def find_object(img, thresholds, sizerange, dist_thresh, erode=False, check_centers=True):
body = nim.threshold(img, thresholds[0], thresholds[1])
if erode is not False:
for i in range(erode):
body = binary_erosion(body)
if check_centers is False:
blobs = nim.find_blobs(body, sizerange=sizerange, aslist=False)
else:
blobs = nim.find_blobs(body, sizerange=sizerange, aslist=True)
body = blobs
if check_centers:
centers = nim.center_of_blob(blobs)
dist = []
for center in centers:
diff = np.linalg.norm( center - np.array(img.shape)/2. )
dist.append(diff)
body = np.zeros_like(img)
for j, d in enumerate(dist):
if d < dist_thresh:
body += blobs[j]
if body.max() > 1:
body /= body.max()
if body is None:
body = np.zeros_like(img)
body = np.array(body*255, dtype=np.uint8)
return body
def calc_saccades(trajec, threshold_lo=300, threshold_hi=100000000, min_angle=10, plot=False):
# thresholds in deg/sec
# min_angle = minimum angle necessary to count as a saccade, deg
fps = 100.
possible_saccade_array = (np.abs(trajec.heading_smooth_diff)*180/np.pi > threshold_lo)*(np.abs(trajec.heading_smooth_diff)*180/np.pi < threshold_hi)
possible_saccades = nim.find_blobs(possible_saccade_array, [3,100])
if len(possible_saccades) == 1:
if np.sum(possible_saccades[0]) == 0:
possible_saccades = []
trajec.all_saccades = []
trajec.saccades = []
trajec.sac_ranges = []
if len(possible_saccades) > 0:
for sac in possible_saccades:
indices = np.where(sac==1)[0].tolist()
if len(indices) > 0:
# expand saccade range to make sure we get full turn
#lo = np.max([indices[0]-5, 0])
#hi = np.min([indices[-1]+5, len(trajec.speed)-2])
#new_indices = np.arange(lo, hi).tolist()
#tmp = np.where( np.abs(trajec.heading_diff_window[new_indices])*180/np.pi > 350 )[0].tolist()
#indices = np.array(new_indices)[ tmp ].tolist()
angle_of_saccade = np.abs(get_angle_of_saccade(trajec, indices)*180./np.pi)
mean_speed = np.mean(trajec.speed[indices])
if len(indices) > 3 and angle_of_saccade > 10: # and mean_speed > 0.005:
trajec.sac_ranges.append(indices)
s_rel = np.argmax( np.abs(trajec.heading_smooth_diff[indices]) )
s = indices[s_rel]
trajec.all_saccades.append(s)
if plot:
fig = plt.figure()
ax = fig.add_subplot(111)
ax.set_aspect('equal')
ax.plot(trajec.positions[:,0], trajec.positions[:,1], '-', color='black', alpha=1, linewidth=1, zorder=1)
for sac in trajec.sac_ranges:
ax.plot(trajec.positions[sac,0], trajec.positions[sac,1], '-', color='red', alpha=1, linewidth=1, zorder=1+1)
for s in trajec.all_saccades:
x = trajec.positions[s, 0]
y = trajec.positions[s, 1]
saccade = patches.Circle( (x, y), radius=0.001, facecolor='blue', edgecolor='none', alpha=1, zorder=3)
ax.add_artist(saccade)
post = patches.Circle( (0, 0), radius=0.009565, facecolor='black', edgecolor='none', alpha=1)
ax.add_artist(post)
fig.savefig('saccade_trajectory.pdf', format='pdf')
def get_bouts_of_constant_speed(movieinfo,
accel_threshold=.005,
duration_threshold=0.2):
duration_threshold_frames = duration_threshold * float(movieinfo.framerate)
#speed_diff = np.diff( np.hstack( (movieinfo.scaled.speed[0], movieinfo.scaled.speed) ) )
#speed_diff_bool_tmp = np.abs(speed_diff) < accel_threshold
nt, alt_speed = interpolate_to_new_framerate(
movieinfo, 5000., movieinfo.trajec.epoch_time,
movieinfo.trajec.velocities[:, 2])
nt, speed = interpolate_to_new_framerate(movieinfo, 5000.,
movieinfo.timestamps,
movieinfo.scaled.speed)
