hongchen/door_detect/scene_processor.py

"""
场景点云处理器
整合 YOLOE 门检测 + 全景图点云转换 + 位姿变换 + 多视角融合 + 3D 门合并
支持 scene0001 格式的场景数据
"""
import os
import sys
import json
import argparse
from pathlib import Path
from dataclasses import dataclass, field
from typing import Dict, List, Optional, Tuple

# 添加当前目录到路径，以便导入 camera_spherical
sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))

import cv2
import numpy as np
import open3d as o3d
from tqdm import tqdm
from ultralytics import YOLOE
from camera_spherical import Intrinsic_Spherical_NP


# ============================================================================
# 数据类
# ============================================================================
@dataclass
class PoseData:
    """位姿数据"""
    uuid: str
    rotation: Dict[str, float]  # w, x, y, z
    translation: Dict[str, float]  # x, y, z


@dataclass
class MaskDetection:
    """单张图像中的 mask 检测结果"""
    image_name: str
    mask_contours: List[List[List[float]]]  # 2D 轮廓像素坐标 [[x, y], ...]
    scores: List[float]
    mask_3d_points: List[np.ndarray]  # 每个 mask 对应的世界坐标系 3D 点


@dataclass
class Door3D:
    """3D 门实例"""
    id: int
    center: np.ndarray  # 中心坐标 [x, y, z]
    bbox_8points: np.ndarray  # 8 个角点 [[x,y,z], ...] (8x3)
    source_detections: List[Dict]  # 来源检测信息


@dataclass
class EntranceInfo:
    """入户门信息"""
    is_detected: bool  # 是否通过检测确定
    door: Optional[Door3D]  # 入户门对象（检测确定时）
    center: np.ndarray  # 入户门中心/估计位置 [x, y, z]
    score: float  # 置信度评分
    method: str  # 确定方法："exterior_class" / "size_score" / "fallback_2d" / "scene_center_estimate"
    reason: str  # 说明文字


# ============================================================================
# 场景处理器
# ============================================================================
class SceneProcessor:
    """场景处理器：整合检测 + 点云生成 + 位姿变换 + 3D 门融合"""

    # 门相关类别 - 统一映射为 "door"
    DOOR_CLASSES = [
          "door", "indoor door", "exterior door",
        "wooden door", "metal door", "glass door", "double door",
        "single door",  "open door", "closed door"
    ]

    # 3D 门合并参数
    MERGE_IOU_THRESH = 0.3       # 3D IoU 阈值
    MERGE_DIST_THRESH = 2.0      # 中心距离阈值 (米)

    # 3D 门过滤参数 - 有效门的物理特性
    DOOR_HEIGHT_MIN = 1.0        # 最小高度 (米)
    DOOR_HEIGHT_MAX = 3.0        # 最大高度 (米)
    DOOR_WIDTH_MIN = 0.3         # 最小宽度 (米)
    DOOR_WIDTH_MAX = 3.0         # 最大宽度 (米)
    DOOR_THICKNESS_MAX = 0.5     # 最大厚度 (米) - 门的深度方向
    GROUND_DIST_THRESH = 0.5     # 门底部距地面最大距离 (米)

    def __init__(
        self,
        scene_folder: str,
        model_path: str = "yoloe-26x-seg.pt",
        conf: float = 0.35,
        iou: float = 0.45,
        voxel_size: float = 0.03,
        depth_scale: float = 256.0,
        depth_min: float = 0.02,
        ground_y: Optional[float] = None,  # 地面 Y 坐标（可选，默认从点云估计）
    ):
        """
        初始化场景处理器

        Args:
            scene_folder: 场景文件夹路径
            model_path: YOLOE 模型路径
            conf: 检测置信度阈值
            iou: NMS IoU 阈值
            voxel_size: 点云体素下采样尺寸
            depth_scale: 深度图缩放因子
            depth_min: 最小有效深度
            ground_y: 地面 Y 坐标（可选，默认从点云自动估计）
        """
        self.scene_folder = Path(scene_folder)
        self.conf = conf
        self.iou = iou
        self.voxel_size = voxel_size
        self.depth_scale = depth_scale
        self.depth_min = depth_min
        self.ground_y = ground_y

        # 子目录
        self.rgb_folder = self.scene_folder / "pano_img"
        self.depth_folder = self.scene_folder / "depth_img"
        self.pose_file = self.scene_folder / "vision.txt"

        # 输出目录
        self.output_folder = self.scene_folder / "output"
        self.detection_folder = self.output_folder / "detections"

