openMVS/libs/MVS/utils/get_world_point.py

'''
获取Mesh的深度图

required:
    - open3d
    - numpy
    - numba
'''

import open3d as o3d
import numpy as np
from numba import njit

@njit
def _process_all_points(Z, width, height, cx, cy, fx, fy, c2w):
    """JIT优化的所有点处理函数"""
    # 手动创建像素网格（替代np.meshgrid）
    u = np.zeros((height, width), dtype=np.float64)
    v = np.zeros((height, width), dtype=np.float64)
    
    for y in range(height):
        for x in range(width):
            u[y, x] = x
            v[y, x] = y
    
    # 创建输出数组
    Z_all = np.zeros_like(Z)
    
    # 手动处理有效点
    for y in range(height):
        for x in range(width):
            if Z[y, x] > 0 and Z[y, x] < 1e9 and not np.isnan(Z[y, x]):
                Z_all[y, x] = Z[y, x]
            else:
                Z_all[y, x] = -10000
    
    # 计算所有点的3D坐标
    X_all = (u - cx) * Z_all / fx
    Y_all = (v - cy) * Z_all / fy
    
    # 创建齐次坐标
    points_3d = np.zeros((height * width, 4))
    points_3d[:, 0] = X_all.reshape(-1)
    points_3d[:, 1] = Y_all.reshape(-1)
    points_3d[:, 2] = Z_all.reshape(-1)
    points_3d[:, 3] = 1.0
    
    # 坐标变换
    return points_3d @ c2w.T

@njit
def _process_valid_points(Z, width, height, cx, cy, fx, fy, c2w):
    """JIT优化的有效点处理函数"""
    # 首先计算有效点的数量
    valid_count = 0
    for y in range(height):
        for x in range(width):
            if Z[y, x] > 0 and Z[y, x] < 1e9 and not np.isnan(Z[y, x]):
                valid_count += 1
    
    # 创建输出数组
    X_valid = np.zeros(valid_count)
    Y_valid = np.zeros(valid_count)
    Z_valid = np.zeros(valid_count)
    
    # 填充有效点数据
    idx = 0
    for y in range(height):
        for x in range(width):
            if Z[y, x] > 0 and Z[y, x] < 1e9 and not np.isnan(Z[y, x]):
                Z_valid[idx] = Z[y, x]
                X_valid[idx] = (x - cx) * Z_valid[idx] / fx
                Y_valid[idx] = (y - cy) * Z_valid[idx] / fy
                idx += 1
    
    # 创建齐次坐标
    points_3d = np.zeros((valid_count, 4))
    points_3d[:, 0] = X_valid
    points_3d[:, 1] = Y_valid
    points_3d[:, 2] = Z_valid
    points_3d[:, 3] = 1.0
    
    # 坐标变换
    return points_3d @ c2w.T

def get_world_points(depth_map: o3d.geometry.Image, K: np.ndarray, c2w: np.ndarray, all_point: bool = False) -> np.ndarray:
    '''
    获取深度图对应的世界坐标点云

    param:
        depth_map: 深度图
        K: 相机内参矩阵
        c2w: 相机坐标系到世界坐标系的变换矩阵
        all_point: 是否返回所有点（包括无效点）
    
    return:
        np.ndarray: 点云坐标（世界坐标系）
            - 当all_point=False时：只包含有效深度值的点云
            - 当all_point=True时：包含所有点的点云，无效点深度值设为10000
    '''
    # 转换为numpy数组
    Z = np.asarray(depth_map)
    height, width = Z.shape
    
    # 获取内参
    fx, fy = K[0, 0], K[1, 1]
    cx, cy = K[0, 2], K[1, 2]
    
    # 确保数据类型兼容性（Numba要求）
    Z = Z.astype(np.float64)
    c2w = c2w.astype(np.float64)
    
    if all_point:
        return _process_all_points(Z, width, height, cx, cy, fx, fy, c2w)
    else:
        return _process_valid_points(Z, width, height, cx, cy, fx, fy, c2w)