Lossless compression and decoding method of depth image
By dividing depth image data into independent image blocks and resetting the encoder state, combined with a thread model and metadata, parallel lossless compression and decoding of depth images are achieved. This solves the serial dependency bottleneck and resource scheduling problem in existing technologies, improves processing efficiency and flexibility, and is applicable to fields such as robot autonomous navigation, augmented reality, and autonomous driving.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WISDOM CORNERSTONE (SHANGHAI) TECHNOLOGY CO LTD
- Filing Date
- 2026-04-01
- Publication Date
- 2026-06-26
AI Technical Summary
Existing depth image compression schemes struggle to balance processing efficiency and computational performance, exhibiting serial dependency bottlenecks and performance fluctuations due to resource scheduling issues. They are particularly difficult to implement in parallel computing and flexible applications in multi-threaded environments.
The sequence data of depth images is divided into multiple independent image blocks according to a predetermined partitioning rule, and the encoder state is reset at the beginning of encoding each image block. Parallel encoding and decoding are achieved by combining a thread model and metadata. Reusable workspaces and context objects are used to dynamically select the thread mode. A secondary compression algorithm is introduced to improve compression efficiency.
It achieves true parallel encoding and decoding, improves throughput, supports ultra-fast random access and on-demand regional decoding, eliminates resource waste, ensures efficient processing response and stability in low computing power environments, and provides hardware adaptation flexibility and storage efficiency.
Smart Images
Figure CN122293876A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of information compression, and more particularly to a lossless compression and decoding method for depth images. Background Technology
[0002] Depth images, as the core data carrier in 3D sensing technology, play an irreplaceable role in fields such as robot autonomous navigation, augmented reality, and environmental perception for autonomous driving. To achieve accurate reconstruction of the physical world, this type of data typically requires extremely high measurement accuracy and update frequency. This makes efficient lossless compression technology a crucial foundation for ensuring the real-time transmission and reliable storage of massive amounts of depth information within limited bandwidth. In applications demanding extreme response speeds, compression systems must not only ensure absolute data integrity but also meet the requirements of high-throughput processing and extremely low latency at the underlying architecture level.
[0003] However, existing data compression schemes often struggle to balance processing efficiency and computational performance when handling depth sequence data with significant spatial correlations. General compression formats, lacking optimization for the statistical characteristics of depth data, result in excessive resource consumption; while specialized depth compression methods are constrained by strict context dependencies, making true parallel computation difficult in multi-threaded environments. Frequent memory allocation and algorithm state initialization operations within the compression loop cause significant latency jitter when processing consecutive frames of data. Furthermore, the lack of flexible local localization capabilities necessitates global decompression even when only data from a specific region of interest is needed, severely limiting the system's application flexibility in computationally limited environments.
[0004] In summary, breaking the bottleneck of serial dependencies between data and eliminating performance fluctuations caused by resource scheduling have become key issues that urgently need to be addressed in the field of high-performance depth image processing. Summary of the Invention
[0005] To overcome the above-mentioned technical defects, the purpose of this invention is to provide a lossless compression and decoding method for depth images.
[0006] This invention discloses a lossless compression and decoding method for depth images. The lossless compression method includes: The sequence data of the depth image is divided into multiple independent image blocks according to a predetermined partitioning rule; Select the corresponding thread model based on the features of the depth image; The thread model calls threads and uses a reversible encoding algorithm to encode each image block independently. At the beginning of encoding each image block, the encoder state is reset so that the encoding of each image block does not depend on the data of the previous image block. Metadata is generated for each image block and associated with the encoded image block for storage. The metadata includes the position index and length information of the corresponding image block, so that the decoder can independently locate and decode any specified image block. Lossless decoding methods include: Read the metadata and decode at least one image patch based on the metadata.
[0007] Preferably, selecting the corresponding thread model based on the features of the depth image includes: When the data size is less than the first threshold In this case, select a single thread to execute directly; When the data size exceeds the first threshold Less than the second threshold At that time, N threads are created, and each thread processes 1-2 image patches; When the data size is greater than the second threshold When choosing a thread pool mode, select the thread pool mode.
