A game data processing method, device and equipment and readable storage medium

Abstract

The application discloses a game data processing method and device, equipment and a readable storage medium, and the game data processing method comprises the following steps: displaying a target map in a game picture used for controlling a first virtual object; the first virtual object belongs to a first camp; the target map is used for displaying a sound icon representing an object behavior category and an object position of a second virtual object; the second virtual object belongs to a second camp; when the target map has the sound icon, an object behavior prompt voice corresponding to the sound icon is acquired; the object behavior prompt voice is used for voice prompting the object behavior category and the object position of the second virtual object; and the object behavior prompt voice is played. By using the application, the accuracy and efficiency of acquiring enemy behavior information in the game can be improved, and the performance form of the game is enriched.

Description

Technical Field

[0001] The present application relates to the field of Internet technology, and in particular to a game data processing method, apparatus, device, and readable storage medium. Background Art

[0002] With the continuous development of mobile communication technology, smart terminals such as mobile phones and tablet computers have occupied a pivotal position in people's daily lives. Nowadays, through smart terminals, people can enjoy leisure and entertainment through electronic games anytime and anywhere, which has greatly enriched people's lifestyles.

[0003] Currently, in some existing multiplayer online competitive games, players can assess enemy behavior by listening to in-game sounds (such as car sounds, gunfire, and footsteps) and observing the prompt icons for different factions on the game map. However, due to the diverse nature of in-game sounds, and the factions often receiving more sounds, relying solely on these sounds can be inaccurate. Using prompt icons on the game's minimap requires players to maintain their focus, and minimaps are typically small, making it difficult to efficiently obtain enemy behavior information. Summary of the Invention

[0004] The embodiments of the present application provide a game data processing method, apparatus, device, and readable storage medium, which can improve the accuracy and efficiency of obtaining enemy behavior information in the game and enrich the game's presentation.

[0005] An embodiment of the present application provides a method for processing game data, including:

[0006] Displaying a target map in a game screen for controlling a first virtual object; the first virtual object belongs to a first camp; the target map is used to display a sound icon representing an object behavior category and an object position of a second virtual object; the second virtual object belongs to a second camp;

[0007] When a sound icon exists on the target map, obtaining an object behavior prompt voice corresponding to the sound icon; the object behavior prompt voice is used to voice prompt the object behavior category and object position of the second virtual object;

[0008] Play the object's behavior prompt voice.

[0009] In one aspect, an embodiment of the present application provides a game data processing device, comprising:

[0010] The display module is configured to display a target map in a game picture used for controlling the first virtual object; the first virtual object belongs to a first camp; and the target map is used to display a sound icon representing an object behavior category and an object position of a second virtual object; the second virtual object belongs to a second camp.

[0011] The voice acquisition module is configured to acquire an object behavior prompt voice corresponding to the sound icon when the target map contains the sound icon; and the object behavior prompt voice is used to voice prompt the object behavior category and the object position of the second virtual object.

[0012] The voice playing module is configured to play the object behavior prompt voice.

[0013] The display module is configured to display the game picture used for controlling the first virtual object in response to a starting operation of a game application; display the target map centered on the first virtual object in the game picture; acquire the object behavior category and the object position of the second virtual object within a region range of the target map; generate the sound icon according to the object behavior category and the object position; and display the sound icon in the target map.

[0014] The voice acquisition module includes:

[0015] The icon identification unit is configured to perform icon identification on the target map to obtain icon categories and icon coordinates corresponding to one or more to-be-detected icons respectively.

[0016] The icon identification unit is configured to input the target map into an icon detection model, perform feature extraction through a feature extraction network in the icon detection model to obtain a picture feature matrix; the picture feature matrix is used to represent different scales of features corresponding to the one or more to-be-detected icons recognized by the icon detection model; and perform non-maximum suppression processing on the picture feature matrix to obtain the icon categories and the icon coordinates corresponding to the one or more to-be-detected icons respectively.

[0017] The filtering unit is configured to perform fixed position filtering on the one or more to-be-detected icons according to the icon categories and the icon coordinates corresponding to the one or more to-be-detected icons respectively.

[0018] The one or more to-be-detected icons include a to-be-detected icon X.

[0019] The filtering unit is specifically configured to: acquire N fixed detection boxes matched with the icon category corresponding to the to-be-detected icon X according to the icon category corresponding to the to-be-detected icon X; the N fixed detection boxes are located at fixed positions in the target map respectively; N is an integer greater than 1; and the filtering unit is specifically configured to: acquire distances between the to-be-detected icon X and the N fixed detection boxes respectively according to icon coordinates corresponding to the to-be-detected icon X, acquire a nearest distance from the N distances, retain the to-be-detected icon X if the nearest distance is less than a distance threshold, and filter out the to-be-detected icon X if the nearest distance is greater than the distance threshold; and when one or more to-be-detected icons are all filtered at fixed positions, the retained to-be-detected icon is determined as a sound icon.

[0020] The category determining unit is configured to determine an icon category corresponding to the sound icon as an object behavior category of the second virtual object when the sound icon is obtained after the one or more to-be-detected icons are filtered at fixed positions.

[0021] The position determining unit is configured to acquire an icon angle of the sound icon in the target map according to the icon category corresponding to the sound icon and the icon coordinates, and determine the icon angle as an object position of the second virtual object.

[0022] The position determining unit is specifically configured to: in the N fixed detection boxes matched with the icon category corresponding to the sound icon, determine a fixed detection box with a nearest distance to the sound icon as a target detection box according to the icon coordinates corresponding to the sound icon; acquire a default azimuth angle corresponding to the target detection box, and determine the default azimuth angle as the icon angle of the sound icon in the target map.

[0023] The voice generating unit is configured to generate an object behavior prompt voice corresponding to the sound icon according to the object behavior category and the object position.

[0024] The object position further includes an estimated object distance between the first virtual object and the second virtual object.

[0025] The voice generating unit includes:

[0026] The distance determining sub-unit is configured to acquire a color attribute of the sound icon, and determine the estimated object distance between the first virtual object and the second virtual object according to the color attribute.

[0027] The distance determination subunit is specifically configured to perform color space conversion on the initial color matrix of the sound icon to obtain a target color matrix, extract foreground pixels in a color filtering interval from the target color matrix, generate a purity mean value of the foreground pixels, determine a color attribute of the sound icon according to the purity mean value, and map the color attribute to the estimated object distance between the first virtual object and the second virtual object; or, the distance determination subunit is specifically configured to perform normalization processing on the initial color matrix of the sound icon to obtain a target color matrix, perform feature extraction on the target color matrix through a convolution layer in the color recognition model to obtain icon edge features corresponding to the sound icon, perform feature integration on the icon edge features through a full connection layer in the color recognition model to obtain a color feature, output a color attribute corresponding to the color feature through an output layer in the color recognition model, and map the color attribute to the estimated object distance between the first virtual object and the second virtual object.

[0028] The distance determination subunit is specifically configured to obtain a target mask identifier having a mapping relationship between an icon category and an icon angle corresponding to the sound icon, perform matching search on the target mask identifier in a mask configuration file, determine a mask picture matched with the target mask identifier in the mask configuration file as a target mask picture, and store, in the mask configuration file, an association relationship between mask identifiers and mask pictures and pixel coordinates of the mask pictures in a target map.

[0029] The voice generation subunit is configured to generate an object behavior prompt voice corresponding to the sound icon according to the icon category, the icon angle, and the estimated object distance corresponding to the sound icon.

[0030] The device further includes:

[0031] The mask generation module is configured to rotate the at least two icon templates to obtain at least two rotated icon sequences; each of the at least two rotated icon sequences includes N rotated icons corresponding to each icon template; the at least two icon templates have different icon categories; the at least two rotated icon sequences are subjected to effective region cropping to obtain at least two cropped rotated icon sequences; the at least two cropped rotated icon sequences are subjected to binarization processing to obtain at least two target rotated icon sequences; the at least two target rotated icon sequences are subjected to erosion processing and dilation processing; in the at least two target rotated icons, a region obtained after erosion is marked as a first chroma, and a region obtained after dilation is marked as a second chroma, to obtain at least two marked rotated icon sequences including the first chroma and the second chroma; in the at least two marked rotated icon sequences, a region that is not marked as the first chroma and is not marked as the second chroma is marked as a third chroma, to obtain a mask picture corresponding to each rotated icon; a mask identifier corresponding to each mask picture is generated according to an icon category and a rotation angle of each rotated icon, a pixel coordinate of each mask picture in a target map, an association relationship between each mask picture and the mask identifier, and the pixel coordinate of each mask picture in the target map are stored in a mask configuration file.

[0032] The number of the sound icons is one or more, and one sound icon corresponds to one object behavior prompt voice;

[0033] The voice playing module includes:

[0034] The voice selection unit is configured to obtain a reminding interval duration and a reminding priority, extract an object behavior prompt voice satisfying the reminding interval duration from the one or more object behavior prompt voices as a to-be-played object behavior prompt voice, and determine a to-be-played object behavior prompt voice with the highest reminding priority in the to-be-played object behavior prompt voice as a target object behavior prompt voice according to the reminding priority;

[0035] The voice playing unit plays the target object behavior prompt voice.

