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
| """
CNN-LSTM睡眠检测模型
参考：arXiv 2024 Survey
ResearchGate CNN-LSTM + PERCLOS + Yawning
架构：
1. CNN提取面部特征
2. LSTM建模时序依赖
3. 多特征融合（PERCLOS + 哈欠频率）
"""
import torch
import torch.nn as nn
import numpy as np
class CNNLSTM_Drowsiness(nn.Module):
"""
CNN-LSTM睡眠检测模型
结构：
- CNN特征提取：ResNet/MobileNet
- LSTM时序建模：捕捉眨眼/哈欠序列
- 多特征融合：PERCLOS + Yawning + Head Pose
"""
def __init__(
self,
cnn_feature_dim: int = 512,
lstm_hidden_dim: int = 256,
lstm_layers: int = 2,
additional_features: int = 10,
num_classes: int = 3,
dropout: float = 0.3
):
super().__init__()
self.cnn = nn.Sequential(
nn.Conv2d(3, 64, 7, stride=2, padding=3),
nn.BatchNorm2d(64),
nn.ReLU(),
nn.MaxPool2d(3, stride=2, padding=1),
nn.Conv2d(64, 128, 3, padding=1),
nn.BatchNorm2d(128),
nn.ReLU(),
nn.MaxPool2d(2),
nn.Conv2d(128, 256, 3, padding=1),
nn.BatchNorm2d(256),
nn.ReLU(),
nn.MaxPool2d(2),
nn.Conv2d(256, cnn_feature_dim, 3, padding=1),
nn.BatchNorm2d(cnn_feature_dim),
nn.ReLU(),
nn.AdaptiveAvgPool2d(1)
)
self.lstm = nn.LSTM(
input_size=cnn_feature_dim + additional_features,
hidden_size=lstm_hidden_dim,
num_layers=lstm_layers,
batch_first=True,
dropout=dropout,
bidirectional=True
)
self.classifier = nn.Sequential(
nn.Linear(lstm_hidden_dim * 2, lstm_hidden_dim),
nn.ReLU(),
nn.Dropout(dropout),
nn.Linear(lstm_hidden_dim, num_classes)
)
def forward(
self,
frame_sequence: torch.Tensor,
additional_features: torch.Tensor
) -> torch.Tensor:
"""
前向传播
Args:
frame_sequence: 视频帧序列, shape=(B, T, C, H, W)
additional_features: PERCLOS/Yawning等特征, shape=(B, T, F)
Returns:
logits: 分类输出, shape=(B, num_classes)
"""
B, T, C, H, W = frame_sequence.shape
cnn_features = []
for t in range(T):
frame = frame_sequence[:, t]
feat = self.cnn(frame)
feat = feat.view(B, -1)
cnn_features.append(feat)
cnn_features = torch.stack(cnn_features, dim=1)
combined_features = torch.cat([
cnn_features, additional_features
], dim=-1)
lstm_out, (h_n, c_n) = self.lstm(combined_features)
final_hidden = lstm_out[:, -1, :]
logits = self.classifier(final_hidden)
return logits
if __name__ == "__main__":
model = CNNLSTM_Drowsiness(
cnn_feature_dim=512,
lstm_hidden_dim=256,
additional_features=10,
num_classes=3
)
frames = torch.randn(4, 30, 3, 112, 112)
features = torch.randn(4, 30, 10)
logits = model(frames, features)
probs = torch.softmax(logits, dim=-1)
print(f"输出shape: {logits.shape}")
print(f"预测类别分布:")
classes = ['正常', '疲劳', '睡眠']
for i in range(4):
for j, cls in enumerate(classes):
print(f" 样本{i+1} {cls}: {probs[i, j].item():.2%}")
|