[0008] Preferably, the number of image blocks is dynamically determined based on the ratio of the total number of pixels in the depth image to the number of available threads, and a minimum block pixel count threshold is set.
[0009] Preferably, in the lossless compression method and / or lossless decoding method, each thread is configured as follows: Each thread is configured with a reusable work area bound to it for storing intermediate data; Reusable workspaces are created when a compression or decoding task is first executed; Reusable workspaces remain alive until the thread exits and can be reused across multiple calls; Reusable workspaces are expanded only when the current capacity is insufficient to meet task requirements.
[0010] Preferably, in the lossless compression method and / or lossless decoding method, each thread is configured as follows: Each thread is configured with a reusable context object bound to it, which is used to maintain the state and configuration parameters of the algorithm; Reusable context objects are created when a compression or decoding task is first performed; The reusable context object remains alive until the thread exits, and subsequent task calls will directly reuse the configured reusable context object.
[0011] Preferably, resetting the encoder state includes: at the start of encoding each image block, resetting the reference value used for incremental encoding to a preset initial value.
[0012] Preferably, the reversible coding algorithm includes variable length coding (RVL), simple incremental coding, or prediction-based coding.
[0013] Preferably, the lossless compression method further includes: The output compressed file is then subjected to secondary compression using algorithms such as LZ4 or Zstandard.
[0014] Preferably, the secondary compression method includes: Archive mode: Collects the compressed files corresponding to the encoded data of all image blocks, and performs a one-time secondary compression on the entire frame compressed file to obtain the highest compression ratio; Low-latency random access mode: Divide image blocks into one or more groups, each group containing one or more contiguous image blocks, and perform independent secondary compression on each group to support selective decompression of specific image blocks or groups.
[0015] Preferably, in the compressed output file, metadata is set at the beginning of the compressed file.
[0016] Compared with existing technologies, the above technical solution has the following advantages: 1. The lossless compression and decoding method for depth images provided in this application divides the sequence data of the depth image into multiple independent image blocks according to a predetermined partitioning rule, and forces a reset of the encoder state at the beginning of encoding each image block, thus mathematically and logically severing the serial context dependency of the pixel sequence in the encoding and decoding process. This achieves true parallel encoding and decoding, unlike the pseudo-multithreading in existing technologies that still require waiting for the results of preceding data. This solution allows each thread to process different image blocks simultaneously without communication overhead, greatly improving throughput. Secondly, with metadata recording location indexes and compressed data length information, the decoder does not need to sequentially scan the file; it can directly and accurately locate and decode any specified target image block using physical offsets, achieving extremely fast random access and on-demand region decoding. Finally, while ensuring lossless restoration of depth measurements, this solution eliminates resource waste caused by redundant calculations, enabling the system to achieve efficient processing response even in low-computing-power environments when facing high frame rate depth streams. 2. This application also constructs an efficient resource scheduling closed loop by forcibly binding the execution thread with the persistent reusable workspace and reusable context object, and dynamically selecting the thread model based on the data scale. For data of different scales, the system can adaptively switch between single-threaded, multi-threaded, or thread pool modes, ensuring the flexibility and rationality of computing resource allocation. The core approach is that the reusable workspace and reusable context object held by each thread remain alive across multiple task calls during the thread's lifetime, and the workspace is only expanded when its capacity is insufficient. This mechanism eliminates the random time overhead caused by frequent memory requests from the operating system and repeated initialization of algorithm states in the real-time compression loop, thus avoiding system-level stuttering during the allocation process. The ultimate technical effect is that the system can achieve zero memory allocation overhead after entering steady-state processing, significantly reducing processing latency jitter and ensuring the smoothness and stability of the output of depth image sequences at extremely high frame rates. 