[0036] The device further includes:

[0037] The detection frame generation module is configured to perform detection frame labeling on sample sound icons in the map sample data, perform superimposed visual processing on the map sample data with the same icon category and the labeled detection frame, obtain a detection frame sample data set corresponding to each icon category, perform rotation enhancement on the detection frame sample data set, and obtain a target detection frame sample data set; the target detection frame sample data set corresponding to the icon category Y includes at least two detection frames; the at least two detection frames in the target detection frame sample data set corresponding to the icon category Y are clustered to obtain N clustering clusters, and N fixed detection frames corresponding to the icon category Y are generated according to the center positions of the N clustering clusters.

[0038] The embodiment of the present application provides a computer device, including: a processor, a memory, a network interface;

[0039] The processor is connected with the memory and the network interface, the network interface is configured to provide a data communication function, the memory is configured to store a computer program, and the processor is configured to call the computer program to execute the method in the embodiment of the present application.

[0040] The embodiment of the present application provides a computer readable storage medium, the computer readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor and executing the method in the embodiment of the present application.

[0041] The embodiment of the present application provides a computer program product or a computer program, the computer program product or the computer program includes computer instructions, the computer instructions are stored in a computer readable storage medium, a processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device executes the method in the embodiment of the present application.

[0042] The embodiment of the application can display a target map in a game picture for controlling a first virtual object, and when the target map has a sound icon for representing an object behavior category and an object position of a second virtual object, an object behavior prompt voice corresponding to the sound icon can be further acquired, and the object behavior prompt voice can be played. As can be seen from the above, in the process that a target user controls the first virtual object to play the game, when the second virtual object belonging to another camp appears near the first virtual object, a corresponding sound icon can be displayed on the target map in the current game picture to indicate the object behavior category and the object position of the second virtual object. In the embodiment of the application, the sound icon can be detected and recognized, so that the sound icon which can only be visualized is converted into a corresponding object behavior prompt voice. By playing the object behavior prompt voice, the target user can directly acquire the behavior information of the enemy near the target user, i.e., the object behavior category and the object position of the second virtual object. Therefore, the player can not need to pay attention to the game mini-map at all times, and the problem of low accuracy and low efficiency in enemy situation judgment caused by the small display size of the game mini-map is reduced. Therefore, the accuracy and efficiency of acquiring the behavior information of the enemy in the game process can be improved, and the performance form of the game can be enriched. BRIEF DESCRIPTION OF DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description only constitute some embodiments of the application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0044] Figure 1 is a system architecture schematic diagram provided by the embodiment of the application;

[0045] Figure 2a-2e is a game data processing scene schematic diagram provided by the embodiment of the application;

[0046] Figure 3 is a game data processing method flow schematic diagram provided by the embodiment of the application;

[0047] Figure 4 is a schematic diagram of different sound icons provided by the embodiment of the application;

[0048] Figure 5 is a icon detection flow schematic diagram provided by the embodiment of the application;

[0049] Figure 6 is a game data processing method flow schematic diagram provided by the embodiment of the application;

[0050] Figure 7a-7b is a scene schematic view of a fixed position filtering provided by an embodiment of the present application;

[0051] Figure 8 is a flow schematic view of a distance identification method provided by an embodiment of the present application;

[0052] Figure 9 is a structure schematic view of a color identification model provided by an embodiment of the present application;

[0053] Figure 10 is a flow schematic view of a data synthesis process provided by an embodiment of the present application;

[0054] Figure 11 is a scene schematic view of data synthesis provided by an embodiment of the present application;

[0055] Figure 12 is a flow schematic view of a data labeling process provided by an embodiment of the present application;

[0056] Figure 13a-13b is a scene schematic view of a data visualization analysis method provided by an embodiment of the present application;

[0057] Figure 14 is a flow schematic view of a mask picture generation process provided by an embodiment of the present application;

[0058] Figure 15 is a flow schematic view of a game data processing method provided by an embodiment of the present application;

[0059] Figure 16 is a structure schematic view of a game data processing apparatus provided by an embodiment of the present application;

[0060] Figure 17 is a structure schematic view of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0061] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0062] Artificial Intelligence (AI) is the theory, method, technology and application system of using digital computer or digital computer controlled machine to simulate, extend and expand human intelligence, perceive environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology of computer science, which attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines, so that machines have the functions of perception, reasoning and decision-making.

[0063] Artificial intelligence technology is a comprehensive discipline, involving a wide range of fields, both hardware and software level technology. Artificial intelligence basic technology generally includes technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction system, mechatronics, etc. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning, etc.

[0064] Computer Vision (CV) is a science that studies how to make machines "see". Further, it refers to using cameras and computers to replace human eyes to identify and measure targets and further process graphics, so that computer processing becomes images more suitable for human eyes to observe or to transmit to instrument detection. As a scientific discipline, computer vision researches related theories and technologies, trying to establish artificial intelligence systems that can obtain information from images or multidimensional data. Computer vision technology usually includes data processing, image recognition, image semantic understanding, image retrieval, OCR, video processing, video semantic understanding, video content / behavior recognition, three-dimensional object reconstruction, 3D technology, virtual reality, augmented reality, simultaneous localization and mapping, etc. It also includes common face recognition, fingerprint recognition and other biometric identification technologies.

[0065] The scheme provided by the embodiments of the present application relates to computer vision technology and deep learning technology of artificial intelligence, and the specific process is described by the following embodiments.

[0066] Please refer to Figure 1, is a schematic diagram of a system architecture provided by an embodiment of the present application. The system architecture may include a business server 100 and a terminal cluster, and the terminal cluster may include: terminal device 200a, terminal device 200b, terminal device 200c, ..., terminal device 200n, wherein a communication connection may exist between the terminal clusters, for example, there is a communication connection between terminal device 200a and terminal device 200b, and there is a communication connection between terminal device 200a and terminal device 200c. At the same time, any terminal device in the terminal cluster may have a communication connection with the business server 100, for example, there is a communication connection between terminal device 200a and the business server 100, wherein the above-mentioned communication connection does not limit the connection method, and may be directly or indirectly connected through a wired communication method, or directly or indirectly connected through a wireless communication method, or through other methods, and the present application does not make any restrictions here.

[0067] It should be understood that Figure 1 Each terminal device in the terminal cluster shown can be installed with an application client. When the application client runs in each terminal device, it can be respectively connected to the above Figure 1 The business servers 100 shown in the figure interact with each other so that the business server 100 can receive business data from each terminal device. Among them, the application client can be a game application, a social application, an instant messaging application, a live broadcast application, a short video application, a video application, a music application, a shopping application, a novel application, a payment application, a browser, and other application clients with the function of displaying text, images, audio, and video data information. Among them, the application client can be an independent client or an embedded sub-client integrated in a client (such as an instant messaging client, a social client, a video client, etc.), which is not limited here. Taking a game application as an example, the business server 100 may be a collection of multiple servers including a gateway server, a scene server, a world server, a database proxy server, an AI server, a chat manager, and the like corresponding to the game application. Therefore, each terminal device may transmit data to the business server 100 through the application client corresponding to the game application. For example, each terminal device may participate in the same game (such as an MMORPG game, full name Massive Multiplayer Online Role-Playing Game) with other terminal devices through the business server 100. During the game, players may control corresponding virtual objects and interact in real time with virtual objects controlled by other players in the virtual space of the game.

[0068] Taking a game application as an example, the business server 100 can provide game login, game data query and the like for the terminal cluster through the communication function, and assuming that the target user logs in a game account through the game application installed on the terminal device A, the terminal device A can receive the business data (for example, information such as a virtual object selected by the target user, an account level, an attribute and the like) returned by the business server 100, and display a corresponding game picture in a screen display area thereof, the game picture being a picture for controlling the virtual object selected by the target user. Similarly, in addition to the target user, other users can also participate in the same game through terminal devices held by the users, and according to the teaming relationship, the virtual objects selected by the users can be divided into different camps. In the game process, when the behavior information of a virtual character controlled by a user changes, the corresponding terminal device will send related business data to the business server 100, and the business server 100 will continuously send the business data of different users to the terminal devices of all related participating users (including the terminal device A and the terminal devices of other users), and then the terminal devices can update the game picture in real time according to the obtained business data. Among them, one picture updating method is to display a sound icon in a target map of the game picture, and the sound icon can be used to represent the object behavior category and object position of the virtual character belonging to the enemy camp, and the method provided by the embodiment of the application can generate the object behavior prompt voice corresponding to the sound icon when the sound icon appears in the target map, and play the object behavior prompt voice, directly prompting the target user of the object behavior category and object position of the virtual character belonging to the enemy camp of different camps through voice.

[0069] Optionally, it can be understood that a plurality of business servers can be included in the system architecture, and one terminal device can be connected with one business server, and each business server can obtain the game picture displayed by the terminal device connected therewith, so as to obtain the corresponding object behavior prompt voice by detecting and identifying the icon of the game picture.

[0070] Optionally, it can be understood that the terminal device can also obtain the game picture, so as to obtain the corresponding object behavior prompt voice by detecting and identifying the icon of the game picture.

