130 and/or storage component 140 may cause the processor 120 to perform one or more processes described herein.

Additionally, or alternatively, hardwired circuitry may be used in place of or in combination with software instructions to perform one or more processes described herein. Thus, embodiments described herein are not limited to any specific combination of hardware circuitry and software.

The number and arrangement of components shown in FIG. 1 are provided as an example. In practice, device 100 may include additional components, fewer components, different components, or differently arranged components than those shown in FIG. 1. Additionally, or alternatively, a set of components (e.g., one or more components) of device 100 may perform one or more functions described as being performed by another set of components of device 100.

Any one of the operations or processes described below (e.g., FIGS. 2, 3, 4, 6, and 7) may be implemented by or using any one of the elements illustrated in FIG. 1.

FIG. 2 is a flow diagram illustrating a method of patch generation in Video-based Point Cloud Coding (V-PCC), according to various embodiments. Generally, V-PCC decomposes an input point cloud into a set of patches. In the MPEG point cloud coding (PCC) test model software “TMC2”, for example, patch generation involves multiple steps.

At 201, the method 200 includes determining a projection plane for each point in a point cloud (original point cloud) based on a defined clustering criterion (e.g., based on associated normal vectors and neighboring points).

At 202, the method 200 includes generating a list of raw points in the original point cloud. Initially, no point in the list belongs to any patch.

At 203, the method 200 includes generating a list of connected components (CCs) based on the points in the list of raw points, by grouping raw points that have the same projection plane and are neighboring each other. A K-D tree may be used to select each raw point's neighboring points. A number of points in each CC in the list of CCs is determined, and if the number of points is less than a threshold number of points, then the CC may be removed from the list of CCs.

At 204, the method 200 includes determining a number of CCs remaining in the list of CCs. If the number of CCs is zero, then the method 200 stops. If the number of CCs is greater than zero, then the method 200 proceeds to 205, and if not, then to 209.

At 205, the method 200 includes generating a near surface s-near, a far surface s-far, and a list of patches for the original point cloud. Generating the near surface s-near, the far surface s-far, and the list of patches is described in more detail in FIGS. 3, 4, 6, and 7.

At 206, the method 200 includes reconstructing a point cloud (reconstructed point cloud) based on points included in the near surface s-near and the far surface s-far, for all patches in the list of patches.

At 207, the method 200 includes building a K-D tree based on the reconstructed point cloud.

At 208, the method 200 includes, for each point in the original point cloud, searching the K-D tree based on the reconstructed point cloud for the point's nearest neighboring point. If a distance between a point in the original point cloud and its nearest neighboring point in the reconstructed point cloud is greater than a user-defined threshold, then the point is classified as a raw point. The method 200 returns to 203.

FIG. 3 is a flowchart illustrating a method of generating a near surface s-near, a far surface s-far, and a list of patches, based on a CC, according to various embodiments. The method 300 shown in FIG. 3 may be performed for each CC in the list of CCs at 205 in FIG. 2.

At 301, the method 300 includes removing outliers. The points in the CC may be projected to the projection plane associated with the CC. For example, a projected coordinate of a 3D point in a two-dimensional (2D) UV plane may be denoted as (u_i, v_i), i=0, 1, . . . , N−1, where N is the number of points in the CC. A minimum value for the projected coordinates may be denoted as:

u_min=min(u₀,u₁, . . . ,u_N-1), and

v_min=min(v₀,v₁, . . . ,v_N-1).

If (u_i−u_min)>U_thresoldor (v_i−v_min)>U_thresold, where U_thresoldand V_thresoldare user-defined thresholds, the i-th point is determined to be an outlier point and is removed from the CC. The resulting CC is referred to as the updated CC and is denoted as CC′.

