An intra-frame encoding and decoding method for three-dimensional sonar point cloud data

Abstract

The invention discloses an intra-frame encoding and decoding method for three-dimensional sonar point cloud data, comprising: encoding operation: after voxelization processing of variable pixel resolution is performed on the position data of the three-dimensional sonar point cloud, one repeated point in each unit voxel is retained, and the position data is octree encoded; for the intensity attribute information of the three-dimensional sonar point cloud data, the voxel attribute information is formed by combining the number of repeated points in the unit voxel; the octree encoding result and the voxel attribute information after quantization are arithmetically encoded to obtain the encoding result; decoding operation: arithmetically decoding the encoding result to obtain the octree encoding result and the voxel attribute information; after octree decoding the octree encoding result, the point cloud data is reconstructed according to the variable pixel resolution, and the repeated points are randomly reconstructed according to the voxel attribute information; the voxel attribute information is dequantized to obtain the intensity attribute information of the point cloud. The method has the advantages of high compression efficiency, fast encoding and decoding speed, and high data restoration quality.

Description

Technical Field

[0001] The invention belongs to the technical field of three-dimensional point cloud encoding and decoding, and in particular relates to an intra-frame encoding and decoding method for three-dimensional sonar point cloud data. Background Art

[0002] The 3D sonar system receives echo data through underwater acoustic sensors, and processes the echo data in real time to obtain original underwater 3D sonar point cloud data. It then transmits the 3D sonar point cloud data to the surface processing system through a watertight cable and displays it in real time on a display device.

[0003] Since the amount of raw 3D sonar point cloud data generated after underwater 3D sonar equipment signal processing is huge, the transmission of raw 3D sonar point cloud data has strict requirements on network transmission performance and cannot guarantee the real-time display of sonar images. Therefore, it is urgent to use appropriate encoding and decoding methods to encode and decode 3D sonar point cloud data to achieve efficient transmission of 3D sonar point cloud data.

[0004] At present, most 3D sonar point cloud data compression schemes use video-based compression algorithms, which project 3D sonar point cloud data into a 2D domain in the form of patches and generate video sequences, and then compress the video sequences through traditional video compression. However, this projection-based compression algorithm is only applicable to dense point cloud data. 3D sonar point cloud data is received by the acoustic wave sensor embedded in the sonar device and then processed. The detection targets are generally seabed landforms or marine life, etc. Therefore, 3D sonar point cloud data is relatively sparse and is not suitable for video-based compression algorithms.

[0005] Patent document CN111899152A discloses a point cloud data compression method based on projection and video stitching, including: analyzing the geometric attribute characteristics of point cloud data to determine the corresponding projection strategy and the optimal projection angle; projecting the projection area at the same projection angle multiple times to obtain a set of two-dimensional images with spatial correlation at the same angle; stitching a set of two-dimensional images into a video file. This method is suitable for dense point cloud data, but not for the compression of submarine three-dimensional sonar point cloud data.

[0006] Patent document CN112530015A discloses a point cloud data compression and storage method based on spatial structural features, including: determining the basic shape types of point cloud data based on the spatial structural features of point cloud data; pre-constructing corresponding three-dimensional vector primitives according to the basic shape types of point cloud data; fitting target point cloud data through pre-constructed vector primitives; editing the color information of the vector primitives, and saving the corresponding vector primitives in the form of point elements, line elements, and surface elements as a three-dimensional vector graphic format. This method combines the spatial structural features of point clouds and is suitable for dense point cloud data, but is not suitable for compression of seabed three-dimensional sonar point cloud data. Summary of the invention

[0007] In view of the above, the purpose of the present invention is to provide an intra-frame encoding and decoding method for three-dimensional sonar point cloud data, which is suitable for the distribution characteristics of three-dimensional sonar point cloud data and has the advantages of high compression efficiency, fast encoding and decoding speed, and high data restoration quality.

