CN107135397B - A kind of panorama video code method and apparatus - Google Patents

Abstract

The present application discloses a panoramic video encoding method and device. This method converts the plane coordinates of a reference point in the current prediction unit in the current prediction unit of the current frame, the nearest neighbor prediction unit of the current frame, the current prediction unit of the reference frame, and the nearest neighbor prediction unit of the reference frame into a spherical surface coordinates, so as to determine the three-dimensional motion vector of the current prediction unit in the three-dimensional space according to the spherical coordinates of the reference point. Furthermore, according to the three-dimensional motion vector, determine the corresponding target reference points of all the target points of the current prediction unit in the current frame in the reference frame, and compensate the target point pixels according to the pixels of the target reference points. Compared with the prior art, the present invention uses a spherical coordinate system to convert a motion model to perform pixel compensation, thereby eliminating the influence of distortion in video, thereby improving video coding efficiency.

Description

A panoramic video encoding method and device

technical field

The present application relates to the field of video coding, and more specifically, to a panoramic video coding method and device.

Background technique

In recent years, virtual reality (VR) has become a hot spot in the field of research and application. The core of VR applications is the panoramic video. When the panoramic video is actually played, it is played on the VR head-mounted display device. Its spatial structure is a three-dimensional spherical surface, but it must be mapped to a plane when it is actually stored on the computer. , the most widely used is the longitude-latitude map mapping method, which maps the spherical surface into a rectangle with a length and width of 2:1, but it will cause serious image shape distortion, which will bring trouble to the motion estimation during encoding.

Traditional general-purpose video coding standards such as H.264 (Advanced Video Coding) and HEVC (High Efficiency Video Coding) both use a translational motion model. This motion model can only describe the translation of objects, but cannot describe rotation and scaling. and other complex movements. However, the panoramic video is mapped to the plane due to distortion, which leads to the problem of low coding efficiency in the translational motion model.

Contents of the invention

In view of this, the present application provides a panoramic video coding method and device, which perform motion estimation and compensation on a panoramic video based on a spherical coordinate system, so as to eliminate the influence of distortion in the video, thereby improving video coding efficiency.

In order to achieve the above purpose, the proposed scheme is as follows:

A panoramic video encoding method, comprising:

determining the most adjacent PU of the current PU;

Acquiring the first plane coordinates, the second plane coordinates, the third plane coordinates and the fourth plane coordinates of the reference point;

Wherein, the first plane coordinates are the coordinates of the reference point in the current prediction unit of the current frame, and the second plane coordinates are the coordinates of the reference point in the most adjacent prediction unit of the current frame, and the The third plane coordinates are the coordinates of the reference point in the current prediction unit of the reference frame, and the fourth plane coordinates are the coordinates of the reference point in the most adjacent prediction unit of the reference frame;

converting the first plane coordinates, the second plane coordinates, the third plane coordinates, and the fourth plane coordinates into spherical coordinates, and calculating the current prediction unit in the three-dimensional space according to the spherical coordinates The three-dimensional motion vector;

For all target points in the current prediction unit of the current frame, determine a target reference point corresponding to the target point in the reference frame according to the three-dimensional motion vector;

Using the pixels of the target reference point to compensate the pixels of the target point, so as to realize the spherical coordinate transformation motion model compensation of the current prediction unit;

Encode the current prediction unit.

Preferably, the determining the most adjacent prediction unit of the current prediction unit includes:

Acquiring a two-dimensional motion vector of the current prediction unit and two-dimensional motion vectors of multiple adjacent prediction units;

comparing the two-dimensional motion vector of the current prediction unit with the two-dimensional motion vector of the adjacent prediction unit;

Taking the neighboring prediction unit closest to the two-dimensional motion vector of the current prediction unit as the nearest neighboring prediction unit of the current prediction unit.

Preferably, said obtaining the first plane coordinates, the second plane coordinates, the third plane coordinates and the fourth plane coordinates of the reference point includes:

acquiring the first plane coordinates and the second plane coordinates of the reference point;

calculating a two-dimensional motion vector difference between the current prediction unit and the nearest neighbor prediction unit;

The third plane coordinate is determined according to the two-dimensional motion vector difference and the first plane coordinate, and the fourth plane coordinate is determined according to the two-dimensional motion vector difference and the second plane coordinate.

