def _point_edge_dist(self, lon, lat, edge):
lat_origin = angle2radian(self.rd_nwk.nodes[edge[0]]['lat'])
lon_origin = angle2radian(self.rd_nwk.nodes[edge[0]]['lon'])
lat_dest = angle2radian(self.rd_nwk.nodes[edge[1]]['lat'])
lon_dest = angle2radian(self.rd_nwk.nodes[edge[1]]['lon'])
a = dist(angle2radian(lat), angle2radian(lon), lat_origin, lon_origin)
b = dist(angle2radian(lat), angle2radian(lon), lat_dest, lon_dest)
c = dist(lat_origin, lon_origin, lat_dest, lon_dest)
# if origin point is the closest
if b**2 > a**2 + c**2:
return a, edge[0] # distance, point
# if destination point is the closest
elif a**2 > b**2 + c**2:
return b, edge[1]
if c == 0:
return a, edge[0]
# otherwise, calculate the Vertical length
p = (a + b + c) / 2
s = math.sqrt(p * math.fabs(p - a) * math.fabs(p - b) *
math.fabs(p - c))
return 2 * s / c, None
def _preprocess(self):
"""
removing points that are within 2 sigma_z of the previous included point
The justification for this step is that until we see a point that is at least 2sigma_z away from
its temporal predecessor, our confidence is low that the apparent movement is due to
actual vehicle movement and not noise
Returns:
"""
new_traj = list()
lat_r_pre = None # latitude in radius
lon_r_pre = None # longitude in radius
for i in range(len(self.trajectory)):
# dist have lat before lon, however dataset provide lon before lat.
lat_r, lon_r = angle2radian(self.trajectory[i][2]), angle2radian(
self.trajectory[i][1])
if lat_r_pre is None or dist(lat_r, lon_r, lat_r_pre,
lon_r_pre) > 2 * self.sigma_z:
# if distance > 2*sigma_z, take this point into calculation, otherwise remove it
new_traj.append(self.trajectory[i])
lat_r_pre = lat_r
lon_r_pre = lon_r
self.trajectory = np.vstack(new_traj)
def _load_geo_and_rel(self):
"""
加载.geo文件，格式[geo_id, type, coordinates, properties(若干列)]
加载.rel文件，格式[rel_id, type, origin_id, destination_id, properties(若干列)],
.rel文件用来表示路网数据
Returns:
self.rd_nwk: networkx.MultiDiGraph
"""
# init road network, which is the result of this function
self.rd_nwk = nx.DiGraph(name="road network")
# load geo and rel file
geofile = pd.read_csv(self.data_path + self.geo_file + '.geo')
relfile = pd.read_csv(self.data_path + self.rel_file + '.rel')
geo_num = geofile.shape[0]
# check type geo in rel file and LineString in geo file
if not ['geo'] == self.config['rel']['including_types']:
raise ValueError('.rel file should include geo type in Map Matching task!')
if not ['LineString'] == self.config['geo']['including_types']:
raise ValueError('.geo file should include LineString type in Map Matching task!')
# get properties
columns = relfile.columns.tolist()[4:]
# use UnionSet to get nodes
node_set = UnionSet(2 * geo_num)
for index, row in relfile.iterrows():
# origin and destination
from_id = int(row[2])
to_id = int(row[3])
node_set.union(from_id, to_id + geo_num)
# generate MultiDigraph
for index, row in geofile.iterrows():
geo_id = int(row['geo_id'])
coordinate = eval(row['coordinates'])
origin_node = node_set.find(geo_id + geo_num)
dest_node = node_set.find(geo_id)
if origin_node not in self.rd_nwk.nodes:
self.rd_nwk.add_node(origin_node, lon=coordinate[0][0], lat=coordinate[0][1])
if dest_node not in self.rd_nwk.nodes:
self.rd_nwk.add_node(dest_node, lon=coordinate[1][0], lat=coordinate[1][1])
# add edge
self.rd_nwk.add_edge(origin_node, dest_node)
feature_dct = dict()
for i, column in enumerate(columns):
feature_dct[column] = row[i + 4]
if 'distance' not in feature_dct.keys():
feature_dct['distance'] = dist(
angle2radian(self.rd_nwk.nodes[origin_node]['lat']),
angle2radian(self.rd_nwk.nodes[origin_node]['lon']),
angle2radian(self.rd_nwk.nodes[dest_node]['lat']),
angle2radian(self.rd_nwk.nodes[dest_node]['lon'])
)
feature_dct['geo_id'] = geo_id
self.rd_nwk.edges[origin_node, dest_node].update(feature_dct)
# logger
self._logger.info("Loaded file " + self.geo_file + '.geo' + ', num_nodes=' + str(geo_num))
self._logger.info("Loaded file " + self.rel_file + '.rel, num_roads=' + str(len(self.rd_nwk)))
def _score_matrix(self):
sub_static_matrices = list()
# for every GPS sample
for i in range(len(self.candidates)):
# current candidates: self.candidates[i]
# prev candidates: self.candidates[j]
# no current candidates
if i == 0:
continue
if len(self.candidates[i]) == 0:
sub_static_matrix = np.zeros((1, 1))
sub_static_matrices.append(sub_static_matrix)
continue
# if there are current candidates, find prev index j
j = i - 1
while j >= 0 and len(self.candidates[j]) == 0:
j -= 1
# j >= 0, calculate score of candidates of GPS sample i
if j < 0:
sub_static_matrix = np.zeros((1, 1))
sub_static_matrices.append(sub_static_matrix)
continue
sub_static_matrix = np.zeros((len(self.candidates[j].items()),
len(self.candidates[i].items())))
n = 0
for edge, dct in self.candidates[i].items():
m = 0
for edge_pre, dct_pre in self.candidates[j].items():
sub_static_matrix[m][n] = dct['N'] * dct_pre['V'][edge] * (
1 if 'T' not in dct_pre.keys() else dct_pre['T'][edge])
m += 1
n += 1
sub_static_matrices.append(sub_static_matrix)
traj_lon_lat = self.trajectory[:, 1:3]
self.traj_points = []
for i in range(len(traj_lon_lat)):
W = []
j = 0
while j < len(traj_lon_lat):
if i == j:
j += 1
continue
lon_i, lat_i = traj_lon_lat[i, :]
lon_j, lat_j = traj_lon_lat[j, :]
euclid_distance = dist(angle2radian(lat_i),
angle2radian(lon_i),
angle2radian(lat_j),
angle2radian(lon_j))
W.append(pow(math.e, -(euclid_distance**2 / self.beta**2)))
j += 1
fai = []
self.sub_static_matrices = sub_static_matrices
self.W = W
for k in range(len(W)):
fai.append(sub_static_matrices[k] * W[k])
traj_point = {
"time": self.trajectory[i][0],
"lon": self.trajectory[i][1],
"lat": self.trajectory[i][2],
"fai": fai,
"candidates": self.candidates[i]
}
self.traj_points.append(
traj_point) # Φ(fai) is the weighted score matrix
def _transmission_probability(self):
"""
Returns:
"""
i = 1
while i < len(self.candidates):
# j and k
j = i - 1
if len(self.candidates[i]) == 0:
k = i + 1
while len(
self.candidates[k]) == 0 and k < len(self.candidates):
k += 1
if k == len(self.candidates):
break
else:
k = i
d = dist(angle2radian(self.trajectory[j][2]),
angle2radian(self.trajectory[j][1]),
angle2radian(self.trajectory[k][2]),
angle2radian(self.trajectory[k][1]))
for edge_j, dct_j in self.candidates[j].items():
for edge_k, dct_k in self.candidates[k].items():
brng_jk = init_bearing(angle2radian(self.trajectory[j][2]),
angle2radian(self.trajectory[j][1]),
angle2radian(self.trajectory[k][2]),
angle2radian(self.trajectory[k][1]))
brng_edge_j = init_bearing(
angle2radian(self.rd_nwk.nodes[edge_j[0]]['lat']),
angle2radian(self.rd_nwk.nodes[edge_j[0]]['lon']),
angle2radian(self.rd_nwk.nodes[edge_j[1]]['lat']),
angle2radian(self.rd_nwk.nodes[edge_j[1]]['lon']),
)
try:
if dct_j['node'] is not None and dct_k[
'node'] is not None:
result = d / nx.astar_path_length(
self.rd_nwk,
dct_j['node'],
dct_k['node'],
weight='distance')
elif dct_j['node'] is not None:
nd2_origin = edge_k[0]
lon, lat = self.rd_nwk.nodes[nd2_origin][
'lon'], self.rd_nwk.nodes[nd2_origin]['lat']
path_len = nx.astar_path_length(self.rd_nwk,
dct_j['node'],
nd2_origin,
weight='distance')
path_len += math.sqrt(
math.fabs(
dist(angle2radian(self.trajectory[k][2]),
angle2radian(self.trajectory[k][1]),
angle2radian(lat), angle2radian(lon))
**2 - dct_k['distance']**2))
if edge_j[1] == dct_j['edge']:
path_len += self.rd_nwk[edge_j[0]][
edge_j[1]]['distance'] * 2
result = d / path_len
elif dct_k['node'] is not None:
nd1_destination = edge_j[1]
lon, lat = self.rd_nwk.nodes[nd1_destination][
'lon'], self.rd_nwk.nodes[nd1_destination][
'lat']
path_len = nx.astar_path_length(self.rd_nwk,
nd1_destination,
dct_k['node'],
weight='distance')
path_len += math.sqrt(
math.fabs(
dist(angle2radian(self.trajectory[j][2]),
angle2radian(self.trajectory[j][1]),
angle2radian(lat), angle2radian(lon))
**2 - dct_j['distance']**2))
if edge_k[1] == dct_k['node']:
path_len += self.rd_nwk[edge_k[0]][
edge_k[1]]['distance'] * 2
result = d / path_len
else:
if edge_j == edge_k and math.fabs(brng_edge_j -
brng_jk) < 90:
result = 1
else:
nd1_destination = edge_j[1]
lon1, lat1 = self.rd_nwk.nodes[nd1_destination]['lon'], \
self.rd_nwk.nodes[nd1_destination]['lat']
nd2_origin = edge_k[0]
lon2, lat2 = self.rd_nwk.nodes[nd2_origin][
'lon'], self.rd_nwk.nodes[nd2_origin][
'lat']
result = d / (nx.astar_path_length(
self.rd_nwk,
nd1_destination,
nd2_origin,
weight='distance'
) + math.sqrt(
math.fabs(
dist(
angle2radian(self.trajectory[j][2]),
angle2radian(self.trajectory[j][1]
), angle2radian(lat1),
angle2radian(lon1))**2 -
dct_j['distance']**2)
) + math.sqrt(
math.fabs(
dist(
angle2radian(self.trajectory[k][2]),
angle2radian(self.trajectory[k][1]
), angle2radian(lat2),
angle2radian(lon2))**2 -
dct_k['distance']**2)))
if 'V' in dct_j.keys():
dct_j['V'][edge_k] = min(result, 1)
else:
dct_j['V'] = {edge_k: min(result, 1)}
except:
if 'V' in dct_j.keys():
dct_j['V'][edge_k] = 0
else:
dct_j['V'] = {edge_k: 0}
i += 1