At 302, the method 300 includes near surface generation. If the CC′ is not empty (a number of points in the CC′ is greater than zero), then for each 3D point in the CC′, its projected coordinate in the 2D UV plane may be denoted as (û_i, {circumflex over (v)}_i), i=0, . . . , {circumflex over (N)}−1, where {circumflex over (N)} is the number of points in the CC′. The method 300 may include determining a near surface s-near that is close to the projection plane associated with the CC′ by selecting points in the CC′. The points in the CC′ may be selected such that if one or more points have the same projected UV coordinate, then the point that has a smallest depth value is selected to be part of the near surface s-near. The depth value of a point may correspond to a distance from the point to the projection plane. The projected coordinates of the selected points may be denoted as (ũ_i, {tilde over (v)}_i), i=0, . . . , Ñ−1, and the depth value associated with each projected coordinate may be denoted as {tilde over (d)}_i, i=0, . . . , Ñ−1, where Ñ is the number of points in the near surface s-near.

At 303, the method 300 includes near surface filtering. For a T×T grid in the projected plane, where T is a user-defined positive integer, a minimum depth value among points in the CC′ that are projected to the T×T grid may be denoted as {tilde over (d)}_min. A minimum depth value among all the points in the CC′ may be denoted as {circumflex over (d)}_min, and a maximum allowed depth representation may be denoted as D_max. For a user-defined threshold d_threshold, and a user-defined surface thickness d_surface, if a point (ũ,{tilde over (v)},{tilde over (d)}) in the near surface S_nearthat is projected to the T×T grid satisfies the condition:

{tilde over (d)}>({tilde over (d)}_min+d_threshold) or ({tilde over (d)}+d_surface)>({circumflex over (d)}_min−D_max),

where (ũ,{tilde over (v)}) is the projected coordinate and {tilde over (d)} is the depth value of the point, then the point is removed from the near surface S_near.

At 305, the method 300 includes far surface generation. The method 300 may include determining a far surface s-far by selecting points in the CC′ such that if one or more points have the same projected UV coordinate, then the point that has a largest depth value is selected to be part of the far surface s-far. For example, the far surface s-far may be initialized to be the same as the selected points (ũi, {tilde over (v)}_i), i=0, . . . , Ñ−1 for the near surface s-near. If multiple points in the CC′ are projected to the same UV coordinate (ũ_i, {tilde over (v)}_i), then the i-th point in the far surface S_faris replaced with the point having the largest depth value such that a distance between this point and the point {tilde over (p)}_iin S_nearis not greater than d_surfaceand the two points' color values are close to each other.

At 304, the method 300 includes generating a list of patches. The method 300 may include adding a patch with the near surface and the far surface to a list of patches for the original point cloud.

FIG. 4 is a flowchart illustrating another method of generating a near surface s-near, a far surface s-far, and a list of patches, based on a CC, according to various embodiments. The method 400 shown in FIG. 4 may be performed for each CC in the list of CCs at 205 in FIG. 2.

At 401, the method 400 includes removing outliers. The points in the CC may be projected to the projection plane associated with the CC. For example, a projected coordinate of a 3D point in a two-dimensional (2D) UV plane may be denoted as (u_i,v_i), i=0, 1, . . . , N−1, where N is the number of points in the CC. A minimum value for the projected coordinates may be denoted as:

u_min=min(u₀,u₁, . . . ,u_N-1), and

v_min=min(v₀,v₁, . . . ,v_N-1).

If (u_i−u_min)>U_thresoldor (v_i−v_min)>U_thresold, where U_thresoldand V_thresoldare user-defined thresholds, the i-th point is determined to be an outlier point and is removed from the CC. The resulting CC is referred to as the updated CC and is denoted as CC′.

At 402, the method 400 includes far surface generation. If the CC′ is not empty (a number of points in the CC′ is greater than zero), then for each 3D point in the CC′, its projected coordinate in the 2D UV plane may be denoted as (û_i, {circumflex over (v)}_i), i=0, . . . , {circumflex over (N)}−1, where {circumflex over (N)} is the number of points in the CC′. The method 300 may include determining a far surface s-far that is farther away from the projection plane associated with the CC′ by selecting points in the CC′ such that if one or more points have the same projected UV coordinate, then the point that has a largest depth value is selected to be part of the far surface s-far. The depth value of a point may correspond to a distance from the point to the projection plane. The projected coordinates of the selected points may be denoted as (ũ₁,{tilde over (v)}_i), i=0, . . . , Ñ−1, and the depth value associated with each projected coordinate may be denoted as {tilde over (d)}_i, i=0, . . . , Ñ−1, where Ñ is the number of points in the far surface s-far.

