Reconstruction

重建

The Reconstruction goes a step further than the camera reconstruction (which only generates multiple 3D vertices) in that it attempts to reconstruct the video using a 3D Point Cloud or a mesh. In addition, vertex colors will be used to color the vertices or the mesh in accordance with the footage.

重建比摄像机重建(只生成多个3D 顶点)更进一步，它试图使用3D 点云或网格重建视频。此外，顶点颜色将被用来颜色顶点或网格根据镜头。

The result of this type of scene reconstruction can, under ideal conditions (see Optimal Footage below), be a dense Point Cloud or a coherent mesh (but actual results often differ because in most cases no ideal footage is available; you will then get patchy and isolated Point Clouds or meshes).

这种类型的场景重建的结果，在理想的条件下(见下面的最佳镜头) ，可以是一个密集的点云或一致的网格(但实际的结果往往不同，因为在大多数情况下没有理想的镜头可用; 然后你会得到片状和孤立的点云或网格)。

The scene reconstruction does not replace a 3D scanner. The Point Clouds or meshes that are created can, for example, serve as orientation in 3D space, for recreating objects and other purposes.

场景重建不能代替三维扫描仪。例如，创建的 Point Clouds 或网格可以作为三维空间的方向，用于重新创建对象和其他目的。

Requirements for scene reconstruction

场景重建的要求

In order to have a successful scene reconstruction you must always first perform a successful camera reconstruction, i.e., a 2D Tracking was done, after which the camera was reconstructed and the Run 3D Solver button was clicked to complete the process (or everything was done automatically by selecting the Full Solve command from the main Tracker menu). It can also be useful to calibrate the camera and features using the Constraint tag, which will correctly define the reconstructed scene’s and Point Cloud’s scale and orientation.

为了有一个成功的场景重建，你必须总是首先执行一个成功的相机重建，例如，一个2D 跟踪完成后，相机重建和运行3D 规划按钮被点击完成过程(或一切都是自动地通过选择完整解决命令从主跟踪器菜单)。使用 Constraint 标签校准摄像机和特征也很有用，它可以正确定义重建场景和 Point Cloud 的尺度和方向。

When is a camera reconstruction successful? This is the case after the previously described process is completed and a series of Auto or User Features have been added as Child objects to the Motion Tracker object. The Features are important because they will later be used as starting points for the scene reconstruction.

什么时候相机重建成功？这是前面描述的过程完成后的情况，一系列的自动或用户功能已经作为子对象添加到运动跟踪对象。这些特征非常重要，因为它们以后将被用作场景重建的起点。

The existing 2D Tracks will then no longer play a role for the scene reconstruction.

现有的2 d 轨道将不再起到场景重建的作用。

Optimal footage

最佳镜头

The optimal footage with which the scene reconstruction works as the following characteristics:

场景重建的最佳镜头具有以下特点:

• High level of image sharpness with as little depth of field or motion blur as possible 高清晰度的图像与尽可能少的景深或运动模糊
• Images should be very detailed and have a high number of even, monotone surfaces. 图像应该是非常详细的，有大量的均匀，单调的表面
• The sequence of images should depict as many sides of an object as possible in a consistent manner and without large jumps. 图像序列应该以一致的方式尽可能多地描述物体的各个侧面，而且不能有大的跳跃
• A sufficient number of 足够数量的Parallaxes 视差 should be present, which is not the case in footage with strong zoom lenses for example. For the same reasons, footage that consists of a single camera movement is also not suitable. 应该是存在的，这是不是在镜头与强大的变焦镜头例如。出于同样的原因，由单个摄像机运动组成的镜头也是不合适的
• Reflective materials such as glass, chrome and wandering specular highlights in the footage should be avoided, if possible. 反射材质，如玻璃，铬和漫游镜头高光应避免在镜头，如果可能的话
• If you want to create your own outdoor footage, it helps to do so under cloudy conditions in order to have a more even, diffuse light. 如果你想创建自己的户外镜头，它有助于在多云的条件下这样做，以便有一个更均匀，漫射光

Useful information

有用资料

• Independent of what the 独立于什么Resampling value is in the Footage menu, the scene reconstruction will always use the full video resolution (if you haven’t defined a different resolution). 值是在胶片菜单，场景重建将始终使用完整的视频分辨率(如果你没有定义不同的分辨率)
• Tracks serve as starting points from which neighboring regions will be reconstructed. 轨道作为重建邻近地区的起点
• Try to have a uniform, adequate Track depth. The more Tracks that exist, the more starting points there will be and the greater the probability that more regions will reconstructed. The exception to the rule: a single feature can suffice to reconstructs a relatively large, continuous surface. 尽量有一个统一的，足够的轨道深度。轨道越多，起点就越多，重建更多区域的概率也就越大。规则的例外: 单个特征就足以重构一个相对较大的连续表面
• Large, uniformly colored regions (e.g., sky) cannot be identified from different camera positions and will therefore remain unconstructed. 大的、均匀着色的区域(例如，天空)无法从不同的摄像机位置识别，因此将保持未构建状态

