JP6053176B2 - Video playback state estimation device, video playback state estimation method, and program - Google Patents

Description

  The present invention relates to a technique for performing reproduction state estimation based on packet capture data passively acquired on a distribution path in progressive download video distribution that performs dynamic rate control.

  In recent years, broadband communication networks have been increasing, and more traffic has been distributed via the Internet. In particular, a progressive download type content distribution system that performs data processing in parallel while downloading content having a large file size has become widespread. In such a content distribution system, data does not arrive at the terminal at a constant speed due to delay variation, packet loss, retransmission, etc. through the network, and fluctuations in transmission timing occur when data arrives. For this reason, in order to absorb the fluctuation on the terminal side, a mechanism is often used in which a reproduction buffer is prepared and the influence of the network is hidden from the upper application.

  In such a content distribution system, when the network throughput temporarily becomes lower than the processing speed of the content data, the playback buffer may fall below a certain amount (playback stop threshold), and playback may be stopped. Since this playback stop greatly affects the user experience quality of the content, a technique for estimating a playback state such as a content playback stop state is required, and various techniques have been proposed (for example, non-patent documents). 1-3).

  In progressive download type content distribution, even a single content is often distributed in a plurality of units. Each divided unit is called a chunk, and such a distribution method is called a chunk-type content distribution method. An operation image of chunk type content distribution is shown in FIG. As shown in FIG. 1, a chunk is acquired from the received packet, stored in a playback buffer in units of chunk, and played (decoded).

Daisuke Ikegami, Yasunori Honda, Koji Yamamoto et al .: "Progressive download service stop time estimation method" IEICE Technical Report, vol. 111, no. 278, CQ2011-59, pp. 91-96, November 2011 . Daisuke Ikegami, Yasunori Honda, Koji Yamamoto: “Examination of Playback State Estimation Method for Chunk-type Video Distribution Service”, IEICE General Conference, B-11-24, March 2012. Yasutoshi Honda, Daisuke Ikegami, Koji Yamamoto: “A video observation state estimation method using Ack observation”, IEICE General Conference, B-11-25, March 2012. Z.I.Botev, J.F.Grotowski, and D.P.Kroese, Kernel density estimation via diffusion, Ann. Statist. 38 (2010), 2916-2957.

  In the prior art described in Non-Patent Documents 1 to 3, the reception byte amount time series of the terminal is grasped from the packet capture data, and input to a virtual buffer model simulating the reproduction buffer of the terminal, thereby estimating the reproduction state Is realized. By using these techniques, it is possible to estimate the playback state of the video of each session from given packet capture data.

  On the other hand, in recent years, with the spread of mobile terminals and tablet terminals, usage forms for viewing progressive download type video by wireless access such as 3G and LTE are increasing on terminals with specifications lower than those of personal computers. In this case, it is not uncommon for the reproduction state to frequently deteriorate due to the deterioration of the network environment accompanying the movement of the user and the fading of radio waves. For this reason, a technique for adaptively delivering video by increasing or decreasing the rate according to the environment has been established and is being discussed in the field of international standardization. The technology aims to solve the trade-off between stopping playback and ensuring video quality by dynamically increasing / decreasing the image quality (decoding rate) of content on the distribution server side based on, for example, estimation of round trip delay on the distribution path. Yes.

  However, the conventional techniques described in Non-Patent Documents 1 to 3 do not support video distribution that performs such adaptive rate control.

  The present invention has been made in view of the above points, and it is an object of the present invention to provide a technique that can appropriately estimate a video reproduction state even in a video distribution service that performs adaptive rate control.

According to an embodiment of the present invention, in a video distribution method for distributing video data from a distribution server to a terminal in units of chunks, a video playback state estimation device that estimates a playback state of video data in a terminal,
Image quality estimating means for estimating the image quality of each chunk based on the chunk length of each chunk acquired from the captured data of the video data;
Chunks are sequentially stored in a virtual buffer at each chunk observation time in the video data, and the chunk with the earliest observation time among the chunks in the virtual buffer is set as a reproduction target. And a buffer processing means for storing the information of the time and the image quality in a storage unit, while erasing and setting the duration of the chunk as the time during which the video of the image quality corresponding to the chunk is being reproduced. A video playback state estimation device is provided.

  According to an embodiment of the present invention, there is provided a technique capable of appropriately estimating a video playback state even in a video distribution service that performs adaptive rate control.