At 403, the method 400 includes far surface filtering. For a T×T grid in the projected plane, where T is a user-defined positive integer, a maximum depth value among points in the CC′ that are projected to the T×T grid may be denoted as {tilde over (d)}_max. A maximum depth value among all the points in the CC′ may be denoted as {circumflex over (d)}_max, and a maximum allowed depth representation may be denoted as D_max. For a user-defined threshold d_threshold, and a user-defined surface thickness d_surface, if a point (ũ,{tilde over (v)},{tilde over (d)}) in the far surface S_farthat is projected to the T×T grid satisfies the condition:

{tilde over (d)}>({tilde over (d)}_max−d_threshold) or ({tilde over (d)}−d_surface)>({circumflex over (d)}_max−D_max),

where (ũ,{tilde over (v)}) is the projected coordinate and {tilde over (d)} is the depth value of the point, then the point is removed from the far surface S_far.

At 405, the method 400 includes near surface generation. The method 400 may include determining a near surface s-near by selecting points in the CC′ such that if one or more points have the same projected UV coordinate, then the point that has a smallest depth value is selected to be part of the near surface s-near. For example, the near surface s-near may be initialized to be the same as the selected points (ũ, {tilde over (v)}_i), i=0, . . . , Ñ−1 for the far surface s-far. If multiple points in the CC′ are projected to the same UV coordinate (ũ_i, {tilde over (v)}_i), then the i-th point in the near surface S_nearis replaced with the point having the smallest depth value such that a distance between this point and the point {tilde over (p)}_iin S_faris not greater than d_surfaceand the two points' color values are close to each other.

At 406, the method 400 includes generating a list of patches. The method 400 may include adding a patch with the near surface and the far surface to a list of patches for the original point cloud.

FIG. 5 illustrates an sample reconstructed point cloud. Referring to 303 in FIG. 3, and 403 in FIG. 4, the grid size T×T is fixed during an encoding process. If the parameter T is small, artifacts may appear. However, when T is large, many small patches may be produced and the coding efficiency may be reduced. As shown in FIG. 5, in an experiment using grid-based patch generation where T=4, coding artifacts may be observed in the marked area 501.

According to various embodiments, the method 300 or the method 400 may include performing adaptive filtering. For example, at 303, the method 300 may include adaptively filtering the generated near surface s-near, and at 403, the method 400 may include adaptively filtering the generated far surface s-far. That is, the parameter T may be adaptively adjusted based on characteristics of the CC.

Characteristics of a CC used to adaptively adjust the parameter T may include statistics associated with a depth of the points in the CC. For example, a depth range, a minimum/maximum depth value, a mean depth value, a depth variation, etc.

Characteristics of a CC used to adaptively adjust the parameter T may include statistics associated with a geometry of the points in the CC. For example, a bounding box, geometry distribution, etc.

Characteristics of a CC used to adaptively adjust the parameter T may include statistics associated with a color of the points in the CC. For example, a color similarity/variation etc.

Characteristics of a CC used to adaptively adjust the parameter T may include statistics associated with other attributes (e.g., reflectance, timestamp, etc.) of the points in the CC.

The parameter T may be adaptively adjusted based on only one characteristics or based on multiple characteristics together. Additionally, one or more thresholds and/or criteria may be defined (e.g., by a user) to determine a value for the parameter T for each CC in a list of CCs.

FIG. 6 is a flowchart illustrating a method of generating a near surface s-near, a far surface s-far, and a list of patches, based on a CC, by performing adaptive filtering, according to various embodiments. The method 600 shown in FIG. 6 may be performed for each CC in the list of CCs at 205 in FIG. 2.

At 601, the method 600 includes removing outliers. The points in the CC may be projected to the projection plane associated with the CC. For example, a projected coordinate of a 3D point in a two-dimensional (2D) UV plane may be denoted as (u_i,v_i), i=0, 1, . . . , N−1, where N is the number of points in the CC. A minimum value for the projected coordinates may be denoted as:

u_min=min(u₀,u₁, . . . ,u_N-1), and

v_min=min(v₀,v₁, . . . ,v_N-1).

