CN103475934B - The Video coding flow control method that a kind of network-oriented is live - Google Patents

Abstract

The Video coding flow control method that a kind of network-oriented is live, including: play end parameter setting step: the duration playing end initial buffer district reception video data is preset as M, is preset as W by playing the minimum reception bandwidth of end；Initialize coding step: after live coding starts, set up simulated cushioned district according to default M, W；Coding step: in an encoding process, periodically prediction is just in size of data and the size of data of frame to be encoded of coded frame；The state of video sequence in calculating simulation buffering area, and adjust quantization parameter and make the data volume non-NULL of video sequence in simulated cushioned district, then proceed by next code；Coded data distributing step: while performing coding step, direct or indirect distributes coded data to broadcasting end；Video playback step: construct decoder at described broadcasting end, starts to receive live at any time, commences play out video, make the bandwidth of the reception live TV stream of described broadcasting end be not less than W simultaneously after receiving data and reach the video data volume of duration M.

Description

The Video coding flow control method that a kind of network-oriented is live

Technical field

The present invention is about a kind of video fluidics, specifically, is the Video coding live about a kind of network-oriented Flow control method and system.

Background technology

It is known that the form of Video service is commonly divided into download, program request, live etc..Download, demand (telecommunication) service feature is Being sent to established video data play end and play back, live broadcast service is then during video data produces in real time, Send to playing end and play back.

Video service can be based on multiple transmission network, such as traditional DVB-C, DMB-T etc., based on internet in recent years regards The business that is taken in small doses at short intervals progressively increases, and the video player on internet is usually the flash play control that PC browser is embedded, and ios operates The control that system is built-in, open source software VLC etc..

Stream control is the important step of Video coding, and its effect is typically the video after ensureing coding and transmission, decoded model Match, and the video quality having had.Video flowing control main models includes HRD, VBV, ABR etc. at present.

HRD (Hypothetical Reference Decoder) model, i.e. hypothetical reference decoding buffer model, including: Encoded picture buffer (CPB), instantaneous decoding process, decoded picture buffer (DPB) and output.

The major parameter of HRD model is defined as follows table:

hrd_parameters(){	C	Descriptor
			cpb_cnt_minusl	0	ue(v)
bit_rate_scale	0	u(4)
			cpb_size_scale	0	u(4)
for(SchedSelIdx=0；SchedSelIdx ＜=cpb_cnt_minus1；SchedSelIdx++){
			bit_rate_value_minus1[SchedSelIdx]	0	ue(v)
cpb_size_value_minus1[SchedSelIdx]	0	ue(v)
			cbr_flag[SchedSelIdx]	0	u(1)
}
			initial_cpb_removal_delay_length_minus1	0	u(5)
cpb_removal_delay_length_minusl	0	u(5)
			dpb_output_delay_length_minus1	0	u(5)
time_offset_length	0	u(5)
			}

Cpb_cnt_minus1+1 specifies the number of the CPB standard that can select, and codomain scope [0,31], when in SPS Not specifying cpb_cnt_minus1, giving tacit consent to this value is 0.Bit_rate_scale and bit_rate_value_minus1 [SchedSeIdx] together, it is intended that the maximum input bit rate of SchedSelIdx CPB.

Together with cpb_size_scale with cpb_size_value_minus1 [SchedSelIdx], it is intended that The CPBSize of SchedSelIdx CPB.

HRD model AUn represents the n-th access unit (access unit, AU) in decoding order, opens from AU0 Begin.HRD requires to verify code stream according to the following rules:

Combination to each BitRate and CPBSize, is required for verifying.The parameter of HRD model, according to SchedSelIdx, initial_cpb_removal_delay [Sched SelIdx], BitRate [SchedSelIdx] and CpbSize [SchedSelIdx] determines.During checking code stream compatibility, following condition need to be met:

1. for AUn, n > 0, define Δ t_g,90(n)=90000*(t_{R, n}(n)-t_af(n-1))

Initial_cpb_removal_delay [SchedSelIdx] needs to meet:

1) if cbr_flag [SchedSelIdx] is equal to 0, then

initial_cpb_removal_delay[SchedSelIdx]<=Ceil(Δt_g,90(n))。

2) otherwise (that is, when cbr_flag [SchedSelIdx] is equal to 1), Floor (Δ t_g,90(n))<=initial_ cpb_removal_delay[SchedSelIdx]<=Ceil(Δt_g,90(n))。

Noting: when any one-frame video data removes from CPB, in CPB, remaining bit number depends on initializing HRD Buffering period SEI message.Encoder must account for this point, and it is any to guarantee no matter decoder selects One buffering period SEI message initializes HRD, and the code stream of encoder output all can meet required pact Bundle condition.

2.CPB overflow represents that the total bit number in CPB has exceeded CPBSize, CPB never should overflow.

3.CPB underflow represents for some AUn, t_r,n(n)<t_af(n).When low_delay_hrd_flag is equal to 0, CPB never should underflow.

The subject matter of HRD model is:

(1) HRD need to provide corresponding initial buffer requirement to each coding AU, and player need to read this information from video flowing And just commence play out after being an actually-received the data volume mated with initial buffer requirement；

(2) need the decoding time (remove time) of dynamic calculation each frame/field during HRD coding, solve simultaneously Code device also needs to control the decoding time of each frame/field data by this value；

By above-mentioned 2 contents it can be seen that HRD model adds the complexity of decoding end, current internet is common The video players such as flash, ios, VLC are not easy to realize the support to this encoding model.