[0008] To achieve the above-mentioned object of the invention, an embodiment provides an intra-frame encoding and decoding method for three-dimensional sonar point cloud data, comprising the following steps:

[0009] Coding operation: Obtain each frame of 3D sonar point cloud data, voxelize the position data of the 3D sonar point cloud according to the variable pixel resolution, consider all points in each unit voxel as duplicate points, retain one duplicate point in each unit voxel, and perform octree encoding on the position data in the voxel space; for the intensity attribute information of the 3D sonar point cloud data, combine the number of duplicate points in the unit voxel to form voxel attribute information and quantize it; perform arithmetic encoding on the octree encoding result and the voxel attribute information after quantization to obtain the encoding result;

[0010] Decoding operation: perform arithmetical decoding on the encoding result to obtain the octree encoding result and voxel attribute information; after octree decoding the octree encoding result, reconstruct the three-dimensional sonar point cloud data according to the variable voxel resolution, and randomly reconstruct the repeated points within each unit voxel according to the voxel attribute information; dequantize the voxel attribute information to obtain the intensity attribute information of the three-dimensional sonar point cloud data.

[0011] Preferably, each frame of acquired three-dimensional sonar point cloud data is formed by a sonar beam array, is unstructured point cloud data between two dimensions and three dimensions, and the point attribute information is intensity attribute information.

[0012] Preferably, the voxelization of the position data of the three-dimensional sonar point cloud according to the variable pixel resolution comprises:

[0013] Determine a critical plane for a sudden change in voxel resolution so that the distribution of the number of unit voxels on both sides of the critical plane is as symmetrical as possible, wherein the critical plane is a two-dimensional plane in a three-dimensional space, the normal vector of the critical plane is the voxel resolution gradient vector, and the voxel resolutions on both sides of the critical plane are different;

[0014] The voxel space origin is determined according to the sonar detection range and the critical plane of the voxel resolution mutation, and the variable voxel resolution is used as the unit length to set the coordinate axis to construct the voxel space, in which the near-field voxel resolution of the voxel space is high and the far-field voxel resolution is low, wherein the critical plane of the voxel resolution mutation divides the sonar detection range into the near field and the far field, the side containing the sonar origin is the near field, and the side not containing the sonar origin is the far field;

[0015] The constructed voxel space is used to voxelize the position data of the three-dimensional sonar point cloud.

[0016] Preferably, when deleting duplicate points within each unit voxel, the number of duplicate points is retained as attribute information of the three-dimensional sonar point cloud data.

[0017] Preferably, when performing octree encoding on the position data in the voxel space, a conditional entropy formula is used to encode the child nodes according to the occupancy of the encoded parent nodes, so as to improve the compression efficiency of the octree encoding.

[0018] Preferably, the voxel attribute information includes the number of repeated points in a unit voxel, the average intensity value and the intensity variance of all point clouds in each unit voxel. That is, after the position data of the point cloud is voxelized, the number of repeated points in the unit voxel is used as one of the attribute information, and combined with the average intensity value and the intensity variance value of all point clouds in the same unit voxel to construct the voxel attribute information.

[0019] Preferably, the randomly reconstructing the repeated points in each unit voxel according to the voxel attribute information comprises:

[0020] For unit voxels containing duplicate points, the number of duplicate points N is read and the N duplicate points are evenly scattered in the space of the corresponding unit voxel by randomly scattering particles to reconstruct the N duplicate points.

[0021] Preferably, the dequantization processing of voxel attribute information includes:

[0022] The number of repeated points N in the unit voxel in the voxel attribute information is read, and the Gaussian probability distribution is calculated with the intensity mean and intensity variance of each unit voxel, and N intensity values ​​that conform to the Gaussian probability distribution are constructed as the attribute values ​​of the point cloud in the corresponding unit voxel.

[0023] Compared with the prior art, the beneficial effects include at least:

[0024] The intra-frame encoding and decoding method of three-dimensional sonar point cloud data provided in the embodiment can realize compression encoding and decoding of three-dimensional point cloud data while ensuring high three-dimensional image restoration by encoding and decoding each frame of data, thereby alleviating the network transmission pressure of three-dimensional point cloud data and being more in line with the imaging characteristics of three-dimensional sonar images, and has very important engineering use value. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0026] Figure 1 A three-dimensional sonar system provided by an embodiment;

[0027] Figure 2 It is a flowchart of an intra-frame encoding and decoding method based on three-dimensional sonar point cloud data provided by an embodiment. DETAILED DESCRIPTION

[0028] To make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific implementation methods described herein are only used to explain the present invention and do not limit the scope of protection of the present invention.

[0029] After research and analysis, the three-dimensional sonar point cloud data obtained by the three-dimensional sonar system has a sparse characteristic. The geometry-based compression algorithm directly preprocesses and compresses the point cloud data in three-dimensional space, usually by dividing the point cloud into spatial structures and encoding the geometric position and attribute information separately. This method is more suitable for compressing three-dimensional sonar point cloud data. Based on the above research and analysis, the embodiment provides an intra-frame encoding and decoding method for three-dimensional sonar point cloud data, which improves compression efficiency, encoding and decoding speed, and data restoration quality through geometric compression.

