TWI704468B - Method and computer program product for translating a game dialogue window - Google Patents

Abstract

A method for translating a game dialog window, includes: finding a location of a dialogue window in a game screen shot; detecting an update frequency of a content of the dialog window; comparing the update frequency with a frequency threshold; and identifying a character content of the captured dialog window and translating by using a translation engine. The frequency threshold is obtained by learning and analyzing at least one historical film material of the game through a deep learning neural network model. When the update frequency is higher than the frequency threshold, a capture speed of the dialog window is increased. When the update frequency is lower than the frequency threshold, the capture speed of the dialog window is decreased.

Description

Method and computer program product for translating game dialogue window

The present invention relates to a real-time translation method, in particular to a method and computer program product for real-time translation of game dialogue windows.

With the advancement of technology, the upsurge of e-sports and the rise of various game platforms, players from all over the world often discuss strategies or tactics through the dialogue window in the game when teaming or individual competitions in the same game, because there are no borders Limitation, English or other languages may appear in the dialog window, and even because the dialog window refreshes too fast, the player cannot translate or understand it too late, so the player needs to spend extra time to view manually through the dialog window scroll The message sent by the person sometimes has no time to watch the content of the dialogue window during the intense attack and defense moments of the game.

Game players often use the translation software DualClip Translator to translate the dialog window in the game screen. The translation software DualClip Translator can use Google or Microsoft's translation engine to translate languages from all over the world. In addition to selecting translations on web pages and translating the entire document, it also supports optical character recognition (OCR) in pictures. Translation, but the game player needs to manually press Alt+S to take a screenshot, and then the translation result will pop up through Alt+Z. Since the translation software DualClip Translator needs to manually select the entire page or a certain section to translate the selected text, it is extremely inconvenient for game players who are stuck in a fierce attack and defense moment.

In order to solve the above-mentioned problems, the present invention provides a method and computer program product for translating the dialog window of a game, through an artificial intelligence (Artificial Intelligence) system analysis to find the position of the dialog window in the game screen, and according to the dialog window Update frequency screenshots and translate.

A method for real-time translation of a game dialog window according to an embodiment of the present invention includes finding the position of the dialog window from the game screen; detecting the update frequency of the dialog window content; comparing the update frequency with a frequency threshold; The character content of the dialog window is captured and translated using a translation engine. The frequency threshold is obtained by learning and analyzing at least one historical video data of the game through a deep learning neural network model. When the update frequency is greater than the frequency threshold, a capture speed for capturing the dialog window is increased; when the update frequency is less than or equal to the frequency threshold, the capture speed for capturing the dialog window is reduced.

Such as the above-mentioned real-time translation method, wherein, learning and analyzing at least one historical video data of the game through the deep learning neural network model to obtain the frequency threshold includes: capturing one of the at least one historical video data of the game The plural pictures in the time series, wherein each of the plural pictures corresponds to each time point in the time series in sequence; according to the plural pictures, at each time point in the time series, look for The position of the dialog window corresponding to each of the plurality of screens; based on the position, a first update frequency of the content of the dialog window corresponding to each of the plurality of screens is detected; and the corresponding time sequence is stored The update frequency of the character content of the dialog window at each time point is in a learning database; and the frequency threshold is calculated according to the update frequency.

As in the above-mentioned real-time translation method, the deep learning neural network model is a Long Short-Term Memory (LSTM) model for predicting the update frequency of the content of the dialogue window.

Such as the above-mentioned instant translation method, wherein when the capturing speed of capturing the dialog window increases, the speed of translating the character content of the dialog window is increased, and the translation result of the character content of the dialog window is increased to be displayed in The speed in the screen of the game.

Such as the above-mentioned real-time translation method, wherein when the capturing speed of capturing the dialog window decreases, the speed of translating the character content of the dialog window is reduced, and the translation result of the character content of the dialog window is reduced to be displayed on The speed in the screen of the game.

The instant translation method described above further includes using a logistic function to standardize the update frequency corresponding to each time point in the time series to reduce the complexity of calculating the frequency threshold.

As in the above-mentioned real-time translation method, the learning database is stored in a memory.

