深度图还原点云

自动摘要: 应用场景 ……..

应用场景

1、嵌体

2、冠的自动生成

PCN深度图还原点云（第一次迭代）

流程

1、模型输入

2、深度图获取

import os
import bpy
import numpy as np
import string
import random
input_dir = 'C:\\Users\\admin\\Desktop\\Dental_Inlay_Data\\input'
output_dir = 'C:\\Users\\admin\\Desktop\\dataset\\ShapeNetCore.v1'
model_list = open("C:/Users/admin/Desktop/dataset/model_list", "w")
length = 16
#生成8位随机数作为文件夹名称
random_num = str(random.randint(0, 99999999)).zfill(8)
f_name = os.path.join(output_dir, random_num)
os.mkdir(f_name)
for i in range(314):
 file_name = str(i+1)
 #创建场景，导入模型
 bpy.ops.import_mesh.ply(filepath=os.path.join(input_dir, (file_name + '.ply')))
 #选中模型
 bpy.ops.object.select_pattern(pattern=file_name, case_sensitive=False, extend=True)
 # 设置场景中心点
 bpy.ops.object.origin_set(type='ORIGIN_CURSOR', center='MEDIAN')
 bpy.ops.object.origin_set(type='GEOMETRY_ORIGIN', center='MEDIAN')
 bpy.ops.transform.resize(value=(0.1,0.1,0.1))
 # 设置导出ply的路径
 #生成31位随机数字加小写字母作为文件夹名称
 a = np.random.choice(list(string.ascii_lowercase + string.digits), length)
 ch_file_name= ''.join(a)
 all_dir = os.path.join(f_name, ch_file_name)
 os.mkdir(all_dir)
# 生成model_list文件
 model_list.write(random_num+'/'+ch_file_name)
 model_list.write('\n')
 bpy.ops.export_scene.obj(filepath=os.path.join(all_dir, ('model.obj')))
 bpy.ops.object.select_all(action='SELECT')
 bpy.ops.object.delete()

3、生成点云

遇到问题

问题1：代码错误

修改：

问题2：牙齿模型太大

解决方法：修改相机参数
缩小模型

问题3：Imath和OpenEXR包的安装

PCN深度图还原点云（第二次迭代）

流程

修改blender为2.79版本采样的深度图成功还原成点云

流程（第二次迭代）

1、模型输入

2、深度图获取

import math
from math import sqrt
import pandas as pd
import open3d as o3d
import numpy as np


pcd = o3d.io.read_point_cloud('/home/zpp/depth_image/models/data0200.ply')
print(pcd)
all_points = np.asarray(pcd.points)
print(all_points)
obb = pcd.get_oriented_bounding_box()
obb.color = (0,1,0)#obb包围盒为绿色
eight_points = np.asarray(obb.get_box_points())
#print(eight_points)
o3d.visualization.draw_geometries([pcd, obb])
pcd_points = np.asarray(pcd.points)
min_x = np.min(eight_points[:, 0])
max_x = np.max(eight_points[:, 0])
min_y = np.min(eight_points[:, 1])
max_y = np.max(eight_points[:, 1])
min_z = np.min(eight_points[:, 2])
max_z = np.max(eight_points[:, 2])
#print(min_x, max_x, min_y, max_y, min_x, max_x)
"""
法向量n
三个点确定一个平面

"""
point1 = np.array([min_x, min_y, max_z])
point2 = np.array([min_x, max_y, max_z])
point3 = np.array([max_x, min_y, max_z])
point4 = np.array([max_x, max_y, max_z])
r = 100
point_d = []
def point_face_d(point1, point2, point3, point4, r, all_points):
    AB = np.asmatrix(point2-point1)
    AC = np.asmatrix(point3-point1)
    N = np.cross(AB, AC)
    A = N[0, 0]
    B = N[0, 1]
    C = N[0, 2]
    D = -(A * point4[0] + B * point4[1] + C * point4[2]-r)
    #function
    for p in all_points:
        d = (A*p[0]+B*p[1]+C*p[2]+D)/sqrt(A**2+B**2+C**2)
        point_d.append(d)
    return r/C
#平移距离
translation_z = point_face_d(point1, point2, point3, point4, r, all_points)
all_points[:, 2] = point_d
print(all_points)
#将距离d进行归一化处理后*255取整
max_d = np.max(point_d)
max_x = np.max(all_points[:, 0])
max_y = np.max(all_points[:, 1])
print(max_x, max_y)
# img_h = math.ceil(point2[1]-point1[1])
# img_w = math.ceil(point3[0]-point1[0])
# img_w = max_x - min_x
# img_h = max_y - min_y
# print(img_w, img_h)
t = 2.54
nor_point_d = []
nor_point_x = []
nor_point_y = []
result = []
for d in point_d:
    nor_point_d.append(round(d * 255 / max_d))
for x in all_points[:, 0]:
    #x = x + abs(min_x)
    nor_point_x.append(round(x * 100 * t / max_x))
for y in all_points[:, 1]:
    y = y + abs(min_y)
    nor_point_y.append(round(y * 100 / max_y))
img_depth = np.zeros((int(140*t), 140), np.uint8)
for idx, d in enumerate(nor_point_d):
    a = nor_point_x[idx]
    b = nor_point_y[idx]
    tmp = (nor_point_x[idx], nor_point_y[idx], nor_point_d[idx])
    result.append(tmp)
    img_depth[a, b] = d
tmp = np.array(result)
h = abs(max_z - min_z) + translation_z
ori_z = []
for z in tmp:
    ori_value = h - z[2]/255*max_d
    ori_z.append(ori_value)
ori_z = np.array(ori_z)
tmp[:, 2] = ori_z
from PIL import Image
im = Image.fromarray(img_depth)
im.save("out1.jpg")

point_cloud = o3d.geometry.PointCloud()
point_cloud.points = o3d.utility.Vector3dVector(tmp)

o3d.io.write_point_cloud('/home/zpp/depth_image/models/data.ply', point_cloud)
o3d.visualization.draw_geometries([point_cloud])