[0071] It can be understood that the method provided in the embodiments of the present application can be executed by a computer device, which includes but is not limited to a terminal device or a service server. The service server can be a stand-alone physical server, a server cluster or a distributed system formed by multiple physical servers, or a cloud server providing basic cloud computing services such as cloud database, cloud service, cloud computing, cloud function, cloud storage, network service, cloud communication, middleware service, domain name service, security service, CDN, and big data and artificial intelligence platform. The terminal device can be a smart terminal that can run a game application, such as a smart phone, a tablet computer, a notebook computer, a desktop computer, a palm computer, a mobile internet device (MID), a wearable device (for example, a smart watch, a smart bracelet, etc.), a smart computer, a smart vehicle, etc. The terminal device and the service server can be connected directly or indirectly through a wired or wireless manner, which is not limited in the embodiments of the present application.

[0072] For ease of understanding, the terminal device 200a and the service server 100 are taken as examples for specific description below.

[0073] Please refer to Figure 2a-2e is a scene schematic diagram of game data processing provided by the embodiments of the present application. The implementation process of the game data processing scene can be performed in the service server 100 as shown in Figure 1 , or in any one of the terminal devices (such as the terminal device 200a, the terminal device 200b, the terminal device 200c, or the terminal device 200n) as shown in Figure 1 , or by the terminal device and the service server together, which is not limited here, and the embodiments of the present application take the terminal device 200a and the service server 100 as an example for description. As shown in Figure 2a , the user A has a binding relationship with the terminal device 200a, and multiple applications (such as game applications, education applications, shopping applications, reading applications, etc.) can be installed on the terminal device 200a. Assuming that the user A wants to open one of the applications, such as a target application A1, the terminal device 200a can display a display interface corresponding to the target application A1 on its screen in response to a trigger operation (such as a click operation) for the target application A1. Assuming that the target application A1 is a game application (such as a game accelerator), the terminal device 200a can display a game acceleration page 300a of the target application A1 on the screen, as shown in Figure 2aAs shown, the game acceleration page 300a is the page corresponding to the "acceleration" control 301a. The game can be configured in the game acceleration page 300a. In addition, the target application A1 can also include functional controls such as "game", "discovery", "short video" and "match". By triggering these controls, the corresponding page can be displayed, which will not be repeated here.

[0074] As shown in the game acceleration page 300a, thumbnails of historical games previously played by user A or thumbnails of recent popular games can be displayed in the display area 304a. The configuration control 302a is an entry related to "voice assistance", and the display area 303a can scroll to display recommended information about the latest features related to "voice assistance". The terminal device 200a can switch the current page to the introduction page of the latest feature by responding to the trigger operation for the text information. Assuming that user A clicks on the configuration control 302a of "voice assistance", the terminal device 200a can respond to the click operation on the configuration control 302a and display the auxiliary function configuration page 300b. In the display area 301b of the auxiliary function configuration page 300b, a voice package list can be displayed. The voice package list can include voice package A, voice package B, voice package C, etc. More voice package options can be displayed by sliding the voice package list. Among them, user A can select his favorite voice pack to download and use in the game by selecting the voice pack list. For example, it can be seen in the display area 301b that the current user A is using the voice pack A selected by the system by default. Furthermore, the auxiliary function list can also be displayed in the auxiliary function configuration page 300b. For example, the auxiliary function list can include the "enemy situation reminder" function, the "room area reminder" function, the "aircraft take-off" function, etc. A brief text introduction to the function can be displayed in the relevant area of ​​each function. If user A wants to understand the function more deeply, he can enter the details page. For example, the terminal device 200a can respond to the trigger operation of the detail control 302b to display the details page 300c corresponding to the "enemy situation reminder" function. The specific form can be seen Figure 2b ,like Figure 2bAs shown, the details page 300c displays a detailed description of the "Enemy Situation Alert" function. When this function is enabled, the terminal device 200a can provide voice alerts about enemy actions, including the corresponding distance and direction. For example, as shown in scene c1, during a game session, when gunfire is heard in the northeast direction near the player, the player will receive a voice alert: "Northeast, someone is shooting nearby." Furthermore, the "Enemy Situation Alert" function also provides alerts for nearby footsteps. For example, as shown in scene c2, during a game session, when someone walks near the player, the player will receive a voice alert: "Someone is nearby" or "Someone is next to you." Furthermore, the "Enemy Situation Alert" function also provides alerts for nearby car sounds. For example, as shown in scene c3, during a game session, when someone drives a car in the southwest direction near the player, the player will receive a voice alert: "Someone is driving next to you in the southwest." Therefore, by enabling the "Enemy Situation Alert" function, user A can quickly and efficiently obtain enemy information during the game. After exiting the details page 300c, the terminal device 200a will return to the display as shown below. Figure 2a The auxiliary function configuration page 300b in the system can then respond to the triggering operation of the "enemy situation reminder" switch control 303b to turn on the "enemy situation reminder" function.

[0075] After user A completes all configurations, the terminal device 200a can exit the auxiliary function configuration page 300b and return to the game acceleration page 300a. Further, the terminal device 200a can respond to the triggering operation of the "start acceleration" control 305a and update the game acceleration page 300a to the game acceleration page 300d, as shown in FIG. Figure 2c As shown, the games installed on the terminal device 200a (such as Game A, Game B, Game C, Game D, etc.) can all start acceleration optimization. It can be understood that if user A does not want to continue to accelerate the game, that is, the "stop acceleration" control 302d can be triggered, and the terminal device 200a will stop the acceleration optimization of the game. In addition, during the acceleration process, a game list can be displayed in the display area 303d. The terminal device 200a can respond to the user A's selection operation for the game list, determine the game selected by user A as the target game, and then respond to the trigger operation of the "enter game" control 301d to start the target game.

[0076] Assuming that user A selects game A, after clicking the "enter game" control 301d, Figure 2dAs shown, the terminal device 200a can initiate a login request by connecting to the business server 100 through the application client of game A. The business server 100 then initiates an identity data verification query and returns the authentication result after completing the data query. If the identity verification is passed, the business server 100 continues to query and return the account status data (such as role, level, attributes, the scene server where the last login was made and coordinates, etc.), and then the account status data can be sent to the corresponding scene server (for ease of understanding, the business server 100 and the scene server here are independently deployed servers). At the same time, the user A online notification can be broadcast to user A’s friends and the player online status detection (control disconnection reconnection and disconnection timeout). Further, after receiving the authentication result, the business server 100 establishes a connection with the corresponding scene server, and user A successfully logs in to the scene server, so that it can be displayed as follows Figure 2d The game screen 300e shown. Figure 2d As shown, the game screen 300e can display the virtual object manipulated by user A (for ease of distinction, this virtual object will be referred to as the first virtual object), the current skills possessed, and the target map corresponding to the area centered on the first virtual object. It will be understood that the first virtual object may not always be displayed in the game screen 300e, but its displayed content will change as the viewing angle of the first virtual object changes. The scene server begins writing all user logs to the log and simultaneously sends user A's related data or query request to the business server 100.

[0077] During the game, the terminal device 200a can respond to the trigger operation (such as click operation, sliding operation, dragging operation, etc.) of user A on the game screen 300e to control the movement of the first virtual object, or switch skills, switch scenes, etc., so that the game screen 300e can be updated according to the trigger operation of user A. Assume that the first virtual object controlled by user A belongs to the first camp. When a second virtual object belonging to the enemy camp (the enemy camp can be referred to as the second camp here) appears near the first virtual object, and the second virtual object performs a specific behavior (such as walking, running, shooting, driving, etc.), a corresponding sound icon will appear on the target map. For example, see Figure 2d In the game screen 300e, when the second virtual object drives a car near the first virtual object, a sound icon corresponding to the car sound can be seen on the target map. The second camp can be one or more camps in addition to the first camp.

[0078] Please also see Figure 2e, the terminal device 200a can generate a car sound icon according to the object behavior category and the object position of the second virtual object, and display the car sound icon at the corresponding position of the target map, so that in order to obtain information related to the second virtual object, the terminal device 200a can input the target map icon detection model, and the icon detection model can output the icon category and icon coordinates corresponding to the car sound icon. Further, according to the icon category and icon coordinates corresponding to the car sound icon, the icon angle of the car sound icon in the target map can be obtained, in addition, the color attribute of the car sound icon can also be obtained, and then the estimated object distance between the first virtual object and the second virtual object can be determined according to the obtained color attribute. Finally, according to the icon category, the icon angle and the estimated object distance corresponding to the car sound icon, the object behavior prompt voice corresponding to the car sound icon can be generated, so that the object behavior prompt voice can be directly played, for example, the generated object behavior prompt voice can be "there is a car in the northeast 100 meters", and the voice prompt can be performed according to the aforementioned voice package A in use, so that the user A can obtain effective enemy behavior information without paying attention to the sound icon appearing on the target map at all times. Optionally, the detection process of the car sound icon can also be performed by the business server 100, and here the terminal device 200a is taken as an example for description, and the embodiments of the present application are not limited thereto.

[0079] The icon detection model can be a lightweight neural network, and the terminal device 200a (or the business server 100) can train the icon detection model by using a video database with a large amount of game videos, and the specific process can be referred to the step S206 in the embodiment corresponding thereto. Figure 6

[0080] It should be noted that for the competition experience type game, the sound icon in the embodiments of the present application can also include a footstep sound icon, a gun sound icon, a silenced gun sound icon, an explosion sound icon, a glass breaking sound icon, etc. There can also be a case that multiple sound icons with different types are displayed simultaneously in the target map at the same time. Here, only a single car sound icon is taken as an example for description, and the detection process of other sound icons appearing on the target map is similar to that of the car sound icon, which will not be described here.

