def __init__(self, gpool): # what's up with gpood. We don't seem to be using it
''' Class Initiator) '''
super(MrGaze_Detector, self).__init__()
self.gpool = gpool
self.cfg = config.LoadConfig(os.path.join(gpool.user_dir))
# # Create a new parser
# self.cfg = ConfigParser.ConfigParser()
# self.cfg = config.InitConfig(self.cfg)
# self.cfg.set('PUPILSEG', 'manualroi', [0.5, 0.5, 1.0])
# Init Cascade Classfier
mrclean_root = utils._package_root()
LBP_path = os.path.join(mrclean_root, 'Cascade/cascade.xml')
logger.debug(' Loading LBP cascade')
self.cascade = cv2.CascadeClassifier(LBP_path)
def __init__(self,
gpool): # what's up with gpood. We don't seem to be using it
''' Class Initiator) '''
super(MrGaze_Detector, self).__init__()
self.gpool = gpool
self.cfg = config.LoadConfig(os.path.join(gpool.user_dir))
# # Create a new parser
# self.cfg = ConfigParser.ConfigParser()
# self.cfg = config.InitConfig(self.cfg)
# self.cfg.set('PUPILSEG', 'manualroi', [0.5, 0.5, 1.0])
# Init Cascade Classfier
mrclean_root = utils._package_root()
LBP_path = os.path.join(mrclean_root, 'Cascade/cascade.xml')
logger.debug(' Loading LBP cascade')
self.cascade = cv2.CascadeClassifier(LBP_path)
def VideoPupilometry(data_dir, subj_sess, v_stub, cfg):
"""
Perform pupil boundary ellipse fitting on entire video
Arguments
----
data_dir : string
Root data directory path.
subj_sess : string
Subject/Session name used for subdirectory within data_dir
v_stub : string
Video filename stub, eg 'cal' or 'gaze'
cfg :
Analysis configuration parameters
Returns
----
pupils : boolean
Completion status (True = successful)
"""
# Output flags
verbose = cfg.getboolean('OUTPUT', 'verbose')
overwrite = cfg.getboolean('OUTPUT','overwrite')
# Video information
vin_ext = cfg.get('VIDEO', 'inputextension')
vout_ext = cfg.get('VIDEO' ,'outputextension')
vin_fps = cfg.getfloat('VIDEO', 'inputfps')
# Full video file paths
ss_dir = os.path.join(data_dir, subj_sess)
vid_dir = os.path.join(ss_dir, 'videos')
res_dir = os.path.join(ss_dir, 'results')
vin_path = os.path.join(vid_dir, v_stub + vin_ext)
vout_path = os.path.join(res_dir, v_stub + '_pupils' + vout_ext)
# Raw and filtered pupilometry CSV file paths
pupils_csv = os.path.join(res_dir, v_stub + '_pupils.csv')
# Check that input video file exists
if not os.path.isfile(vin_path):
print('* %s does not exist - returning' % vin_path)
return False
# Set up the LBP cascade classifier
LBP_path = os.path.join(utils._package_root(), 'Cascade/cascade.xml')
print(' Loading LBP cascade')
cascade = cv2.CascadeClassifier(LBP_path)
if cascade.empty():
print('* LBP cascade is empty - mrgaze installation problem')
return False
# Check for output CSV existance and overwrite flag
if os.path.isfile(pupils_csv):
print('+ Pupilometry output already exists - checking overwrite flag')
if overwrite:
print('+ Overwrite allowed - continuing')
else:
print('+ Overwrite forbidden - skipping pupilometry')
return True
#
# Input video
#
print(' Opening input video stream')
try:
vin_stream = cv2.VideoCapture(vin_path)
except:
print('* Problem opening input video stream - skipping pupilometry')
return False
if not vin_stream.isOpened():
print('* Video input stream not opened - skipping pupilometry')
return False
# Video FPS from metadata
# TODO: may not work with Quicktime videos
# fps = vin_stream.get(cv2.cv.CV_CAP_PROP_FPS)
# Total number of frames in video file
nf = vin_stream.get(cv2.CAP_PROP_FRAME_COUNT)
print(' Video has %d frames at %0.3f fps' % (nf, vin_fps))
# Read first preprocessed video frame from stream
keep_going, frame_orig = media.LoadVideoFrame(vin_stream, cfg)
if keep_going:
frame, art_power = media.Preproc(frame_orig, cfg)
else:
art_power = 0.0
# Get size of preprocessed frame for output video setup
nx, ny = frame.shape[1], frame.shape[0]
#
# Output video
#
print(' Opening output video stream')
# Output video codec (MP4V - poor quality compression)
fourcc = cv2.VideoWriter_fourcc('m','p','4','v')
try:
vout_stream = cv2.VideoWriter(vout_path, fourcc, 30, (nx, ny), True)
except:
print('* Problem creating output video stream - skipping pupilometry')
return False
if not vout_stream.isOpened():
print('* Output video not opened - skipping pupilometry')
return False
# Open pupilometry CSV file to write
try:
pupils_stream = open(pupils_csv, 'w')
