eye_tracking_experiment/facemesh_estimation.py at main · cortictechnology/eye_tracking_experiment · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
import mediapipe as mp
import numpy as np
import cv2
from data import triangulation
from PIL import Image

COLOR = [(0, 153, 0), (234, 187, 105), (80, 190, 168), (0, 0, 255)]

# index of crop top left: 70
# index of crop bottom right: 340

class FaceMeshEstimation:
    def __init__(self, min_detection_confidence=0.5, min_tracking_confidence=0.5, static_mode=False, smooth_factor=0.1):
        self.min_detection_confidence = min_detection_confidence
        self.min_tracking_confidence = min_tracking_confidence
        self.mp_face_mesh = mp.solutions.face_mesh
        self.face_mesh_estimator = self.mp_face_mesh.FaceMesh(static_image_mode=static_mode,
                                                              refine_landmarks=True,
                                                              min_detection_confidence=self.min_detection_confidence,
                                                              min_tracking_confidence=self.min_tracking_confidence)
        self.landmarks = None
        self.face = None
        self.smooth_factor = smooth_factor

    def get_facemesh(self, frame, convert_rgb=False):
        frame_rgb = frame
        if convert_rgb:
            cv2_im_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
            frame_rgb = np.asarray(Image.fromarray(cv2_im_rgb), dtype=np.uint8)
        results = self.face_mesh_estimator.process(frame_rgb)
        multi_face_landmarks = results.multi_face_landmarks

        if multi_face_landmarks:
            face_landmarks = results.multi_face_landmarks[0]
            face_landmarks = np.array(
                [(lm.x, lm.y, lm.z) for lm in face_landmarks.landmark]
            )
            if self.landmarks is None:
                self.landmarks = face_landmarks.T
            else:
                self.landmarks = face_landmarks.T * (1 - self.smooth_factor) + self.smooth_factor * self.landmarks
            self.face = [[np.min(self.landmarks[0, :]), np.min(self.landmarks[1, :]), np.max(self.landmarks[0, :]), np.max(self.landmarks[1, :])]]
        return self.landmarks, self.face

    def draw_disconnected_rect(self, frame, pt1, pt2, color, thickness):
        img = frame
        width = pt2[0] - pt1[0]
        height = pt2[1] - pt1[1]
        line_width = min(20, width // 4)
        line_height = min(20, height // 4)
        line_length = max(line_width, line_height)
        cv2.line(img, pt1, (pt1[0] + line_length, pt1[1]), color, thickness)
        cv2.line(img, pt1, (pt1[0], pt1[1] + line_length), color, thickness)
        cv2.line(
            img, (pt2[0] - line_length, pt1[1]
                  ), (pt2[0], pt1[1]), color, thickness
        )
        cv2.line(
            img, (pt2[0], pt1[1]), (pt2[0], pt1[1] +
                                    line_length), color, thickness
        )
        cv2.line(
            img, (pt1[0], pt2[1]), (pt1[0] +
                                    line_length, pt2[1]), color, thickness
        )
        cv2.line(
            img, (pt1[0], pt2[1] - line_length), (pt1[0],
                                                  pt2[1]), color, thickness
        )
        cv2.line(img, pt2, (pt2[0] - line_length, pt2[1]), color, thickness)
        cv2.line(img, (pt2[0], pt2[1] - line_length), pt2, color, thickness)
        return img

    def norm2abs(self, x_y, frame_size, pad_w, pad_h):
        x = int(x_y[0] * frame_size - pad_w)
        y = int(x_y[1] * frame_size - pad_h)
        return (x, y)

    def draw_detected_face(self, frame, detected_faces):
        if frame is not None and detected_faces is not None:
            frame_width = frame.shape[1]
            frame_height = frame.shape[0]
            for face_coordinates in detected_faces:
                x1 = int(face_coordinates[0] * frame_width)
                y1 = int(face_coordinates[1] * frame_height)
                x2 = int(face_coordinates[2] * frame_width)
                y2 = int(face_coordinates[3] * frame_height)
                frame = self.draw_disconnected_rect(frame, (x1, y1), (x2, y2), (234, 105, 105), 2)
        return frame

    def draw_facemesh(self, frame, landmarks):
        frame_width = frame.shape[1]
        frame_height = frame.shape[0]
        for i in range(int(triangulation.shape[0] / 3)):
            pt1 = np.array(landmarks[0:2, triangulation[i * 3]], np.float32)
            pt2 = np.array(landmarks[0:2, triangulation[i * 3 + 1]], np.float32)
            pt3 = np.array(landmarks[0:2, triangulation[i * 3 + 2]], np.float32)
            pt1[0] = int(pt1[0] * frame_width)
            pt1[1] = int(pt1[1] * frame_height)
            pt2[0] = int(pt2[0] * frame_width)
            pt2[1] = int(pt2[1] * frame_height)
            pt3[0] = int(pt3[0] * frame_width)
            pt3[1] = int(pt3[1] * frame_height)
            pt1 = pt1.astype(np.int32)
            pt2 = pt2.astype(np.int32)
            pt3 = pt3.astype(np.int32)
            cv2.line(frame, (pt1[0], pt1[1]), (pt2[0], pt2[1]), COLOR[1], thickness=1)
            cv2.line(frame, (pt2[0], pt2[1]), (pt3[0], pt3[1]), COLOR[1], thickness=1)
            cv2.line(frame, (pt3[0], pt3[1]), (pt1[0], pt1[1]), COLOR[1], thickness=1)
        return frame

    def crop_eye_image(self, frame, xmin, xmax, ymin, ymax):
        if xmin < 0:
            xmin = 0
        if ymin < 0:
            ymin = 0
        if xmax >= frame.shape[1]:
            xmax = frame.shape[1] - 1
        if ymax >= frame.shape[0]:
            ymax = frame.shape[0] - 1
        crop_frame_width = xmax - xmin
        crop_frame_height = ymax - ymin
        cropped_eye_image = frame[ymin:ymax, xmin:xmax]

        return crop_frame_width, crop_frame_height, cropped_eye_image