CN101068366A - Method and multiplexer based on H.264 multiplex video transcoding multiplexing - Google Patents

Abstract

An escape-code complexing method based on H.264 multi-path video includes applying fast conversion means of MPEG-2 to H.264 code and utilizing H.264 macro-block mode to select relativity to MPEG-2 motion compensation residual error as well as utilizing motion compensation residual error and MB mode as well as mapped H.264 macro-block mode obtained by MPEG-2 decoding to synthesize TS stream and to input multi-path MPEG-2 program stream to escape code complex server by SI interface through PCI bus and outputting escape-coded and complexed single H.264 video stream through PCI bus in ASI interface mode.

Description

Method and multiplexer based on H.264 multiplex video transcoding multiplexing

technical field

The invention belongs to the field of video compression encoding and multiplexing in digital television. In particular, it relates to a method and a multiplexer based on H.264 multiplex video transcoding and multiplexing.

Background technique

In recent years, mobile digital TV has developed rapidly in China, but the image bandwidth restricts the expansion of digital video services. In order to give consideration to both code stream transmission efficiency and video image quality, the usual transmission rate of the system is 6-10Mbps. However, most digital TV video recording programs adopt the MPEG-2 video compression standard, and the image size is relatively large. For example, the standard definition MPEG-2 bit rate is about 4Mbps, and the high definition MPEG-2 bit rate is about 10Mbps. The bandwidth of mobile digital TV users is generally difficult to meet the real-time transmission of multiple high-bit-rate video streams. In order to enable users to watch more mobile digital TV programs smoothly under the condition of lower bandwidth, it is necessary to reduce the bit rate of video streams. . Coupled with the limitation of storage capacity and the emergence of various digital TV terminals, digital TV users have increasingly urgent demands for efficient video coding technology.

Before the digital TV source adopts low-bit-rate, high-definition compression coding standards, there are two current solutions to the above problems. One is to compress high-bit-rate digital videos such as MPEG-2 into Low bit rate MPEG-2 digital video; the second is to transcode high bit rate MPEG-2 digital video into H.264 digital video. The first method will lead to a substantial decline in image quality, which is obviously not desirable, and the second method will obtain higher compression efficiency and lower transmission bit rate without reducing the image quality.

Compared with MPEG-2, H.264 can increase the compression efficiency by more than 4 times under the same image quality. It can be seen that the above-mentioned second method is preferable. However, as a pure video compression standard, H.264 has no content about audio and video synthesis and multiplexing transmission. At present, there is no dedicated equipment to realize video transcoding and multiplexing from MPEG-2 to H.264. Considering that the original MPEG-2 front-end equipment of the TV station is large and very expensive, it is unrealistic to abandon a large number of existing MPEG-2 front-end equipment, including digital video cameras and non-linear image program editors. How to ensure the image quality while greatly reducing the bandwidth of the image, that is, building a dedicated device for transcoding and multiplexing has become a top priority.

The method and multiplexer based on H.264 multiplex video transcoding and multiplexing are not involved in the prior art. Such as CN1745573 image pickup device and moving picture shooting method thereof, the image pickup device working under the moving picture shooting mode, wherein before the moving picture shooting starts, it is indicated by the shutter button on the key input part (12), and the control part ( The clock frequency of 10) is set to a common frequency, thereby reducing power consumption in the monitoring state to prolong battery life, and wherein, when the start of motion picture shooting is instructed, the clock frequency is greatly increased by the clock conversion control part (101), Therefore, during the decoding process of moving picture data, the MPEG converter (7) can access SDRAM (8) storing YUV data at high speed, such as reference data, search data, etc., and can compress the moving picture in real time.

CN1567271 MPEG code stream conversion collection method and device with high-speed network interface, data filtering, PID modification, service information insertion and code rate conversion of transmission stream are realized in the device, and the device has a high-speed Ethernet interface for transmitting the transformed target stream to the computer. Realize the direct acquisition of the code stream, and also process the code stream.

CN1633180 Multi-description video coding method based on transformation and data fusion, including implementing transformation 1 to n on the signal to be coded; respectively performing quantization and entropy coding on the signal after transformation 1 to n; respectively quantizing according to the respective paths 1 to n and entropy-encoded signals 1~n are decoded; the decoded signals 1~n are respectively inversely transformed; after inverse transformation, side descriptions 1~n are respectively obtained, and the 1~n inversely transformed data are fused into a central description and other steps. It can combine multi-description coding and video coding based on transformation and data fusion. For a set of video sequences, this coding method can generate multiple MPEG streams, and can restore a distorted image from each stream. video sequence; when multiple streams are received, a video sequence with less distortion will be restored.