It is a figure which shows the operation | movement image of chunk type | mold delivery. It is a function block diagram of the video reproduction state estimation apparatus 100 which concerns on embodiment of this invention. It is a function block diagram of the video reproduction state estimation apparatus 200 which concerns on embodiment of this invention. It is a figure which shows the example which drawn the basic area and probability density function of the chunk length which belong to each image quality obtained by the chunk length density estimation part 1. FIG. It is a flowchart which shows the procedure 1 of image quality determination. It is a figure which shows the example of the image quality at the time of performing the procedure 1 and procedure 2 of image quality determination. 3 is a flowchart showing processing of a buffer processing unit 3. It is a figure which shows the example of a buffer process in case there exists an image quality change. It is a figure which shows the example of a reproduction | regeneration state estimation result. In Example 1 (download direction), it is a figure which shows the example of the sequence number which the chunk acquisition time estimation part 232 acquires. 6 is a sequence diagram of processing in Example 1. FIG. 10 is a diagram illustrating a determination example of Example 1. FIG. In Example 2 (upload direction), it is a figure which shows the example of the Ack number which the chunk acquisition time estimation part 232 acquires. FIG. 10 is a sequence diagram of processing in Example 2. 10 is a diagram illustrating a determination example of Example 2. FIG.

  Embodiments of the present invention will be described below with reference to the drawings. The embodiment described below is only an example, and the embodiment to which the present invention is applied is not limited to the following embodiment. In addition, specific numerical values shown in the following embodiments are merely examples, and are not limited thereto.

In the following description, α represents a margin between basic areas of chunk lengths for each image quality. N represents the number of image quality levels set for the target service. For example, N = 3 when there are three image quality settings for the target service: high image quality, medium image quality, and low image quality. M represents the number of chunks in the packet capture data. N _ij is the number of chunks in which the image quality i is reproduced in the virtual buffer model when the image quality i is changed to the image quality j in the buffer process.

(Device configuration)
<Video Playback State Estimation Device 100>
FIG. 2 is a functional configuration diagram of the video reproduction state estimation apparatus 100 according to the embodiment of the present invention. As shown in FIG. 2, the video playback state estimation device 100 according to the present embodiment includes a chunk length density estimation unit 1, an image quality change detection unit 2, a determination result storage unit 21, a buffer processing unit 3, and a playback state storage unit. 31 is provided. As a user of this apparatus, for example, a network provider that provides the service or a service provider is assumed. The details of the operation of each functional unit of the video reproduction state estimation device 100 will be described later, but the outline is as follows.

  The chunk length density estimation unit 1 receives the number of image quality levels of the target service, the chunk length distribution observation value (frequency distribution) of each image quality, and the margin parameter α, and the chunk length basic region and probability density belonging to each image quality Output function.

  The image quality change detection unit 2 receives the chunk information of the packet capture data (chunk length (byte amount), chunk observation time, etc. for each chunk), the basic area of the chunk length belonging to each image quality, and the probability density function, and inputs each chunk. Is estimated (determined), and a determination result (image quality for each chunk) is output. The determination result output by the image quality change detection unit 2 is stored in the determination result storage unit 21. In the determination result storage unit 21, for example, the chunk number and the image quality of the chunk are stored. The image quality change detection unit 2 is a functional unit that estimates image quality, and may be referred to as an image quality estimation unit.

The buffer processing unit 3 includes chunk information of the packet capture data (chunk length for each chunk, chunk observation time, etc.), the image quality for each chunk, the remaining chunk number matrix (N _ij ) of the old image quality when the image quality is changed, Using the duration (playback time length) per chunk and the viewing end time _TE as input, the playback state is estimated and the estimated playback state is output. The output playback state is stored in the playback state storage unit 31 together with the playback time information.

  In the present embodiment, the video playback state estimation device 100 includes the chunk length density estimation unit 1, and the video playback state estimation device 100 obtains the chunk length basic region and probability density function belonging to each image quality. The reproduction state estimation device 100 may not include the chunk length density estimation unit 1. That is, the basic region of the chunk length and the probability density function belonging to each image quality may be obtained outside the video reproduction state estimation apparatus 100 and given to the video reproduction state estimation apparatus 100 as inputs.

<Video Playback State Estimation Device 200: Example Including Chunk State Estimation Unit>
In the present embodiment, the packet is captured outside the video reproduction state estimation apparatus 100, chunk information such as chunk length and chunk observation time is calculated from the captured data, and the calculated value is stored in the video reproduction state estimation apparatus 100. Although given as an input, the video reproduction state estimation device 100 may include a function of capturing a packet and calculating chunk information from the captured data. FIG. 3 shows an example of the device configuration in this case (denoted as a video reproduction state estimation device 200). In FIG. 3, a functional unit corresponding to the video playback state estimation device 100 of FIG. 2 is denoted as a video playback state estimation unit 100.

  The video playback state estimation device 200 shown in the figure is connected to a distribution server 300 that performs chunk-based delivery and a terminal 400, and includes a packet data acquisition unit 210, a storage unit 220, a chunk state estimation unit 230, and a video playback state. The estimation unit 100 is included. The terminal 400 has at least a playback buffer and a playback application.

