Dynamic point cloud compression encoding unit optimization division and dynamic point cloud compression method

Abstract

The application discloses a coding unit optimization division method for dynamic point cloud compression, extracts attribute features and geometric features of multiple coding units, constructs a data set for training and testing multiple division identification models which correspond to different division modes respectively, calculates the gradient and variance of a current coding unit to preliminarily screen the division mode of the current coding unit, further extracts the attribute features and geometric features of the current coding unit, selects a trained division identification model to test the screened division mode one by one, screens out an optimal division mode from the test results, and the optimal division mode is used for the division of the current coding unit. The application also discloses a dynamic point cloud fast compression method.

Description

Technical Field

[0001] This invention relates to the technical field of point cloud coding, specifically to a method for optimizing the partitioning of coding units and a method for dynamic point cloud compression. Background Technology

[0002] With the rapid development of 3D data acquisition technology, point cloud data has been widely used in fields such as autonomous driving, virtual reality, and augmented reality. Compared to traditional 2D image data, point cloud data can more comprehensively and accurately describe the shape, structure, and dynamic changes of 3D objects and scenes, thus providing a more immersive interactive experience. However, due to the high precision and rich detail of point cloud data, its data volume is extremely large, posing a significant challenge to storage and transmission. Therefore, how to efficiently compress point cloud data has become a research hotspot. To address this issue, MPEG proposed a video-based point cloud compression (V-PCC) scheme, such as... Figure 1 As shown, this scheme improves coding efficiency by projecting a 3D dynamic point cloud onto a 2D video frame and compressing the projected data using existing video coding technologies (such as H.265 / HEVC, H.266 / VVC, etc.). V-PCC relies on traditional video encoders, which are mainly designed for ordinary video data and fail to fully utilize the spatial structure characteristics of point cloud data, resulting in high computational complexity and affecting coding efficiency in real-time application scenarios.

[0003] In the V-PCC framework, the partitioning method of the coding unit (CU) directly affects the final compression efficiency and encoding quality. Currently, V-PCC defaults to using H.265 / HEVC as the basic encoder, where the CU partitioning employs a quadtree (QT) structure, determining the optimal block partitioning method by traversing different levels of partitioning patterns. With the evolution of video coding technology, the latest H.266 / VVC standard significantly improves compression efficiency compared to H.265 / HEVC. H.266 / VVC uses a more flexible multi-branch tree partitioning pattern, such as... Figure 2As shown, combined structures such as quadtrees (QT), binary trees (BT), and ternary trees (TT) diversify the CU partitioning methods, thereby improving coding efficiency. To find the optimal CU partition for the current frame, H.266 needs to traverse all possible partitioning scenarios and calculate the rate-distortion cost (RD-cost) of each partitioning scheme, ultimately selecting the partitioning pattern with the minimum RD-cost as the optimal solution. Although this method improves compression performance, it also significantly increases the coding computation, resulting in a long block partitioning process, which is detrimental to real-time coding applications. Therefore, to reduce the computational complexity of coding unit partitioning and improve coding efficiency, it is necessary to study an efficient and fast partitioning method that reduces unnecessary partitioning pattern searches while ensuring coding quality, thereby effectively reducing coding computation overhead and improving the real-time compression performance of dynamic point clouds. Summary of the Invention

[0004] This invention provides an optimized coding unit partitioning method and a dynamic point cloud compression method for dynamic point cloud compression, which solves the shortcomings of existing partitioning methods, such as significantly increasing the amount of coding computation, resulting in a long block partitioning process, which is not conducive to real-time coding applications.

[0005] This invention can be achieved through the following technical solutions:

[0006] A method for optimizing the partitioning of coding units for dynamic point cloud compression includes the following steps:

[0007] Step 1: Train the segmentation and recognition model

[0008] Extract the attribute features and geometric features of multiple coding units to construct a dataset for training and testing multiple segmentation and recognition models, each corresponding to a different segmentation pattern;

[0009] Step 2: Establish a surveillance image dataset

[0010] The gradient and variance of the current coding unit are calculated to initially screen the partitioning patterns of the current coding unit. Then, the attribute features and geometric features of the current coding unit are extracted. The trained partitioning recognition model is selected to test the partitioning patterns one by one, and the optimal partitioning pattern is selected for partitioning the current coding unit.