Limitations

局限性

• Higher values for settings or high footage resolution can require a lot of memory. Internal testing showed that far more than 10 GB can be required. 更高的设置值或高镜头分辨率可能需要大量的内存。内部测试显示需要的空间远远超过10gb
• Depending on the settings, render times of several to very many minutes may be required. 根据设置，渲染时间可能需要几到几分钟
• A good camera reconstruction is required for a scene reconstruction. If the video quality is not good enough for correctly calculating a camera reconstruction, the scene reconstruction that uses it as a base will not turn out particularly well. 场景重建需要一个好的相机重建。如果视频质量不够好，无法正确计算摄像机的重建，那么用它作为基础的场景重建就不会有特别好的结果
• Regions that are not seen by the camera can, of course, not be reconstructed accurately. The same applies for regions in which no or too few 3D features were calculated. 当然，摄像机无法看到的区域无法精确重建。这同样适用于没有计算或计算的三维特征太少的区域

How it works

它是如何工作的

The algorithm that is used is complex and will only be described here in short (since the parameter naming will otherwise be incomprehensible when patches and cells are explained). Fortunately, the Quality setting adjusts the settings according to its own quality criteria.

所使用的算法非常复杂，只在这里简短地描述(因为在解释补丁和单元格时，参数命名将变得难以理解)。幸运的是，质量设置根据自己的质量标准来调整设置。

Generally speaking, what happens when a scene construction is run?

一般来说，运行场景构造时会发生什么？

1. A range of frames will be selected (at least 10; more if the camera movement is faster) 一个帧的范围将被选择(至少10; 如果相机移动更快，更多)
   
2. Each frame has a camera position (ascertained through the previously executed 3D camera reconstruction) 每一帧都有一个摄像机的位置(通过先前执行的3D 摄像机重建确定)
   
3. Each of these frames will be divided into square cells using a grid. 每一帧都将使用网格被划分为方形单元格
   
4. A 3D point in the scene that is visible from various camera positions has a 2D position in each corresponding frame. This point in turn lies within a cell. 从不同的摄像机位置可以看到的场景中的3D 点在每个相应的帧中都有一个2D 位置。这一点依次位于一个单元格内
   
5. The previously completed 3D camera reconstruction has already generated 3D points (the 3D features). These are used to create start patches. Patches are points in space from which the Point Cloud will later be generated. Each patch represents a small, flat surface region. 之前完成的3D 摄像机重建已经生成了3D 点(3D 特征)。这些用于创建启动补丁。补丁是空间中的点，点云稍后将从中生成。每个补丁代表一个小，平坦的表面区域
   
6. A patch and corresponding cell will be defined for each frame. 将为每个帧定义一个补丁和相应的单元格
   
7. An iterative expansion process will now take place during which additional patches will be created based on the existing patches (these lie on the 3D surface to be reconstructed). 现在将进行一个迭代扩展过程，在此过程中，将基于现有的补丁创建额外的补丁(这些补丁位于要重建的3D 表面上)
   
8. New patches will be assigned to the cells and the cell grid will slowly become more full. The aim is to place a patch in every cell (which rarely actually happens). 新的补丁将分配给细胞和细胞网格将慢慢变得更加充分。其目的是在每个细胞中放置一个补丁(这种情况实际上很少发生)
   
9. The expansion of new patches to this point was aimed towards empty cells. 新的补丁扩大到这一点是针对空的细胞
   
10. The expansion process continues until no new patches can be generated that meet the various internal filter requirements (each patch that is created is run through a series of tests; if these tests are not passed, the patch will be discarded) or the cels from all frames have at least one patch. 扩展过程持续进行，直到没有新的补丁可以满足各种内部过滤器的要求(创建的每个补丁都要经过一系列测试; 如果这些测试没有通过，补丁将被丢弃) ，或者所有帧的 cels 都至少有一个补丁
    

In a nutshell, the algorithm uses existing 3D vertices (the camera reconstruction features in the first expansion process; the patches of previous expansions in subsequent expansion processes) as a starting position and expands outwardly across a 3D surface. How well this works depends on the footage or scene (as well as the camera reconstruction).