  In the present embodiment, the distribution server 300 and the terminal 400 communicate by TCP / IP, and the distribution server 300 divides one content into a plurality of chunks and distributes it to the terminal 400. Also, when viewed at Layer 4, the chunk is divided and transmitted to the terminal 400 as a packet (segment).

  The terminal 400 has a playback buffer. When a predetermined number of chunks is stored in the playback buffer, the terminal 400 starts playback. During playback, data necessary for playback is read from the playback buffer and stored in the playback buffer. When the number of chunks that have been played falls below the playback stop threshold, playback stops. When the number of chunks exceeds the playback start threshold, playback starts.

  The packet data acquisition unit 210 acquires packet data (including TCP packet header information) distributed from the distribution server 300 to the terminal 400 and stores it in the storage unit 220. In the present embodiment, the packet data acquisition unit 210 also acquires a TCP Ack packet transmitted from the terminal 400 to the distribution server 300 and stores it in the storage unit 220. The Ack packet may be referred to as a TCP packet in a broad sense.

  The chunk state estimation unit 230 includes a chunk time length estimation unit 231 and a chunk acquisition time estimation unit 232.

  The chunk time length estimation unit 231 creates a list of playback time lengths (durations) of each chunk based on packet data stored in the storage unit 220 and information such as a known content rule in advance. The reproduction time length list can be held as a one-dimensional array, for example, or may be held as a two-dimensional array corresponding to the chunk number.

  The chunk acquisition time estimation unit 232 estimates the time when reception of each chunk is completed from the packet data stored in the storage unit 220, creates a list of acquisition times (also referred to as observation times) of each chunk, Store in memory etc. The list of acquisition times can be stored as a one-dimensional array, for example, or may be stored as a two-dimensional array corresponding to the chunk number. Note that the chunk length (byte amount) for each chunk can be calculated from the acquisition time estimated here.

  In the present embodiment, the chunk acquisition time estimation unit 232 estimates the acquisition time of each chunk by using the TCP header information, details of which will be described later.

  The video playback state estimation device 200 has a configuration including a packet data acquisition unit 210, a storage unit 220, a chunk state estimation unit 230, and a video playback state estimation unit 100. In addition to this, the video playback state estimation device The configuration may be such that a chunk state estimation unit 230 is added to 100 (FIG. 2). That is, in this case, packet data is captured externally, and the video playback state estimation apparatus 100 performs chunk state estimation and video playback state estimation.

  The video reproduction state estimation apparatus according to the present embodiment can be realized by causing a computer to execute a program describing the processing contents described in the present embodiment. More specifically, the function of each unit of the video playback state estimation device is implemented by each unit using hardware resources such as a CPU, a memory, and a hard disk built in the computer constituting the video playback state estimation device. It can be realized by executing a program corresponding to the process. The program can be recorded on a computer-readable recording medium (portable memory or the like), stored, or distributed. It is also possible to provide the program through a network such as the Internet or electronic mail.

(Operation example of video reproduction state estimation apparatus 100)
Hereinafter, an operation example of the video reproduction state estimation apparatus 100 (FIG. 2) will be described in detail.

<Chunk length density estimation unit 1>
The user of this apparatus inputs the number of image quality of the target service, the chunk length distribution observation value (frequency distribution) of each image quality, and the margin parameter α to the chunk length density estimation unit 1. These values are values obtained in advance by observation or the like.

The chunk length density estimation unit 1 first sets the basic area so that there is no overlap for the chunk length (byte amount) belonging to each image quality. That is, the number of image quality is N, and the basic regions of image quality i (i = 1, 2,..., N) are respectively K _i = [a _i , b _i ] (i = 1, 2,..., N) And That is, the chunk length of image quality i is not less than a _{i and not} more than b _i . Here, for the simplicity of the following description, the image quality i is in descending order of the chunk length. That is, the smaller the value of i (the better the image quality), the longer the chunk length, and b ₁ ≧ a ₁ ≧ b ₂ ≧ a ₂ ... B _N ≧ a _N.

In addition, _{ai + 1} −b _i ≧ α (i = 1, 2,..., N−1) is established by the margin parameter α. Note that the basic area is set as a closed section that does not overlap between image quality.

  The chunk length density estimation unit 1 further estimates the probability density function of the chunk length belonging to each image quality as a smooth function. As this method, estimation by a kernel density function may be applied, or a method found in Non-Patent Document 4 may be applied.

  The chunk length density estimation unit 1 outputs the basic region and probability density function of the chunk length belonging to each image quality obtained by the above process. The above process can be performed outside the apparatus in advance, and the result can be used as an input to the image quality change detection unit 2.

