def save_global_map():
globalmappos = np.empty([0, 2])
mapfactors = np.full(len(pynclt.sessions), np.nan)
poleparams = np.empty([0, 6])
for isession, s in enumerate(pynclt.sessions):
print(s)
session = pynclt.session(s)
istart, imid, iend = get_map_indices(session)
localmappos = session.T_w_r_gt_velo[imid, :2, 3]
if globalmappos.size == 0:
imaps = range(localmappos.shape[0])
else:
imaps = []
for imap in range(localmappos.shape[0]):
distance = np.linalg.norm(
localmappos[imap] - globalmappos, axis=1).min()
if distance > remapdistance:
imaps.append(imap)
globalmappos = np.vstack([globalmappos, localmappos[imaps]])
mapfactors[isession] = np.true_divide(len(imaps), len(imid))
with progressbar.ProgressBar(max_value=len(imaps)) as bar:
for iimap, imap in enumerate(imaps):
scans = []
for iscan in range(istart[imap], iend[imap]):
xyz, _ = session.get_velo(iscan)
scan = o3.PointCloud()
scan.points = o3.Vector3dVector(xyz)
scans.append(scan)
T_w_mc = np.identity(4)
T_w_mc[:3, 3] = session.T_w_r_gt_velo[imid[imap], :3, 3]
T_w_m = T_w_mc.dot(T_mc_m)
T_m_w = util.invert_ht(T_w_m)
T_m_r = np.matmul(
T_m_w, session.T_w_r_gt_velo[istart[imap]:iend[imap]])
occupancymap = mapping.occupancymap(
scans, T_m_r, mapshape, mapsize)
localpoleparams = poles.detect_poles(occupancymap, mapsize)
localpoleparams[:, :2] += T_w_m[:2, 3]
poleparams = np.vstack([poleparams, localpoleparams])
bar.update(iimap)
xy = poleparams[:, :2]
a = poleparams[:, [4]]
boxes = np.hstack([xy - 0.5 * a, xy + 0.5 * a])
clustermeans = np.empty([0, 5])
scores = []
for ci in cluster.cluster_boxes(boxes):
ci = list(ci)
if len(ci) < n_mapdetections:
continue
clustermeans = np.vstack([clustermeans, np.average(
poleparams[ci, :-1], axis=0, weights=poleparams[ci, -1])])
scores.append(np.mean(poleparams[ci, -1]))
clustermeans = np.hstack([clustermeans, np.array(scores).reshape([-1, 1])])
globalmapfile = os.path.join(pynclt.resultdir, get_globalmapname() + '.npz')
np.savez(globalmapfile,
polemeans=clustermeans, mapfactors=mapfactors, mappos=globalmappos)
plot_global_map(globalmapfile)
def save_global_map(seq):
sequence = dataset.sequence(seq)
seqdir = os.path.join(result_dir, '{:03d}'.format(seq))
util.makedirs(seqdir)
istart, imid, iend = get_map_indices(sequence)
poleparams = np.empty([0, 6])
with np.load(os.path.join(seqdir, localmapfile),
allow_pickle=True) as data:
for i, map in enumerate(data['maps']):
T_w_m = sequence.poses[map['imid']].dot(T_cam0_mc).dot(T_mc_m)
localpoleparams = map['poleparams']
h = np.diff(localpoleparams[:, 2:4], axis=1).squeeze()
npoles = localpoleparams.shape[0]
polepos_m = np.hstack(
[localpoleparams[:, :3],
np.ones([npoles, 1])]).T
polepos_w = np.matmul(T_w_m, polepos_m)[:3].T
localpoleparams[:, :3] = polepos_w
localpoleparams[:, 3] = polepos_w[:, 2] + h