[0011] Furthermore, the gradient and variance of the current coding unit in the horizontal and vertical directions are calculated respectively, and the partitioning patterns are initially screened using the following partitioning rules:

[0012] If the vertical gradient is significantly greater than the horizontal gradient, then the vertical partitioning mode is disabled.

[0013] If the vertical gradient is not significantly greater than the horizontal gradient, then variance is used to assist in the judgment.

[0014] If the horizontal gradient is significantly greater than the vertical gradient, then the horizontal partitioning mode is disabled.

[0015] If the horizontal gradient is not significantly greater than the vertical gradient, then variance is used to assist in the judgment.

[0016] If the horizontal and vertical gradients are relatively balanced, then all partitioning patterns are retained.

[0017] Furthermore, let G be the horizontal gradient, vertical gradient, horizontal variance, and vertical variance of the current coding unit. hor G ver V hor V ver Then the gradient ratio Thresholds th1 and th2 are set, with th1 < th2. The division rules are set as follows:

[0018] (vi) When and If the vertical gradient is significantly greater than the horizontal gradient, then the vertical partitioning mode is directly disabled.

[0019] (vii) When If the vertical gradient is large but not significant enough, variance can be used to help determine its significance.

[0020] (3) When V hor >th1 indicates that the variance ratio of the horizontal sub-blocks is relatively large, that is, the texture changes in the horizontal direction are relatively large, so the vertical partitioning mode is disabled.

[0021] (4) If V hor If the variance ratio of the sub-blocks in the horizontal direction is ≤th1, it means that the texture in the horizontal direction is relatively uniform. In this case, all partitioning patterns are retained and the process proceeds to the next encoding stage.

[0022] (viii) When G div >th1 and G div When the value is >th2, it indicates that the horizontal gradient is significantly greater than the vertical gradient, so the horizontal partitioning mode is directly disabled.

[0023] (ix) When th1 < G div When the value is ≤th2, it indicates that the horizontal gradient is large but not significant enough. Variance can be used to help determine this.

[0024] (3) If V ver >th1 indicates that the directional texture changes significantly, so the horizontal division mode is disabled;

[0025] (4) If Vver If ≤th1, it means that the texture in the vertical direction is relatively uniform, so all division patterns are retained;

[0026] (x) When If the gradients in the horizontal and vertical directions are relatively balanced, then all partitioning patterns should be retained.

[0027] Furthermore, the trained segmentation recognition model is selected to test the selected segmentation patterns one by one. If the output of the current segmentation recognition model is less than the threshold, the next segmentation recognition model is selected to continue testing until the output of the current segmentation recognition model is not less than the threshold. Then, the segmentation pattern corresponding to the current segmentation recognition model is considered to be the optimal segmentation pattern.

[0028] Furthermore, the partitioning modes include binary tree horizontal partitioning mode BTH, binary tree vertical partitioning mode BTV, ternary tree horizontal partitioning mode TTH, ternary tree vertical partitioning mode TTV, and no partitioning mode. The partitioning recognition models corresponding to the first four partitioning modes are BTH neural network model, BTV neural network model, TTH neural network model, and TTV neural network model, respectively.

[0029] Furthermore, several frames of data were encoded using the original V-PCC encoder on the dynamic point cloud under different quantization parameters. The final selected partitioning mode is the optimal partitioning mode. The attribute features and geometric features of the corresponding encoding units were extracted to construct a dataset for training and testing the partitioning recognition model corresponding to the optimal partitioning mode.

[0030] Furthermore, the attribute features include the aspect ratio, rate-distortion superiority, and orientation consistency of the current coding unit, and the geometric features include the absolute average linearity and absolute average curvature of the point cloud region corresponding to the current coding unit.

[0031] A fast compression method for dynamic point clouds based on the above-described method for optimizing the partitioning of coding units for dynamic point cloud compression is proposed. When performing dynamic point cloud compression, the method described above for optimizing the partitioning of coding units for dynamic point cloud compression is used to partition the coding units.