FIG. 4 shows an example in which the chunk length basic region and the probability density function belonging to each image quality obtained by the chunk length density estimation unit 1 are drawn. The example shown in FIG. 4 is an example of N = 3 (high image quality, medium image quality, low image quality), and for the margin parameter α = 10 [kbyte], the basic region has high image quality I ₁ = [200, 300], Medium image quality I ₂ = [100, 150] and low image quality I ₃ = [50, 90] (both units are [kbyte]).

<Image quality change detection unit 2: Procedure 1>
Next, the image quality change detection unit 2 receives the chunk information (chunk length, observation time, etc.) of the packet capture data, the basic area of the chunk length belonging to each image quality, and the probability density function as follows. Detect image quality changes. This is procedure 1.

Hereinafter, M is the total number of chunks in the data. In the data, the image quality change detection unit 2 acquires each chunk length in the order of observation time, and the flowchart of FIG. 5 shows the chunk length l _j of the j (j = 1, 2,..., M) th chunk. The image quality is determined according to the procedure shown in FIG.

If l _j ≧ b ₁ (Yes in step 101), that is, if the chunk length l _j is greater than or equal to the maximum chunk length of the basic region with the highest image quality, the image quality Q (j) of the chunk is set to 1 ( Step 102). If l _j ≦ a _N (No in step 101, Yes in step 103), that is, if the chunk length l _j is less than or equal to the minimum chunk length of the basic region having the lowest image quality, the image quality of the chunk is set to N. (Step 104).

In the case of No in step 103, it is determined whether the chunk belongs to any basic area (step 105), that is, whether the following image quality i ₀ exists.

i₀が存在する場合（ステップ１０５のＹｅｓ）、チャンクjは画質i₀に該当すると判定する（ステップ１０６）。本判定ルールで判定不能な場合には「判定不能」とする（ステップ１０７）。

If i ₀ exists (Yes in step 105), it is determined that the chunk j corresponds to the image quality i ₀ (step 106). If the determination rule cannot be determined, “determination is impossible” (step 107).

  The image quality change detection unit 2 stores the determination result for each chunk in the determination result storage unit 21 in the order of the chunk observation time for each chunk.

<Image quality change detection unit 2: Procedure 2>
In the present embodiment, the image quality change detection unit 2 performs the maximum likelihood determination of the image quality as follows for the chunk that cannot be determined by the above-described determination rule (procedure 1). This is procedure 2.

  The image quality change detection unit 2 extracts a connected component of chunks whose image quality cannot be determined in the procedure 1 from the determination result of the procedure 1 stored in the determination result storage unit 21. Here, the connected component means chunks having consecutive numbers. For example, if the chunk number whose image quality could not be determined in step 1 is {2, 3, 5, 6, 7}, the connected components are the two components (2, 3) and (5, 6, 7). It is.

  The image quality change detection unit 2 performs the following procedure for each connected component of the chunk whose determination result is undetermined in step 1.

  Assuming that there are m chunks (m is 1 or more) belonging to a single connected component,

で表す。これらの各チャンクについて、それぞれ画質を振らせ、チャンク長確率密度推定部１で算出した確率密度関数に基づき、確率密度を算出する。

Represented by For each of these chunks, the image quality is varied, and the probability density is calculated based on the probability density function calculated by the chunk length probability density estimation unit 1.

For example, the probability density when the chunk length of the chunk C _l belongs to the image quality q is expressed as f (l; q). Further, the estimated image quality of the chunk C _l is represented by Q (l). At this time, the chunk group

に対する最も確からしい画質の組合せ Q : l → Q(l) を、以下の目的関数を最小化するように選択する：

The most probable image quality combination for Q: l → Q (l) is chosen to minimize the following objective function:

ただし、c_i (i = 1, 2, . . . , 4) は重み付け定数、Q(0), Q(m + 1) はそれぞれ当該連結成分前後の画質決定チャンクの画質である。

Here, c _i (i = 1, 2,..., 4) is a weighting constant, and Q (0) and Q (m + 1) are the image quality determination chunks before and after the connected component, respectively.

  By the procedure 1 and procedure 2, the image quality change detection unit 2 can calculate the estimated image quality of each chunk in the capture data for all the chunks. The image quality change detection unit 2 outputs the image quality for each chunk in the order of the chunk observation time for each chunk.

FIG. 6A shows an example of the image quality determination result of each chunk stored in the determination result storage unit 21 in procedure 1. As a result of performing maximum likelihood estimation defined by the expression of procedure 2 on this result, the determination result shown in FIG. 6B is obtained. However, in this example, c _i = 1 (i = 1, 12, 3, 4). For example, in the chunks indicated by C ₀ , C ₁ , C ₂ , C ₃ , and C ₄ in FIG. 6A, the image quality of C ₁ , C ₂ , and C ₃ cannot be determined. By the estimation, as shown in FIG. 6B, the image quality of C ₁ , C ₂ , and C ₃ is estimated.

