Video index detection method and device, computer device, and storage medium

Abstract

Embodiments of the present application relate to a video index detection method and device, computer equipment and a storage medium, the method comprising: in the process of monitoring the video index of a target video, for a plurality of preset time windows, extracting the time sequence characteristics of the video index corresponding to each time window to obtain a time sequence characteristic set corresponding to the video index; inputting the time sequence characteristic set into an initial model to fit each time sequence characteristic through the initial model to obtain a fitting value corresponding to each time window; generating a fitting model according to the plurality of fitting values; and detecting whether the video index of the target video according to the fitting model. Thus, the time sequence characteristics of the video index can be extracted from the perspective of a plurality of time windows of different lengths, an effective time window generation model is determined and the detection of the video index is performed in real time, the problem of easy missed reporting of sparse video monitoring indexes is solved, and the accuracy of video index detection is improved.

Description

Technical Field

[0001] The present invention relates to the field of video playback technology, and in particular to a method, apparatus, computer device and storage medium for detecting video indicators. Background Technology

[0002] In video monitoring scenarios, there is a common need for monitoring sparse monitoring metrics, such as monitoring the success rate of specific interfaces with low access volume. However, due to the fine time granularity of the monitoring, the success rate metric fluctuates significantly, and its drawbacks are obvious. Because such monitoring metrics lack temporal continuity, it is difficult to discover patterns in the time series. Anomaly detection for video metrics with sparse time monitoring is often difficult and prone to missed detections.

[0003] The main reason why sparse monitoring indicators are prone to missed reports is their sparsity: due to the sparsity of time, the monitoring lacks temporal regularity, and conventional models cannot learn its true temporal steady state.

[0004] Taking the business scenario of domestic video websites as an example, consider monitoring their real-time playback success rate to obtain the actual playback experience of online users. Because relatively real-time success rate data is needed, the interval between each metric delivery should not be too long. Furthermore, to prevent being influenced by past data, success rates within a shorter timeframe are generally delivered. However, since the number of users in the early morning is extremely low compared to other time periods, the delivery of these success rates often becomes sparse. Moreover, with a very small base, the success rate metric becomes abnormally volatile, losing its effectiveness in reflecting the actual playback experience. For example, if there are only 10 users in the early morning, and 2 of them fail to play, the playback success rate will only be 80%. However, from a business perspective, those two users are likely inactive users or bots, constantly experiencing playback failures throughout the day, and do not represent the real user experience. But with fewer playbacks in the early morning, the playback success rate will be significantly reduced.

[0005] Therefore, accurately detecting whether video metrics are correct when monitoring video has become an urgent problem to be solved. Summary of the Invention

[0006] In view of this, in order to solve the above-mentioned technical problems or some of the technical problems, the present invention provides a method, apparatus, computer equipment and storage medium for detecting video indicators.

[0007] In a first aspect, embodiments of the present invention provide a method for detecting video metrics, comprising:

[0008] During the monitoring of video metrics of the target video, for multiple preset time windows, the temporal features of the video metrics corresponding to each time window are extracted to obtain the temporal feature set corresponding to the video metrics.

[0009] The time series feature set is input into the initial model, and the initial model is used to fit each time series feature to obtain the fitting value corresponding to each time window;

[0010] A fitting model is generated based on the multiple fitted values;

[0011] The correctness of the video metrics of the target video is determined based on the fitted model.

[0012] In one possible implementation, the video metric is the playback success rate of the target video during playback, and the step of extracting the temporal features of the video metric corresponding to each of the multiple preset time windows includes:

[0013] Obtain multiple time windows corresponding to the playback success rate;

[0014] During the playback of the target video, the playback success rate of the target video within each time window is obtained;

[0015] Each time window is slid-processed once at preset intervals;

[0016] Each time a sliding process is performed, the playback success rate of the target video within the time window after the sliding process is obtained, and a set of playback success rates corresponding to each time window is obtained.

[0017] The set of playback success rates is used as the temporal feature corresponding to each time window.

[0018] In one possible implementation, the initial model is a long short-term memory network model, and the fitting process for each of the temporal features using the initial model includes:

[0019] Based on the time when the playback success rate is obtained, extract the time-series feature training samples corresponding to each time window from the time-series feature set;

[0020] Multiple time-series feature training samples are input into the long short-term memory network model for model training, so that the long short-term memory network model can perform fitting processing on the time-series feature training samples corresponding to each time window.

[0021] In one possible implementation, generating a fitting model based on the plurality of fitted values ​​includes:

[0022] The target time window is determined from the multiple time windows based on the multiple fitted values;

[0023] The target time window is used as the time window corresponding to the video metric in the long short-term memory network model for model training until the model converges, resulting in a well-trained fitted model.