speed_diff = np.diff(np.hstack((speed[0], speed)))
speed_diff_bool_tmp = np.abs(speed_diff) < accel_threshold
alt_speed_diff = np.diff(np.hstack((alt_speed[0], alt_speed)))
alt_speed_diff_bool_tmp = np.abs(alt_speed_diff) < .005
bool_tmp = alt_speed_diff_bool_tmp * speed_diff_bool_tmp
continuous_sequences = nim.find_blobs(
bool_tmp, sizerange=[duration_threshold_frames, np.inf], dilate=1)
calc_pitch_estimate(movieinfo)
pitch = []
speed = []
for i, sequence in enumerate(continuous_sequences):
indices = np.where(sequence == 1)[0].tolist()
if np.sum(sequence) > 0:
print movieinfo.id, len(sequence)
r1 = np.median(movieinfo.scaled.speed[indices])
r2 = np.median(movieinfo.scaled.pitchangle[indices])
if not np.isnan(r1) and not np.isnan(r2):
pitch.append(r2)
speed.append(r1)
return pitch, speed
def get_flydra_speed_during_fixation_means(trajec,
fixation_threshold_degrees=5,
fixation_duration_threshold=0.1):
fixation_indices = get_flydra_indices_during_fixation(
trajec, fixation_threshold_degrees, fixation_duration_threshold)
if len(fixation_indices) > 0:
diffarr = np.diff(np.hstack((fixation_indices[0], fixation_indices)))
continuous_sequences = nim.find_blobs(diffarr)
mean = []
std = []
for sequence in continuous_sequences:
indices = np.where(sequence == 1)[0].tolist()
mean.append(np.mean(trajec.speed[indices]))
std.append(np.std(trajec.speed[indices]))
return mean, std
else:
return [], []
def get_continuous_data(fixation):
fixation_angle_bool_tmp = np.abs(fixation) < fixation_threshold_radians
dist_bool = trajec.dist_to_stim_r > 0.01
altitude_bool_high = trajec.positions[:, 2] < -.01
altitude_bool_low = trajec.positions[:, 2] > -.1
fixation_bool_tmp = fixation_angle_bool_tmp * altitude_bool_high * altitude_bool_low * dist_bool
#fixation_diff = np.diff( np.hstack( (fixation[0], fixation) ) )
#fixation_diff_bool_tmp = fixation_diff < 0.0005
#fixation_bool_tmp = fixation_angle_bool_tmp#*diff_bool_tmp
#fixation_indices_tmp = np.where( fixation_angle_bool_tmp == True )[0].tolist()
#fixation_indices_tmp = np.where( np.abs(fixation) < fixation_threshold_radians )[0].tolist()
continuous_sequences = nim.find_blobs(
fixation_bool_tmp,
sizerange=[fixation_duration_threshold_frames, np.inf],
dilate=False)
fixation_indices = []
for sequence in continuous_sequences:
indices = np.where(sequence == 1)[0].tolist()
fixation_indices = np.hstack((fixation_indices, indices))
return fixation_indices.tolist()
def get_continuous_data(fixation_lower):
lower_fixation_angle_bool_tmp = np.abs(
fixation_lower) < fixation_threshold_radians
fixation_lower_diff = np.diff(
np.hstack((fixation_lower[0], fixation_lower)))
lower_fixation_diff_bool_tmp = np.abs(fixation_lower_diff) < 0.01745
lower_fixation_bool_tmp = lower_fixation_angle_bool_tmp #lower_fixation_diff_bool_tmp #
lower_fixation_indices_tmp = np.where(
lower_fixation_bool_tmp == True)[0].tolist()
#lower_fixation_indices_tmp = np.where( np.abs(fixation_lower) < fixation_threshold_radians )[0].tolist()
continuous_sequences = nim.find_blobs(
lower_fixation_indices_tmp,
sizerange=[fixation_duration_threshold_frames, np.inf],
dilate=1)
lower_fixation_indices = []
for sequence in continuous_sequences:
indices = np.where(sequence == 1)[0].tolist()
lower_fixation_indices = np.hstack(
(lower_fixation_indices,
np.array(lower_fixation_indices_tmp)[indices]))
return lower_fixation_indices.tolist()
def background_subtraction(self, raw, save_raw=False):