  As described above, the image quality change detection unit 2 estimates the image quality of each chunk based on the data (basic region and the like described above) that associates the image quality with the chunk length range and the chunk length of each chunk.

<Buffer processing unit 3>
Next, the operation of the buffer processing unit 3 will be described. The input to the buffer processing unit 3 is the chunk information of packet capture data (observation time for each chunk, etc.), the image quality for each chunk, the duration (reproduction time length) per single chunk of each image quality, and the old when the image quality was changed remaining chunks number of the matrix of picture quality {Nij} (i, j = 1, 2,..., N) ( when changing from the image quality i to quality j), and a viewing end time T _E. T _E is the packet capture start time (i.e., chunk acquisition start time) when the zero represents the viewing end time.

  The buffer processing unit 3 has a virtual buffer (virtual buffer model). The buffer processing unit 3 inputs and accumulates the chunk in the virtual buffer model at the observation time of the chunk in the packet capture data, estimates the number of chunks in the buffer based on the input duration of each chunk, The reproduction state (in this example, the image quality of the chunk being reproduced or the reproduction stop state) is recorded in the reproduction state accumulation unit 31 every second. Here, the duration of each chunk may be different for each image quality.

  Each chunk accumulated in the buffer model disappears in the order of arrival after the input duration has elapsed, but the chunk consumption time is the time during which the video of the image quality is being played, that is, the playback state. Recording is performed in the storage unit 31. By executing this processing in the order of the chunk observation times, the number of chunks and the playback state in the playback buffer of the terminal can be estimated. In other words, the buffer processing unit 3 sequentially accumulates chunks in the virtual buffer at each chunk observation time in the video data, sets the chunk with the earliest observation time among the chunks in the virtual buffer as a reproduction target, and The chunk is erased after the lapse of the duration, and the duration information of the chunk and the image quality are stored in the playback state storage unit 31 as the duration of the video of the image quality corresponding to the chunk being played.

  FIG. 7 shows a processing flow of the buffer processing unit 3. As shown in FIG. 7, i is the number of arriving chunks up to the present, j is the number of erased chunks up to the present, t is the current time, d is the next chunk erasure time, and st is the playback state (in this example, , The image quality of the chunk being played back or the playback stopped state), and ch is the number of chunks in the buffer.

The buffer processing unit 3 initializes variables (step 201). Then, whenever one of the playback state recording time (every s seconds), chunk erasure time (d), or new chunk arrival time (t _{i + 1} ) arrives, the corresponding processing is performed (step 202).

  When the current time is the playback state recording time, the current playback state (st) is recorded in the playback state storage unit 31 together with the current time (step 203).

  At the chunk erasure time in step 204, if there is a chunk in the buffer (when ch is positive), the chunk having the earliest arrival time among the chunks existing in the buffer is erased. That is, 1 is subtracted from the number of chunks in the buffer, and the number of erased chunks up to the present is increased by 1. Furthermore, the erase time of the chunk with the earliest arrival time among the chunks existing in the buffer after the chunk erase is defined as the next chunk erase time d. Then, the image quality of the chunk being reproduced (the chunk to be erased next) is entered in the reproduction state st. If the number of chunks is 0 after erasing the chunks, the playback state st is “stopped”.

  The chunk erasure time is obtained from the chunk duration. For example, when chunk 1 is erased at time T, and the duration of chunk 2 having the earliest arrival time among the chunks existing in the buffer is R, the next chunk erase time is T + R. .

  At the chunk arrival time in step 205, if the image quality of the newly arrived chunk has not changed from the image quality of the previously arrived chunk (No in step 205), the number of chunks in the buffer is increased by 1, and the arrival up to the present time Increase the number of chunks by one. Further, the erase time of the chunk having the earliest arrival time among the chunks existing in the buffer is set as the next chunk erase time d (step 206). When a chunk already exists in the buffer when the chunk arrives, the next chunk erasure time d is not changed when the chunk arrives.

When the image quality of a newly arrived chunk is changed from the image quality of a previously arrived chunk at the chunk arrival time (Yes in step 205), the time at which the new image quality chunk is observed is Then, the old image quality chunks remaining in the buffer are reproduced up to N _ij from the next chunk of the chunk being reproduced at that time, and then changed to a new image quality (step 207). In the case of N _ij = 0, as soon as the playback of the chunk being played back is completed at the time when the chunk of the new image quality is observed, all the chunks in the buffer are cleared and playback of the new image quality chunk is started.

  The operation of recording the image quality of the chunk being played at each time in the playback state storage unit 31 together with the time information is the same as described above.