Preferably, converting the first plane coordinates, the second plane coordinates, the third plane coordinates and the fourth plane coordinates into spherical coordinates respectively includes:

formula based z=Rcosθ calculates the spherical coordinates;

where θ represents the latitude in spherical coordinates, and denote the longitude in spherical coordinates, and

(u, v) represent the plane coordinates of the reference point, H represents the height of the image, W represents the width of the image, and R represents the depth of the reference point.

Preferably, the calculating the three-dimensional motion vector of the current prediction unit in three-dimensional space according to the spherical coordinates includes:

calculating a first three-dimensional motion vector according to the first spherical coordinates corresponding to the first plane coordinates and the third spherical coordinates corresponding to the third plane coordinates;

calculating a second three-dimensional motion vector according to the second spherical coordinates corresponding to the second plane coordinates and the fourth spherical coordinates corresponding to the fourth plane coordinates;

Using the equality of the first 3D motion vector and the second 3D motion vector and the preset depth of the reference point in the current prediction unit of the reference frame to respectively determine the depth of the reference point in the current prediction unit of the current frame, in the depth in the nearest PU of the current frame and the depth in the nearest PU of the reference frame;

calculating a three-dimensional motion vector of the current prediction unit in three-dimensional space according to a preset depth of the reference point in the current prediction unit of the reference frame and a depth of the reference point in the current prediction unit of the current frame;

Or calculate the 3D motion vector of the current prediction unit in the 3D space according to the depth of the reference point in the most adjacent PU of the reference frame and the depth of the reference point in the most adjacent PU of the current frame.

Preferably, said encoding said current prediction unit includes;

Calculating the first cost function value of the current prediction unit after spherical coordinate transformation motion model compensation;

Calculating the second cost function value of the current prediction unit after translation model compensation;

Comparing the first cost function value and the second cost function value, and selecting a compensation mode for the current prediction unit according to the comparison result.

A panoramic video encoding device, comprising:

screening unit, used to determine the most adjacent prediction unit of the current prediction unit;

a coordinate acquisition unit, configured to acquire the first plane coordinates, the second plane coordinates, the third plane coordinates and the fourth plane coordinates of the reference point;

Wherein, the first plane coordinates are the coordinates of the reference point in the current prediction unit of the current frame, and the second plane coordinates are the coordinates of the reference point in the most adjacent prediction unit of the current frame, and the The third plane coordinates are the coordinates of the reference point in the current prediction unit of the reference frame, and the fourth plane coordinates are the coordinates of the reference point in the most adjacent prediction unit of the reference frame;

a coordinate conversion unit, configured to respectively convert the first plane coordinates, the second plane coordinates, the third plane coordinates, and the fourth plane coordinates into spherical coordinates;

A three-dimensional motion vector calculation unit, configured to calculate the three-dimensional motion vector of the current prediction unit in three-dimensional space according to the spherical coordinates;

A target reference point determination unit, configured to, for all target points in the current prediction unit of the current frame, determine the target reference point corresponding to the target point in the reference frame according to the three-dimensional motion vector;

A compensation unit, configured to use the pixels of the target reference point to compensate the pixels of the target point, so as to realize the compensation of the spherical coordinate transformation motion model of the current prediction unit;

A coding unit, configured to code the current prediction unit.

Preferably, the screening unit includes:

A two-dimensional motion vector search module, configured to obtain the two-dimensional motion vector of the current prediction unit and the two-dimensional motion vectors of multiple adjacent prediction units;

A comparison module, configured to compare the two-dimensional motion vector of the current prediction unit with the two-dimensional motion vector of the adjacent prediction unit;

The prediction unit selection module is configured to use the adjacent prediction unit closest to the two-dimensional motion vector of the current prediction unit as the most adjacent prediction unit of the current prediction unit.

