File size: 15,473 Bytes
19f420a
 
 
 
3cfc6a1
19f420a
f6146b6
19f420a
8054642
 
19f420a
f6146b6
 
 
 
19f420a
 
8054642
19f420a
f6146b6
 
 
 
19f420a
 
 
8054642
3cfc6a1
 
 
8054642
 
 
3cfc6a1
19f420a
 
f6146b6
 
8054642
f6146b6
 
 
8054642
3cfc6a1
 
8054642
f6146b6
8054642
f6146b6
 
8054642
f6146b6
8054642
f37553c
19f420a
8054642
3cfc6a1
 
19f420a
f6146b6
 
8054642
 
3cfc6a1
 
 
87a985c
8054642
 
 
 
 
 
 
 
f6146b6
 
8054642
 
3cfc6a1
f6146b6
8054642
 
f6146b6
f37553c
8054642
f6146b6
8054642
f37553c
8054642
 
 
f6146b6
 
8054642
f37553c
8054642
19f420a
8054642
 
 
f37553c
 
19f420a
8054642
 
 
 
 
f37553c
 
 
8054642
f37553c
 
 
 
19f420a
8054642
 
 
 
 
 
 
 
 
 
 
 
 
 
3cfc6a1
 
f37553c
 
8054642
 
 
 
 
 
 
 
 
f37553c
8054642
3cfc6a1
8054642
 
 
f37553c
3cfc6a1
8054642
f6146b6
8054642
 
 
 
f37553c
8054642
 
f6146b6
8054642
 
f6146b6
 
 
 
 
 
 
 
 
 
f37553c
f6146b6
 
 
 
 
8054642
 
 
f6146b6
8054642
f6146b6
 
19f420a
8054642
f37553c
 
19f420a
f6146b6
8054642
ea86682
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
import cv2
import mediapipe as mp
import numpy as np
import math
from src.detection.base_processor import BaseProcessor

# --- Helper Functions (Unchanged) ---
def calculate_ear(eye_landmarks, frame_shape):
    """Calculates the Eye Aspect Ratio for a single eye."""
    # Note: frame_shape is (height, width)
    coords = np.array([(lm.x * frame_shape[1], lm.y * frame_shape[0]) for lm in eye_landmarks])
    v1 = np.linalg.norm(coords[1] - coords[5])
    v2 = np.linalg.norm(coords[2] - coords[4])
    h1 = np.linalg.norm(coords[0] - coords[3])
    return (v1 + v2) / (2.0 * h1) if h1 > 0 else 0.0

def calculate_mar(mouth_landmarks, frame_shape):
    """Calculates the Mouth Aspect Ratio."""
    coords = np.array([(lm.x * frame_shape[1], lm.y * frame_shape[0]) for lm in mouth_landmarks])
    v1 = np.linalg.norm(coords[1] - coords[7])
    v2 = np.linalg.norm(coords[2] - coords[6])
    v3 = np.linalg.norm(coords[3] - coords[5])
    h1 = np.linalg.norm(coords[0] - coords[4])
    return (v1 + v2 + v3) / (2.0 * h1) if h1 > 0 else 0.0

class GeometricProcessor(BaseProcessor):
    # Landmark indices for eyes and mouth
    L_EYE = [362, 385, 387, 263, 373, 380]
    R_EYE = [33, 160, 158, 133, 153, 144]
    MOUTH = [61, 291, 39, 181, 0, 17, 84, 178]
    
    # Landmark indices for Head Pose Estimation
    HEAD_POSE_LANDMARKS = [1, 152, 263, 33, 287, 57] # Nose tip, Chin, Left eye left corner, Right eye right corner, Left mouth corner, Right mouth corner

    def __init__(self, config):
        self.settings = config['geometric_settings']
        self.face_mesh = mp.solutions.face_mesh.FaceMesh(
            max_num_faces=1,
            refine_landmarks=False, # Set to True for more detailed landmarks around eyes/lips, at a slight performance cost
            min_detection_confidence=0.5,
            min_tracking_confidence=0.5)

        self.downscale_factor = self.settings.get('downscale_factor', 0.35)
        self.default_skip = max(1, self.settings.get("skip_frames", 2))