  As described above, when the image quality of the chunks sequentially stored in the virtual buffer is changed, the buffer processing unit 3 retains the pre-image quality change chunk that remains up to a predetermined number according to the combination of image quality before and after the image quality change. After playback, the chunk after the image quality change is set as the playback target. If the predetermined number is 0, the buffer processing unit 3 immediately deletes all the chunks in the virtual buffer immediately after erasing the chunks being played back at the time of observing the changed image quality chunks. Clear and set the changed image quality chunk as the playback target.

FIGS. 8A and 8B show examples of buffer processing when there is a change in image quality. FIG. 8A shows an example of the change from the image quality 1 to the image quality 2. The chunk of the image quality 2 arrives during the reproduction of the chunk 10 of the image quality 1, and after the reproduction from the chunk 10 being reproduced to the chunk 12 , Indicating that the image quality is changed to a new one. This is the case for N ₁₂ = 2.

FIG. 8B shows an example of the change from the image quality 2 to the image quality 1. When the image quality 1 chunk arrives during the reproduction of the image quality 2 chunk 11 and the reproduction of the image quality 2 chunk 11 finishes, the buffer is displayed. All the chunks are cleared, and playback of the chunk with image quality 1 is started. This is the case when N ₂₁ = 0. That is, in the example of FIGS. 8A and 8B, the buffer chunks are consumed more quickly at the time of transition to high image quality.

In the manner described above, continue to record the playback state to the playback state storage unit 31 to the viewing end time T _E. Also, if T _E is larger than the chunk downloads end time, i.e., the case where even after the download completion was continued viewing, when was reproduced state at the end of the chunk downloading time, reproduced to the buffer chunk is depleted The state is recorded in the reproduction state storage unit 31.

If the case was a reproduction stop state at the time immediately after the chunk download end time, since the reproduction stop state continues, as the end of the viewing end time T _E, play state accumulation between them of state as all "stop playback" Record in part 31.

  FIG. 9 shows an example of a reproduction state estimation result output from the buffer processing unit 3 and recorded in the reproduction state storage unit 31. As shown in FIG. 9, as a result of comparison with the actual value, it can be confirmed that the reproduction state can be estimated with high accuracy.

(Example of operation of chunk acquisition time estimation unit 232)
In the present invention, how to estimate the chunk acquisition time (observation time) from the packet capture data is not limited to a specific method, and for example, the chunk acquisition time can be estimated from an HTTP GET signal. However, there is a problem that the information of the HTTP GET signal cannot be grasped when the encryption measure in L7 is taken by HTTPS or the like. In this embodiment, by using the TCP header information, Chunk acquisition time is estimated. This operation is executed by the chunk acquisition time estimation unit 232 shown in FIG.

  The chunk acquisition time estimation unit 232 determines the chunk download completion timing using the information of the TCP packet (layer 4) acquired by the packet data acquisition unit 210, and sets the time at the timing as the chunk acquisition time. In the present embodiment, the following two examples will be described as methods for determining the chunk download completion timing using information of the TCP packet. Here, “TCP packet” means a packet including a TCP header and a TCP payload.

<Example 1: Example using sequence number>
In Example 1, the sequence number is acquired from the header of the TCP packet in the download direction (distribution server 300-> terminal 400). If the sequence number continues to increase by a certain constant, it is determined that chunk downloading is in progress, It is determined that the chunk download is completed when the number is increased.

  Since a TCP packet is configured with a maximum packet size that can pass through the path, the payload length of the TCP packet is constant during downloading a certain chunk, and is the maximum payload length. Only the last TCP packet for downloading the chunk has a different payload length. In the TCP packet, the maximum payload length is a value that can be calculated from the header information as (MTU (Maximum Transmission Unit) -header length). The MTU is a value indicating the maximum value of data that can be transmitted in one transfer. For example, in the case of Ethernet (registered trademark), since the MTU is 1500 and the header length is 40, the maximum payload length is 1460.

  In Example 1, using the above characteristics, it is determined that the chunk download has been completed with a packet in which the increase in the sequence number of the TCP packet is uneven.

  The sequence number of the TCP packet is a number indicating the byte number of the transmission data at the head of the data (payload) stored in the TCP packet. Therefore, as described above, when data having the maximum payload length is sequentially delivered, the sequence number increases by the maximum payload length. However, in the last part of the chunk divided into packets, the data is shorter than the maximum payload length, so that the maximum payload length is not increased. Note that this method cannot be applied when the chunk size is an integral multiple of the maximum payload length.

  Hereinafter, the operation of the chunk acquisition time estimation unit 232 in Example 1 will be described with reference to FIGS. 10, 11, and 12.

  The chunk acquisition time estimation unit 232 sequentially refers to the TCP packets in the download direction stored in the storage unit 220 in time series, and acquires the sequence number from the header. For example, the chunk acquisition time estimation unit 232 acquires a sequence number for each packet as shown in FIG.