3. By resetting the incremental encoding reference value to a preset initial value at the beginning of each image block encoding, each image block is endowed with fully self-contained characteristics, thus supporting the aforementioned parallelism and random access capabilities at the algorithm implementation level. This method ensures that the reconstruction process of any target image block is a closed-loop mathematical operation, no longer constrained by the decoding state of the preceding image block, allowing the decoder to independently reconstruct data from any block boundary within the file. The accompanying pluggable encoding design supports various algorithms such as variable length coding (RVL), simple incremental coding, or prediction-based coding, enabling the system to perform targeted redundancy removal for spatial coherence or zero-region features generated by different sensors. This scheme provides extremely strong hardware adaptation flexibility while maintaining the consistency of the block architecture, ensuring that the system can continuously output optimal lossless compression performance based on the statistical characteristics of depth data. 4. Furthermore, a secondary compression stage based on LZ4 or Zstandard is introduced on top of the first-stage encoding, offering a choice between archiving mode and low-latency random access mode. Archiving mode maximizes the utilization of global entropy redundancy through one-time processing of the entire frame, providing extreme compression efficiency for long-term storage; while low-latency random access mode improves compression ratio while retaining second-level retrieval capability for specific areas through independent compression of image blocks. To further improve read efficiency, the scheme sets metadata at the beginning of the compressed file, constructing an instantly accessible global offset map. The decoder can know the physical distribution of each image block in advance without accessing the massive compressed image data segment, thus minimizing the addressing latency of non-linear jump decoding. This collaborative optimization from multi-level compression to physical layout achieves a dual solution of maximizing storage space utilization and increasing local data retrieval speed. Attached Figure Description
[0017] Figure 1 A flowchart illustrating the lossless compression and decoding method for depth images provided in this application. Detailed Implementation
[0018] The advantages of the present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments.
[0019] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.
[0020] The terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. The singular forms “a,” “the,” and “the” as used in this disclosure and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.
[0021] It should be understood that although the terms first, second, third, etc., may be used in this disclosure to describe various information, such information should not be limited to these terms. These terms are used only to distinguish information of the same type from one another. For example, without departing from the scope of this disclosure, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."
[0022] In the description of this invention, it should be understood that the terms "longitudinal", "lateral", "up", "down", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.
[0023] In the description of this invention, unless otherwise specified and limited, it should be noted that the terms "installation", "connection" and "linking" should be interpreted broadly. For example, they can refer to mechanical or electrical connections, or internal connections between two components. They can be direct connections or indirect connections through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms according to the specific circumstances.
[0024] In the following description, suffixes such as "module," "part," or "unit" used to denote elements are used only for the convenience of the description of the invention and have no specific meaning in themselves. Therefore, "module" and "part" can be used interchangeably.
[0025] Please see Figure 1 , Figure 1 A flowchart illustrating the lossless compression and decoding method for depth images provided in this application.
[0026] like Figure 1 As shown, this invention discloses a lossless compression and decoding method for depth images. The lossless compression method includes: The sequence data of the depth image is divided into multiple independent image blocks according to a predetermined partitioning rule; Select the corresponding thread model based on the features of the depth image; The thread model calls threads and uses a reversible encoding algorithm to encode each image block independently. At the beginning of encoding each image block, the encoder state is reset so that the encoding of each image block does not depend on the data of the previous image block. Metadata is generated for each image block and associated with the encoded image block for storage. The metadata includes the location index of the corresponding image block and the compressed data length information, so that the decoder can independently locate and decode any specified image block. Lossless decoding methods include: Read the metadata and decode at least one image patch based on the metadata.
[0027] First, some background information is needed: Depth images, also known as depth maps or distance images, represent the distances from a sensor to various points in a scene. Unlike traditional RGB images, which typically use 8 bits per color channel, depth images usually use 16-bit values to represent depth, providing sufficient accuracy for applications such as robotics, augmented / virtual reality (AR / VR), autonomous vehicles, 3D scanning, and gesture recognition. A typical 1024×768 depth image requires approximately 1.5MB of raw storage space (640 pixels × 480 pixels × 2 bytes per pixel, or higher resolution: 1024×768 × 2 = 1,572,864 bytes). For real-time applications running at 30-120 frames per second, each sensor generates data at a rate of 45-180MB / s. Multi-sensor systems and high-resolution depth cameras exacerbate this storage and bandwidth challenge.