Contents of the invention

The invention proposes adding a dedicated H.264 video transcoding and multiplexing server on the basis of the original MPEG-2 mobile digital TV, and adopts a video transcoding algorithm in a transform domain to reduce transcoding complexity. The transcoding from MPEG-2 to H.264, the multiplexing and demultiplexing of H.264 video and audio, and the multiplexing and demultiplexing of multiple H.264 programs are realized by software.

The technical scheme adopted in the realization of this transcoding multiplexer is as follows: based on the method and multiplexer of H.264 multi-channel video transcoding multiplexing, the input is a multi-channel MPEG-2 single program stream, and the output is one H.264 Multi-program stream, realizing video transcoding from MPEG-2 to H.264, demultiplexing and multiplexing of audio and video, and multiplexing of multiple H.264 programs. The video transcoding includes code rate, resolution and format conversion . The video transcoding algorithm from MPEG-2 to H.264 adopts a machine learning-based transcoding algorithm to realize adjustable bit rate and resolution, and different algorithms are used within and between frames. The fast conversion method from MPEG-2 to H.264 code is as follows. When synthesizing the TS stream, rewrite the PID value according to a certain rule, so as to avoid the decoder not being able to decode correctly caused by the PID conflict. When synthesizing the TS stream, the stream type field of the PMT table is modified accordingly. Multiple MPEG-2 program streams are input to the transcoding and multiplexing server through the PCI bus through the ASI interface, and the single H.264 video stream after transcoding and multiplexing is output through the PCI bus through the ASI interface, and uses the half-full signal provided by the FIFO To read FIFO data or write FIFO to avoid frequent CPU access to the PCI interface. The transcoding from MPEG-2 to H.264 and the multiplexing of H.264 video stream and audio stream after multiple transcoding are completed in the same server.

The TS stream is composed of the encoded elementary data stream (ES) packaged according to a certain format to form a PES package, and then some system information is added. At the sending end, the PES package of the elementary stream is completed by the audio/video encoder, and the multiplexer receives Audio and video data streams and auxiliary data streams at the encoding end are interleaved into a single TS stream according to a certain multiplexing method. In order to realize audio and video synchronization, various time marks and system control information must also be added to the code stream. For the receiving end, it is just the opposite of the sending end process.

Video transcoding from MPEG-2 to H.264: Transcoding from MPEG-2 video to H.264 video, there are currently two main architectures: cascade system transcoding (CPDT) based on pixel domain and DCT domain-based transcoding Transcoding (DDT). Transcoding based on the cascade system in the pixel domain is to completely decode it first, process it in the pixel domain, and then re-encode. Since the encoding part and the decoding part are completely independent in structure during the secondary encoding, transcoding has great flexibility, but the motion vector and encoding mode of the macroblock data are recalculated, and the transcoding efficiency is low. Realized by software, it is difficult to meet real-time requirements. DCT domain-based transcoding (DDT) directly re-estimates DCT coefficients and motion loss in the DCT domain. The computational complexity is low, but the flexibility is limited. When it is required to change the motion vector, code rate, resolution, etc., it is very difficult Difficult to adopt this architecture.

The method for quickly converting MPEG-2 to H.264 codes of the present invention utilizes the correlation between H.264 macroblock mode selection and MPEG-2 motion compensation residuals to convert the H.264 macroblock mode selection problem into data classification The problem is that the motion compensation residual, MB mode, and coded block mode (CBPC) obtained by MPEG-2 decoding are directly mapped to the macroblock mode of H.264; when MPEG-2 code is decoded, relevant MB information is saved, including MB Coding mode, coded block type (CBPC), mean and variance of MB residuals (calculated separately in 4×4 sub-MB, a total of 16 mean and variance), after decoding, use standard H.264 encoder to YUV Image encoding, and save the H.264MB encoding mode, use the machine learning algorithm to obtain a decision tree for the classification of the H.264 encoding mode; when decoding the MPEG-2 stream, obtain the MC residual and macroblock mode of MPEG-2 , coded block mode (CBPC), and calculate the mean and variance of the 4×4 sub-block MC residuals; obtain the macroblock coding mode in H.264 through a decision tree; when encoding H.264, directly encode the MB coding mode Assignment; the input of the H.264 encoder is the MPEG-2 decoded YUV data and the MB coding mode, and does not use the MPEG-2 motion vector. In the motion estimation, the MB coding mode obtained by the decision tree is used. The block diagram of its transcoding algorithm is shown in Figure 1.