For above-mentioned (2nd) point, also add the complexity of coding side, simultaneously because to remove time in HRD model The adjustment of numerical value, also can make video playback fluency be affected, and the program that particularly motion amplitude is bigger, such as physical culture, joy Happy etc..

In practical business, VBV (Video Buffer Verifier, video buffer verifier) model is more used, Its basic characteristics be the initial buffer of player be fixing data volume (vbv buffer size): VBV model regulation player During startup, from the beginning of the first frame data, by speed R (n)=Rmax, to VBV buffer fillings data, filling process continues vbv_ The delay second, vbv_delay is defined in picture header, and now decoding process starts.

But at present the major video player (flash player, VLC) of internet be generally based on buffered data time Between, i.e. buffering area the first frame data and the time difference of last frame data rather than data volume (i.e. how many bit), at the beginning of arranging Beginning buffering area, i.e. the time tolerance of the video data of buffering area commences play out after reaching setting value again.Use VBV model User is when live broadcast service, if calculate the buffer time of buffering area by vbv buffer size/vbv bitrate, and Being set to initial transmission duration, also cannot ensure smoothness, because in practical business, the bandwidth of code stream is non-constant, so Player is buffered, by the vbv buffer size/vbv bitrate time, the video sequence obtained, and its data volume cannot ensure Equal to vbv buffersize.If the strictest restriction vbv buffer size, it is possible to the possibility of card is greatly reduced, But coding quality also can be greatly reduced.

HRD, VBV model more of towards tradition broadcasting and TV business, in actual application its decoding end frequently be ASIC core The schemes such as sheet.

In Video coding, more use also has ABR (Average Bit Rate, mean bit rate) Flow Control Model, ABR Model does not has strict definition, and concrete methods of realizing is the most different, it is common that the most strictly limit the fluctuation range of code stream, but stream controls mesh Mark is so that the average bit rate of video is drawn close to given target bit rate, and ABR model is usually used in the program of not face transmission and compiles/turn Code.

Summary of the invention

The present invention provides the Video coding flow control method that a kind of network-oriented is live, ensures that internet video is live with this Fluency.

To achieve these goals, the embodiment of the present invention provides the Video coding flow control method that a kind of network-oriented is live, This Video coding flow control method includes:

Play end parameter setting step: the duration playing end initial buffer district reception video data is preset as M, will play Minimum reception bandwidth is held to be preset as W；

Initialize coding step: after live coding starts, set up simulated cushioned district according to default M, W.

Coding step: in an encoding process, periodically prediction is just at size of data and the number of frame to be encoded of coded frame According to size；Calculate the state of video sequence in described simulated cushioned district, and adjust quantization parameter and make in described simulated cushioned district The data volume non-NULL of video sequence, then proceeds by next code；

Coded data distributing step: while performing described coding step, direct or indirect divides to described broadcasting end Send out coded data；

Video playback step: construct decoder at described broadcasting end, can put at any time and start to receive live, receive Data commence play out video after reaching the video data volume of duration M, make the bandwidth receiving live TV stream of described broadcasting end not simultaneously Less than W.

In one embodiment, described method also includes: in initializing coding step, reaches described M at coding duration After, the video data of coding is put into described simulated cushioned district, the data volume of the video sequence in simulated cushioned district will be introduced into It is defined as SN, now, SN=0.

In one embodiment, described method also includes: in initializing coding step, the average bandwidth arranging video is A, and with A for target bandwidth, encode by ABR model.

In one embodiment, described method also includes: in the encoding step, with A for target bandwidth, based on ABR model Calculate quantization parameter, adjust quantization parameter according to simulated cushioned district numerical value the most again.

In one embodiment, described method also includes: in the encoding step, after each frame Video coding completes, it was predicted that Just in size of data and the size of data of frame to be encoded of coded frame；Calculate the state of video sequence in described simulated cushioned district, And adjust quantization parameter and make the data volume non-NULL of video sequence in described simulated cushioned district, then proceed by next code；

Further, described coding step also includes:

Step 1: after the i-th frame coding, by live limited data transmission model calculate live under the conditions of described mould Intend buffer state；Limited data transmission model: video sequence is when transmission, and transmission process should meet transmission data volume and be less than Transmission bandwidth limits, and the difference of currently transmitted video time stamp video time stamp initial with transmission, after deducting transmission process time, It is not to be exceeded a predetermined constant.

Described limited data transmission model is:

SN_i=SN_i-1+F_i,

If SN_i＞ D_i* W, then SN_i-=D_i* W,

Otherwise SN_i=0；

Wherein, SN_iFor being introduced into the data volume of the i-th frame video sequence in described simulated cushioned district, F_iFor i-th frame of video Size of data；The duration that i-th frame of video is corresponding:Unit is the second；During the DTS of i-th frame of video Between stamp for DTS_i。

Further, described prediction is just in size of data and the size of data of frame to be encoded of coded frame；Calculate described mould Intend the state of video sequence in buffering area, and adjust quantization parameter and make the data volume of video sequence in described simulated cushioned district non- Sky, including:

Step 2: set frame of video F to be encoded_i+a+1The value of intermediate parameters q of quantization parameter:

$q_{i+a+1}=q_{i+a}+q_{i+a}\&times;({\mathrm{\&sigma;}}_i-A*\underset{j=0}{\overset{i}{\mathrm{\&Sigma;}}}{D}_j)/(A*\max (1,\sqrt{\underset{j=0}{\overset{i}{\mathrm{\&Sigma;}}}{D}_j}))$