[0028] However, existing data compression schemes often struggle to balance processing efficiency and computational performance when handling depth sequence data with significant spatial correlations. General compression formats, lacking optimization for the statistical characteristics of depth data, result in excessive resource consumption; while specialized depth compression methods are constrained by strict context dependencies, making true parallel computation difficult in multi-threaded environments. Frequent memory allocation and algorithm state initialization operations within the compression loop cause significant latency jitter when processing consecutive frames of data. Furthermore, the lack of flexible local localization capabilities necessitates global decompression even when only data from a specific region of interest is needed, severely limiting the system's application flexibility in computationally limited environments.
[0029] Therefore, the difference in the lossless compression and decoding method for depth images provided in this application lies in dividing the sequence data of the depth image into multiple independent image blocks according to a predetermined partitioning rule, and forcibly resetting the encoder state at the beginning of encoding each image block, thus mathematically and logically severing the serial context dependency of the pixel sequence in the encoding and decoding process. This achieves true parallel encoding and decoding, unlike the pseudo-multithreading in existing technologies that still need to wait for the results of preceding data. This scheme allows each thread to process different image blocks simultaneously without communication overhead, greatly improving throughput. Secondly, with the help of metadata that records the position index and compressed data length information, the decoder does not need to scan the file sequentially. It can directly and accurately locate and decode any specified target image block through physical offset, realizing extremely fast random access and on-demand region decoding. Finally, while ensuring lossless restoration of depth measurement values, this scheme eliminates the waste of resources caused by redundant calculations, enabling the system to achieve efficient processing response even in low-computing-power environments when facing high frame rate depth streams.
[0030] Based on the above scheme, the lossless compression and decoding method for depth images provided in this application can be achieved in low computing power scenarios: the compression ratio is 8-25% of the original size.
[0031] Based on the configuration, the compression speed is 220-1000+ MB / s, and the decompression speed is 550-2500+ MB / s.
[0032] Real-time capability for full HD depth imaging, up to 120+ FPS.
[0033] Low memory footprint suitable for embedded systems It supports region-selective decoding, eliminating the need to process the entire file.
[0034] Lossless compression with perfect data integrity.
[0035] The above is a basic description of the lossless compression and decoding method for depth images provided in this application. The specific implementation of each step in this method will be explained below.
[0036] First, there is no limit to the number of image blocks that can be divided.
[0037] In one possible implementation, the number of image block divisions is dynamically determined based on the ratio of the total number of pixels in the depth image to the number of available threads, and a minimum block pixel count threshold is set. For example, the minimum block pixel count threshold can be 16384 pixels.
[0038] The number of image blocks is dynamically determined based on the total number of pixels in the depth image and the number of available threads. Combined with the minimum block pixel count threshold, this prevents metadata redundancy caused by excessive block division and compression efficiency reduction caused by encoding reset, ensuring the optimal trade-off between parallel granularity and compression ratio.
[0039] Secondly, after dividing the image into blocks, the specific implementation method of the corresponding thread model is not limited.
[0040] In one possible implementation, selecting the appropriate thread model based on the features of the depth image includes: When the data size is less than the first threshold In this case, select a single thread to execute directly; When the data size exceeds the first threshold Less than the second threshold At that time, N threads are created, and each thread processes 1-2 image patches; When the data size is greater than the second threshold When choosing a thread pool mode, select the thread pool mode.
[0041] This can be understood as dynamically matching single-threaded, multi-threaded, or thread pool models based on the data scale of the depth image, achieving precise matching of computing resources and task load. For small-scale tasks, single-threaded execution is directly selected, thus avoiding the redundant overhead of frequently creating threads. For medium-scale and / or bursty tasks, N threads are created directly, with each thread assigned to process 1-2 image patches, allowing each thread to stably process image patches without switching. For large-scale and / or continuous load tasks, a thread pool model is adopted, dynamically allocating threads and combining worker thread stealing to dynamically adapt the relationship between threads and image patches.
[0042] Furthermore, after a thread is invoked in the thread model, the thread's resource management is also unrestricted.
[0043] In one possible implementation, the threads in the lossless compression and / or lossless decoding methods are configured as follows: Each thread is configured with a reusable work area bound to it for storing intermediate data; Reusable workspaces are created when a compression or decoding task is first executed; Reusable workspaces remain alive until the thread exits and can be reused across multiple calls; Reusable workspaces are expanded only when the current capacity is insufficient to meet task requirements.