A computer program product according to an embodiment of the present invention is used to translate a game dialog window. The computer program product is loaded into the program by a computer to execute a window positioning command, a window detection command, a value comparison command, and a Recognize translation instructions. The window positioning instruction enables a processor of the computer to find the position of the dialog window from the screen of the game. The window detection command enables the processor to detect the update frequency of the content of the dialog window. The numerical comparison instruction enables the processor to compare the update frequency with a frequency threshold obtained by the processor through a deep learning neural network model to learn and analyze at least one historical video data of the game. When the update frequency is greater than the frequency threshold, the processor increases a capture speed for capturing the dialog window; when the update frequency is less than or equal to the frequency threshold, the processor decreases the capture speed for capturing the dialog window . The recognition and translation command enables the processor to recognize and translate the captured character content of the dialog window.

Such as the above-mentioned computer program product, wherein at least one historical video data of the game is learned and analyzed through the deep learning neural network model to obtain the frequency threshold, and the processor execution includes a retrieval command, a search command, and a detection Test instructions, a storage instruction, and an arithmetic instruction. The capture instruction causes the processor to capture a plurality of frames in a time series in the at least one historical video data of the game, wherein each of the plurality of frames corresponds to each of the time series in sequence Point in time. The search command causes the processor to search for the position of the dialog window corresponding to each of the plurality of frames at each time point in the time sequence according to the plurality of frames. The detection command enables the processor to detect a first update frequency of the content of the dialog window corresponding to each of the plural frames according to the position. The storage instruction causes the processor to store the update frequency of the character content of the dialog window corresponding to each time point in the time series in a learning database. The operation instruction enables the processor to calculate the frequency threshold according to the update frequency.

Such as the computer program product described above, wherein the deep learning neural network model is a long short-term memory model for predicting the update frequency of the content of the dialog window.

Such as the computer program product described above, wherein when the capturing speed of capturing the dialog window increases, the processor increases the speed of translating the character content of the dialog window, and increases the translation of the character content of the dialog window The result is displayed in the game's screen speed.

Such as the computer program product described above, wherein when the capturing speed of capturing the dialog window is reduced, the processor reduces the speed of translating the character content of the dialog window, and reduces the translation of the character content of the dialog window The result is displayed in the game's screen speed.

Such as the computer program product described above, wherein when the capturing speed of capturing the dialog window is reduced, the processor reduces the speed of translating the character content of the dialog window, and reduces the translation of the character content of the dialog window The result is displayed in the game's screen speed.

As in the above-mentioned computer program product, the processor execution further includes a standardized instruction, so that the processor uses a logic function to standardize the update frequency corresponding to each time point in the time series to reduce the calculation of the frequency threshold. Complexity.

Such as the computer program product mentioned above, wherein the learning database is stored in a memory.

Figure 1 is an application flow chart of the instant translation method for the game dialogue window according to the embodiment of the present invention. As shown in Figure 1, the instant translation method of the game dialogue window of the present invention uses video object detection to find the position of the dialogue window from the game screen (step S100). The instant translation method of the game dialog window of the present invention is executed by a computer device and a plurality of code codes running on a processor of the computer device. The plural code codes correspond to different functions, such as image capture function, Image recognition function, or translation function, etc. The image analysis technology is a piece of code that can run on a processor in a computer device, can analyze the image input to the computer device, and perform a predetermined action on a special object in the image.

In this embodiment, the instant translation method of the game dialog window of the present invention captures the screen when the e-sports player is performing the game, and uses the above-mentioned image analysis technology to find the position of the dialog window from the captured screen. For example, since the screens of games are mostly images, the real-time translation method of the game dialogue window of the present invention can set the above-mentioned image analysis technology to capture the area with the highest distribution density of characters (text) in the game screen. Or an area with frequent character update frequency, this area can correspond to the position of the dialogue window in the game screen. Figure 2 is a schematic diagram of a game dialog window according to an embodiment of the present invention. As shown in Figure 2, in the game screen 200, the mark 202 is the position of the dialog window.

