Dapalm | IMS技术前沿

乘员分类与自适应约束：Euro NCAP 2026要求下的传感器融合方案

技术背景

• 核心要求： Euro NCAP 2026 Protocol v1.2 — 气囊必须按乘员大小/位置/姿态自适应
• 关键产业： IEE Smart Sensing BodySense™、BeBop Sensors织物压力、InCabin Europe 2026
• 标准： FMVSS 208(美国)、Euro NCAP(欧洲)、GB 14167(中国)
• 成本目标： 织物传感器~$7-10（InCabin 2026报告：现有技术的一半成本）

Euro NCAP 2026 乘员分类要求清单

要求编号	场景	检测要求	时限	分值
OC-01	成人 vs 儿童	重量+体型分类	≤2秒	2分
OC-02	后向儿童座椅	自动识别+气囊禁用	≤2秒	3分
OC-03	体型小成人(前排)	气囊低力度展开	≤2秒	1分
OC-04	脚搭仪表台	OOP检测+警告	≤5秒	2分
OC-05	上身靠近仪表台	<20cm检测+警告	≤3秒	2分
OC-06	后排乘员存在	存在+安全带状态	实时	2分
OC-07	eCall乘员数据	碰撞时传输人数	实时	3分

传感器融合方案

多传感器分类架构

graph TB
    subgraph 传感器层
        A[座椅重量传感器<br/>压阻式 4点]
        B[座椅压力阵列<br/>16x16织物]
        C[OMS摄像头<br/>RGB-IR]
        D[mmWave雷达<br/>60GHz]
        E[安全带张力<br/>应变片]
    end
    
    subgraph 特征提取
        A --> F[重量±2kg]
        B --> G[压力分布图]
        C --> H[人体关键点<br/>体型估计]
        D --> I[呼吸率<br/>存在检测]
        E --> J[安全带状态]
    end
    
    subgraph 融合分类
        F --> K[Bayesian融合]
        G --> K
        H --> K
        I --> K
        J --> K
        K --> L{乘员分类}
    end
    
    subgraph 输出
        L --> M[成人/儿童/儿童座椅]
        L --> N[体型分类]
        L --> O[姿态/OOP]
        L --> P[气囊策略]
    end

传感器BOM

传感器	型号/规格	覆盖场景	单价	来源
座椅重量	4点压阻式传感器	OC-01/02	$5	IEE/通用
压力阵列	16×16织物压力	OC-04/05	$7	BeBop/AIQ
OMS摄像头	OV2311 RGB-IR	OC-03/04/05	$8	OmniVision
60GHz雷达	TI IWR6843AOP	OC-06/呼吸	$15	TI
安全带张力	应变片	安全带状态	$3	通用
总计	-	-	$38	-

技术实现

多传感器乘员分类引擎

"""
Euro NCAP 2026 乘员分类系统

融合: 重量 + 压力分布 + 摄像头 + 雷达 + 安全带
依赖: pip install numpy scipy

输出:
- 乘员类别: 成人/儿童/儿童座椅/空座
- 体型: 小/中/大
- 姿态: 正常/OOP
- 气囊策略: 正常/低力度/禁用
"""

import numpy as np
from typing import Dict, Tuple, Optional
from dataclasses import dataclass
from enum import IntEnum

class OccupantCategory(IntEnum):
    EMPTY = 0
    CHILD_SEAT_REAR = 1    # 后向儿童座椅
    CHILD_SEAT_FORWARD = 2 # 前向儿童座椅
    CHILD = 3              # 儿童(<36kg)
    SMALL_ADULT = 4        # 小体型成人(<50kg)
    ADULT = 5              # 成人(50-100kg)
    LARGE_ADULT = 6        # 大体型成人(>100kg)

class AirbagStrategy(IntEnum):
    DISABLE = 0    # 禁用
    LOW_POWER = 1  # 低力度
    NORMAL = 2     # 正常

@dataclass
class SensorReadings:
    """传感器读数"""
    weight_kg: float = 0.0
    pressure_map: Optional[np.ndarray] = None  # 16x16
    body_keypoints: Optional[np.ndarray] = None  # 17个关键点
    breathing_rate: float = 0.0  # 雷达呼吸率
    presence_detected: bool = False  # 雷达存在检测
    seatbelt_buckled: bool = False
    seatbelt_position: str = 'correct'

class OccupantClassifier:
    """
    多传感器乘员分类器
    
    方法: 加权证据融合
    每个传感器提供独立证据，加权融合后决策
    """
    
    def __init__(self):
        self.weight_thresholds = {
            'empty': 2.0,      # <2kg = 空座
            'child': 36.0,     # <36kg = 儿童
            'small': 50.0,     # <50kg = 小成人
            'large': 100.0,    # >100kg = 大成人
        }
    
    def classify_from_weight(self, weight: float) -> Tuple[OccupantCategory, float]:
        """重量分类"""
        if weight < self.weight_thresholds['empty']:
            return OccupantCategory.EMPTY, 0.9
        elif weight < self.weight_thresholds['child']:
            return OccupantCategory.CHILD, 0.7
        elif weight < self.weight_thresholds['small']:
            return OccupantCategory.SMALL_ADULT, 0.65
        elif weight < self.weight_thresholds['large']:
            return OccupantCategory.ADULT, 0.8
        else:
            return OccupantCategory.LARGE_ADULT, 0.75
    