except:
print('* Problem opening pupilometry CSV file - skipping pupilometry')
return False
#
# Main Video Frame Loop
#
# Print verbose column headers
if verbose:
print('')
print(' %10s %10s %10s %10s %10s %10s' % (
'Time (s)', '% Done', 'Area', 'Blink', 'Artifact', 'FPS'))
# Init frame counter
fc = 0
# Init processing timer
t0 = time.time()
while keep_going:
# Current video time in seconds
t = fc / vin_fps
# -------------------------------------
# Pass this frame to pupilometry engine
# -------------------------------------
pupil_ellipse, roi_rect, blink, glint, frame_rgb = engine.PupilometryEngine(frame, cascade, cfg)
# Derive pupilometry parameters
px, py, area = engine.PupilometryPars(pupil_ellipse, glint, cfg)
# Write data line to pupilometry CSV file
pupils_stream.write(
'%0.3f,%0.3f,%0.3f,%0.3f,%d,%0.3f,\n' %
(t, area, px, py, blink, art_power)
)
# Write output video frame
vout_stream.write(frame_rgb)
# Read next frame (if available)
keep_going, frame_orig = media.LoadVideoFrame(vin_stream, cfg)
if keep_going:
frame, art_power = media.Preproc(frame_orig, cfg)
else:
art_power = 0.0
# Increment frame counter
fc = fc + 1
# Report processing FPS
if verbose:
if fc % 100 == 0:
perc_done = fc / float(nf) * 100.0
pfps = fc / (time.time() - t0)
print(' %10.1f %10.1f %10.1f %10d %10.3f %10.1f' % (
t, perc_done, area, blink, art_power, pfps))
# Clean up
cv2.destroyAllWindows()
vin_stream.release()
vout_stream.release()
pupils_stream.close()
# Return pupilometry timeseries
return t, px, py, area, blink, art_power
def LivePupilometry(data_dir, live_eyetracking=False):
"""
Perform pupil boundary ellipse fitting on camera feed
Arguments
----
data_dir : string
Root data directory path.
cfg :
Analysis configuration parameters
Returns
----
pupils : boolean
Completion status (True = successful)
"""
# If user did not provide a root data directory, we use HOME/mrgaze
if data_dir == '':
data_dir = os.path.join(os.getenv("HOME"), 'mrgaze')
# Full video file paths
hostname = os.uname()[1]
username = getpass.getuser()
ss_dir = os.path.join(data_dir, "%s_%s_%s" % (hostname, username, int(time.time())))
else:
ss_dir = data_dir
# Load Configuration
cfg = config.LoadConfig(data_dir)
cfg_ts = time.time()
# Output flags
verbose = cfg.getboolean('OUTPUT', 'verbose')
overwrite = cfg.getboolean('OUTPUT', 'overwrite')
# Video information
# vin_ext = cfg.get('VIDEO', 'inputextension')
vout_ext = cfg.get('VIDEO' ,'outputextension')
# vin_fps = cfg.getfloat('VIDEO', 'inputfps')
# Flag for freeze frame
freeze_frame = False
vid_dir = os.path.join(ss_dir, 'videos')
res_dir = os.path.join(ss_dir, 'results')
vout_path = os.path.join(vid_dir, 'gaze' + vout_ext)
cal_vout_path = os.path.join(vid_dir, 'cal' + vout_ext)
# if we do live eye-tracking, we read in what would be the output of the live eye-tracking
if not live_eyetracking:
vin_path = vout_path
cal_vin_path = cal_vout_path
else:
vin_path = 0
# Raw and filtered pupilometry CSV file paths
cal_pupils_csv = os.path.join(res_dir, 'cal_pupils.csv')
pupils_csv = os.path.join(res_dir, 'gaze_pupils.csv')
# Check that output directory exists
if not os.path.isdir(res_dir):
os.makedirs(res_dir)
print('* %s does not exist - creating' % res_dir)
if not os.path.isdir(vid_dir):
os.makedirs(vid_dir)
print('* %s does not exist - creating' % vid_dir)
# Set up the LBP cascade classifier
LBP_path = os.path.join(utils._package_root(), 'Cascade/cascade.xml')
print(' Loading LBP cascade')
cascade = cv2.CascadeClassifier(LBP_path)
if cascade.empty():
print('* LBP cascade is empty - mrgaze installation problem')
return False
# Check for output CSV existance and overwrite flag
if os.path.isfile(pupils_csv):
print('+ Pupilometry output already exists - checking overwrite flag')
if overwrite:
print('+ Overwrite allowed - continuing')
else:
print('+ Overwrite forbidden - skipping pupilometry')
return True
#
# Camera Input
#
print(' Opening camera stream')
try:
if not live_eyetracking:
vin_stream = cv2.VideoCapture(vin_path)
cal_vin_stream = cv2.VideoCapture(cal_vin_path)
else:
vin_stream = cv2.VideoCapture(vin_path)
cal_vin_stream = vin_stream
except:
print('* Problem opening input video stream - skipping pupilometry')
return False
while not vin_stream.isOpened():
print("Waiting for Camera.")