According to q calculating quantization parameter: $Q{P}_{i+a+1}=12+6*\mathrm{lo}{g}_2\left(\frac{q_{i+a+1}}{0.85}\right)$

Wherein, a is the quantity of current encoder system in parallel coded frame, and usual span is between 4-8；B is 10～30 Between constant, the sequence of frames of video length of presentation code forecast analysis, the present embodiment is set to 20；From the 1st frame to the i-th frame one The accumulation size of data of i frame is σ altogether_i=σ_i-1+F_i。

Step 3: based on q_i+1To q_i+a, video residual error and Video coding historical information predict just in the size of coded frame: F_i+1、F_i+2、......、F_i+a；

Step 4: δ is set_i+a+1=0；

Wherein, δ represents the degree that q adjusts.

Step 5: with q_i+a+1+δ_i+a+1* Δ is prediction q value, it was predicted that frame to be encoded: F_i+a+1、F_i+a+2、……、F_i+a+b；

Step 6: according to limited data transmission model SN '_i+a+b=SN_i+F_i+1+F_i+2+…+F_i+a+bIf, ${{\mathrm{SN}}^{\&prime;}}_{i+a+b}>\frac{({\mathrm{DTS}}_{i+a+b}-{\mathrm{DTS}}_i)}{{10}^7}*W,$ Then ${{\mathrm{SN}}^{\&prime;}}_{i+a+b}-=\frac{({\mathrm{DTS}}_{i+a+b}-{\mathrm{DTS}}_i)}{{10}^7}*W;$ Otherwise SN '_i+a+b=0；DB′_i+a+b= M-SN′_i+a+b/ W；As DB '_i+a+bDuring ＞ 0, recalculate quantization parameter So After carry out the coding of subsequent frame.

Wherein, time a length of DB of video sequence in buffering area is simulated after i-th coding video frames completes_i, it was predicted that i-th Coding video frames completes time a length of DB ' of rear video sequence_i。

In one embodiment, in described i-th frame coding step, as described DB '_i+a+b＜=0, then δ_i+a+1+=1；If δ_i+a+1 ＜ Max Δ/Δ, performs above-mentioned steps 5 and step 6 again.

Wherein, Max Δ is pre-set constant, and the q representing the most maximum adjusts amplitude, can be set to 20, and Δ is normal Amount, represents the minimum of a value of q change, can be set to 0.05.

Further, if δ_i+a+1>=Max Δ/Δ, provides the underflow prompting of simulated cushioned district, assignment δ simultaneously_i+a+1=Max Δ/ Δ, then recalculates quantization parameterThen the volume of subsequent frame is carried out Code.

In one embodiment, in described coding step, as described DB '_i+a+b＜=0, then define δ_i+a+1It is mapped to DB′_i+a+bFunction be: y=f (x)；Y implication is δ_i+a+1During=x, the DB ' calculated by preceding method_i+a+bValue, x is integer；

Definition P-1 rank multinomial: y=g (x)=t₀+t₁*x+t₂*x²+…+t_p-1*x^p-1, carry out approach f (x) with g (x)；P For predicting the target order of fitting function, for constant, 5 ＜ P ＜ Max Δs/Δ.

Y=g (x) is calculated sample point, and is fitted: make x=0,1,2 ..., P-1, calculate y=y₀, y₁, y₂..., y_p-1, At y=y₀, y₁, y₂..., y_p-1Calculating during, if for the first time y occurring_k＞ 0, then make δ_i+a+1=k, recalculates quantization parameterThen the coding of subsequent frame is carried out；

Otherwise, if all y_k＜=0, by x=0,1,2 ..., P-1, and the y=y of correspondence₀, y₁, y₂..., y_p-1, as Reference sample point, carries out least square fitting, draws each term coefficient of polynomial function: t₀, t₁, t₂..., t_p-1；

Seek minimum integer x,Inequality g (x) ＞ 0 is set up；

By Newton-decline method to equation g (x)=t₀+t₁*x+t₂*x²+…+t_p-1*x^p-1=0 is iterated solving, wherein with X=P is iterative initial value, and 1 is iteration precision, and concentrates at solution of equation, chooses minimum of a value x of satisfied following 2 conditions_root:

1) in intervalIn；

2) first derivative g ' (the x)=t to g (x)₁+t₂*2x+…+t_p-1*(p-1)*x^p-2, and second dervative g of g (x) " (x)=t₂*2+t₃*3x+…+t_p-1*(p-1)*(p-2)*x^p-3, meet: g ' (x_root) ≠ 0 or g " (x_root)>0；

If there is not the x meeting condition_root, then buffer underflow prompting, assignment δ simultaneously are given_i+a+1=Max Δ/Δ, so After recalculate quantization parameter ${\mathrm{QP}}_{i+a+1}=12+6*{\log }_2\left(\frac{q_{i+a+1}+{\mathrm{\&delta;}}_{i+a+1}*\mathrm{\&Delta;}}{0.85}\right),$ Then the coding of subsequent frame is carried out.

Further, if there is x_root, then:

If g ' is (x_root) > 0, then make δ_i+a+1=INT(x_root)+1；If g ' is (x_root) < 0, then make δ_i+a+1=INT(x_root)；If g ' (x_root)=0, then g " (x_root) > 0, make δ_i+a+1=INT(x_root)；

Recalculate quantization parameter ${\mathrm{QP}}_{i+a+1}=12+6*{\log }_2\left(\frac{q_{i+a+1}+{\mathrm{\&delta;}}_{i+a+1}*\mathrm{\&Delta;}}{0.85}\right),$ Then the volume of subsequent frame is carried out Code；

Wherein, INT (x) represents the maximum integer being not more than x.