    def classify_from_pressure(self, pressure_map: np.ndarray) -> Tuple[OccupantCategory, float]:
        """压力分布分类"""
        total_pressure = np.sum(pressure_map)
        pressure_area = np.sum(pressure_map > 0.1)  # 有效接触面积
        
        if total_pressure < 1.0:
            return OccupantCategory.EMPTY, 0.85
        
        # 后向儿童座椅特征：压力集中在中部+头部区域有压力
        center_pressure = np.sum(pressure_map[4:12, 6:10])
        upper_pressure = np.sum(pressure_map[:4, 4:12])
        
        if center_pressure > total_pressure * 0.6 and upper_pressure > 5:
            return OccupantCategory.CHILD_SEAT_REAR, 0.6
        
        # 按接触面积分类
        if pressure_area < 50:
            return OccupantCategory.CHILD, 0.6
        elif pressure_area < 100:
            return OccupantCategory.SMALL_ADULT, 0.55
        elif pressure_area < 200:
            return OccupantCategory.ADULT, 0.7
        else:
            return OccupantCategory.LARGE_ADULT, 0.65
    
    def classify_from_camera(self, keypoints: np.ndarray) -> Tuple[OccupantCategory, float]:
        """摄像头体型估计"""
        if keypoints is None or len(keypoints) < 5:
            return OccupantCategory.ADULT, 0.3  # 默认假设
        
        # 肩宽估计
        left_shoulder = keypoints[5]  # 左肩
        right_shoulder = keypoints[6]  # 右肩
        shoulder_width = np.sqrt(
            (left_shoulder[0] - right_shoulder[0])**2 +
            (left_shoulder[1] - right_shoulder[1])**2
        )
        
        # 坐高估计
        nose = keypoints[0]
        hip = keypoints[11]
        sitting_height = np.sqrt(
            (nose[0] - hip[0])**2 + (nose[1] - hip[1])**2
        )
        
        # 分类
        if shoulder_width < 80 and sitting_height < 150:
            return OccupantCategory.CHILD, 0.6
        elif shoulder_width < 100:
            return OccupantCategory.SMALL_ADULT, 0.5
        elif shoulder_width > 160:
            return OccupantCategory.LARGE_ADULT, 0.55
        else:
            return OccupantCategory.ADULT, 0.6
    
    def detect_child_seat(self, weight: float, 
                           pressure_map: np.ndarray,
                           breathing_rate: float) -> Tuple[bool, str, float]:
        """
        儿童座椅检测
        
        特征:
        - 重量异常轻但压力分布有座椅轮廓
        - 呼吸率偏高（婴幼儿30-40次/分 vs 成人12-20次/分）
        - 压力分布模式不同于人体
        """
        # 后向儿童座椅
        if weight < 15 and breathing_rate > 25:
            return True, 'rear_facing', 0.7
        
        # 压力分布异常（座椅模式）
        if pressure_map is not None:
            center = pressure_map[6:10, 6:10]
            edges = pressure_map[:2, :].sum() + pressure_map[-2:, :].sum()
            if center.mean() > edges * 0.3:
                return True, 'forward_facing', 0.5
        
        return False, '', 0.0
    
    def detect_oop(self, pressure_map: np.ndarray,
                    keypoints: np.ndarray,
                    distance_to_dash: float) -> Dict:
        """
        OOP异常姿态检测
        """
        oop_results = {}
        
        # 脚搭仪表台
        if keypoints is not None and len(keypoints) > 15:
            left_ankle = keypoints[15]
            right_ankle = keypoints[16]
            if left_ankle[1] < 100 or right_ankle[1] < 100:  # y坐标小说明在前方
                oop_results['feet_on_dash'] = True
        
        # 上身靠近仪表台
        if distance_to_dash < 20:
            oop_results['too_close'] = True
            oop_results['distance'] = distance_to_dash
        
        # 前伏
        if pressure_map is not None:
            upper = pressure_map[:6, :].sum()
            lower = pressure_map[10:, :].sum()
            if upper > lower * 1.5:
                oop_results['forward_bent'] = True
        
        return oop_results
    
    def determine_airbag_strategy(self, category: OccupantCategory,
                                    oop: Dict) -> AirbagStrategy:
        """确定气囊策略"""
        if category == OccupantCategory.CHILD_SEAT_REAR:
            return AirbagStrategy.DISABLE
        elif category == OccupantCategory.CHILD:
            return AirbagStrategy.DISABLE
        elif category == OccupantCategory.SMALL_ADULT:
            return AirbagStrategy.LOW_POWER
        elif oop.get('too_close') or oop.get('feet_on_dash'):
            return AirbagStrategy.LOW_POWER
        else:
            return AirbagStrategy.NORMAL
    
    def classify(self, readings: SensorReadings) -> Dict:
        """
        完整乘员分类
        """
        # 各传感器独立分类
        w_cat, w_conf = self.classify_from_weight(readings.weight_kg)
        p_cat = OccupantCategory.ADULT
        p_conf = 0.3
        if readings.pressure_map is not None:
            p_cat, p_conf = self.classify_from_pressure(readings.pressure_map)
        
        c_cat = OccupantCategory.ADULT
        c_conf = 0.3
        if readings.body_keypoints is not None:
            c_cat, c_conf = self.classify_from_camera(readings.body_keypoints)
        
