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
| import numpy as np
from sklearn.neighbors import KNeighborsClassifier
from sklearn.preprocessing import StandardScaler
class DrowsinessFeatureExtractor:
"""
Mercedes-Benz疲劳检测特征提取器
论文方法：从标准座舱摄像头提取35个特征
分3大类：眨眼动态 + 头部运动 + PERCLOS
"""
def extract_blink_features(self, eye_tracking_data):
"""
眨眼动态特征（~15个）
"""
blink_events = self._detect_blinks(eye_tracking_data)
features = {
'blink_frequency': len(blink_events) / (len(eye_tracking_data) / 30),
'blink_interval_mean': np.mean([b['interval'] for b in blink_events]),
'blink_interval_std': np.std([b['interval'] for b in blink_events]),
'blink_duration_mean': np.mean([b['duration'] for b in blink_events]),
'blink_duration_std': np.std([b['duration'] for b in blink_events]),
'blink_duration_max': np.max([b['duration'] for b in blink_events]),
'closing_velocity_mean': np.mean([b['closing_speed'] for b in blink_events]),
'closing_velocity_std': np.std([b['closing_speed'] for b in blink_events]),
'reopening_velocity_mean': np.mean([b['reopening_speed'] for b in blink_events]),
'reopening_velocity_std': np.std([b['reopening_speed'] for b in blink_events]),
'eyelid_cleft_max': np.max([b['max_openness'] for b in blink_events]),
'eyelid_cleft_mean': np.mean([b['max_openness'] for b in blink_events]),
'eyelid_cleft_std': np.std([b['max_openness'] for b in blink_events]),
'reopening_duration_mean': np.mean([b['reopen_duration'] for b in blink_events]),
'reopening_duration_std': np.std([b['reopen_duration'] for b in blink_events]),
}
return features
def extract_head_features(self, head_pose_data):
"""
头部运动特征（~10个）
"""
roll = head_pose_data['roll']
pitch = head_pose_data['pitch']
yaw = head_pose_data['yaw']
features = {
'head_nodding_freq': self._count_nods(pitch) / (len(pitch) / 30),
'head_nodding_amplitude': np.std(np.diff(pitch)),
'head_nodding_mean': np.mean(pitch),
'head_nodding_std': np.std(pitch),
'head_bobbing_freq': self._count_bobs(roll) / (len(roll) / 30),
'head_bobbing_amplitude': np.std(np.diff(roll)),
'head_bobbing_mean': np.mean(roll),
'head_bobbing_std': np.std(roll),
'head_movement_total': np.mean(np.sqrt(
np.diff(pitch)**2 + np.diff(roll)**2 + np.diff(yaw)**2
)),
'head_stillness_ratio': self._calc_stillness_ratio(pitch, roll, yaw),
}
return features
def extract_perclos_features(self, eye_openness):
"""
PERCLOS特征（~10个）
"""
threshold_80 = 0.2
features = {
'perclos_60s': self._calc_perclos(eye_openness, window=60, threshold=threshold_80),
'perclos_30s': self._calc_perclos(eye_openness, window=30, threshold=threshold_80),
'perclos_120s': self._calc_perclos(eye_openness, window=120, threshold=threshold_80),
'eye_openness_mean': np.mean(eye_openness),
'eye_openness_std': np.std(eye_openness),
'eye_openness_min': np.min(eye_openness),
'closed_time_ratio': np.mean(eye_openness < threshold_80),
'longest_closure': self._longest_closure(eye_openness, threshold_80),
'closure_count': np.sum(np.diff((eye_openness < threshold_80).astype(int)) > 0),
'mean_closure_duration': self._mean_closure_duration(eye_openness, threshold_80),
}
return features
def _detect_blinks(self, data):
"""检测眨眼事件"""
blinks = []
is_closed = data < 0.2
transitions = np.diff(is_closed.astype(int))
close_idx = np.where(transitions == 1)[0]
open_idx = np.where(transitions == -1)[0]
for c, o in zip(close_idx, open_idx):
blinks.append({
'duration': (o - c) / 30,
'closing_speed': 0.3 / (c - max(close_idx[close_idx < c-1], default=c-5) + 1),
'reopening_speed': 0.3 / (o - c),
'max_openness': 0.8,
'reopen_duration': (o - c) / 30,
'interval': 0
})
return blinks
def _calc_perclos(self, data, window, threshold):
"""计算PERCLOS"""
n = len(data)
w = window * 30
values = []
for i in range(0, n - w, w):
values.append(np.mean(data[i:i+w] < threshold) * 100)
return np.mean(values) if values else 0
def _count_nods(self, pitch):
return np.sum(np.abs(np.diff(pitch)) > 5)
def _count_bobs(self, roll):
return np.sum(np.abs(np.diff(roll)) > 5)
def _calc_stillness_ratio(self, p, r, y):
movement = np.sqrt(np.diff(p)**2 + np.diff(r)**2 + np.diff(y)**2)
return np.mean(movement < 1)
def _longest_closure(self, data, threshold):
is_closed = data < threshold
max_len = current = 0
for c in is_closed:
if c:
current += 1
max_len = max(max_len, current)
else:
current = 0
return max_len / 30
def _mean_closure_duration(self, data, threshold):
is_closed = data < threshold
transitions = np.diff(is_closed.astype(int))
close_idx = np.where(transitions == 1)[0]
open_idx = np.where(transitions == -1)[0]
durations = [(o - c) / 30 for c, o in zip(close_idx, open_idx)]
return np.mean(durations) if durations else 0
class MercedesDrowsinessClassifier:
"""
Mercedes-Benz k-NN疲劳分类器
论文核心模型
二分类: Awake vs Drowsy (84.2% BA)
三分类: Awake/Questionable/Drowsy (70.0% with OvO)
"""
def __init__(self, n_neighbors=100):
self.scaler = StandardScaler()
self.knn_binary = KNeighborsClassifier(
n_neighbors=n_neighbors,
weights='distance',
metric='euclidean'
)
self.knn_multiclass = KNeighborsClassifier(
n_neighbors=n_neighbors,
weights='distance',
metric='euclidean'
)
self.feature_extractor = DrowsinessFeatureExtractor()
def fit(self, features, labels):
"""训练"""
features_scaled = self.scaler.fit_transform(features)
self.knn_binary.fit(features_scaled, labels)
def predict(self, features):
"""预测"""
features_scaled = self.scaler.transform(features)
return self.knn_binary.predict(features_scaled)
def explain(self, features, k=5):
"""
可解释性：返回k个最近邻
安全关键系统的可审计需求
"""
features_scaled = self.scalaler.transform(features)
distances, indices = self.knn_binary.kneighbors(features_scaled, k)
return {
'prediction': self.knn_binary.predict(features_scaled),
'neighbors': indices,
'distances': distances,
'explanation': f"决策基于{k}个最相似的历史样本"
}
if __name__ == "__main__":
np.random.seed(42)
n_samples = 1000
n_features = 35
X = np.random.randn(n_samples, n_features)
y = np.random.choice(['awake', 'drowsy'], n_samples, p=[0.6, 0.4])
clf = MercedesDrowsinessClassifier(n_neighbors=100)
clf.fit(X, y)
pred = clf.predict(X[:5])
print(f"预测结果: {pred}")
print(f"\n特征数: {n_features}")
print("论文发现: ~10个特征足够（Sequential Forward Selection）")
print("关键特征: reopening_duration, eyelid_cleft, head_nodding")
|