        # 加载位姿
        self.poses = self._load_poses()

        # 初始化 YOLOE 模型
        print(f"加载 YOLOE 模型：{model_path}")
        self.model = YOLOE(model_path)
        self.model.set_classes(self.DOOR_CLASSES)
        print(f"检测类别 (统一为 door): {self.DOOR_CLASSES}")

    def _load_poses(self) -> Dict[str, PoseData]:
        """从 vision.txt 加载位姿信息"""
        if not self.pose_file.exists():
            raise FileNotFoundError(f"位姿文件不存在：{self.pose_file}")

        with open(self.pose_file, 'r') as f:
            data = json.load(f)

        poses = {}
        for loc in data.get('sweepLocations', []):
            uuid = str(loc['uuid'])
            poses[uuid] = PoseData(
                uuid=uuid,
                rotation=loc['pose']['rotation'],
                translation=loc['pose']['translation']
            )
        print(f"加载 {len(poses)} 个位姿")
        return poses

    def _build_pose_matrix(self, pose: PoseData) -> np.ndarray:
        """构建 4x4 位姿变换矩阵"""
        R = o3d.geometry.get_rotation_matrix_from_quaternion(
            np.array([pose.rotation['w'], pose.rotation['x'],
                      pose.rotation['y'], pose.rotation['z']])
        )
        t = np.array([
            pose.translation['x'],
            pose.translation['y'],
            pose.translation['z']
        ])

        T = np.eye(4)
        T[:3, :3] = R
        T[:3, 3] = t
        return T

    def _mask_to_3d_points(
        self,
        mask: np.ndarray,
        depth: np.ndarray,
        pose_matrix: np.ndarray
    ) -> Optional[np.ndarray]:
        """
        将 2D mask 映射到世界坐标系 3D 点

        Args:
            mask: 二值 mask (H, W)
            depth: 深度图 (H, W)
            pose_matrix: 4x4 位姿矩阵

        Returns:
            世界坐标系下的 3D 点 (N, 3)
        """
        H, W = depth.shape
        sph = Intrinsic_Spherical_NP(W, H)

        # 获取 mask 内的像素
        ys, xs = np.where(mask > 0)
        if len(xs) == 0:
            return None

        # 有效深度掩码
        valid = depth[ys, xs] > self.depth_min
        if not np.any(valid):
            return None

        xs, ys = xs[valid], ys[valid]
        depths = depth[ys, xs]

        # 计算方向向量
        bx, by, bz = sph.bearing([xs.astype(np.float64), ys.astype(np.float64)])
        bx, by, bz = np.array(bx), np.array(by), np.array(bz)

        # 相机坐标系
        pts_cam = np.stack([bx * depths, by * depths, bz * depths], axis=1)

        # Z 轴 180 度翻转
        R_z180 = np.diag([-1.0, -1.0, 1.0])
        pts_cam = pts_cam @ R_z180.T

        # 世界坐标系
        pts_w = (pose_matrix[:3, :3] @ pts_cam.T).T + pose_matrix[:3, 3]
        return pts_w

    def _extract_mask_contours(self, masks) -> Tuple[List[List[List[float]]], List[np.ndarray]]:
        """
        从 YOLOE mask 结果提取轮廓

        Args:
            masks: YOLOE masks (H, W, N)

        Returns:
            (轮廓列表, 对应 mask 数组) 每个 mask 只保留最大轮廓，保证与 scores 一一对应
        """
        contours = []
        mask_arrays = []
        if masks is None:
            return contours, mask_arrays

        masks_np = masks.cpu().numpy()
        for i in range(masks_np.shape[0]):
            mask = masks_np[i]
            # 二值化
            mask_bin = (mask > 0.5).astype(np.uint8) * 255
            # 提取轮廓
            cnts, _ = cv2.findContours(mask_bin, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
            if not cnts:
                continue
            # 只保留面积最大的轮廓，确保与 score 一一对应
            largest = max(cnts, key=cv2.contourArea)
            if len(largest) >= 3:
                # 简化轮廓
                epsilon = 0.02 * cv2.arcLength(largest, True)
                approx = cv2.approxPolyDP(largest, epsilon, True)
                contour = approx.reshape(-1, 2).astype(float).tolist()
                contours.append(contour)
                mask_arrays.append((mask_bin > 0).astype(np.uint8))
            else:
                # 点数不足时回退到完整 mask
                mask_arrays.append((mask_bin > 0).astype(np.uint8))
        return contours, mask_arrays

    def detect_single_image(
        self,
        img_path: str,
        depth: np.ndarray,
        pose_matrix: np.ndarray,
        save_path: Optional[str] = None
    ) -> MaskDetection:
        """
        检测单张图像并提取 mask 轮廓和 3D 点

        Args:
            img_path: 图像路径
            depth: 深度图
            pose_matrix: 位姿矩阵
            save_path: 保存路径

        Returns:
            MaskDetection 对象
        """
        results = self.model.predict(
            img_path,
            imgsz=(1024, 2048),
            conf=self.conf,
            iou=self.iou,
            max_det=50,
            augment=True,
            retina_masks=True,
            half=False,
            verbose=False,
        )

        result = results[0]
        scores = []
        contours = []
        mask_3d_points = []

        if result.masks is not None:
            masks = result.masks.data

            # 提取轮廓（每个 mask 只保留最大轮廓，与 scores 一一对应）
            contours, mask_arrays = self._extract_mask_contours(masks)

            # 获取分数
            scores = result.boxes.conf.cpu().numpy().tolist()

            # 每个 mask 转 3D 点
            H, W = depth.shape
            for mask_bin in mask_arrays:
                mask_resized = cv2.resize(mask_bin, (W, H), interpolation=cv2.INTER_NEAREST)
                pts_3d = self._mask_to_3d_points(mask_resized, depth, pose_matrix)
                if pts_3d is not None and len(pts_3d) > 0:
                    mask_3d_points.append(pts_3d)

