feat: pupil dilation work start

This commit is contained in:
Prohurtz 2023-10-05 09:29:54 -05:00
parent 6b7f5a63ca
commit e694054cb4
7 changed files with 626 additions and 112 deletions

View File

@ -16,8 +16,8 @@ WAIT_TIME = 0.1
# header-type (2 bytes)
# packet-size (2 bytes)
# packet (packet-size bytes)
ETVR_HEADER = b'\xff\xa0'
ETVR_HEADER_FRAME = b'\xff\xa1'
ETVR_HEADER = b"\xff\xa0"
ETVR_HEADER_FRAME = b"\xff\xa1"
ETVR_HEADER_LEN = 6
@ -26,15 +26,16 @@ class CameraState(Enum):
CONNECTED = 1
DISCONNECTED = 2
class Camera:
def __init__(
self,
config: EyeTrackConfig,
camera_index: int,
cancellation_event: "threading.Event",
capture_event: "threading.Event",
camera_status_outgoing: "queue.Queue[CameraState]",
camera_output_outgoing: "queue.Queue",
self,
config: EyeTrackConfig,
camera_index: int,
cancellation_event: "threading.Event",
capture_event: "threading.Event",
camera_status_outgoing: "queue.Queue[CameraState]",
camera_output_outgoing: "queue.Queue(maxsize=2)",
):
self.camera_status = CameraState.CONNECTING
self.config = config
@ -53,7 +54,7 @@ class Camera:
self.fps = 0
self.bps = 0
self.start = True
self.buffer = b''
self.buffer = b""
self.pf_fps = 0
self.prevft = 0
self.newft = 0
@ -70,8 +71,10 @@ class Camera:
def run(self):
OPENCV_PARAMS = [
cv2.CAP_PROP_OPEN_TIMEOUT_MSEC, 5000,
cv2.CAP_PROP_READ_TIMEOUT_MSEC, 5000,
cv2.CAP_PROP_OPEN_TIMEOUT_MSEC,
5000,
cv2.CAP_PROP_READ_TIMEOUT_MSEC,
5000,
]
while True:
if self.cancellation_event.is_set():
@ -80,25 +83,23 @@ class Camera:
should_push = True
# If things aren't open, retry until they are. Don't let read requests come in any earlier
# than this, otherwise we can deadlock ourselves.
if (
self.config.capture_source != None and self.config.capture_source != ""
):
if self.config.capture_source != None and self.config.capture_source != "":
if "COM" in str(self.current_capture_source):
if (
self.serial_connection is None
or self.camera_status == CameraState.DISCONNECTED
or self.config.capture_source != self.current_capture_source
self.serial_connection is None
or self.camera_status == CameraState.DISCONNECTED
or self.config.capture_source != self.current_capture_source
):
port = self.config.capture_source
self.current_capture_source = port
self.start_serial_connection(port)
else:
if (
self.cv2_camera is None
or not self.cv2_camera.isOpened()
or self.camera_status == CameraState.DISCONNECTED
or self.config.capture_source != self.current_capture_source
self.cv2_camera is None
or not self.cv2_camera.isOpened()
or self.camera_status == CameraState.DISCONNECTED
or self.config.capture_source != self.current_capture_source
):
print(self.error_message.format(self.config.capture_source))
# This requires a wait, otherwise we can error and possible screw up the camera
@ -106,7 +107,7 @@ class Camera:
if self.cancellation_event.wait(WAIT_TIME):
return
self.current_capture_source = self.config.capture_source
# self.cv2_camera = cv2.VideoCapture(self.current_capture_source)
# self.cv2_camera = cv2.VideoCapture(self.current_capture_source)
self.cv2_camera = cv2.VideoCapture()
self.cv2_camera.setExceptionMode(True)
@ -137,7 +138,9 @@ class Camera:
ret, image = self.cv2_camera.read()
height, width = image.shape[:2] # Calculate the aspect ratio
if int(width) > 680:
aspect_ratio = float(width) / float(height) # Determine the new height based on the desired maximum width
aspect_ratio = float(width) / float(
height
) # Determine the new height based on the desired maximum width
new_height = int(680 / aspect_ratio)