简而言之，该算法使用现有的3D 顶点(第一次展开过程中的摄像机重建特征; 后续展开过程中的先前展开的补丁)作为起始位置，并向外展开穿过3D 表面。这种方法的效果如何取决于镜头或场景(以及相机的重建)。

The patches are assigned to a polygon object named Scenepointcloud. Each vertex has a color that was ascertained from the footage that is added to the polygon object as a Vertex Color tag. If desired, a "meshing" can be done, which will generate a polygon surface from the reconstructed vertices.

补丁被分配给一个名为 Scenepointcloud 的多边形对象。每个顶点都有一种颜色，这种颜色是从作为顶点颜色标记添加到多边形对象的镜头中确定的。如果需要，一个“啮合”可以完成，这将产生一个多边形曲面从重建顶点。

Preset

预置

Here you will find various presets that define the settings below according to the available quality levels. If you can’t achieve acceptable results using these presents then you have to assume that you footage is not optimal (see also Optimal Footage).

在这里你会发现各种预设定义下面的设置根据可用的质量水平。如果你不能达到可接受的结果使用这些礼物，那么你必须假设你的镜头不是最佳的(也看到最佳镜头)。

Advanced Settings

高级设置

Iterations[1..3]

迭代[1. . 3]

If you take a quick look at the order of how the internal processes function you will see that a series of iterative tasks are processed. These are primarily the creation of new patches in empty regions, optimizing and filtering of "bad" patches - and then repeating these processes. This setting defines the number of iterations that should take place.

如果您快速查看一下内部流程的功能顺序，您将看到处理了一系列迭代任务。这些主要是在空白区域创建新的补丁，优化和过滤“坏”补丁，然后重复这些过程。此设置定义应该进行的迭代次数。

The higher the value, the fewer "gaps" that the Point Cloud will exhibit and the longer the render times will be.

值越高，点云显示的“间隙”就越少，渲染时间也就越长。

Footage Subsampling[0..8]

镜头二取样[0. . 8]

Two settings that work independently of one another are available for the 3D camera reconstruction (Resampling) and the scene reconstruction that define the resolution with which the footage will be used for reconstruction. The Footage Subsampling setting is used for scene reconstruction.

在3D 摄像机重建(Resampling)和场景重建中有两个独立工作的设置，这两个设置定义了用于重建的镜头的分辨率。在场景重建中，采用了镜头分采样设置。

If the footage should be processed internally with a lower resolution for scene reconstruction, this value should be increased. A value of 0 reflects the original resolution (see next paragraph, 1 half the resolution (in sions, i.e., a square pexel region of 4 pixels will be compiled), etc. The scene reconstruction calculates faster with increasing resolution but the Point Cloud density will abate accordingly.

如果镜头内部处理的分辨率较低，场景重建，这个值应该增加。值0反映原始分辨率(见下一段，分辨率的一半(在表达式中，即编译4像素的正方形 pexel 区域)等等。随着分辨率的提高，场景重建计算速度越来越快，但是点云密度会相应减小。

This setting can be used (e.g., values greater than 1) if you have footage with a very high resolution loaded. This will help reduce render times somewhat.

如果你有一个非常高分辨率的视频，可以使用这个设置(例如，大于1的值)。这将有助于减少渲染时间。

Note that a value of 0 for the original resolution will not necessarily produce better results than if set to 1. Internally, deviating resolutions will be used, depending on the camera’s distance from the patch. The resolution to be defined only applies for an explicit camera to patch distance (which is defined using the "reference camera"). The defined resolution will be used for all distances. If the distances between cameras and patch vary greatly, lower as well as higher resolutions will automatically be applied. However, if the value is set to 0, higher resolutions cannot be applied.

注意，原始分辨率为0的值并不一定比设置为1的结果更好。在内部，将使用不同的分辨率，这取决于相机从补丁的距离。定义的分辨率只适用于一个显式相机补丁距离(这是定义使用“参考相机”)。定义的分辨率将用于所有的距离。如果相机和补丁之间的距离变化很大，低分辨率和高分辨率将自动应用。但是，如果该值设置为0，则不能应用更高的分辨率。

This mechanism makes it possible to more easily compare patches from various distances.

这种机制使得比较不同距离的补丁变得更加容易。

Point Density[1..8]

点密度[1. . 8]

As explained under Function, each frame in the footage is divided into a grid. The algorithm attempts to calculate a 3D point for the Point Cloud in each cell of the grid. Ideally (with optimal footage), this will work (in most cases, however, gaps will remain).

