sustaining_gazes_tnc/head_pose.py
2018-04-29 15:09:15 +02:00

445 lines
19 KiB
Python

#!/usr/bin/env python
import cv2
import dlib
import numpy as np
import os
import pickle
import logging
from scipy.ndimage.filters import gaussian_filter
from PIL import Image, ImageDraw,ImageTk
import pandas as pd
import seaborn as sns
from matplotlib.backends.backend_tkagg import FigureCanvasTkAgg
from matplotlib.figure import Figure
import sys
if sys.version_info[0] < 3:
import Tkinter as Tk
else:
import tkinter as Tk
import time
logging.basicConfig( format='%(asctime)-15s %(name)s %(levelname)s: %(message)s' )
logger = logging.getLogger(__name__)
# Read Image
c = cv2.VideoCapture(0)
# im = cv2.imread("headPose.jpg");
predictor_path = "shape_predictor_68_face_landmarks.dat"
detector = dlib.get_frontal_face_detector()
predictor = dlib.shape_predictor(predictor_path)
screenDrawCorners = np.array([[10,60], [90, 60], [10, 110], [90, 110]])
# metrics matrix
metricsSize = [1920,1080]
dataframe = pd.DataFrame(columns=['x','y'])
metrics = np.zeros(metricsSize)
screenDrawCorners = np.array([[0,0], [1919,0], [0, 1079], [1919,1079]])
def create_perspective_transform_matrix(src, dst):
""" Creates a perspective transformation matrix which transforms points
in quadrilateral ``src`` to the corresponding points on quadrilateral
``dst``.
Will raise a ``np.linalg.LinAlgError`` on invalid input.
"""
# See:
# * http://xenia.media.mit.edu/~cwren/interpolator/
# * http://stackoverflow.com/a/14178717/71522
in_matrix = []
for (x, y), (X, Y) in zip(src, dst):
in_matrix.extend([
[x, y, 1, 0, 0, 0, -X * x, -X * y],
[0, 0, 0, x, y, 1, -Y * x, -Y * y],
])
A = np.matrix(in_matrix, dtype=np.float)
B = np.array(dst).reshape(8)
af = np.dot(np.linalg.inv(A.T * A) * A.T, B)
m = np.append(np.array(af).reshape(8), 1).reshape((3, 3))
logger.info("Created transformmatrix: src %s dst %s m %s", src, dst, m)
return m
# got this amazing thing from here: https://stackoverflow.com/a/24088499
def create_perspective_transform(src, dst, round=False, splat_args=False):
""" Returns a function which will transform points in quadrilateral
``src`` to the corresponding points on quadrilateral ``dst``::
>>> transform = create_perspective_transform(
... [(0, 0), (10, 0), (10, 10), (0, 10)],
... [(50, 50), (100, 50), (100, 100), (50, 100)],
... )
>>> transform((5, 5))
(74.99999999999639, 74.999999999999957)
If ``round`` is ``True`` then points will be rounded to the nearest
integer and integer values will be returned.
>>> transform = create_perspective_transform(
... [(0, 0), (10, 0), (10, 10), (0, 10)],
... [(50, 50), (100, 50), (100, 100), (50, 100)],
... round=True,
... )
>>> transform((5, 5))
(75, 75)
If ``splat_args`` is ``True`` the function will accept two arguments
instead of a tuple.