[0032] The beneficial technical effects of this invention are as follows:

[0033] 1. By calculating the gradient and variance of the current coding unit, the directional characteristics of the point cloud coding unit are analyzed, the candidate list of partitioning modes is initially updated, redundant mode searches are reduced, and the rationality of partitioning decisions is improved.

[0034] 2. Drawing on the experience of 2D video coding optimization and combining the characteristics of 3D point cloud data, the attribute features and geometric features of the point cloud are calculated and input into the neural network model for prediction. Based on the prediction results, it is determined whether the test of the current partitioning mode can be skipped, that is, the test of the current partitioning mode is terminated in advance. This early termination mechanism significantly reduces the amount of computation while maintaining a high compression ratio and restoration quality, reducing coding complexity, effectively improving coding efficiency and significantly shortening coding time, thus improving the running speed of V-PCC.

[0035] 3. It adopts a lightweight neural network model with low computational overhead, which can be efficiently integrated into the V-PCC reference coding software, reducing dependence on hardware resources.

[0036] This application embodiment adopts H.266 as the encoding complexity optimization method for the internal two-dimensional encoder of V-PCC. H.266 (VVC) is the latest video coding standard, which has higher coding efficiency, stronger adaptability to different types of two-dimensional video content, and greater flexibility in encoder architecture selection. Attached Figure Description

[0037] Figure 1 A schematic diagram of the existing VPCC encoding process;

[0038] Figure 2 A schematic diagram of six partitioning modes for the existing H.266 / VVC encoding method;

[0039] Figure 3 This is a schematic diagram of the overall process of the present invention. Detailed Implementation

[0040] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0041] To address the shortcomings of existing technologies, this paper provides a method for optimizing the partitioning of coding units for dynamic point cloud compression based on texture features and machine learning. This method is developed by deeply analyzing the process and characteristics of V-PCC dynamic point cloud encoding and drawing on mature experience in optimizing the complexity of traditional video coding. It comprehensively considers the geometric characteristics of point cloud encoding and the characteristics of traditional 2D encoding. This method reduces time overhead while ensuring prediction accuracy. First, the directional characteristics of the current coding unit are evaluated by calculating gradients and variances to initially optimize the candidate partitioning pattern list and reduce computational complexity. Then, for the partitioning patterns in the optimized candidate list, the aspect ratio, rate-distortion advantage, segmentation direction consistency, absolute average linearity, and absolute average curvature of the current coding unit are extracted as features. These features are input into the trained neural network model corresponding to the partitioning pattern. Based on the network output, the method predicts whether the test of the current partitioning pattern (including binary tree horizontal partitioning BTH, binary tree vertical partitioning BTV, ternary tree horizontal partitioning TTH, and ternary tree vertical partitioning TTV) can be skipped in advance until the optimal partitioning pattern is found. This significantly improves the running speed of V-PCC while ensuring the quality of compressed point clouds.

[0042] Specifically, see the appendix. Figure 3 This application provides a flowchart of a coding unit optimization partitioning method for dynamic point cloud compression based on texture features and machine learning, according to an embodiment of this application. The fast dynamic point cloud compression method based on texture features and machine learning, according to an embodiment of this application, includes the following steps:

[0043] Step 1: Train the segmentation and recognition model

[0044] In the process of encoding dynamic point clouds using V-PCC, a progressively refined partitioning strategy is employed to encode the point cloud video. Specifically, encoding starts with larger basic coding units, and based on the distribution of their texture complexity, it dynamically determines whether further partitioning is needed and how, progressively dividing them into smaller coding units. This partitioned encoding method can adapt to the texture characteristics of different regions, improving encoding efficiency while ensuring reconstruction quality.

[0045] Therefore, we use the original V-PCC encoder to encode several frames of data, such as 10 frames, of dynamic point clouds under different quantization parameters. The final selected partitioning mode is the optimal partitioning mode. We extract the attribute features and geometric features of the corresponding encoding units to construct a dataset for training and testing the partitioning recognition model corresponding to the optimal partitioning mode.

