sequence
sequence_index = 0
sequences.append({"file_list": [
file_list[0]], "directions": [directions[0]], "latlons": [latlons[0]], "capture_times": [capture_times[0]]})
if len(file_list) >= 1:
# diff in capture time
capture_deltas = [
t2 - t1 for t1, t2 in zip(capture_times, capture_times[1:])]
# distance between consecutive images
distances = [gps_distance(ll1, ll2)
for ll1, ll2 in zip(latlons, latlons[1:])]
# if cutoff time is given use that, else assume cutoff is
# 1.5x median time delta
if cutoff_time is None:
if verbose:
print(
"Warning, sequence cut-off time is None and will therefore be derived based on the median time delta between the consecutive images.")
median = sorted(capture_deltas)[
len(capture_deltas) // 2]
if type(median) is not int:
median = median.total_seconds()
cutoff_time = 1.5 * median
else:
cutoff_time = float(cutoff_time)
cut = 0
for i, filepath in enumerate(file_list[1:]):
cut_time = capture_deltas[i].total_seconds(
) > cutoff_time
cut_distance = distances[i] > cutoff_distance
if cut_time or cut_distance:
cut += 1
# delta too big, start new sequence
sequence_index += 1
sequences.append({"file_list": [
filepath], "directions": [directions[1:][i]], "latlons": [latlons[1:][i]], "capture_times": [capture_times[1:][i]]})
if verbose:
if cut_distance:
print('Cut {}: Delta in distance {} meters is bigger than cutoff_distance {} meters at {}'.format(
cut, distances[i], cutoff_distance, file_list[i + 1]))
elif cut_time:
print('Cut {}: Delta in time {} seconds is bigger then cutoff_time {} seconds at {}'.format(
cut, capture_deltas[i].total_seconds(), cutoff_time, file_list[i + 1]))
else:
# delta not too big, continue with current
# group
sequences[sequence_index]["file_list"].append(
filepath)
sequences[sequence_index]["directions"].append(
directions[1:][i])
sequences[sequence_index]["latlons"].append(
latlons[1:][i])
sequences[sequence_index]["capture_times"].append(
capture_times[1:][i])
return sequences
def space_distance(a, b):
return geo.gps_distance(a[1:3], b[1:3])