        if save_path:
            os.makedirs(os.path.dirname(save_path), exist_ok=True)
            result.save(save_path)

        return MaskDetection(
            image_name=os.path.basename(img_path),
            mask_contours=contours,
            scores=scores,
            mask_3d_points=mask_3d_points
        )

    def _axis_aligned_bbox(self, points: np.ndarray) -> Tuple[np.ndarray, np.ndarray]:
        """计算轴对齐包围盒 (min, max)"""
        lo = np.min(points, axis=0)
        hi = np.max(points, axis=0)
        return lo, hi

    def _bbox_8corners(self, bbox_min: np.ndarray, bbox_max: np.ndarray) -> np.ndarray:
        """从 bbox min/max 获取 8 个角点"""
        cx, cy, cz = bbox_min
        ex, ey, ez = bbox_max
        return np.array([
            [cx, cy, cz], [ex, cy, cz], [ex, ey, cz], [cx, ey, cz],
            [cx, cy, ez], [ex, cy, ez], [ex, ey, ez], [cx, ey, ez],
        ])

    def _bbox_iou_3d(self, b1, b2) -> float:
        """3D IoU 计算"""
        lo = np.maximum(b1[0], b2[0])
        hi = np.minimum(b1[1], b2[1])
        inter = np.prod(np.maximum(hi - lo, 0))
        vol1 = np.prod(b1[1] - b1[0])
        vol2 = np.prod(b2[1] - b2[0])
        union = vol1 + vol2 - inter
        return inter / union if union > 0 else 0.0

    def _merge_3d_doors(self, door_candidates: List[Dict]) -> List[Door3D]:
        """
        合并重叠或接近的 3D 门 - 使用并查集确保完全合并

        合并逻辑：
        1. 构建连通图：如果两个门满足合并条件，则它们连通
        2. 使用并查集找出所有连通分量
        3. 每个连通分量合并为一个门

        Args:
            door_candidates: 候选门列表

        Returns:
            合并后的 Door3D 列表
        """
        if not door_candidates:
            return []

        n = len(door_candidates)
        if n == 1:
            # 只有一个候选，直接返回
            d = door_candidates[0]
            return [Door3D(
                id=0,
                center=(d['bbox_min'] + d['bbox_max']) / 2,
                bbox_8points=self._bbox_8corners(d['bbox_min'], d['bbox_max']),
                source_detections=[d['source']]
            )]

        # ========== 并查集 ==========
        parent = list(range(n))

        def find(x):
            if parent[x] != x:
                parent[x] = find(parent[x])  # 路径压缩
            return parent[x]

        def union(x, y):
            px, py = find(x), find(y)
            if px != py:
                parent[px] = py

        # ========== 构建连通关系 ==========
        # 检查所有门对，满足条件的合并
        for i in range(n):
            for j in range(i + 1, n):
                ci = (door_candidates[i]['bbox_min'] + door_candidates[i]['bbox_max']) / 2
                cj = (door_candidates[j]['bbox_min'] + door_candidates[j]['bbox_max']) / 2
                dist = np.linalg.norm(ci - cj)
                iou = self._bbox_iou_3d(
                    (door_candidates[i]['bbox_min'], door_candidates[i]['bbox_max']),
                    (door_candidates[j]['bbox_min'], door_candidates[j]['bbox_max'])
                )
                if dist < self.MERGE_DIST_THRESH and iou > self.MERGE_IOU_THRESH:
                    union(i, j)

        # ========== 按连通分量分组 ==========
        from collections import defaultdict
        groups = defaultdict(list)
        for i in range(n):
            groups[find(i)].append(door_candidates[i])

        # ========== 合并每个组 ==========
        doors = []
        for door_id, members in enumerate(groups.values()):
            if not members:
                continue

            # 合并所有成员的 bbox
            bbox_min = np.min([m['bbox_min'] for m in members], axis=0)
            bbox_max = np.max([m['bbox_max'] for m in members], axis=0)
            sources = [m['source'] for m in members]

            doors.append(Door3D(
                id=door_id,
                center=(bbox_min + bbox_max) / 2,
                bbox_8points=self._bbox_8corners(bbox_min, bbox_max),
                source_detections=sources
            ))

        return doors

    def _estimate_ground_y(self, combined_pc: o3d.geometry.PointCloud) -> float:
        """
        从点云估计地面 Y 坐标

        假设：地面是场景中最低的大面积平面
        方法：取所有点中 Y 坐标的下 1% 分位数

        Args:
            combined_pc: 完整场景点云

        Returns:
            估计的地面 Y 坐标
        """
        points = np.asarray(combined_pc.points)
        if len(points) == 0:
            return 0.0

        # 取 Y 坐标（假设 Y 轴向上）的下 1% 分位数，避免家具腿等低矮物体干扰
        ground_y = np.percentile(points[:, 1], 1)
        return ground_y

    def _filter_door_by_properties(self, door: Door3D, ground_y: float) -> Tuple[bool, List[str]]:
        """
        根据物理特性过滤门