[0081] Optionally, in addition to starting the "enemy situation reminder" function through the game accelerator, the "enemy situation reminder" function can also be directly implanted into the game application, so that the "enemy situation reminder" function can be directly called through the game application process after starting the game application, to realize voice reminding of the category, distance and direction of the enemy behavior. Correspondingly, the user can also select whether to start the "enemy situation reminder" function through the game application.

[0082] ​It can be known from the above that the embodiment of the application provides a game enemy situation identification method based on image recognition based on a deep neural network. In the process in which a target user (such as a user A) controls a first virtual object to play a game, when a second virtual object belonging to another camp appears near the first virtual object, a corresponding sound icon can be displayed on a target map in a current game screen to indicate an object behavior category and an object position of the second virtual object. Further, the sound icon can be detected and identified, so as to convert the originally visualized sound icon into a corresponding object behavior prompt voice. The object behavior prompt voice can be played to enable the target user to directly obtain enemy behavior information near the target user, that is, the object behavior category and the object position of the second virtual object. Therefore, the player can not need to pay attention to a game mini-map at all times, and the problem of low accuracy and low efficiency of enemy situation judgment caused by a small display size of the game mini-map is reduced. Therefore, the accuracy and the efficiency of obtaining enemy behavior information in the process of playing the game can be improved, and the performance form of the game can be enriched.

[0083] Please refer to Figure 3 , Figure 3 is a flowchart of a game data processing method provided by the embodiment of the application. The game data processing method can be executed by a computer device. The computer device can include a terminal device or a service server as described in Figure 1 , and here, the terminal device is taken as an example. As shown in Figure 3 , the game data processing method can at least include the following steps S101-S103:

[0084] Step S101, displaying a target map in a game screen used for controlling a first virtual object; the first virtual object belongs to a first camp; the target map is used for displaying a sound icon representing an object behavior category and an object position of a second virtual object; the second virtual object belongs to a second camp;

[0085] Specifically, the terminal device can first open a game acceleration application, and then display an auxiliary function configuration page (such as an auxiliary function configuration page 300b in Figure 2a ) in response to a trigger operation (such as a click operation) on a configuration control (such as a configuration control 302a in the game acceleration page 300a in Figure 2a ). The auxiliary function configuration page can include an object behavior prompt control (such as an object behavior prompt control 302b in the auxiliary function configuration page 300b in Figure 2a ). Figure 2athe object behavior voice prompt function can be used to detect the sound icon in the target map during the game process, and trigger the acquisition of the object behavior prompt voice corresponding to the sound icon. After the configuration is completed, the terminal device can exit the auxiliary function configuration page, return to the game configuration page, and can enter the game application in response to the start operation of the game application in the game configuration page, and display the game screen for controlling the first virtual object, wherein the content displayed by the game screen corresponds to the perspective of the first virtual object. Further, the terminal device can display the target map centered on the first virtual object in the game screen, and the target map can be displayed in a certain area in the game screen, and the object behavior category and object position of the second virtual object in the area range contained by the target map are acquired, so that the sound icon can be generated according to the above-mentioned object behavior category and object position, and the sound icon is displayed in the target map, that is, when the second virtual object appears in the area range centered on the first virtual object (that is, the target map) and the second virtual object performs a certain behavior and emits a corresponding sound, the corresponding sound icon will be displayed in the target map in addition to the corresponding sound effect played in the game screen. The specific process can be referred to the related description in the above-mentioned Figure 2a-2e corresponding embodiments. As Figure 2e The target map shown in the target map is the position of the first virtual object at the center point of the target map, and the sound icon, terrain, road, important place, position mark of other virtual objects in the area, etc. can be displayed on the target map.

[0086] In an embodiment, the sound icon includes but is not limited to a car sound icon, a footstep sound icon, a gun sound icon, a silencer gun sound icon, an explosion sound icon, and a glass breaking sound icon, which respectively represent the behaviors of driving a car, walking, shooting a gun, shooting a silencer gun, throwing a bomb, and breaking glass, that is, the object behavior category of the second virtual object. It should be noted that during the game process, the explosion sound represents that an enemy (that is, the second virtual object) or a teammate (a virtual object belonging to the first camp) throws a bomb and explodes in a certain place, so the explosion position cannot represent the position of the enemy, so the explosion sound icon is not reminded in the embodiment of the present application; and the glass breaking sound represents that an enemy or a teammate breaks glass, and it cannot be determined that it is broken by the enemy, so the glass breaking sound icon is not reminded, so after identifying that the sound icon is an explosion sound icon or a glass breaking sound icon, the terminal device can no longer perform the next step of identification. For details, please refer to Figure 4 is a schematic diagram of different sound icons provided by an embodiment of the present application. As Figure 4As shown, different types of sound icons have different icon styles, which can represent different object behavior categories, and the icon angle of the sound icon in the target map can represent the orientation information of the second virtual object relative to the first virtual object, and the color attribute (specifically the lightness of the color) corresponding to the sound icon can also be used to represent the estimated object distance between the first virtual object and the second virtual object. The above icon angle and estimated object distance can constitute the object position of the second virtual object. As can be seen, detecting and identifying the sound icon can obtain the object behavior category and object position of the second virtual object. It can be understood that in different game products, the sound icon can also have other types and styles, which are not limited by the present application.

[0087] In the game match, the virtual objects (which can include virtual characters and virtual items) selected by the players can be divided into different camps. For ease of understanding and differentiation, it is assumed that the first virtual object belongs to the first camp and the second virtual object belongs to the second camp, i.e., the second virtual object refers to the collective term of virtual objects in a different camp from the first virtual object. Therefore, the number of second virtual objects can include one or more, and the embodiments of the present application are not limited.

[0088] It should be noted that the game acceleration application and the game application can be independent applications. For example, the game acceleration application can be an embedded application in the game application, and the embodiments of the present application do not limit this. Figure 2a-2e In the corresponding embodiments, the user A can click the "enter game" control 301d in the target application A1 (i.e., the game acceleration application) to launch the game A (i.e., the game application) in the target application A1. Optionally, the game acceleration application and the game application can be integrated into one application. For example, the game acceleration application can be an embedded application in the game application, and the embodiments of the present application do not limit this.

[0089] In step S102, when the target map has a sound icon, the object behavior prompt voice corresponding to the sound icon is obtained. The object behavior prompt voice is used to voice prompt the object behavior category and object position of the second virtual object.

[0090] Specifically, as described above, the terminal device will detect the content of the target map in real time. When the second virtual object appears in the target map and performs a certain behavior and emits a corresponding sound, the terminal device will render and draw the corresponding sound icon at the corresponding position in the target map. At the same time, the terminal device can detect the game picture in real time, i.e., perform icon recognition on the target map to detect whether there is a sound icon in the target map. For specific processes, please refer to Figure 5 is a flowchart of an icon detection process provided by an embodiment of the present application. As shown in Figure 5As shown, the terminal device first identifies the position of the target map in the game screen, and extracts the target map from the game screen based on the identified position information. The target map can then be input into the icon detection model, and feature extraction is performed through the feature extraction network in the icon detection model to obtain a picture feature matrix, wherein the picture feature matrix is ​​used to characterize the features of different scales corresponding to one or more icons to be detected identified by the icon detection model, and the features of different scales can be used to detect the detection frame subsequently. In an embodiment of the present application, the feature extraction network can extract picture feature matrices with different scales through branches with different network structures. For example, two picture feature matrices can be obtained through two different branches in the feature extraction network, each picture feature matrix containing information about the icon category, icon coordinates, and confidence. Further, the picture feature matrix can be subjected to non-maximum suppression (NMS) processing to obtain the icon category, icon coordinates, and confidence corresponding to one or more icons to be detected. Among them, the icon coordinates refer to the position coordinates of the icon to be detected on the target map, so the icon coordinates are a set of relative coordinates. Among them, the confidence can be used to filter out repeated and redundant detection frames during the non-maximum suppression processing.

[0091] Furthermore, one or more icons to be detected can be filtered at a fixed position based on the icon categories and icon coordinates corresponding to the one or more icons to be detected, and the falsely detected icons to be detected can be filtered out to obtain a sound icon, and then the icon category corresponding to the sound icon can be determined as the object behavior category of the second virtual object. At the same time, the icon angle of the sound icon in the target map can be obtained based on the icon category and icon coordinates corresponding to the sound icon, and the icon angle can be determined as the object position of the second virtual object. In addition, the terminal device can also obtain the color attribute of the sound icon, and the estimated object distance between the first virtual object and the second virtual object can be obtained based on the color attribute, so that the icon angle and the estimated object distance can be used together as the object position of the second virtual object. Finally, the object behavior prompt voice corresponding to the sound icon can be generated based on the object behavior category and the object position. Among them, the icon angle refers to the relative angle of the sound icon corresponding to the first virtual object at the center position of the target map.

[0092] Step S103: Play the object behavior prompt voice.