mask_center_0 = int(self.mask_center[0])
mask_center_1 = int(self.mask_center[1])
mask_0_lo = max(0, mask_center_0 - self.mask_radius)
mask_0_hi = min(self.width, mask_center_0 + self.mask_radius)
mask_1_lo = max(0, mask_center_1 - self.mask_radius)
mask_1_hi = min(self.width, mask_center_1 + self.mask_radius)
if self.tracking_mask is not None:
tracking_mask = copy.copy(self.tracking_mask)
tracking_mask[mask_0_lo:mask_0_hi, mask_1_lo:mask_1_hi] = 1
if self.dynamic_tracking_mask is True:
# TODO: currently dynamic tracking and such only works for square format cameras
#print 'dynamic tracking'
masked_img = raw[mask_0_lo:mask_0_hi, mask_1_lo:mask_1_hi]
masked_background = self.background[mask_0_lo:mask_0_hi, mask_1_lo:mask_1_hi]
else:
masked_img = raw
masked_background = self.background
'''
if masked_img.shape[0] < 100 or masked_img.shape[1] < 100:
#print 'no uframe'
self.dynamic_tracking_mask = False
uframe = uFrame()
return uframe, None
'''
absdiff = nim.absdiff(masked_img, masked_background)
if self.dynamic_tracking_mask is False and self.tracking_mask is not None:
absdiff *= tracking_mask
#absdiff = nim.auto_adjust_levels(absdiff)
#print 'shape: ', np.shape(absdiff)
#threshold = max( 10, absdiff.max() - THRESHRANGE )
#print 'dynamic threshold: ', threshold
#diffthresh = nim.threshold(absdiff, threshold)
#print 'max absdiff: ', absdiff.max()
diffthresh = nim.threshold(absdiff, 15, threshold_hi=255)
# abort early if there is no info:
s = np.sum(diffthresh)
if s < 10:
uframe = uFrame()
#print 'no uframe, early abort, sum: ', s
self.dynamic_tracking_mask = False
return uframe, None
blobs = nim.find_blobs(diffthresh, self.blob_size_range)
if blobs is None:
self.dynamic_tracking_mask = False
masked_img = raw
masked_background = self.background
absdiff = nim.absdiff(masked_img, masked_background)
diffthresh = nim.threshold(absdiff, threshold)
if self.dynamic_tracking_mask is False and self.tracking_mask is not None:
diffthresh *= tracking_mask
blobs = nim.find_blobs(diffthresh, self.blob_size_range)
if blobs is None:
uframe = uFrame()
if save_raw is False:
#print 'no uframe'
self.dynamic_tracking_mask = False
return uframe, None
else:
frame = Frame(raw, absdiff, diffthresh)
return uframe, frame
nblobs = blobs.max()
#print 'n blobs: ', nblobs
if nblobs > 1:
blobs = nim.find_biggest_blob(blobs)
center = nim.center_of_blob(blobs)
#print 'center: ', center
if np.isnan(center)[0] or np.isnan(center)[1]:
uframe = uFrame()
#print 'no uframe, NaN center!'
self.dynamic_tracking_mask = False
return uframe, None
if center[0] < 1 or center[1] < 1:
uframe = uFrame()
#print 'no uframe, NaN center!'
self.dynamic_tracking_mask = False
return uframe, None
#print 'center found'
if 1:
limlo_x = max( int(center[0])-ROI_RADIUS, 0 )
limlo_y = max( int(center[1])-ROI_RADIUS, 0 )
limhi_x = min( int(center[0])+ROI_RADIUS, masked_img.shape[0] )
limhi_y = min( int(center[1])+ROI_RADIUS, masked_img.shape[1] )
# TODO: right now object entering or leaving frame doesnt work perfectly
uimg = masked_img[limlo_x:limhi_x, limlo_y:limhi_y]
uimg = copy.copy(uimg)
uimg_absdiff = absdiff[limlo_x:limhi_x, limlo_y:limhi_y]
uimg_absdiff = copy.copy(uimg_absdiff)
uimg_diffthresh = blobs[limlo_x:limhi_x, limlo_y:limhi_y]
uimg_diffthresh = copy.copy(uimg_diffthresh)
if self.dynamic_tracking_mask is True:
tmp = np.array([masked_img.shape[0]/2., masked_img.shape[1]/2.])