Preferably, the coordinate acquisition unit includes:

a coordinate acquisition module, configured to acquire the first plane coordinates and the second plane coordinates of the reference point;

A first calculation module, configured to calculate a two-dimensional motion vector difference between the current prediction unit and the nearest neighbor prediction unit;

A second calculation module, configured to determine the third plane coordinates according to the two-dimensional motion vector difference and the first plane coordinates, and determine the third plane coordinates according to the two-dimensional motion vector difference and the second plane coordinates Describe the fourth plane coordinates.

Preferably, the coordinate conversion unit is specifically used for:

formula based z=Rcosθ calculates the spherical coordinates;

where θ represents the latitude in spherical coordinates, and denote the longitude in spherical coordinates, and

(u, v) represent the plane coordinates of the reference point, H represents the height of the image, W represents the width of the image, and R represents the depth of the reference point.

Preferably, the three-dimensional motion vector calculation unit includes:

A third calculation module, configured to calculate a first three-dimensional motion vector according to the first spherical coordinates corresponding to the first plane coordinates and the third spherical coordinates corresponding to the third plane coordinates;

A fourth calculation module, configured to calculate a second three-dimensional motion vector according to the second spherical coordinates corresponding to the second plane coordinates and the fourth spherical coordinates corresponding to the fourth plane coordinates;

The fifth calculation module is used to respectively determine the current position of the reference point in the current prediction unit of the reference frame by using the equality of the first three-dimensional motion vector and the second three-dimensional motion vector and the preset depth of the reference point in the current prediction unit of the reference frame. the depth in the PU, the depth in the nearest PU of the current frame, and the depth in the nearest PU of the reference frame;

The sixth calculation module is used to calculate the three-dimensional motion vector of the current prediction unit in the three-dimensional space according to the preset depth of the reference point in the current prediction unit of the reference frame and the depth of the reference point in the current prediction unit of the current frame;

Or calculate the 3D motion vector of the current prediction unit in the 3D space according to the depth of the reference point in the most adjacent PU of the reference frame and the depth of the reference point in the most adjacent PU of the current frame.

Preferably, the encoding unit includes;

The first cost function calculation module is used to calculate the first cost function value of the current prediction unit after being compensated by the spherical coordinate transformation motion model;

The second cost function calculation module is used to calculate the second cost function value of the current prediction unit after the compensation of the translation model;

A comparing module, configured to compare the first cost function value and the second cost function value, and select a compensation mode for the current prediction unit according to the comparison result.

It can be known from the above technical solution that the present application discloses a panoramic video coding method and device. This method converts the plane coordinates of a reference point in the current prediction unit in the current prediction unit of the current frame, the nearest neighbor prediction unit of the current frame, the current prediction unit of the reference frame, and the nearest neighbor prediction unit of the reference frame into a spherical surface coordinates, so as to determine the three-dimensional motion vector of the current prediction unit in the three-dimensional space according to the spherical coordinates of the reference point. Furthermore, according to the three-dimensional motion vector, determine the corresponding target reference points of all the target points of the current prediction unit in the current frame in the reference frame, and compensate the target point pixels according to the pixels of the target reference points. Compared with the prior art, the present invention uses a spherical coordinate system to convert a motion model to perform pixel compensation, thereby eliminating the influence of distortion in video, thereby improving video coding efficiency.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only It is an embodiment of the present invention, and those skilled in the art can also obtain other drawings according to the provided drawings without creative work.

FIG. 1 shows a schematic flow diagram of a panoramic video encoding method disclosed in an embodiment of the present invention;

Fig. 2 shows a position map of the nearest neighbor prediction unit disclosed by another embodiment of the present invention;

Fig. 3 shows a schematic diagram of a spherical model;

Fig. 4 shows a schematic structural diagram of a panoramic video encoding device disclosed in another embodiment of the present invention.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

Referring to FIG. 1 , a schematic flowchart of a panoramic video encoding method disclosed by an embodiment of the present invention is shown.

In this embodiment, the method includes:

As can be seen from Figure 1, the method includes:

S101: Determine the most adjacent prediction unit of the current prediction unit.

For a prediction unit, search is first performed in its domain, and the search positions are a ₀ , a ₁ , b ₀ , b ₁ and b ₂ as shown in FIG. 2 .

