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
| import numpy as np
class GazeCalibrationManager:
"""
DMS视线估计校准管理器
核心问题：
1. 实验室训练的模型在实车部署时有系统偏移
2. 偏移量在不同驾驶会话间变化
3. 需要在线重校准
校准方案：
1. 出厂校准：生产线固定点校准
2. 会话校准：每次启动驾驶时快速校准
3. 在线校准：行驶中持续微调
"""
def __init__(self):
self.offset = None
self.calibration_history = []
self.confidence_threshold = 0.5
def factory_calibration(self, gaze_samples: np.ndarray,
target_points: np.ndarray) -> dict:
"""
出厂校准
Args:
gaze_samples: (N, 2) 校准点视线预测
target_points: (N, 2) 真实目标点
Returns:
{'offset': (2,), 'residual_error': float}
"""
offset = np.mean(target_points - gaze_samples, axis=0)
residual = np.std(target_points - gaze_samples - offset, axis=0)
self.offset = offset
return {
'offset': offset.tolist(),
'residual_error_rad': float(np.linalg.norm(residual)),
'residual_error_deg': float(np.degrees(np.linalg.norm(residual))),
'n_samples': len(gaze_samples)
}
def session_calibration(self, gaze_samples: np.ndarray,
target_points: np.ndarray) -> dict:
"""
会话校准（每次启动驾驶）
使用少量点快速校准
"""
if len(gaze_samples) < 3:
return {'status': 'insufficient_data'}
session_offset = np.mean(target_points - gaze_samples, axis=0)
if self.offset is not None:
alpha = 0.3
adjusted_offset = (1 - alpha) * self.offset + alpha * session_offset
else:
adjusted_offset = session_offset
self.offset = adjusted_offset
return {
'offset': adjusted_offset.tolist(),
'status': 'calibrated',
'n_samples': len(gaze_samples)
}
def apply_calibration(self, gaze: np.ndarray) -> np.ndarray:
"""应用校准偏移"""
if self.offset is None:
return gaze
return gaze + self.offset
def online_update(self, gaze: np.ndarray,
road_gaze_estimate: np.ndarray,
confidence: float = 0.5):
"""
在线更新（行驶中）
使用道路上下文作为弱监督：
- 直道行驶时，视线应在前方道路
- 车道保持时，视线应在道路中心区域
"""
if confidence < self.confidence_threshold:
return
delta = road_gaze_estimate - gaze
alpha = 0.001
self.offset = (1 - alpha) * self.offset + alpha * delta
if __name__ == "__main__":
calibrator = GazeCalibrationManager()
np.random.seed(42)
n = 9
factory_gaze = np.random.randn(n, 2) * 0.05
factory_targets = np.array([
[-0.5, -0.3], [-0.5, 0], [-0.5, 0.3],
[0, -0.3], [0, 0], [0, 0.3],
[0.5, -0.3], [0.5, 0], [0.5, 0.3],
])
result = calibrator.factory_calibration(factory_gaze, factory_targets)
print("出厂校准:", result)
session_gaze = factory_gaze + np.array([0.05, -0.03])
result = calibrator.session_calibration(session_gaze[:5], factory_targets[:5])
print("会话校准:", result)
test_gaze = np.array([0.1, 0.05])
calibrated = calibrator.apply_calibration(test_gaze)
print(f"原始视线: {test_gaze} → 校准后: {calibrated}")
|
