206 lines
7.5 KiB
Python
206 lines
7.5 KiB
Python
import os
|
|
import numpy as np
|
|
import math
|
|
import cv2
|
|
import onnxruntime
|
|
import time
|
|
|
|
|
|
|
|
def clamp_to_im(pt, w, h): #8 times per frame, but that only accounts for 0.005ms
|
|
x=max(pt[0],0)
|
|
y=max(pt[1],0)
|
|
if x >= w:
|
|
x = w-1
|
|
if y >= h:
|
|
y = h-1
|
|
return (int(x),int(y+1))
|
|
|
|
def rotate(origin, point, a): #runs 22 times per frame
|
|
x, y = point - origin
|
|
cosa = math.cos(-a)
|
|
sina = math.sin(-a)
|
|
qx = origin[0] + cosa * x - sina * y
|
|
qy = origin[1] + sina * x + cosa * y
|
|
return qx, qy
|
|
|
|
def clamp (value, minimum, maxium):
|
|
return max(min(value,maxium),minimum)
|
|
|
|
def rotate_image(image, a, center): #twice per frame, 0.2ms - 0.25ms each, improved to 0.15ms - 0.25ms
|
|
M = cv2.getRotationMatrix2D((float(center[0]), float(center[1])), math.degrees(a), 1.0)
|
|
(h, w) = image.shape[:2]
|
|
image = cv2.warpAffine(image, M, (w, h), flags = cv2.INTER_CUBIC)
|
|
return image
|
|
|
|
|
|
|
|
|
|
class Eye():
|
|
def __init__(self,index):
|
|
|
|
self.eye_tracking_frames = 0
|
|
|
|
self.index = index
|
|
self.state = [1.,0.,0.,0.]
|
|
self.image = None
|
|
self.info = None
|
|
self.mean = np.float32(np.array([-2.1179, -2.0357, -1.8044]))
|
|
self.std = np.float32(np.array([0.0171, 0.0175, 0.0174]))
|
|
self.results = None
|
|
self.condifence = 0
|
|
self.standardDeviation = 0.
|
|
self.averageEyeConfidence = 1.
|
|
self.averageEyeConfidenceVariance = 0
|
|
self.lastEyeState = [0.,0.]
|
|
self.innerPoint = None
|
|
self.outerPoint = None
|
|
|
|
def prepare_eye(self, faceFrame):
|
|
self.state = [1.,0.,0.,0.]
|
|
im = faceFrame
|
|
(x1, y1), (x2, y2), a = self.corners_to_eye( im.shape[1], im.shape[0])
|
|
#rotating an image is expensive and reduces clarity
|
|
#so I just don't if it's a relatively small angle
|
|
if math.degrees(a) > 7.5 and math.degrees(a) < 352.5:
|
|
im = rotate_image(im, a, self.outerPoint)
|
|
|
|
im = im[int(y1):int(y2), int(x1):int(x2)]
|
|
if np.prod(im.shape) < 1:
|
|
return None, None, None, None, None, None
|
|
if self.index == 1:
|
|
im = cv2.flip(im, 1)
|
|
scale = [(x2 - x1)/ 32., (y2 - y1)/ 32.]
|
|
im = cv2.resize(im, (32, 32), interpolation=cv2.INTER_CUBIC)
|
|
im = im.astype(np.float32) * self.std + self.mean
|
|
im = np.expand_dims(im, 0)
|
|
self.image = np.transpose(im, (0,3,2,1))
|
|
self.info = [x1, y1, scale, self.outerPoint, a]
|
|
return
|
|
|
|
def corners_to_eye(self, w, h):
|
|
c1 = np.array(self.outerPoint)
|
|
c2 = np.array(self.innerPoint)
|
|
a = math.atan2(*(c2 - c1)[::-1]) % (2*math.pi)
|
|
c2 = rotate(c1, c2, a)
|
|
center = (c1 + c2) / 2.0
|
|
radius = np.linalg.norm(c1 - c2)
|
|
radius = [radius * 0.7, radius * 0.6]
|
|
upper_left = clamp_to_im(center - radius, w, h)
|
|
lower_right = clamp_to_im(center + radius, w, h)
|
|
return upper_left, lower_right, a
|
|
|
|
def calculateEye(self):
|
|
|
|
e_x, e_y, scale, reference, angles = self.info
|
|
confidenceThreshold = self.averageEyeConfidence - 2 * self.standardDeviation
|
|
|
|
m = self.results[0].argmax()
|
|
x = m // 8
|
|
y = m % 8
|
|
|
|
p=self.results[1][x, y]
|
|
p = clamp(p, 0.00001, 0.9999)
|
|
off_x = math.log(p/(1-p))
|
|
eye_x = 4.0 *(x + off_x)
|
|
|
|
p=self.results[2][x, y]
|
|
p = clamp(p, 0.00001, 0.9999)
|
|
off_y = math.log(p/(1-p))
|
|
eye_y = 4.0 * (y + off_y)
|
|
|
|
#I'm proud of this
|
|
#if eye movements are below 2 standard deviations of the average the movement is severely limited
|
|
#the eyes are handled independenly because my testing showed that one eye tended to have a higher condifence than the other
|
|
limit = max((self.results[0][x,y]- 0.5)*(2/confidenceThreshold),0)