        过滤条件：
        1. 高度在合理范围内 (1.0m - 3.0m)
        2. 宽度在合理范围内 (0.3m - 3.0m)
        3. 厚度不超过阈值 (≤ 0.5m)
        4. 门底部接近地面 (≤ 0.5m)

        Args:
            door: 3D 门对象
            ground_y: 地面 Y 坐标

        Returns:
            (是否通过过滤，拒绝原因列表)
        """
        reasons = []

        # 计算 bounding box 尺寸
        size = door.bbox_8points.max(axis=0) - door.bbox_8points.min(axis=0)
        height = size[1]  # Y 方向是高度
        width = max(size[0], size[2])  # 宽度是 X/Z 中较大的
        thickness = min(size[0], size[2])  # 厚度是 X/Z 中较小的

        # 门底部 Y 坐标
        door_bottom_y = door.bbox_8points[:, 1].min()

        # 条件 1: 高度检查
        if height < self.DOOR_HEIGHT_MIN:
            reasons.append(f"高度过小 ({height:.2f}m < {self.DOOR_HEIGHT_MIN}m)")
        elif height > self.DOOR_HEIGHT_MAX:
            reasons.append(f"高度过大 ({height:.2f}m > {self.DOOR_HEIGHT_MAX}m)")

        # 条件 2: 宽度检查
        if width < self.DOOR_WIDTH_MIN:
            reasons.append(f"宽度过小 ({width:.2f}m < {self.DOOR_WIDTH_MIN}m)")
        elif width > self.DOOR_WIDTH_MAX:
            reasons.append(f"宽度过大 ({width:.2f}m > {self.DOOR_WIDTH_MAX}m)")

        # 条件 3: 厚度检查
        if thickness > self.DOOR_THICKNESS_MAX:
            reasons.append(f"厚度过大 ({thickness:.2f}m > {self.DOOR_THICKNESS_MAX}m)")

        # 条件 4: 地面贴合检查
        dist_to_ground = door_bottom_y - ground_y
        if dist_to_ground > self.GROUND_DIST_THRESH:
            reasons.append(f"距地面过远 ({dist_to_ground:.2f}m > {self.GROUND_DIST_THRESH}m)")
        elif dist_to_ground < -self.GROUND_DIST_THRESH:
            reasons.append(f"嵌入地面过深 ({abs(dist_to_ground):.2f}m)")

        passed = len(reasons) == 0
        return passed, reasons

    def _score_entrance_door(self, door: Door3D, ground_y: float, all_centers: np.ndarray) -> Dict:
        """
        为每个门计算"入户门可能性"评分

        评分维度（满分 100）：
        1. 尺寸评分 (30分): 入户门通常较大，高度约 2.0-2.4m，宽度约 0.9-1.2m
        2. 地面贴合 (20分): 入户门底部贴近地面
        3. 边缘位置 (25分): 入户门在建筑外围，离场景中心较远
        4. 厚度评分 (15分): 入户门通常较厚（实心门）
        5. 多视角支持 (10分): 被多个视角检测到的门更可信

        Args:
            door: 3D 门对象
            ground_y: 地面 Y 坐标
            all_centers: 所有门的中心坐标

        Returns:
            评分详情 {"total": float, "details": Dict[str, float], "reason": str}
        """
        size = door.bbox_8points.max(axis=0) - door.bbox_8points.min(axis=0)
        height = size[1]
        width = max(size[0], size[2])
        thickness = min(size[0], size[2])
        door_bottom_y = door.bbox_8points[:, 1].min()

        # 1. 尺寸评分 (30分) - 理想入户门尺寸 高2.1m 宽1.0m
        height_score = max(0, 15 - abs(height - 2.1) * 10)
        width_score = max(0, 15 - abs(width - 1.0) * 12)
        size_score = height_score + width_score

        # 2. 地面贴合 (20分) - 底部越接近地面分越高
        dist_to_ground = abs(door_bottom_y - ground_y)
        ground_score = max(0, 20 - dist_to_ground * 40)

        # 3. 边缘位置 (25分) - 离所有门的中心越远，越可能是入户门
        if len(all_centers) > 1:
            dists_to_others = np.linalg.norm(all_centers - door.center, axis=1)
            max_dist = dists_to_others.max()
            avg_dist = dists_to_others.mean()
            # 平均距离越大越可能是入户门
            edge_score = min(25, avg_dist * 10)
        else:
            edge_score = 15  # 只有一个门时给中等分

        # 4. 厚度评分 (15分) - 入户门通常更厚
        # 理想厚度 0.04-0.1m，但点云中由于包含门框，会显得更厚
        if 0.05 <= thickness <= 0.3:
            thickness_score = 15
        elif thickness < 0.05:
            thickness_score = 5
        else:
            thickness_score = max(0, 15 - (thickness - 0.3) * 30)