First, the two-dimensional motion vector of the current prediction unit is compared with the two-dimensional motion vectors of the prediction units where the above five candidate positions are located. Determine the candidate position with the closest distance to the two-dimensional motion vector, and use the prediction unit where the candidate position is located as the most adjacent prediction unit of the current prediction unit.

S102: Obtain first plane coordinates, second plane coordinates, third plane coordinates, and fourth plane coordinates of the reference point.

The current frame contains 2 prediction units (the current prediction unit and the most adjacent prediction unit), and the coordinates of the reference point (usually the center point of the prediction unit is selected as the reference point) in the current prediction unit of the current frame is the first plane coordinate c ₁ =(u _1c ,v _1c ), the coordinates of the reference point in the nearest neighbor unit of the current frame are the second plane coordinates c ₂ =(u _2c ,v _2c ), the coordinates of the reference point in the current prediction unit of the reference frame is the third plane coordinate a ₁ =(u ₁ , v ₁ ), and the coordinate of the reference point in the most adjacent prediction unit of the reference frame is the fourth plane coordinate a ₁ =(u ₁ , v ₁ ).

S103: Transform the first plane coordinates, the second plane coordinates, the third plane coordinates, and the fourth plane coordinates into spherical coordinates, respectively.

formula based z=Rcosθ calculates the spherical coordinates.

Referring to Fig. 3, a schematic diagram of a spherical model is shown. where θ represents the latitude in spherical coordinates, and denote the longitude in spherical coordinates, and

(u, v) represent the plane coordinates of the reference point, H represents the height of the image, W represents the width of the image, and R represents the depth of the reference point.

We can easily calculate the latitude θ and longitude of each point by using the longitude-latitude map projection formula The depth of the four points is unknown, we assume r _a1 , r _a2 , r _c1 and r _c2 , then the coordinates of the four points in three-dimensional space can be obtained by using spherical coordinate transformation as follows.

S104: Calculate a three-dimensional motion vector of the current prediction unit in three-dimensional space according to the spherical coordinates.

In this embodiment, we set: a. All points in one prediction unit share one depth. b. If the two-dimensional motion vectors of adjacent prediction units are similar, they should correspond to the same object. c. The movement of points on the same object relative to the camera is also the same.

Based on the above three settings, it can be calculated that two three-dimensional motion vectors can be calculated and the two three-dimensional motion vectors are equal:

In this way, the following equations can be obtained

r _a1 cosθ _a1 -r _c1 cosθ _c1 =r _a2 cosθ _a2 -r _c2 cosθ _c2

Here we stipulate r _a1 = 1 to write the above equations in matrix form:

but:

r = Φ ^-1 a

It can be seen that what we estimate is the relative depth of the current PU rather than the absolute depth, but the relative

Depth is good enough for motion compensation. When Φ is irreversible, set r _a2 =r _c1 =r _c2 =1.

After estimating all the depths, the MV of the current prediction unit in the three-dimensional space can be calculated

value

Δz＝r _c1 cosθ _c1 -r _a1 cosθ _a1

S105: For all target points in the current prediction unit of the current frame, determine target reference points corresponding to the target points in the reference frame according to the 3D motion vector.

For any target point p in the current frame, use spherical coordinate transformation to obtain p ^3D in three-dimensional space,

Utilize p ^3D Plus You can get the coordinates of the target reference point for point p on the reference frame

S106: Use the pixels of the target reference point to compensate the pixels of the target point, so as to realize the spherical coordinate transformation motion model compensation of the current prediction unit.

right Using the inverse transformation of spherical coordinates, we can get the pixel p _org = (u _org ,v _org ) in the reference frame:

Compensate the current point p with the obtained pixel value of the pixel point p _org in the reference frame. Here, if the coordinate (u org,vorg) of p(org) is not an integer, then we interpolate on the reference frame to get the pixel value of p(org).

S107: Encode the current prediction unit.

Specifically, calculate the first cost function value of the current prediction unit after being compensated by the spherical coordinate transformation motion model;

Calculating the second cost function value of the current prediction unit after translation model compensation;

Comparing the first cost function value and the second cost function value, and selecting a compensation mode for the current prediction unit according to the comparison result.