The method of obtaining the decision tree is: the classification of the decision tree should follow the following principles:

1) Classifiers that divide the input sequence into Intra, Skip, Inter16×16 and Inter8×8;

2) Divide Inter16×16 into 16×16, 16×8, 8×16 classifiers;

3) Divide inter8×8 into 8×8, 8×4, 4×8, 4×4 classifiers.

The decision tree generation should follow the following principles:

1) If the MC of MPEG-2MB is not coded, that is, there is no non-zero MV, and the four 8×8 blocks have no coding coefficients, H.264MB will be coded into 16×16, and the optimal mode needs to be selected through the second-level discrimination of the decision tree ;

2) If the MPEG-2 MB is in intra mode, then in H.264, the MB is coded as intra or inter8×8. If it is coded as intra, the algorithm terminates; Excellent mode;

3) If MPEG-2MB is in skip mode, in H.264, this MB is also in skip mode.

4) The decision tree is generated by WEKA data mining tool. The file format of WEKA's data mining program is ARFF (Attribute-Relation File Format). An ARFF file is written in ASCII code, reflecting the interrelationships between a set of attributes. It generally includes two different sections: 1) the file header, including the name, attribute and type of the relationship; 2) the data.

5) The training set consists of high bit rate MPEG-2 sequences, excluding B frames. The decision set is obtained by H.264 re-encoding after decoding the MPEG-2 code stream. In the H.264 encoding process, the quantization parameter is 25, and the macroblock encoding mode is obtained through RD optimization.

The transcoding decision tree includes three levels, using three different WEKA trees, as shown in Figure 2:

1) Classifiers that divide the input sequence into Intra, Skip, Inter16×16 and Inter8×8;

2) Divide Inter16×16 into 16×16, 16×8, 8×16 classifiers;

3) Divide inter8×8 into 8×8, 8×4, 4×8, 4×4 classifiers.

The first WEKA decision tree, the training data set uses the mean and variance of 16 4×4 sub-block residuals in a MPEG-2 macro block, macro block mode (skip, intra and 3 kinds of non-intra, respectively with 0 , 1, 2, 4, 8 logo), coding block mode (CBPC) and H.264 MB coding mode.

The instance row samples of the ARFF data segment are used to train the decision tree model, and one row represents a macroblock sample.

For the second decision tree, the training sample set uses the mean and variance of 16 4×4 sub-block residuals in a MPEG-2 macroblock, macroblock mode (3 kinds of non-intra), coded block mode (CBPC) and H.264MB 16×16 sub-coding mode (16×16, 16×8, 8×16). This decision tree determines the final encoding mode of inter16×16.

For the third decision tree, the training sample set uses the mean and variance of 4 4×4 sub-block residuals in a MPEG-2 macroblock, macroblock mode (3 kinds of non-intra), coding block mode (CBPC) and H.264MB 8×8 sub-coding mode (8×8, 8×4, 4×8, 4×4).

Based on these training files, a decision tree is generated using the J48 algorithm through the WEKA data mining tool. The J48 algorithm was proposed by Ross Quinlan and has a wide range of applications in the field of data mining.

TS stream multiplexing

For the H.264 video of the multi-channel program after transcoding, and the audio of the original program, the multiplexing and synchronization of audio and video data are realized according to the MPEG-2 system layer, and the multi-channel program is synthesized into one TS stream (transport stream) for transmission. The TS stream is composed of the encoded elementary data stream (ES) packaged according to a certain format to form a PES package, and then some system information is added. At the sending end, the PES package of the elementary stream is completed by the audio/video encoder, and the multiplexer receives Audio and video data streams and auxiliary data streams at the encoding end are interleaved into a single TS stream according to a certain multiplexing method. In order to realize audio and video synchronization, various time marks and system control information must also be added to the code stream. For the receiving end, it is just the opposite of the sending end process.

