import numpy as np
import cv2
from typing import Tuple

class AptivAOC:
    """
    Aptiv 摄像头乘员分类系统
    
    基于车内摄像头实现乘员分类
    """
    
    def __init__(self):
        # 加载模型（假设已训练）
        self.detector = self._load_detector()
        self.pose_estimator = self._load_pose_estimator()
        self.classifier = self._load_classifier()
        
    def classify_occupant(self, frame: np.ndarray) -> Tuple[str, float]:
        """
        分类乘员
        
        Args:
            frame: 输入图像（座舱摄像头）
            
        Returns:
            occupant_type: "adult" | "child" | "empty" | "pet"
            confidence: 置信度 [0, 1]
        """
        # 1. 检测人体
        detections = self.detector.detect(frame)
        
        if len(detections) == 0:
            return "empty", 0.95
        
        # 取最大检测框（假设为主乘员）
        main_detection = max(detections, key=lambda d: d['confidence'])
        
        # 2. 提取关键点
        keypoints = self.pose_estimator.estimate(frame, main_detection['bbox'])
        
        # 3. 提取特征
        features = self._extract_features(frame, main_detection, keypoints)
        
        # 4. 分类
        occupant_type, confidence = self.classifier.predict(features)
        
        return occupant_type, confidence
    
    def _extract_features(self, 
                          frame: np.ndarray,
                          detection: dict,
                          keypoints: np.ndarray) -> np.ndarray:
        """
        提取特征
        
        Args:
            frame: 输入图像
            detection: 检测结果
            keypoints: 关键点 (N, 3), (x, y, confidence)
            
        Returns:
            features: 特征向量
        """
        features = []
        
        # 1. 边界框大小（相对图像尺寸）
        bbox = detection['bbox']
        bbox_area = (bbox[2] - bbox[0]) * (bbox[3] - bbox[1])
        image_area = frame.shape[0] * frame.shape[1]
        bbox_ratio = bbox_area / image_area
        features.append(bbox_ratio)
        
        # 2. 头部大小（相对边界框）
        head_keypoints = keypoints[0:5]  # 假设前5个为头部关键点
        head_width = np.max(head_keypoints[:, 0]) - np.min(head_keypoints[:, 0])
        head_height = np.max(head_keypoints[:, 1]) - np.min(head_keypoints[:, 1])
        head_area = head_width * head_height
        head_ratio = head_area / bbox_area
        features.append(head_ratio)
        
        # 3. 身体比例（头身比）
        body_height = np.max(keypoints[:, 1]) - np.min(keypoints[:, 1])
        head_body_ratio = head_height / body_height if body_height > 0 else 0
        features.append(head_body_ratio)
        
        # 4. 肩宽（相对身高）
        shoulder_width = np.linalg.norm(keypoints[5] - keypoints[6])  # 左右肩关键点
        shoulder_ratio = shoulder_width / body_height if body_height > 0 else 0
        features.append(shoulder_ratio)
        
        return np.array(features)


# 测试
if __name__ == "__main__":
    aoc = AptivAOC()
    
    # 模拟输入
    frame = np.zeros((480, 640, 3), dtype=np.uint8)
    
    # 分类
    occupant_type, confidence = aoc.classify_occupant(frame)
    
    print(f"乘员类型：{occupant_type}")
    print(f"置信度：{confidence:.2%}")