CN114945096A

CN114945096A - Lossless decompression algorithm based on CPU + GPU heterogeneous platform and storage medium

Info

Publication number: CN114945096A
Application number: CN202210582027.0A
Authority: CN
Inventors: 邵云峰; 曹桂平; 董宁
Original assignee: Hefei Eko Photoelectric Technology Co ltd
Current assignee: Hefei Eko Photoelectric Technology Co ltd
Priority date: 2022-05-26
Filing date: 2022-05-26
Publication date: 2022-08-26

Abstract

The invention relates to a lossless decompression algorithm and a storage medium based on a CPU + GPU heterogeneous platform, which are based on an industrial camera comprising two CPU resources and a GPU resource, when one industrial camera is connected with a system, two cores with the lowest resource utilization rate are searched from a PC end to serve as processing cores, wherein a main core is responsible for communicating with the industrial camera, receiving data code streams and scheduling and managing GPU thread resources; receiving image original data decoded by the GPU from the core, and displaying and storing the image original data; GPU resources are applied and scheduled by a main core, if M rows of data need to be decoded at the same time, M GPU threads are applied, each thread independently decodes one row of data, and the decoded data are transmitted to a CPU slave core for displaying and storing. The invention transmits the decompression work with larger calculation amount to the GPU for completion, saves the CPU resource of the host end, simultaneously, the heterogeneous architecture of the CPU and the GPU enables the image code stream transmission and the data decoding work to be completed in parallel, and saves the total data processing time.

Description

Lossless decompression algorithm and storage medium based on CPU + GPU heterogeneous platform

Technical Field

The invention relates to the technical field of image processing, in particular to a lossless decompression algorithm and a storage medium based on a CPU + GPU heterogeneous platform.

Background

In order to realize higher transmission bandwidth, an industrial camera generally adopts a data compression mode to transmit images, and after the compressed data is transmitted to a host, a CPU completes a decompression process. Because the same host computer may be connected with a plurality of cameras simultaneously, a plurality of compressed code streams need to be processed simultaneously and parallelly, the processing speed of the single-core CPU cannot meet the requirement of decoding a plurality of code streams, and the decoding failure is caused.

Image compression refers to a technique of lossy or lossless representation of an original pixel matrix with a small number of bits, and is also called image encoding. The basic principle of image compression is that there is redundancy in the original data. The redundancy mainly comprises: spatial redundancy: as information correlation between adjacent pixels in the image; time redundancy: as a correlation between different frames in the image sequence; spectrum redundancy: appearing as a correlation of different color planes or spectral bands.

There are many methods for image compression, which can be mainly classified into two categories: lossy compression and lossless compression. Lossy compression typically has a relatively high compression ratio, but loses image detail; the compression rate of lossless compression is low, but the integrity of image information can be guaranteed.

In order to obtain a higher transmission bandwidth and ensure image quality during the use of the industrial camera, a lossless compression mode is generally adopted for data transmission, and compressed image data is decoded at a host. If the number of the compressed code streams transmitted at the same time is too large, the CPU at the host end may not decode in time, and thus an error may be generated.

The current more common solutions are as follows:

1. a lossy compression algorithm is used for replacing lossless compression, the typical representative algorithm is a JPEG algorithm, the compression rate of lossy compression is generally high, so that the transmission data volume is small, but the problems are that image detail information is lost, the calculation amount is increased, and CPU resources are occupied.

2. Using a prediction-entropy coding technique, such as reference Starosolski R.simple Fast and Adaptive Lossless Image Compression Algorithm [ J ]. Software Practice and Exerce, 2007,37(1):65-91 SFALC, the basic flow of which comprises two parts of prediction and coding, in the prediction stage, obtaining the predicted value of a pixel according to the surrounding pixel point information of the pixel to be coded, and then calculating the predicted value and the pixel true value to obtain a pixel residual error, wherein compared with the pixel original value, the dynamic range of the pixel residual error is greatly reduced, and the encoding work is more suitable; in the encoding process, an entropy encoding technique, such as huffman encoding or arithmetic encoding, is generally selected, and the residual is encoded to obtain compressed code stream data.

Specifically, the main disadvantages of the prior art are three. The first is that the code stream data receiving and decoding work are completed on the same CPU, so the data processing flow is serial, the decoding work can be started only by receiving the code stream data, the whole work efficiency is low and the decoding time is long; secondly, CPU resources are occupied, and decoding work usually needs to use more CPU resources, so that the number of code streams which can be decoded at the same time is limited by the number of CPUs at the host end and the host frequency; and thirdly, data is transmitted by using a lossy compression mode, which can cause image detail information loss and texture loss and is not suitable for detecting scenes with higher precision requirements.

