def load_sdd_dir(path: str, **kwargs):
search_filter_str = "**/annotations.txt"
if not path.endswith("/"):
search_filter_str = "/" + search_filter_str
files_list = sorted(glob.glob(path + search_filter_str, recursive=True))
scales_yaml_file = os.path.join(path, 'estimated_scales.yaml')
with open(scales_yaml_file, 'r') as f:
scales_yaml_content = yaml.load(f, Loader=yaml.FullLoader)
partial_datasets = []
for file in files_list:
dir_names = file.split('/')
scene_name = dir_names[-3]
scene_video_id = dir_names[-2]
scale = scales_yaml_content[scene_name][scene_video_id]['scale']
partial_dataset = load_sdd(file,
scale=scale,
scene_id=scene_name +
scene_video_id.replace('video', ''))
partial_datasets.append(partial_dataset.data)
traj_dataset = TrajDataset()
traj_dataset.data = pd.concat(partial_datasets)
fps = 30
sampling_rate = kwargs.get('sampling_rate', 1)
use_kalman = kwargs.get('use_kalman', False)
traj_dataset.postprocess(fps=fps,
sampling_rate=sampling_rate,
use_kalman=use_kalman)
return traj_dataset
def ttc(all_frames, name, trajlets):
all_ttc = []
Rp = 0.33 #assume pedestrians radius is 0.33
new_frames = []
for frame in all_frames:
frame.reset_index(inplace=True)
#if there is only one pedestrian at that time, or encounter invalid vel value
if len(frame.index) < 2 or frame['vel_x'].isnull().values.any(
) or frame['vel_y'].isnull().values.any():
continue
#calculate ttc for each pair
x_4d = np.stack((frame.pos_x.values, frame.pos_y.values,
frame.vel_x.values, frame.vel_y.values),
axis=1)
DCA, TTCA = DCA_MTX(x_4d)
for i in range(len(TTCA)):
#find out ttc of one agent
ttc = [
TTCA[i][j] for j in range(len(TTCA[i]))
if DCA[i][j] < 2 * Rp and TTCA[i][j] > 0
]
#find out min ttc for one agent
if len(ttc) > 0:
min_ttc = np.min(ttc)
frame.loc[i, 'ttc'] = min_ttc
min_dca = np.min([j for j in DCA[i] if j > 0])
frame.loc[i, 'dca'] = min_dca
new_frames.append(frame)
new_frames = pd.concat(new_frames)
new_traj = TrajDataset()
new_traj.data = new_frames
trajs = new_traj.get_trajectories(label="pedestrian")
trajlets[name] = split_trajectories(trajs, to_numpy=False)
#average local density o each trajlet
avg_traj_ttc = []
avg_traj_dca = []
for trajlet in trajlets[name]:
avg_traj_ttc.append(np.min(trajlet['ttc'].dropna())) #min of min
avg_traj_dca.append(np.min(trajlet['dca'].dropna())) #min of min
return avg_traj_ttc, avg_traj_dca
def local_density(all_frames,trajlets,name):
#define local density function
#for all pedestrians at that time, find its distance to NN
distNN = []
dens_t = []
a=1
new_frames = []
for frame in all_frames:
if len(frame)>1:
#find pairwise min distance
distNN.append([])
dens_t.append([])
dist = squareform(pdist(frame[['pos_x','pos_y']].values))
pair_dist = []
for pi in dist:
pair_dist.append(np.array(pi))
min_pi = [j for j in pi if j>0.01]
if len(min_pi) == 0:
min_dist = 0.01
else:
min_dist = np.min(min_pi)
distNN[-1].append(min_dist)
#calculate local density for agent pj
for pj in range(len(dist)):
dens_t_i = 1/(2*np.pi)*np.sum(1/((a*np.array(distNN[-1]))**2)*np.exp(-np.divide((pair_dist[pj]**2),(2*(a*np.array(distNN[-1]))**2))))
dens_t[-1].append(dens_t_i)
frame.loc[frame.index[pj],'p_local'] = dens_t_i
new_frames.append(frame)
new_frames = pd.concat(new_frames)
new_traj = TrajDataset()
new_traj.data = new_frames
trajs = new_traj.get_trajectories(label="pedestrian")
trajlets[name] = split_trajectories(trajs, to_numpy=False)
#average local density for each trajlet
avg_traj_plocal=[]
for trajlet in trajlets[name]:
avg_traj_plocal.append(np.max(trajlet['p_local']))
return avg_traj_plocal
def get_datasets(opentraj_root, dataset_names):
datasets = {}
# Make a temp dir to store and load trajdatasets (no postprocess anymore)
trajdataset_dir = os.path.join(opentraj_root, 'trajdatasets__temp')