In one embodiment, described method also includes, in video playback step, to flash player, by arranging The member property NetConnection::bufferTime of NetConnection class specifies flash play control commencing play out Before, need to cache the video data of M duration, specifically include:

Varnc:NetConnection=new NetConnection；

Var flvbuftime:Number=stage.loaderInfo.parameters [" buftime "]；

Nc.bufferTime=flvbuftime, flvbuftime are M.

Having the beneficial effects that of the embodiment of the present invention:

Stream control target is combined with average bandwidth, i.e. ensures to play the smooth of end, also ensure that average discharge is less than Expection, can control to distribute the flow cost of link.

Owing to invention defines data volume SN of the video sequence being introduced into simulated cushioned district, every frame video need not be recorded Specifying information, this specifying information is including but not limited to the timestamp of every frame video, byte number, frame type etc., it is only necessary to record is not The byte number in the simulated cushioned district sent, can calculate the duration of video in simulated cushioned district.

By the Video coding flow control method of the present invention, coding side is without using HRD and VBV model, not changing broadcasting end The decoding time, can realize play end smooth broadcasting.

Employing limited data transmission model is to adapt to towards video real-time live broadcast business, as used traditional mode, Then during reality is live, owing to coding side data are to produce in real time, therefore can cause sending data volume not enough, Jin Eryu Simulation process produces deviation, and causes the risk of actual appearance card.

The Video coding flow control method of the present invention simplifies flow control algorithm, particularly without being decoded counting again of time Calculate, it is simple to the realization of coding side, the video simultaneously playing end also can be play by video time stamp, it is ensured that result of broadcast.

The Video coding flow control method of the present invention uses the time span of buffers video data rather than data volume is as initially Buffering foundation, it is easy to the realization of internet main flow player, and need not be able to use the strictest while ensureing fluency Stream control limits, and then ensure that video quality.

The invention can ensure that broadcasting end random time point after live coding starts starts to receive video flowing, all can ensure Smooth playing.

The Flow Control Model of the present invention does not consider that the overflow that conventional flow control model considers limits, and simplifies flow control procedure.

Accompanying drawing explanation

In order to be illustrated more clearly that the embodiment of the present invention or technical scheme of the prior art, embodiment will be described below The accompanying drawing used required in is briefly described, it should be apparent that, the accompanying drawing in describing below is only some of the present invention Embodiment, for those of ordinary skill in the art, on the premise of not paying creative work, it is also possible to according to these Accompanying drawing obtains other accompanying drawing.

Fig. 1 is the Video coding flow control method flow chart that embodiment of the present invention network-oriented is live；

Fig. 2 is the method flow diagram after the embodiment of the present invention every frame Video coding.

Detailed description of the invention

Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete Describe, it is clear that described embodiment is only a part of embodiment of the present invention rather than whole embodiments wholely.Based on Embodiment in the present invention, it is every other that those of ordinary skill in the art are obtained under not making creative work premise Embodiment, broadly falls into the scope of protection of the invention.

As it is shown in figure 1, the embodiment of the present invention provides the Video coding flow control method that a kind of network-oriented is live, this video is compiled Code flow control method includes:

S101: play end parameter setting step: the duration playing end initial buffer district reception video data is preset as M, It is preset as W by playing the minimum reception bandwidth of end；

S102: initialize coding step: after live coding starts, set up simulated cushioned district according to default M, W；

S103: coding step: in an encoding process, periodically prediction is just at size of data and the frame to be encoded of coded frame Size of data；Calculate the state of video sequence in described simulated cushioned district, and adjust quantization parameter and make described simulated cushioned In district, the data volume non-NULL of video sequence, then proceeds by next code；

S104: coded data distributing step: while performing coding step, direct or indirect divides to described broadcasting end Send out coded data；

S105: video playback step: construct decoder at described broadcasting end, reaches the video counts of duration M in reception data Commence play out video according to after amount, make the bandwidth receiving live TV stream of described broadcasting end be not less than W simultaneously.

By the flow process of Fig. 1, in the embodiment of the present invention, the duration of end initial buffer district reception video data will be play And play the minimum reception bandwidth of end, video frame rate is preset as respectively M and W；Simulated cushioned district is set up according to default M, W；Week The prediction of phase property just video sequence in the size of data of coded frame and the size of data of frame to be encoded calculating simulation buffering area State, adjust quantization parameter make the data volume non-NULL of video sequence in described simulated cushioned district；Performing described coding step Distribute coded data to described broadcasting end while Zhou, and playing end structure decoder, reach regarding of duration M in receiving data Frequently commence play out video after data volume, and make the bandwidth receiving live TV stream of broadcasting end be not less than W.Play end simple structure, Need not rely on HRD, VBV model, need not change the decoding time playing end, Flow Control Model limits regardless of overflow, sets up mould Intend also need not recording during buffering area the specifying information of every frame video, it is only necessary to the byte number in the simulated cushioned district that record does not sends, i.e. The live smoothness of internet video can be realized.

When S101 is embodied as, before live beginning, need to arrange following two parameters: play end initial buffer district and receive The duration M of video data and broadcasting minimum reception bandwidth W of end.It is preferred that coding average bandwidth A can also be arranged.