Disclosure of Invention

The lossless decompression algorithm based on the CPU + GPU heterogeneous platform can solve the technical problems.

In order to realize the purpose, the invention adopts the following technical scheme:

a lossless decompression algorithm based on a CPU + GPU heterogeneous platform is based on an industrial camera comprising two CPU resources and a GPU resource, wherein one of the two CPU resources is a master core, and the other CPU resource is a slave core; which is characterized by comprising the following steps of,

when one industrial camera is connected with the system, two cores with the lowest resource utilization rate are searched from the PC end to serve as processing cores, wherein the main core is responsible for communicating with the industrial camera, receiving data code streams and scheduling and managing GPU thread resources; receiving image original data decoded by the GPU from the core, and displaying and storing the image original data;

GPU resources are applied and scheduled by a main core, if M rows of data need to be decoded at the same time, M GPU threads are applied, each thread independently decodes one row of data, and the decoded data are transmitted to a CPU slave core for displaying and storing.

Further, the method comprises the following steps:

s01, connecting the camera with the host end, and selecting two CPUs with the lowest CPU utilization rates as a main core and a slave core by the host end according to the current CPU utilization rate;

s02, the camera sends a decoding table to the main core;

s03, after each M rows of data are compressed by the camera, sending code stream data to the main core;

s04, applying for M GPU thread resources when the main core receives M lines of code stream data, wherein the number of GPU threads is the same as the number of lines of images to be decoded, and each GPU thread can independently decode one line of image data;

s05, each GPU thread independently decodes a line of data according to a decoding table and a decoding algorithm to obtain original image data;

s06, the GPU transmits the decoded M rows of data to a slave core, and the slave core is responsible for image display and storage;

and S07, repeating the steps S02-S06 until all the image data are transmitted.

Further, the steps S04, S05, and S06 are executed concurrently, assuming that the execution times of the steps S04, S05, and S06 are all T, and the whole system starts to work from time T0, the master core will complete the receiving work of the first M lines of code stream data within [ T0, T1], and at this time, neither the GPU decoding thread nor the slave core is yet working; in the time of [ T1, T2], the main core finishes the receiving work of the second M lines of code stream data, and simultaneously, the GPU decoding thread starts the decoding work of the first M lines of code stream data, and the slave core does not work; and in the time [ T2, T3], the main core finishes the receiving work of the third M-line code stream data, the GPU decoding thread starts the decoding work of the second M-line code stream data, the core starts the first M-line decoding, the display and storage work of the image data is carried out, and the steps S04, S05 and S06 are executed concurrently at all the later times.

Furthermore, the step of GPU decoding comprises inputting code stream images, carrying out Huffman decoding, then carrying out residual error recovery processing, and finally transmitting the decoded M lines of image data to a slave core for display and storage.

Furthermore, for an input code stream image, each GPU thread can decode a single line of data, the decoding processes among the GPU cores are the same, the decoded data have no correlation, and the decoding work can be started at the same time.

Furthermore, the Huffman decoding step is that the input code stream data is coded according to a Huffman coding mode, a coding table and a decoding table are sent to the GPU by the main core during initialization, and the GPU completes the Huffman decoding work according to the decoding table.

Further, the residual error recovering step is that the data after huffman decoding is an image residual error, and if the huffman decoded data is R, the original image data X is (R + a) mod 256, where a is a left-side pixel point of X, R is residual error data after S2 decoding, X is original image data, and when X is leftmost edge data of the image, a is 0;

the overall process of residual recovery is from the leftmost pixel of a line of image data, all the way back to the end of the line of image pixels.

In yet another aspect, the present invention also discloses a computer readable storage medium storing a computer program, which when executed by a processor causes the processor to perform the steps of the method as described above.

According to the technical scheme, the lossless decompression algorithm based on the CPU + GPU heterogeneous platform can simultaneously solve the three problems. The mode that the CPU receives the code stream and the GPU decodes the code stream is used, so that data receiving and image decoding can be executed concurrently, and the overall processing time is saved; the GPU is used for replacing the CPU to finish decoding work, so that the use requirement on the system CPU is reduced, and a plurality of code stream data can be processed concurrently; and the lossless compression mode is used for transmitting the code stream data, so that the detail information of the image is ensured not to be lost.

Specifically, the lossless decompression algorithm based on the CPU + GPU heterogeneous platform uses the CPU to complete the transmission work of image code stream data and the management scheduling work of GPU multithreading, and uses the GPU to complete the decoding work and the lossless decompression work of the image code stream. The decompression work with larger calculation amount but single calculation method is completed by the GPU, so that CPU resources at the host end are saved, and meanwhile, the heterogeneous architecture of the CPU and the GPU enables the image code stream transmission and data decoding work to be completed in parallel, so that the total data processing time is saved.