|
|
if self.results[0][x,y] < confidenceThreshold:
|
|
Delta = eye_y - self.lastEyeState[0]
|
|
eye_y = self.lastEyeState[0] + clamp(Delta, -limit, limit)
|
|
Delta = eye_x - self.lastEyeState[1]
|
|
eye_x = self.lastEyeState[1] + clamp(Delta, -limit, limit)
|
|
|
|
self.lastEyeState = [eye_y, eye_x]
|
|
|
|
if self.index == 1:
|
|
eye_x = (32. - eye_x)
|
|
eye_x = e_x + scale[0] * eye_x
|
|
eye_y = e_y + scale[1] * eye_y
|
|
|
|
eye_x, eye_y = rotate(reference, (eye_x, eye_y), -angles)
|
|
eye_x, eye_y = (eye_x, eye_y) + self.offset
|
|
|
|
self.condifence = self.results[0][x,y]
|
|
self.calculateStandardDeviation()
|
|
self.state = [1.0, eye_y, eye_x, self.condifence]
|
|
return
|
|
|
|
def calculateStandardDeviation(self):
|
|
self.eye_tracking_frames+=1
|
|
|
|
avgRatio = 1/self.eye_tracking_frames
|
|
self.averageEyeConfidence = (self.averageEyeConfidence*(1-avgRatio))+ (self.condifence * avgRatio)
|
|
|
|
avgRatio = 1/(self.eye_tracking_frames + 1)
|
|
self.averageEyeConfidenceVariance = pow(((self.condifence - self.averageEyeConfidence) * avgRatio) + (self.averageEyeConfidenceVariance * (1-avgRatio)), 2)
|
|
|
|
self.standardDeviation = math.sqrt(self.averageEyeConfidenceVariance)
|
|
|
|
|
|
class EyeTracker():
|
|
def __init__(self):
|
|
self.leftEye = Eye(1)
|
|
self.rightEye = Eye(0)
|
|
self.mean = np.float32(np.array([-2.1179, -2.0357, -1.8044]))
|
|
self.std = np.float32(np.array([0.0171, 0.0175, 0.0174]))
|
|
self.faceFrame = None
|
|
self.offset = None
|
|
self.faceCenter = None
|
|
self.faceRadius = None
|
|
|
|
model_base_path = os.path.join(os.path.dirname(__file__), os.path.join("models"))
|
|
providersList = onnxruntime.capi._pybind_state.get_available_providers()
|
|
options = onnxruntime.SessionOptions()
|
|
options.inter_op_num_threads = 1
|
|
options.intra_op_num_threads = 1
|
|
options.execution_mode = onnxruntime.ExecutionMode.ORT_SEQUENTIAL
|
|
options.graph_optimization_level = onnxruntime.GraphOptimizationLevel.ORT_ENABLE_ALL
|
|
options.log_severity_level = 3
|
|
self.gaze_model = onnxruntime.InferenceSession(os.path.join(model_base_path, "mnv3_gaze32_split_opt.onnx"), sess_options=options, providers=providersList)
|
|
|
|
def get_eye_state(self,frame, lms):
|
|
|
|
lms = self.extract_face(frame, np.array(lms)[:,0:2][:,::-1])
|
|
|
|
self.rightEye.offset = self.offset
|
|
self.leftEye.offset = self.offset
|
|
self.rightEye.innerPoint = lms[39,0:2]
|
|
self.rightEye.outerPoint = lms[36,0:2]
|
|
self.leftEye.innerPoint = lms[45,0:2]
|
|
self.leftEye.outerPoint = lms[42,0:2]
|
|
|
|
self.rightEye.prepare_eye(self.faceFrame)
|
|
self.leftEye.prepare_eye(self.faceFrame)
|
|
|
|
if self.rightEye.image is None or self.leftEye.image is None:
|
|
return [[1.,0.,0.,0.],[1.,0.,0.,0.]] #Early exit if one of the eyes doesn't have data
|
|
both_eyes = np.concatenate((self.rightEye.image, self.leftEye.image))
|
|
|
|
self.rightEye.results, self.leftEye.results = self.gaze_model.run([], {"input": both_eyes})[0]
|
|
|
|
self.rightEye.calculateEye()
|
|
self.leftEye.calculateEye()
|
|
|
|
return [self.rightEye.state, self.leftEye.state]
|
|
|
|
def extract_face(self, frame, lms):
|
|
x1, y1 = lms.min(0)
|
|
x2, y2 = lms.max(0)
|
|
self.faceRadius = np.array([(x2 - x1), (y2 - y1)])*0.6
|
|
self.faceCenter = (np.array((x1, y1)) + np.array((x2, y2))) / 2.0
|
|
w, h, _ = frame.shape
|
|
x1, y1 = clamp_to_im(self.faceCenter - self.faceRadius , h, w)
|
|
x2, y2 = clamp_to_im(self.faceCenter + self.faceRadius + 1, h, w)
|
|
self.offset = np.array((x1, y1))
|
|
lms = (lms[:, 0:2] - self.offset).astype(int)
|
|
self.faceFrame = frame[y1:y2, x1:x2]
|
|
return lms
|