阅读说明：本技术 基于平面拟合中心坐标投影的点云空间可伸缩编码几何重构方法 (Point cloud space scalable coding geometric reconstruction method based on plane fitting center coordinate projection ) 是由 万帅 陈章 王哲诚 于 2021-07-04 设计创作，主要内容包括：本发明涉及一种基于平面拟合中心坐标投影的点云空间可伸缩编码几何重构方法,属于视频编解码技术领域。通过每个结点与周围26邻居结点构成的局部空间,利用邻居结点信息拟合当前结点的局部平面,再求解当前结点的中心点在该平面上投影点的坐标。解决了空间可伸缩后几何重构过程中的几何误差较大的问题。(The invention relates to a point cloud space scalable coding geometric reconstruction method based on plane fitting center coordinate projection, and belongs to the technical field of video coding and decoding. And fitting a local plane of the current node by using the information of the neighbor nodes through a local space formed by each node and the surrounding 26 neighbor nodes, and then solving the coordinates of the projection point of the central point of the current node on the plane. The problem of the geometric error is great in the space scalable back geometry reconstruction process is solved.)

1. A point cloud space scalable coding geometric reconstruction method based on plane fitting center coordinate projection is characterized in that: and fitting a local plane of the current node by using the information of the neighbor nodes through a local space formed by each node and the surrounding 26 neighbor nodes, and then solving the coordinates of the projection point of the central point of the current node on the plane.

2. The method of claim 1, wherein the surrounding 26 neighbor nodes comprise 6 coplanar neighbor nodes, 12 common edge neighbor nodes, and 8 common point neighbor nodes.

3. The method for point cloud space scalable coding geometric reconstruction based on plane fitting center coordinate projection according to claim 1, wherein fitting the local plane of the current node by using the neighbor node information specifically comprises:

1) k neighbor node plane fitting judgment

Retrieving 26 neighbor situations of the current neighbor node, and setting the number of neighbors existing in the 26 neighbors as: neighNum;

when the neighbor num is larger than or equal to K, calculating K neighbor nodes closest to the node center position coordinate by using the neighbor num nodes existing in 26 neighbor nodes around the node, and calculating the neighbor node center position coordinate by using the P point coordinate, the octreeSize of the node side length of the octree and the position of the neighbor;

2) least square method plane fitting

The final equation for plane fitting by using the least square method plane fitting principle and the central position coordinates of N1-Nk is as follows:

a₀*x+a₁*y+a₂-z＝0

wherein x_i、y_i、z_iAnd the coordinates of the center position of the neighbor node.

4. The method of claim 3, wherein the K value is equal to the number of neighbors searched in the normal vector of the plane of the point cloud image, and is 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16.

5. The method of claim 1, wherein the method comprises the following steps: solving the coordinates of the projection point of the central point of the current node on the plane as follows:

wherein: x is the number of₀、y₀、z₀The coordinates of the current geometric center position, namely the center position of the cube; x is the number of₁、y₁、z₁And fitting the geometric coordinates calculated by the geometric reconstruction method for the plane before constraint.

Technical Field

The invention relates to the technical field of video coding and decoding, in particular to a point cloud space scalable coding geometric reconstruction method based on plane fitting center coordinate projection.

Background

In a point cloud G-PCC encoder framework, the slice is independently encoded after the input point cloud is divided. In slice, the geometric information of the point cloud and the attribute information corresponding to the points in the point cloud are encoded separately. The G-PCC encoder first encodes the geometry information. The encoder performs coordinate conversion on the geometric information to enable the point clouds to be contained in a bounding box; and then, quantization is carried out, wherein the quantization step mainly plays a role of scaling, as the quantization rounding ensures that the geometric information of a part of points is the same, whether to remove the repeated points is determined according to the parameters, the process of not removing the repeated points is called geometric lossless compression, the process of removing the repeated points is called geometric lossy compression, and the process of quantizing and removing the repeated points is also called a voxelization process. Next, the bounding box is divided based on octree. The geometric lossless compression process and the geometric lossy compression process have geometric reconstruction processes when the octree division is completed.

In the point cloud G-PCC decoder framework, the geometric bit stream information of the point cloud and the attribute bit stream corresponding to the points in the point cloud are decoded separately. The G-PCC decoder first decodes the geometry bitstream. The decoder performs arithmetic decoding on the geometric bit stream to decode a bounding box of the point cloud and occupied bits (1 is non-empty and 0 is empty) based on octree; decoding of geometric information is divided into two frames based on octree and trisup (triangular patch set) according to the difference of the hierarchy depth of octree partitioning during encoding.

spatial scalability (spatial scalability) is an important function of G-PCC, generating point cloud thumbnails by decoding partial point cloud bitstream information, and currently only works in decoding frameworks based on octree geometry information. skip Layer is an octree level with less decoding at the decoding end. As shown in fig. 1, the octree geometry is encoded to the K-th layer, and when the spatial scalability process is not performed, the K-th layer is completely decoded; and partial decoding is carried out when the spatial scalability process is carried out until the M Layer is finished (M is K-skip Layer).