It can be known from the above embodiments that the present application discloses a panoramic video coding method. This method converts the plane coordinates of a reference point in the current prediction unit in the current prediction unit of the current frame, the nearest neighbor prediction unit of the current frame, the current prediction unit of the reference frame, and the nearest neighbor prediction unit of the reference frame into a spherical surface coordinates, so as to determine the three-dimensional motion vector of the current prediction unit in the three-dimensional space according to the spherical coordinates of the reference point. Furthermore, according to the three-dimensional motion vector, determine the corresponding target reference points of all the target points of the current prediction unit in the current frame in the reference frame, and compensate the target point pixels according to the pixels of the target reference points. Compared with the prior art, the present invention uses a spherical coordinate system to convert a motion model to perform pixel compensation, thereby eliminating the influence of distortion in video, thereby improving video coding efficiency.

It should be noted that, in the above embodiment, in order to reduce the computational complexity, when determining the two-dimensional motion vectors of the current prediction unit and the most adjacent prediction unit in step S101, our two-dimensional motion vector search is divided into two times, The translation model is used for motion compensation in the first two-dimensional motion vector search, and the spherical coordinate-based model is used in the second time, but in the second search, we use the motion vector of the first search as the basic two-dimensional motion Vector, which narrows the scope of the second search to a 3×3 pixel area. In addition, if the basic two-dimensional motion vector of the first search satisfies the following conditions:

|Δu|+|Δv|≤4

Just do a fractional pixel search on the second search. One thing to note is that the MV precision we use is 1/4 pixel precision, but we can expand it to arbitrary precision when we do spherical coordinate transformation.

In addition, in other embodiments disclosed in the present invention, a special mode is called merge mode (MergeMode), that is, the current prediction unit uses the two-dimensional motion vector of the position of the merge candidate (Merge Candidate), and the position of the merge candidate in the spatial domain is also In the five positions in Figure 2, in order to prevent the adjacent prediction unit selected by the prediction unit using the merge mode from always being a merge candidate in our model, a new merge mode is adopted:

As shown in Figure 2, if the current PU uses a ₁ as its best merging candidate, it means that the 2D motion vector of the PU is equal to MV1. This leads to the fact that the prediction unit where a1 is located must become the most adjacent prediction unit.

In order to avoid this situation, we do not directly set MV equal to MV1, but search for the MV with the closest Euclidean distance to MV1 in other merging candidate positions, such as MV3, then use MV1 and MV3 to determine the current prediction unit. Two-dimensional motion vector, set to:

Referring to FIG. 4 , a schematic structural diagram of a panoramic video encoding device disclosed in another embodiment of the present invention is shown.

As can be seen from FIG. 4 , the device includes: a screening unit 1 , a coordinate acquisition unit 2 , a coordinate conversion unit 3 , a three-dimensional motion vector calculation unit 4 , a target reference point determination unit 5 , a compensation unit 6 and an encoding unit 7 .

Wherein, the screening unit 1 is used to determine the most adjacent prediction unit of the current prediction unit.

The coordinate acquisition unit 2 is used to acquire the first plane coordinates, the second plane coordinates, the third plane coordinates and the fourth plane coordinates of the reference point.

Wherein, the first plane coordinates are the coordinates of the reference point in the current prediction unit of the current frame, and the second plane coordinates are the coordinates of the reference point in the most adjacent prediction unit of the current frame, and the The third plane coordinates are the coordinates of the reference point in the current prediction unit of the reference frame, and the fourth plane coordinates are the coordinates of the reference point in the most adjacent prediction unit of the reference frame;

The coordinate conversion unit 3 is used to respectively convert the first plane coordinates, the second plane coordinates, the third plane coordinates and the fourth plane coordinates into spherical coordinates.

The 3D motion vector calculation unit 4 is configured to calculate a 3D motion vector of the current prediction unit in a 3D space according to the spherical coordinates.

The target reference point determining unit 5 is used for determining, for all target points in the current prediction unit of the current frame, the target reference points corresponding to the target points in the reference frame according to the 3D motion vector.