[0024] In one possible implementation, determining the target time window from the plurality of time windows based on the fitted value includes:

[0025] Based on the time of acquisition of the playback success rate, extract the time-series feature test samples corresponding to each time window from the time-series feature set;

[0026] The fitting residual corresponding to each time window is determined based on the time-series feature test samples and the fitted values;

[0027] The time window with the smallest fitting residual is taken as the target time window, and the fitting residual of the target time window is less than a set threshold.

[0028] In one possible implementation, determining the fitting residual corresponding to each time window based on the time-series feature test samples and the fitted values ​​includes:

[0029] Based on the fitted values, the predicted time-series features corresponding to each time-series feature test sample are determined;

[0030] The difference between each time-series feature test sample and the corresponding predicted time-series feature is calculated, and the resulting difference is used as the fitting residual of the corresponding time window.

[0031] In one possible implementation, the step of detecting whether the video metrics of the target video are correct based on the fitted model includes:

[0032] Obtain the first value of the video metrics of the target video;

[0033] The second value of the video metric of the target video is output through the fitting model;

[0034] If the difference between the first value and the second value is less than a set threshold within a preset time period, the video indicator is determined to be correct.

[0035] If the difference between the first value and the second value is greater than or equal to a set threshold within a preset time period, the video indicator is determined to be incorrect.

[0036] Secondly, embodiments of the present invention provide a video indicator detection device, comprising:

[0037] The extraction module is used to extract the temporal features of the video indicators corresponding to each of the multiple preset time windows during the monitoring of video indicators of the target video, and obtain the temporal feature set corresponding to the video indicator.

[0038] The processing module is used to input the time series feature set into the initial model, so as to perform fitting processing on each of the time series features through the initial model to obtain the fitting value corresponding to each time window;

[0039] A generation module is used to generate a fitting model based on multiple fitting values;

[0040] The detection module is used to detect whether the video metrics of the target video are correct based on the fitted model.

[0041] Thirdly, embodiments of the present invention provide a computer device, including: a processor and a memory, wherein the processor is configured to execute a video metric detection program stored in the memory to implement the video metric detection method described in any one of the first aspects above.

[0042] Fourthly, embodiments of the present invention provide a storage medium storing one or more programs, which can be executed by one or more processors to implement the video metric detection method described in any one of the first aspects.

[0043] The video metric detection scheme provided in this invention involves extracting temporal features of the video metric corresponding to each of multiple preset time windows during the monitoring of video metrics of a target video, thereby obtaining a temporal feature set corresponding to the video metric. The temporal feature set is then input into an initial model to fit each temporal feature, obtaining a fitted value for each time window. A fitting model is generated based on the multiple fitted values. Finally, the correctness of the video metric of the target video is detected based on the fitting model. This approach allows for the extraction of temporal features of video metrics from multiple time windows of varying lengths, enabling the determination of an effective time window model for video metric detection. This solves the problem of missed detections due to sparse video surveillance metrics and improves the accuracy of video metric detection. Attached Figure Description

[0044] Figure 1 A flowchart illustrating a video metric detection method provided in an embodiment of the present invention;

[0045] Figure 2 A flowchart illustrating another method for detecting video metrics provided in an embodiment of the present invention;

[0046] Figure 3 A schematic diagram illustrating the principle of a video metric detection method provided in an embodiment of the present invention;

[0047] Figure 4 This is a schematic diagram of the structure of a video indicator detection device provided in an embodiment of the present invention;

[0048] Figure 5 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] To facilitate understanding of the embodiments of the present invention, further explanations and descriptions will be provided below with reference to the accompanying drawings and specific embodiments. These embodiments do not constitute a limitation on the embodiments of the present invention.

[0051] Figure 1 This is a flowchart illustrating a video metric detection method provided in an embodiment of the present invention, as shown below. Figure 1 As shown, the method specifically includes:

[0052] S11. During the monitoring of video metrics of the target video, for multiple preset time windows, extract the temporal features of the video metrics corresponding to each time window to obtain the temporal feature set corresponding to the video metrics.

[0053] The video metric detection method provided in this invention is applied to a computer device, which may include, but is not limited to, a server, a desktop computer, etc. Specifically, in the scenario of monitoring video metrics, the temporal features of the video metrics are obtained through a sliding time window, and an initial model is trained based on the temporal features to obtain a trained fitting model to detect whether the video metrics are correct.

[0054] In this embodiment, the target video is delivered to multiple clients for playback. During the delivery and playback process, video metrics of the target video are monitored. These video metrics can be the playback success rate over a specific period, reflecting the current playback status of the target video. Since there may be periods with low user numbers during video delivery and playback—for example, delivery may occur once per minute, but no users are present during that minute—the playback success rate metric cannot be obtained for that minute. Therefore, multiple time windows are pre-defined, and video metrics are extracted for each time window. For example, time windows could be 1 minute, 10 minutes, 30 minutes, 1 hour, 3 hours, 6 hours, etc. The playback success rate within each time window is extracted; that is, the playback success rates of different users within each time window are aggregated to obtain the playback success rate for that time window, which serves as the video metric for that time window.