When S102 is embodied as, after live coding starts, set up simulated cushioned district according to default M, W, and when coding After length reaches described M, being put into by the video data of coding in described simulated cushioned district B, now in B, video sequence duration DB is M, is defined as SN by the data volume being introduced into the video sequence in simulated cushioned district, now, and SN=0.Due to this step defines into Enter data volume SN of the video sequence in simulated cushioned district, the specifying information of every frame video need not be recorded, this specifying information comprise but It is not limited to the timestamp of every frame video, byte number, frame type etc., it is only necessary to the byte number in the simulated cushioned district that record does not sends, i.e. The duration of video in simulated cushioned district can be calculated.

In this step, initial simulated cushioned district is set to the video sequence with M duration, rather than a fixing video Data volume.In this step, it is also possible to the average bandwidth arranging video is A, and with A for target bandwidth, compiles by ABR model Code.

After S102 completes, tri-steps of S103, S104 and S105 can be entered, from the flow chart of Fig. 1 it can be seen that Tri-steps of S103, S104 and S105 do not have sequencing, and this three steps are described separately below:

In S103, after each frame Video coding completes, need prediction just at size of data and the frame to be encoded of coded frame Size of data, the state of video sequence in calculating simulation buffering area, and adjust quantization parameter and make in described simulated cushioned district The data volume non-NULL of video sequence, the most just carries out next code.

Concrete, when adjusting quantization parameter, with A for target bandwidth, first calculate quantization parameter by ABR model, the most again Quantization parameter is adjusted according to simulated cushioned district numerical value.Stream control target is combined with average bandwidth, i.e. ensures smooth, also ensure Average discharge, less than expection, can control to distribute the flow cost of link.

During as in figure 2 it is shown, S103 is embodied as, coding step can also include:

S201: after the i-th frame coding, by live limited data transmission model calculate live under the conditions of simulated cushioned Zone state, limited data transmission model: video sequence is when transmission, and transmission process should meet transmission data volume less than transmission belt Tolerance system, and the difference of currently transmitted video time stamp video time stamp initial with transmission, after deducting transmission process time, should not surpass Cross a predetermined constant.This limited data transmission model is:

SN_i=SN_i-1+F_i,

If SN_i＞ D_i* W, then SN_i-=D_i* W,

Otherwise SN_i=0；

Wherein, SN_iFor being introduced into the data volume of the i-th frame video sequence in described simulated cushioned district, F_iFor i-th frame of video Size of data；The duration that i-th frame of video is corresponding:Unit is the second；The DTS time of i-th frame of video Stamp is DTS_i。

Due to the fact that employing limited data transmission model, be adapted to towards video real-time live broadcast business.As used biography The mode of system, then, during reality is live, owing to coding side data are to produce in real time, therefore can produce data volume not Foot, and then produce deviation with simulation process, and cause the risk of actual appearance card.Therefore, the drawing of limited data transmission model Enter to solve the above-mentioned problems in the prior art.

After S201, it was predicted that just in size of data and the size of data of frame to be encoded of coded frame, as in figure 2 it is shown, in advance Survey just in size of data and the size of data of frame to be encoded of coded frame, calculate the shape of video sequence in described simulated cushioned district State, and adjust quantization parameter and make the data volume non-NULL of video sequence in described simulated cushioned district, specifically include S202, S203, S204 and S205:

S202: need by abr mode, sets frame of video F to be encoded_i+a+1Q value:

$q_{i+a+1}=q_{i+a}+q_{i+a}\&times;({\mathrm{\&sigma;}}_i-A*{\mathrm{\&Sigma;}}_{j=0}^i{D}_j)/(A*\max (1,\sqrt{{\mathrm{\&Sigma;}}_{j=0}^i{D}_j}));$

According to q calculating quantization parameter: ${\mathrm{QP}}_{i+a+1}=12+6*\mathrm{lo}{g}_2\left(\frac{q_{i+a+1}}{0.85}\right).$

Wherein, a is the quantity of current encoder system in parallel coded frame, and usual span is between 4-8；B is 10～30 Between constant, the sequence of frames of video length of presentation code forecast analysis, the present embodiment is set to 20；From the 1st frame to the i-th frame one The accumulation size of data of i frame is σ altogether_i=σ_i-1+F_i。

S203: based on q_i+1To q_i+a, video residual error and Video coding historical information predict just at the size of coded frame: F_i+1、 F_i+2、......、F_i+a；

S204: δ is set_i+a+1=0。

Wherein, δ represents the degree that q adjusts.

S205: with q_i+a+1+δ_i+a+1* Δ is prediction q value, it was predicted that frame to be encoded: F_i+a+1、F_i+a+2、......、F_i+a+b。

As in figure 2 it is shown, in S103, the duration of video sequence in calculating simulation buffering area, specifically include:

S206: after drawing, according to S203 and S205, the prediction size that i+1 arrives these frames of i+a+b, passes according to finite data Defeated model SN'_i+a+b=SN_i+F_i+1+F_i+2+…+F_i+a+bIf,ThenOtherwise SN'_i+a+b=0；DB′_i+a+b=M-SN'_i+a+b/W。

As in figure 2 it is shown, in S103, in calculating simulation buffering area after the duration of video sequence, also include:

S207: judge DB '_i+a+bWhether ＞ 0 sets up, if it is, enter S208；If DB '_i+a+b＜=0, then, and δ_i+a+1+=1； Judge δ_i+a+1With the size (S210) of Max Δ/Δ, if δ_i+a+1＜ Max Δ/Δ, then with q_i+a+1+δ_i+a+1* Δ is prediction q value, in advance Survey frame to be encoded: F_i+a+1、F_i+a+2、......、F_i+a+b；According to limited data transmission model SN '_i+a+b=SN_i+F_i+1+F_i+2+…+ F_i+a+bIf, ${{\mathrm{SN}}^{\&prime;}}_{i+a+b}>\frac{({\mathrm{DTS}}_{i+a+b}-{\mathrm{DTS}}_i)}{{10}^7}*W,$ Then ${{\mathrm{SN}}^{\&prime;}}_{i+a+b}-=\frac{({\mathrm{DTS}}_{i+a+b}-{\mathrm{DTS}}_i)}{{10}^7}*W;$ Otherwise SN '_i+a+b=0； DB′_i+a+b=M-SN′_i+a+b/W；As DB '_i+a+bDuring ＞ 0, recalculate quantization parameter Then the coding of subsequent frame is carried out.

If δ_i+a+1>=Max Δ/Δ, then enter S209, then jump to S208.Wherein, Max Δ is pre-set normal Amount, the q representing the most maximum adjusts amplitude, and Δ is constant, represents the minimum of a value of q change.

S208: recalculate the quantization parameter of frame of video ${\mathrm{QP}}_{i+a+1}=12+6*{\log }_2\left(\frac{q_{i+a+1}+{\mathrm{\&delta;}}_{i+a+1}*\mathrm{\&Delta;}}{0.85}\right),$ Then open Begin the actual coding of a new frame, repeats S103 afterwards.

S209: provide buffer underflow prompting, represents that this code stream exists pause possible, assignment δ simultaneously_i+a+1=MaxΔ/Δ。

S210: judge δ_i+a+1Size with Max Δ/Δ.

In a preferred embodiment, in order to reduce amount of calculation, S207, S209 can be amended as follows:

As DB '_i+a+b＞ 0, enters S208；

As described DB '_i+a+b＜=0, then define δ_i+a+1It is mapped to DB '_i+a+bFunction be: y=f (x)；Y implication is δ_i+a+1=x Time, through the DB ' that S205 and S206 calculates_i+a+bValue, wherein, x implication is δ_i+a+1, x is integer.

From S205 and S206, when x is the biggest, it was predicted that q value is the biggest, from H.264 coding criterion, it was predicted that to be encoded Frame F_i+a+1、F_i+a+2、......、F_i+a+bThe least, SN'_i+a+bThe least, therefore DB '_i+a+bThe biggest, i.e. y is the biggest.Namely Say that function y=f (x) is monotonically increasing function.

For reducing amount of calculation, rapid solving meets DB '_i+a+bThe minimum δ of ＞ 0_i+a+1, definition P-1 rank multinomial: y=g (x)= t₀+t₁*x+t₂*x²+…+t_p-1*x^p-1, carry out approach f (x) with g (x)；

To y=g (x), calculate sample point, and be fitted: make x=0,1,2 ..., P-1, calculate y=y₀, y₁, y₂..., y_p-1, at y=y₀, y₁, y₂..., y_p-1Calculating during, if for the first time y occurring_k＞ 0, then make δ_i+a+1=k, enters S208, weight Newly calculate quantization parameter Then carrying out i+1 frame coding, wherein, x contains Justice is δ_i+a+1, x is integer, and P is the target order of prediction fitting function, for constant, 5 ＜ P ＜ Max Δs/Δ.

Otherwise, if all y_k＜=0, by x=0,1,2 ..., P-1, and the y=y of correspondence₀, y₁, y₂..., y_p-1, as Reference sample point, carries out least square fitting, draws each term coefficient of polynomial function: t₀, t₁, t₂..., t_p-1；

For calculating target, find the δ of minimum_i+a+1So that DB '_i+a+b＞ 0, and meetPermissible Mathematical description is:

Seek minimum integer x,Inequality g (x) ＞ 0 is set up；

After obtaining approximating function g (x) of f (x), calculate target and be further simplified as: seek minimum integer x,Inequality g (x) ＞ 0 is set up.

The method solved according to unitary inequality of higher degree, the disaggregation space of inequality g (x) ＞ 0 is by the solution of equation g (x)=0 Collection, sign of inequality direction and g (x) first derivative at equattion root determines, first to g (x)=0 equation solution.

Then, by Newton-decline method to equation g (x)=t₀+t₁*x+t₂*x²+…+t_p-1*x^p-1=0 is iterated solving (wherein with x=P as iterative initial value, 1 is iteration precision), and concentrate at solution of equation, choose the minimum of a value of satisfied following 2 conditions x_root:

1) in intervalIn；

2) first derivative g to g (x) ' (x)=t₁+t₂*2x+…+t_p-1*(p-1)*x^p-2, and second dervative g of g (x) " (x)=t₂*2+t₃*3x+…+t_p-1*(p-1)*(p-2)*x^p-3, meet: g ' (x_root) ≠ 0 or g " (x_root)>0；

If there is not the x meeting condition_root, then buffer underflow prompting, assignment δ simultaneously are given_i+a+1=Max Δ/Δ, so Rear entrance S208 recalculates quantization parameterThen i+1 frame volume is carried out Code.

If there is the x meeting condition_root, then:

If g ' is (x_root) > 0, then make δ_i+a+1=INT(x_root)+1；If g ' is (x_root) < 0, then make δ_i+a+1=INT(x_root)；If g ' (x_root)=0, then g " (x_root) > 0, make δ_i+a+1=INT(x_root)；Quantization parameter is recalculated subsequently into S208Then i+1 frame coding is carried out；Wherein, INT (x) represents no more than x Maximum integer.