The compensation unit 6 is used to use the pixels of the target reference point to compensate the pixels of the target point, so as to realize the compensation of the spherical coordinate transformation motion model of the current prediction unit.

The encoding unit 7 is configured to encode the current prediction unit.

Specifically, the screening unit includes:

A two-dimensional motion vector search module, configured to obtain the two-dimensional motion vector of the current prediction unit and the two-dimensional motion vectors of multiple adjacent prediction units.

A comparing module, configured to compare the two-dimensional motion vector of the current prediction unit with the two-dimensional motion vector of the adjacent prediction unit.

The prediction unit selection module is configured to use the adjacent prediction unit closest to the two-dimensional motion vector of the current prediction unit as the most adjacent prediction unit of the current prediction unit.

The coordinate acquisition unit includes:

A coordinate acquisition module, configured to acquire the first plane coordinates and the second plane coordinates of the reference point.

A first calculation module, configured to calculate a two-dimensional motion vector difference between the current prediction unit and the nearest neighbor prediction unit.

A second calculation module, configured to determine the third plane coordinates according to the two-dimensional motion vector difference and the first plane coordinates, and determine the third plane coordinates according to the two-dimensional motion vector difference and the second plane coordinates Describe the coordinates of the fourth plane.

The coordinate conversion unit is specifically used for:

formula based z=Rcosθ calculates the spherical coordinates;

where θ represents the latitude in spherical coordinates, and denote the longitude in spherical coordinates, and

(u, v) represent the plane coordinates of the reference point, H represents the height of the image, W represents the width of the image, and R represents the depth of the reference point.

The three-dimensional motion vector calculation unit includes:

A third calculation module, configured to calculate a first three-dimensional motion vector according to the first spherical coordinates corresponding to the first plane coordinates and the third spherical coordinates corresponding to the third plane coordinates.

A fourth calculation module, configured to calculate a second three-dimensional motion vector according to the second spherical coordinates corresponding to the second plane coordinates and the fourth spherical coordinates corresponding to the fourth plane coordinates.

The fifth calculation module is used to respectively determine the current position of the reference point in the current prediction unit of the reference frame by using the equality of the first three-dimensional motion vector and the second three-dimensional motion vector and the preset depth of the reference point in the current prediction unit of the reference frame. The depth in the PU, the depth in the nearest PU of the current frame, and the depth in the nearest PU of the reference frame.

The sixth calculation module is used to calculate the three-dimensional motion vector of the current prediction unit in the three-dimensional space according to the preset depth of the reference point in the current prediction unit of the reference frame and the depth of the reference point in the current prediction unit of the current frame; Or calculate the 3D motion vector of the current prediction unit in the 3D space according to the depth of the reference point in the most adjacent PU of the reference frame and the depth of the reference point in the most adjacent PU of the current frame.

The encoding unit includes;

The first cost function calculation module is used to calculate the first cost function value of the current prediction unit after being compensated by the spherical coordinate transformation motion model;

The second cost function calculation module is used to calculate the second cost function value of the current prediction unit after the compensation of the translation model;

A comparing module, configured to compare the first cost function value and the second cost function value, and select a compensation mode for the current prediction unit according to the comparison result.

It should be noted that the device embodiment corresponds to the method embodiment, and its execution process and execution principle are the same, so details are not repeated here.

Finally, it should also be noted that in this text, relational terms such as first and second etc. are only used to distinguish one entity or operation from another, and do not necessarily require or imply that these entities or operations, any such actual relationship or order exists. Furthermore, the term "comprises", "comprises" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method, article or apparatus comprising a set of elements includes not only those elements, but also includes elements not expressly listed. other elements of or also include elements inherent in such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article or apparatus comprising said element.

Each embodiment in this specification is described in a progressive manner, each embodiment focuses on the difference from other embodiments, and the same and similar parts of each embodiment can be referred to each other.