        # 5. 多视角支持 (10分)
        source_count = len(door.source_detections)
        view_score = min(10, source_count * 3)

        total = size_score + ground_score + edge_score + thickness_score + view_score

        reasons = [
            f"尺寸={size_score:.0f}/30 (高{height:.2f}m 宽{width:.2f}m)",
            f"地面={ground_score:.0f}/20 (距地{dist_to_ground:.2f}m)",
            f"边缘={edge_score:.0f}/25",
            f"厚度={thickness_score:.0f}/15 ({thickness:.2f}m)",
            f"视角={view_score:.0f}/10 ({source_count}个)",
        ]

        return {
            "total": round(total, 1),
            "details": {
                "size": round(size_score, 1),
                "ground": round(ground_score, 1),
                "edge": round(edge_score, 1),
                "thickness": round(thickness_score, 1),
                "view": round(view_score, 1),
            },
            "reason": " | ".join(reasons),
        }

    def _identify_entrance_door(
        self,
        valid_doors: List[Door3D],
        ground_y: float,
        all_detections: List[MaskDetection],
        combined_pc: o3d.geometry.PointCloud,
    ) -> EntranceInfo:
        """
        从多个门中识别入户门，无检测时提供兜底策略

        策略优先级：
        1. 如果 YOLOE 检测到 "exterior door" 类别 → 直接使用
        2. 有多个有效门 → 按评分选择最高分
        3. 无有效门但有 2D 检测 → 用置信度最高的 2D 检测的 3D 投影位置
        4. 完全无检测 → 用场景几何估计（场景边界中心+地面高度）

        Args:
            valid_doors: 通过物理特性过滤的门
            ground_y: 地面 Y 坐标
            all_detections: 所有 2D 检测结果
            combined_pc: 合并后的场景点云

        Returns:
            EntranceInfo 入户门信息
        """
        # ===== 策略 1: 检查是否有 exterior door 检测结果 =====
        for det in all_detections:
            for src in det.source_detections if hasattr(det, 'source_detections') else []:
                pass  # source_detections 在 Door3D 上，不在 MaskDetection 上
            # MaskDetection 只有 scores，没有类别信息
            # YOLOE predict 返回的 result.boxes.cls 包含类别 ID
            pass

        # ===== 策略 2: 有多个有效门，按评分选择 =====
        if len(valid_doors) >= 1:
            all_centers = np.array([d.center for d in valid_doors])
            scored_doors = []
            for door in valid_doors:
                score_info = self._score_entrance_door(door, ground_y, all_centers)
                scored_doors.append((door, score_info))

            # 按总分排序
            scored_doors.sort(key=lambda x: x[1]["total"], reverse=True)
            best_door, best_score = scored_doors[0]

            method = "size_score"
            reason = f"多门评分选择（共{len(valid_doors)}个门）: {best_score['reason']}"

            if len(valid_doors) == 1:
                method = "size_score"
                reason = f"唯一有效门: {best_score['reason']}"

            print(f"\n入户门识别 - 选择门{best_door.id}")
            print(f"  方法: {method}")
            print(f"  评分: {best_score['total']}/100")
            print(f"  中心: {best_door.center.round(3)}")
            print(f"  原因: {reason}")

            return EntranceInfo(
                is_detected=True,
                door=best_door,
                center=best_door.center.copy(),
                score=best_score["total"] / 100.0,
                method=method,
                reason=reason,
            )

        # ===== 策略 3: 无有效门但有 2D 检测 =====
        # 找置信度最高的 2D 检测
        best_2d_score = 0.0
        best_2d_det = None
        for det in all_detections:
            if det.scores and len(det.mask_3d_points) > 0:
                max_score = max(det.scores)
                if max_score > best_2d_score:
                    best_2d_score = max_score
                    best_2d_det = det

        if best_2d_det is not None and len(best_2d_det.mask_3d_points) > 0:
            # 使用最高分 mask 的 3D 点中心作为入户门估计位置
            best_mask_idx = best_2d_det.scores.index(best_2d_score)
            if best_mask_idx < len(best_2d_det.mask_3d_points):
                est_center = np.mean(best_2d_det.mask_3d_points[best_mask_idx], axis=0)
                print(f"\n入户门识别 - 使用最高分 2D 检测兜底")
                print(f"  方法: fallback_2d")
                print(f"  图像: {best_2d_det.image_name}")
                print(f"  置信度: {best_2d_score:.3f}")
                print(f"  估计中心: {est_center.round(3)}")

                return EntranceInfo(
                    is_detected=False,
                    door=None,
                    center=est_center,
                    score=best_2d_score,
                    method="fallback_2d",
                    reason=f"最高分 2D 检测 ({best_2d_det.image_name}, conf={best_2d_score:.3f})",
                )

        # ===== 策略 4: 完全无检测，用场景几何估计 =====
        print(f"\n入户门识别 - 无任何检测，使用场景几何估计")
        points = np.asarray(combined_pc.points)
        if len(points) > 0:
            # 取点云的水平中心 + 最低点作为入口估计
            est_x = np.median(points[:, 0])
            est_y = ground_y + 1.0  # 地面以上 1m（典型门把手高度附近）
            est_z = np.median(points[:, 2])
            # 找离中心最远的方向（可能是入口方向）
            center_2d = np.array([est_x, est_z])
            dists = np.linalg.norm(points[:, [0, 2]] - center_2d, axis=1)
            far_idx = np.argsort(dists)[-len(dists)//10:]  # 最远 10% 的点
            far_points = points[far_idx]
            est_x = np.median(far_points[:, 0])
            est_z = np.median(far_points[:, 2])
        else:
            est_x, est_y, est_z = 0.0, ground_y + 1.0, 0.0