[0030] Figure 1 The three-dimensional sonar system provided by the embodiment. Figure 1 As shown, the three-dimensional sonar system provided in the embodiment is divided into two parts, an underwater part and an above-water part. The underwater part includes an acoustic sensor and a real-time signal processing system, and the above-water part includes a processing system and a display device. The underwater signal real-time processing system and the above-water processing system are connected via a watertight cable. After the sonar signal is received by the acoustic sensor of the underwater part, it is processed by the real-time signal processing system to obtain the original underwater three-dimensional sonar point cloud data, and then the three-dimensional sonar point cloud data is transmitted to the above-water processing system through the watertight cable for processing and displayed in real time on the display device.

[0031] Figure 2 1 is a flow chart of the encoding and decoding of the intra-frame encoding and decoding method based on three-dimensional sonar point cloud data provided by the embodiment. Figure 2 As shown, the intra-frame encoding and decoding method based on three-dimensional sonar point cloud data provided by the embodiment includes an encoding operation at the encoding end and a decoding operation at the decoding end, wherein the encoding operation is performed at Figure 1 The real-time signal processing system in the three-dimensional sonar system shown in the figure is realized, and the decoding operation is performed in Figure 1 The processing system in the three-dimensional sonar system shown in the figure is implemented. The encoding operation and the decoding operation are described in detail below.

[0032] like Figure 2 As shown, the encoding operation includes:

[0033] S101, acquiring each frame of three-dimensional sonar point cloud data, and voxelizing the position data of the three-dimensional sonar point cloud according to a variable pixel resolution.

[0034] In the embodiment, each frame of three-dimensional sonar point cloud data obtained is formed by a sonar beam array, is unstructured point cloud data between two dimensions and three dimensions, and the point attribute information is intensity attribute information.

[0035] For the unstructured 3D sonar point cloud data obtained, first determine the appropriate critical plane for the voxel resolution mutation, so that the distribution of the number of unit voxels on both sides of the critical plane is as symmetrical as possible. Take a 3D sonar with a sonar detection range of 50°×50°×120m and an angular resolution of 0.4° as an example. Its sonar detection range is a spherical cone in 3D space. The distance between the critical plane and the sonar origin is approximately 1 / 3 of the sonar detection range. The actual distance between the critical plane and the sonar origin is 40m.

[0036] Then, the voxel space origin and the far-field and near-field ranges are determined according to the sonar detection range and the critical plane of the voxel resolution mutation, and a voxel space with high near-field voxel resolution and low far-field voxel resolution is constructed.

[0037] In the embodiment, the origin of the voxel space is the intersection of the critical plane and the central axis of the spherical cone formed by the sonar detection range. The near field is the range from the origin to the detection distance of 40m, and the far field is the range from 40m to the maximum sonar detection distance of 120m, and the interface between the two is the critical plane. The voxel resolution of the near field area is generally a multiple of 2 compared to the resolution of the far field. In order to make the number of repeated points in a unit voxel no more than 5, the side length of the near field voxel block is selected to be 2m, and the side length of the far field voxel block is selected to be 4m. By constructing a voxel space with high near-field voxel resolution and low far-field voxel resolution, the details of the three-dimensional sonar image after encoding and decoding can be effectively retained, and some redundant information of the distant point cloud can be ignored, which is conducive to improving the imaging quality.

[0038] Finally, the position data of the 3D sonar point cloud is voxelized, and the number of repeated points in the same unit voxel is retained as attribute information.

[0039] S102, retaining one duplicate point in each unit voxel, constructing voxel attribute information and then performing quantization processing.

[0040] In the embodiment, all points in each unit voxel are considered to be duplicate points. In order to reduce data redundancy, the duplicate points in the same unit voxel after voxelization are deleted, and only one duplicate point is retained. At the same time, the number of duplicate points in the unit voxel is recorded, and the number of duplicate points in the unit voxel is used as one of the voxel attribute information, combined with the average value and intensity variance value of all point clouds in the same unit voxel, to construct voxel attribute information. The attribute information is quantized to achieve attribute encoding.

[0041] S103, performing octree encoding on the position data in the voxel space.