center = np.array(self.mask_center) + np.array(center) - tmp
uimg_indices = [mask_0_lo+limlo_x, mask_0_lo+limhi_x, mask_1_lo+limlo_y, mask_1_lo+limhi_y]
else:
uimg_indices = [limlo_x, limhi_x, limlo_y, limhi_y]
self.dynamic_tracking_mask = True
self.mask_center = center
uframe = uFrame(center, uimg, uimg_absdiff, uimg_diffthresh)
uframe.blobsize = blobs.sum()
uframe.indices = uimg_indices
if save_raw is False:
del(raw)
del(absdiff)
del(diffthresh)
del(blobs)
return uframe, None
frame = Frame(masked_img, absdiff, diffthresh)
return uframe, frame
def calc_saccades_z(trajec, threshold_lo=200, threshold_hi=100000000, min_angle=10, plot=False):
# thresholds in deg/sec
# min_angle = minimum angle necessary to count as a saccade, deg
fps = 100.
heading_norollover_for_axes = floris_math.remove_angular_rollover(np.arctan2(trajec.speed_xy[:], trajec.velocities[:,2]), 3)
## kalman
data = heading_norollover_for_axes.reshape([len(heading_norollover_for_axes),1])
ss = 3 # state size
os = 1 # observation size
F = np.array([ [1,1,0], # process update
[0,1,1],
[0,0,1]],
dtype=np.float)
H = np.array([ [1,0,0]], # observation matrix
dtype=np.float)
Q = np.eye(ss) # process noise
Q[0,0] = .01
Q[1,1] = .01
Q[2,2] = .01
R = 1*np.eye(os) # observation noise
initx = np.array([data[0,0], data[1,0]-data[0,0], 0], dtype=np.float)
initv = 0*np.eye(ss)
xsmooth,Vsmooth = kalman_math.kalman_smoother(data, F, H, Q, R, initx, initv, plot=False)
heading_norollover_smooth_for_axes = xsmooth[:,0]
heading_smooth_diff_for_axes = xsmooth[:,1]*trajec.fps
heading_for_axes = floris_math.fix_angular_rollover(heading_norollover_for_axes)
heading_smooth_for_axes = floris_math.fix_angular_rollover(heading_norollover_smooth_for_axes)
trajec.heading_altitude_smooth = heading_smooth_for_axes
trajec.heading_altitude_smooth_diff = heading_smooth_diff_for_axes
## saccades
possible_saccade_array = (np.abs(trajec.heading_altitude_smooth_diff)*180/np.pi > threshold_lo)*(np.abs(trajec.heading_altitude_smooth_diff)*180/np.pi < threshold_hi)
possible_saccades = nim.find_blobs(possible_saccade_array, [3,100])
if len(possible_saccades) == 1:
if np.sum(possible_saccades[0]) == 0:
possible_saccades = []
trajec.saccades_z = []
if len(possible_saccades) > 0:
for sac in possible_saccades:
indices = np.where(sac==1)[0].tolist()
if len(indices) > 0:
# expand saccade range to make sure we get full turn
#lo = np.max([indices[0]-5, 0])
#hi = np.min([indices[-1]+5, len(trajec.speed)-2])
#new_indices = np.arange(lo, hi).tolist()
#tmp = np.where( np.abs(trajec.heading_diff_window[new_indices])*180/np.pi > 350 )[0].tolist()
#indices = np.array(new_indices)[ tmp ].tolist()
angle_of_saccade = np.abs(get_angle_of_saccade_z(trajec, indices)*180./np.pi)
mean_speed = np.mean(trajec.speed[indices])
if len(indices) > 3 and angle_of_saccade > 10: # and mean_speed > 0.005:
trajec.saccades_z.append(indices)
if plot:
fig = plt.figure()
ax = fig.add_subplot(111)
ax.plot(trajec.positions[:,0], trajec.positions[:,2], color='black')
for sac in trajec.saccades_z:
ax.plot(trajec.positions[sac,0], trajec.positions[sac,2], color='red')