Then, in step S102, the instant translation method of the game dialog window of the present invention also uses the above-mentioned video analysis technology to detect the update frequency f of the dialog window content (step S102). For example, suppose that the capture speed of the game screen is 5 frames per second. In other words, the time difference between each adjacent frame is 200ms, and the maximum text message that can be contained in the dialog window in the game screen The number is 10 messages. For example, the first text message, the second text message, the third text message,... to the tenth text message have been displayed on the game screen, a total of 10 messages. At this time, if another player submits the eleventh text message, the first message displayed on the game screen will be automatically removed. Suppose that in the first frame of the screen, there are 5 text messages in the dialog window, and in the second frame, the dialog window becomes 10 text messages. In one embodiment, the original five text messages in the first frame are retained in the second frame, and five additional text messages are added. Then the frame from the first frame to the second frame The update frequency of the characters in the dialog window is 25 times per second, that is, (10-5)/0.2=25. In another embodiment, similarly, in the first frame of the screen, there are 5 text messages in the dialog window, and in the second frame, the dialog window becomes 10 text messages, but in the second frame If no original message is retained in the first frame, the update frequency of the characters in the dialog window from the first frame to the second frame is 75 times per second, that is, (5+10)/0.2=75. Generally speaking, in the course of the game, when the enemy and our opponents meet each other, members of the same team need to inform their teammates of the relevant battle information through the dialogue window. At this time, the update frequency of the dialogue window will increase significantly.

The real-time translation method of the game dialogue window of the present invention compares the detected update frequency ( f ) with a frequency threshold ( f _th ) (step S104 ), wherein the frequency threshold ( f _th ) is obtained through a deep learning nerve The network model is obtained by studying and analyzing at least one historical video data of the game. In this embodiment, the deep learning neural network model is a Long Short-Term Memory (LSTM) model. The long and short-term memory model is suitable for processing and predicting an important event with a very long interval and delay in a time series. In other words, this case is based on applying the collected historical data of update frequency versus time axis to the aforementioned long and short-term memory model to obtain the frequency threshold ( f _th ) corresponding to the time point that can be used to predict the occurrence of battle events. The important event is the point in time when the battle event occurs. As e-sports players are currently executing the game, it is impossible to predict when a battle event will occur. However, in the historical game live video of the same game, the time point of the battle event and the update frequency of the dialogue window before and after the battle event are known. Therefore, the long and short-term memory model calculates the frequency threshold ( f _th ) based on the update frequency of the dialog window of the game screen before the time point when the battle event occurred in the historical game live movie, as long as the update frequency of the dialog window of the game screen is higher than the When a trigger time point of the frequency threshold ( f _th ) occurs, it means that a battle event will occur after the trigger time point.

For example, based on the many historical video data of the same game or the live video of the game obtained through the Application Programming Interface (API), the long-term and short-term memory model found that as long as the update frequency of the dialog window at a certain point in time ( f ) is greater than a frequency threshold ( f _th ) (for example, 20 times per second), a battle event will occur within a period of time after this point in time. In other words, the long and short-term memory model learns through experience (that is, accumulates many historical video data as a learning database). When the update frequency of the dialog window is greater than the frequency threshold (for example, 20 times per second), the time point can be predicted In the subsequent period, the update frequency ( f ) of the dialog window will be greater than the frequency threshold ( f _th ), and it can be predicted that a battle event will occur in a period after the time point. The frequency threshold ( f _th ) is used as a basis for adjusting the speed of the game screen capture, adjusting the translation speed of the dialogue window content, and adjusting the display speed of the translation result.

The instant translation method of the game dialogue window of the present invention judges whether the update frequency ( f ) of the dialogue window is greater than the frequency threshold ( f _th ) (step S106 ). When the update frequency ( f ) of the dialog window is greater than the frequency threshold ( f _th ), the dialog window screen is accelerated (step S108 ). When the update frequency ( f ) of the dialogue window is less than or equal to the frequency threshold ( f _th ), the dialogue window frame is slowed down (step S110). For example, when the update frequency ( f ) of the dialog window (e.g. 25 times per second) is greater than the frequency threshold ( f _th ) (e.g. 20 times per second), the rate of capturing the dialog window is increased (e.g., by Capture 5 frames per second to capture 8 frames per second). When the update frequency ( f ) of the dialog window (e.g. 15 times per second) is less than or equal to the frequency threshold ( f _th ) (e.g. 20 times per second), the rate of capturing the dialog window is reduced (e.g., from Taking 5 frames of pictures becomes 3 frames of pictures per second).

Next, the instant translation method of the game dialog window of the present invention recognizes the character content of the captured dialog window (step S112). In this embodiment, the present invention uses an optical character recognition engine Tesseract OCR Engine to recognize characters in the dialog window. In the next step S114, the instant translation method of the game dialogue window of the present invention uses a translation engine for translation. In this embodiment, the present invention uses Google or Microsoft's translation engine to automatically recognize the characters to translate into the language required by the e-sports players.