poleparams = np.vstack([poleparams, localpoleparams])
xy = poleparams[:, :2]
a = poleparams[:, [4]]
boxes = np.hstack([xy - 0.5 * a, xy + 0.5 * a])
clustermeans = np.zeros([0, 5])
clustercovs = np.zeros([0, 5, 5])
clusterviz = []
for ci in cluster.cluster_boxes(boxes):
ci = list(ci)
if len(ci) < 1:
continue
clustermeans = np.vstack([
clustermeans,
np.average(poleparams[ci, :-1], axis=0, weights=poleparams[ci, -1])
])
np.savez(os.path.join(seqdir, globalmapfile),
polemeans=clustermeans,
polecovs=clustercovs)
def localize(sessionname, visualize=False):
print(sessionname)
mapdata = np.load(
os.path.join(pynclt.resultdir,
get_globalmapname() + '.npz'))
polemap = mapdata['polemeans'][:, :2]
polevar = 1.50
session = pynclt.session(sessionname)
locdata = np.load(os.path.join(session.dir, get_localmapfile()),
allow_pickle=True)['maps']
polepos_m = []
polepos_w = []
for i in range(len(locdata)):
n = locdata[i]['poleparams'].shape[0]
pad = np.hstack([np.zeros([n, 1]), np.ones([n, 1])])
polepos_m.append(np.hstack([locdata[i]['poleparams'][:, :2], pad]).T)
polepos_w.append(locdata[i]['T_w_m'].dot(polepos_m[i]))
istart = 0
# igps = np.searchsorted(session.t_gps, session.t_relodo[istart]) + [-4, 1]
# igps = np.clip(igps, 0, session.gps.shape[0] - 1)
# T_w_r_start = pynclt.T_w_o
# T_w_r_start[:2, 3] = np.mean(session.gps[igps], axis=0)
T_w_r_start = util.project_xy(
session.get_T_w_r_gt(session.t_relodo[istart]).dot(T_r_mc)).dot(T_mc_r)
filter = particlefilter.particlefilter(500,
T_w_r_start,
2.5,
np.radians(5.0),
polemap,
polevar,
T_w_o=T_mc_r)
# Init: particlefilter(count = #particles, start: init pose, posrange: for init, angrange: for init,\
# polemeans: global map data, polevar, T_w_o=np.identity(4))
filter.estimatetype = 'best'
filter.minneff = 0.5
# filter = inEKF.inEKF(T_w_r_start, polemap, polevar, T_w_o = T_mc_r)
if visualize:
plt.ion()
figure = plt.figure()
nplots = 1
mapaxes = figure.add_subplot(nplots, 1, 1)
mapaxes.set_aspect('equal')
mapaxes.scatter(polemap[:, 0], polemap[:, 1], s=5, c='b', marker='s')
x_gt, y_gt = session.T_w_r_gt[::20, :2, 3].T
mapaxes.plot(x_gt, y_gt, 'g')
particles = mapaxes.scatter([], [], s=1, c='r')
arrow = mapaxes.arrow(0.0,
0.0,
1.0,
0.0,
length_includes_head=True,
head_width=0.7,
head_length=1.0,
color='k')
arrowdata = np.hstack(
[arrow.get_xy(), np.zeros([8, 1]),
np.ones([8, 1])]).T
locpoles = mapaxes.scatter([], [], s=30, c='k', marker='x')
viewoffset = 25.0
# weightaxes = figure.add_subplot(nplots, 1, 2)
# gridsize = 50
# offset = 5.0
# visfilter = particlefilter.particlefilter(gridsize**2,
# np.identity(4), 0.0, 0.0, polemap,
# polevar, T_w_o=pynclt.T_w_o)
# gridcoord = np.linspace(-offset, offset, gridsize)
# x, y = np.meshgrid(gridcoord, gridcoord)
# dxy = np.hstack([x.reshape([-1, 1]), y.reshape([-1, 1])])
# weightimage = weightaxes.matshow(np.zeros([gridsize, gridsize]),
# extent=(-offset, offset, -offset, offset))
# histaxes = figure.add_subplot(nplots, 1, 3)
imap = 0