        est_center = np.array([est_x, est_y, est_z])
        print(f"  方法: scene_center_estimate")
        print(f"  估计中心: {est_center.round(3)}")

        return EntranceInfo(
            is_detected=False,
            door=None,
            center=est_center,
            score=0.0,
            method="scene_center_estimate",
            reason="无检测结果，场景几何估计",
        )

    def _rgb_depth_to_pointcloud(
        self,
        rgb: np.ndarray,
        depth: np.ndarray,
        pose_matrix: np.ndarray
    ) -> o3d.geometry.PointCloud:
        """将 RGB-D 转换为世界坐标系点云"""
        H, W = depth.shape
        sph = Intrinsic_Spherical_NP(W, H)

        px, py = np.meshgrid(np.arange(W), np.arange(H))
        px_flat = px.flatten().astype(np.float64)
        py_flat = py.flatten().astype(np.float64)

        bx, by, bz = sph.bearing([px_flat, py_flat])
        bx, by, bz = np.array(bx), np.array(by), np.array(bz)

        mask = depth.flatten() > self.depth_min
        d = depth.flatten()[mask]

        if len(d) == 0:
            return o3d.geometry.PointCloud()

        pts_cam = np.stack([bx[mask] * d, by[mask] * d, bz[mask] * d], axis=1)
        R_z180 = np.diag([-1.0, -1.0, 1.0])
        pts_cam = pts_cam @ R_z180.T
        pts_w = (pose_matrix[:3, :3] @ pts_cam.T).T + pose_matrix[:3, 3]

        if rgb.shape[:2] != depth.shape:
            rgb_d = cv2.resize(rgb, (W, H), interpolation=cv2.INTER_LINEAR)
        else:
            rgb_d = rgb
        colors = rgb_d.reshape(-1, 3)[mask].astype(np.float64) / 255.0

        pc = o3d.geometry.PointCloud()
        pc.points = o3d.utility.Vector3dVector(pts_w)
        pc.colors = o3d.utility.Vector3dVector(colors)
        return pc

    def process_scene(self):
        """处理整个场景"""
        # 创建输出目录
        self.output_folder.mkdir(parents=True, exist_ok=True)
        self.detection_folder.mkdir(parents=True, exist_ok=True)

        rgb_files = sorted(
            self.rgb_folder.glob("*.jpg"),
            key=lambda x: int(x.stem)
        )

        if not rgb_files:
            raise FileNotFoundError(f"在 {self.rgb_folder} 中未找到 RGB 图像")

        print(f"找到 {len(rgb_files)} 张全景图，开始处理...")

        combined_pc = o3d.geometry.PointCloud()
        all_detections: List[MaskDetection] = []
        door_candidates = []  # 3D 门候选

        for rgb_file in tqdm(rgb_files, desc="处理场景"):
            idx = rgb_file.stem
            pose = self.poses.get(idx)

            if pose is None:
                print(f"  ⚠️ 警告：{rgb_file.name} 无位姿信息，跳过")
                continue

            depth_path = self.depth_folder / f"{idx}.png"
            if not depth_path.exists():
                print(f"  ⚠️ 警告：深度图不存在 {depth_path}，跳过")
                continue

            rgb = cv2.cvtColor(cv2.imread(str(rgb_file)), cv2.COLOR_BGR2RGB)
            depth = cv2.imread(str(depth_path), cv2.IMREAD_UNCHANGED).astype(np.float32) / self.depth_scale
            pose_matrix = self._build_pose_matrix(pose)

            # 1. YOLOE 检测 + mask 3D 映射
            save_det_path = self.detection_folder / f"{idx}_det.jpg"
            det_result = self.detect_single_image(
                str(rgb_file), depth, pose_matrix, str(save_det_path)
            )
            all_detections.append(det_result)

            det_count = len(det_result.mask_contours)
            tqdm.write(f"  {rgb_file.name}: 检测 {det_count} 个门")

            # 2. 收集 3D 门候选
            for i, pts_3d in enumerate(det_result.mask_3d_points):
                if len(pts_3d) > 10:  # 至少 10 个点
                    bbox_min, bbox_max = self._axis_aligned_bbox(pts_3d)
                    door_candidates.append({
                        'bbox_min': bbox_min,
                        'bbox_max': bbox_max,
                        'points_3d': pts_3d,
                        'source': {
                            'image': rgb_file.name,
                            'score': det_result.scores[i] if i < len(det_result.scores) else 0.0
                        }
                    })

            # 3. RGB-D 转完整点云
            pc = self._rgb_depth_to_pointcloud(rgb, depth, pose_matrix)
            combined_pc += pc

        # 最终下采样
        print(f"\n融合前点数：{len(combined_pc.points)}")
        combined_pc = combined_pc.voxel_down_sample(self.voxel_size)
        print(f"融合后点数：{len(combined_pc.points)}")

        # 保存合并点云
        merge_ply_path = self.output_folder / "merged.ply"
        o3d.io.write_point_cloud(str(merge_ply_path), combined_pc)
        print(f"保存合并点云：{merge_ply_path}")