[0044] Those skilled in the art will understand that a reusable workspace is used for temporary storage of intermediate data during compression / decompression. In this application, it is bound to a thread and adheres to the above management principles. By binding a workspace to the execution thread, which is reusable across calls and expands on demand, the random latency caused by frequent memory allocation and deallocation in real-time processing loops is eliminated. This proprietary memory management approach avoids resource contention between threads, enabling zero-overhead memory allocation after the system enters steady-state processing, significantly improving the smoothness of processing.
[0045] Furthermore, in one possible implementation, in the lossless compression method and / or lossless decoding method, each thread is configured as follows: Each thread is configured with a reusable context object bound to it, which is used to maintain the state and configuration parameters of the algorithm; Reusable context objects are created when a compression or decoding task is first performed; The reusable context object remains alive until the thread exits, and subsequent task calls will directly reuse the configured reusable context object.
[0046] This section describes the resource management methods for another core resource: the reusable context object. The thread-bound context object enables the algorithm state and configuration parameters to be initialized once and reused multiple times, avoiding the complex algorithm environment reconstruction process before each task execution. This not only reduces the startup time of a single task but also ensures extremely fast response times when multiple threads execute secondary compression or decoding tasks in parallel through a lock-free design.
[0047] By employing the above methods, the essential core resources required by threads can be reused across cycles, thereby eliminating a significant amount of memory allocation overhead during compression and decoding. Furthermore, it avoids the locking and synchronization operations required for shared buffers between threads, thus greatly reducing resource overhead and latency.
[0048] Finally, those skilled in the art will understand that the specific reversible coding algorithm used in the lossless compression and decoding method for depth images provided in this application is not limited.
[0049] In one possible implementation, reversible coding algorithms include variable length coding (RVL), simple incremental coding, or prediction-based coding.
[0050] The system offers a variety of reversible encoding algorithms, enabling it to be optimized for different 3D sensing device output characteristics (such as spatial coherence and zero-region). This decoupled design ensures that the algorithm layer can continuously iterate based on the actual data distribution to maintain optimal compression performance without changing the overall modular architecture of the system.
[0051] Furthermore, in one possible implementation, resetting the encoder state includes: at the start of encoding each image block, resetting the reference value used for incremental encoding to a preset initial value.
[0052] At the beginning of each image block, the incremental encoding reference value is reset to a preset initial value, giving each image block a self-contained characteristic from the underlying algorithm implementation. This method is the physical prerequisite for realizing the aforementioned random access and parallel computing, allowing the decoder to independently reconstruct data from any block boundary within the file without having to backtrack to historical states.
[0053] It should be noted that the specific location of the stored metadata in the compressed file is also not limited.
[0054] In one possible implementation, metadata is set at the beginning of the compressed output file.
[0055] By setting metadata at the beginning of the compressed file, a globally available offset map is constructed. The decoder can know the physical distribution of each image patch in advance without scanning a large compressed data segment, thereby minimizing the latency of non-linear jump decoding and achieving efficient collaboration between data reading and computation processing.
[0056] The above is an exemplary description of a complete compression process according to this application. It is understood that secondary compression can be used to further improve the compression ratio and reduce file size.
[0057] In one possible implementation, the lossless compression method also includes: The output compressed file is then subjected to secondary compression using algorithms such as LZ4 or Zstandard.
[0058] A second compression stage is introduced based on the first-stage encoding, further reducing data redundancy by utilizing a higher-order entropy encoding algorithm. This approach, through the synergy of two-stage compression, achieves the technical effect of further improving storage space utilization while maintaining processing speed.
[0059] Furthermore, in one possible implementation, the secondary compression method includes: Archive mode: Collects the compressed files corresponding to the encoded data of all image blocks, and performs a one-time secondary compression on the entire frame compressed file to obtain the highest compression ratio; Low-latency random access mode: Divide image blocks into one or more groups, each group containing one or more contiguous image blocks, and perform independent secondary compression on each group to support selective decompression of specific image blocks or groups.