while imap < locdata.shape[0] - 1 and session.t_velo[
locdata[imap]['iend']] < session.t_relodo[istart]:
imap += 1
T_w_r_est = np.full([session.t_relodo.size, 4, 4], np.nan)
#steps = 5000
x_sigma_contour = np.zeros(session.t_relodo.size)
y_sigma_contour = np.zeros(session.t_relodo.size)
p_sigma_contour = np.zeros(session.t_relodo.size)
x_err = np.zeros(session.t_relodo.size)
y_err = np.zeros(session.t_relodo.size)
p_err = np.zeros(session.t_relodo.size)
with progressbar.ProgressBar(max_value=session.t_relodo.size) as bar:
for i in range(istart, session.t_relodo.size):
#for i in range(istart, istart + steps):
relodocov = np.empty([3, 3])
relodocov[:2, :2] = session.relodocov[i, :2, :2]
relodocov[:, 2] = session.relodocov[i, [0, 1, 5], 5]
relodocov[2, :] = session.relodocov[i, 5, [0, 1, 5]]
filter.update_motion(
session.relodo[i], relodocov
) ### relodocov #propagate: session.relodo[i]=[x,y,p] in R^3
T_w_r_est[i] = filter.estimate_pose() ## estimate pose
t_now = session.t_relodo[i]
if imap < locdata.shape[0]:
t_end = session.t_velo[locdata[imap]['iend']]
if t_now >= t_end:
imaps = range(imap, np.clip(imap - n_localmaps, -1, None),
-1)
xy = np.hstack([polepos_w[j][:2] for j in imaps]).T
a = np.vstack([ld['poleparams'][:, [4]] \
for ld in locdata[imaps]])
boxes = np.hstack([xy - 0.5 * a, xy + 0.5 * a])
ipoles = set(range(polepos_w[imap].shape[1]))
iactive = set()
for ci in cluster.cluster_boxes(boxes):
if len(ci) >= n_locdetections:
iactive |= set(ipoles) & ci
iactive = list(iactive)
# print('{}.'.format(
# len(iactive) - polepos_w[imap].shape[1]))
if iactive:
t_mid = session.t_velo[locdata[imap]['imid']]
T_w_r_mid = util.project_xy(
session.get_T_w_r_odo(t_mid).dot(T_r_mc)).dot(
T_mc_r)
T_w_r_now = util.project_xy(
session.get_T_w_r_odo(t_now).dot(T_r_mc)).dot(
T_mc_r)
T_r_now_r_mid = util.invert_ht(T_w_r_now).dot(
T_w_r_mid)
polepos_r_now = T_r_now_r_mid.dot(T_r_m).dot(
polepos_m[imap]
[:, iactive]) # online poles(landmarks): lumbda
filter.update_measurement(
polepos_r_now[:2].T) ### measurement update
T_w_r_est[i] = filter.estimate_pose() ### estimate
if visualize:
polepos_w_est = T_w_r_est[i].dot(polepos_r_now)
locpoles.set_offsets(polepos_w_est[:2].T)
# T_w_r_gt_now = session.get_T_w_r_gt(t_now)
# T_w_r_gt_now = np.tile(
# T_w_r_gt_now, [gridsize**2, 1, 1])
# T_w_r_gt_now[:, :2, 3] += dxy
# visfilter.particles = T_w_r_gt_now
# visfilter.weights[:] = 1.0 / visfilter.count
# visfilter.update_measurement(
# polepos_r_now[:2].T, resample=False)
# weightimage.set_array(np.flipud(
# visfilter.weights.reshape(
# [gridsize, gridsize])))
# weightimage.autoscale()
imap += 1
# estimattion error
T_w_r_gt_i = util.project_xy(