>>> transform = create_perspective_transform(
... [(0, 0), (10, 0), (10, 10), (0, 10)],
... [(50, 50), (100, 50), (100, 100), (50, 100)],
... splat_args=True,
... )
>>> transform(5, 5)
(74.99999999999639, 74.999999999999957)
If the input values yield an invalid transformation matrix an identity
function will be returned and the ``error`` attribute will be set to a
description of the error::
>>> tranform = create_perspective_transform(
... np.zeros((4, 2)),
... np.zeros((4, 2)),
... )
>>> transform((5, 5))
(5.0, 5.0)
>>> transform.error
'invalid input quads (...): Singular matrix
"""
try:
transform_matrix = create_perspective_transform_matrix(src, dst)
error = None
except np.linalg.LinAlgError as e:
transform_matrix = np.identity(3, dtype=np.float)
error = "invalid input quads (%s and %s): %s" %(src, dst, e)
error = error.replace("\n", "")
to_eval = "def perspective_transform(%s):\n" %(
splat_args and "*pt" or "pt",
)
to_eval += " res = np.dot(transform_matrix, ((pt[0], ), (pt[1], ), (1, )))\n"
to_eval += " res = res / res[2]\n"
if round:
to_eval += " return (int(round(res[0][0])), int(round(res[1][0])))\n"
else:
to_eval += " return (res[0][0], res[1][0])\n"
locals = {
"transform_matrix": transform_matrix,
}
locals.update(globals())
exec to_eval in locals, locals
res = locals["perspective_transform"]
res.matrix = transform_matrix
res.error = error
return res
def coordinatesToSrc(coordinates):
return np.array([coordinates['tl'], coordinates['tr'],coordinates['bl'], coordinates['br']])
# coordinates of the screen boundaries
if os.path.exists("coordinates.p"):
coordinates = pickle.load(open("coordinates.p", "rb"))
transform = create_perspective_transform(coordinatesToSrc(coordinates), screenDrawCorners)
a = [np.array([ 1312.15541183]), np.array([ 244.56278002]), 0]
logger.info("Loaded coordinates: %s", coordinatesToSrc(coordinates))
logger.debug("Corners: %s", screenDrawCorners)
logger.debug("Test point %s", a)
logger.debug("Transformed point %s", transform(a[0:2]))
# exit()
else:
coordinates = {'tl': None, 'tr': None, 'bl': None, 'br': None}
transform = None
windowRoot = Tk.Toplevel()
windowSize = (1000,1000)
windowRoot.geometry('%dx%d+%d+%d' % (windowSize[0],windowSize[1],0,0))
figure = Figure(figsize=(16, 9), dpi=100)
axes = figure.add_subplot(111)
axes.set_title('Tk embedding')
axes.set_xlabel('X axis label')
axes.set_ylabel('Y label')
# canvas = Tk.Canvas(windowRoot,width=1000,height=1000)
canvas = FigureCanvasTkAgg(figure,master=windowRoot)
canvas.show()
canvas.get_tk_widget().pack(side=Tk.TOP, fill=Tk.BOTH, expand=1)
imageWindowRoot = Tk.Toplevel()
imageWindowSize = (1000,1000)
imageWindowRoot.geometry('%dx%d+%d+%d' % (imageWindowSize[0],imageWindowSize[1],0,0))
imageCanvas = Tk.Canvas(imageWindowRoot,width=1000,height=1000)
while True:
t1 = time.time()
_, im = c.read()
size = im.shape
t2 = time.time()
logger.debug("Captured frame in %fs", t2-t1)
# Docs: Ask the detector to find the bounding boxes of each face. The 1 in the
# second argument indicates that we should upsample the image 1 time. This
# will make everything bigger and allow us to detect more faces.
dets = detector(im, 1)
t3 = time.time()
logger.debug("Number of faces detected: {} - took {}s".format(len(dets), t3-t2))
# We use this later for calibrating
currentPoint = None
currentPoints = []
if len(dets) > 0:
for d in dets:
td1 = time.time()
shape = predictor(im, d)
td2 = time.time()
logger.debug("Found face points in %fs", td2-td1)
#2D image points. If you change the image, you need to change vector
image_points = np.array([
(shape.part(30).x,shape.part(30).y), # Nose tip
(shape.part(8).x,shape.part(8).y), # Chin
(shape.part(36).x,shape.part(36).y), # Left eye left corner
(shape.part(45).x,shape.part(45).y), # Right eye right corne
(shape.part(48).x,shape.part(48).y), # Left Mouth corner
(shape.part(54).x,shape.part(54).y) # Right mouth corner
], dtype="double")