In step S116, the instant translation method of the game dialogue window of the present invention presents the translation result in the dialogue window in the game screen of the e-sports player, and then completes the method of instant translation of the characters in the game dialogue window. In this embodiment, when the capturing speed of capturing the dialog window increases, the instant translation method of the game dialog window of the present invention also increases the speed of translating the character content of the dialog window, and increases the speed of translating the character content of the dialog window. The speed at which the translation result is displayed on the game screen. Conversely, when the capturing speed of capturing the dialog window decreases, the instant translation method for the game dialog window of the present invention reduces the speed of translating the character content of the dialog window, and reduces the display of the translation result of the character content of the dialog window. The speed in the game screen. In this embodiment, the instant translation method of the game dialog window of the present invention can finally transmit the translated text to the mobile device, or present it to the e-sports player in another section of the game screen.

FIG. 3 is a training flowchart of the instant translation method for the game dialogue window according to the embodiment of the present invention. The purpose of the training process in Fig. 3 is to obtain the required frequency threshold ( f _th ) in step S104 in Fig. 1. As shown in Figure 3, the real-time translation method of the game dialog window of the present invention captures multiple screens in a time series in a historical video data of the game through an application programming interface (API), or captures multiple histories of the game Multiple pictures in multiple time series in the video data (step S300). A corresponding time series can be obtained for each different game live video, and the time series is the time length of the captured game live video (including all videos or fragments). Each time point in the time sequence is the time point at which the game screen is captured, and each time point can correspond to a game sample screen. For example, the real-time translation method of the game dialogue window of the present invention captures a first game live video with a total time of 10 seconds at a rate of 5 frames per second, and captures a total of a video at a rate of 5 frames per second. A live video of the second game at 20 seconds. In other words, the instant translation method of the game dialogue window of the present invention captures a total of 50 frames of the game screen of the first game live video, the time interval of the game screens of the two adjacent frames is 200ms, and captures a total of 100 frames For the game screen of the second live game video, the time interval between two adjacent game screens is also 200ms. In this embodiment, the instant translation method of the game dialogue window of the present invention is to obtain the live video and audio data of the players of the same game through YouTube and Twitch API.

In step S302, the instant translation method of the game dialog window of the present invention is based on the plural images captured in step 200, and at each time point in the time sequence, searches for each of the plural images. Corresponds to the position of the dialog window. For example, the real-time translation method of the game dialogue window of the present invention uses the video analysis technology to find every 50 frames of the captured first game live video and 100 frames of the second live game video. The position of the dialog window of a frame. In the instant translation method of the game dialogue window of the present invention, the method of finding the position of the dialogue window in step S302 is the same as that of step S100 in FIG.

Then, in step S304, the instant translation method of the game dialog window of the present invention detects the update frequency f of the content of the corresponding dialog window according to the position of the dialog window in each game screen. In other words, in each game screen (game sampling screen), the update frequency f of the dialogue window content corresponding to the game screen can be obtained. In the method for real-time translation of the game dialog window of the present invention, the method of detecting the update frequency f in step S304 is the same as that in step S102, so it will not be repeated. After that, the instant translation method of the game dialogue window of the present invention stores the detected update frequency f of the dialogue window character content at each time point in a time sequence into a learning database (step S306). The learning database is stored in a memory in the computer device that executes the instant translation method of the game dialogue window of the present invention.

Figure 4 is a diagram showing the relationship between the update frequency f of the dialog window and the time (axis) according to the embodiment of the present invention. As shown in Figure 4, at the 30th second of the game live video, the update frequency f of the corresponding dialog window is 15 times. It is worth noting that the dialog window update frequency f versus time relationship diagram in Figure 4 can correspond to a live video. In other words, each different game live video can get a corresponding dialog window update frequency f versus time relationship. Figure. Therefore, the learning database stores a plurality of dialogue windows update frequency f versus time relationship graph data, and each dialogue window update frequency f versus time relationship graph corresponds to a different game live video.