[0046] Four partitioning and recognition models were constructed for the binary tree horizontal partitioning pattern (BTH), binary tree vertical partitioning pattern (BTV), ternary tree horizontal partitioning pattern (TTH), and ternary tree vertical partitioning pattern (TTV), respectively. These models are the BTH neural network model, the BTV neural network model, the TTH neural network model, and the TTV neural network model. Each neural network has three fully connected layers: the first fully connected layer has 15 neurons, the second fully connected layer has 10 neurons, and the third fully connected layer has 5 neurons.

[0047] Encoding dynamic point cloud video sequences begins by dividing a single frame into several Coding Tree Units (CTUs). Each CTU is processed as an independent coding unit. Following the original partitioning strategy of the coding standard, different partitioning modes are applied to each CTU for refinement. During this process, the attribute features and geometric features of each coding unit are extracted, and the final partitioning mode corresponding to each coding unit is recorded as the optimal partitioning mode. By extracting the attribute features, geometric features, and final partitioning mode of each coding unit, a dataset containing the features of coding units and their corresponding optimal partitioning modes is constructed. These features are used as input data to train a neural network, laying the foundation for subsequent prediction of partitioning modes.

[0048] The extracted dataset is preprocessed and then fed into the corresponding neural network model for training to learn the final decision of the partitioning of coding units under different modes. The four neural network models obtained by the final training will be used to predict the early termination of the partitioning mode during the encoding process, so as to select the optimal partitioning model.

[0049] Step 2: During the encoding process, acquire the grayscale image of the current encoding unit and calculate the gradient features and variance features.

[0050] Extract the grayscale image of the current coding unit, and calculate its directional gradient ratio G based on the grayscale image. div And the ratio of the variance of the sub-blocks in the horizontal and vertical directions (V hor and V ver This is used to quantitatively assess the differences in texture complexity in different directions.

[0051] S21. Obtain the grayscale image matrix G of the current coding unit, which is defined as:

[0052] G={g i，j}for i∈[x,x+W],j∈[y,y+H]

[0053] Where W and H represent the width and height of the region, respectively, g i，j It is the gray value at position (i, j).

[0054] S22. The gradient is used to measure the texture complexity of the current coding unit. The horizontal gradient G of the current coding unit is extracted. hor Vertical gradient G ver The gradient ratio G is calculated in this way. div Its calculation formula is defined as follows:

[0055]

[0056] S23, Sub-block variance σ of the coding unit 2 The texture complexity used to measure the current coding unit is defined as follows:

[0057]

[0058] Where, x represents i Pixel values ​​within a sub-block This represents the average value of the sub-block, where N is the number of pixels.

[0059] The horizontal sub-block variance ratio V of the current coding unit is calculated using the following formula. hor V = Variance ratio of vertical sub-blocks ver Its definition is:

[0060]

[0061] in, This represents the variance of the upper half of the sub-block of the coding unit. This represents the variance of the lower half of the sub-block of the coding unit. This represents the variance of the left half of the sub-block of the coding unit. This represents the variance of the right half of the sub-block of the coding unit.

[0062] Step 3: Evaluate the directionality of the partitioning patterns based on gradient and variance characteristics, and preliminarily optimize the candidate partitioning patterns accordingly. Create a candidate list of the following partitioning patterns: binary tree horizontal partitioning pattern (BTH), binary tree vertical partitioning pattern (BTV), ternary tree horizontal partitioning pattern (TTH), ternary tree vertical partitioning pattern (TTV), and no partitioning pattern. Perform preliminary screening using the following partitioning rules:

[0063] If the vertical gradient is significantly greater than the horizontal gradient, then the vertical partitioning mode is disabled.

[0064] If the vertical gradient is not significantly greater than the horizontal gradient, then variance is used to assist in the judgment.

[0065] If the horizontal gradient is significantly greater than the vertical gradient, then the horizontal partitioning mode is disabled.

[0066] If the horizontal gradient is not significantly greater than the vertical gradient, then variance is used to assist in the judgment.

[0067] If the horizontal and vertical gradients are relatively balanced, then all partitioning patterns are retained.

[0068] Specifically, let G be the horizontal gradient, vertical gradient, horizontal variance, and vertical variance of the current coding unit. hor G ver V hor V ver Then the gradient ratio Thresholds th1 and th2 are set, with th1 < th2. The division rules are set as follows:

[0069] (a) When and If the vertical gradient is significantly greater than the horizontal gradient, then the vertical partitioning mode (BTV and TTV) update mode candidate list is directly disabled.