Specifically, the advantages of the invention are as follows:

the invention has the first advantage that the heterogeneous architecture of the CPU + GPU is used for separating the image code stream transmission work and the image decoding work, so that the image transmission work, the image decoding work and the image display work can be executed concurrently, and the whole image processing time is shortened.

The second advantage of the invention is that the GPU is used to complete the image decoding work, the number of GPU cores is much more than that of the CPU, thus a plurality of lines of image data can be decoded concurrently, the utilization rate of the CPU is reduced, and the concurrent efficiency of the system is improved.

The third advantage of the invention is that the lossless decompression algorithm is used, the lossless decompression algorithm does not lose image detail information, the image texture is reserved, and the method is suitable for scenes with high detection precision or high image quality requirement.

Drawings

FIG. 1 is a system block diagram of an embodiment of the invention;

FIG. 2 is a flowchart of the operation of an embodiment of the present invention;

FIG. 3 is a flowchart illustrating an embodiment of a GPU decoding method.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention.

FIG. 1 is a basic block diagram of a system according to an embodiment of the present invention, to which two CPU resources and one GPU resource are allocated for each industrial camera. One of the two CPU resources is a master core and the other is a slave core. When one industrial camera is connected with the system, two cores with the lowest resource utilization rate are searched from the PC end to serve as processing cores, wherein the main core is responsible for communicating with the industrial camera, receiving data code streams and scheduling and managing GPU thread resources; and receiving the image raw data decoded by the GPU from the core, and displaying and storing the image raw data. GPU resources are applied and scheduled by a main core, if M rows of data need to be decoded at the same time, M GPU threads are applied, each thread independently decodes one row of data, and the decoded data are transmitted to a CPU slave core for displaying and storing.

Specifically, the lossless decompression algorithm based on the CPU + GPU heterogeneous platform according to the embodiment of the present invention, as shown in fig. 2, includes the following steps:

step S01, connecting the camera with the host, and selecting two CPUs with the lowest CPU utilization rates as a main core and a slave core by the host according to the current CPU utilization rate;

step S02, the camera sends a decoding table to the main core;

step S03, after each M rows of data are compressed by the camera, sending code stream data to the main core;

step S04, applying for M GPU thread resources when the main core receives M lines of code stream data, wherein the number of GPU threads is the same as the number of lines of images to be decoded, and each GPU thread can independently decode one line of image data;

step S05, each GPU thread decodes a line of data according to a decoding table and a decoding algorithm to obtain original image data;

and step S06, the GPU transmits the decoded M line data to the slave core, and the slave core is responsible for image display and storage.

And step S07, repeating the steps S02-S06 until all the image data are transmitted.

It should be noted that steps S04, S05 and S06 above can be executed concurrently, assuming that the execution time of steps S04, S05 and S06 are all T, and the whole system starts to operate from time T0. Then in time [ T0, T1], the master core will complete the receiving work of the first M lines of code stream data, and at this time, neither the GPU decoding thread nor the slave core is yet working; in time [ T1, T2], the main core completes the receiving work of the second M-line code stream data, and meanwhile, the GPU decoding thread starts the decoding work of the first M-line code stream data, and the slave core does not work; and in the time of [ T2, T3], the main core finishes the receiving work of the third M-line code stream data, the GPU decoding thread starts the decoding work of the second M-line code stream data, and the display and storage work of the image data is carried out after the core starts the first M-line decoding. All the time steps S04, S05 and S06 are executed concurrently, and the overall processing efficiency of the system is improved.

Fig. 3 shows a GPU decoding flow chart, which includes the following specific flows:

step S11, inputting a code stream image: for each GPU thread, a single line of data is decoded, the decoding processes among GPU cores are the same, the decoded data do not have correlation, and the decoding work can be started at the same time;

step S12, huffman decoding: the input code stream data is coded according to a Huffman coding mode, a coding table and a decoding table are sent to the GPU by the main core during initialization, and the GPU completes Huffman decoding work according to the decoding table;

step S13, residual error recovery: the huffman-decoded data is an image residual, and assuming that the decoded data in step S12 is R, the original image data X is (R + a) mod 256, where a is a left-side pixel of X, R is the decoded residual data in step S2, and X is the original image data, and it is noted that when X is the leftmost edge data of the image, a is 0. The overall process of residual recovery starts with the leftmost pixel of a line of image data and is recovered to the end of the line of image pixels.

And step S04, transmitting the decoded M lines of image data to the slave core for display and storage.