[0055] Furthermore, for each time window, a time-sliding process is performed with a preset time step. Each time the process is repeated, the video metrics corresponding to each time window are obtained, resulting in temporally continuous video metrics, which serve as the temporal features corresponding to the time window. In other words, the temporal features are temporally continuous video metrics. The temporal features of each time window corresponding to the target video are used as a set of temporal features. For example, taking a 1-minute monitoring interval as an example, the playback success rate is calculated every minute. However, due to the absence of users or a low number of users in some time periods, some minutes may have no success rate or the success rate metric may not be normal. Therefore, different time windows are selected to calculate the playback success rate for that period. With a time step of 1 minute and a time window of 1 hour, the time window is slid every 1 minute, meaning the playback success rate for the most recent hour is calculated every 1 minute. The number of users in 1 hour is sufficient, ensuring a success rate metric for each minute, making the data continuous and resulting in temporally continuous temporal features.

[0056] S12. Input the set of time-series features into the initial model, and use the initial model to fit each time-series feature to obtain the fitting value corresponding to each time window.

[0057] In this embodiment, the initial model is a model used to fit time series features, which can be an LSTM model. The time series feature set is input into the initial model, and the time series features under different time windows are fitted. After fitting, a model that can predict future time series features based on historical time series features can be obtained. This model can output the fitting value corresponding to each time window, which is used to characterize the future time series features predicted based on historical time series features.

[0058] S13. Generate a fitting model based on the multiple fitted values.

[0059] In this embodiment, after fitting, the fitted value and the fitting residual of the test data under each time window are obtained using the pre-acquired test data. The test data can be the actual temporal features of each time window within the first historical period. The initial model is used to fit the temporal features of the second period before the first period, and the fitted value corresponding to the first period is predicted as the predicted temporal features of the first period. Then, the fitting residual of the fitted value and the corresponding actual temporal features of each time window is determined. The time window with the smallest fitting residual is selected as the time window in the actual model training, and the model is trained based on the window to generate a trained fitted model.

[0060] For example, if the fitting residuals of time series features with rated time windows of 10 minutes, 30 minutes, and 1 hour are all less than a set threshold, they are considered stable and predictable time series features. In this case, the smallest time window of 10 minutes is selected as the final actual aggregation window for the video metric, and the model generated with the 10-minute time window is used as the final fitting model.

[0061] S14. Detect whether the video metrics of the target video are correct based on the fitted model.

[0062] In this embodiment, the generated fitting model can predict future temporal features based on historical temporal features. Using the current temporal features of the target video as the actual value, the fitting model can predict the current temporal features of the video metric using historical temporal features. The deviation between the predicted and actual values ​​is calculated. Under normal circumstances, the deviation will be less than a set threshold. Only when the data trend differs from historical data, i.e., when an anomaly occurs, will the deviation exceed the set threshold. If several consecutive deviations exceed the set threshold, it is determined that the currently acquired video metric is incorrect, indicating an anomaly. The anomaly may be due to the sparsity of video metrics within the current monitoring period, leading to a lack of temporal regularity in the monitoring. If an error in the video metric is determined, it is recorded and the developers are alerted. When the deviation is less than or equal to the set threshold, it is determined that the currently acquired video metric is correct and there is no anomaly.

[0063] The video metric detection method provided in this invention involves, during the monitoring of video metrics of a target video, extracting temporal features of the video metric corresponding to each of multiple preset time windows to obtain a temporal feature set corresponding to the video metric; inputting the temporal feature set into an initial model to fit each temporal feature to obtain a fitted value corresponding to each time window; generating a fitting model based on the multiple fitted values; and detecting whether the video metric of the target video is correct based on the fitting model. Therefore, it is possible to extract temporal features of video metrics from multiple time windows of different lengths, determine an effective time window to generate a model, and perform real-time video metric detection, solving the problem of missed detections of sparse video surveillance metrics and improving the accuracy of video metric detection.

[0064] Figure 2 A flowchart illustrating another video metric detection method provided in this embodiment of the invention is shown below. Figure 2 As shown, the method specifically includes:

[0065] S21. During the monitoring of video metrics of the target video, multiple time windows corresponding to the playback success rate are obtained; during the playback of the target video, the playback success rate of the target video within each time window is obtained; and each time window is slid-processed once at a preset interval.

[0066] In this embodiment, the playback success rate of the target video during playback is described using video metrics. Multiple time windows (e.g., 10 minutes, 30 minutes, 1 hour, etc.) are pre-set for monitoring the playback success rate. During the process of delivering the target video to different clients for playback, the success rate of each delivery of the target video is obtained. The first delivery number and the second successful playback number of the target video within each time window are obtained. The second number is divided by the first number to obtain the playback success rate of the target video in the current time window. Thus, the playback success rate within each time window is obtained.