image = cv2.resize(image, (680, new_height))
if not ret:
@ -162,11 +165,13 @@ class Camera:
self.fl.pop(0)
self.fl.append(self.fps)
self.fps = sum(self.fl) / len(self.fl)
#self.bps = image.nbytes
# self.bps = image.nbytes
if should_push:
self.push_image_to_queue(image, frame_number, self.fps)
except:
print(f"{Fore.YELLOW}[WARN] Capture source problem, assuming camera disconnected, waiting for reconnect.{Fore.RESET}")
print(
f"{Fore.YELLOW}[WARN] Capture source problem, assuming camera disconnected, waiting for reconnect.{Fore.RESET}"
)
self.camera_status = CameraState.DISCONNECTED
pass
@ -186,8 +191,8 @@ class Camera:
def get_next_jpeg_frame(self):
beg, end = self.get_next_packet_bounds()
jpeg = self.buffer[beg+ETVR_HEADER_LEN:end+ETVR_HEADER_LEN]
self.buffer = self.buffer[end+ETVR_HEADER_LEN:]
jpeg = self.buffer[beg + ETVR_HEADER_LEN : end + ETVR_HEADER_LEN]
self.buffer = self.buffer[end + ETVR_HEADER_LEN :]
return jpeg
def get_serial_camera_picture(self, should_push):
@ -199,16 +204,22 @@ class Camera:
jpeg = self.get_next_jpeg_frame()
if jpeg:
# Create jpeg frame from byte string
image = cv2.imdecode(np.fromstring(jpeg, dtype=np.uint8), cv2.IMREAD_UNCHANGED)
image = cv2.imdecode(
np.fromstring(jpeg, dtype=np.uint8), cv2.IMREAD_UNCHANGED
)
if image is None:
print(f"{Fore.YELLOW}[WARN] Frame drop. Corrupted JPEG.{Fore.RESET}")
print(
f"{Fore.YELLOW}[WARN] Frame drop. Corrupted JPEG.{Fore.RESET}"
)
return
# Discard the serial buffer. This is due to the fact that it
# may build up some outdated frames. A bit of a workaround here tbh.
if conn.in_waiting >= 32768:
print(f"{Fore.CYAN}[INFO] Discarding the serial buffer ({conn.in_waiting} bytes){Fore.RESET}")
print(
f"{Fore.CYAN}[INFO] Discarding the serial buffer ({conn.in_waiting} bytes){Fore.RESET}"
)
conn.reset_input_buffer()
self.buffer = b''
self.buffer = b""
# Calculate the fps.
current_frame_time = time.time()
delta_time = current_frame_time - self.last_frame_time
@ -230,7 +241,9 @@ class Camera:
if should_push:
self.push_image_to_queue(image, self.frame_number, self.fps)
except Exception:
print(f"{Fore.YELLOW}[WARN] Serial capture source problem, assuming camera disconnected, waiting for reconnect.{Fore.RESET}")
print(
f"{Fore.YELLOW}[WARN] Serial capture source problem, assuming camera disconnected, waiting for reconnect.{Fore.RESET}"
)
conn.close()
self.camera_status = CameraState.DISCONNECTED
pass
@ -248,15 +261,14 @@ class Camera:
return
try:
conn = serial.Serial(
baudrate = 3000000,
port = port,
xonxoff=False,
dsrdtr=False,
rtscts=False)
baudrate=3000000, port=port, xonxoff=False, dsrdtr=False, rtscts=False
)
# Set explicit buffer size for serial.
conn.set_buffer_size(rx_size = 32768, tx_size = 32768)
conn.set_buffer_size(rx_size=32768, tx_size=32768)
print(f"{Fore.CYAN}[INFO] ETVR Serial Tracker device connected on {port}{Fore.RESET}")
print(
f"{Fore.CYAN}[INFO] ETVR Serial Tracker device connected on {port}{Fore.RESET}"
)
self.serial_connection = conn
self.camera_status = CameraState.CONNECTED
except Exception:
@ -269,6 +281,7 @@ class Camera:
qsize = self.camera_output_outgoing.qsize()
if qsize > 1:
print(
f"{Fore.YELLOW}[WARN] CAPTURE QUEUE BACKPRESSURE OF {qsize}. CHECK FOR CRASH OR TIMING ISSUES IN ALGORITHM.{Fore.RESET}")
f"{Fore.YELLOW}[WARN] CAPTURE QUEUE BACKPRESSURE OF {qsize}. CHECK FOR CRASH OR TIMING ISSUES IN ALGORITHM.{Fore.RESET}"
)
self.camera_output_outgoing.put((image, frame_number, fps))
self.capture_event.clear()
self.capture_event.clear()