[0070] (ii) When If the vertical gradient is large but not significant enough, variance can be used to help determine its significance.

[0071] (1) When V hor >th1 indicates that the variance ratio of the horizontal sub-blocks is large, that is, the texture changes in the horizontal direction are large. In this case, the vertical partitioning mode (BTV and TTV) update mode candidate list is disabled.

[0072] (2) If V hor If the variance ratio of the sub-blocks in the horizontal direction is ≤th1, it means that the texture in the horizontal direction is relatively uniform. In this case, all partitioning patterns are retained and the process proceeds to the next encoding stage.

[0073] (III) When G div >th1 and G div When the gradient is greater than th2, it indicates that the horizontal gradient is significantly greater than the vertical gradient, so the horizontal partitioning mode (BTH and TTH) update mode candidate list is directly disabled.

[0074] (iv) When th1 < G div When the value is ≤th2, it indicates that the horizontal gradient is large but not significant enough. Variance can be used to help determine this.

[0075] (5) If V ver >th1 indicates that the directional texture changes significantly, so the horizontal directional division mode (BTH and TTH) is disabled to update the mode candidate list;

[0076] (6) If V ver If ≤th1, it means that the texture in the vertical direction is relatively uniform, so all division patterns are retained;

[0077] (v) When If the gradients in the horizontal and vertical directions are relatively balanced, then all partitioning patterns should be retained.

[0078] Note: Retaining all partitioning modes means not disabling any partitioning modes. All partitioning modes in the original mode candidate list, such as BTH, TTH, BTV, TTV, QT, and the unsplit mode, can participate in the subsequent encoding process.

[0079] Step 4: Traverse the partitioning patterns in the candidate pattern list. When the partitioning pattern is BTH, TTH, BTV and TTV, extract the attribute features and geometric features of the current coding unit, input them into the trained partitioning recognition model corresponding to the current partitioning pattern for testing, and determine whether to terminate the test of the partitioning pattern in advance or select the next partitioning pattern to continue testing based on the predicted value output by the partitioning recognition model, until the optimal partitioning pattern is found.

[0080] If the current partitioning mode is QT or no partitioning mode, then continue coding according to the original V-PCC process.

[0081] S41. Feature extraction of the current coding unit:

[0082] The attribute features include the aspect ratio, rate-distortion superiority, and orientation consistency of the current coding unit, while the geometric features include the absolute average linearity and absolute average curvature of the point cloud region corresponding to the current coding unit. These features collectively characterize the multidimensional information of the current coding unit in terms of texture, structure, and geometry, and can comprehensively reflect its coding complexity and partitioning characteristics.

[0083] I. Absolute average linearity, as detailed below:

[0084] Step 1: For a point in the point cloud, select its neighborhood point set. These points are selected by a fixed radius or a fixed number of neighborhoods.

[0085] Step 2: For these neighborhood points, use Principal Component Analysis (PCA) to fit the local plane. PCA obtains the three principal component directions in 3D space by eigenvalue decomposition of the covariance matrix of the point set, specifically:

[0086] 1. Let P be a point in the neighborhood. j =(x j y j , z j The corresponding coordinates are P. j =[x j y j , z j ] T Then the covariance matrix C of the neighborhood point set is:

[0087]

[0088] Among them is The center point (mean point) of the neighborhood point set.

[0089] 2. Perform eigenvalue decomposition on the covariance matrix C to obtain three eigenvalues ​​λ1'λ2'λ3 and their corresponding eigenvectors v1, v2, v3. These three eigenvectors represent the directions of the three principal components in the point cloud, where λ1≥λ2≥λ3 are the eigenvalues ​​of the covariance matrix C arranged in descending order.

[0090] 3. From PCA, we can see that the direction corresponding to the principal eigenvector v1 is the direction of maximum variance, v2 is the direction of the second largest variance, and v3 is the direction of minimum variance. For a local point set, the principal component v1 and the secondary principal component v2 determine a fitting plane. Using these eigenvectors and eigenvalues, we can fit the local plane and define its equation:

[0091] ax + by + cz + d = 0

[0092] 4. For each point P i =(x i y i , z i ), calculate its distance D to the fitted plane. i Point P i The formula for distance to a plane is:

[0093]

[0094] Where a, b, and c are the components of the plane normal vector, and d is the plane offset.