key = utils._waitKey(500)
if key == 'ESC':
print("User Abort.")
break
if not vin_stream.isOpened():
print('* Video input stream not opened - skipping pupilometry')
return False
if not cal_vin_stream.isOpened():
print('* Calibration video input stream not opened - skipping pupilometry')
return False
# Video FPS from metadata
# TODO: may not work with Quicktime videos
# fps = vin_stream.get(cv2.cv.CV_CAP_PROP_FPS)
# fps = cfg.getfloat('CAMERA', 'fps')
# Desired time between frames in milliseconds
# time_bw_frames = 1000.0 / fps
vin_stream.set(cv2.cv.CV_CAP_PROP_FRAME_WIDTH, 320)
vin_stream.set(cv2.cv.CV_CAP_PROP_FRAME_HEIGHT, 240)
vin_stream.set(cv2.cv.CV_CAP_PROP_FPS, 30)
# Total number of frames in video file
# nf = vin_stream.get(cv2.cv.CV_CAP_PROP_FRAME_COUNT)
# print(' Video has %d frames at %0.3f fps' % (nf, vin_fps))
# Read first preprocessed video frame from stream
keep_going, frame_orig = media.LoadVideoFrame(vin_stream, cfg)
if keep_going:
frame, art_power = media.Preproc(frame_orig, cfg)
else:
art_power = 0.0
# Get size of preprocessed frame for output video setup
nx, ny = frame.shape[1], frame.shape[0]
# By default we start in non-calibration mode
# switch between gaze/cal modes by pressing key "c"
do_cal = False
while keep_going:
if do_cal == False:
#
# Output video
#
if live_eyetracking:
print(' Opening output video stream')
# Output video codec (MP4V - poor quality compression)
# TODO : Find a better multiplatform codec
fourcc = cv2.cv.CV_FOURCC('m','p','4','v')
try:
vout_stream = cv2.VideoWriter(vout_path, fourcc, 30, (nx, ny), True)
except:
print('* Problem creating output video stream - skipping pupilometry')
return False
if not vout_stream.isOpened():
print('* Output video not opened - skipping pupilometry')
return False
# Open pupilometry CSV file to write
try:
pupils_stream = open(pupils_csv, 'w')
except:
print('* Problem opening pupilometry CSV file - skipping pupilometry')
return False
#
# Main Video Frame Loop
#
# Print verbose column headers
if verbose:
print('')
print(' %10s %10s %10s %10s %10s' % (
'Time (s)', 'Area', 'Blink', 'Artifact', 'FPS'))
# Init frame counter
fc = 0
# Init processing timer
t0 = time.time()
t = t0
while keep_going:
# check whether config file has been updated, reload of that is the case
if fc % 30 == 0:
cfg_mtime = os.path.getmtime(os.path.join(data_dir, 'mrgaze.cfg'))
if cfg_mtime > cfg_ts:
print("Updating Configuration")
cfg = config.LoadConfig(data_dir)
cfg_ts = time.time()
# Current video time in seconds
t = time.time()
# -------------------------------------
# Pass this frame to pupilometry engine
# -------------------------------------
# b4_engine = time.time()
pupil_ellipse, roi_rect, blink, glint, frame_rgb = engine.PupilometryEngine(frame, cascade, cfg)
# print "Enging took %s ms" % (time.time() - b4_engine)
# Derive pupilometry parameters
px, py, area = engine.PupilometryPars(pupil_ellipse, glint, cfg)
# Write data line to pupilometry CSV file
pupils_stream.write(
'%0.4f,%0.3f,%0.3f,%0.3f,%d,%0.3f,\n' %
(t, area, px, py, blink, art_power)
)
if live_eyetracking:
# Write output video frame
vout_stream.write(frame_orig)
# Read next frame, unless we want to figure out the correct settings for this frame
if not freeze_frame:
keep_going, frame_orig = media.LoadVideoFrame(vin_stream, cfg)
if keep_going:
frame, art_power = media.Preproc(frame_orig, cfg)
else:
art_power = 0.0
# Increment frame counter
fc = fc + 1
# Report processing FPS
if verbose:
if fc % 100 == 0:
pfps = fc / (time.time() - t0)
print(' %10.1f %10.1f %10d %10.3f %10.1f' % (
t, area, blink, art_power, pfps))
t0 = time.time()
fc = 0
# wait whether user pressed esc to exit the experiment
key = utils._waitKey(1)
if key == 'ESC':
# Clean up
if live_eyetracking:
vout_stream.release()
pupils_stream.close()
keep_going = False
elif key == 'c':
# Clean up
if live_eyetracking:
vout_stream.release()
pupils_stream.close()
do_cal = True
print("Starting calibration.")