        # --- FIX: Caching states for efficiency ---
        self.frame_counter = 0
        # Initialize with safe defaults
        self.last_indicators = {"drowsiness_level": "Initializing...", "lighting": "Good", "details": {}}
        self.last_landmarks = None
        self.last_drawn_frame = None # Cache the fully drawn frame

        # Drowsiness event counters
        self.counters = { "eye_closure": 0, "yawning": 0, "head_nod": 0, "looking_away": 0 }

        # Pre-allocated buffer for solvePnP
        self.zeros_4x1 = np.zeros((4, 1), np.float32)

    def process_frame(self, frame):
        self.frame_counter += 1
        
        # --- FIX: More efficient frame skipping ---
        # Adaptive skipping: process more frequently if drowsiness is detected.
        last_level = self.last_indicators.get("drowsiness_level", "Awake")
        skip_n = 1 if last_level != "Awake" else self.default_skip

        if self.frame_counter % skip_n != 0:
            # If we have a cached frame, return it to avoid re-drawing.
            if self.last_drawn_frame is not None:
                return self.last_drawn_frame, self.last_indicators
            # Fallback if the first frame was skipped (unlikely but safe)
            else:
                return frame.copy(), self.last_indicators

        # --- CORE FRAME PROCESSING ---
        original_frame = frame.copy()
        h_orig, w_orig, _ = original_frame.shape
        
        # Optimization: Downscale frame for faster processing
        small_frame = cv2.resize(original_frame, (0, 0), fx=self.downscale_factor, fy=self.downscale_factor, interpolation=cv2.INTER_AREA)
        h, w, _ = small_frame.shape

        # All processing is done on the `small_frame` for speed.
        gray = cv2.cvtColor(small_frame, cv2.COLOR_BGR2GRAY)
        brightness = np.mean(gray)
        
        drowsiness_indicators = {"drowsiness_level": "Awake", "lighting": "Good", "details": {}}
        face_landmarks_data = None

        if brightness < self.settings['low_light_thresh']:
            drowsiness_indicators["lighting"] = "Low"
        else:
            # Convert the SMALL frame to RGB for MediaPipe
            img_rgb = cv2.cvtColor(small_frame, cv2.COLOR_BGR2RGB)
            img_rgb.flags.writeable = False # Performance enhancement
            results = self.face_mesh.process(img_rgb)
            img_rgb.flags.writeable = True

            if results.multi_face_landmarks:
                face_landmarks_data = results.multi_face_landmarks[0]
                landmarks = face_landmarks_data.landmark
                score = 0
                weights = self.settings['indicator_weights']

                # --- Drowsiness Calculations (on small frame dimensions 'h', 'w') ---
                ear_left = calculate_ear([landmarks[i] for i in self.L_EYE],(h,w))
                ear_right = calculate_ear([landmarks[i] for i in self.R_EYE],(h,w))
                ear = (ear_left + ear_right) / 2.0
                
                if ear < self.settings['eye_ar_thresh']: self.counters['eye_closure']+=1
                else: self.counters['eye_closure']=0
                if self.counters['eye_closure'] >= self.settings['eye_ar_consec_frames']: score += weights['eye_closure']

                mar = calculate_mar([landmarks[i] for i in self.MOUTH], (h, w))
                if mar > self.settings['yawn_mar_thresh']: self.counters['yawning']+=1
                else: self.counters['yawning']=0
                if self.counters['yawning'] >= self.settings['yawn_consec_frames']: score += weights['yawning']

                # --- Head Pose Estimation (on small frame dimensions 'h', 'w') ---
                face_3d_model = 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 corner
                    [-150.0, -150.0, -125.0],    # Left Mouth corner
                    [150.0, -150.0, -125.0]      # Right mouth corner
                ], dtype=np.float32)
                
                face_2d_points = np.array([(landmarks[i].x * w, landmarks[i].y * h) for i in self.HEAD_POSE_LANDMARKS], dtype=np.float32)
                cam_matrix = np.array([[w, 0, w/2], [0, w, h/2], [0, 0, 1]], dtype=np.float32)
                