[0067] Further preset durations are set, and each time window is slid forward at preset intervals. For example, if the preset duration is 1 minute, the current time is 3 o'clock, and the time window is 10 minutes, the corresponding time period for obtaining playback success rate is from 2:50 to 3 o'clock. After a 1-minute interval, the current time is 3:01, and the time window is slid forward by 1 minute. The corresponding time period for obtaining playback success rate for the 10-minute time window is from 2:51 to 3:01.

[0068] S22. For each sliding process, obtain the playback success rate of the target video within the time window after the sliding process, and obtain the playback success rate set corresponding to each time window; use the playback success rate set as the temporal feature corresponding to each time window.

[0069] In this embodiment, after performing a sliding process once for each time window, the current playback success rate within each time window is obtained. After a preset number of sliding processes, multiple playback success rates corresponding to each time window are obtained as a set of playback success rates. Since the playback success rates in the set of playback success rates corresponding to each time window are obtained continuously, the set of playback success rates corresponding to each time window is used as the temporal feature corresponding to each time window. This temporal feature is a feature that is continuous in time. A temporal feature set is generated from the temporal features corresponding to each time window.

[0070] S23. Based on the time of acquisition of the playback success rate, extract the time-series feature training samples corresponding to each time window from the time-series feature set; input multiple time-series feature training samples into the long short-term memory network model for model training, so as to obtain the fitting value corresponding to each time window by fitting the time-series feature training samples corresponding to each time window through the long short-term memory network model.

[0071] In this embodiment, after acquiring the temporal features over a period of time, the acquisition time of each playback success rate corresponding to the temporal features is determined. The playback success rates are sorted according to the acquisition time from the initial acquisition time to the current time. Multiple playback success rates within a period of time starting from the initial acquisition time are extracted as training samples of the temporal features corresponding to each time window for model training.

[0072] Furthermore, the initial model used for fitting is the LSTM model, also known as the Long Short-Term Memory network model. The temporal feature training samples are input into the LSTM network model for model training. The LSTM network model is then used for fitting, and the resulting model can predict the fitting value corresponding to each time window in the future. The fitting value represents the predicted video index in the future period.

[0073] S24. Determine a target time window from multiple time windows based on multiple fitted values; use the target time window as the time window corresponding to the video index in the long short-term memory network model for model training until the model converges, and obtain a trained fitted model.

[0074] In this embodiment, determining the target time window from multiple time windows based on the fitted value includes: extracting time-series feature test samples corresponding to each time window from the time-series feature set according to the time of acquisition of the playback success rate; determining the fitting residual corresponding to each time window based on the time-series feature test samples and the fitted value; and taking the time window with the smallest fitting residual as the target time window, wherein the fitting residual of the target time window is less than a set threshold.

[0075] Specifically, after extracting the time-series feature training samples corresponding to each time window using the method in step S23, the remaining time-series features in the time-series feature set are used as time-series feature test samples. In other words, the time-series feature test samples are the true values ​​corresponding to the time-series features predicted by the model, which are used to subtract the fitted values ​​predicted by the model, thereby determining the fitting residuals.

[0076] Furthermore, determining the fitting residual corresponding to each time window based on the time-series feature test sample and the fitted value includes: determining the predicted time-series feature corresponding to each time-series feature test sample based on the fitted value; subtracting each time-series feature test sample from the corresponding predicted time-series feature, and using the difference as the fitting residual of the corresponding time window.

[0077] Since the model can output fitted values ​​to predict the time series features in the future after inputting the test samples of the time series features, the fitted values ​​generated for each time window are used as the predicted time series features for each test sample of the time series features. The difference between each test sample of the time series features and the corresponding predicted time series features is used to obtain the fitting residual for each time window.

[0078] The residuals that are less than a set threshold are selected from the fitted residuals, and the time window corresponding to the smallest fitted residual among the fitted residuals that are less than the set threshold is taken as the target time window.

[0079] Furthermore, the target time window is used as the time window for obtaining the playback success rate in the long short-term memory network model for model training. The model is iterated continuously until it converges, and the fitted model trained with the target time window is used as the final trained fitted model.

[0080] For example, given the temporal features of the same video metric over 10 days, the temporal features of the first 8 days are selected as training samples, and the temporal features of the last 2 days are used as test samples. Taking an LSTM model as an example, the input to this model is a historical temporal feature set, and the output is the video metric for the next time period. The model is fitted to the training samples of the temporal features, resulting in a model that predicts future video metrics based on historical temporal features. That is, by inputting the training samples of the first 8 days into the model and iterating continuously, the model can output the fitted values ​​of the predicted video metrics for the last 2 days. Furthermore, the fitted values ​​of the video metrics for the last 2 days can be used as predicted temporal features. By subtracting the predicted temporal features corresponding to different time windows from the actual test samples of the last 2 days, the fitting residuals for different time windows can be obtained. The time window with the smallest fitting residual that is less than a set threshold is selected. For example, if the fitting residuals of time windows of 10 minutes, 30 minutes, and 1 hour are all less than the set threshold, then the smallest 10-minute time window is selected as the actual time window used to obtain the playback success rate. The fitting model trained within the 10-minute time window is the final fitting model. Then, the deviation between the real-time video metrics and the predicted video metrics can be calculated using prediction methods to detect anomalies.