# 3D model points.
model_points = np.array([
(0.0, 0.0, 0.0), # Nose tip
(0.0, -330.0, -65.0), # Chin
(-225.0, 170.0, -135.0), # Left eye left corner
(225.0, 170.0, -135.0), # Right eye right corne
(-150.0, -150.0, -125.0), # Left Mouth corner
(150.0, -150.0, -125.0) # Right mouth corner
])
# Camera internals
focal_length = size[1]
center = (size[1]/2, size[0]/2)
camera_matrix = np.array(
[[focal_length, 0, center[0]],
[0, focal_length, center[1]],
[0, 0, 1]], dtype = "double"
)
# print ("Camera Matrix :\n {0}".format(camera_matrix))
dist_coeffs = np.zeros((4,1)) # Assuming no lens distortion
(success, rotation_vector, translation_vector) = cv2.solvePnP(model_points, image_points, camera_matrix, dist_coeffs, flags=cv2.SOLVEPNP_ITERATIVE)
if not success:
print("Error determening PnP", success)
continue
logger.debug ("Rotation Vector:\n %s", rotation_vector)
print ("Translation Vector:\n {0}".format(translation_vector))
# Project a 3D point (0, 0, 1000.0) onto the image plane.
# We use this to draw a line sticking out of the nose
(nose_end_point2D, jacobian) = cv2.projectPoints(np.array([(0.0, 0.0, 1000.0)]), rotation_vector, translation_vector, camera_matrix, dist_coeffs)
for p in image_points:
cv2.circle(im, (int(p[0]), int(p[1])), 3, (0,0,255), -1)
p1 = ( int(image_points[0][0]), int(image_points[0][1]))
p2 = ( int(nose_end_point2D[0][0][0]), int(nose_end_point2D[0][0][1]))
cv2.line(im, p1, p2, (255,0,0), 2)
rotMatrix = np.zeros([3,3])
cv2.Rodrigues(rotation_vector, rotMatrix, jacobian=0)
# Find rotation: https://stackoverflow.com/a/15029416
rx = np.arctan2(rotMatrix[2,1], rotMatrix[2,2])
ry = np.arctan2(-rotMatrix[2,0], np.sqrt(np.square(rotMatrix[2,1]) + np.square(rotMatrix[2,2])))
rz = np.arctan2(rotMatrix[1,0],rotMatrix[0,0])
print("rotation", rx, ry, rz)
ry = - np.arcsin(rotMatrix[0,2])
rx = np.arctan2(rotMatrix[1,2]/np.cos(ry), rotMatrix[2,2]/np.cos(ry))
rz = np.arctan2(rotMatrix[0,1]/np.cos(ry), rotMatrix[0,0]/np.cos(ry))
print("rotation ml", rx, ry, rz) # seems better?