        # 合并 3D 门
        print(f"\n3D 门候选：{len(door_candidates)}")
        doors_3d = self._merge_3d_doors(door_candidates)
        print(f"合并后门数量：{len(doors_3d)}")

        # 估计地面 Y 坐标
        if self.ground_y is not None:
            ground_y = self.ground_y
        else:
            ground_y = self._estimate_ground_y(combined_pc)
        print(f"估计地面 Y 坐标：{ground_y:.3f}")

        # 过滤不符合物理特性的门
        print("\n过滤 3D 门...")
        valid_doors = []
        filtered_doors = []
        for door in doors_3d:
            passed, reasons = self._filter_door_by_properties(door, ground_y)
            if passed:
                valid_doors.append(door)
            else:
                filtered_doors.append((door, reasons))

        print(f"通过过滤：{len(valid_doors)} 个门")
        if len(filtered_doors) > 0:
            print(f"被过滤：{len(filtered_doors)} 个门")
            for door, reasons in filtered_doors:
                print(f"  门{door.id} (中心={door.center.round(2)}):")
                for reason in reasons:
                    print(f"    - {reason}")

        # 识别入户门
        print("\n" + "=" * 40)
        print("入户门识别")
        print("=" * 40)
        entrance_info = self._identify_entrance_door(
            valid_doors, ground_y, all_detections, combined_pc
        )

        # 保存检测结果（包含入户门信息）
        self._save_detections(all_detections, valid_doors, filtered_doors, entrance_info)

        return combined_pc, valid_doors, entrance_info

    def _save_detections(self, detections: List[MaskDetection], doors_3d: List[Door3D],
                         filtered_doors: List[Tuple[Door3D, List[str]]] = None,
                         entrance_info: Optional[EntranceInfo] = None):
        """保存检测结果"""
        # 只保存有检测到的图像
        detected_results = []
        for d in detections:
            if len(d.mask_contours) > 0:
                detected_results.append({
                    "image": d.image_name,
                    "count": len(d.mask_contours),
                    "mask_contours": d.mask_contours,
                    "scores": d.scores
                })

        # 3D 门信息 - 正确处理多来源合并
        doors_3d_data = []
        for door in doors_3d:
            # 统计来源信息
            source_count = len(door.source_detections)
            scores = [s['score'] for s in door.source_detections if 'score' in s]
            avg_score = sum(scores) / len(scores) if scores else 0.0

            # 计算门的尺寸
            size = door.bbox_8points.max(axis=0) - door.bbox_8points.min(axis=0)

            doors_3d_data.append({
                "id": door.id,
                "center": door.center.tolist(),
                "bbox_8points": door.bbox_8points.tolist(),
                "size": np.round(size, 4).tolist(),  # [宽，高，厚]
                "source_count": source_count,  # 来源数量
                "avg_score": round(avg_score, 4),  # 平均置信度
                "sources": door.source_detections  # 详细来源列表
            })

        # 被过滤的门信息
        filtered_data = []
        if filtered_doors and len(filtered_doors) > 0:
            for door, reasons in filtered_doors:
                source_count = len(door.source_detections)
                scores = [s['score'] for s in door.source_detections if 'score' in s]
                avg_score = sum(scores) / len(scores) if scores else 0.0
                size = door.bbox_8points.max(axis=0) - door.bbox_8points.min(axis=0)

                filtered_data.append({
                    "id": door.id,
                    "center": door.center.tolist(),
                    "size": np.round(size, 4).tolist(),
                    "avg_score": round(avg_score, 4),
                    "source_count": source_count,
                    "reject_reasons": reasons
                })

        # 入户门信息
        entrance_data = None
        if entrance_info:
            entrance_data = {
                "is_detected": entrance_info.is_detected,
                "center": entrance_info.center.tolist(),
                "score": round(entrance_info.score, 4),
                "method": entrance_info.method,
                "reason": entrance_info.reason,
            }
            if entrance_info.door is not None:
                entrance_data["door_id"] = entrance_info.door.id

        output = {
            "total_images_processed": len(detections),
            "images_with_doors": len(detected_results),
            "total_2d_detections": sum(r["count"] for r in detected_results),
            "total_3d_doors": len(doors_3d),
            "filtered_3d_doors": len(filtered_doors) if filtered_doors else 0,
            "entrance_door": entrance_data,
            "detected_images": detected_results,
            "3d_doors": doors_3d_data,
            "filtered_doors": filtered_data
        }

        json_path = self.output_folder / "detections.json"
        with open(json_path, 'w', encoding='utf-8') as f:
            json.dump(output, f, indent=2, ensure_ascii=False)
        print(f"保存检测结果：{json_path}")