[0095] Step 3: Linearity measures the linearity of the local structure of a point cloud. We can obtain linearity by calculating the standard deviation of the point set from the fitted plane. Let the distances from the point set to the plane be D1, D2, ..., D... n Then the standard deviation σ plane The calculation formula is:

[0096]

[0097] in, It is the mean distance of all points to the plane:

[0098] Step 4: Calculate the linearity L of the current point, specifically:

[0099]

[0100] Where d maxThis represents the maximum distance from a point in the point cloud to the fitted plane.

[0101] Step 5: Calculate the absolute average linearity of the current coding region. Specifically:

[0102]

[0103] Where L i L represents the linearity of the i-th point, N is the total number of points, and |L i | represents the absolute value of the linearity at each point.

[0104] II. Absolute mean curvature Specifically:

[0105]

[0106] Wherein, λ1, λ2, and λ3 are the eigenvalues ​​of the covariance matrix C, corresponding to the main direction of change of the local point cloud.

[0107] S42. Determine the optimal partitioning pattern

[0108] After preprocessing, the aforementioned geometric and attribute features are fed as input data into the trained segmentation and recognition model. The neural network learns and analyzes these features to further predict whether the current segmentation pattern is the optimal choice for the coding unit. If not, the process terminates early, providing data support for improving coding efficiency and performance.

[0109] Specifically, based on the current segmentation mode, select the corresponding segmentation recognition model (BTH, BTV, TTH, and TTV), input the extracted feature data, obtain a probability value, and compare it with the set threshold.

[0110] If the probability value is not less than the threshold, the current partitioning mode is determined to be the optimal partitioning mode, and the encoding can continue according to the original V-PCC process in this mode;

[0111] If the probability value is less than the threshold, the test of the current partitioning mode is terminated, and the test of the next partitioning mode in the candidate mode list is directly entered. In this way, when performing partitioning mode test, it is not necessary to perform encoding or even complete rate-distortion cost calculation to determine whether the current test should be terminated, that is, to give the recognition result of the optimal partitioning mode. This is convenient and fast, and can greatly save encoding time.

[0112] In addition, the present invention also provides a method for fast compression of dynamic point clouds, wherein the coding units are divided using the coding unit optimization partitioning method for dynamic point cloud compression as described above.

[0113] The following simulation experiment verifies the encoding performance of the dynamic point cloud fast compression method based on texture features and machine learning proposed in this embodiment.

[0114] To evaluate the feasibility and effectiveness of the above method, independent tests were conducted using the dynamic 3D point cloud coding reference software TMC2-v18.0 and the VVC reference software VTM-v13.0 as test platforms. The test sequences included five different sequences provided by 8i: Soldier, Longdress, Loot, Queen, dancer, and Basketball_player. The coding quantization parameter combination (QPs) was set to ([32, 42], [28, 37], [24, 32], [20, 27], [16, 22]), and the coding configuration was All Intra (AI).

[0115] Table 1. Optimization of Dynamic Point Cloud Complexity and Encoding Performance

[0116]

[0117] Experimental results show that this method exhibits good overall performance in dynamic point cloud encoding. Regarding geometric errors, the average changes in D1 and D2 are 0.2% and 0.4%, respectively, indicating that the geometric data reconstruction accuracy remains relatively stable. Regarding color errors, the average changes in Luma, Cb, and Cr are 1.1%, 1.2%, and 0.6%, respectively, indicating high accuracy in color component reconstruction. Simultaneously, the encoding time (ΔT) is reduced by an average of 40.10%, demonstrating a significant time optimization effect. Overall, the results show that this method effectively balances geometric and color reconstruction quality while improving encoding efficiency, demonstrating good performance.