[0042] In an embodiment, the voxelized three-dimensional sonar point cloud data is divided and encoded using an octree structure. Specifically, the octree structure is constructed by recursively dividing the point cloud bounding box (the smallest cube that surrounds all point clouds and the origin). At each level, each node block is divided into 8 child node blocks, and 0 and 1 are used to represent the occupancy of the 8 child node blocks. Among them, 1 indicates that the child nodes of the current coding node are occupied, and 0 indicates that the child nodes of the current coding node are not occupied (i.e., empty nodes). When performing octree partitioning, only child node blocks that are non-empty and larger than 1×1×1 in size (i.e., non-unit voxels) will continue to be divided until they are divided into leaf nodes. Due to the use of variable voxel resolution, the 1×1×1 child node blocks in the near distance are smaller in size than the 1×1×1 child node blocks in the far distance. Generally speaking, the volume of a nearby 1×1×1 child node block is 1 / 8 of the volume of a distant 1×1×1 child node block. Within the critical spatial range where the resolution changes, the farther unit child node block is the parent node block of the closer unit child node block. This is conducive to unified octree division within the point cloud bounding box and realizes different divisions for distant and near locations.

[0043] In the embodiment, the conditional entropy formula shown below is used to encode the child nodes according to the occupancy of the encoded parent node:

[0044] H(b)=H(b 0 )+H(b 1 |b 0 )+…+H(b 7 |b 6 …b 0 )

[0045] Among them, b 7 b 6 …b 0 An eight-bit integer with a value between 1 and 255 is formed, which represents the occupancy of the eight child node blocks divided from a node block b. 0 The binary encoder is used to encode directly, and then the encoded sub-node blocks are used to encode the remaining sub-node blocks. And so on, the encoding result of node block b can be obtained. This process realizes geometric coding.

[0046] S104, performing arithmetic coding on the octree coding result and the voxel attribute information after quantization to obtain a coding result.

[0047] In the embodiment, arithmetic coding is used to further compress the octree encoding result (i.e., the geometric position of the point cloud data after octree encoding) and the voxel attribute information after quantization processing (the number of repeated points in the voxel, the combination of the intensity mean and the variance), so as to complete the encoding of the three-dimensional sonar point cloud data, and obtain the compressed three-dimensional sonar point cloud data as the compressed encoding result, and the compressed encoding result is uploaded to the processing system of the water part through the watertight cable in the form of a data stream.

[0048] like Figure 2 As shown, the decoding operation includes:

[0049] S201, performing arithmetic decoding on the encoding result.

[0050] In an embodiment, octree encoding result data and voxel attribute information of the point cloud geometric position are obtained by performing arithmetical decoding on the encoding result, wherein the voxel attribute information includes the number of repeated points in the voxel, the intensity mean and the intensity variance.

[0051] S202, performing inverse quantization processing on the voxel attribute information to obtain intensity attribute information of the three-dimensional sonar point cloud data.

[0052] In the embodiment, after arithmetic decoding, the voxel attribute information is restored by attribute inverse quantization to obtain point cloud intensity information. Specifically, the number of repeated points N in the voxel is read and the intensity mean and intensity variance of the voxel are used as the mathematical expectation and variance of the Gaussian probability distribution, the Gaussian distribution function of the point cloud intensity in the voxel is calculated, and N intensity values ​​that conform to the Gaussian probability distribution are constructed as the attribute values ​​of the point cloud corresponding to the voxel block. This process completes the attribute decoding.

[0053] S203, performing octree decoding on the octree encoding result data.

[0054] In the embodiment, the octree encoding result data containing the geometric position of the point cloud is octree decoded to reconstruct the point cloud space. Specifically, the decoding is performed according to the above conditional entropy formula, and the geometric position information of the point cloud in the voxel space is reconstructed.

[0055] S204, reconstructing the three-dimensional sonar point cloud data according to the variable voxel resolution in the voxel space, and repeating the point construction.

[0056] In the embodiment, the point cloud is reconstructed according to the variable pixel resolution of the critical plane, near field area and far field area of ​​the voxel space. For the unit voxel containing repeated points, the number of repeated points N is read and the N points are evenly scattered in the space of the corresponding unit voxel by randomly scattering particles to reconstruct the N repeated points. Through the variable resolution point cloud reconstruction and repeated point construction, the geometric decoding of the position information of the three-dimensional point cloud is completed.

[0057] The decoding of the three-dimensional point cloud data is completed through the position information of geometric decoding and the color information of attribute decoding.