[0093] Specifically, since the number of sound icons in the detected target map at the same time can be one or more, and one sound icon corresponds to one object behavior prompt voice, the embodiment of the application provides a reminding scheduling method, which judges whether voice prompt is needed at present by designing a reminding interval length and a reminding priority, and optimizes the object behavior type that needs to be reminded, and then reminds the player through voice on the client of the game application. The specific process is as follows: the terminal device first acquires the reminding interval length and the reminding priority, then extracts the object behavior prompt voice that meets the reminding interval length from one or more object behavior prompt voices as the to-be-played object behavior prompt voice, and then determines the to-be-played object behavior prompt voice with the highest reminding priority in the to-be-played object behavior prompt voice as the target object behavior prompt voice, and then the target object behavior prompt voice can be played. The application adopts the reminding priority and the reminding interval length to schedule the voice reminder, which can avoid the frequent voice reminder from affecting the game experience in the intense battle. For example, if a sound icon of a silencer gunshot sound appears in the northeast direction (relative to the first virtual object) in the target map, and a sound icon of a footstep sound appears in the northwest direction, the terminal device can identify the icon categories of the two sound icons as the silencer gunshot sound icon and the footstep sound icon by the method described in step S102, and can detect that the icon angle corresponding to the silencer gunshot sound icon is 45° and the icon angle corresponding to the footstep sound icon is 345°, assuming that the reminding priority is set as silencer gunshot sound greater than footstep sound, and the time interval from the last voice reminder is greater than the set reminding interval length, the terminal device can perform voice reminder for the silencer gunshot sound icon, and the corresponding playing content is the voice of "northeast, someone shoots nearby".

[0094] It can be understood that the developer can realize diversified voice prompt rules by adjusting the reminding interval length and the reminding priority.

[0095] In an implementation, the reminding priority can be set as: gunshot (including silencer gunshot) is greater than footstep sound, and footstep sound is greater than vehicle sound, and the reminding interval length is set as: the reminding interval length between sound icons of the same category is 15 seconds, and the reminding interval length between sound icons of different categories is 10 seconds. The detection frequency of the game picture is 1 time per second, when a sound icon is detected, whether the time interval between the current time stamp and the time stamp corresponding to the last voice reminder has reached the set reminding interval length is judged according to the setting of the reminding priority, only when the reminding interval length reaches and there is a sound icon in the target map, the object behavior prompt voice that needs to be played is selected in the order of the reminding priority.

[0096] The embodiment of the application mainly uses icon recognition + fixed position filtering + icon color recognition for cascade. Alternatively, a detection model can be used to identify the icon category and color attribute of each type of sound icon, form 8 icon categories (including 4 sound categories and 2 color attributes for each category), and perform detection. Subsequently, a color recognition model can not be used, and only a fixed position filtering module (used to perform the above process of filtering one or more icons to be detected in a fixed position) is needed. In addition, in the embodiment of the application, data analysis obtains the conclusion that each type of sound icon has N angles. The model can also directly identify the category of the screenshot of the N fixed detection boxes of each type of sound icon, so as to determine whether there is a sound icon to be detected at this position. When N = 12 and there are 4 categories of sound icons, 48 fixed detection boxes need to be identified each time. Since the identification model is small, the inference speed is fast, and in terms of performance, it can be easily deployed on a related client.

[0097] The embodiment of the application can display a target map in a game picture for controlling a first virtual object, and when the target map has a sound icon for representing an object behavior category and an object position of a second virtual object, an object behavior prompt voice corresponding to the sound icon can be further obtained, and the object behavior prompt voice can be played. As can be seen above, in the process in which a target user controls a first virtual object to play a game, when a second virtual object belonging to another camp appears near the first virtual object, a corresponding sound icon can be displayed on a target map in a current game picture to indicate an object behavior category and an object position of the second virtual object. The embodiment of the application can use an AI target detection and image recognition-based method to accurately detect and identify a sound icon on a target map to determine whether there is an enemy situation around the first virtual object, and use a voice to remind the behavior category, distance, and angle. That is, the originally visualized sound icon can be converted into a corresponding object behavior prompt voice. By playing the object behavior prompt voice, a player (i.e., a target user) can directly obtain enemy behavior information near the player, i.e., an object behavior category and an object position of a second virtual object. Therefore, the player can not have to pay attention to a target map at all times and can not have to concentrate on listening to enemy sound in a game. Compared with a scheme without a voice reminder, the embodiment of the application can reduce the problem of low accuracy and low efficiency in judging an enemy situation caused by a small display size of a game mini-map, and can improve the accuracy and efficiency of obtaining enemy behavior information in a game process, and can enrich the performance of a game and improve the game experience.

[0098] See Figure 6 is a flowchart of a game data processing method provided by the embodiment of the application. The game data processing method can be executed by a computer device, which can include, for example, Figure 1The terminal device or the service server. Figure 6 The game data processing method can include the following steps:

[0099] In step S201, a target map in a game picture for controlling a first virtual object is obtained, an icon detection model is used to perform icon recognition on the target map, and one or more to-be-detected icons are obtained, and the one or more to-be-detected icons correspond to icon categories and icon coordinates, respectively; the first virtual object belongs to a first camp; the target map is used to display a sound icon representing a behavior category and a position of a second virtual object; and the second virtual object belongs to a second camp.

[0100] Specifically, after the computer device opens a game application, a game picture for controlling a first virtual object can be displayed. The game picture can display a target map centered on the first virtual object. The game picture and the target map are updated in real time. When a second virtual object appears in the target map and performs a certain behavior and emits a corresponding sound, the computer device can render and draw a corresponding sound icon at a corresponding position in the target map. Correspondingly, the computer device can obtain the game picture in real time, and then cut out the target map from the game picture. Then, the computer device can input the target map into an icon detection model for detection to obtain icon categories and icon coordinates corresponding to one or more to-be-detected icons, respectively. It can be understood that the first virtual object and the second virtual object belong to different camps.

[0101] The target map input into the icon detection model is represented in an RGB color representation, i.e., the target map includes three color channels of R (red), G (green), and B (blue). For example, the computer device can read a color matrix of the target map, and the size of the color matrix can be limited to 416*416*3. After feature extraction, a first picture feature matrix of 13*13*27 and a second picture feature matrix of 26*26*27 (both of which belong to picture feature matrices) can be obtained. Since multiple detection boxes can be generated in the icon recognition process to frame the to-be-detected icons that can be sound icons, non-maximum suppression processing is required. The purpose of the non-maximum suppression processing is to remove redundant detection boxes, extract independent and maximum confidence detection boxes from the prediction results as the detection results, and finally map the detection results to icon categories and icon coordinates, respectively. The RGB color mode is an industry standard for colors. It obtains various colors by changing the three color channels of red, green, and blue and superimposing them on each other. This standard almost includes all colors that can be perceived by human vision, and it is one of the most widely used color systems.

[0102] The specific process of this step can be referred to in the above Figure 3Steps S101 and S102 in the corresponding embodiment will not be described in detail here.

[0103] It should be noted that before executing this step, step S206 may be executed first to obtain the icon detection model.

[0104] Step S202, performing fixed position filtering on the one or more icons to be detected based on the icon categories and icon coordinates corresponding to the one or more icons to be detected, obtaining a sound icon, and determining the icon category corresponding to the sound icon as the object behavior category of the second virtual object;

[0105] Specifically, since the icon to be detected identified in step S201 is not necessarily a sound icon, the icon detection model may have false detections. Therefore, in order to filter out falsely detected icons, one or more icons to be detected can be filtered at fixed positions based on the icon categories and icon coordinates corresponding to the one or more icons to be detected. The specific process can be: assuming that the one or more icons to be detected include icon X to be detected, N fixed detection frames that match the icon category corresponding to icon X to be detected can be obtained based on the icon category corresponding to icon X to be detected. Then, based on the icon coordinates corresponding to icon X to be detected, the distances between icon X to be detected and the N fixed detection frames can be calculated, and the closest distance is selected from the N distances. The closest distance is then compared with a set distance threshold: if the closest distance is less than the distance threshold, icon X to be detected is considered to be a sound icon, and therefore icon X to be detected can be retained; if the closest distance is greater than the distance threshold, icon X to be detected is considered to be a false detection, and therefore icon X to be detected needs to be filtered out. The detection process for other icons to be detected is the same as that for icon X to be detected, and will not be repeated here. When all icons to be detected have completed fixed position filtering, the icons to be detected that are finally retained can be determined as sound icons. Among them, the fixed position filtering process can be implemented in the icon detection model. The distance threshold can be set according to actual needs, and the embodiment of the present application does not limit its numerical value. Among them, the fixed detection frames can be pre-generated by the method described in step S207. They are respectively located at fixed positions in the target map, and each icon category will correspond to N fixed detection frames respectively. Therefore, before executing step S202, step S207 can be executed first to generate a fixed detection frame.