View File

@ -10,9 +10,11 @@ from camera import Camera, CameraState
from osc import EyeId
import cv2
import sys
from utils.misc_utils import PlaySound, SND_FILENAME,SND_ASYNC, resource_path
from utils.misc_utils import PlaySound, SND_FILENAME, SND_ASYNC, resource_path
import traceback
import numpy as np
class CameraWidget:
def __init__(self, widget_id: EyeId, main_config: EyeTrackConfig, osc_queue: Queue):
self.gui_camera_addr = f"-CAMERAADDR{widget_id}-"
@ -46,7 +48,9 @@ class CameraWidget:
elif self.eye_id == EyeId.LEFT:
self.config = main_config.left_eye
else:
raise RuntimeError("\033[91m[WARN] Cannot have a camera widget represent both eyes!\033[0m")
raise RuntimeError(
"\033[91m[WARN] Cannot have a camera widget represent both eyes!\033[0m"
)
self.cancellation_event = Event()
# Set the event until start is called, otherwise we can block if shutdown is called.
@ -80,14 +84,14 @@ class CameraWidget:
self.roi_layout = [
[
sg.Graph(
sg.Graph(
(640, 480),
(0, 480),
(640, 0),
key=self.gui_roi_selection,
drag_submits=True,
enable_events=True,
background_color='#424042',
background_color="#424042",
)
]
]
@ -95,34 +99,50 @@ class CameraWidget:
# Define the window's contents
self.tracking_layout = [
[
sg.Text("Rotation", background_color='#424042'),
sg.Text("Rotation", background_color="#424042"),
sg.Slider(
range=(0, 360),
default_value=self.config.rotation_angle,
orientation="h",
key=self.gui_rotation_slider,
background_color='#424042',
tooltip = "Adjust the rotation of your cameras, make them level.",
background_color="#424042",
tooltip="Adjust the rotation of your cameras, make them level.",
),
],
[
sg.Button("Start Calibration", key=self.gui_restart_calibration, button_color='#6f4ca1', tooltip = "Start eye calibration. Look all arround to all extreams without blinking until sound is heard.",),
sg.Button("Stop Calibration", key=self.gui_stop_calibration, button_color='#6f4ca1', tooltip = "Stop eye calibration manualy.",),
sg.Button("Recenter Eyes", key=self.gui_recenter_eyes, button_color='#6f4ca1', tooltip = "Make your eyes center again.",),
sg.Button(
"Start Calibration",
key=self.gui_restart_calibration,
button_color="#6f4ca1",
tooltip="Start eye calibration. Look all arround to all extreams without blinking until sound is heard.",
),
sg.Button(
"Stop Calibration",
key=self.gui_stop_calibration,
button_color="#6f4ca1",
tooltip="Stop eye calibration manualy.",
),
sg.Button(
"Recenter Eyes",
key=self.gui_recenter_eyes,
button_color="#6f4ca1",
tooltip="Make your eyes center again.",
),
],
[
sg.Text("Mode:", background_color='#424042'),
sg.Text("Calibrating", key=self.gui_mode_readout, background_color='#424042'),
sg.Text("", key=self.gui_tracking_fps, background_color='#424042'),
sg.Text("", key=self.gui_tracking_bps, background_color='#424042'),
# sg.Checkbox(
# "Circle crop:",
# default=self.config.gui_circular_crop,
# key=self.gui_circular_crop,
# background_color='#424042',
# tooltip = "Circle crop only applies to RANSAC3D and Blob.",
# ),
sg.Text("Mode:", background_color="#424042"),
sg.Text(
"Calibrating", key=self.gui_mode_readout, background_color="#424042"
),
sg.Text("", key=self.gui_tracking_fps, background_color="#424042"),
sg.Text("", key=self.gui_tracking_bps, background_color="#424042"),
# sg.Checkbox(
# "Circle crop:",
# default=self.config.gui_circular_crop,
# key=self.gui_circular_crop,
# background_color='#424042',
# tooltip = "Circle crop only applies to RANSAC3D and Blob.",
# ),
],
[sg.Image(filename="", key=self.gui_tracking_image)],
[
@ -135,25 +155,57 @@ class CameraWidget:
drag_submits=True,
enable_events=True,
),
sg.Text("Please set an Eye Cropping.", key=self.gui_roi_message, background_color='#424042', visible=False),
sg.Text(
"Please set an Eye Cropping.",
key=self.gui_roi_message,
background_color="#424042",
visible=False,
),
],
]
self.widget_layout = [
[
sg.Text("Camera Address", background_color='#424042'),
sg.InputText(self.config.capture_source, key=self.gui_camera_addr, tooltip = "Enter the IP address or UVC port of your camera. (Include the 'http://')",),
sg.Text("Camera Address", background_color="#424042"),
sg.InputText(
self.config.capture_source,
key=self.gui_camera_addr,
tooltip="Enter the IP address or UVC port of your camera. (Include the 'http://')",
),
],
[
sg.Button("Save and Restart Tracking", key=self.gui_save_tracking_button, button_color='#6f4ca1'),
sg.Button(
"Save and Restart Tracking",
key=self.gui_save_tracking_button,
button_color="#6f4ca1",
),
],
[
sg.Button("Tracking Mode", key=self.gui_tracking_button, button_color='#6f4ca1', tooltip = "Go here to track your eye.",),
sg.Button("Cropping Mode", key=self.gui_roi_button, button_color='#6f4ca1', tooltip = "Go here to crop out your eye.",),
sg.Button(
"Tracking Mode",
key=self.gui_tracking_button,
button_color="#6f4ca1",
tooltip="Go here to track your eye.",
),
sg.Button(
"Cropping Mode",
key=self.gui_roi_button,