[0058] The intra-frame encoding and decoding method of 3D sonar point cloud data provided in the above embodiment has the characteristics of high compression efficiency, fast encoding and decoding speed, and high image restoration quality. The method is suitable for encoding and decoding compression of point cloud data collected by 3D sonar, and has high compression efficiency. The method can be applied to encoding and decoding of point cloud data of 3D sonars with a spherical cone detection range, which is conducive to realizing efficient transmission of 3D sonar point clouds.

[0059] The specific implementation methods described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for intra-frame encoding and decoding of three-dimensional sonar point cloud data, characterized in that: The following steps are involved: Coding operation: Obtain each frame of 3D sonar point cloud data, voxelize the position data of the 3D sonar point cloud according to the variable pixel resolution. At this time, all points in each unit voxel are repeated points, and one repeated point in each unit voxel is retained. The position data in the voxel space is octree encoded; for the intensity attribute information of the 3D sonar point cloud data, the voxel attribute information is formed by combining the number of repeated points in the unit voxel and quantized; Performing arithmetic coding on the octree coding result and the voxel attribute information after quantization to obtain a coding result; The position data of the three-dimensional sonar point cloud is voxelized according to the variable pixel resolution, including: determining a critical plane of voxel resolution mutation so that the distribution of the number of unit voxels on both sides of the critical plane is as symmetrical as possible, wherein the critical plane is a two-dimensional plane in the three-dimensional space, the normal vector of the critical plane is the voxel resolution gradient vector, and the voxel resolutions on both sides of the critical plane are different; determining the origin of the voxel space according to the sonar detection range and the critical plane of the voxel resolution mutation, setting the coordinate axis with the variable pixel resolution as the unit length, and constructing a voxel space, wherein the near-field voxel resolution of the voxel space is high and the far-field voxel resolution is low, wherein the critical plane of the voxel resolution mutation divides the sonar detection range into the near field and the far field, the side containing the sonar origin is the near field, and the side not containing the sonar origin is the far field; using the constructed voxel space to voxelize the position data of the three-dimensional sonar point cloud; Decoding operation: perform arithmetical decoding on the encoding result to obtain the octree encoding result and voxel attribute information; perform inverse quantization on the voxel attribute information, perform octree decoding on the octree encoding result, reconstruct the three-dimensional sonar point cloud data according to the variable voxel resolution, and randomly reconstruct the repeated points within each unit voxel according to the voxel attribute information to obtain the intensity attribute information of the three-dimensional sonar point cloud data.

2. The intra-frame encoding and decoding method of three-dimensional sonar point cloud data according to claim 1, characterized in that: Each frame of 3D sonar point cloud data acquired is formed by a sonar beam array and is unstructured point cloud data between 2D and 3D. The point attribute information is intensity attribute information.

3. The intra-frame encoding and decoding method of three-dimensional sonar point cloud data according to claim 1, characterized in that: When deleting duplicate points within each unit voxel, the number of duplicate points is retained as attribute information of the three-dimensional sonar point cloud data.

4. The intra-frame encoding and decoding method of three-dimensional sonar point cloud data according to claim 1, characterized in that: When performing octree encoding on the position data in the voxel space, the conditional entropy formula is used to encode the child nodes according to the occupancy of the encoded parent nodes.

5. The intra-frame encoding and decoding method of three-dimensional sonar point cloud data according to claim 1, characterized in that: The voxel attribute information includes the number of repeated points in a unit voxel, and the intensity average and intensity variance of all point clouds in each unit voxel.

6. The intra-frame encoding and decoding method of three-dimensional sonar point cloud data according to claim 1, characterized in that: The randomly reconstructing the repeated points in each unit voxel according to the voxel attribute information includes: For unit voxels containing duplicate points, the number of duplicate points N is read and the N duplicate points are evenly scattered in the space of the corresponding unit voxel by randomly scattering particles to reconstruct the N duplicate points.

7. The intra-frame encoding and decoding method of three-dimensional sonar point cloud data according to claim 1, characterized in that: The dequantization processing of voxel attribute information includes: The number of repeated points N in the unit voxel in the voxel attribute information is read, and the Gaussian probability distribution is calculated with the intensity mean and intensity variance of each unit voxel, and N intensity values ​​that conform to the Gaussian probability distribution are constructed as the attribute values ​​of the point cloud in the corresponding unit voxel.

Images