session.get_T_w_r_gt(session.t_relodo[i]).dot(T_r_mc)).dot(
T_mc_r) # ground truth pose
x_err[i] = T_w_r_est[i, 0, 3] - T_w_r_gt_i[0, 3]
y_err[i] = T_w_r_est[i, 1, 3] - T_w_r_gt_i[1, 3]
p_err[i] = np.arctan2(T_w_r_est[i, 1, 0],
T_w_r_est[i, 0, 0]) - np.arctan2(
T_w_r_gt_i[1, 0], T_w_r_gt_i[0, 0])
p_err[i] = np.arctan2(np.sin(p_err[i]),
np.cos(p_err[i])) # wrap to [-pi, pi]
# 3-sigma contour
if isinstance(filter, inEKF.inEKF):
x_sigma_contour[i] = 3 * np.sqrt(filter.Sigma[1, 1])
y_sigma_contour[i] = 3 * np.sqrt(filter.Sigma[2, 2])
p_sigma_contour[i] = 3 * np.sqrt(filter.Sigma[0, 0])
elif isinstance(filter, particlefilter.particlefilter):
x_sigma_contour[i] = 3 * np.sqrt(filter.Sigma[0, 0])
y_sigma_contour[i] = 3 * np.sqrt(filter.Sigma[1, 1])
p_sigma_contour[i] = 3 * np.sqrt(filter.Sigma[2, 2])
if visualize:
## particles.set_offsets(filter.particles[:, :2, 3])
arrow.set_xy(T_w_r_est[i].dot(arrowdata)[:2].T)
x, y = T_w_r_est[i, :2, 3]
mapaxes.set_xlim(left=x - viewoffset, right=x + viewoffset)
mapaxes.set_ylim(bottom=y - viewoffset, top=y + viewoffset)
# histaxes.cla()
# histaxes.hist(filter.weights,
# bins=50, range=(0.0, np.max(filter.weights)))
figure.canvas.draw_idle()
figure.canvas.flush_events()
bar.update(i)
# Plot the NEES (normalized estimation error squared) graph
fig = plt.figure()
ax1 = plt.subplot(311)
plt.plot(x_err, 'r', label="Deviance from Ground Truth")
plt.plot(x_sigma_contour, 'b', label='3-Sigma Contour')
plt.plot(-x_sigma_contour, 'b')
plt.ylabel('x error')
plt.xlabel('step')
# ax.legend(loc='upper right', bbox_to_anchor=(1,0))
ax2 = plt.subplot(312)
plt.plot(y_err, 'r')
plt.plot(y_sigma_contour, 'b')
plt.plot(-y_sigma_contour, 'b')
plt.ylabel('y error')
plt.xlabel('step')
ax3 = plt.subplot(313)
plt.plot(p_err, 'r')
plt.plot(p_sigma_contour, 'b')
plt.plot(-p_sigma_contour, 'b')
plt.ylabel('theta error')
plt.xlabel('step')
handles, labels = ax1.get_legend_handles_labels()
fig.legend(handles, labels, 'upper right')
plt.savefig(os.path.join(pynclt.resultdir, sessionname + '_NEES.png'))
#plt.savefig("NEES.png")
filename = os.path.join(session.dir, get_locfileprefix() \
+ datetime.datetime.now().strftime('_%Y-%m-%d_%H-%M-%S.npz'))
np.savez(filename, T_w_r_est=T_w_r_est)
def localize(sessionname, visualize=False):
print(sessionname)
mapdata = np.load(os.path.join(pynclt.resultdir, get_globalmapname() + '.npz'))
polemap = mapdata['polemeans'][:, :2]
polevar = 1.50
session = pynclt.session(sessionname)
locdata = np.load(os.path.join(session.dir, get_localmapfile()), allow_pickle=True)['maps']
polepos_m = []
polepos_w = []
for i in range(len(locdata)):
n = locdata[i]['poleparams'].shape[0]
pad = np.hstack([np.zeros([n, 1]), np.ones([n, 1])])
polepos_m.append(np.hstack([locdata[i]['poleparams'][:, :2], pad]).T)