button_color="#6f4ca1",
tooltip="Go here to crop out your eye.",
),
],
[
sg.Column(self.tracking_layout, key=self.gui_tracking_layout, background_color='#424042'),
sg.Column(self.roi_layout, key=self.gui_roi_layout, background_color='#424042', visible=False),
sg.Column(
self.tracking_layout,
key=self.gui_tracking_layout,
background_color="#424042",
),
sg.Column(
self.roi_layout,
key=self.gui_roi_layout,
background_color="#424042",
visible=False,
),
],
]
@ -203,7 +255,11 @@ class CameraWidget:
event == self.gui_save_tracking_button
and values[self.gui_camera_addr] != self.config.capture_source
):
print("\033[94m[INFO] New value: {}\033[0m".format(values[self.gui_camera_addr]))
print(
"\033[94m[INFO] New value: {}\033[0m".format(
values[self.gui_camera_addr]
)
)
try:
# Try storing ints as ints, for those using wired cameras.
self.config.capture_source = int(values[self.gui_camera_addr])
@ -211,20 +267,24 @@ class CameraWidget:
if values[self.gui_camera_addr] == "":
self.config.capture_source = None
else:
if len(values[self.gui_camera_addr]) > 5 and "http" not in values[self.gui_camera_addr] and ".mp4" not in values[self.gui_camera_addr]: # If http is not in camera address, add it.
self.config.capture_source = f"http://{values[self.gui_camera_addr]}/"
if (
len(values[self.gui_camera_addr]) > 5
and "http" not in values[self.gui_camera_addr]
and ".mp4" not in values[self.gui_camera_addr]
): # If http is not in camera address, add it.
self.config.capture_source = (
f"http://{values[self.gui_camera_addr]}/"
)
else:
self.config.capture_source = values[self.gui_camera_addr]
changed = True
if self.config.rotation_angle != values[self.gui_rotation_slider]:
self.config.rotation_angle = int(values[self.gui_rotation_slider])
changed = True
# if self.config.gui_circular_crop != values[self.gui_circular_crop]:
# self.config.gui_circular_crop = values[self.gui_circular_crop]
# if self.config.gui_circular_crop != values[self.gui_circular_crop]:
# self.config.gui_circular_crop = values[self.gui_circular_crop]
# changed = True
if changed:
@ -248,16 +308,15 @@ class CameraWidget:
# Event for mouse button up in ROI mode
self.is_mouse_up = True
if self.x1 < 0:
self.x1 = 0
self.x1 = 0
if self.y1 < 0:
self.y1 = 0
self.y1 = 0
if abs(self.x0 - self.x1) != 0 and abs(self.y0 - self.y1) != 0:
self.config.roi_window_x = min([self.x0, self.x1])
self.config.roi_window_y = min([self.y0, self.y1])
self.config.roi_window_w = abs(self.x0 - self.x1)
self.config.roi_window_h = abs(self.y0 - self.y1)
self.main_config.save()
if event == self.gui_roi_selection:
# Event for mouse button down or mouse drag in ROI mode
@ -269,7 +328,7 @@ class CameraWidget:
if event == self.gui_restart_calibration:
self.ransac.calibration_frame_counter = self.settings.calibration_samples
self.ransac.ibo.clear_filter()
PlaySound(resource_path('Audio/start.wav'), SND_FILENAME | SND_ASYNC)
PlaySound(resource_path("Audio/start.wav"), SND_FILENAME | SND_ASYNC)
if event == self.gui_stop_calibration:
self.ransac.calibration_frame_counter = 0
@ -280,8 +339,8 @@ class CameraWidget:
needs_roi_set = self.config.roi_window_h <= 0 or self.config.roi_window_w <= 0
# TODO: Refactor if statements below...
window[self.gui_tracking_fps].update('')
window[self.gui_tracking_bps].update('')
window[self.gui_tracking_fps].update("")
window[self.gui_tracking_bps].update("")
if self.config.capture_source is None or self.config.capture_source == "":
window[self.gui_mode_readout].update("Waiting for camera address")
window[self.gui_roi_message].update(visible=False)
@ -329,6 +388,7 @@ class CameraWidget:
window[self.gui_roi_message].update(visible=False)
window[self.gui_output_graph].update(visible=True)
(maybe_image, eye_info) = self.image_queue.get(block=False)
imgbytes = cv2.imencode(".ppm", maybe_image)[1].tobytes()
window[self.gui_tracking_image].update(data=imgbytes)
@ -336,13 +396,15 @@ class CameraWidget:
graph = window[self.gui_output_graph]
graph.erase()
if eye_info.info_type != EyeInfoOrigin.FAILURE: #and not eye_info.blink:
if (
eye_info.info_type != EyeInfoOrigin.FAILURE
): # and not eye_info.blink:
graph.update(background_color="white")
if not np.isnan(eye_info.x) and not np.isnan(eye_info.y):
graph.draw_circle(
(eye_info.x * -100, eye_info.y * -100),
20,
15],
fill_color="black",
line_color="white",
)
@ -356,12 +418,20 @@ class CameraWidget:
if not np.isnan(eye_info.blink):
graph.draw_line((-100, eye_info.blink * 2 * 200), (-100, 100), color="#6f4ca1", width=10)
graph.draw_line(
(-100, abs(eye_info.blink) * 2 * 200),
(-100, 100),
color="#6f4ca1",
width=10,
)
else:
graph.draw_line((-100, 0.5 * 200), (-100, 100), color="#6f4ca1", width=10)
graph.draw_line(
(-100, 0.5 * 200), (-100, 100), color="#6f4ca1", width=10
)
if eye_info.blink <= 0.0:
graph.update(background_color="#6f4ca1")
elif eye_info.info_type == EyeInfoOrigin.FAILURE:
graph.update(background_color="red")
# Relay information to OSC

