hongchen/door_detect/multi_points_pano2world.py

import os
import json
from tqdm import tqdm
import cv2
import numpy as np
import open3d as o3d
import torch
from camera_spherical import Intrinsic_Spherical
from tqdm import tqdm

import open3d as o3d
import numpy as np
import glob
import os
import pathlib


def merge_ply_files(directory_path, output_filename):
    directory_path = pathlib.Path(directory_path)
    ply_files = list(directory_path.rglob('*.ply'))

    if not ply_files:
        print(f"⚠️ 警告：在目录 '{directory_path}' 中没有找到任何 .ply 文件。")
        return None

    # 初始化一个空的点云对象用于存储合并后的数据
    combined_cloud = o3d.geometry.PointCloud()

    print(f"--- 找到 {len(ply_files)} 个 .ply 文件，开始合并... ---")

    for i, file_path in enumerate(ply_files):
        print(f"✅ 正在读取文件 {i + 1}/{len(ply_files)}: {os.path.basename(file_path)}")
        try:
            # 1. 读取点云文件
            current_cloud = o3d.io.read_point_cloud(file_path)
            # current_cloud = current_cloud.voxel_down_sample(voxel_size=0.03)

            # 2. 合并点云数据 (使用 '+' 运算符实现简单的连接)
            combined_cloud += current_cloud

        except Exception as e:
            print(f"❌ 读取文件 {file_path} 时发生错误: {e}")
            continue

    # 3. 可选：保存合并后的点云文件
    if len(combined_cloud.points) > 0:
        o3d.io.write_point_cloud(output_filename, combined_cloud)
        print(f"\n🎉 合并完成！总共包含 {len(combined_cloud.points)} 个点。")
        print(f"文件已保存为: {output_filename}")

        # 4. 可选：显示合并后的点云
        # o3d.visualization.draw_geometries([combined_cloud])

        return combined_cloud
    else:
        print("\n⚠️ 警告：合并后的点云为空，没有数据可保存。")
        return None


def rgbd_vis(image0, depthmap0, save_ply_path, pose, vis=False):
    mask_flatten = (depthmap0 > 0.02).flatten()

    depthmap0 = torch.from_numpy(depthmap0)

    pixel_x, pixel_y = np.meshgrid(range(image0.shape[1]), range(image0.shape[0]))

    sph_cam = Intrinsic_Spherical(image0.shape[1], image0.shape[0])

    bearing = sph_cam.bearing([pixel_x, pixel_y])

    point_cloud_0 = [depthmap0 * bearing[0], depthmap0 * bearing[1], depthmap0 * bearing[2]]

    point_cloud_0 = np.transpose(
        [point_cloud_0[0].flatten().numpy(), point_cloud_0[1].flatten().numpy(), point_cloud_0[2].flatten().numpy()])

    point_cloud_0 = np.transpose(point_cloud_0)

    x0 = point_cloud_0[0].flatten()
    y0 = point_cloud_0[1].flatten()
    z0 = point_cloud_0[2].flatten()

    r0 = image0[pixel_y, pixel_x, 0].flatten()
    g0 = image0[pixel_y, pixel_x, 1].flatten()
    b0 = image0[pixel_y, pixel_x, 2].flatten()

    point_cloud_data0 = np.stack([x0, y0, z0, r0, g0, b0], axis=1)

    # open3d显示保存
    points = np.stack([x0, y0, z0], axis=1)
    color = np.stack([r0, g0, b0], axis=1) / 255.0

    points = points[mask_flatten]
    color = color[mask_flatten]

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

    if pose:
        # Apply 180-degree rotation around Z-axis locally first
        R_z_180 = np.array([[-1, 0, 0],
                            [0, -1, 0],
                            [0, 0, 1]])
        T_z_180 = np.eye(4)
        T_z_180[:3, :3] = R_z_180
        pc.transform(T_z_180)

        # Then apply the pose transformation from the file
        rotation_q = pose['rotation']
        translation_v = pose['translation']
        quat = np.array([rotation_q['w'], rotation_q['x'], rotation_q['y'], rotation_q['z']])
        rotation_m = o3d.geometry.get_rotation_matrix_from_quaternion(quat)
        trans = np.eye(4)
        trans[:3, :3] = rotation_m
        trans[:3, 3] = [translation_v['x'], translation_v['y'], translation_v['z']]
        pc.transform(trans)

    pc = pc.voxel_down_sample(voxel_size=0.03)

    o3d.io.write_point_cloud(save_ply_path, pc, write_ascii=True)
    if vis:
        o3d.visualization.draw_geometries([pc])


if __name__ == '__main__':
    pano_folder = "pano"
    rgb_folder = os.path.join(pano_folder, 'rgb')
    depth_folder = os.path.join(pano_folder, 'depth')
    ply_folder = os.path.join(pano_folder, 'ply')
    json_path = os.path.join(pano_folder, 'inf.json')
    merge_ply_name = os.path.join(pano_folder, "merge.ply")

    with open(json_path, 'r') as f:
        pose_data = json.load(f)

    pose_lookup = {item['uuid']: item['pose'] for item in pose_data['sweepLocations']}

    os.makedirs(ply_folder, exist_ok=True)
    rgb_names = os.listdir(rgb_folder)
    for rgb_name in tqdm(rgb_names):
        flag = rgb_name.split('.')[0]
        pose = pose_lookup.get(flag)
        if not pose:
            print(f"Warning: Pose not found for {rgb_name}. Skipping transformation.")

        depth_name = flag + ".png"
        rgb_name_path = os.path.join(rgb_folder, rgb_name)
        save_ply_path = os.path.join(ply_folder, f"{flag}.ply")
        depth_name_path = os.path.join(depth_folder, depth_name)

        image0 = cv2.imread(rgb_name_path)
        image0 = cv2.resize(image0, (2048, 1024))
        image0 = cv2.cvtColor(image0, cv2.COLOR_BGR2RGB)
        depthmap0 = (cv2.imread(depth_name_path, cv2.IMREAD_UNCHANGED)) / 256.
        rgbd_vis(image0, depthmap0, save_ply_path, pose, vis=False)

    merge_ply_files(ply_folder, merge_ply_name)