python实现插值填充有序均匀点云空洞（2/2）

最新推荐文章于 2023-11-09 18:53:11 发布

三尺流流

最新推荐文章于 2023-11-09 18:53:11 发布

阅读量1k

点赞数

文章标签： python 开发语言计算机视觉边缘计算图像处理

本文链接：https://blog.csdn.net/qq_55433334/article/details/129397423

版权

点云处理前沿理论复现专栏收录该内容

15 篇文章 28 订阅

订阅专栏

在上一节我们讨论了运用均值插值来对有空洞缺失的点云进行填充，

可以看到，对应缺失坐标的点云都补充出来了。但是后来就有人对这个逻辑的完整性提出了质疑，见下图：

图中通过单轴的插值获得的填充数据虽然在x轴方向上确实接上了，但是在y轴方向上却可以说是毫不相干，一点也不光滑。如何设计一种填充方案，使得两个方向的连接都比较平滑呢？

在这一节，我们将对处理逻辑进行一定的调整，使得填充的点云数据不管在哪个方向看过去都是相对平滑的连接。同时，也将类似下图这种“矩阵型”的列表数据构造方法在代码中用函数的方式表示出来。

“矩阵型”结构构造

我们将一个列表中嵌套多行或者多列的点云数据，用这个新的列表来表示“矩阵型”的方阵点云。在上一节的讨论中，出现了以下新型的按行排列的点云数据形式：

point_data_line = [
        [[1, 1, 2], [1, 2, 1], [1, 3, 2], [1, 4, 3], [1, 5, 2], [1, 6, 1], [1, 7, 2], [1, 8, 3], [1, 9, 5], [1, 10, 7]],
        [[2, 1, 1], [2, 2, 1], [2, 3, 2], [2, 4, 4], [2, 5, 3], [2, 6, 3], [2, 7, 4], [2, 8, 2], [2, 9, 4], [2, 10, 6]],
        [[3, 1, 1], [3, 2, 2], [3, 3, 3], [3, 4, 3], [3, 8, 1], [3, 9, 2], [3, 10, 5]],
        [[4, 1, 2], [4, 2, 3], [4, 3, 2], [4, 9, 6], [4, 10, 4]],
        [[5, 1, 4], [5, 2, 5], [5, 9, 2], [5, 10, 4]],
        [[6, 1, 6], [6, 2, 7], [6, 9, 4], [6, 10, 5]],
        [[7, 1, 5], [7, 2, 6], [7, 9, 3], [7, 10, 7]],
        [[8, 1, 5], [8, 2, 7], [8, 3, 5], [8, 9, 5], [8, 10, 7]],
        [[9, 1, 3], [9, 2, 4], [9, 3, 6], [9, 4, 7], [9, 5, 8], [9, 6, 6], [9, 7, 8], [9, 8, 7], [9, 9, 6], [9, 10, 8]],
        [[10, 1, 5], [10, 2, 5], [10, 3, 7], [10, 4, 6], [10, 5, 4], [10, 6, 5], [10, 7, 6], [10, 8, 8], [10, 9, 7],
         [10, 10, 9]]]

由于我们的点云列表数据有的行数据存在缺失，所以其并非真正意义上的矩阵。只能自己手写构造，其方法如下：

1.找到行或者列坐标相同的点。

2.这些点添加进新的空列表。

3.此列表所有点按照另一坐标轴从小到大排列。

4.按这个步骤依次从1到10构造新的行或者列并添加进整体的“矩阵型”。

均值平滑

如图：

在靠近边缘点，可能由于每行每列边界点z轴之间的差值大小变化而产生较大的偏差。由于不是直接拟合曲面函数，这种误差不好避免。为尽量减少这种“锯齿状”的误差，我们可以同时结合x轴方向与y轴方向的插值来实现。

这里我们直接将两个方向的插值数据按照xoy平面坐标做一个z轴的均值计算，以达到尽量平滑连接空洞边缘的效果。

逻辑流

1. 导入原始点云

2. 分别构造x，y方向的新“矩阵型”结构

3. 分别搜索x，y方向的点云空洞

4. 分别插值填充x，y方向的点云空洞

5. 按照坐标对两种方向的插值求均值，平滑填充点云

完整代码

import numpy as np
import matplotlib.pyplot as plt


point_data = [[1, 1, 2], [1, 2, 1], [1, 3, 2], [1, 4, 3], [1, 5, 2], [1, 6, 1], [1, 7, 2], [1, 8, 3], [1, 9, 5], [1, 10, 7],
                  [2, 1, 1], [2, 2, 1], [2, 3, 2], [2, 4, 4], [2, 5, 3], [2, 6, 3], [2, 7, 4], [2, 8, 2], [2, 9, 4], [2, 10, 6],
                  [3, 1, 1], [3, 2, 2], [3, 3, 3], [3, 4, 3], [3, 8, 1], [3, 9, 2], [3, 10, 5],
                  [4, 1, 2], [4, 2, 3], [4, 3, 2], [4, 9, 6], [4, 10, 4],
                  [5, 1, 4], [5, 2, 5], [5, 9, 2], [5, 10, 4],
                  [6, 1, 6], [6, 2, 7], [6, 9, 4], [6, 10, 5],
                  [7, 1, 5], [7, 2, 6], [7, 9, 3], [7, 10, 7],
                  [8, 1, 5], [8, 2, 7], [8, 3, 5], [8, 9, 5], [8, 10, 7],
                  [9, 1, 3], [9, 2, 4], [9, 3, 6], [9, 4, 7], [9, 5, 8], [9, 6, 6], [9, 7, 8], [9, 8, 7], [9, 9, 6], [9, 10, 8],
                  [10, 1, 5], [10, 2, 5], [10, 3, 7], [10, 4, 6], [10, 5, 4], [10, 6, 5], [10, 7, 6], [10, 8, 8], [10, 9, 7], [10, 10, 9]]