# draw little floorplan for x: 10 -> 50 maps to z: 0 -> 10000, x: -2000 -> 2000
mapPosX = int((translation_vector[0] + 500) / 1000 * 40)
mapPosY = int((translation_vector[1] + 500) / 1000 * 40)
mapPosZ = int((translation_vector[2] + 0 ) / 10000 * 40)
cv2.circle(im, (mapPosZ + 10, mapPosX + 10), 2, (0,0,255), -1)
cv2.circle(im, (mapPosZ + 60, mapPosY + 10), 2, (0,0,255), -1)
# make it an _amazing_ stick figurine for the side view
cv2.line(im, (mapPosZ + 60, mapPosY + 10), (mapPosZ + 60, mapPosY + 20), (0,0,255), 1)
cv2.line(im, (mapPosZ + 60, mapPosY + 20), (mapPosZ + 55, mapPosY + 25), (0,0,255), 1)
cv2.line(im, (mapPosZ + 60, mapPosY + 20), (mapPosZ + 65, mapPosY + 25), (0,0,255), 1)
cv2.line(im, (mapPosZ + 60, mapPosY + 15), (mapPosZ + 55, mapPosY + 10), (0,0,255), 1)
cv2.line(im, (mapPosZ + 60, mapPosY + 15), (mapPosZ + 65, mapPosY + 10), (0,0,255), 1)
# draw rotation vector
cv2.circle(im, (mapPosZ + 60, mapPosY + 10), 2, (0,0,255), -1)
viewDirectionVector = np.dot(np.array([0.0, 0.0, 1000.0]), rotMatrix)
cv2.line(im, (mapPosZ + 10, mapPosX + 10), (mapPosZ + 10 + int(viewDirectionVector[2] * 100), mapPosX + 10 + int(viewDirectionVector[0] * 100)), (255,255,0), 1)
cv2.line(im, (mapPosZ + 60, mapPosY + 10), (mapPosZ + 60 + int(viewDirectionVector[2] * 100), mapPosY + 10 - int(viewDirectionVector[1] * 100)), (255,0,255), 1)
# Translation vector gives position in space:
# x, y z: 0,0,0 is center of camera
# line: (x,y,z) = f(a) = (t1 + r1*a, t2+r2*a, t3 + r3*a)
# Screen: (x,y,z) = (x,y,0)
# Interesection:
# x = t1 + r1 * a
# y = t2 + r2 * a
# z = t3 * r3 * a = 0
# => a = -t3 / r3
# substitute found a in x,y
a = - translation_vector[2] / rotation_vector[2]
x = translation_vector[0] + rotation_vector[0] * a
y = translation_vector[1] + rotation_vector[1] * a
a = - translation_vector[2] / viewDirectionVector[2]
x = translation_vector[0] + viewDirectionVector[0] * a
y = translation_vector[1] + viewDirectionVector[1] * a
point = np.array([x,y])
currentPoint = point
currentPoints.append(point)
td3 = time.time()
logger.debug("Timer: All other face drawing stuff in %fs", td3-td2)
# TODO only draw nose line now, so we can change color depending whether on screen or not
# processed all faces, now draw on screen:
te1 = time.time()
# draw little floorplan for 10 -> 50, sideplan 60 -> 100 (40x40 px)
cv2.rectangle(im, (9, 9), (51, 51), (255,255,255), 1)
cv2.rectangle(im, (59, 9), (101, 51), (255,255,255), 1)
cv2.line(im, (10,10), (10,50), (200,200,200), 2)
cv2.line(im, (60,10), (60,50), (200,200,200), 2)
# screen is 16:10
cv2.rectangle(im, (9, 59), (91, 111), (255,255,255), 1)
if transform is None:
cv2.putText(im, "1", (10,70), cv2.FONT_HERSHEY_PLAIN, .7, (255,255,255) if coordinates['tl'] is not None else (0,0,255))
cv2.putText(im, "2", (85,70), cv2.FONT_HERSHEY_PLAIN, .7, (255,255,255) if coordinates['tr'] is not None else (0,0,255))
cv2.putText(im, "3", (10,110), cv2.FONT_HERSHEY_PLAIN, .7, (255,255,255) if coordinates['bl'] is not None else (0,0,255))
cv2.putText(im, "4", (85,110), cv2.FONT_HERSHEY_PLAIN, .7, (255,255,255) if coordinates['br'] is not None else (0,0,255))
tm1 = 0
tm2 = 0
tm3 = 0
tm4 = 0
else:
tm1 = time.time()
newMetrics = np.zeros(metricsSize)
tm2 = time.time()
for point in currentPoints:
# check if within coordinates:
# dot1 = np.dot(coordinates['tl'] - point, coordinates['tl'] - coordinates['br'])
# dot2 = np.dot(coordinates['bl'] - point, coordinates['tl'] - coordinates['br'])
# pointIn3 = [point[0], point[1], 0]
# targetPoint = np.dot(pointIn3, transformationMatrix)
# print("Looking at", pointIn3, np.dot( transformationMatrix, pointIn3))
targetPoint = transform(point)