View File

@ -0,0 +1,399 @@
"""
------------------------------------------------------------------------------------------------------
,@@@@@@
@@@@@@@@@@@ @@@
@@@@@@@@@@@@ @@@@@@@@@@@
@@@@@@@@@@@@@ @@@@@@@@@@@@@@
@@@@@@@/ ,@@@@@@@@@@@@@
/@@@@@@@@@@@@@@@ @@@@@@@@
@@@@@@@@@@@@@@@@@@@@@@@@ @@@@@
@@@@@@@@ @@@@@
,@@@ @@@@&
@@@@@@. @@@@
@@@ @@@@@@@@@/ @@@@@
,@@@. @@@@@@((@ @@@@(
//@@@ ,, @@@@ @@@@@
@@@( @@@@@@@
@@@ @ @@@@@@@@#
@@@@@@@@@@@@@@@@@
@@@@@@@@@@@@@(
Ellipse Based Pupil Dilation By: Prohurtz, PallasNeko (Optimization)
Algorithm App Implementations By: Prohurtz
Copyright (c) 2023 EyeTrackVR <3
------------------------------------------------------------------------------------------------------
"""
import numpy
import numpy as np
import time
import os
import cv2
from enums import EyeLR
from one_euro_filter import OneEuroFilter
from utils.img_utils import safe_crop
from enum import IntEnum
import psutil
import sys
process = psutil.Process(os.getpid()) # set process priority to low
try: # medium chance this does absolutely nothing but eh
sys.getwindowsversion()
except AttributeError:
process.nice(0) # UNIX: 0 low 10 high
process.nice()
else:
process.nice(psutil.BELOW_NORMAL_PRIORITY_CLASS) # Windows
process.nice()
class EyeId(IntEnum):
RIGHT = 0
LEFT = 1
BOTH = 2
SETTINGS = 3
# Note.
# OpenCV on Windows will generate an error if the file path contains non-ASCII characters when using cv2.imread(), cv2.imwrite(), etc.
# https://stackoverflow.com/questions/43185605/how-do-i-read-an-image-from-a-path-with-unicode-characters
# https://github.com/opencv/opencv/issues/18305
def csv2data(frameshape, filepath):
# For data checking
frameshape = (frameshape[0], frameshape[1] + 1)
out = np.zeros(frameshape, dtype=np.uint32)
xy_list = []
val_list = []
with open(filepath, mode="r", encoding="utf-8") as in_f:
# Skip header.
_ = in_f.readline()
for s in in_f:
xyval = [int(val) for val in s.strip().split(",")]
xy_list.append((xyval[0], xyval[1]))
val_list.append(xyval[2])
xy_list = np.array(xy_list)
val_list = np.array(val_list)
out[xy_list[:, 1], xy_list[:, 0]] = val_list[:]
return out
def data2csv(data_u32, filepath):
# For data checking
nonzero_index = np.nonzero(data_u32) # (row,col)
data_list = data_u32[nonzero_index].tolist()
datalines = [
"{},{},{}\n".format(x, y, val) for y, x, val in zip(*nonzero_index, data_list)
]
with open(filepath, "w", encoding="utf-8") as out_f:
out_f.write("x,y,eyedilation\n")
out_f.writelines(datalines)
return
def u32_1ch_to_u16_3ch(img):
out = np.zeros((*img.shape[:2], 3), dtype=np.uint16)
# https://github.com/numpy/numpy/issues/2524
# https://stackoverflow.com/questions/52782511/why-is-numpy-slower-than-python-for-left-bit-shifts
out[:, :, 0] = img & np.uint32(65535)
out[:, :, 1] = (img >> np.uint32(16)) & np.uint32(65535)
return out
def u16_3ch_to_u32_1ch(img):
# The image format with the most bits that can be displayed on Windows without additional software and that opencv can handle is PNG's uint16
out = img[:, :, 0].astype(np.float64) # float64 = max 2^53
cv2.add(
out, img[:, :, 1].astype(np.float64) * np.float64(65536), dst=out
) # opencv did not have uint32 type
return out.astype(np.uint32) # cast
def newdata(frameshape):
print("\033[94m[INFO] Initialise data for dilation.\033[0m")
return np.zeros(frameshape, dtype=np.uint32)
# EBPD
class EllipseBasedPupilDilation:
def __init__(self, eye_id):
# todo: It is necessary to consider whether the filename can be changed in the configuration file, etc.
if eye_id in [EyeId.LEFT]:
self.imgfile = "EBPD_LEFT.png"
else:
pass
if eye_id in [EyeId.RIGHT]:
self.imgfile = "EBPD_RIGHT.png"
else:
pass
# self.data[0, -1] = maxval, [1, -1] = rotation, [2, -1] = x, [3, -1] = y
self.data = None
self.lct = None
self.maxval = 0
# self.img_roi = self.now_roi == {"rotation": 0, "x": 0, "y": 0}
self.img_roi = np.zeros(3, dtype=np.int32)
self.now_roi = np.zeros(3, dtype=np.int32)
self.prev_val = 0.5
self.avg_dilation = 0.0
self.old = []
self.color = []
self.x = []
self.fc = 0
self.filterlist = []
self.averageList = []
self.openlist = []
self.eye_id = eye_id
self.maxinten = 0
self.tri_filter = []
# try:
# min_cutoff = float(self.settings.gui_min_cutoff) # 0.0004
# beta = float(self.settings.gui_speed_coefficient) # 0.9
# except:
print("\033[93m[WARN] OneEuroFilter values must be a legal number.\033[0m")
min_cutoff = 0.0004
beta = 0.9
noisy_point = np.array([1, 1])
self.one_euro_filter = OneEuroFilter(
noisy_point, min_cutoff=min_cutoff, beta=beta
)
def check(self, frameshape):
# 0 in data is used as the initial value.
# When assigning a value, +1 is added to the value to be assigned.
self.load(frameshape)
# self.maxval = self.data[0, -1]
if self.lct is None:
self.lct = time.time()
def load(self, frameshape):
req_newdata = False
# Not very clever, but increase the width by 1px to save the maximum value.
frameshape = (frameshape[0], frameshape[1] + 1)
if self.data is None:
print(
f"\033[92m[INFO] Loaded data for pupil dilation: {self.imgfile}\033[0m"
)
if os.path.isfile(self.imgfile):
try:
img = cv2.imread(self.imgfile, flags=cv2.IMREAD_UNCHANGED)
# check code: cv2.absdiff(img,u32_1ch_to_u16_3ch(u16_3ch_to_u32_1ch(img)))
if img.shape[:2] != frameshape:
print("[WARN] Size does not match the input frame.")
req_newdata = True
else:
self.data = u16_3ch_to_u32_1ch(img)
self.img_roi[:] = self.data[1:4, -1]
if not np.array_equal(self.img_roi, self.now_roi):
# If the ROI recorded in the image file differs from the current ROI
req_newdata = True
else:
self.maxval = self.data[0, -1]
except:
print("[ERROR] File read error: {}".format(self.imgfile))
req_newdata = True
else:
print("\033[94m[INFO] File does not exist.\033[0m")
req_newdata = True