break
elif key == 'f':
freeze_frame = not freeze_frame
else: # do calibration
#
# Output video
#
if live_eyetracking:
print(' Opening output video stream')
# Output video codec (MP4V - poor quality compression)
# TODO : Find a better multiplatform codec
fourcc = cv2.cv.CV_FOURCC('m','p','4','v')
try:
cal_vout_stream = cv2.VideoWriter(cal_vout_path, fourcc, 30, (nx, ny), True)
except:
print('* Problem creating output video stream - skipping pupilometry')
return False
if not cal_vout_stream.isOpened():
print('* Output video not opened - skipping pupilometry')
return False
# Open pupilometry CSV file to write
try:
cal_pupils_stream = open(cal_pupils_csv, 'w')
except:
print('* Problem opening pupilometry CSV file - skipping pupilometry')
return False
#
# Main Video Frame Loop
#
# Print verbose column headers
if verbose:
print('')
print(' %10s %10s %10s %10s %10s' % (
'Time (s)', 'Area', 'Blink', 'Artifact', 'FPS'))
# Init frame counter
fc = 0
# Init processing timer
t0 = time.time()
t = t0
while keep_going:
# check whether config file has been updated, reload of that is the case
if fc % 30 == 0:
cfg_mtime = os.path.getmtime(os.path.join(data_dir, 'mrgaze.cfg'))
if cfg_mtime > cfg_ts:
print("Updating Configuration")
cfg = config.LoadConfig(data_dir)
cfg_ts = time.time()
# Current video time in seconds
t = time.time()
# -------------------------------------
# Pass this frame to pupilometry engine
# -------------------------------------
# b4_engine = time.time()
pupil_ellipse, roi_rect, blink, glint, frame_rgb = engine.PupilometryEngine(frame, cascade, cfg)
# print "Engine took %s ms" % (time.time() - b4_engine)
# Derive pupilometry parameters
px, py, area = engine.PupilometryPars(pupil_ellipse, glint, cfg)
# Write data line to pupilometry CSV file
cal_pupils_stream.write(
'%0.4f,%0.3f,%0.3f,%0.3f,%d,%0.3f,\n' %
(t, area, px, py, blink, art_power)
)
# Write output video frame
if live_eyetracking:
cal_vout_stream.write(frame_orig)
# Read next frame (if available)
# if verbose:
# b4_frame = time.time()
keep_going, frame_orig = media.LoadVideoFrame(vin_stream, cfg)
if keep_going:
frame, art_power = media.Preproc(frame_orig, cfg)
else:
art_power = 0.0
#if verbose:
# print "Time to load frame: %s" % (time.time() - b4_frame)
# Increment frame counter
fc = fc + 1
# Report processing FPS
if verbose:
if fc % 100 == 0:
pfps = fc / (time.time() - t0)
print(' %10.1f %10.1f %10d %10.3f %10.1f' % (
t, area, blink, art_power, pfps))
t0 = time.time()
fc = 0
# wait whether user pressed esc to exit the experiment
key = utils._waitKey(1)
if key == 'ESC':
keep_going = False
# Clean up
if live_eyetracking:
cal_vout_stream.release()
cal_pupils_stream.close()
elif key == 'v' or not keep_going:
do_cal = False
print("Stopping calibration.")
# Clean up
if live_eyetracking:
cal_vout_stream.release()
cal_pupils_stream.close()
break
print(' Create calibration model')
C, central_fix = calibrate.AutoCalibrate(res_dir, cfg)
if not C.any():
print('* Empty calibration matrix detected - skipping')
try:
print(' Calibrate pupilometry')
calibrate.ApplyCalibration(ss_dir, C, central_fix, cfg)
except UnboundLocalError:
print(' No calibration data found')
cv2.destroyAllWindows()
vin_stream.release()
print('')
print(' Generate Report')
print(' ---------------')
report.WriteReport(ss_dir, cfg)
# Return pupilometry timeseries
return t, px, py, area, blink, art_power