正如在函数下解释的那样，连续镜头中的每一帧都被划分为一个网格。该算法试图计算网格中每个单元中点云的三维点。理想情况下(最佳镜头) ，这将工作(在大多数情况下，然而，差距将继续)。

In any case, the Point Density value defines the size of the individual cells of the grid. The larger the value, the smaller and more dense the cells will be and the more dense the Point Cloud will be. The maximum value of 8, the cell size will equal the pixel size, whereby, ideally, one point will lie on each pixel of the footage. Of course the render times will increase accordingly with an increasing Point Density value.

在任何情况下，点密度值定义了网格中单个单元格的大小。值越大，单元格越小，密度越大，点云密度越大。最大值为8，单元格大小将等于像素大小，即，理想情况下，一个点将位于每个像素的镜头。当然渲染时间会随着点密度值的增加而增加。

Patch Size[3..2147483647]

补丁大小[3. . 2147483647]

This algorithm tries to recognize small regions in the footage at specific times in order to define and better optimize the position and rotation of the patches. The Patch Size value defines the square size of these regions in pixels. In principle, this value reflects the Pattern Size in the 2D Tracking menu only that it’s for scene reconstruction:

该算法尝试在特定的时间段识别镜头中的小区域，以便更好地定义和优化补丁的位置和旋转。补丁大小值以像素为单位定义这些区域的正方形大小。原则上，这个值反映了2 d 跟踪菜单中的模式大小，只是用于场景重建:

Again: Each patch is subsequently represented by a 3D point in the Point Cloud.

再次强调: 每个补丁随后在 Point Cloud 中用一个3D 点来表示。

Higher values will increase the quality of the Point Cloud accordingly, with correspondingly longer render times. Note in the image above how, for example, the distortion of the window frame and bars improves with increasing Patch Size values.

更高的值将相应地提高点云的质量，相应地延长渲染时间。在上面的图像中注意，例如，窗口框架和条形图的失真是如何随着补丁大小值的增加而改善的。

Min Texture Detail[0..100%]

最小纹理细节[0. . 100% ]

This value controls a threshold value that uses a patch’s pixel variation as a reference. If no real changes take place within a patch’s surface, i.e., if it is basically monotone, without high contrasts, a higher Min Texture Detail value will prevent a patch from being created (= no reconstruction at this location).

此值控制一个阈值，该阈值使用补丁的像素变化作为参考。如果补丁的表面没有发生真正的变化，也就是说，如果它基本上是单调的，没有高对比度，一个较高的最小纹理细节值将阻止补丁被创建(= 在这个位置没有重建)。

The purpose of this setting is to prevent patches being created in regions with low texture details (e.g., constant blue sky) by carefully increasing this setting’s value. Such regions are very difficult to process and are therefore prone to error.

此设置的目的是通过小心地增加此设置的值，防止在纹理细节较低的区域(例如，恒定的蓝天)创建补丁。这些区域很难处理，因此容易出错。

If you have footage with a high level of texture detail, increasing this value can lead to improved quality for the Point Cloud.

如果你的素材有很高的纹理细节，增加这个值可以提高点云的质量。

To sum it up, higher values lead to fewer but higher-quality patches/reconstruction points; lower values create more but lower-quality patches/reconstruction points.

综上所述，较高的值导致较少但较高质量的补丁/重构点; 较低的值导致较多但较低质量的补丁/重构点。

The default value of 0.5% should be good enough for most footage.

默认值0.5% 对于大多数镜头来说应该足够好了。

Filter Small Groups

过滤小组

If this option is enabled, the scene reconstruction will try to remove small, isolated patch regions. Note in the image above how many small, isolated point clusters were filtered out.

如果启用此选项，场景重建将尝试删除小的、孤立的补丁区域。请注意，在上面的图片中，有多少小的、孤立的点群被过滤掉了。

The effectiveness of this setting, however, depends a lot on the footage and can therefore also be disabled since patches can end up being removed unnecessarily.

然而，这种设置的有效性很大程度上取决于视频片段，因此也可能被禁用，因为补丁最终可能被删除，而不必要。

Min Angle[0..180°]

最小角度[0. . 180 ° ]

Which angle is affected? Imagine a straight line between the first camera and the patch and between the second camera and the patch: both lines meet at the patch. The angle between these lines is the angle that is affected.

哪个角度受影响？想象一下在第一台摄像机和补丁之间以及第二台摄像机和补丁之间有一条直线: 两条直线在补丁处相交。这些线之间的角度是受影响的角度。

Patches are processed from different camera positions (those from the camera reconstruction). This setting ensures that a minimum angle between the cameras is maintained to prevent patches from being processed by two neighboring cameras that have possibly not moved (the patch will be nearly identical).