Returning to Figure 3, the instant translation method of the game dialogue window of the present invention uses a logistic function to normalize the update frequency f corresponding to each time point in a time series to reduce the frequency threshold f _th Complexity (step 208). In this embodiment, the present invention uses a Sigmoid function to normalize the update frequency f . The S function is one of many logic functions. Figure 5 is a diagram showing the relationship between update frequency and time of the standardized dialog window according to an embodiment of the present invention. Fig. 5 is a schematic diagram obtained by normalizing the update frequency f versus time graph of Fig. 4. As shown in Figure 5, the instant translation method of the game dialog window of the present invention maps the Y-axis parameter frequency (number of times) in the original Figure 4 to an integer between 0 and 1, and the X-axis parameter time is taken to multiple The maximum time of the live video of the game. If the length of the video is insufficient, the Y-axis parameter frequency (number) is set to 0. For example, at the 30th second of a live game video, at point A, the normalized value of the update frequency f of the corresponding dialog window is 0.6 (score 0.6), that is, 60%.

For example, the real-time translation method of the game dialog window of the present invention captures a section of the first live game video with a total time of 10 seconds at a rate of 5 frames per second, and captures a section of total video at a rate of 5 frames per second. The second live video of the game at 20 seconds. In other words, the instant translation method of the game dialogue window of the present invention captures a total of 50 frames of the game screen of the first game live video, the time interval of the game screens of the two adjacent frames is 200ms, and captures a total of 100 frames For the game screen of the second live game video, the time interval between two adjacent game screens is also 200ms. Therefore, when the instant translation method of the game dialogue window of the present invention is standardized, since the time length of the second live game video is greater than the time length of the first live game video, the time length of the second live game video is used It is the normalized X-axis parameter time.

For example, since the first live game video is insufficient in time length, the instant translation method of the game dialogue window of the present invention aligns the first frame of the 50 frames of the first live game video to the second live game video The first frame of the game, and the 50 frames of the first live game video are stuffed into the time length of the 100 frames of the second live game, so the instant translation method of the game dialogue window of the present invention is standardized when , The Y-axis parameter frequency (number of times) of the last 50 frames of the first live game video will be set to 0.

The instant translation method of the game dialogue window of the present invention reads the data of a plurality of dialogue window update frequency f versus time relationship graphs from the learning database, and after standardizing it, the step S310 is started. In step S310, the instant translation method of the game dialog window of the present invention uses a long and short-term memory model, and takes the standardized update frequency f versus time relationship data of a plurality of dialog windows as input, that is, calculates according to the update frequency f The frequency threshold ( f _th ). Because in the historical game live video, the time point of the battle event and the update frequency of the dialogue window before and after the battle event are known. Therefore, the long and short-term memory model calculates the frequency threshold ( f _th ) based on the update frequency of the dialog window of the game screen before the time point when the battle event occurred in the historical game live movie, as long as the update frequency of the dialog window of the game screen is higher than the When a trigger time point of the frequency threshold ( f _th ) occurs, it means that a battle event will occur after the trigger time point. In this embodiment, as shown in Figure 5, when the long-term and short-term memory model calculates a time point that exceeds a frequency threshold of 60% (ie at the 30th second in Figure 5), it represents the update of the dialog window in the game screen The frequency f is increased, that is, a battle event may occur at the 30th second, so the speed of capturing the dialog window starts to increase. Therefore, after the training process shown in Fig. 3, the long- and short-term memory model uses, for example, 60% as the frequency threshold used in the method of the present invention.

Referring to Figures 1 and 3 at the same time, in step S312, when an e-sports player is not satisfied with the translation speed presented by the instant translation method of the game dialogue window of the present invention, they can directly use the instant translation of the game dialogue window of the present invention. A user interface of the translation method manually adjusts the frequency threshold ( f _th ), or re-executes the training process (ie, steps S300 to S310) to obtain a new frequency threshold ( f _th ). If the e-sports player is satisfied with the translation speed presented by the instant translation method of the game dialogue window of the present invention, step S314 is entered to end the procedure. For example, referring to Figure 5, when the original frequency threshold ( f _th ) is 60%, e-sports players feel that the speed of the translation results presented by the instant translation method of the game dialogue window of the present invention still cannot keep up. The talking speed of teammates in the dialogue window. At this time, the e-sports player can manually adjust the frequency threshold ( f _th ) to 40% through the user interface, so that the instant translation method of the game dialogue window of the present invention is at the 26th second in Fig. 5 B. Start to increase the speed of capturing game screens. In other words, referring to Figure 5, when the frequency threshold ( f _th ) is adjusted from 60% to 40%, the instant translation method of the game dialog window of the present invention increases the speed of capturing the game screen and increases the character content of the dialog window. The speed of translation and the start time of increasing the speed of displaying the translation result of the character content of the dialogue window on the game screen will be 4 seconds earlier, so that the speed of rendering the translation result can keep up with the conversation speed of the actual dialogue window.