A transport stream can be composed of multiple programs, and each program can be composed of multiple streams, including video streams, audio streams, program specific information streams (PSI), and so on. There are four types of PSI: Program Association Table (PAT), Program Mapping Table (PMT), Network Information Table (NIT) and Conditional Access Table (CAT). The multiplexer packs the transcoded H.264 video and original audio in the transport stream format. The length of the TS packet is 188 bytes, which is divided into two parts: the header and the payload. The 4-byte prefix of the packet header is the link header, including the synchronization byte 0×47 and the data packet identification PID. From the PID, it can be judged whether the data type behind it is a video stream, audio stream, PSI or other data packets. The packet payload is the actual content of the packet, depending on the situation, a PES packet or a PSI packet can be placed.

PSI is used to describe the composition structure of the transport stream, and plays an extremely important role in the system, especially the PAT table and PMT table in multiplexing. The PAT table shows how many sets of programs there are in one TS stream, and the corresponding relationship between it and the PID of the PMT table; the PMT table gives the specific composition of a set of programs and the corresponding relationship with PIDs such as video and audio.

In the transcoding multiplexer, the MPEG-2 transport stream (SPTS) of multiple single programs is transcoded by software and multiplexed into one H.264 transport stream (MPTS) of multiple programs. Its system composition block diagram is shown in the figure 3.

Multiple single-program MPEG-2 TS streams are connected through the ASI interface, and the program data is transmitted to the transcoding and multiplexing server through the PCI bus. The main function of the server is to receive 4-channel MPEG-2 single-channel program transport stream, convert its video into H.264 video, and then multiplex it into a multi-channel program transport stream, remove empty packets, and rewrite the PID value and stream type field; extract and process any received PSI and service information (SI), and integrate it with such locally generated data; in addition, the system clock STC is also required to carry out re-identification processing of the program clock reference PGR. In order to complete the above functions and improve the working speed of the system as much as possible, the following points are considered in the specific implementation:

1) In order to prevent the host CPU from frequently accessing the PCI interface, the CPU reads FIFO data or writes FIFO using the half-full signal provided by FIFO. For the input FIFO, an interrupt is generated when it is half full, and the CPU responds to the interrupt, and reads the data in the FIFO into the memory buffer at one time; for the output FIFO, the situation is similar, and the FIFO is written to half full at one time.

2) Recognition of program sync word. To obtain the data of a program, you must first find the sync word of the TS stream data packet. Since the sync header does not meet the unique transparency principle, that is, the payload may be exactly its value, so it needs to be searched and detected.

3) Resolution of PID conflicts. PID is the unique identifier of the payload type in the TS stream. The PID values of different branch MPEG-2 code streams may be the same, if not modified, it will often lead to incorrect decoding. The solution is to rewrite the PID values according to certain rules when synthesizing TS streams. For example, if the PID of program 1 is 100, then every time a program is detected, the new PID will be increased by 1, and so on.

4) Modification of stream type. Since the video format of the input MPEG-2 TS stream is MPEG-2, and the video format of the recomposed TS stream is H.264, it is necessary to modify the stream type field of the PMT table accordingly. The flow type field is 0×02, and the modified flow type field is 0×1b.

TS stream demultiplexing

The process of demultiplexing and multiplexing of TS streams is just opposite, and its process is shown in Figure 4. The receiving end establishes a PAT table by detecting a packet with a PID of 0, and obtains the PID of the PMT table of each program contained in the TS stream from the PAT table, thereby establishing a PMT table. Finally, the PID of the audio and video package corresponding to each program is obtained from the PMT table. The receiving end puts the corresponding audio and video data into the buffer through these PIDs for decoding by the audio and video decoder.

Description of drawings

Figure 1 is a block diagram of the video transcoding algorithm from MPEG-2 to H.264.

Figure 2 MPEG-2 to H.264 video transcoder decision tree block diagram.

Fig. 3 is a block diagram of transcoding and multiplexing of multiple single-program transport streams.

Fig. 4 is a flowchart of TS stream demultiplexing.

Fig. 5 is a block diagram of the application of video transcoding in mobile digital TV.

FIG. 6 is a diagram showing the correspondence relationship between table PIDs of TS streams.

Detailed ways

In the mobile digital TV system based on MPEG-2, the video content mainly comes from the MPEG-2 program library, satellite TV, and live video programs. Multiple MPEG-2 program streams are multiplexed through the multiplexer and then channel coded. Modulation, and then digital TV wireless transmission.