If (u_i−u_min)>U_thresoldor (v_i−v_min)>U_thresold, where U_thresoldand V_thresoldare user-defined thresholds, the i-th point is determined to be an outlier point and is removed from the CC. The resulting CC is referred to as the updated CC and is denoted as CC′.

At 602, the method 600 includes near surface generation. If the CC′ is not empty (a number of points in the CC′ is greater than zero), then for each 3D point in the CC′, its projected coordinate in the 2D UV plane may be denoted as (û_i,{circumflex over (v)}_i), i=0, . . . , {circumflex over (N)}−1, where {circumflex over (N)} is the number of points in the CC′. The method 600 may include determining a near surface s-near that is close to the projection plane associated with the CC′ by selecting points in the CC′. The points in the CC′ may be selected such that if one or more points have the same projected UV coordinate, then the point that has a smallest depth value is selected to be part of the near surface s-near. The depth value of a point may correspond to a distance from the point to the projection plane. The projected coordinates of the selected points may be denoted as (ũ_i,{tilde over (v)}_i), i=0, . . . , Ñ−1, and the depth value associated with each projected coordinate may be denoted as {tilde over (d)}_i, i=0, . . . , Ñ−1, where Ñ is the number of points in the near surface s-near.

At 603, the method 600 includes near surface adaptive filtering. In one embodiment, the method 600 determines a range of depth for points in the CC. For example, the method 600 may include identifying a smallest depth value, denoted as d_min, and a largest depth value, denoted as d_max. The range of depth may be determined according to:

d_range=|d_max−d_min|,

where the operation |x| is the absolute value of x. The method 600 may include determining the parameter T based on the following:

T
=

{

T
1

if
⁢

d
range

>

d

T
⁢
h

T
2

else

,

where d_This a positive integer threshold value (e.g., 64, 128, etc.), T₁and T₂are two positive integer values such that T₁>T₂(e.g., T₁=32 and T₁=4). Here, the method 600 selects a larger value for T when d_rangeis relatively large and the artifacts are likely to appear.

For a T×T grid in the projected plane, based on the determined value for T, a minimum depth value among points in the CC′ that are projected to the T×T grid may be denoted as {tilde over (d)}_min. A minimum depth value among all the points in the CC′ may be denoted as {circumflex over (d)}_min, and a maximum allowed depth representation may be denoted as D_maxFor a user-defined threshold d_threshold, and a user-defined surface thickness d_surface, if a point (ũ, {tilde over (v)}, {tilde over (d)}) in the near surface S_nearthat is projected to the T×T grid satisfies the condition:

{tilde over (d)}>({tilde over (d)}_max−d_threshold) or ({tilde over (d)}−d_surface)>({circumflex over (d)}_max−D_max),

where (ũ, {tilde over (v)}) is the projected coordinate and {tilde over (d)} is the depth value of the point, then the point is removed from the near surface S_near.

At 605, the method 600 includes far surface generation. The method 600 may include determining a far surface s-far by selecting points in the CC′ such that if one or more points have the same projected UV coordinate, then the point that has a largest depth value is selected to be part of the far surface s-far. For example, the far surface s-far may be initialized to be the same as the selected points (ũ_i, {tilde over (v)}_i), i=0, . . . , Ñ−1 for the near surface s-near. If multiple points in the CC′ are projected to the same UV coordinate (ũ_i, {tilde over (v)}_i), then the i-th point in the far surface S_faris replaced with the point having the largest depth value such that a distance between this point and the point {tilde over (p)}_iin S_nearis not greater than d_surfaceand the two points' color values are close to each other.

At 606, the method 600 includes generating a list of patches. The method 600 may include adding a patch with the near surface and the far surface to a list of patches for the original point cloud.

FIG. 7 is a flowchart illustrating another method of generating a near surface s-near, a far surface s-far, and a list of patches, based on a CC, by performing adaptive filtering, according to various embodiments. The method