Its function is controlled by the parameter scalable _ shifting _ enabled _ flag. When the scalable _ shifting _ enabled _ flag is equal to 0, the spatial scalability function is not performed; when the scalable _ shifting _ enabled _ flag is 1, the spatial scalability function is performed.

In the geometric information decoding framework based on the octree, the current point cloud in the space is calculated according to a bounding boxThe non-empty subcubes are further divided into eight equal parts according to the space occupying bits, the division is stopped when the leaf nodes obtained by the division are the unit cubes of 1 × 1 × 1, but if the decoding process is spatially scalable, the leaf nodes are divided into the specified skip Layer to generate 2 skip layers^skipLayer×2^skipLayer×2^skipLayerThe cube of (1). Then, generating geometric coordinates through the cube, wherein the geometric coordinates are coordinates of the left front lower corner of the cube for a unit cube with leaf nodes of 1 multiplied by 1, and the geometric coordinates are coordinates of 2 for the leaf nodes^skipLayer×2^skipLayer×2^skipLayerAccording to the different skip Layer, different reconstruction strategies are adopted for the geometrical coordinates of the unit cube.

The current scalable coding geometry reconstruction method in the standard is proposed by Hyejung Hur, Sejin Oh of LGelectronics Inc in proposal m52315 at 1 month 2020 and is received by the MPEG G-PCC standard (i.e., MPEG-I (ISO/IEC23090) Part 9). The technical scheme is that different reconstruction strategies are adopted according to different levels of the skip Layer.

The specific implementation of the decoding end is described as follows:

when the skip Layer is equal to 1, the geometric coordinate is the coordinate of the lower left front corner of the cube, as shown in fig. 3.Q point position;

when the skip Layer is greater than 1, the geometric coordinate is the coordinate of the center position of the cube, such as the position of a P point in FIG. 3;

the technique is currently in the standard appendix C.

C.3 decoded position shift procedure

When mingeotnesizelog 2 is greater than 1, the process operates as follows for each slice of the current point cloud image in the octree-based geometric information decoding framework:

mask＝(-1)<<MinGeomNodeSizeLog2

for(i＝0；i<PointCount；i++){

PointPos[i][0]＝(PointPos[i][0]&mask)+(1<<(MinGeomNodeSizeLog2-1))

PointPos[i][1]＝(PointPos[i][1]&mask)+(1<<(MinGeomNodeSizeLog2-1))

PointPos[i][2]＝(PointPos[i][2]&mask)+(1<<(MinGeomNodeSizeLog2-1))

}

MinGeomNodeSizeLog2 is the minimum node side length of the current octree, and the value is equal to skip Layer;

PointCount is the total number of nodes when the current slice is decoded to the level of MinGeomNodeSizeLog 2;

PointPos [ i ] [0] is the x-axis coordinate of the current node reconstruction geometric point;

PointPos [ i ] [1] is the y-axis coordinate of the current node reconstruction geometric point;

PointPos [ i ] [2] is the z-axis coordinate of the current node reconstruction geometric point;

mask is the intermediate mask;

the initial value of PointPos [ i ] [0] PointPos [ i ] [1] PointPos [ i ] [2] is the coordinates of the left/front/lower corner of the cube with the current side length of MinGeomNodeSizeLog2 node, as shown in FIG. 3.Q, and through the shift operation of the codes, the value of PointPos [ i ] [0] PointPos [ i ] [1] PointPos [ i ] [2] is equal to the coordinates of the center position of the cube of the current node, as shown in FIG. 3.P point.

Currently, G-PCC measures the process geometry error using the following two methods.

(1) Using the point-to-point distance representation, the point-to-point geometric error measure calculation process is shown in FIG. 5, which is a black dot (b)_i) Red point (a) as the point generated after point cloud expansion coding and decoding_j) The point in the original point cloud with the closest distance. The difference between the coordinates of the black dot and the red dot (E (i, j) ═ b)_i-a_j) Is a point-to-point error vector. The length of the error vector is a point-to-point geometric error, namely:b is a sparse point cloud after expansion coding, A is an original point cloud, and the point-to-point distance of B belongs to all points iWith N_BFor the number of points in point cloud B, the point-to-point error D1 for the entire point cloud is defined as:

(2) using the point-to-plane distance representation, error vector E (i, j) is oriented normal to N_jProjecting to obtain a new error vectorThus, the point-to-plane error is calculated as:

in the above technology, only part of the geometric bitstream information is decoded, and all points in the node range are represented by one geometric point in the node space with the side length of the skip Layer, so that the geometric reconstruction process after the spatial scalability is performed is a lossy process.