  On the premise of the above operation, the chunk acquisition time estimation unit 232 performs chunk download completion determination processing in the sequence diagram shown in FIG.

  The chunk acquisition time estimation unit 232 acquires the maximum payload length (step 311). The acquisition of the maximum payload length does not need to be performed every time a TCP packet is received. For example, it may be performed every predetermined number of packets or every TCP connection.

  Subsequently, the chunk acquisition time estimation unit 232 calculates an increase value, that is, the payload length of the TCP packet by subtracting the sequence number of the previous TCP packet from the sequence number of the TCP packet that is the current target ( Step 312).

  Then, the chunk acquisition time estimation unit 232 compares the payload length calculated in step 312 with the maximum payload length, and checks whether they match (is the same value) (step 313). If yes (Yes in step 313), the sequence number of the next TCP packet is acquired, and the processing from step 312 is repeated using this as the current sequence number.

  When the payload length and the maximum payload length do not match, that is, when the increase value and the maximum payload length do not match (No in Step 313), it is determined that the chunk download is completed (Step 314), and the chunk acquisition time estimation unit For example, 232 holds the capture time (which can be acquired from the time stamp) of the TCP packet at that time in a storage unit such as a memory as the chunk acquisition time.

  For example, in the example of the input information shown in FIG. 10, as shown in FIG. 12, the packet 16 increases by a constant value (1416 bytes) until the packet 15, but the packet 16 is only 712 bytes (≠ 1416 bytes) from the packet 15. Since it has not increased, the reception timing of the packet 16 is determined as the chunk download completion timing.

<Example 2: Example using confirmation response (Ack) number>
In Example 2, the Ack number is acquired from the header of the TCP packet in the upload direction (terminal 400-> distribution server 300), and the Ack number is an integer multiple of a constant (the integer here is an integer of 1 or more). If the number continues to increase, it is determined that the chunk is being downloaded, and it is determined that the chunk download has been completed when the number is increased other than an integer multiple.

  In TCP communication, the packet receiving side (terminal 400) performs an operation of returning a TCP packet (referred to as an Ack packet) including an Ack number obtained by adding the number of bytes of received data (payload) to the sequence number of the received packet. . However, in TCP communication, there is a mode called a delayed response (Delayed ACK) that collectively returns responses for a plurality of packets. For example, when returning the Ack number for two packets, a number obtained by adding the payload length for two packets to the sequence number of the packet received first among the two packets is returned as the Ack number.

  As described in Example 1, during chunk download, the sequence number of the TCP packet increases uniformly with the maximum payload length, so when returning Ack to each packet, the Ack number is incremented by the maximum payload length. In the case of increasing and returning Ack for a plurality of packets collectively, it is increased by an integral multiple (the number of packets) of the maximum payload length. Therefore, as described above, in Example 2, when the Ack number continues to increase by an integer multiple of a certain constant (the integer here is an integer of 1 or more), it is determined that chunk downloading is in progress, and other than the integer multiple It is determined that the chunk download has been completed when the number is increased.

  Hereinafter, the operation of the chunk acquisition time estimation unit 232 in Example 2 will be described with reference to FIGS. 13, 14, and 15.

  The chunk acquisition time estimation unit 232 sequentially refers to the Ack packets in the upload direction in the storage unit 220 in time series, and acquires the Ack number from the header. For example, the chunk acquisition time estimation unit 232 acquires an Ack number for each packet as shown in FIG.

  On the premise of the above operation, the chunk acquisition time estimation unit 232 performs chunk download completion determination processing according to the sequence diagram shown in FIG.

  The chunk acquisition time estimation unit 232 acquires the maximum payload length (step 321). The maximum payload length need not be acquired every time an Ack packet is acquired. For example, it may be performed every predetermined number of packets or every TCP connection.

  Subsequently, the chunk acquisition time estimation unit 232 calculates an increase value of the Ack number by subtracting the Ack number of the previous Ack packet from the Ack number of the current Ack packet (step 322). This increased value corresponds to the total payload length of the packets received from the previous Ack packet transmission until the current Ack packet transmission.

  Then, the chunk acquisition time estimation unit 232 compares the increase value calculated in step 322 with the maximum payload length, and checks whether the increase value is an integer (an integer greater than or equal to 1) times the maximum payload length ( In step 323), if it is an integral multiple (Yes in step 323), the Ack number of the next Ack packet is acquired, and the process from step 322 is repeated using this as the current Ack number.

  When the increase value is not an integer multiple of the maximum payload length (No in Step 323), it is determined that the chunk download is completed (Step 324), and the chunk acquisition time estimation unit 232, for example, captures the Ack packet at that time ( Can be acquired from the time stamp) as a chunk acquisition time in a storage means such as a memory.