if not os.path.exists(trajdataset_dir): os.makedirs(trajdataset_dir)
for dataset_name in dataset_names:
dataset_h5_file = os.path.join(trajdataset_dir, dataset_name + '.h5')
if os.path.exists(dataset_h5_file):
datasets[dataset_name] = TrajDataset()
datasets[dataset_name].data = pd.read_pickle(dataset_h5_file)
datasets[dataset_name].title = dataset_name
print("loading dataset from pre-processed file: ", dataset_h5_file)
continue
print("Loading dataset:", dataset_name)
# ========== ETH ==============
if 'eth-univ' == dataset_name.lower():
eth_univ_root = os.path.join(opentraj_root, 'datasets/ETH/seq_eth/obsmat.txt')
datasets[dataset_name] = load_eth(eth_univ_root, title=dataset_name, scene_id='Univ',
use_kalman=True)
elif 'eth-hotel' == dataset_name.lower():
eth_hotel_root = os.path.join(opentraj_root, 'datasets/ETH/seq_hotel/obsmat.txt')
datasets[dataset_name] = load_eth(eth_hotel_root, title=dataset_name, scene_id='Hotel')
# ******************************
# ========== UCY ==============
elif 'ucy-zara' == dataset_name.lower(): # all 3 zara sequences
zara01_dir = os.path.join(opentraj_root, 'datasets/UCY/zara01')
zara02_dir = os.path.join(opentraj_root, 'datasets/UCY/zara02')
zara03_dir = os.path.join(opentraj_root, 'datasets/UCY/zara03')
zara_01_ds = load_crowds(zara01_dir + '/annotation.vsp',
homog_file=zara01_dir + '/H.txt',
scene_id='1', use_kalman=True)
zara_02_ds = load_crowds(zara02_dir + '/annotation.vsp',
homog_file=zara02_dir + '/H.txt',
scene_id='2', use_kalman=True)
zara_03_ds = load_crowds(zara03_dir + '/annotation.vsp',
homog_file=zara03_dir + '/H.txt',
scene_id='3', use_kalman=True)
datasets[dataset_name] = merge_datasets([zara_01_ds, zara_02_ds, zara_03_ds], dataset_name)
elif 'ucy-univ' == dataset_name.lower(): # all 3 sequences
st001_dir = os.path.join(opentraj_root, 'datasets/UCY/students01')
st003_dir = os.path.join(opentraj_root, 'datasets/UCY/students03')
uni_ex_dir = os.path.join(opentraj_root, 'datasets/UCY/uni_examples')
#st001_ds = load_Crowds(st001_dir + '/students001.txt',homog_file=st001_dir + '/H.txt',scene_id='1',use_kalman=True)
st001_ds = load_crowds(st001_dir + '/annotation.vsp',
homog_file=st003_dir + '/H.txt',
scene_id='1', use_kalman=True)
st003_ds = load_crowds(st003_dir + '/annotation.vsp',
homog_file=st003_dir + '/H.txt',
scene_id='3', use_kalman=True)
uni_ex_ds = load_crowds(uni_ex_dir + '/annotation.vsp',
homog_file=st003_dir + '/H.txt',
scene_id='ex', use_kalman=True)
datasets[dataset_name] = merge_datasets([st001_ds, st003_ds, uni_ex_ds], dataset_name)
elif 'ucy-zara1' == dataset_name.lower():
zara01_root = os.path.join(opentraj_root, 'datasets/UCY/zara01/obsmat.txt')
datasets[dataset_name] = load_eth(zara01_root, title=dataset_name)
elif 'ucy-zara2' == dataset_name.lower():
zara02_root = os.path.join(opentraj_root, 'datasets/UCY/zara02/obsmat.txt')
datasets[dataset_name] = load_eth(zara02_root, title=dataset_name)
elif 'ucy-univ3' == dataset_name.lower():
students03_root = os.path.join(opentraj_root, 'datasets/UCY/students03/obsmat.txt')
datasets[dataset_name] = load_eth(students03_root, title=dataset_name)
# ******************************
# ========== HERMES ==============
elif 'bn' in dataset_name.lower().split('-'):
[_, exp_flow, cor_size] = dataset_name.split('-')
if exp_flow == '1d' and cor_size == 'w180': # 'Bottleneck-udf-180'
bottleneck_path = os.path.join(opentraj_root, 'datasets/HERMES/Corridor-1D/uo-180-180-120.txt')
elif exp_flow == '2d' and cor_size == 'w160': # 'Bottleneck-bdf-160'
bottleneck_path = os.path.join(opentraj_root, "datasets/HERMES/Corridor-2D/bo-360-160-160.txt")
else:
"Unknown Bottleneck dataset!"
