如何基于Open3D移除点云pcd1中位于pcd2凸包内的点

C++ 实现方案 #

步骤说明 #

1. 加载目标点云pcd1和pcd2
2. 计算pcd2的凸包结构
3. 遍历pcd1的每个点，判断是否处于pcd2的凸包内部
4. 保留凸包外部的点，生成过滤后的点云

代码示例 #

#include <open3d/Open3D.h>

int main() {
    // 加载点云文件
    auto pcd1 = open3d::io::CreatePointCloudFromFile("pcd1.pcd");
    auto pcd2 = open3d::io::CreatePointCloudFromFile("pcd2.pcd");

    // 计算pcd2的凸包，返回凸包网格和顶点索引
    open3d::geometry::TriangleMesh convex_hull_mesh;
    std::vector<size_t> hull_vertex_indices;
    std::tie(convex_hull_mesh, hull_vertex_indices) = pcd2->ComputeConvexHull();

    // 提取凸包的顶点集合
    open3d::geometry::PointCloud hull_points;
    hull_points.points_ = convex_hull_mesh.vertices_;

    // 初始化过滤后的点云
    auto filtered_pcd = std::make_shared<open3d::geometry::PointCloud>();

    // 遍历pcd1的所有点，筛选凸包外部的点
    for (size_t i = 0; i < pcd1->points_.size(); ++i) {
        const auto& point = pcd1->points_[i];
        // 判断点是否在凸包内部
        bool is_inside = open3d::geometry::PointInConvexHull(point, hull_points.points_);
        if (!is_inside) {
            filtered_pcd->points_.push_back(point);
            // 同步复制颜色、法线等附加属性（如果存在）
            if (!pcd1->colors_.empty()) {
                filtered_pcd->colors_.push_back(pcd1->colors_[i]);
            }
            if (!pcd1->normals_.empty()) {
                filtered_pcd->normals_.push_back(pcd1->normals_[i]);
            }
        }
    }

    // 保存过滤后的点云
    open3d::io::WritePointCloud("filtered_pcd1.pcd", *filtered_pcd);
    return 0;
}

注意事项 #

• 如果点云存在噪声，可先对pcd2执行VoxelDownSample或StatisticalOutlierRemoval滤波，避免凸包被噪声点干扰
• 若需更精准的判断（比如排除凸包边界上的点），可在判断时给点添加微小偏移量

Python 实现方案 #

步骤说明 #

1. 加载点云并计算pcd2的凸包
2. 批量判断pcd1中所有点是否在凸包内
3. 通过索引筛选保留凸包外部的点

代码示例 #

import open3d as o3d

# 加载点云
pcd1 = o3d.io.read_point_cloud("pcd1.pcd")
pcd2 = o3d.io.read_point_cloud("pcd2.pcd")

# 计算pcd2的凸包
convex_hull_mesh, hull_indices = pcd2.compute_convex_hull()
# 提取凸包顶点
hull_points = o3d.geometry.PointCloud()
hull_points.points = convex_hull_mesh.vertices

# 批量判断pcd1中的点是否在凸包内
points_array = o3d.utility.Vector3dVector(pcd1.points)
is_inside_list = o3d.geometry.PointCloud.hull_point_in_convex_hull(hull_points, points_array)

# 筛选出凸包外部的点的索引
filtered_indices = [idx for idx, is_inside in enumerate(is_inside_list) if not is_inside]
# 根据索引生成过滤后的点云
filtered_pcd = pcd1.select_by_index(filtered_indices)

# 保存结果
o3d.io.write_point_cloud("filtered_pcd1.pcd", filtered_pcd)

补充方案（凸包法备选） #

如果凸包法出现误判（比如凸包包含了pcd1中不属于pcd2的点），可以改用KDTree最近邻匹配：

# 构建pcd2的KDTree
kdtree = o3d.geometry.KDTreeFlann(pcd2)
threshold = 1e-6  # 点匹配距离阈值，根据点云精度调整

filtered_indices = []
for idx, point in enumerate(pcd1.points):
    # 查找最近邻
    [k, idx_nn, dists] = kdtree.search_knn_vector_3d(point, 1)
    if dists[0] > threshold:
        filtered_indices.append(idx)

filtered_pcd = pcd1.select_by_index(filtered_indices)

内容的提问来源于stack exchange，提问作者mikkelsen1996