                _, rvec, _ = cv2.solvePnP(face_3d_model, face_2d_points, cam_matrix, self.zeros_4x1, flags=cv2.SOLVEPNP_EPNP)
                rmat, _ = cv2.Rodrigues(rvec)
                angles, _, _, _, _, _ = cv2.RQDecomp3x3(rmat)
                pitch, yaw = angles[0], angles[1]

                if pitch > self.settings['head_nod_thresh']: self.counters['head_nod']+=1
                else: self.counters['head_nod']=0
                if self.counters['head_nod'] >= self.settings['head_pose_consec_frames']: score += weights['head_nod']

                if abs(yaw) > self.settings['head_look_away_thresh']: self.counters['looking_away']+=1
                else: self.counters['looking_away']=0
                if self.counters['looking_away'] >= self.settings['head_pose_consec_frames']: score += weights['looking_away']

                # Determine final drowsiness level based on score
                levels = self.settings['drowsiness_levels']
                if score >= levels['very_drowsy_threshold']:
                    drowsiness_indicators['drowsiness_level'] = "Very Drowsy"
                elif score >= levels['slightly_drowsy_threshold']:
                    drowsiness_indicators['drowsiness_level'] = "Slightly Drowsy"

                drowsiness_indicators['details']['Score'] = score

        # --- Update state for next frame (skipped or processed) ---
        self.last_indicators = drowsiness_indicators
        self.last_landmarks = face_landmarks_data

        # --- Draw visuals on the ORIGINAL frame for high-quality output ---
        processed_frame = self.draw_visuals(original_frame, drowsiness_indicators, face_landmarks_data)
        
        # --- FIX: Cache the newly drawn frame ---
        self.last_drawn_frame = processed_frame

        # --- FIX: Return only the two values expected by the Gradio app ---
        return processed_frame, drowsiness_indicators

    def draw_visuals(self, frame, indicators, landmarks_data=None):
        """Helper function to draw all visualizations on the frame."""
        h, w, _ = frame.shape
        level = indicators['drowsiness_level']
        score_val = indicators.get("details", {}).get("Score", 0)
        color = (0, 255, 0) # Green for Awake

        if indicators['lighting'] == "Low":
            color = (0, 165, 255) # Orange
            cv2.putText(frame, "LOW LIGHT", (w // 2 - 120, h // 2), cv2.FONT_HERSHEY_SIMPLEX, 2, color, 3, cv2.LINE_AA)
        elif level == "Slightly Drowsy": color = (0, 255, 255) # Yellow
        elif level == "Very Drowsy": color = (0, 0, 255) # Red

        # Draw landmarks if they were detected
        if landmarks_data:
            landmarks = landmarks_data.landmark
            eye_mouth_landmarks_indices = self.L_EYE + self.R_EYE + self.MOUTH
            for idx in eye_mouth_landmarks_indices:
                lm = landmarks[idx]
                # Scale landmark coordinates to the full-sized frame
                x, y = int(lm.x * w), int(lm.y * h)
                cv2.circle(frame, (x, y), 2, (0, 255, 0), -1)

        cv2.rectangle(frame, (0, 0), (w - 1, h - 1), color, 10)
        status_text = f"Status: {level} (Score: {score_val:.2f})"
        cv2.putText(frame, status_text, (20, 40), cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 255, 255), 2, cv2.LINE_AA)

        return frame

    def analyse_frame(self, frame):
        # self.frame_counter += 1
        
        # --- FIX: More efficient frame skipping ---
        # # Adaptive skipping: process more frequently if drowsiness is detected.
        # last_level = self.last_indicators.get("drowsiness_level", "Awake")
        # skip_n = 1 if last_level != "Awake" else self.default_skip

        # if self.frame_counter % skip_n != 0:
        #     # If we have a cached frame, return it to avoid re-drawing.
        #     if self.last_drawn_frame is not None:
        #         return self.last_drawn_frame, self.last_indicators
        #     # Fallback if the first frame was skipped (unlikely but safe)
        #     else:
        #         return frame.copy(), self.last_indicators

        # --- CORE FRAME PROCESSING ---
        original_frame = frame.copy()
        h_orig, w_orig, _ = original_frame.shape
        
        # Optimization: Downscale frame for faster processing
        small_frame = cv2.resize(original_frame, (0, 0), fx=self.downscale_factor, fy=self.downscale_factor, interpolation=cv2.INTER_AREA)
        h, w, _ = small_frame.shape