continue
datasets[dataset_name] = load_bottleneck(bottleneck_path, sampling_rate=6,
use_kalman=True,
title=dataset_name)
# ******************************
# ========== PETS ==============
elif 'pets-s2l1' == dataset_name.lower():
pets_root = os.path.join(opentraj_root, 'datasets/PETS-2009/data')
datasets[dataset_name] = load_pets(os.path.join(pets_root, 'annotations/PETS2009-S2L1.xml'), #Pat:was PETS2009-S2L2
calib_path=os.path.join(pets_root, 'calibration/View_001.xml'),
sampling_rate=2,
title=dataset_name)
# ******************************
# ========== GC ==============
elif 'gc' == dataset_name.lower():
gc_root = os.path.join(opentraj_root, 'datasets/GC/Annotation')
datasets[dataset_name] = load_gcs(gc_root, world_coord=True, title=dataset_name,
use_kalman=True
)
# ******************************
# ========== InD ==============
elif 'ind-1' == dataset_name.lower():
ind_root = os.path.join(opentraj_root, 'datasets/InD/inD-dataset-v1.0/data')
file_ids = range(7, 17 + 1) # location_id = 1
ind_1_datasets = []
for id in file_ids:
dataset_i = load_ind(os.path.join(ind_root, '%02d_tracks.csv' % id),
scene_id='1-%02d' %id,
sampling_rate=10,
use_kalman=True)
ind_1_datasets.append(dataset_i)
datasets[dataset_name] = merge_datasets(ind_1_datasets, new_title=dataset_name)
elif 'ind-2' == dataset_name.lower():
ind_root = os.path.join(opentraj_root, 'datasets/InD/inD-dataset-v1.0/data')
file_ids = range(18, 29 + 1) # location_id = 1
ind_2_datasets = []
for id in file_ids:
dataset_i = load_ind(os.path.join(ind_root, '%02d_tracks.csv' % id),
scene_id='1-%02d' % id,
sampling_rate=10,
use_kalman=True)
ind_2_datasets.append(dataset_i)
datasets[dataset_name] = merge_datasets(ind_2_datasets, new_title=dataset_name)
elif 'ind-3' == dataset_name.lower():
ind_root = os.path.join(opentraj_root, 'datasets/InD/inD-dataset-v1.0/data')
file_ids = range(30, 32 + 1) # location_id = 1
ind_3_datasets = []
for id in file_ids:
dataset_i = load_ind(os.path.join(ind_root, '%02d_tracks.csv' % id),
scene_id='1-%02d' % id,
sampling_rate=10,
use_kalman=True)
ind_3_datasets.append(dataset_i)
datasets[dataset_name] = merge_datasets(ind_3_datasets, new_title=dataset_name)
elif 'ind-4' == dataset_name.lower():
ind_root = os.path.join(opentraj_root, 'datasets/InD/inD-dataset-v1.0/data')
file_ids = range(0, 6 + 1) # location_id = 1
ind_4_datasets = []
for id in file_ids:
dataset_i = load_ind(os.path.join(ind_root, '%02d_tracks.csv' % id),
scene_id='1-%02d' % id,
sampling_rate=10,
use_kalman=True)
ind_4_datasets.append(dataset_i)
datasets[dataset_name] = merge_datasets(ind_4_datasets, new_title=dataset_name)
# ******************************
# ========== KITTI ==============
elif 'kitti' == dataset_name.lower():
kitti_root = os.path.join(opentraj_root, 'datasets/KITTI/data')
datasets[dataset_name] = load_kitti(kitti_root, title=dataset_name,
use_kalman=True,
sampling_rate=1) # FixMe: apparently original_fps = 2.5
# ******************************
# ========== L-CAS ==============
elif 'lcas-minerva' == dataset_name.lower():
lcas_root = os.path.join(opentraj_root, 'datasets/L-CAS/data')