S104 performs while S103 performs, the actual video data produced by coding, to playing end distribution, so that broadcasting Put end and play video.

Broadcasting end can receive live video stream to be carried out while performing S103, S104, when S105 is embodied as, plays end Structure decoder so that it is when receiving live at any time, just should open after reception data reach the video data volume of duration M Beginning to play video, player removes video data by the DTS timestamp of video from buffering area and decodes, and the most also should ensure that broadcasting The bandwidth receiving live TV stream of end is not less than fixed value W.

S105, in video playback step, can use flash play control, to flash player, with q_i+a+1+ δ_i+a+1* △ is prediction q value, it was predicted that frame to be encoded: F_i+a+1、F_i+a+2、......、F_i+a+bStep in, by arrange The member property NetConnection::bufferTime of NetConnection class specifies flash play control commencing play out Before, need to cache the video data of M duration, specifically include:

By in M assignment to bufferTime:

Varnc:NetConnection=new NetConnection；

Var flvbuftime:Number=stage.loaderInfo.parameters [" buftime "]；

Nc.bufferTime=flvbuftime, flvbuftime are M.

After as above step, it is ensured that when broadcasting end puts reception live video at any time, all can ensure that buffering area is non- Sky, gets final product smooth playing.

It addition, the present invention is also equipped with following technique effect:

By the Video coding flow control method of the present invention, coding side is without using HRD model, not changing and play the decoding of end Time, can realize playing the smooth broadcasting of end.

The Video coding flow control method of the present invention simplifies flow control algorithm, particularly without being decoded counting again of time Calculate, it is simple to the realization of coding side, the video simultaneously playing end also can be play by video time stamp, it is ensured that result of broadcast.

The Video coding flow control method of the present invention uses the time span of buffers video data rather than data volume is as initially Buffering foundation, it is easy to the realization of internet main flow player.

The Flow Control Model of the present invention does not consider that overflow limits, and simplifies flow control procedure.

One of ordinary skill in the art will appreciate that all or part of step realizing in above-described embodiment method can be led to Program of crossing completes to instruct relevant hardware, and this program can be stored in a computer read/write memory medium, such as ROM/RAM, magnetic disc, CD etc..

Particular embodiments described above, has been carried out the purpose of the present invention, technical scheme and beneficial effect the most in detail Describe in detail bright, be it should be understood that the specific embodiment that the foregoing is only the present invention, the guarantor being not intended to limit the present invention Protect scope, all within the spirit and principles in the present invention, any modification, equivalent substitution and improvement etc. done, should be included in this Within the protection domain of invention.

Claims (12)

1. the Video coding flow control method that a network-oriented is live, it is characterised in that described Video coding flow control method includes:

Play end parameter setting step: the duration playing end initial buffer district and receiving video data is preset as M, by broadcasting end Low reception bandwidth is preset as W；

Initialize coding step: after live coding starts, set up simulated cushioned district according to default M, W；

Coding step: in an encoding process, periodically prediction are just big in the size of data of coded frame and the data of frame to be encoded Little；Calculate the state of video sequence in described simulated cushioned district, and adjust quantization parameter and make video in described simulated cushioned district The data volume non-NULL of sequence, then proceeds by next code；Wherein calculate the state of video sequence in described simulated cushioned district Method as follows:

Step 1: after the i-th frame coding, by live limited data transmission model calculate live under the conditions of described simulation delay Rush zone state；Described limited data transmission model is:

SN_i=SN_i-1+F_i,

If SN_i＞ D_i* W, then SN_i-=D_i* W,

Otherwise SN_i=0；Wherein, F_iFor the size of data of i-th frame of video, the duration that i-th frame of video is corresponding:The DTS timestamp of i-th frame of video is DTS_i, SN_iIt is to be introduced into described simulation after the i-th frame encodes to delay Rush the data volume of the video sequence in district；

Coded data distributing step: while performing described coding step, direct or indirect compiles to the distribution of described broadcasting end Code data；

Video playback step: construct decoder at described broadcasting end, starts to receive live at any time, reaches in reception data After the video data volume of duration M, commence play out video, make the bandwidth receiving live TV stream of described broadcasting end be not less than W simultaneously.

Method the most according to claim 1, it is characterised in that described method also includes: in initializing coding step, After coding duration reaches described M, the video data of coding is put into described simulated cushioned district, simulated cushioned district will be introduced into The data volume of video sequence be defined as SN, now, SN=0.

Method the most according to claim 1, it is characterised in that described method also includes: in initializing coding step, The average bandwidth arranging video is A, and with A for target bandwidth, encodes by ABR model.

Method the most according to claim 1, it is characterised in that described method also includes: in the encoding step, presets and regards The average bandwidth of frequency is A, with A for target bandwidth, first calculates quantization parameter by ABR model, the most again according to simulated cushioned district Numerical value adjusts quantization parameter.