In another embodiment, when the e-sports player feels that the speed of the translation result presented by the instant translation method of the game dialogue window of the present invention still cannot keep up with the conversation speed of the teammates in the dialogue window, the e-sports player can finish After the current game, instruct the instant translation method of the game dialog window of the present invention to re-execute the training process (ie, steps S300 to S310). Since the update frequency of the dialog window corresponding to the current game live video will also be stored in the learning database during the re-training process, the long-short-term memory model can be based on the maternal sample with more data (ie update The frequency f samples) are recalculated to a new frequency threshold ( f _th ). Generally speaking, the more samples of the update frequency f , the more accurate the frequency threshold ( f _th ) calculated by the long- and short-term memory model will be, that is, the more accurate the e-sports player's real-time translation needs during the game.

The present invention further discloses a computer program product for translating game dialogue windows. The computer program product is loaded into the program by a computer to execute a window positioning command, a window detection command, a value comparison command, and an identification translation command . The window positioning command can cause a processor of the computer to execute step S100 in FIG. 1. The window detection command can cause the processor to execute step S102 in FIG. 1. The numerical comparison instruction can cause the processor to execute steps S104 to S108, or steps S104, S106, and S110 in FIG. The recognition and translation instruction can cause the processor to execute steps S112 to S116 in FIG. 1.

In this embodiment, the numerical comparison instruction causes the processor to compare the update frequency f of the dialog window of the game screen with a frequency threshold ( f _th ). Wherein, the computer program product of the present invention learns and analyzes at least one historical video data of the game through the deep learning neural network model to obtain the frequency threshold, and the processor executes including a retrieval command, a search command, and a detection Instruction, a storage instruction, and an arithmetic instruction. The fetching instruction can make the processor execute step S300 in FIG. 3. The search command can cause the processor to execute step S302 in FIG. 3. The detection instruction can cause the processor to execute step S304 in FIG. 3. The storage instruction can cause the processor to execute step S306 in FIG. 3. The operation instruction can cause the processor to perform step S310 in FIG. 3. The processor execution further includes a standardized instruction, which causes the processor to execute step S308 in FIG. 3.

The steps or actions performed by the computer program product of the present invention can directly correspond to the method for real-time translation of the game dialogue window disclosed in the present invention. The method for real-time translation of the game dialogue window disclosed in the present invention has been It is described in detail in paragraphs [0021]~[0038] of the specification of this case, so it is not repeated here. The method and computer program product for real-time translation of the game dialogue window disclosed in this case can effectively reduce the time for e-sports players to translate characters in the dialogue window in the game, and reduce the inconvenience during game operation.

Although the embodiments of the present invention are as described above, we should understand that what is presented above is only an example, not a limitation. According to this embodiment, many changes of the above exemplary embodiment can be implemented without violating the spirit and scope of the invention. Therefore, the breadth and scope of the present invention should not be limited by the embodiments described above. More precisely, the scope of the present invention should be defined by the following patented scope and its equivalents.

S100: step S102: step S104: step S106: step S108: step S110: step S112: step S114: step S116: step f : update frequency f _th : frequency threshold 200: game screen 202: mark S300: step S302: step S304 : Step S306: Step S308: Step S310: Step S312: Step S314: Step A, B: Point

Figure 1 is an application flow chart of the instant translation method for the game dialogue window according to the embodiment of the present invention; Figure 2 is a schematic diagram of the game dialogue window according to the embodiment of the present invention; Figure 3 is the game dialogue window according to the embodiment of the present invention A training flow chart of the real-time translation method of the present invention; Figure 4 is a diagram of the dialog window update frequency versus time in an embodiment of the present invention; Figure 5 is a standardized dialog window update frequency vs. time diagram of the embodiment of the present invention.