补丁是从不同的摄像机位置处理(那些从摄像机重建)。这种设置确保了摄像机之间保持最小的角度，以防止补丁被相邻的两个摄像机处理，可能没有移动(补丁将几乎完全相同)。

The default value works well for most footage. For videos in which the camera moves only very slightly it can help to lower the value slightly. This will, however, make the reconstruction less precise. Don’t be fooled if the view fills up with patches and reconstructed points when lower values are used - quantity does not always equal quality.

默认值适用于大多数镜头。对于摄像机只是轻微移动的视频，它可以帮助稍微降低价值。然而，这将使重建不那么精确。不要被欺骗，如果视图填补补丁和重建点时，较低的值使用-数量并不总是等于质量。

To sum it up, higher values lead to fewer but higher-quality patches/reconstruction points; lower values create more but lower-quality patches/reconstruction points. Values that are too high will result in no Point Cloud being created at all.

综上所述，较高的值导致较少但较高质量的补丁/重构点; 较低的值导致较多但较低质量的补丁/重构点。如果值太高，则不会创建任何 Point Cloud。

Meshing

网格化

After the settings in the Advanced Settings menu have been defined, a polygon mesh can be created using these points.

在“高级设置”菜单中定义了设置后，可以使用这些点创建多边形网格。

Area Weight[0..100%]
Photometric Weight[0..100%]

面积重量[0. . 100% ]光度重量[0. . 100% ]

Both settings have a similar effect.

两种设置都有相似的效果。

With increasing values, these settings will attempt to replace larger triangles with smaller triangles that cover the same surface.

随着数值的增加，这些设置将尝试用覆盖同一表面的小三角形替换大三角形。

Faulty surfaces will be averted, if possible. If very large values are defined, correspondingly less mesh will remain, depending on the scene.

如果可能的话，将避免错误的表面。如果定义了非常大的值，则根据场景的不同，保留的网格也会相应减少。

The Photometric Weight setting works a little smarter than the Area Weight setting because it also uses the footage information.

配光权重设置比面积权重设置更聪明一点，因为它也使用了画面信息。

Generate Mesh

生成网格

If a scene reconstruction has already taken place, i.e., you’ve already clicked on the Create Point Cloud button, this data will be used to generate a corresponding polygon mesh. The Area Weight and Photometric Weight settings can be changed at any time and clikcing on Generate Mesh will generate a new mesh without having to perform a completely new scene reconstruction.

如果已经进行了场景重建，比如，你已经点击了创建点云按钮，这些数据将用于生成相应的多边形网格。区域重量和光度重量设置可以随时改变，并且在生成网格上的打火将生成一个新的网格，而不需要执行一个全新的场景重建。

If no scene reconstruction has taken place, this will be done and the mesh will be generated. The polygon object’s individual points will correspond to reconsructed 3D positions in space.

如果没有进行场景重建，这将完成和网格将生成。多边形对象的单个点将对应于空间中重新配置的三维位置。

It’s recommended to first generate a point cloud until you’re satisfied with the density and coverage before turning your attention to the mesh.

建议首先生成一个点云，直到你满意的密度和覆盖率，然后再把你的注意力转向网格。

Tip: Assign a Phong tag to the polygon mesh to smooth it.

提示: 为多边形网格分配一个 Phong 标签来平滑它。

Problematic regions in the mesh contain gaps in the Point Cloud. These will initially be closed, regardless of their size, which can lead to very large triangles being created and in turn does not look very good. The previous two settings can be used to lessen this effect.

网格中有问题的区域包含点云中的缺口。不管它们的大小如何，它们最初都会被关闭，这会导致非常大的三角形被创建出来，反过来看起来也不是很好。前面的两个设置可以用来减少这种影响。

Clicking on this button starts the scene reconstruction. The result, if successful, will be a scene point cloud with a Vertex Color tag (in which each color of each point in the footage is saved).

点击这个按钮开始场景重建。结果，如果成功，将是一个场景点云与顶点颜色标签(其中每个点的每个颜色在镜头保存)。

Depending on the footage and the settings, the calculation can take a few to several minutes to complete.

根据镜头和设置，计算可能需要几分钟到几分钟才能完成。

This button will be grayed out if one or more prerequisites are not met, for example:

如果一个或多个先决条件没有满足，这个按钮将变成灰色，例如:

• no camera reconstruction has been done 没有进行过摄像机重建
• the footage was not found 录像没有找到