polepos_w.append(locdata[i]['T_w_m'].dot(polepos_m[i]))
istart = 0
# igps = np.searchsorted(session.t_gps, session.t_relodo[istart]) + [-4, 1]
# igps = np.clip(igps, 0, session.gps.shape[0] - 1)
# T_w_r_start = pynclt.T_w_o
# T_w_r_start[:2, 3] = np.mean(session.gps[igps], axis=0)
T_w_r_start = util.project_xy(
session.get_T_w_r_gt(session.t_relodo[istart]).dot(T_r_mc)).dot(T_mc_r)
filter = particlefilter.particlefilter(5000,
T_w_r_start, 2.5, np.radians(5.0), polemap, polevar, T_w_o=T_mc_r)
filter.estimatetype = 'best'
filter.minneff = 0.5
if visualize:
plt.ion()
figure = plt.figure()
nplots = 1
mapaxes = figure.add_subplot(nplots, 1, 1)
mapaxes.set_aspect('equal')
mapaxes.scatter(polemap[:, 0], polemap[:, 1], s=5, c='b', marker='s')
x_gt, y_gt = session.T_w_r_gt[::20, :2, 3].T
mapaxes.plot(x_gt, y_gt, 'g')
particles = mapaxes.scatter([], [], s=1, c='r')
arrow = mapaxes.arrow(0.0, 0.0, 1.0, 0.0, length_includes_head=True,
head_width=0.7, head_length=1.0, color='k')
arrowdata = np.hstack(
[arrow.get_xy(), np.zeros([8, 1]), np.ones([8, 1])]).T
locpoles = mapaxes.scatter([], [], s=30, c='k', marker='x')
viewoffset = 25.0
# weightaxes = figure.add_subplot(nplots, 1, 2)
# gridsize = 50
# offset = 5.0
# visfilter = particlefilter.particlefilter(gridsize**2,
# np.identity(4), 0.0, 0.0, polemap,
# polevar, T_w_o=pynclt.T_w_o)
# gridcoord = np.linspace(-offset, offset, gridsize)
# x, y = np.meshgrid(gridcoord, gridcoord)
# dxy = np.hstack([x.reshape([-1, 1]), y.reshape([-1, 1])])
# weightimage = weightaxes.matshow(np.zeros([gridsize, gridsize]),
# extent=(-offset, offset, -offset, offset))
# histaxes = figure.add_subplot(nplots, 1, 3)
imap = 0
while imap < locdata.shape[0] - 1 and \
session.t_velo[locdata[imap]['iend']] < session.t_relodo[istart]:
imap += 1
T_w_r_est = np.full([session.t_relodo.size, 4, 4], np.nan)
with progressbar.ProgressBar(max_value=session.t_relodo.size) as bar:
for i in range(istart, session.t_relodo.size):
relodocov = np.empty([3, 3])
relodocov[:2, :2] = session.relodocov[i, :2, :2]
relodocov[:, 2] = session.relodocov[i, [0, 1, 5], 5]
relodocov[2, :] = session.relodocov[i, 5, [0, 1, 5]]
filter.update_motion(session.relodo[i], relodocov * 2.0**2)
T_w_r_est[i] = filter.estimate_pose()
t_now = session.t_relodo[i]
if imap < locdata.shape[0]:
t_end = session.t_velo[locdata[imap]['iend']]
if t_now >= t_end:
imaps = range(imap, np.clip(imap-n_localmaps, -1, None), -1)
xy = np.hstack([polepos_w[j][:2] for j in imaps]).T
a = np.vstack([ld['poleparams'][:, [4]] \
for ld in locdata[imaps]])
boxes = np.hstack([xy - 0.5 * a, xy + 0.5 * a])
ipoles = set(range(polepos_w[imap].shape[1]))
iactive = set()
for ci in cluster.cluster_boxes(boxes):
if len(ci) >= n_locdetections:
iactive |= set(ipoles) & ci
iactive = list(iactive)