        # All processing is done on the `small_frame` for speed.
        gray = cv2.cvtColor(small_frame, cv2.COLOR_BGR2GRAY)
        brightness = np.mean(gray)
        
        drowsiness_indicators = {"drowsiness_level": "Awake", "lighting": "Good", "details": {}}
        face_landmarks_data = None

        if brightness < self.settings['low_light_thresh']:
            drowsiness_indicators["lighting"] = "Low"
        else:
            # Convert the SMALL frame to RGB for MediaPipe
            img_rgb = cv2.cvtColor(small_frame, cv2.COLOR_BGR2RGB)
            img_rgb.flags.writeable = False # Performance enhancement
            results = self.face_mesh.process(img_rgb)
            img_rgb.flags.writeable = True

            if results.multi_face_landmarks:
                face_landmarks_data = results.multi_face_landmarks[0]
                landmarks = face_landmarks_data.landmark
                score = 0
                weights = self.settings['indicator_weights']

                # --- Drowsiness Calculations (on small frame dimensions 'h', 'w') ---
                ear_left = calculate_ear([landmarks[i] for i in self.L_EYE],(h,w))
                ear_right = calculate_ear([landmarks[i] for i in self.R_EYE],(h,w))
                ear = (ear_left + ear_right) / 2.0
                
                if ear < self.settings['eye_ar_thresh']: self.counters['eye_closure']+=1
                else: self.counters['eye_closure']=0
                if self.counters['eye_closure'] >= self.settings['eye_ar_consec_frames']: score += weights['eye_closure']

                mar = calculate_mar([landmarks[i] for i in self.MOUTH], (h, w))
                if mar > self.settings['yawn_mar_thresh']: self.counters['yawning']+=1
                else: self.counters['yawning']=0
                if self.counters['yawning'] >= self.settings['yawn_consec_frames']: score += weights['yawning']

                # --- Head Pose Estimation (on small frame dimensions 'h', 'w') ---
                face_3d_model = 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 corner
                    [-150.0, -150.0, -125.0],    # Left Mouth corner
                    [150.0, -150.0, -125.0]      # Right mouth corner
                ], dtype=np.float32)
                
                face_2d_points = np.array([(landmarks[i].x * w, landmarks[i].y * h) for i in self.HEAD_POSE_LANDMARKS], dtype=np.float32)
                cam_matrix = np.array([[w, 0, w/2], [0, w, h/2], [0, 0, 1]], dtype=np.float32)
                
                _, rvec, _ = cv2.solvePnP(face_3d_model, face_2d_points, cam_matrix, self.zeros_4x1, flags=cv2.SOLVEPNP_EPNP)
                rmat, _ = cv2.Rodrigues(rvec)
                angles, _, _, _, _, _ = cv2.RQDecomp3x3(rmat)
                pitch, yaw = angles[0], angles[1]

                if pitch > self.settings['head_nod_thresh']: self.counters['head_nod']+=1
                else: self.counters['head_nod']=0
                if self.counters['head_nod'] >= self.settings['head_pose_consec_frames']: score += weights['head_nod']

                if abs(yaw) > self.settings['head_look_away_thresh']: self.counters['looking_away']+=1
                else: self.counters['looking_away']=0
                if self.counters['looking_away'] >= self.settings['head_pose_consec_frames']: score += weights['looking_away']

                # Determine final drowsiness level based on score
                levels = self.settings['drowsiness_levels']
                if score >= levels['very_drowsy_threshold']:
                    drowsiness_indicators['drowsiness_level'] = "Very Drowsy"
                elif score >= levels['slightly_drowsy_threshold']:
                    drowsiness_indicators['drowsiness_level'] = "Slightly Drowsy"

                drowsiness_indicators['details']['Score'] = score

        # --- Update state for next frame (skipped or processed) ---
        self.last_indicators = drowsiness_indicators
        self.last_landmarks = face_landmarks_data

        # --- Draw visuals on the ORIGINAL frame for high-quality output ---
        # processed_frame = self.draw_visuals(original_frame, drowsiness_indicators, face_landmarks_data)
        
        # --- FIX: Cache the newly drawn frame ---
        # self.last_drawn_frame = processed_frame

        # --- FIX: Return only the two values expected by the Gradio app ---
        return drowsiness_indicators