        # 打印汇总
        print("\n" + "="*50)
        print("检测汇总")
        print("="*50)
        print(f"  处理图像数：{output['total_images_processed']}")
        print(f"  检测到门的图像：{output['images_with_doors']}")
        print(f"  2D 检测总数：{output['total_2d_detections']}")
        print(f"  3D 门数量：{output['total_3d_doors']}")
        if len(filtered_doors) > 0 if filtered_doors else False:
            print(f"  被过滤的门：{output['filtered_3d_doors']}")

        if entrance_data:
            det_tag = "已检测" if entrance_data["is_detected"] else "估计"
            print(f"\n  入户门 [{det_tag}]:")
            print(f"    方法: {entrance_data['method']}")
            print(f"    中心: {entrance_data['center']}")
            print(f"    评分: {entrance_data['score']}")
            print(f"    原因: {entrance_data['reason']}")

        if doors_3d:
            print("\n  有效 3D 门信息:")
            for door in doors_3d:
                size = door.bbox_8points.max(axis=0) - door.bbox_8points.min(axis=0)
                sources_info = f"{len(door.source_detections)} 个视角"
                if len(door.source_detections) > 1:
                    scores_str = ", ".join([f"{s['image']}:{s['score']:.2f}" for s in door.source_detections])
                    sources_info += f" ({scores_str})"
                print(f"    门{door.id}: 中心={door.center.round(3)}, 尺寸={size.round(3)}")
                print(f"           来源：{sources_info}")

        if filtered_doors and len(filtered_doors) > 0:
            print("\n  被过滤的门 (不符合物理特性):")
            for door, reasons in filtered_doors:
                size = door.bbox_8points.max(axis=0) - door.bbox_8points.min(axis=0)
                print(f"    门{door.id}: 中心={door.center.round(3)}, 尺寸={size.round(3)}")
                for reason in reasons:
                    print(f"      - {reason}")
        print("="*50)


# ============================================================================
# 主函数
# ============================================================================
def main():
    parser = argparse.ArgumentParser(
        description="场景点云处理器 - YOLOE 检测 + 点云融合 + 3D 门合并",
        formatter_class=argparse.RawDescriptionHelpFormatter,
        epilog="""
示例:
  # 处理 scene0001 场景
  python scene_processor.py -s scene0001

  # 指定模型和参数
  python scene_processor.py -s scene0001 --model yoloe-26x-seg.pt --conf 0.4

  # 调整点云精度
  python scene_processor.py -s scene0001 --voxel-size 0.02

  # 指定地面 Y 坐标（默认自动估计）
  python scene_processor.py -s scene0001 --ground-y -1.5

  # 调整门尺寸过滤阈值
  python scene_processor.py -s scene0001 --door-height-min 1.2 --door-width-min 0.5
        """
    )
    parser.add_argument(
        "--scene", "-s",
        type=str,
        default="scene0001",
        help="场景文件夹 (默认：scene0001)"
    )
    parser.add_argument(
        "--model", "-m",
        type=str,
        default="yoloe-26x-seg.pt",
        help="YOLOE 模型路径"
    )
    parser.add_argument(
        "--conf",
        type=float,
        default=0.35,
        help="置信度阈值 (默认：0.35)"
    )
    parser.add_argument(
        "--iou",
        type=float,
        default=0.45,
        help="NMS IoU 阈值 (默认：0.45)"
    )
    parser.add_argument(
        "--voxel-size",
        type=float,
        default=0.03,
        help="点云体素大小 (默认：0.03)"
    )
    parser.add_argument(
        "--depth-scale",
        type=float,
        default=256.0,
        help="深度图缩放因子 (默认：256.0)"
    )
    parser.add_argument(
        "--ground-y",
        type=float,
        default=None,
        help="地面 Y 坐标 (默认：从点云自动估计)"
    )
    parser.add_argument(
        "--door-height-min",
        type=float,
        default=1.0,
        help="门最小高度 (默认：1.0 米)"
    )
    parser.add_argument(
        "--door-height-max",
        type=float,
        default=3.0,
        help="门最大高度 (默认：3.0 米)"
    )
    parser.add_argument(
        "--door-width-min",
        type=float,
        default=0.3,
        help="门最小宽度 (默认：0.3 米)"
    )
    parser.add_argument(
        "--door-width-max",
        type=float,
        default=3.0,
        help="门最大宽度 (默认：3.0 米)"
    )
    parser.add_argument(
        "--door-thickness-max",
        type=float,
        default=0.5,
        help="门最大厚度 (默认：0.5 米)"
    )
    parser.add_argument(
        "--ground-dist-thresh",
        type=float,
        default=0.5,
        help="门底部距地面最大距离 (默认：0.5 米)"
    )

    args = parser.parse_args()

    if not Path(args.scene).exists():
        print(f"❌ 场景文件夹不存在：{args.scene}")
        return

    processor = SceneProcessor(
        scene_folder=args.scene,
        model_path=args.model,
        conf=args.conf,
        iou=args.iou,
        voxel_size=args.voxel_size,
        depth_scale=args.depth_scale,
        ground_y=args.ground_y,
    )

    # 更新过滤参数
    processor.DOOR_HEIGHT_MIN = args.door_height_min
    processor.DOOR_HEIGHT_MAX = args.door_height_max
    processor.DOOR_WIDTH_MIN = args.door_width_min
    processor.DOOR_WIDTH_MAX = args.door_width_max
    processor.DOOR_THICKNESS_MAX = args.door_thickness_max
    processor.GROUND_DIST_THRESH = args.ground_dist_thresh

    processor.process_scene()


if __name__ == "__main__":
    main()