S100: steps

S102: Step

S104: Step

S106: Step

S108: Step

S110: Step

S112: Step

S114: Step

S116: Step

f : update frequency

f _th : frequency threshold

Claims (14)

A real-time translation method for a game dialogue window, comprising: finding the position of the dialogue window from the game screen; detecting the update frequency of the dialogue window content; comparing the update frequency with a frequency threshold; the frequency threshold is through a depth The learning neural network model is obtained by learning and analyzing at least one historical video data of the game; when the update frequency is greater than the frequency threshold, increase a capture speed for capturing the dialog window; when the update frequency is less than or equal to the frequency threshold , Reducing the speed of capturing the dialog window; and recognizing the character content of the captured dialog window, and using a translation engine for translation. According to the real-time translation method described in claim 1, wherein the at least one historical video data of the game is learned and analyzed through the deep learning neural network model to obtain the frequency threshold, including: extracting the at least one of the game A plurality of pictures in a time series in a historical film data, wherein each of the plurality of pictures corresponds to each time point in the time series in sequence; according to the plurality of pictures, in the time series At each time point, search for the position of the dialog window corresponding to each of the plural screens; according to the position, detect a first update frequency of the contents of the dialog window corresponding to each of the plural frames ; Store the first update frequency corresponding to the character content of the dialogue window at each time point in the time series in a learning database; and calculate through the deep learning neural network model according to the first update frequency Out the frequency threshold. The real-time translation method described in item 1 of the scope of patent application, wherein the deep learning neural network model is a Long Short-Term Memory (LSTM) model for predicting the update frequency of the content of the dialog window . The instant translation method described in the first item of the scope of patent application, wherein when the capturing speed of capturing the dialog window increases, the speed of translating the character content of the dialog window is increased, and the amount of the dialog window is increased The speed at which the translation result of the character content is displayed on the screen of the game. For example, the instant translation method described in item 1 of the scope of patent application, wherein when the capturing speed of capturing the dialog window decreases, the speed of translating the character content of the dialog window is reduced, and the amount of the dialog window is reduced. The speed at which the translation result of the character content is displayed on the screen of the game. For example, the instant translation method described in item 2 of the scope of patent application further includes: using a logistic function to standardize the update frequency corresponding to each time point in the time series to reduce the calculation of the frequency threshold Complexity. The real-time translation method described in item 2 of the scope of patent application, wherein the learning data library is stored in a memory. A computer program product for translating a game dialog window. The computer program product is loaded into the program by a computer to execute: a window positioning command, which enables a processor of the computer to find the dialog window from the game screen Position; a window detection command to enable the processor to detect the update frequency of the content of the dialog window; a numerical comparison command to enable the processor to compare the update frequency with a frequency threshold; the frequency threshold is passed by the processor A deep learning neural network model is obtained by learning and analyzing at least one historical video data of the game; when the update frequency is greater than the frequency threshold, the processor increases a capture speed for capturing the dialog window; when the update frequency is less than When it is equal to the frequency threshold, the processor reduces the capturing speed of capturing the dialog window; and a recognition translation command enables the processor to recognize the captured character content of the dialog window and perform translation. For example, the computer program product described in claim 8, wherein the deep learning neural network model is used to learn and analyze at least one historical video data of the game to obtain the frequency threshold, and the processor execution includes: a capture Instructions to enable the processor to capture a plurality of frames in a time series in the at least one historical video data of the game, wherein each of the plurality of frames corresponds to each time point in the time series in sequence ; A search command that causes the processor to find the position of the dialog window corresponding to each of the plurality of screens at each time point in the time series according to the plurality of screens; a detection command, Allow the processor to detect a first update frequency of the content of the dialog window corresponding to each of the plurality of screens according to the position; a storage instruction to cause the processor to store each time in the corresponding time series The first update frequency of the character content of the dialog window clicked is in a learning database; and an arithmetic instruction enables the processor to calculate the frequency threshold through the deep learning neural network model according to the first update frequency . For the computer program product described in item 8 of the scope of patent application, wherein the deep learning neural network model is a long-short-term memory model for predicting the update frequency of the content of the dialog window. For example, the computer program product described in item 8 of the scope of patent application, wherein when the capturing speed of capturing the dialog window increases, the processor increases the speed of translating the character content of the dialog window, and increases the The speed at which the translation result of the character content of the dialog window is displayed on the screen of the game. For example, the computer program product described in item 8 of the scope of patent application, wherein, when the capturing speed of capturing the dialog window is reduced, the processor reduces the speed of translating the character content of the dialog window, and reduces the The speed at which the translation result of the character content of the dialog window is displayed on the screen of the game. For the computer program product described in item 9 of the scope of the patent application, the processor execution further includes: a standardized instruction, so that the processor uses a logistic function to correspond to the each time point in the time series The update frequency is standardized to reduce the complexity of calculating the frequency threshold. For the computer program product described in item 9 of the scope of patent application, the learning database is stored in a memory.

Images