[0081] S25. Obtain the first value of the video index of the target video; output the second value of the video index of the target video through the fitting model; when the difference between the first value and the second value within a preset time period is less than a set threshold, determine that the video index is correct; when the difference between the first value and the second value within a preset time period is greater than or equal to the set threshold, determine that the video index is incorrect.

[0082] In this embodiment, since the generated fitting model can predict future temporal characteristics based on historical temporal features, for the real-time acquired playback success rate, the fitting model can use historical temporal features to predict the current playback success rate. The deviation between the predicted value and the actual value is calculated. Under normal circumstances, the deviation will be less than a set threshold. Only when the data trend differs from historical conditions, i.e., when an anomaly occurs, will the deviation exceed the set threshold. Therefore, if the deviation values ​​at several consecutive points are greater than the set threshold, it is determined that the currently detected video metric is incorrect.

[0083] Specifically, the first value is the actual playback success rate obtained in real time within a preset time period, and the second value is the predicted playback success rate predicted by the fitting model. The difference between the actual value and the predicted value is determined. If the difference is less than a set threshold, the video metric is determined to be correct. If the difference is greater than or equal to the set threshold within the preset time period, the video metric is determined to be incorrect, that is, the video metric obtained in real time is abnormal.

[0084] As an example, such as Figure 3 The diagram illustrates the principle of a video metric detection method according to an embodiment of the present invention. It involves acquiring a set of historical monitoring time-series features of video metrics for a target video, dividing the time-series feature set into a training set and a test set for offline model training, aggregating the training set using sliding time windows of fixed lengths to determine the playback success rate, fitting the time-series features extracted from the sliding time windows using LSTM or other fitting models, and using the trained or fitted model to predict each time-series feature for the test set. Based on the prediction results, the best-fitting model and the n time-series features with the corresponding minimum window size are selected as the monitoring features. During real-time detection, the determined multi-scale time-series features are extracted in real-time and compared with the predicted time-series features. Anomalies are determined based on the degree of deviation from the prediction results.

[0085] The video metric detection method provided by this invention acquires multiple time windows corresponding to the playback success rate during the monitoring of video metrics of a target video; during the playback of the target video, the playback success rate of the target video within each time window is acquired; each time window is slid-processed once at preset intervals; after each sliding process, the playback success rate of the target video within the slid-processed time window is acquired, resulting in a playback success rate set corresponding to each time window; the playback success rate set is used as the temporal feature corresponding to each time window; based on the acquisition time of the playback success rate, temporal feature training samples corresponding to each time window are extracted from the temporal feature set; and multiple temporal feature training samples are input into a Long Short-Term Memory (LSTM) network. The model is trained by fitting training samples of temporal features corresponding to each time window using a Long Short-Term Memory (LSTM) network model, obtaining a fitted value for each time window. Based on multiple fitted values, a target time window is determined from multiple time windows. This target time window is then used as the time window corresponding to the video metric in the LSM network model for training until the model converges, resulting in a well-trained fitted model. The first value of the video metric for the target video is obtained. The second value of the video metric for the target video is output through the fitted model. If the difference between the first and second values ​​within a preset time period is less than a set threshold, the video metric is determined to be correct; if the difference is greater than or equal to the set threshold, the video metric is determined to be incorrect. This allows for the extraction of temporal features from multiple sliding time windows of varying lengths, automatically selecting effective time windows and temporal features for video metric detection, solving the problem of missed detections due to sparse monitoring metrics, and improving the accuracy of video metric detection.

[0086] Figure 4 This is a schematic diagram of the structure of a video indicator detection device provided in an embodiment of the present invention, as shown below. Figure 4 As shown, the device specifically includes:

[0087] Extraction module 41 is used to extract the temporal features of the video indicators corresponding to each of the multiple preset time windows during the monitoring of video indicators of the target video, and obtain the temporal feature set corresponding to the video indicator.

[0088] Processing module 42 is used to input the time series feature set into an initial model, so as to perform fitting processing on each time series feature through the initial model to obtain the fitting value corresponding to each time window;

[0089] Generation module 43 is used to generate a fitting model based on the multiple fitting values;

[0090] The detection module 44 is used to detect whether the video metrics of the target video are correct based on the fitting model.

[0091] In one possible implementation, the acquisition module 45 is used to acquire multiple time windows corresponding to the playback success rate;

[0092] During the playback of the target video, the playback success rate of the target video within each time window is obtained;

[0093] The processing module is specifically used to perform a sliding process on each time window once at a preset time interval;

[0094] The acquisition module is specifically used to acquire the playback success rate of the target video within the time window after each sliding process, and to obtain a set of playback success rates corresponding to each time window.