[0060] This can be understood as follows: This application provides two modes for users to choose from during secondary compression. The archive mode maximizes the utilization of global redundancy of the entire frame data, meeting the demand for extreme compression ratios in long-term storage; while the low-latency random access mode uses independent compression of groups, which preserves local decoding capabilities while maintaining compression efficiency, providing diversified strategy support for real-time streaming and region of interest retrieval.
[0061] It should be noted that the embodiments of the present invention have better implementability and are not intended to limit the present invention in any way. Any person skilled in the art may use the above-disclosed technical content to change or modify it into equivalent effective embodiments. However, any modifications or equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of the technical solution of the present invention.
Claims
1. A lossless compression and decoding method for depth images, characterized in that, The lossless compression method includes: The sequence data of the depth image is divided into multiple independent image blocks according to a predetermined partitioning rule; Based on the features of the depth image, select the corresponding thread model; Based on the thread model, the thread is invoked, and a reversible encoding algorithm is used to encode each image block independently; wherein, the encoder state is reset at the beginning of encoding for each image block, so that the encoding of each image block does not depend on the data of the previous image block; Metadata is generated for each image block, and the metadata is associated with the encoded image block and stored. The metadata includes the location index of the corresponding image block and the compressed data length information, so that the decoder can independently locate and decode any specified image block. The lossless decoding method includes: Read the metadata and decode at least one of the image blocks based on the metadata.
2. The lossless compression and decoding method for depth images as described in claim 1, characterized in that, The step of selecting the corresponding thread model based on the features of the depth image includes: When the data size is less than the first threshold In this case, select a single thread to execute directly; When the data size exceeds the first threshold Less than the second threshold At that time, N threads are created, and each thread processes 1-2 of the image blocks; When the data size is greater than the second threshold When choosing a thread pool mode, select the thread pool mode.
3. The lossless compression and decoding method for depth images as described in claim 1, characterized in that, The number of image blocks is dynamically determined based on the ratio of the total number of pixels in the depth image to the number of available threads, and a minimum block pixel count threshold is set.
4. The lossless compression and decoding method for depth images as described in claim 1, characterized in that, In the lossless compression method and / or the lossless decoding method, each thread is configured as follows: Each thread is configured with a reusable work area bound to it for storing intermediate data; The reusable workspace is created when a compression or decoding task is first executed; The reusable workspace remains alive until the thread exits and can be reused across multiple calls; The reusable workspace will only be expanded when the current capacity is insufficient to meet the task requirements.
5. The lossless compression and decoding method for depth images as described in claim 1, characterized in that, In the lossless compression method and / or the lossless decoding method, each thread is configured as follows: Each thread is configured with a reusable context object bound to it, which is used to maintain the state and configuration parameters of the algorithm; The reusable context object is created when the compression or decoding task is first executed; The reusable context object remains alive until the thread exits, and subsequent task calls will directly reuse the configured reusable context object.
6. The lossless compression and decoding method for depth images as described in claim 1, characterized in that, The reset encoder state includes: at the start of encoding each image block, resetting the reference value used for incremental encoding to a preset initial value.
7. The lossless compression and decoding method for depth images as described in claim 1, characterized in that, The reversible coding algorithms include run-length coding, simple incremental coding, or prediction-based coding.
8. The lossless compression and decoding method for depth images as described in claim 1, characterized in that, The lossless compression method further includes: The output compressed file is then subjected to secondary compression using algorithms such as LZ4 or Zstandard.
9. The lossless compression and decoding method for depth images as described in claim 8, characterized in that, The secondary compression method includes: Archive mode: Collect the compressed files corresponding to the encoded data of all the image blocks, and perform a one-time secondary compression on the entire frame of the compressed file to obtain the highest compression ratio; Low-latency random access mode: The image block is divided into one or more groups, each group containing one or more consecutive image blocks, and each group is subjected to independent secondary compression to support selective decompression of specific image blocks or groups.
10. The lossless compression and decoding method for depth images as described in claim 1, characterized in that, In the compressed output file, the metadata is set at the beginning of the compressed file.