Method the most according to claim 1, it is characterised in that described prediction is just in the size of data of coded frame and to be encoded The size of data of frame, calculates the state of video sequence in described simulated cushioned district, and adjusts quantization parameter described simulation is delayed Rush the data volume non-NULL of video sequence in district, including:

Step 2: set frame of video F to be encoded_i+a+1The value of intermediate parameters q of quantization parameter:

According to q calculating quantization parameter:

Step 3: based on q_i+1To q_i+a, video residual error and Video coding historical information predict just at the size of coded frame: F_i+1、 F_i+2、……、F_i+a；

Step 4: δ is set_i+a+1=0；

Step 5: with q_i+a+1+δ_i+a+1* Δ is prediction q value, it was predicted that frame to be encoded: F_i+a+1、F_i+a+2、……、F_i+a+b；

Wherein, Δ is constant, represents the minimum of a value of q change, F_i+a+bBe the size of data of the i-th+a+b frame of video, from the 1st frame to The accumulation size of data of the i-th frame i frame altogether is σ_i=σ_i-1+F_i, a is the quantity of current encoder system in parallel coded frame, and b represents volume The sequence of frames of video length of code forecast analysis, δ represents the degree that q adjusts, and A is the average bandwidth of pre-setting video.

Method the most according to claim 5, it is characterised in that the shape of video sequence in the described simulated cushioned district of described calculating State, and adjust quantization parameter and make the data volume non-NULL of video sequence in described simulated cushioned district, sequel after then proceeding by Code, also includes:

Step 6: according to limited data transmission model: SN '_i+a+b=SN_i+F_i+1+F_i+2+…+F_i+a+bIf,ThenOtherwise SN '_i+a+b=0；DB′_i+a+b =M-SN '_i+a+b/W；As DB '_i+a+b> 0 time, recalculate quantization parameter Then the coding of subsequent frame is carried out；

Wherein, time a length of DB of video sequence in buffering area is simulated after i-th coding video frames completes_i, it was predicted that i-th video Frame has encoded time a length of DB ' of rear video sequence_i。

Method the most according to claim 6, it is characterised in that in described coding step, as described DB '_i+a+B≤0, then δ_i+a+1+=1；

If δ_i+a+1＜ Max Δ/Δ, performs step 5 and step 6；

Wherein, Max Δ is pre-set constant, and the q representing the most maximum adjusts amplitude.

Method the most according to claim 7, it is characterised in that if δ_i+a+1>=Max Δ/Δ, provides simulated cushioned district underflow and carries Show, simultaneously assignment δ_i+a+1=Max Δ/Δ, then recalculates quantization parameter Then the coding of subsequent frame is carried out.

Method the most according to claim 6, it is characterised in that in described coding step, as described DB '_i+a+b≤ 0, then Definition δ_i+a+1It is mapped to DB '_i+a+bFunction be: y=f (x)；Y implication is δ_i+a+1During=x, the DB ' calculated_i+a+bValue；

Definition P-1 rank multinomial: y=g (x)=t₀+t₁*x+t₂*x²+…+t_p-1*x^p-1, carry out approach f (x) with g (x)；

Y=g (x) is calculated sample point, and is fitted: make x=0,1,2 ..., P-1, calculate y=y₀, y₁, y₂..., y_p-1, Y=y₀, y₁, y₂..., y_p-1Calculating during, if for the first time y occurring_k＞ 0, then make δ_i+a+1=k, recalculates quantization parameter QP_i+a+1=12+6*Then i+1 frame coding is carried out；

Wherein, x implication is δ_i+a+1, x is integer, and P is the target order of prediction fitting function, for constant, 5 < P < Max Δs/Δ.

Method the most according to claim 9, it is characterised in that if all y_k＜=0, by x=0,1,2 ..., P-1, and Corresponding y=y₀, y₁, y₂..., y_p-1, as reference sample point, carry out least square fitting, draw each of polynomial function Term coefficient: t₀, t₁, t₂..., t_p-1；

Seek minimum integer x,Inequality g (x) ＞ 0 is set up；

By Newton-decline method to equation g (x)=t₀+t₁*x+t₂*x²+…+t_p-1*x^p-1=0 is iterated solving, wherein with x= P is iterative initial value, and 1 is iteration precision, and concentrates at solution of equation, chooses minimum of a value x of satisfied following 2 conditions_root:

A) in intervalIn；

B) first derivative g to g (x) ' (x)=t₁+t₂*2x+…+t_p-1*(p-1)*x^p-2, and second dervative g of g (x) " (x)= t₂*2+t₃*3x+…+t_p-1*(p-1)*(p-2)*x^p-3, meet: g ' (x_root) ≠ 0 or g " (x_root) ＞ 0；

If there is not the x meeting condition_root, then buffer underflow prompting, assignment δ simultaneously are given_i+a+1=Max Δ/Δ, then weighs Newly calculate quantization parameterThen i+1 frame coding is carried out.

11. methods according to claim 10, it is characterised in that if there is x_root, then:

If g ' is (x_root) ＞ 0, then make δ_i+a+1=INT (x_root)+1；If g ' is (x_root) ＜ 0, then make δ_i+a+1=INT (x_root)；If g ' (x_root)=0, then g " (x_root) ＞ 0, make δ_i+a+1=INT (x_root)；

Recalculate quantization parameterThen i+1 frame coding is carried out；

Wherein, INT (x) represents the maximum integer being not more than x.

12. methods according to claim 1, it is characterised in that described method also includes, in video playback step, To flash player, specify by arranging the member property NetConnection::bufferTime of NetConnection class Flash play control, before commencing play out, needs to cache the video data of M duration, specifically includes:

By in M assignment to bufferTime:

Varnc:NetConnection=new NetConnection；

Var flvbuftime:Number=stage.loaderInfo.parameters [" buftime "]；

Nc.bufferTime=flvbuftime, flvbuftime are M.