[0095] The set of playback success rates is used as the temporal feature corresponding to each time window.

[0096] In one possible implementation, the extraction module is specifically used to extract time-series feature training samples corresponding to each time window from the time-series feature set based on the time when the playback success rate is obtained.

[0097] The processing module is specifically used to input multiple time-series feature training samples into the long short-term memory network model for model training, so as to perform fitting processing on the time-series feature training samples corresponding to each time window through the long short-term memory network model.

[0098] In one possible implementation, the determining module 46 is configured to determine a target time window from the plurality of time windows based on the plurality of fitted values;

[0099] The processing module is specifically used to train the model by using the target time window as the time window corresponding to the video index in the long short-term memory network model until the model converges and a well-trained fitted model is obtained.

[0100] In one possible implementation, the extraction module is specifically used to extract time-series feature test samples corresponding to each time window from the time-series feature set based on the time when the playback success rate is obtained.

[0101] The determining module is specifically used to determine the fitting residual corresponding to each time window based on the time-series feature test samples and the fitting values.

[0102] The time window with the smallest fitting residual is taken as the target time window, and the fitting residual of the target time window is less than a set threshold.

[0103] In one possible implementation, the determining module is specifically used to determine the predicted temporal feature corresponding to each of the temporal feature test samples based on the fitted value;

[0104] The difference between each time-series feature test sample and the corresponding predicted time-series feature is calculated, and the resulting difference is used as the fitting residual of the corresponding time window.

[0105] In one possible implementation, the acquisition module is specifically used to acquire a first value of the video metric of the target video;

[0106] The detection module is specifically used to output a second value of the video metric of the target video through the fitting model;

[0107] If the difference between the first value and the second value is less than a set threshold within a preset time period, the video indicator is determined to be correct.

[0108] If the difference between the first value and the second value is greater than or equal to a set threshold within a preset time period, the video indicator is determined to be incorrect.

[0109] The video indicator detection device provided in this embodiment of the invention can be as follows: Figure 4 The apparatus shown can perform, as Figure 1-3 All steps of the method for detecting mid-length video metrics, thereby achieving Figure 1-3 For details on the technical effectiveness of the video metric detection method shown, please refer to [link / reference]. Figure 1 The relevant descriptions are presented concisely and will not be elaborated upon here.

[0110] Figure 5 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Figure 5The illustrated electronic device 500 includes at least one processor 501, a memory 502, at least one network interface 504, and other user interfaces 503. The various components in the computer device 500 are coupled together via a bus system 505. It is understood that the bus system 505 is used to implement communication between these components. In addition to a data bus, the bus system 505 also includes a power bus, a control bus, and a status signal bus. However, for clarity, ... Figure 5 The general designated all buses as Bus System 505.

[0111] The user interface 503 may include a display, keyboard, or clicking device (e.g., mouse, trackball, touchpad, or touchscreen).

[0112] It is understood that the memory 502 in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), and Direct Rambus RAM (DRRAM). The memory 502 described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0113] In some implementations, memory 502 stores elements, executable units or data structures, or subsets thereof, or extended sets thereof: operating system 5021 and application program 5022.

[0114] The operating system 5021 includes various system programs, such as the framework layer, core library layer, and driver layer, used to implement various basic business functions and handle hardware-based tasks. The application program 5022 includes various applications, such as a media player and a browser, used to implement various application functions. The program implementing the method of this embodiment can be included in the application program 5022.

[0115] In this embodiment of the invention, by calling the program or instructions stored in memory 502, specifically the program or instructions stored in application program 5022, processor 501 executes the method steps provided in each method embodiment, including, for example:

[0116] During the monitoring of video metrics of the target video, for multiple preset time windows, the temporal features of the video metrics corresponding to each time window are extracted to obtain the temporal feature set corresponding to the video metrics.

[0117] The time series feature set is input into the initial model, and the initial model is used to fit each time series feature to obtain the fitting value corresponding to each time window;

[0118] A fitting model is generated based on the multiple fitted values;

[0119] The correctness of the video metrics of the target video is determined based on the fitted model.

[0120] In one possible implementation, multiple time windows corresponding to the playback success rate are obtained;

[0121] During the playback of the target video, the playback success rate of the target video within each time window is obtained;

[0122] Each time window is slid-processed once at preset intervals;

[0123] Each time a sliding process is performed, the playback success rate of the target video within the time window after the sliding process is obtained, and a set of playback success rates corresponding to each time window is obtained.

[0124] The set of playback success rates is used as the temporal feature corresponding to each time window.

[0125] In one possible implementation, based on the time at which the playback success rate is obtained, time-series feature training samples corresponding to each time window are extracted from the time-series feature set;

[0126] Multiple time-series feature training samples are input into the long short-term memory network model for model training, so that the long short-term memory network model can perform fitting processing on the time-series feature training samples corresponding to each time window.