        # 儿童座椅检测
        is_child_seat, seat_type, cs_conf = self.detect_child_seat(
            readings.weight_kg,
            readings.pressure_map,
            readings.breathing_rate
        )
        
        # 证据融合（加权投票）
        weights = {'weight': 0.35, 'pressure': 0.30, 'camera': 0.35}
        
        if is_child_seat and cs_conf > 0.5:
            final_category = (OccupantCategory.CHILD_SEAT_REAR 
                            if seat_type == 'rear_facing'
                            else OccupantCategory.CHILD_SEAT_FORWARD)
            final_conf = cs_conf
        else:
            # 加权投票
            votes = {}
            for cat, conf, w in [
                (w_cat, w_conf, weights['weight']),
                (p_cat, p_conf, weights['pressure']),
                (c_cat, c_conf, weights['camera']),
            ]:
                if cat not in votes:
                    votes[cat] = 0
                votes[cat] += conf * w
            
            final_category = max(votes, key=votes.get)
            final_conf = votes[final_category]
        
        # OOP检测
        distance = 999.0
        if readings.body_keypoints is not None:
            # 简化距离估计
            distance = 30.0  # 实际应从深度图计算
        
        oop = self.detect_oop(
            readings.pressure_map if readings.pressure_map is not None else np.zeros((16,16)),
            readings.body_keypoints,
            distance
        )
        
        # 气囊策略
        airbag = self.determine_airbag_strategy(final_category, oop)
        
        return {
            'category': final_category.name,
            'confidence': round(final_conf, 2),
            'weight_kg': readings.weight_kg,
            'child_seat': is_child_seat,
            'seat_type': seat_type,
            'oop': oop,
            'airbag_strategy': airbag.name,
            'seatbelt_status': 'buckled' if readings.seatbelt_buckled else 'unbuckled',
        }


# 测试
if __name__ == "__main__":
    np.random.seed(42)
    classifier = OccupantClassifier()
    
    # 成人
    adult = SensorReadings(
        weight_kg=75,
        pressure_map=np.random.rand(16, 16) * 0.3 + 0.3,
        body_keypoints=np.random.rand(17, 2) * 200,
        breathing_rate=16,
        presence_detected=True,
        seatbelt_buckled=True,
    )
    
    # 儿童
    child = SensorReadings(
        weight_kg=22,
        pressure_map=np.random.rand(16, 16) * 0.2 + 0.1,
        body_keypoints=np.random.rand(17, 2) * 100,
        breathing_rate=28,
        presence_detected=True,
        seatbelt_buckled=True,
    )
    
    # 后向儿童座椅
    infant = SensorReadings(
        weight_kg=8,
        pressure_map=np.random.rand(16, 16) * 0.1,
        breathing_rate=35,
        presence_detected=True,
    )
    
    print("=== 成人 ===")
    r = classifier.classify(adult)
    for k, v in r.items():
        print(f"  {k}: {v}")
    
    print(f"\n=== 儿童 ===")
    r = classifier.classify(child)
    for k, v in r.items():
        print(f"  {k}: {v}")
    
    print(f"\n=== 后向儿童座椅 ===")
    r = classifier.classify(infant)
    for k, v in r.items():
        print(f"  {k}: {v}")

织物压力传感器方案

InCabin Europe 2026 产业报告

“最成熟的应用是新一代座椅占用分类，使用座椅装饰下方薄织物传感器，成本约为现有技术的一半” — InCabin 2026

参数	织物传感器(新)	重量传感器(现有)
厚度	<2mm	~10mm
成本	~$7	~$15
铺设	座椅面下方全铺	座椅底部4点
分辨率	16×16=256点	4点
分类能力	体型+姿态	仅重量
耐久	>10年	>10年
水洗	✅	❌
供应商	BeBop/AIQ/IEE	IEE/H.B. Fuller

参考文献

1. Euro NCAP Protocol v1.2: https://cdn.euroncap.com/cars/assets/Euro_NCAP_Protocol_Safe_Driving_Occupant_Monitoring_v1_2_aebbc7361f.pdf
2. Smart Eye: https://smarteye.se/blog/euro-ncap-2026-new-standards-for-occupant-monitoring-and-adaptive-restraints/
3. InCabin 2026: https://incabin.com/blog/incabin-europe-2026-press-conference-recap/
4. IEE BodySense: https://iee-sensing.com/automotive/safety-and-comfort/occupant-classification/
5. Anyverse OEM指南: https://anyverse.ai/euro-ncap-2026-in-cabin-monitoring-oem-guidelines-to-readiness/
6. EDAG儿童安全: https://insights.edag.com/en/child-safety-euro-ncap-ratings