# print('{}.'.format(
# len(iactive) - polepos_w[imap].shape[1]))
if iactive:
t_mid = session.t_velo[locdata[imap]['imid']]
T_w_r_mid = util.project_xy(session.get_T_w_r_odo(
t_mid).dot(T_r_mc)).dot(T_mc_r)
T_w_r_now = util.project_xy(session.get_T_w_r_odo(
t_now).dot(T_r_mc)).dot(T_mc_r)
T_r_now_r_mid = util.invert_ht(T_w_r_now).dot(T_w_r_mid)
polepos_r_now = T_r_now_r_mid.dot(T_r_m).dot(
polepos_m[imap][:, iactive])
filter.update_measurement(polepos_r_now[:2].T)
T_w_r_est[i] = filter.estimate_pose()
if visualize:
polepos_w_est = T_w_r_est[i].dot(polepos_r_now)
locpoles.set_offsets(polepos_w_est[:2].T)
# T_w_r_gt_now = session.get_T_w_r_gt(t_now)
# T_w_r_gt_now = np.tile(
# T_w_r_gt_now, [gridsize**2, 1, 1])
# T_w_r_gt_now[:, :2, 3] += dxy
# visfilter.particles = T_w_r_gt_now
# visfilter.weights[:] = 1.0 / visfilter.count
# visfilter.update_measurement(
# polepos_r_now[:2].T, resample=False)
# weightimage.set_array(np.flipud(
# visfilter.weights.reshape(
# [gridsize, gridsize])))
# weightimage.autoscale()
imap += 1
if visualize:
particles.set_offsets(filter.particles[:, :2, 3])
arrow.set_xy(T_w_r_est[i].dot(arrowdata)[:2].T)
x, y = T_w_r_est[i, :2, 3]
mapaxes.set_xlim(left=x - viewoffset, right=x + viewoffset)
mapaxes.set_ylim(bottom=y - viewoffset, top=y + viewoffset)
# histaxes.cla()
# histaxes.hist(filter.weights,
# bins=50, range=(0.0, np.max(filter.weights)))
figure.canvas.draw_idle()
figure.canvas.flush_events()
bar.update(i)
filename = os.path.join(session.dir, get_locfileprefix() \
+ datetime.datetime.now().strftime('_%Y-%m-%d_%H-%M-%S.npz'))
np.savez(filename, T_w_r_est=T_w_r_est)
def localize(sessionname, visualize=False):
print(sessionname)
mapdata = np.load(
os.path.join(pynclt.resultdir,
get_globalmapname() + '.npz'))
polemap = mapdata['polemeans'][:, :2]
polevar = 1.50
session = pynclt.session(sessionname)
figuresdir = os.path.join(
pynclt.resultdir, session.dir,
'Figures_{:.0f}_{:.0f}_{:.0f}'.format(n_mapdetections,
10 * poles.minscore,
poles.polesides[-1]))
util.makedirs(figuresdir)
locdata = np.load(os.path.join(session.dir, get_localmapfile()),
allow_pickle=True)['maps']
polepos_m = []
polepos_w = []
for i in range(len(locdata)):
n = locdata[i]['poleparams'].shape[0]
pad = np.hstack([np.zeros([n, 1]), np.ones([n, 1])])
polepos_m.append(np.hstack([locdata[i]['poleparams'][:, :2], pad]).T)
polepos_w.append(locdata[i]['T_w_m'].dot(polepos_m[i]))
istart = 0
T_w_r_start = util.project_xy(
session.get_T_w_r_gt(session.t_relodo[istart]).dot(T_r_mc)).dot(T_mc_r)
filter = particlefilter.particlefilter(5000,
T_w_r_start,
2.5,
np.radians(5.0),
polemap,
polevar,
T_w_o=T_mc_r)
filter.estimatetype = 'best'
filter.minneff = 0.5
if visualize:
plt.ion()
figure = plt.figure()
nplots = 1
mapaxes = figure.add_subplot(nplots, 1, 1)
mapaxes.set_aspect('equal')