datasets[dataset_name] = load_lcas(lcas_root, title=dataset_name,
use_kalman=True,
sampling_rate=1) # FixMe: apparently original_fps = 2.5
# ******************************
# ========== Wild-Track ==============
elif 'wildtrack' == dataset_name.lower():
wildtrack_root = os.path.join(opentraj_root, 'datasets/Wild-Track/annotations_positions')
datasets[dataset_name] = load_wildtrack(wildtrack_root, title=dataset_name,
use_kalman=True,
sampling_rate=1) # original_annot_framerate=2
# ******************************
# ========== Edinburgh ==============
elif 'edinburgh' in dataset_name.lower():
edinburgh_dir = os.path.join(opentraj_root, 'datasets/Edinburgh/annotations')
if 'edinburgh' == dataset_name.lower(): # all files
# edinburgh_path = edinburgh_dir
# select 1-10 Sep
Ed_selected_days = ['01Sep', '02Sep', '04Sep', '05Sep', '06Sep', '10Sep']
partial_ds = []
for selected_day in Ed_selected_days:
edinburgh_path = os.path.join(edinburgh_dir, 'tracks.%s.txt' % selected_day)
partial_ds.append(load_edinburgh(edinburgh_path, title=dataset_name,
use_kalman=True, scene_id=selected_day,
sampling_rate=4) # original_framerate=9
)
merge_datasets(partial_ds)
else:
seq_date = dataset_name.split('-')[1]
edinburgh_path = os.path.join(edinburgh_dir, 'tracks.%s.txt' %seq_date)
datasets[dataset_name] = load_edinburgh(edinburgh_path, title=dataset_name,
use_kalman=True,
sampling_rate=4) # original_framerate=9
# ******************************
# ========== Town-Center ==============
elif 'towncenter' == dataset_name.lower():
towncenter_root = os.path.join(opentraj_root, 'datasets/Town-Center')
# FixMe: might need Kalman Smoother
datasets[dataset_name] = load_town_center(towncenter_root + '/TownCentre-groundtruth-top.txt',
calib_path=towncenter_root + '/TownCentre-calibration-ci.txt',
title=dataset_name,
use_kalman=True,
sampling_rate=10) # original_framerate=25
# ******************************
# ========== SDD ==============
elif 'sdd-' in dataset_name.lower():
scene_name = dataset_name.split('-')[1]
sdd_root = os.path.join(opentraj_root, 'datasets', 'SDD')
annot_files_sdd = sorted(glob.glob(sdd_root + '/' + scene_name + "/**/annotations.txt", recursive=True))
sdd_scales_yaml_file = os.path.join(sdd_root, 'estimated_scales.yaml')
with open(sdd_scales_yaml_file, 'r') as f:
scales_yaml_content = yaml.load(f, Loader=yaml.FullLoader)
scene_datasets = []
for file_name in annot_files_sdd:
filename_parts = file_name.split('/')
scene_name = filename_parts[-3]
scene_video_id = filename_parts[-2]
scale = scales_yaml_content[scene_name][scene_video_id]['scale']
sdd_dataset_i = load_sdd(file_name, scale=scale,
scene_id=scene_name + scene_video_id.replace('video', ''),
drop_lost_frames=False,
use_kalman=True,
sampling_rate=12) # original_framerate=30
scene_datasets.append(sdd_dataset_i)
scene_dataset = merge_datasets(scene_datasets, dataset_name)
datasets[dataset_name] = scene_dataset
# ******************************
else:
print("Error! invalid dataset name:", dataset_name)
# save to h5 file
datasets[dataset_name].data.to_pickle(dataset_h5_file)
print("saving dataset into pre-processed file: ", dataset_h5_file)
return datasets