else:
if self.data.shape != frameshape or not np.array_equal(
self.img_roi, self.now_roi
):
# If the ROI recorded in the image file differs from the current ROI
# todo: Using the previous and current frame sizes and centre positions from the original, etc., the data can be ported to some extent, but there may be many areas where code changes are required.
print("[INFO] \033[94mFrame size changed.\033[0m")
req_newdata = True
if req_newdata:
self.data = newdata(frameshape)
self.maxval = 0
self.img_roi = self.now_roi.copy()
# data2csv(self.data, "a.csv")
# csv2data(frameshape,"a.csv")
def save(self):
self.data[0, -1] = self.maxval
self.data[1:4, -1] = self.now_roi
cv2.imwrite(self.imgfile, u32_1ch_to_u16_3ch(self.data))
# print("SAVED: {}".format(self.imgfile))
def change_roi(self, roiinfo: dict):
self.now_roi[:] = [v for v in roiinfo.values()]
def clear_filter(self):
self.data = None
self.filterlist.clear()
self.averageList.clear()
if os.path.exists(self.imgfile):
os.remove(self.imgfile)
def intense(self, w, h, x, y, frame, filterSamples, outputSamples):
# x,y = 0~(frame.shape[1 or 0]-1), frame = 1-channel frame cropped by ROI
self.check(frame.shape)
int_x, int_y = int(x), int(y)
if int_x < 0 or int_y < 0:
return self.prev_val
upper_x = min(
int_x + 25, frame.shape[1] - 1
) # TODO make this a setting NEEDS TO BE BASED ON HSF RADIUS if possible
lower_x = max(int_x - 25, 0)
upper_y = min(int_y + 25, frame.shape[0] - 1)
lower_y = max(int_y - 25, 0)
# frame_crop = frame[lower_y:upper_y, lower_x:upper_x]
# frame = safe_crop(frame, lower_x, lower_y, upper_x, upper_y, False)
# ret_, th = cv2.threshold(frame_crop, 80, 1.0, cv2.THRESH_BINARY_INV, dst=frame_crop)
frame_crop = frame
# ret, f = cv2.threshold(frame, 80, 255, cv2.THRESH_BINARY)
# ret, frame_crop = cv2.threshold(frame_crop, 80, 255, cv2.THRESH_BINARY)
# The same can be done with cv2.integral, but since there is only one area of the rectangle for which we want to know the total value, there is no advantage in terms of computational complexity.
pupil_area = numpy.pi * (w / 2) * (h / 2)
# cv2.imshow('e', frame)
# if cv2.waitKey(10) == 27:
# exit()
if len(self.filterlist) < filterSamples:
self.filterlist.append(pupil_area)
else:
self.filterlist.pop(0)
self.filterlist.append(pupil_area)
try:
if pupil_area >= np.percentile(
self.filterlist, 98
): # filter abnormally high values
# print('filter, assume blink')
pupil_area = self.maxval
# if intensity <= np.percentile( # TODO test this
# self.filterlist, 0.3
# ): # filter abnormally low values
# print('filter, assume blink')
# intensity = self.data[int_y, int_x]
except:
pass
# self.tri_filter.append(intensity)
# if len(self.tri_filter) > 3:
# self.tri_filter.pop(0)
# intensity = sum(self.tri_filter) / 3
# avg_color_per_row = np.average(frame_crop, axis=0)
# avg_color = np.average(avg_color_per_row, axis=0)
# ar, ag, ab = avg_color
# intensity = int(ar * 8) #higher = closed
# if cv2.waitKey(1) & 0xFF == ord("q"):
# pass
# numpy:np.sum(),ndarray.sum()
# opencv:cv2.sumElems()
# I don't know which is faster.
changed = False
newval_flg = False
oob = False
if int_x >= frame.shape[1]:
int_x = frame.shape[1] - 1
oob = True
# print('CAUGHT X OUT OF BOUNDS')
if int_x < 0:
int_x = True
oob = True
# print('CAUGHT X UNDER BOUNDS')
if int_y >= frame.shape[0]:
int_y = frame.shape[0] - 1
oob = True
# print('CAUGHT Y OUT OF BOUNDS')
if int_y < 0:
int_y = 1
oob = True
# print('CAUGHT Y UNDER BOUNDS')
if oob != True and self.data.any():
data_val = self.data[int_y, int_x]
else:
data_val = 0
# max pupil per cord
if data_val == 0:
# The value of the specified coordinates has not yet been recorded.
self.data[int_y, int_x] = pupil_area
changed = True
newval_flg = True
else:
if (
pupil_area < data_val
): # if current intensity value is less (more pupil), save that
self.data[int_y, int_x] = pupil_area # set value
changed = True
else:
pupil_areaa = max(
data_val + 5000, 1
) # if current intensity value is not less use this is an agressive adjust, test
self.data[int_y, int_x] = pupil_areaa # set value
changed = True
# min pupil global
if self.maxval == 0: # that value is not yet saved
self.maxval = pupil_area # set value at 0 index
else:
if (
pupil_area > self.maxval
): # if current intensity value is more (less pupil), save that NOTE: we have the
self.maxval = pupil_area - 5 # set value at 0 index
else:
pupil_aread = max(
(self.maxval - 5), 1
) # continuously adjust closed intensity, will be set when user blink, used to allow eyes to close when lighting changes
self.maxval = pupil_aread # set value at 0 index
# print(intensityd, intensity)
if newval_flg:
# Do the same thing as in the original version.
eyedilation = self.prev_val # 0.9
else:
maxp = float(self.data[int_y, int_x])
minp = float(self.maxval)
eyedilation = (pupil_area - maxp) / (
minp - maxp
) # for whatever reason when input and maxp are too close it outputs high
eyedilation = 1 - eyedilation
if outputSamples > 0:
if len(self.averageList) < outputSamples:
self.averageList.append(eyedilation)
else:
self.averageList.pop(0)
self.averageList.append(eyedilation)
eyedilation = np.average(self.averageList)
if eyedilation > 1: # clamp values
eyedilation = 1.0
if eyedilation < 0:
eyedilation = 0.0
if changed and (
(time.time() - self.lct) > 5
): # save every 5 seconds if something changed to save disk usage
self.save()
self.lct = time.time()
self.prev_val = eyedilation
try:
noisy_point = np.array(
[float(eyedilation), float(eyedilation)]
) # fliter our values with a One Euro Filter
point_hat = self.one_euro_filter(noisy_point)
eyedilationx = point_hat[0]
eyedilationy = point_hat[1]
eyedilation = (eyedilationx + eyedilationy) / 2
except:
pass
return eyedilation