[0106] In one embodiment, a data visualization analysis method is used to manually annotate multiple map sample images containing sound icons with sound icons, and the annotated frames are overlaid and visualized. It can be determined that each sound icon has 12 fixed detection frames (i.e., N=12), and the interval angle between adjacent fixed detection frames is 30 degrees. Please also refer to Figure 7a-7b , is a schematic diagram of a fixed position filtering scenario provided by an embodiment of the present application. Figure 7aAs shown, in the scenario of detecting footsteps, after inputting the target map 400a into the icon detection model, it can be detected that the target map 400a contains a to-be-detected icon A, a to-be-detected icon B and a to-be-detected icon C, and the icon categories and icon coordinates corresponding to the three to-be-detected icons can be identified. Assuming that the three to-be-detected icons are all identified as a footsteps icon, 12 fixed detection boxes corresponding to the footsteps icon can be further obtained, such as Figure 7a As shown, the 12 fixed detection boxes are located at fixed positions in the target map 400a, and the distances between the 12 fixed detection boxes and the to-be-detected icons A, B and C can be calculated by the distance formula , wherein (x', y') represents the coordinates of the fixed detection box, and (x, y) represents the coordinates of the to-be-detected icon. The nearest distance between each to-be-detected icon and the fixed detection box can be found. In the embodiments of the present application, the distance threshold can be set to 8 pixels. The size relationship between the nearest distance and the distance threshold can be determined to detect whether the current to-be-detected icon is a sound icon. For example, the distance between the to-be-detected icon C and the fixed detection box P can be calculated by the distance formula to be the nearest distance d, that is, the fixed detection box P is closest to the to-be-detected icon C. When the nearest distance d is less than 8 pixels, it can be determined that the to-be-detected icon C is a sound icon. After the above process, it can be finally determined that the to-be-detected icons A and B are both false sound icons, so they can be filtered out, and the to-be-detected icon C is a true footsteps icon. For details, please refer to Figure 7b In the scenario of detecting gunshots, the icon detection model can detect that the target map 400b contains a to-be-detected icon D and a to-be-detected icon E, and both of them are identified as a gunshot icon. Therefore, 12 fixed detection boxes corresponding to the gunshot icon can also be obtained, and after the same fixed position filtering process, it can be finally determined that the to-be-detected icon D is a true gunshot icon, and the to-be-detected icon F is a false gunshot icon. It can be understood that the fixed position filtering is a process of filtering out to-be-detected icons far away from the fixed detection boxes obtained by analysis. Through fixed position filtering, false to-be-detected icons can be removed, thereby improving the accuracy of model recognition.

[0107] Further, the icon category corresponding to the sound icon filtered by the fixed position can be determined as the object behavior category of the second virtual object. For example, when the sound icon is a car sound icon, it indicates that the second virtual object is driving a vehicle (such as a car, a tank, etc.), and when the sound icon is a footstep sound icon, it indicates that the second virtual object is walking (or running, jumping, etc.). It can be understood that one icon category corresponds to one object behavior category. In the embodiments of the present application, the gun sound icon and the silenced gun sound icon both indicate that the second virtual object is shooting, but the sound position corresponding to the silenced gun sound icon is closer to the first virtual object.

[0108] As described above, the embodiments of the present application abandon the traditional template matching method to detect the sound icon. The template matching method adopts a sliding window and a multi-angle matching strategy, which is very time-consuming, and the matching accuracy is very low due to the large change of the target map background. The embodiments of the present application use an icon detection model to detect the icon and perform fixed position filtering optimization, so that a detection result with high accuracy can be obtained.

[0109] In step S203, the icon angle of the sound icon in the target map is obtained according to the icon category corresponding to the sound icon and the icon coordinates.

[0110] Specifically, as described above in the fixed position filtering process, the computer device can select the fixed detection frame closest to the sound icon from the N fixed detection frames matched with the icon category corresponding to the sound icon to determine as the target detection frame, and then the detection frame serial number corresponding to the target detection frame can be obtained. The default azimuth angle corresponding to the target detection frame can be obtained by querying the detection frame serial number, and thus the default azimuth angle can be determined as the icon angle of the sound icon in the target map. In an embodiment, 12 fixed detection frames are configured for each sound icon, and the interval angle between adjacent fixed detection frames is 30 degrees. At this time, the default azimuth angle corresponding to the fixed detection frame can include 12 angles, which are 15°, 45°, 75°, 105°, 135°, 165°, 195°, 225°, 255°, 285°, 315°, and 345°. For example, in the above Figure 7a In the corresponding embodiment, the fixed detection frame P is closest to the to-be-detected icon C, and thus the fixed detection frame P can be used as the target detection frame. Assuming that the detection frame serial number of the fixed detection frame P is Q, the default azimuth angle associated with the detection frame serial number Q can be determined as the icon angle of the to-be-detected icon C. For example, in the target map 400a Figure 7a of the above, the icon angle of the to-be-detected icon C is 255°. Alternatively, the angle between the line connecting the sound icon and the first virtual object located at the center position of the target map and the north direction of the target map can also be calculated as the icon angle of the sound icon in the target map.

[0111] In step S204, the color attribute of the sound icon is acquired, the estimated object distance between the first virtual object and the second virtual object is determined according to the color attribute, and the estimated object distance and the icon angle are determined as the object position of the second virtual object.

[0112] Specifically, the application provides three configurable and selectable distance recognition methods based on icon foreground color recognition, which are feature threshold method, mask perspective method and image recognition method. The three methods can perform color recognition on the sound icon to obtain the estimated object distance between the first virtual object and the second virtual object, and the specific use scheme can be obtained according to the configuration.

[0113] Optionally, please see Figure 8 is a flowchart of a distance recognition method provided by an embodiment of the application. As shown in Figure 8As shown, the computer device can detect and intercept the sound icon from the target map. Since the image pixels are saved in a matrix manner, in order to facilitate subsequent color recognition, the initial color matrix of the sound icon can be converted in color space by using the HSV color conversion mode to obtain a target color matrix (i.e., the HSV numerical matrix corresponding to the sound icon). The initial color matrix is stored in the order of B, G and R (or in the order of R, G and B) in RGB color mode, while the target color matrix is stored in the order of H, S and V in HSV color space model (also known as hexagonal pyramid model), and the parameters of the color in the HSV color space model are hue (H), saturation (S) and value (V). In the embodiments of the present application, the conversion script for converting the initial color matrix into the target color matrix according to the color space conversion rule can be: hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV), where img represents the initial color matrix and hsv represents the target color matrix. Further, the foreground pixels within the color filtering interval can be extracted from the target color matrix, and the purity mean of the foreground pixels is generated, and then the color attribute of the sound icon can be determined according to the purity mean, so that the color attribute can be mapped to the estimated object distance between the first virtual object and the second virtual object. The color filtering interval can be set in advance according to actual needs, which is used to filter the foreground content with color values within the color filtering interval by setting the hue threshold range, the saturation threshold range and the value threshold range. Preferably, the matrix filtering method can be used to obtain the foreground pixels with foreground color label true and background color label false, and the corresponding matrix screening script is: sat = np.zeros_like(h); sat[np.where(((h>h_threshold[0])&(h<h_threshold[1]))&((s>s_threshold[0])&(s<s_threshold[1])&(v>v_threshold[0])&(v<v_threshold[1])))]=1, where the sat matrix obtained as the screening result can be used to determine the pixels corresponding to the positions with value 1 (indicating true) in the sat matrix as foreground pixels, where h_threshold[0] and h_threshold[1] represent the minimum value and the maximum value of the hue threshold range respectively, s_threshold[0] and s_threshold[1] represent the minimum value and the maximum value of the saturation threshold range respectively, and v_threshold[0] and v_threshold[1] represent the minimum value and the maximum value of the value threshold range respectively.wherein the purity mean refers to the mean value of the saturation of the foreground pixels, and the purity mean is calculated by taking the average of the saturation s of all the pixels corresponding to the positions with a value of 1 in the sat matrix. For example, Figure 8 As shown in the figure, in a preferred embodiment, the hue threshold range can be set as (9, 17), the saturation threshold range can be set as (100, 250), and the lightness threshold range can be set as (100, 235). The foreground content corresponding to the sound icon can be obtained through the screening of the three threshold ranges, and then the purity mean of all the foreground pixels contained in the foreground content can be calculated. According to the purity mean and the set color threshold (color_threshold), the color attribute of the sound icon can be determined, wherein the color attribute can include dark and light, that is, the color attribute of the foreground color of the sound icon can be determined according to the color threshold: when the purity mean is greater than the color threshold, the color attribute of the sound icon can be determined as dark; when the purity mean is less than the color threshold, the color attribute of the sound icon can be determined as light. Further, the color depth can be mapped to the distance: if the color attribute is dark, the estimated object distance is near distance, that is, the behavior of the second virtual object occurs in the near distance; if the color attribute is light, the estimated object distance is far distance, that is, the behavior of the second virtual object occurs in the far distance. This distance recognition method can be referred to as the feature threshold method.

[0114] wherein the estimated object distance obtained by mapping the color attribute is only divided into near distance and far distance, the near distance can represent a circular region range in the game world with the first virtual object as the center and 300 meters as the radius, and the far distance can represent a circular region range in the game world with the first virtual object as the center and 500 meters as the radius.