The above description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention will not be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (10)

1. A panoramic video encoding method, comprising: determining the most adjacent PU of the current PU; Acquiring the first plane coordinates, the second plane coordinates, the third plane coordinates and the fourth plane coordinates of the reference point; Wherein, the first plane coordinates are the coordinates of the reference point in the current prediction unit of the current frame, and the second plane coordinates are the coordinates of the reference point in the most adjacent prediction unit of the current frame, and the The third plane coordinates are the coordinates of the reference point in the current prediction unit of the reference frame, and the fourth plane coordinates are the coordinates of the reference point in the most adjacent prediction unit of the reference frame; converting the first plane coordinates, the second plane coordinates, the third plane coordinates, and the fourth plane coordinates into spherical coordinates, and calculating the current prediction unit in the three-dimensional space according to the spherical coordinates The three-dimensional motion vector; For all target points in the current prediction unit of the current frame, determine a target reference point corresponding to the target point in the reference frame according to the three-dimensional motion vector; Using the pixels of the target reference point to compensate the pixels of the target point, so as to realize the spherical coordinate transformation motion model compensation of the current prediction unit; encoding the current prediction unit; Wherein, the determining the most adjacent prediction unit of the current prediction unit includes: Acquiring a two-dimensional motion vector of the current prediction unit and two-dimensional motion vectors of multiple adjacent prediction units; comparing the two-dimensional motion vector of the current prediction unit with the two-dimensional motion vector of the adjacent prediction unit; Taking the neighboring prediction unit closest to the two-dimensional motion vector of the current prediction unit as the nearest neighboring prediction unit of the current prediction unit. 2. The method according to claim 1, wherein said obtaining the first plane coordinates, the second plane coordinates, the third plane coordinates and the fourth plane coordinates of the reference point comprises: acquiring the first plane coordinates and the second plane coordinates of the reference point; calculating a two-dimensional motion vector difference between the current prediction unit and the nearest neighbor prediction unit; The third plane coordinate is determined according to the two-dimensional motion vector difference and the first plane coordinate, and the fourth plane coordinate is determined according to the two-dimensional motion vector difference and the second plane coordinate. 3. The method according to claim 1, wherein said converting said first plane coordinates, said second plane coordinates, said third plane coordinates and said fourth plane coordinates into spherical coordinates respectively ,include: formula based z=Rcosθ calculates the spherical coordinates; where θ represents the latitude in spherical coordinates, and denote the longitude in spherical coordinates, and (u, v) represent the plane coordinates of the reference point, H represents the height of the image, W represents the width of the image, and R represents the depth of the reference point. 4. The method according to claim 3, wherein the calculating the three-dimensional motion vector of the current prediction unit in three-dimensional space according to the spherical coordinates comprises: calculating a first three-dimensional motion vector according to the first spherical coordinates corresponding to the first plane coordinates and the third spherical coordinates corresponding to the third plane coordinates; calculating a second three-dimensional motion vector according to the second spherical coordinates corresponding to the second plane coordinates and the fourth spherical coordinates corresponding to the fourth plane coordinates; Using the equality of the first 3D motion vector and the second 3D motion vector and the preset depth of the reference point in the current prediction unit of the reference frame to respectively determine the depth of the reference point in the current prediction unit of the current frame, in the depth in the nearest PU of the current frame and the depth in the nearest PU of the reference frame; calculating a three-dimensional motion vector of the current prediction unit in three-dimensional space according to a preset depth of the reference point in the current prediction unit of the reference frame and a depth of the reference point in the current prediction unit of the current frame; Or calculate the 3D motion vector of the current prediction unit in the 3D space according to the depth of the reference point in the most adjacent PU of the reference frame and the depth of the reference point in the most adjacent PU of the current frame. 5. The method according to claim 1, wherein said encoding said current prediction unit comprises; Calculating the first cost function value of the current prediction unit after spherical coordinate transformation motion model compensation; Calculating the second cost function value of the current prediction unit after translation model compensation; Comparing the first cost function value and the second cost function value, and selecting a compensation mode for the current prediction unit according to the comparison result. 