[0118] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for optimizing the partitioning of coding units for dynamic point cloud compression, characterized in that... Includes the following steps: Step 1: Train the segmentation and recognition model Extract the attribute features and geometric features of multiple coding units to construct a dataset for training and testing multiple segmentation and recognition models, each corresponding to a different segmentation pattern; Step 2: Establish a surveillance image dataset The gradient and variance of the current coding unit are calculated to initially screen the partitioning patterns of the current coding unit. Then, the attribute features and geometric features of the current coding unit are extracted. The trained partitioning and recognition model is selected to test the partitioning patterns one by one and select the optimal partitioning pattern for the partitioning of the current coding unit. Calculate the gradient and variance of the current coding unit in the horizontal and vertical directions respectively, and use the following partitioning rules to initially screen the partitioning patterns: If the vertical gradient is significantly greater than the horizontal gradient, then the vertical partitioning mode is disabled. If the vertical gradient is not significantly greater than the horizontal gradient, then variance is used to assist in the judgment. If the horizontal gradient is significantly greater than the vertical gradient, then the horizontal partitioning mode is disabled. If the horizontal gradient is not significantly greater than the vertical gradient, then variance is used to assist in the judgment. If the horizontal and vertical gradients are relatively balanced, then all partitioning patterns are retained. Let the horizontal gradient, vertical gradient, horizontal variance, and vertical variance of the current coding unit be respectively... , , , Then the gradient ratio Set threshold , ,and The partitioning rules are set as follows: (a) When and If the vertical gradient is significantly greater than the horizontal gradient, then the vertical partitioning mode is directly disabled. (ii) When If the vertical gradient is large but not significant enough, variance can be used to help determine its significance. (1) When If the variance ratio of the horizontal sub-blocks is large, that is, the texture changes in the horizontal direction are large, then the vertical partitioning mode should be disabled. (2) If If the variance ratio of the sub-blocks in the horizontal direction is small, that is, the texture in the horizontal direction is relatively uniform, then all partitioning patterns are retained and the process proceeds to the next encoding stage. (III) When and If the horizontal gradient is significantly greater than the vertical gradient, then the horizontal partitioning mode is directly disabled. (iv) When If the horizontal gradient is large but not significant enough, variance can be used to help determine its significance. (1) If If the direction texture changes significantly, then the horizontal division mode should be disabled. (2) If If the vertical texture is relatively uniform, then all division patterns will be retained. (v) When If the gradients in the horizontal and vertical directions are relatively balanced, then all partitioning patterns should be retained.

2. The coding unit optimization partitioning method for dynamic point cloud compression according to claim 1, characterized in that: The trained segmentation recognition model is selected and tested one by one with the selected segmentation patterns. If the output of the current segmentation recognition model is less than the threshold, the next segmentation recognition model is selected and the test continues until the output of the current segmentation recognition model is not less than the threshold. Then the segmentation pattern corresponding to the current segmentation recognition model is considered to be the optimal segmentation pattern.

3. The coding unit optimization partitioning method for dynamic point cloud compression according to claim 2, characterized in that: The partitioning modes include binary tree horizontal partitioning mode BTH, binary tree vertical partitioning mode BTV, ternary tree horizontal partitioning mode TTH, ternary tree vertical partitioning mode TTV, and no partitioning mode. The partitioning recognition models corresponding to the first four partitioning modes are BTH neural network model, BTV neural network model, TTH neural network model, and TTV neural network model, respectively.

4. The coding unit optimization partitioning method for dynamic point cloud compression according to claim 1, characterized in that: Several frames of data are encoded using the original V-PCC encoder on dynamic point clouds under different quantization parameters. The final selected partitioning mode is the optimal partitioning mode. The attribute features and geometric features of the corresponding encoding units are extracted to construct a dataset for training and testing the partitioning recognition model corresponding to the optimal partitioning mode.

5. The coding unit optimization partitioning method for dynamic point cloud compression according to claim 4, characterized in that: The attribute features include the aspect ratio, rate-distortion superiority, and orientation consistency of the current coding unit, and the geometric features include the absolute average linearity and absolute average curvature of the point cloud region corresponding to the current coding unit.

6. A fast compression method for dynamic point clouds based on the coding unit optimization partitioning method for dynamic point cloud compression as described in claim 1, characterized in that: When performing dynamic point cloud compression, the coding unit is divided using the coding unit optimization partitioning method for dynamic point cloud compression as described in claims 1-5.

Images