[0115] Optionally, according to the data visualization analysis in step S207, it can be obtained that each type of sound icon has N fixed detection boxes and corresponding N default azimuth angles on the target map. Since the game client renders and draws the same type of sound icon at a specific position in the target map, the result is the same, that is, for the same sound icon, the foreground area is also the same at the same position in the target map. Therefore, in order to obtain the position of the foreground pixel, the embodiment of the present application can pre-acquire the foreground and background separation mask image corresponding to the sound icon at each fixed position and the pixel coordinates of the mask image in the target map based on the above-mentioned prior knowledge, and then obtain the nearest icon coordinates and icon angle corresponding to the detected sound icon in actual application. According to the mask image corresponding to the icon angle, the foreground pixels and background pixels can be directly separated, so that the purity mean of the foreground pixels can be generated to perform color distance. The distance mapping, this distance recognition method can be called the mask perspective method, and the specific process can be: first, the computer device obtains a target mask identifier with a mapping relationship between the icon category and the icon angle corresponding to the sound icon, and then a matching search can be performed in the mask configuration file according to the target mask identifier, so that the mask image matching the target mask identifier in the mask configuration file can be determined as the target mask image, and further, the target mask image is parsed to obtain the foreground pixel coordinates of the sound icon in the target map, and the foreground pixels corresponding to the sound icon can be determined according to the foreground pixel coordinates, and then the average value of the saturation of the foreground pixels can be calculated as the purity mean, and the color attribute of the sound icon can be determined according to the purity mean, so that the color attribute can be mapped to the estimated object distance between the first virtual object and the second virtual object. The judgment and mapping process of the color attribute are consistent with the relevant process in the above-mentioned feature threshold method, and will not be repeated here. Among them, the above mask configuration file is used to store the association between the mask identifier (also known as the mask id) and the mask image, as well as the pixel coordinates of the mask image in the target map. It can be understood that each sound icon corresponds to N positions, and there will be corresponding N mask images. For example, in the above Figure 4 In the corresponding embodiment, the sound icons to be detected are vehicle sound icons, footsteps icons, gunshot icons, and silenced gunshot icons. If each sound icon can appear in 12 locations on the target map, then in this scenario, 12 mask images will be generated for each sound icon, ultimately resulting in 48 mask images. The mask identifier corresponding to each mask image is unique. Therefore, before executing this step, step S208 can be executed to generate the mask image.

[0116] Optionally, the embodiment of the present application also supports the use of an end-to-end neural network classification model (which may be called a color recognition model) to directly identify the color attributes of the sound icon. This distance recognition method may be called an image recognition method. Figure 9 , is a structural diagram of a color recognition model provided in an embodiment of the present application. Figure 9 As shown, the process of distance recognition by the color recognition model can be specifically as follows: after the computer device obtains the sound icon to be recognized, it can obtain the initial color matrix of the sound icon (i.e., the RGB value matrix corresponding to each pixel). Since the processing order of the image color is not exactly the same when executing this step on different platforms, the initial color matrix that does not meet the color processing order needs to be converted into a channel first. For example, the color processing order is set to R channel, G channel, B channel, and on the Android platform and the Linux platform, the order of reading the color is B channel, G channel, R channel. Therefore, on these two platforms, the initial color matrix needs to be converted into a matrix stored in RGB order; on the iOS platform, the order of reading the color is R channel, G channel, B channel, so the initial color matrix does not need to be converted into a channel on this platform. Further, the initial color matrix that meets the color processing order is normalized to obtain a target color matrix. For example, all the values ​​in the initial color matrix can be divided by 255 at the same time to obtain a target color matrix whose values ​​are all floating-point numbers between 0 and 1. Then, the target color matrix can be input into the color recognition model for processing. The target color matrix first passes through a 3*3 convolution layer. In this convolution layer, the target color matrix is ​​sequentially subjected to convolution operations, nonlinear factors are introduced using the linear unit activation function (ReLU), and maximum pooling processing (maxpooling) to obtain intermediate features. The intermediate features are then input into the next 5*5 convolution layer. After the same convolution operation, activation, and maximum pooling processing, the icon edge features corresponding to the sound icon can be output. Furthermore, the icon edge features are integrated through the fully connected layer in the color recognition model to obtain color features, and then logsoftmax is used as the final activation function for standardization. The color attributes corresponding to the color features can be output through the output layer in the color recognition model, so that the color attributes can be mapped to the estimated object distance between the first virtual object and the second virtual object. Among them, the mapping process of color attributes is consistent with the mapping process in the above-mentioned feature threshold method, and will not be repeated here. Please refer to Figure 9 ,like Figure 9As shown, the color recognition model provided by the embodiments of the present application can output three color attribute categories, including dark color, light color, and non-icon. In a preferred embodiment, the initial color matrix can be limited to a matrix with a size of 32*32*3. The task of the color recognition model is to identify the color depth (including dark color, light color, and non-icon) of the foreground area corresponding to each sound icon. Through two convolution layers, the flattened icon edge features with a dimension of 1568 can be extracted. Then, through a fully connected layer and an activation layer, an output vector with a length of 3 can be obtained. The sum of the three values in the output vector is equal to 1. Then, the maximum value in the output vector can be taken as the confidence, and the index value of the maximum value can be taken as the category number. Thus, the color attribute of the sound icon can be obtained. The category number "0" represents dark color, the category number "1" represents light color, and the category number "2" represents non-icon. When an icon is identified as a non-icon, it is filtered out again. It can be understood that increasing the identification of non- icons can further filter the detected false icons in actual applications, thereby obtaining more accurate identification results. It should be noted that when using the color recognition model, the model can be converted into an ONNX format (Open Neural Network Exchange, open neural network exchange format). Then, the DNN module (Deep Neural Network module, OpenCV module specially used to realize the functions related to deep neural networks) of OpenCV is used to load the color recognition model in ONNX format and perform prediction. When OpenCV loads the model, the DNN module is used to rewrite the model, so that the running efficiency of the model is higher. The ONNX format is an open file format designed for machine learning, which can be used to store trained models. It enables different artificial intelligence frameworks (such as Pytorch and MXNet) to store model data in the same format and interact with each other. OpenCV is a cross-platform computer vision and machine learning software library based on the BSD license (open source), which can run on Linux, Windows, Android, and Mac OS operating systems. It should be noted that before this step is performed, the color recognition model can be obtained by performing step S209.

[0117] As known from the above, the image recognition method based on data pre-labeling can obtain the color recognition model end-to-end, and the feature threshold method is used for automatic data pre-labeling, which saves the labeling cost and improves the labeling efficiency.

[0118] It should be noted that the above three distance recognition methods based on icon foreground color recognition can be configured by the R&D personnel as needed. The feature threshold method, the mask perspective method and the image recognition method can be selected to obtain the color information of the sound icon, so as to obtain the estimated object distance. Preferably, the image classification method can obtain more accurate classification effect. Compared with the color recognition method based on pixel mean value, the feature threshold method provided in the embodiment of the application can extract the sound icon foreground region and use the purity mean value for color judgment. Compared with using the color of the entire sound icon, the feature threshold method can more accurately identify the real icon color. The mask perspective method can obtain the foreground pixels of the sound icon according to the pre-set foreground region, so that the icon color can be obtained faster and more accurately. The image recognition method can accurately identify the icon color, so as to realize accurate distance recognition. The color recognition model is very small, only 72K, the inference speed is very fast, the inference time is less than 1ms, the storage and calculation resources are very small, and it is relatively easy to deploy and use on the related client.

[0119] Finally, the estimated object distance and the icon angle obtained in the above step S203 can be used as the object position of the second virtual object.

[0120] In step S205, an object behavior prompt voice corresponding to the sound icon is generated according to the object behavior category and the object position, and the object behavior prompt voice is played.

[0121] Specifically, the computer device can generate an object behavior prompt voice corresponding to the sound icon according to the icon category corresponding to the sound icon, the icon angle and the estimated object distance. Optionally, the object behavior prompt text can be generated according to the icon category, the icon angle and the estimated object distance, and then the object behavior prompt text is converted into the object behavior prompt voice by the voice synthesis model. Optionally, if the number of sound icons is multiple, the object behavior prompt voice corresponding to each sound icon can be generated first, and then the voice to be reminded is selected from the multiple object behavior prompt voices according to the designed prompting scheduling rule, or the object behavior prompt text corresponding to each sound icon can be generated first, and then the text to be reminded is selected from the multiple object behavior prompt texts according to the designed prompting scheduling rule, and the object behavior prompt voice is converted from the text to be reminded.

[0122] In addition, since the icon coordinates are relative position coordinates of the sound icon on the target map, the specific position information of the second virtual object in the virtual game world cannot be directly represented, and therefore the icon coordinates can be converted into object coordinates of the second virtual object in the virtual game world through coordinate transformation, and then the object coordinates, the icon angle, and the estimated object distance can be determined as the object position of the second virtual object, and a larger amount of information of the object behavior prompt voice can be generated based on the icon category, the icon angle, the object coordinates, and the estimated object distance.

[0123] Further, according to the set reminding interval length and the reminding priority, the target object behavior prompt voice can be obtained from one or more object behavior prompt voices, and then the target object behavior prompt voice is played, and the specific process can be referred to the above Figure 3 The step S103 in the corresponding embodiment.

[0124] For the steps S201-S205, the embodiments of the present application can further include preprocessing steps S206-S209, which are specifically described as follows:

[0125] In step S206, the synthesized annotation data is obtained, the synthesized annotation data is input into the basic detection model for training to obtain a target basic detection model, the target basic detection model is used to perform icon annotation on the map sample data, and the annotated map sample data is input into the initial icon detection model for training to obtain an icon detection model.