View File

@ -59,6 +59,7 @@ from blink import *
from utils.img_utils import circle_crop
from eye import EyeInfo, EyeInfoOrigin
from intensity_based_openness import *
from ellipse_based_pupil_dilation import *
def run_once(f):
@ -140,6 +141,7 @@ class EyeProcessor:
self.er_daddy = None
self.er_leap = None
self.ibo = IntensityBasedOpeness(self.eye_id)
self.ebpd = EllipseBasedPupilDilation(self.eye_id)
self.roi_include_set = {"rotation_angle", "roi_window_x", "roi_window_y"}
self.failed = 0
self.skip_blink_detect = False
@ -164,6 +166,8 @@ class EyeProcessor:
self.ran_blink_check_for_file = True
self.bd_blink = False
self.current_algo = EyeInfoOrigin.HSRAC
self.puipil_width = 0.0
self.pupil_height = 0.0
try:
min_cutoff = float(self.settings.gui_min_cutoff) # 0.0004
@ -249,7 +253,18 @@ class EyeProcessor:
pass
def UPDATE(self):
# print(self.eyeopen)
# print(self.eyeopen)
self.pupil_dilation = self.ebpd.intense(
self.pupil_width,
self.pupil_height,
self.rawx,
self.rawy,
self.current_image_white,
self.settings.ibo_filter_samples,
self.settings.ibo_average_output_samples,
)
print(self.pupil_dilation)
if self.settings.gui_BLINK:
self.eyeopen = BLINK(self)
@ -288,7 +303,7 @@ class EyeProcessor:
self.rawy,
self.eyeopen,
) = self.er_leap.run(self.current_image_gray)
# print(self.eyeopen)
# print(self.eyeopen)
if (
len(self.prev_y_list) >= 100
@ -301,9 +316,9 @@ class EyeProcessor:
# print(abs(self.eyeopen - self.past_blink))
blink_vec = min(abs(self.eyeopen - self.past_blink), 1) # clamp to 1
#if blink_vec >= 0.2:
# if blink_vec >= 0.2:
if blink_vec >= 0.15 or blink_vec == 0.0 and (self.out_y - self.prev_y) < 0.0:
#self.out_x = sum(self.prev_x_list) / len(self.prev_x_list)
# self.out_x = sum(self.prev_x_list) / len(self.prev_x_list)
self.out_y = sum(self.prev_y_list) / len(self.prev_y_list)
# print('AVG', self.out_y, len(self.prev_y_list))
@ -313,7 +328,13 @@ class EyeProcessor:
self.output_images_and_update(
self.thresh,
EyeInfo(self.current_algo, self.out_x, self.out_y, 0, self.eyeopen),
EyeInfo(
self.current_algo,
self.out_x,
self.out_y,
self.pupil_dilation,
self.eyeopen,
),
)
if self.settings.gui_RANSACBLINK and self.eyeopen == 0.0:
pass
@ -331,7 +352,8 @@ class EyeProcessor:
self.thresh = self.current_image_gray.copy()
self.out_x, self.out_y = cal.cal_osc(self, self.rawx, self.rawy)
self.current_algorithm = EyeInfoOrigin.LEAP
# print(self.eyeopen)
# print(self.eyeopen)
def DADDYM(self):
# todo: We should have a proper variable for drawing.
@ -361,10 +383,17 @@ class EyeProcessor:
self.rawx, self.rawy, self.thresh, self.radius = self.er_hsf.run(
self.current_image_gray
)
self.rawx, self.rawy, self.thresh, ranblink = RANSAC3D(self, True)
(
self.rawx,
self.rawy,
self.thresh,
ranblink,
self.pupil_width,
self.pupil_height,
) = RANSAC3D(self, True)
if self.settings.gui_RANSACBLINK: # might be redundant
self.eyeopen = ranblink
# print("RANBLINK", ranblink)
# print("RANBLINK", ranblink)
# print(self.radius)
# if self.prev_x is None:

View File

@ -13,7 +13,7 @@ from utils.misc_utils import is_nt, resource_path
if is_nt:
from winotify import Notification
os.system('color') # init ANSI color
os.system("color") # init ANSI color
# Random environment variable to speed up webcam opening on the MSMF backend.
# https://github.com/opencv/opencv/issues/17687
@ -30,7 +30,7 @@ SETTINGS_RADIO_NAME = "-SETTINGSRADIO-"
ALGO_SETTINGS_RADIO_NAME = "-ALGOSETTINGSRADIO-"
page_url = "https://github.com/RedHawk989/EyeTrackVR/releases/latest"
appversion = "EyeTrackApp 0.2.0 BETA 7"
appversion = "EyeTrackApp 0.2.0 BETA 8"
def main():
@ -49,7 +49,7 @@ def main():
)
latestversion = response.json()["name"]
if (
appversion == latestversion
appversion == latestversion
): # If what we scraped and hardcoded versions are same, assume we are up to date.
print(f"\033[92m[INFO] App is the latest version! [{latestversion}]\033[0m")
else:
@ -59,7 +59,7 @@ def main():
try:
if is_nt:
cwd = os.getcwd()
#icon = cwd + "\Images\logo.ico"
# icon = cwd + "\Images\logo.ico"
icon = resource_path("Images/logo.ico")
toast = Notification(
app_id="EyeTrackApp",
@ -183,7 +183,10 @@ def main():
# Create the window
window = sg.Window(
f"{appversion}", layout, icon=resource_path("Images/logo.ico"), background_color="#292929"
f"{appversion}",
layout,
icon=resource_path("Images/logo.ico"),
background_color="#292929",
)
# GUI Render loop
@ -259,8 +262,10 @@ def main():
config.eye_display_id = EyeId.SETTINGS
config.save()
elif values[ALGO_SETTINGS_RADIO_NAME] and config.eye_display_id != EyeId.ALGOSETTINGS:
elif (
values[ALGO_SETTINGS_RADIO_NAME]
and config.eye_display_id != EyeId.ALGOSETTINGS
):
eyes[0].stop()
eyes[1].stop()
settings[0].stop()
@ -282,6 +287,5 @@ def main():
setting.render(window, event, values)
if __name__ == "__main__":
main()

View File

@ -85,8 +85,8 @@ class LEAP_C(object):
else:
self.model_path = resource_path("Models/mommy072623.onnx")
self.interval = 1 # FPS print update rate
self.low_priority = False # set process priority to low may cause issues
self.print_fps = True
self.low_priority = True # set process priority to low (may cause issues when unfocusing? reported by one, not reproducable)
self.print_fps = False
# Init variables
self.frames = 0
self.queues = []

View File

@ -427,7 +427,7 @@ def RANSAC3D(self, hsrac_en):
with open("RANSAC_BLINK_RIGHT.cfg", "w") as file:
for item in self.blink_list:
file.write(str(item) + "\n")
# print("SAVE")
# print("SAVE")
# self.blink_list.pop(0)
self.blink_list.append(abs(perscalarw - perscalarh))
@ -438,7 +438,6 @@ def RANSAC3D(self, hsrac_en):
if abs(perscalarw - perscalarh) >= np.percentile(self.blink_list, 92):
blink = 0.0
try:
cv2.drawContours(
self.current_image_gray, contours, -1, (255, 0, 0), 1
@ -490,7 +489,7 @@ def RANSAC3D(self, hsrac_en):
thresh = cv2.resize(thresh, (x, y))
try:
self.failed = 0 # we have succeded, continue with this
return cx, cy, thresh, blink
return cx, cy, thresh, blink, w, h
except:
self.failed = self.failed + 1 # we have failed, move onto next algo
return 0, 0, thresh, blink
return 0, 0, thresh, blink, 0, 0