6. A panoramic video encoding device, comprising: screening unit, used to determine the most adjacent prediction unit of the current prediction unit; a coordinate acquisition unit, configured to acquire the first plane coordinates, the second plane coordinates, the third plane coordinates and the fourth plane coordinates of the reference point; Wherein, the first plane coordinates are the coordinates of the reference point in the current prediction unit of the current frame, and the second plane coordinates are the coordinates of the reference point in the most adjacent prediction unit of the current frame, and the The third plane coordinates are the coordinates of the reference point in the current prediction unit of the reference frame, and the fourth plane coordinates are the coordinates of the reference point in the most adjacent prediction unit of the reference frame; the coordinate conversion unit is configured to respectively converting the first plane coordinates, the second plane coordinates, the third plane coordinates, and the fourth plane coordinates into spherical coordinates; A three-dimensional motion vector calculation unit, configured to calculate the three-dimensional motion vector of the current prediction unit in three-dimensional space according to the spherical coordinates; A target reference point determination unit, configured to, for all target points in the current prediction unit of the current frame, determine the target reference point corresponding to the target point in the reference frame according to the three-dimensional motion vector; A compensation unit, configured to use the pixels of the target reference point to compensate the pixels of the target point, so as to realize the compensation of the spherical coordinate transformation motion model of the current prediction unit; a coding unit, configured to code the current prediction unit; Wherein, the screening unit includes: A two-dimensional motion vector search module, configured to obtain the two-dimensional motion vector of the current prediction unit and the two-dimensional motion vectors of multiple adjacent prediction units; A comparison module, configured to compare the two-dimensional motion vector of the current prediction unit with the two-dimensional motion vector of the adjacent prediction unit; The prediction unit selection module is configured to use the adjacent prediction unit closest to the two-dimensional motion vector of the current prediction unit as the most adjacent prediction unit of the current prediction unit. 7. The device according to claim 6, wherein the coordinate acquisition unit comprises: a coordinate acquisition module, configured to acquire the first plane coordinates and the second plane coordinates of the reference point; A first calculation module, configured to calculate a two-dimensional motion vector difference between the current prediction unit and the nearest neighbor prediction unit; A second calculation module, configured to determine the third plane coordinates according to the two-dimensional motion vector difference and the first plane coordinates, and determine the third plane coordinates according to the two-dimensional motion vector difference and the second plane coordinates Describe the coordinates of the fourth plane. 8. The device according to claim 6, wherein the coordinate conversion unit is specifically used for: formula based z=Rcosθ calculates the spherical coordinates; where θ represents the latitude in spherical coordinates, and denote the longitude in spherical coordinates, and (u, v) represent the plane coordinates of the reference point, H represents the height of the image, W represents the width of the image, and R represents the depth of the reference point. 9. The device according to claim 6, wherein the three-dimensional motion vector calculation unit comprises: A third calculation module, configured to calculate a first three-dimensional motion vector according to the first spherical coordinates corresponding to the first plane coordinates and the third spherical coordinates corresponding to the third plane coordinates; A fourth calculation module, configured to calculate a second three-dimensional motion vector according to the second spherical coordinates corresponding to the second plane coordinates and the fourth spherical coordinates corresponding to the fourth plane coordinates; The fifth calculation module is used to respectively determine the current position of the reference point in the current prediction unit of the reference frame by using the equality of the first three-dimensional motion vector and the second three-dimensional motion vector and the preset depth of the reference point in the current prediction unit of the reference frame. the depth in the PU, the depth in the nearest PU of the current frame, and the depth in the nearest PU of the reference frame; The sixth calculation module is used to calculate the three-dimensional motion vector of the current prediction unit in the three-dimensional space according to the preset depth of the reference point in the current prediction unit of the reference frame and the depth of the reference point in the current prediction unit of the current frame; Or calculate the 3D motion vector of the current prediction unit in the 3D space according to the depth of the reference point in the most adjacent PU of the reference frame and the depth of the reference point in the most adjacent PU of the current frame. 10. The device according to claim 6, wherein the encoding unit comprises; The first cost function calculation module is used to calculate the first cost function value of the current prediction unit after being compensated by the spherical coordinate transformation motion model; The second cost function calculation module is used to calculate the second cost function value of the current prediction unit after the compensation of the translation model; A comparing module, configured to compare the first cost function value and the second cost function value, and select a compensation mode for the current prediction unit according to the comparison result.