[0127] In one possible implementation, a target time window is determined from a plurality of time windows based on a plurality of said fitted values;

[0128] The target time window is used as the time window corresponding to the video metric in the long short-term memory network model for model training until the model converges, resulting in a well-trained fitted model.

[0129] In one possible implementation, based on the time at which the playback success rate is obtained, a time-series feature test sample corresponding to each time window is extracted from the time-series feature set;

[0130] The fitting residual corresponding to each time window is determined based on the time-series feature test samples and the fitted values;

[0131] The time window with the smallest fitting residual is taken as the target time window, and the fitting residual of the target time window is less than a set threshold.

[0132] In one possible implementation, the predicted temporal feature corresponding to each of the temporal feature test samples is determined based on the fitted value;

[0133] The difference between each time-series feature test sample and the corresponding predicted time-series feature is calculated, and the resulting difference is used as the fitting residual of the corresponding time window.

[0134] In one possible implementation, a first value of the video metric of the target video is obtained;

[0135] The second value of the video metric of the target video is output through the fitting model;

[0136] If the difference between the first value and the second value is less than a set threshold within a preset time period, the video indicator is determined to be correct.

[0137] If the difference between the first value and the second value is greater than or equal to a set threshold within a preset time period, the video indicator is determined to be incorrect.

[0138] The methods disclosed in the above embodiments of the present invention can be applied to or implemented by processor 501. Processor 501 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware in processor 501 or by instructions in the form of software. The processor 501 may be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of the present invention can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software units in the decoding processor. The software units may be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory 502. Processor 501 reads the information in memory 502 and, in conjunction with its hardware, completes the steps of the above method.

[0139] It is understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, microcode, or a combination thereof. For hardware implementation, the processing unit can be implemented in one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described herein, or combinations thereof.

[0140] For software implementation, the techniques described herein can be implemented by units that perform the functions described herein. The software code can be stored in memory and executed by a processor. The memory can be implemented in the processor or external to the processor.

[0141] The computer device provided in this embodiment may be as follows: Figure 5 The computer device shown can perform, for example Figure 1-3 All steps of the method for detecting mid-length video metrics, thereby achieving Figure 1-3 For details on the technical effectiveness of the video metric detection method shown, please refer to [link / reference]. Figure 1-3 The relevant descriptions are presented concisely and will not be elaborated upon here.

[0142] This invention also provides a storage medium (computer-readable storage medium). This storage medium stores one or more programs. The storage medium may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as read-only memory, flash memory, hard disk, or solid-state drive; the memory may also include combinations of the above types of memory.

[0143] One or more programs in the storage medium can be executed by one or more processors to implement the above-described method for detecting video metrics executed on the device side.

[0144] The processor is used to execute a detection program for video metrics stored in the memory to implement the following steps of a video metric detection method executed on the device side:

[0145] During the monitoring of video metrics of the target video, for multiple preset time windows, the temporal features of the video metrics corresponding to each time window are extracted to obtain the temporal feature set corresponding to the video metrics.

[0146] The time series feature set is input into the initial model, and the initial model is used to fit each time series feature to obtain the fitting value corresponding to each time window;

[0147] A fitting model is generated based on the multiple fitted values;

[0148] The correctness of the video metrics of the target video is determined based on the fitted model.

[0149] In one possible implementation, multiple time windows corresponding to the playback success rate are obtained;

[0150] During the playback of the target video, the playback success rate of the target video within each time window is obtained;

[0151] Each time window is slid-processed once at preset intervals;

[0152] Each time a sliding process is performed, the playback success rate of the target video within the time window after the sliding process is obtained, and a set of playback success rates corresponding to each time window is obtained.

[0153] The set of playback success rates is used as the temporal feature corresponding to each time window.

[0154] In one possible implementation, based on the time at which the playback success rate is obtained, time-series feature training samples corresponding to each time window are extracted from the time-series feature set;

[0155] Multiple time-series feature training samples are input into the long short-term memory network model for model training, so that the long short-term memory network model can perform fitting processing on the time-series feature training samples corresponding to each time window.

[0156] In one possible implementation, a target time window is determined from a plurality of time windows based on a plurality of said fitted values;

[0157] The target time window is used as the time window corresponding to the video metric in the long short-term memory network model for model training until the model converges, resulting in a well-trained fitted model.

[0158] In one possible implementation, based on the time at which the playback success rate is obtained, a time-series feature test sample corresponding to each time window is extracted from the time-series feature set;

[0159] The fitting residual corresponding to each time window is determined based on the time-series feature test samples and the fitted values;

[0160] The time window with the smallest fitting residual is taken as the target time window, and the fitting residual of the target time window is less than a set threshold.