print("Looking at", point, targetPoint)
# cv2.circle(im, (int(targetPoint[0]), int(targetPoint[1])), 2, (0,255,0), -1)
# from 1920x1080 to 80x50
miniTargetPoint = (int(targetPoint[0] / 1920 * 80 + 10), int(targetPoint[1] / 1080 * 50 + 60))
cv2.circle(im, miniTargetPoint, 2, (0,255,0), -1)
targetInt = (int(targetPoint[0]), int(targetPoint[1]))
# check if point fits on screen:
# if so, measure it
if targetInt[0] >= 0 and targetInt[1] >= 0 and targetInt[0] < metricsSize[0] and targetInt[1] < metricsSize[1]:
dataframe = dataframe.append({'x':targetInt[0],'y':targetInt[1]}, ignore_index=True)
newMetrics[targetInt[0],targetInt[1]] += 1
# after we collected all new metrics, blur them foor smoothness
# and add to all metrics collected
tm3 = time.time()
metrics = metrics + gaussian_filter(newMetrics, sigma = 8)
tm4 = time.time()
logger.debug("Updated matrix with blur in %f", tm4 - tm3 + tm2 - tm1)
# Display webcam image with overlays
te2 = time.time()
logger.debug("Drew on screen in %fs", te2-te1)
cv2.imshow("Output", im)
te3 = time.time()
logger.debug("showed webcam image in %fs", te3-te2)
# blur smooth the heatmap
# logger.debug("Max blurred metrics: %f", np.max(metrics))
# update the heatmap output
tm21 = time.time()
normalisedMetrics = metrics / (np.max(metrics)/255)
tm22 = time.time()
logger.debug("Max normalised metrics: %f", np.max(normalisedMetrics))
# print(normalisedMetrics)
tm23 = time.time()
image = Image.fromarray(normalisedMetrics)
wpercent = (imageWindowSize[0] / float(image.size[0]))
hsize = int((float(image.size[1]) * float(wpercent)))
image = image.resize((imageWindowSize[0], hsize))
tkpi = ImageTk.PhotoImage(image)
imageCanvas.delete("IMG")
imagesprite = imageCanvas.create_image(500,500,image=tkpi, tags="IMG")
imageWindowRoot.update()
tm24 = time.time()
logger.debug("PIL iamge generated in %fs", tm24 - tm23)
logger.debug("Total matrix time is %fs", tm4 - tm3 + tm2 - tm1 + tm24 - tm21)
te4 = time.time()
axes.clear()
if(len(dataframe) > 2):
g = sns.kdeplot(dataframe['x'], dataframe['y'],ax=axes, n_levels=30, shade=True, cmap="rainbow")
canvas.draw()
windowRoot.update()
te5 = time.time()
logger.debug("Drew graph & updated window in %fs", te5-te4)
# (optionally) very slowly fade out previous metrics:
# metrics = metrics * .999
keyPress = cv2.waitKey(5)
if keyPress==27:
break
elif keyPress == ord('d'):
logger.setLevel(logging.DEBUG)
elif keyPress > -1 and currentPoint is not None:
recalculate = False
if keyPress == ord('1'):
coordinates['tl'] = currentPoint
recalculate = True
elif keyPress == ord('2'):
coordinates['tr'] = currentPoint
recalculate = True
elif keyPress == ord('3'):
coordinates['bl'] = currentPoint
recalculate = True
elif keyPress == ord('4'):
coordinates['br'] = currentPoint
recalculate = True
elif keyPress == ord('t') and transform is not None:
print("Coordinates", coordinates)
print("Drawing area", screenDrawCorners)
print("Test point %s", currentPoint )
print("Transformed point %s", transform(currentPoint))
if recalculate is True and not any (x is None for x in coordinates.values()):
logger.debug(coordinates.values())
pickle.dump( coordinates, open( "coordinates.p", "wb" ) )
logger.info("Saved coordinates")
transform = create_perspective_transform(coordinatesToSrc(coordinates), screenDrawCorners)
cv2.destroyAllWindows()