After introducing the H.264-based video transcoding multiplexer, the system architecture is shown in Figure 5. It actually moves the MPEG-2 program library and MPEG-2 program stream forward. On the one hand, the H.264 video program library is established through static transcoding for selection by the playback system; The MPEG-2 program stream is dynamically transcoded in real time to reduce the bit rate of the video stream, change the spatial resolution and frame rate of the video stream, and adapt to the transmission requirements of the backend. After transcoding, multiple sets of H.264 programs are synthesized into one TS stream for transmission through software multiplexing.

The multi-channel single-program MPEG-2 TS stream is connected to the video transcoding multiplexer through the ASI interface, and the program data is transmitted to the transcoding multiplexing server through the PCI bus. The server receives multiple MPEG-2 single-channel program transport streams, converts its video into H.264 video, and then multiplexes it into a multi-channel program transport stream, and outputs it through the ASI interface.

In the input MPEG-2 single-channel program stream, the PID of the first program detected is 100, and every time a program is detected thereafter, the new PID is added by 1 when TS stream is synthesized. Since the video format of the input MPEG-2 TS stream is MPEG-2, and the video format of the recomposed TS stream is H.264, it is necessary to modify the stream type field of the PMT table accordingly. The type field is 0×02, and the modified stream type field is 0×1b.

Decision tree-based classification employed in the fast conversion method of MPEG-2 to H.264 codes:

The open source data mining tool WEKA is used to analyze the mean and variance, coding mode, and coding block type (CBPC) of MPEG-2 macroblock residuals to obtain the H.264 macroblock coding mode. The decision tree of this transcoder includes 3 WEKA decision trees, marked in gray in Figure 2. The first WEKA decision tree is used to distinguish skip, Intra, 8×8, and 16×16 modes. If it is 8×8 mode or 16×16 mode, use the second or third decision tree to determine the final decision of the MB model. The decision level of the mean and variance in the decision tree is calculated by the WEKA tool. A decision tree works as follows:

Node 1: Input to this node is an MPEG-2 encoded MB. By detecting the residual size of MPEG-2MB, MB encoding methods are divided into 4 categories: skip, Intra, 8×8 or 16×16. The Intra decision-making process is not discussed in the patent, and other cases need to be classified according to the previous classification. When building a decision tree, the following rules are used:

1) If the MC of MPEG-2MB is not coded, that is, there is no non-zero MV, the four 8×8 blocks have no coded coefficients. H.264MB will be encoded as 16×16. It is necessary to select the optimal mode through the secondary discrimination of the decision tree.

2) If the MPEG-2 MB is in intra mode, then in H.264, the MB is coded as intra or inter8×8. If it is encoded as intra, the algorithm terminates; if it is inter8×8, it needs to go through a second-level decision to select the optimal mode.

3) If MPEG-2MB is in skip mode, in H.264, this MB is also in skip mode.

Node 2: Input This node is 16×16MB separated from node 1. This node uses the second WEKA decision tree to classify the H.264MB mode (16×16, 16×8 or 8×16). Detect whether the 16×8 or 8×16 sub-block generates a better prediction. If it is judged as 16×8 or 8×16, it is the final coding mode. Otherwise, node 4 will continue to judge.

Node 3: The 8×8 MB split by Node 1 is input to this node. This node uses the third WEKA decision tree to select the optimal mode for the H.264 8×8 sub-macroblock: 8×8, 8×4, 4×8, 4×4. The decision tree is executed four times, and the four 8×8 sub-blocks in one macroblock are discriminated once, and this part only uses the mean value and variance of the four 4×4 blocks in the 8×8 sub-block.

Node 4: The input to this node is the skip pattern block split from node 1 or the 16×16 pattern block split from node 2. This node estimates the H.264 16×16 mode (excluding 16×8 and 8×16 modes), and the optimal mode is skip or inter16×16.