# 构造行
def point_data_row(point_data):
    point_row = []
    for i_row in range(10):
        row = []
        for i_column in range(10):
            for i in range(len(point_data)):
                if point_data[i][0] == i_row + 1 and point_data[i][1] == i_column + 1:
                    row.append(point_data[i])
        point_row.append(row)
    return point_row

# 构造列
def point_data_column(point_data):
    point_column = []
    for i_column in range(10):
        column = []
        for i_row in range(10):
            for i in range(len(point_data)):
                if point_data[i][1] == i_column + 1 and point_data[i][0] == i_row + 1:
                    column.append(point_data[i])
        point_column.append(column)
    return point_column


# 行填充
def boundary_row(point_data_line):
    boundary_points_row = []
    for i in range(len(point_data_line)):
        num_row_pre = 1
        for i_row in range(len(point_data_line[i])):
            num_row = point_data_line[i][i_row][1]
            if num_row != num_row_pre:
                boundary_points_first = point_data_line[i][i_row - 1]
                boundary_points_last = point_data_line[i][i_row]
                boundary_points_row.append([boundary_points_first, boundary_points_last])
                break
            num_row_pre += 1
    return boundary_points_row


def fillp_row(boundary_points):
    copy_point_data = []
    for i in point_data:
        copy_point_data.append(i)
    for i in range(len(boundary_points)):
        numpoints = boundary_points[i][1][1] - boundary_points[i][0][1] + 1
        fill_points = np.linspace(boundary_points[i][0], boundary_points[i][1], numpoints)
        for p in fill_points:
            copy_point_data.append(list(p))
    point_data_fill_row = copy_point_data
    return point_data_fill_row

# 列填充
def boundary_column(point_data_line):
    boundary_points_column = []
    for i in range(len(point_data_line)):
        num_row_pre = 1
        for i_row in range(len(point_data_line[i])):
            num_row = point_data_line[i][i_row][0]
            if num_row != num_row_pre:
                boundary_points_first = point_data_line[i][i_row - 1]
                boundary_points_last = point_data_line[i][i_row]
                boundary_points_column.append([boundary_points_first, boundary_points_last])
                break
            num_row_pre += 1
    return boundary_points_column


def fillp_column(boundary_points):
    copy_point_data = []
    for i in point_data:
        copy_point_data.append(i)
    for i in range(len(boundary_points)):
        numpoints = boundary_points[i][1][0] - boundary_points[i][0][0] + 1
        fill_points = np.linspace(boundary_points[i][0], boundary_points[i][1], numpoints)
        for p in fill_points:
            copy_point_data.append(list(p))
    point_data_fill_column = copy_point_data
    return point_data_fill_column

def mean_fill(point_data_fill_row,point_data_fill_column):
    point_data_fill = []
    for i in range(len(point_data_fill_row)):
        for j in range(len(point_data_fill_column)):
            if point_data_fill_row[i][0] == point_data_fill_column[j][0] and point_data_fill_row[i][1] == point_data_fill_column[j][1]:
                point_data_fill.append([point_data_fill_row[i][0],point_data_fill_row[i][1],(point_data_fill_row[i][2]+point_data_fill_column[j][2])/2])

    return point_data_fill

def view(point_data, point_data_fill):
    fig1 = plt.figure()
    ax1 = fig1.add_subplot(111, projection='3d')
    ax1.set_xlabel("x")
    ax1.set_ylabel("y")
    ax1.set_zlabel("z")
    for i in range(len(point_data_fill)):
        xf = point_data_fill[i][0]
        yf = point_data_fill[i][1]
        zf = point_data_fill[i][2]
        ax1.scatter(xf, yf, zf, color='b', marker='.')
    for i in range(len(point_data)):
        x = point_data[i][0]
        y = point_data[i][1]
        z = point_data[i][2]
        ax1.scatter(x, y, z, color='gray', marker='o')
    plt.show()
    return None


if __name__ == '__main__':
    # 行列格式构造
    point_data_row = point_data_row(point_data)
    point_data_column = point_data_column(point_data)
    # x，y方向分别插值
    boundary_points_row = boundary_row(point_data_row)
    point_data_fill_row = fillp_row(boundary_points_row)
    boundary_points_column = boundary_column(point_data_column)
    point_data_fill_column = fillp_column(boundary_points_column)
    # 差值均值与绘图
    point_data_fill = mean_fill(point_data_fill_row, point_data_fill_column)
    view(point_data, point_data_fill)