[0161] In one possible implementation, the predicted temporal feature corresponding to each of the temporal feature test samples is determined based on the fitted value;

[0162] The difference between each time-series feature test sample and the corresponding predicted time-series feature is calculated, and the resulting difference is used as the fitting residual of the corresponding time window.

[0163] In one possible implementation, a first value of the video metric of the target video is obtained;

[0164] The second value of the video metric of the target video is output through the fitting model;

[0165] If the difference between the first value and the second value is less than a set threshold within a preset time period, the video indicator is determined to be correct.

[0166] If the difference between the first value and the second value is greater than or equal to a set threshold within a preset time period, the video indicator is determined to be incorrect.

[0167] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0168] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented in hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0169] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting video metrics, characterized in that, include: During the monitoring of video metrics of the target video, for multiple preset time windows, the temporal features of the video metrics corresponding to each time window are extracted to obtain the temporal feature set corresponding to the video metrics. The time series feature set is input into the initial model, and the initial model is used to fit each time series feature to obtain the fitting value corresponding to each time window; A fitting model is generated based on the multiple fitted values; The correctness of the video metrics of the target video is determined based on the fitted model. The video metric is the playback success rate of the target video during playback. The step of extracting the temporal features of the video metric corresponding to each of the multiple preset time windows includes: Obtain multiple time windows corresponding to the playback success rate; During the playback of the target video, the playback success rate of the target video within each time window is obtained; Each time window is slid-processed once at preset intervals; Each time a sliding process is performed, the playback success rate of the target video within the time window after the sliding process is obtained, and a set of playback success rates corresponding to each time window is obtained. The set of playback success rates is used as the temporal feature corresponding to each time window.

2. The method according to claim 1, characterized in that, The initial model is a long short-term memory network model. The fitting process for each temporal feature using the initial model includes: Based on the time when the playback success rate is obtained, extract the time-series feature training samples corresponding to each time window from the time-series feature set; Multiple time-series feature training samples are input into the long short-term memory network model for model training, so that the long short-term memory network model can perform fitting processing on the time-series feature training samples corresponding to each time window.

3. The method according to claim 2, characterized in that, The step of generating a fitting model based on multiple fitting values ​​includes: The target time window is determined from the multiple time windows based on the multiple fitted values; The target time window is used as the time window corresponding to the video metric in the long short-term memory network model for model training until the model converges, resulting in a well-trained fitted model.

4. The method according to claim 3, characterized in that, Determining the target time window from the plurality of time windows based on the fitted value includes: Based on the time of acquisition of the playback success rate, extract the time-series feature test samples corresponding to each time window from the time-series feature set; The fitting residual corresponding to each time window is determined based on the time-series feature test samples and the fitted values; The time window with the smallest fitting residual is taken as the target time window, and the fitting residual of the target time window is less than a set threshold.

5. The method according to claim 4, characterized in that, The step of determining the fitting residual corresponding to each time window based on the time-series feature test samples and the fitted values ​​includes: Based on the fitted values, the predicted time-series features corresponding to each time-series feature test sample are determined; The difference between each time-series feature test sample and the corresponding predicted time-series feature is calculated, and the resulting difference is used as the fitting residual of the corresponding time window.

6. The method according to claim 5, characterized in that, The step of detecting whether the video metrics of the target video are correct based on the fitted model includes: Obtain the first value of the video metrics of the target video; The second value of the video metric of the target video is output through the fitting model; If the difference between the first value and the second value is less than a set threshold within a preset time period, the video indicator is determined to be correct. If the difference between the first value and the second value is greater than or equal to a set threshold within a preset time period, the video indicator is determined to be incorrect.

7. A device for detecting video metrics, characterized in that, include: The extraction module is used to extract the temporal features of the video indicators corresponding to each of the multiple preset time windows during the monitoring of video indicators of the target video, and obtain the temporal feature set corresponding to the video indicator. The processing module is used to input the time series feature set into the initial model, so as to perform fitting processing on each of the time series features through the initial model to obtain the fitting value corresponding to each time window; A generation module is used to generate a fitting model based on multiple fitting values; The detection module is used to detect whether the video metrics of the target video are correct based on the fitted model; The video metric is the playback success rate of the target video during playback. The step of extracting the temporal features of the video metric corresponding to each of the multiple preset time windows includes: Obtain multiple time windows corresponding to the playback success rate; During the playback of the target video, the playback success rate of the target video within each time window is obtained; Each time window is slid-processed once at preset intervals; Each time a sliding process is performed, the playback success rate of the target video within the time window after the sliding process is obtained, and a set of playback success rates corresponding to each time window is obtained. The set of playback success rates is used as the temporal feature corresponding to each time window.

8. A computer device, characterized in that, include: A processor and a memory, the processor being configured to execute a program stored in the memory to implement the video metric detection method according to any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium stores one or more programs, which can be executed by one or more processors to implement the video metric detection method according to any one of claims 1 to 6.

Images