  For example, in the example of the input information shown in FIG. 13, as shown in FIG. 15, the packet 9 is increased by an integer multiple of 1416 bytes, but the packet 9 is 712 bytes from the packet 8 (an integer multiple of 1416 bytes). Therefore, the packet 9 acquisition timing is determined as the chunk download completion timing.

(Summary of the embodiment, effects, etc.)
According to the technology according to the present embodiment, for example, as shown in FIG. 9, in a progressive download video distribution service that dynamically performs rate control, not only playback / stop but also playback state estimation including image quality is performed. It can be realized with high accuracy.

  Here, for example, the conventional techniques described in Non-Patent Documents 1 to 3 do not refer to means for identifying chunks of video data to be distributed based on TCP header information. In commercial services, the chunk identification, chunk number, and image quality may be obtained by analyzing the HTTP payload depending on the implementation. However, this method is implementation-dependent, and cannot be applied when such information is not described in the HTTP payload or when the distribution service is implemented by HTTPS.

  On the other hand, in the technology according to the present embodiment, the observation time, chunk length, etc. of the chunk of video data to be distributed can be identified based on the header information of the TCP packet, and when there is no description in the HTTP payload, Even when it is mounted, it is possible to determine the image quality for each chunk, and it is possible to detect an image quality change location.

  In addition, since the conventional techniques described in Non-Patent Documents 1 to 3 do not take into account the image quality change, there is a problem that the number of chunks in the buffer after the image quality change occurs is different from the actual one. In this technique, as described above, since the change in image quality is taken into consideration, the number of chunks in the buffer and the playback state can be accurately estimated even when a progressive download video distribution service with dynamic rate control is targeted. In other words, in this embodiment, it is possible to process the chunks in the buffer based on different algorithms depending on the combination of the respective image quality before and after the image quality change, and change the image quality by simulating the image quality switching process in the buffer. It is possible to provide a technique for realizing reproduction state estimation with high accuracy even when it occurs.

  The present invention is not limited to the above-described embodiments, and various modifications and applications are possible within the scope of the claims.

100 Video playback state estimation device, video playback state estimation unit 1 chunk length density estimation unit 2 image quality change detection unit 21 determination result storage unit 3 buffer processing unit 31 playback state storage unit 200 video playback state estimation device 210 packet data acquisition unit 220 storage Unit 230 chunk state estimation unit 231 chunk time length estimation unit 232 chunk acquisition time designation unit

Claims (7)

In a video distribution method for distributing video data from a distribution server to a terminal in units of chunks, a video playback state estimation device for estimating a playback state of video data in a terminal,
Image quality estimating means for estimating the image quality of each chunk based on the chunk length of each chunk acquired from the captured data of the video data;
Chunks are sequentially stored in a virtual buffer at each chunk observation time in the video data, and the chunk with the earliest observation time among the chunks in the virtual buffer is set as a reproduction target. And a buffer processing means for storing the information of the time and the image quality in a storage unit, with the duration of the chunk being the time during which a video of the image quality corresponding to the chunk is being played back. Video playback state estimation device. 2. The video reproduction according to claim 1, wherein the image quality estimation unit estimates the image quality of each chunk based on data in which image quality is associated with a range of chunk lengths and the chunk length of each chunk. State estimation device. When the image quality of the chunks sequentially stored in the virtual buffer is changed, the buffer processing means, after reproducing the pre-image quality change chunks remaining up to a predetermined number according to the combination of the image quality before and after the image quality change, The video playback state estimation device according to claim 1 or 2, wherein a chunk after the image quality change is set as a playback target. When the predetermined number is 0, the buffer processing unit clears all the chunks in the virtual buffer immediately after erasing the chunks being played back at the time when the chunks with the changed image quality are observed. The video playback state estimation apparatus according to claim 3, wherein the changed image quality chunk is a playback target. 5. The video playback state estimation device according to claim 1, further comprising means for estimating an observation time of each chunk based on header information of a TCP packet in the capture data. In a video distribution method for distributing video data from a distribution server to a terminal in units of chunks, a video playback state estimation method executed by a video playback state estimation device that estimates a playback state of video data in a terminal,
An image quality estimation step of estimating the image quality of each chunk based on the chunk length of each chunk acquired from the captured data of the video data;
Chunks are sequentially stored in a virtual buffer at each chunk observation time in the video data, and the chunk with the earliest observation time among the chunks in the virtual buffer is set as a reproduction target. And a buffer processing step for storing the information of the time and the image quality in the storage means, with the duration of the chunk being the time during which the video of the image quality corresponding to the chunk is being played back. Video playback state estimation method.   The program for functioning a computer as each means in the image | video reproduction | regeneration state estimation apparatus of any one of Claims 1 thru | or 5.

Images