The judgment of the MB mode and the selection of the threshold value are determined by the quantization parameter (QP) of H.264. With the difference of the QP, the threshold value of the mean value and the variance are also different. There are two ways to solve this situation: 1) Generate a decision tree for each QP, and select the corresponding decision tree according to the QP value used during H.264 encoding; 2) Generate only one decision tree, according to the QP The value adjusts the threshold for the mean and variance. For the first method, 52 different decision trees need to be generated in one transcoder, and each needs 3 WEKA decision trees, so a total of 156 WEKA decision trees are needed. In H.264, the QP value has a certain relationship with the quantization step size. For every 6 increase in QP, the quantization step size doubles. Therefore, the threshold value of the mean and variance can be adjusted through this relationship. In this transcoder, the second method is used. A decision tree with QP of 25 is generated, and other QP values can be realized by adjusting the threshold level. When the QP increases by 6, the threshold value increases by 2.5%, otherwise it decreases by 2.5%.

In the demultiplexing of the TS stream at the receiving end, the PAT table is established by detecting the packet with a PID of 0, and the PMT table of the PMT table of each program contained in the TS stream is obtained from the PAT table, thereby establishing the PMT table. Finally, the PID of the audio and video package corresponding to each program is obtained from the PMT table, as shown in FIG. 6 . The receiving end puts the corresponding audio and video data into the buffer through these PIDs, and is decoded by the audio and video decoder.

Rewrite the PID value according to a certain rule when synthesizing TS streams. For example, if the PID of program 1 is 100, the new PID will be increased by 1 for each program detected in the future, and so on; Corresponding modification, the stream type field of MPEG-2 before modification is 0×02, and the stream type field after modification is 0×1b.

The basic data stream (ES) is packaged according to a certain format to form a PES package, and then some system information (such as service information (SI), system clock information, etc.) is added to form it.

PSI is used to describe the composition structure of the transport stream. In multiplexing, the PAT table shows how many sets of programs there are in a TS stream, and the corresponding relationship between it and the PMT table PID; the PMT table gives a set of programs. The specific composition of the program and the corresponding relationship with PIDs such as video and audio; and the modification of the stream type: since the video format of the input MPEG-2TS stream is MPEG-2, the video format of the re-synthesized TS stream is H.264 , modify the stream type field of the PMT table accordingly, the stream type field of MPEG-2 before modification is 0×02, and the stream type field after modification is 0×1b.

Multi-channel single-program MPEG-2 TS streams are connected through the ASI interface, and the program data is transmitted to the transcoding and multiplexing server through the PCI bus; the server receives 4-channel MPEG-2 single-channel program transport streams and converts its video into H .264 video, and then multiplex it into a transport stream of a multi-channel program, and remove empty packets, rewrite the PID value and stream type field; extract and process any received PSI and service information (SI), and combine it with the local The resulting data are integrated. In addition, it is necessary to use the system clock STC to re-identify the program clock reference PCR. When the TS stream is demultiplexed, the receiving end establishes the PAT table by detecting the PID of 0 packets, and obtains the PID of the PMT table of each program contained in the TS stream from the PAT table, thereby establishing the PMT table; The PID of the audio and video package corresponding to the program. The receiving end puts the corresponding audio and video data into the buffer through these PIDs, and is decoded by the audio and video decoder.

Claims (10)