mapaxes.scatter(polemap[:, 0], polemap[:, 1], s=10, c='b', marker='s')
x_gt, y_gt = session.T_w_r_gt[::20, :2, 3].T
mapaxes.plot(x_gt, y_gt, 'g')
particles = mapaxes.scatter([], [], s=1, c='r')
arrow = mapaxes.arrow(0.0,
0.0,
4.0,
0.0,
length_includes_head=True,
head_width=1.2,
head_length=1.5,
color='k')
arrowdata = np.hstack(
[arrow.get_xy(), np.zeros([8, 1]),
np.ones([8, 1])]).T
locpoles = mapaxes.scatter([], [], s=30, c='y', marker='^')
viewoffset = 25.0
imap = 0
while imap < locdata.shape[0] - 1 and \
session.t_velo[locdata[imap]['iend']] < session.t_relodo[istart]:
imap += 1
T_w_r_est = np.full([session.t_relodo.size, 4, 4], np.nan)
with progressbar.ProgressBar(max_value=session.t_relodo.size) as bar:
for i in range(istart, session.t_relodo.size):
relodocov = np.empty([3, 3])
relodocov[:2, :2] = session.relodocov[i, :2, :2]
relodocov[:, 2] = session.relodocov[i, [0, 1, 5], 5]
relodocov[2, :] = session.relodocov[i, 5, [0, 1, 5]]
filter.update_motion(session.relodo[i], relodocov * 2.0**2)
T_w_r_est[i] = filter.estimate_pose()
t_now = session.t_relodo[i]
if imap < locdata.shape[0]:
t_end = session.t_velo[locdata[imap]['iend']]
if t_now >= t_end:
imaps = range(imap, np.clip(imap - n_localmaps, -1, None),
-1)
xy = np.hstack([polepos_w[j][:2] for j in imaps]).T
a = np.vstack([ld['poleparams'][:, [4]] \
for ld in locdata[imaps]])
boxes = np.hstack([xy - 0.5 * a, xy + 0.5 * a])
ipoles = set(range(polepos_w[imap].shape[1]))
iactive = set()
for ci in cluster.cluster_boxes(boxes):
if len(ci) >= n_locdetections:
iactive |= set(ipoles) & ci
iactive = list(iactive)
if iactive:
t_mid = session.t_velo[locdata[imap]['imid']]
T_w_r_mid = util.project_xy(
session.get_T_w_r_odo(t_mid).dot(T_r_mc)).dot(
T_mc_r)
T_w_r_now = util.project_xy(
session.get_T_w_r_odo(t_now).dot(T_r_mc)).dot(
T_mc_r)
T_r_now_r_mid = util.invert_ht(T_w_r_now).dot(
T_w_r_mid)
polepos_r_now = T_r_now_r_mid.dot(T_r_m).dot(
polepos_m[imap][:, iactive])
filter.update_measurement(polepos_r_now[:2].T)
T_w_r_est[i] = filter.estimate_pose()
if visualize:
polepos_w_est = T_w_r_est[i].dot(polepos_r_now)
locpoles.set_offsets(polepos_w_est[:2].T)
imap += 1
if visualize:
particles.set_offsets(filter.particles[:, :2, 3])
arrow.set_xy(T_w_r_est[i].dot(arrowdata)[:2].T)
x, y = T_w_r_est[i, :2, 3]
mapaxes.set_xlim(left=x - viewoffset, right=x + viewoffset)
mapaxes.set_ylim(bottom=y - viewoffset, top=y + viewoffset)
# Save figures for generating GIF
if i % 25 == 0:
filename = sessionname + '_' + str(i) + '_'
figure.savefig(os.path.join(figuresdir, filename + '.png'))
figure.canvas.draw_idle()
figure.canvas.flush_events()
bar.update(i)
filename = os.path.join(session.dir, get_locfileprefix() \
+ datetime.datetime.now().strftime('_%Y-%m-%d_%H-%M-%S.npz'))
np.savez(filename, T_w_r_est=T_w_r_est)