Disclosure of Invention

Technical problem to be solved

The method aims to solve the problems that in the existing geometric reconstruction process after space scalability, the difference of distribution conditions of internal points of different nodes is not considered, and the geometric error in the geometric reconstruction process after space scalability is large. The invention provides a point cloud space scalable coding geometric reconstruction method based on plane fitting center coordinate projection.

Technical scheme

A point cloud space scalable coding geometric reconstruction method based on plane fitting center coordinate projection is characterized in that: and fitting a local plane of the current node by using the information of the neighbor nodes through a local space formed by each node and the surrounding 26 neighbor nodes, and then solving the coordinates of the projection point of the central point of the current node on the plane.

The further technical scheme of the invention is as follows: the surrounding 26 neighbor nodes comprise 6 coplanar neighbor nodes, 12 common-edge neighbor nodes and 8 common-point neighbor nodes.

The further technical scheme of the invention is as follows: the fitting of the local plane of the current node by using the neighbor node information specifically comprises the following steps:

1) k neighbor node plane fitting judgment

Retrieving 26 neighbor situations of the current neighbor node, and setting the number of neighbors existing in the 26 neighbors as: neighNum;

when the neighbor num is larger than or equal to K, calculating K neighbor nodes closest to the node center position coordinate by using the neighbor num nodes existing in 26 neighbor nodes around the node, and calculating the neighbor node center position coordinate by using the P point coordinate, the octreeSize of the node side length of the octree and the position of the neighbor;

2) least square method plane fitting

The final equation for plane fitting by using the least square method plane fitting principle and the central position coordinates of N1-Nk is as follows:

a₀*x+a₁*y+a₂-z＝0

wherein x_i、y_i、z_iAnd the coordinates of the center position of the neighbor node.

The further technical scheme of the invention is as follows: the method is characterized in that the K value is equal to the number of searching neighbors of a point cloud image plane normal vector, and can be 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 and 16.

The further technical scheme of the invention is as follows: solving the coordinates of the projection point of the central point of the current node on the plane as follows:

wherein:

wherein: x is the number of₀、y₀、z₀The coordinates of the current geometric center position, namely the center position of the cube; x is the number of₁、y₁、z₁And fitting the geometric coordinates calculated by the geometric reconstruction method for the plane before constraint.

Advantageous effects

The point cloud space scalable coding geometric reconstruction method based on plane fitting center coordinate projection provided by the invention makes full use of the spatial correlation of the point cloud, so that the error of geometric reconstruction after spatial scalability is reduced. The geometric information PSNR represents: compared with the prior art, under the condition of the same code rate, the geometric error brought by the method provided by the invention is reduced (PSNR is a positive value) or increased (PSNR is a negative value) than that brought by the prior art.

Drawings

The drawings are only for purposes of illustrating particular embodiments and are not to be construed as limiting the invention, wherein like reference numerals are used to designate like parts throughout.

FIG. 1 is a schematic view of a spatial scalability;

FIG. 2G-PCC decoder framework diagram;

FIG. 3 a scalable post-coding geometric reconstruction method;

FIG. 4 is a schematic diagram of the location of the present invention in the frame of a point cloud G-PCC decoder;

FIG. 5 is a schematic of a point-to-point error;

FIG. 6 is a schematic diagram of the center positions of nodes and neighboring nodes;

FIG. 7 is a schematic diagram of a plane fitting geometric reconstruction method;

FIG. 8 is a schematic diagram of a plane fitting geometric reconstruction method.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

The terms and expressions referred to in the present invention are used for the following explanations:

1) point Cloud Compression (PCC)

2) Geometry-based Point Cloud Compression (G-PCC)

3) Sheet (slice)

4) Bounding box (bounding box)

5) OctreeImage (octree)

6) Intra-frame prediction (intra prediction)

7) Triangular patch set (triangle soup, trisup)

8) Context-based Adaptive Binary Arithmetic Coding (CABAC)

9) Block (block)

10) Crossing point (vertex)

11) Level of Detail (Level of Detail, LOD)

12) Region Adaptive Hierarchical Transform (RAHT)