1. Based on the H.264 multi-channel video transcoding and multiplexing method, it is characterized in that the input is multiple MPEG-2 single program streams, and the output is one H.264 multi-program stream, realizing video from MPEG-2 to H.264 Transcoding, demultiplexing and multiplexing of audio and video, and multiplexing of multiple H.264 programs. The video transcoding includes bit rate, resolution and format conversion; the video transcoding algorithm from MPEG-2 to H.264 adopts A fast conversion method from MPEG-2 to H.264 codes, using the correlation between H.264 macroblock mode selection and MPEG-2 motion compensation residuals, transforms the H.264 macroblock mode selection problem into a data classification problem, The motion compensation residual, MB mode, and coded block mode (CBPC) obtained by MPEG-2 decoding are directly mapped to the macroblock mode of H.264; when MPEG-2 code is decoded, relevant MB information is saved, including MB coding mode , coded block type (CBPC), mean and variance of MB residual, after decoding, use standard H.264 encoder to encode YUV image, and save H.264 MB encoding mode, use machine learning algorithm to obtain decision tree, use Classification of H.264 encoding modes; when decoding MPEG-2 code stream, obtain MPEG-2 MC residual, macroblock mode, coded block mode (CBPC), and calculate 4×4 sub-block MC residual Mean value and variance; Obtain the macroblock coding mode in H.264 through the decision tree; when encoding H.264, directly assign the coding mode of MB; the input of the H.264 encoder is the YUV data after MPEG-2 decoding and MB Coding mode: In motion estimation, use the MB coding mode obtained by the decision tree; realize adjustable code rate and resolution, and use different algorithms within and between frames; and synthesize TS streams, in multiple MPEG-2 program streams The transcoding and multiplexing server is input through the PCI bus through the ASI interface, and the single-channel H.264 video stream after transcoding and multiplexing is output through the PCI bus through the ASI interface. 2. The method for transcoding and multiplexing based on H.264 multi-channel video according to claim 1, characterized in that the CPU reads the FIFO data or writes the FIFO for the half-full signal provided by the FIFO; for the input FIFO, half-full When an interrupt is generated, the CPU responds to the interrupt and reads the data in the FIFO into the memory buffer at one time; for the output FIFO, writes the FIFO to half full at one time. 3. The method for transcoding and multiplexing based on H.264 multiplex video according to claim 1, characterized in that the PID value is rewritten when synthesizing TS streams. 4. The method for transcoding and multiplexing based on H.264 multi-channel video according to claim 1, characterized in that the TS stream is packaged according to a certain format by the encoded elementary data stream (ES) to form a PES package, and then added to the system At the sending end, the PES packaging of the basic stream is completed by the audio/video encoder, and the multiplexer receives the audio, video data stream and auxiliary data stream at the encoding end, and interleaves them into a single TS according to a certain multiplexing method Stream; add various time marks and system control information to the code stream; for the receiving end, it is just the opposite of the sending end process. 5. The method according to claim 1 based on H.264 multiplex video transcoding and multiplexing, characterized in that the transport stream can be composed of multiple programs, and each program can be composed of multiple streams, including video streams, Audio stream, Program Specific Information Stream (PSI); PSI has four types: Program Association Table (PAT), Program Mapping Table (PMT), Network Information Table (NIT) and Conditional Access Table (CAT); the multiplexer will The transcoded H.264 video and original audio are packaged in transport stream format. 6. The method for transcoding and multiplexing based on H.264 multi-channel video according to claim 5, wherein the length of the TS packet is 188 bytes, which is divided into two parts: header and packet load; the 4-byte prefix of the header is a link word Header, including the synchronization byte 0x47 and the data packet identification PID, judge the type of data loaded behind it from the PID, whether it is a video stream, audio stream, PSI or other data packets; the packet load is the actual content of the packet, and place the PES packet or PSI packet . 7. The method for multiplexing based on H.264 multi-channel video transcoding according to claim 1, characterized in that PSI is used to describe the composition structure of the transport stream, and one TS is given in the PAT table in the multiplexing How many sets of programs are there in the stream, and the corresponding relationship between it and the PID in the PMT table; the PMT table gives the specific composition of a set of programs and the corresponding relationship with PIDs such as video and audio; and the modification of the stream type: due to the input The video format of the MPEG-2 TS stream is MPEG-2, and the video format of the re-synthesized TS stream is H.264. The stream type field of the PMT table is modified accordingly. The stream type field of MPEG-2 before modification is 0x02, the modified stream type field is 0x1b. 8. The method for transcoding and multiplexing based on H.264 multi-channel video according to claim 1, characterized in that the multi-channel single-program MPEG-2 TS streams are connected through the ASI interface, and the program data is transmitted through the PCI bus. For the transcoding and multiplexing server; the server receives 4 MPEG-2 single-channel program transport streams, converts its video into H.264 video, and then multiplexes it into a multi-channel program transport stream, removes empty packets, and rewrites the PID Value and Flow Type fields; extract and process any received PSI and Service Information (SI), integrating it with locally generated such data. 9. The method for transcoding and multiplexing based on H.264 multiplexed video according to claim 8, characterized in that the system clock STC is also required to re-identify the program clock reference PCR. 10. The method for transcoding and multiplexing based on H.264 multi-channel video according to claim 1, characterized in that when the TS stream is demultiplexed, the receiving end establishes a PAT table by detecting a packet with a PID of 0, and obtains from the PAT table The TS stream contains the PID of the PMT table of each program, thereby establishing the PMT table; finally, the PID of the audio and video package corresponding to each program is obtained from the PMT table. The receiving end puts the corresponding audio and video data into the buffer through these PIDs, and is decoded by the audio and video decoder.

Images