The decoding process of the GPU comprises the following steps: the method comprises a Huffman decoding process and a residual error recovery process, wherein the two processes can not cause data information loss, so the method belongs to a lossless coding algorithm.

In summary, in the lossless decompression algorithm based on the CPU + GPU heterogeneous platform according to the embodiment of the present invention, the CPU is used to complete the transmission of image code stream data and the GPU multithread management scheduling, and the GPU is used to complete the decoding and lossless decompression of image code streams. The decompression work with larger calculation amount but single calculation method is completed by the GPU, so that the CPU resource at the host end is saved, and meanwhile, the heterogeneous architecture of the CPU and the GPU enables the image code stream transmission and data decoding work to be completed in parallel, so that the total data processing time is saved.

In yet another aspect, the present invention also discloses a computer readable storage medium storing a computer program which, when executed by a processor, causes the processor to perform the steps of any of the methods described above.

In yet another aspect, the present invention also discloses a computer device comprising a memory and a processor, the memory storing a computer program, the computer program, when executed by the processor, causing the processor to perform the steps of any of the methods as described above.

In a further embodiment provided by the present application, there is also provided a computer program product comprising instructions which, when run on a computer, cause the computer to perform the steps of any of the methods of the above embodiments.

It is understood that the system provided by the embodiment of the present invention corresponds to the method provided by the embodiment of the present invention, and the explanation, the example and the beneficial effects of the related contents can refer to the corresponding parts in the method.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by a computer program, which can be stored in a non-volatile computer-readable storage medium, and can include the processes of the embodiments of the methods described above when the program is executed. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory, among others. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), synchronous Link (Synchlink) DRAM (SLDRAM), Rambus (Rambus) direct RAM (RDRAM), direct bused dynamic RAM (DRDRAM), and bused dynamic RAM (RDRAM).

The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.

The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims

1. A lossless decompression algorithm based on a CPU + GPU heterogeneous platform is based on an industrial camera comprising two CPU resources and a GPU resource, wherein one of the two CPU resources is a master core, and the other CPU resource is a slave core; the method is characterized by comprising the following steps:

2. The lossless decompression algorithm based on the CPU + GPU heterogeneous platform according to claim 1, wherein: the method comprises the following steps:

s02, the camera sends a decoding table to the main core;

and S07, repeating the steps S02-S06 until all the image data are transmitted.

3. The lossless decompression algorithm based on the CPU + GPU heterogeneous platform according to claim 2, wherein: the steps S04, S05 and S06 are executed concurrently, and assuming that the execution times of the steps S04, S05 and S06 are all T and the entire system starts to work from the time of T0, the master core will complete the receiving work of the first M-line code stream data in the time [ T0, T1], and at this time, neither the GPU decoding thread nor the slave core is still working; in the time of [ T1, T2], the main core finishes the receiving work of the second M lines of code stream data, and simultaneously, the GPU decoding thread starts the decoding work of the first M lines of code stream data, and the slave core does not work; and in the time [ T2, T3], the main core finishes the receiving work of the third M-line code stream data, the GPU decoding thread starts the decoding work of the second M-line code stream data, the core starts the first M-line decoding, the display and storage work of the image data is carried out, and the steps S04, S05 and S06 are executed concurrently at all the later times.

4. The lossless decompression algorithm based on the CPU + GPU heterogeneous platform according to claim 1, wherein: and the GPU decoding step comprises the steps of inputting a code stream image, carrying out Huffman decoding, then carrying out residual error recovery processing, and finally transmitting the decoded M lines of image data to the slave core for display and storage.

5. The lossless decompression algorithm based on the CPU + GPU heterogeneous platform according to claim 4, wherein: for an input code stream image, each GPU thread can decode a single line of data, the decoding processes among the GPU cores are the same, the decoded data have no correlation, and the decoding work can be started at the same time.

6. The lossless decompression algorithm based on the CPU + GPU heterogeneous platform according to claim 4, wherein: the Huffman decoding step is that the input code stream data is coded according to a Huffman coding mode, a coding table and a decoding table are sent to the GPU by the main core during initialization, and the GPU completes Huffman decoding work according to the decoding table.

7. The lossless decompression algorithm based on the CPU + GPU heterogeneous platform according to claim 4, wherein: the residual error recovery step is that the data after huffman decoding is an image residual error, and if the huffman decoding data is R, the original image data X is (R + a) mod 256, where a is a left-side pixel point of X, R is residual error data after S2 decoding, X is the original image data, and when X is the leftmost edge data of the image, a is 0;

8. A computer-readable storage medium, storing a computer program which, when executed by a processor, causes the processor to carry out the steps of the method according to any one of claims 1 to 7.