13) Skipping Layer (skip Layer)

14) Spatial Scalability (Spatial Scalability)

15) Moving Picture Experts Group (MPEG)

16) International Organization for Standardization (ISO)

17) International Electrotechnical Commission (IEC)

18) Log2 logarithm of Minimum geometric Node side length (Minimum Geometry Node Size Log2, MinGeomNodeSizeLog2)

19) Direct code Point (Direct Point Count )

The invention provides a brand-new point cloud space scalable coding geometric reconstruction method based on plane fitting center coordinate projection, which is characterized in that a local plane of a current node is fitted by using neighbor node information through a local space formed by each node and 26 surrounding neighbor nodes, and then the coordinates of projection points of the center point of the current node on the plane are solved. The specific process is as follows:

N_jis a point a_jThe normal vector of (a) is a normal vector using a point a_jAnd neighbor points (K points in total) of the local plane are fitted with the normal vector of the local plane, and after scalable is carried out, each specific position information of the point cloud cannot be obtained at a decoding end, and the accurate local plane and plane normal vector cannot be calculated, so that neighbor occupation information of the current node is adopted to carry out approximate fitting on the local plane, and the local plane equation is calculated.

K neighbor node plane fitting judgment

Retrieving 26 neighbor situations of the current neighbor node, and setting the number of neighbors existing in the 26 neighbors as: neighNum;

when neighNum is larger than or equal to K, K neighbor nodes closest to the node center position coordinate (point P in figure 6 is the node center position) are calculated by using neighNum nodes existing in 26 neighbor nodes around the node, and the neighbor node center position coordinate (point N1-Nk in figure 6 are the neighbor node center position coordinates) is calculated by using the point P coordinate, octreeSize node side length and neighbor positions.

2. Least square method plane fitting

And performing plane fitting by using the least square method plane fitting principle through the central position coordinates of N1-Nk. According to the least square method plane fitting principle and a general plane formula:

z＝a₀*x+a₁*y+a₂

the least square method shows that:

S＝min∑[(a₀*x_i+a₁*y_i+a₂)-z_i]²

for the above formula, respectively take a₀,a₁,a₂Partial derivatives of (a):

and then, converting the formula into a matrix form after shifting:

then, a is obtained by the Kramer rule according to the formula₀,a₁,a₂A determinant expression of (a);

namely: the final equation for plane fitting by using the least square method plane fitting principle and the central position coordinates of N1-Nk is as follows:

a₀*x+a₁*y+a₂-z＝0

3. calculation of geometrical reconstruction coordinates after plane fitting

Then calculating nodes according to the fitted plane equationCenter position P (x)₀,y₀,z₀) Projected point P on the fitting plane₁(x₁,y₁,z₁). As shown in fig. 7.

The normal vector of the plane can be obtained by fitting the above formula to the general formula of the plane equationThus, P is known₁(x₁,y₁,z₁) The coordinates satisfy the following equation:

wherein:

the point P₁(x₁,y₁,z₁) Namely the geometric coordinates calculated by the plane fitting geometric reconstruction method.

4. Building geometric coordinate value constraints

Due to the point P₁(x₁,y₁,z₁) Must be inside the space of the node, i.e. the following constraints are satisfied:

setting point P (x)₀,y₀,z₀) The distance to the fitting plane is L;

then:

(1) when L is less than or equal to octreeSize/2, the constraint condition is satisfied, and a point P is₁(x₁,y₁,z₁) Inside the node space.

(2) When L is>octreeSize/2, point P appears₁(x₁，y₁，z₁) Appearing outside the node space, the plane fitting geometric coordinates at this time are corrected as follows.

As shown in FIG. 8, a point P is set₂(x₂,y₂,z₂) Is PP₁And connecting a point on the line, wherein the point is the closest point to the fitting plane in the node space.

Therefore, point P₂(x₂,y₂,z₂) Satisfies the following conditions:

namely:

point P₂(x₂,y₂,z₂) I.e. at L>And during octreeSize/2, fitting the corrected plane with the geometric coordinates of the geometric reconstruction method. In the following, the case where the skip Layer is 3 different from the skip Layer is compared.

D1-PSNR is calculated as follows:

D2-PSNR is calculated as follows:

(where p is the peak constant value for each reference point cloud defined in the table, determined from the point cloud sequence, as shown in Table 2. bold.)

TABLE 1 PSNR for geometric information

TABLE 2 constant value of point cloud sequence peak

While the invention has been described with reference to specific embodiments, the invention is not limited thereto, and various equivalent modifications or substitutions can be easily made by those skilled in the art within the technical scope of the present disclosure.