[0126] Specifically, please refer to Figure 10 is a flowchart of a data synthesis process provided by the embodiments of the present application. As shown in Figure 10As shown, the computer device can obtain preset fixed parameters, including an initial icon angle, an initial distance (referring to the distance between the icon template and the center point of the map template), and a zooming coefficient (which can also be referred to as a zooming ratio parameter). In an embodiment, the initial icon angle can be limited to a range of 0°-359°, the initial distance can be limited to a range of 96-105 pixels, and the zooming coefficient can be limited to a range of 0.8-1.25. The initial icon angle, the initial distance, and the zooming coefficient can be randomly generated within the corresponding range. According to the initial icon angle, the icon template can be rotated to obtain an initial rotated icon. First, a square with a side length equal to the diagonal length of the icon template is generated. Then, the coordinates of each pixel of the icon template after rotation are calculated. The x coordinate and the y coordinate of the pixel after rotation can be calculated by using the sum algorithm of the sine angle sin(a+b)=sin(a)*cos(b)+cos(a)*sin(b) and the sum algorithm of the cosine angle cos(a+b)=cos(a)*cos(b)–sin(a)*sin(b), as the rotation coordinates, and the rotation coordinates are filled as the corresponding coordinate color of the icon template, so as to obtain the initial rotated icon. Further, the pixel coordinates of the initial rotated icon in the map template can be calculated according to the initial icon angle and the initial distance. The pixel coordinates can be obtained by using the invariant point coordinates and the relative position calculation, and combining the sum algorithm of the sine angle and the sum algorithm of the cosine angle. In addition, the size of the icon template can be adjusted according to the zooming coefficient. Further, the initial rotated icon can be superimposed on the area indicated by the pixel coordinates in the map template, replacing the original picture content in the area, to obtain the synthesized annotation data. For details, see Figure 11 FIG. 1 is a schematic diagram of a scene for data synthesis, provided by an embodiment of the present application. As shown in Figure 11 FIG. 1, using a map template without any sound icon and an icon template representing gun sound, after the above data synthesis process, the synthesized annotation data superimposed with the gun sound icon can be finally obtained. It can be understood that various synthesized annotation data can be generated by using the above data synthesis method.

[0127] Further, the computer device also needs to calculate the coordinates of each pixel in the effective area of the synthesized annotation data. The effective area refers to the area obtained by filtering the more transparent areas formed by the icon rotation in the synthesized annotation data. Further, the obtained coordinates and the storage path of the synthesized annotation data can be saved to a target file. The target file can be an XML (Extensible Markup Language) file. Using the XML format to store data can facilitate data interaction. The XML file can include many fields. Therefore, the coordinates need to be written into the target field (such as the objects field) related to the coordinates. The coordinate information can be conveniently obtained by subsequent field query.

[0128] Furthermore, in order to obtain a better target basic detection model for icon pre-labeling, the map sample data can also be manually labeled to obtain manually labeled data. Since there are slight differences between manually labeled data and synthetically labeled data, the manually labeled data and the synthetically labeled data can be input into the basic detection model for training, thereby obtaining a target basic detection model. For example, 4000 synthetically labeled data can be synthesized, and then 200 manually labeled data can be obtained through manual labeling. The 200 manually labeled data are rotated and enhanced by 90°, 180°, and 270° respectively to obtain 800 manually labeled data. The above 4000 synthetically labeled data and 800 manually labeled data are input into the basic detection model as input data for training. Under low threshold conditions, the target basic detection model has a recall rate of 100% for the detection box in the input data and a precision rate of up to 98%.

[0129] Furthermore, you can use the trained target basic detection model to batch detect real map sample data and write the annotation results into the annotation file. Please refer to Figure 12 , is a flow chart of a data labeling process provided by an embodiment of the present application. Figure 12 As shown, the specific process is: the computer device obtains multiple game videos, and parses the game videos to obtain multiple picture frames, and then the multiple picture frames can be cropped to obtain map sample data. The map sample data can then be input into the target basic detection model, and the sound icons in the map sample data are detected by the target basic detection model. Then, through fixed position filtering and fixed position correction, the annotated map sample data can be obtained, wherein fixed position filtering refers to filtering the annotation box generated by the target basic detection model according to the corresponding fixed detection box. The specific process can be seen in the above step S202. Fixed position correction refers to replacing the coordinates of the annotation box with the coordinates of the fixed detection box closest to the annotation box. The annotated map sample data can then be cleaned to obtain basic map sample data. Among them, data cleaning refers to cropping and storing the detected sound icons, and then manually scanning and correcting them in batches in the category folder, and finally reconstructing the annotated data according to the original image name, category, and coordinates. The specific process can be:

[0130] a) First, the sound icons in the annotated map sample data are cropped and stored: all the annotation boxes in each annotated map sample data are cropped to obtain the target icon, and the target icon is stored separately according to the category folder named after the category. When storing, the coordinate information of the annotation box is written in the file name. For example, it can be stored as originpath_xmin_xmax_ymin_ymax.jpg, where originpath represents the file name of the original annotated map sample data (referred to as the original large image), and xmin, xmax, ymin, and ymax represent the minimum x coordinate, maximum x coordinate, minimum y coordinate, and maximum y coordinate of the annotation box of the target icon, respectively.

[0131] b) Perform manual correction: Manually scan the images in each category folder, delete any non-icon images found, and move any images found to the correct category folder.

[0132] c) Reconstructing the annotation data: For the corrected target icon, find the original image, category, and coordinates, and write them back to the annotation file corresponding to the original image. The annotation file can be in XML format.

[0133] Furthermore, the basic map sample data can be rotated and enhanced to obtain the target map sample data. For example, each basic map sample data can be rotated and enhanced by 90°, 180°, and 270°, respectively, that is, 4 images can be obtained for each basic map sample data. The target map sample data can then be input into the initial icon detection model, and the initial icon detection model outputs the predicted icon category and predicted icon coordinates corresponding to the target map sample data. A category loss function is generated based on the predicted icon category, and a position loss function is generated based on the predicted icon coordinates. A target loss function can be generated based on the category loss function and the position loss function. The model parameters in the initial icon detection model are then corrected using the target loss function to obtain an icon detection model. Subsequently, the icon detection model can be used to perform icon recognition on the input target map, thereby obtaining the icon category and icon coordinates corresponding to the sound icon in the target map. Under optimal conditions, the recall rate and precision rate of the icon detection model can both reach 100%, which is very effective and can accurately detect sound icons on the target map.

[0134] From the above, it can be seen that in the existing scheme, an end-to-end target detection model based on manually labeled data can be used for icon recognition. However, the acquisition of sample materials for this method is very time-consuming, and the workload of manual screening and labeling is large. There will be errors in manually labeled icons, and there will be some false detections. In the absence of fixed position filtering, there will be many false detections and missed detections above and below the classification threshold. The embodiment of the present application uses synthetic labeled data to train the target basic detection model, and then uses the target basic detection model to label the data. Then, data cleaning is performed to obtain real labeled data for training to obtain an icon detection model. This method of automatic data labeling can obtain accurate labeled data with less human participation, thereby improving the efficiency and quality of labeling. The fixed position filtering method is then used to filter out the labeled boxes with large errors in the position, and very accurate detection results can be obtained in actual applications.

[0135] It should be noted that the icon detection model is based on a deep neural network. Specifically, the lightweight detection network YoloV4-Tiny can be used to build a lightweight backbone network EfficientNet-lite structure, followed by two branches to obtain image feature matrices of different scales. Among them, YoloV4-Tiny is a simplified version of YoloV4 (a real-time, high-precision target detection model). Compared with YoloV4, it has fewer structures but faster calculation speed. EfficientNet-lite is the end-side version of EfficientNet (a fast and high-precision model that uses depth, width, and input image resolution joint adjustment technology) launched by Google. It provides 5 different versions (EfficientNet-lite0~4), allowing users to choose between latency, number of parameters, and accuracy according to actual application scenarios and resource conditions. In the icon detection model provided in the embodiment of the present application, the feature extraction network is formed by stacking multiple CBL modules and Res modules, wherein the CBL module may include one or more CBL units connected in sequence, and the Res module may include one or more CBL units with residual connections. Each CBL unit includes a convolution layer, a normalization layer, and an activation layer connected in sequence. The convolution layer performs a convolution operation (Convolution, Conv) on the input data to extract features. The normalization layer performs a batch normalization operation (BatchNormalization) on the input data. The batch normalization operation helps solve the gradient vanishing and gradient exploding problems in the back propagation process, reduces the sensitivity to some hyperparameters (such as the learning rate, the size range of network parameters, and the choice of activation function), and when each batch is normalized separately, it plays a certain regularization effect, thereby achieving better convergence speed and convergence effect. The activation layer uses the leaky rectified linear unit (Leaky ReLU) as the activation function. The Leaky ReLU function is a variant of the rectified linear unit (ReLU) function. The output of this function has a very small slope for negative inputs, which solves the problem that neurons do not learn after the ReLU function enters the negative interval.

[0136] It can be understood that the icon detection model and the target basic detection model can be integrated into a model, which can automatically label icons on the input target map and perform icon detection and recognition.

[0137] Step S207: marking the map sample data with a detection frame, and performing overlay visualization processing on the marked map sample data to generate a fixed detection frame;

[0138] Specifically, the computer device can obtain map sample data from multiple game videos and perform detection frame annotation on the sample sound icons in the map sample data. For example, Lableme software (an image annotation software with a graphical interface that can annotate images in the form of polygons, rectangles, circles, polylines, line segments, and points, and is commonly used for target detection, image segmentation, and other tasks) can be used to perform rectangular annotation on the sample sound icons on the map sample data, and then the annotated map sample data with the same icon category can be overlaid and visualized to obtain a detection frame sample data set corresponding to each icon category. In the embodiment of the present application, the overlaid visualization process is also called a data visualization analysis method. Please refer to Figure 13a-13b , is a scenario diagram of a data visualization analysis method provided by an embodiment of the present application. Figure 13a As shown, all sample sound icons contained in the map sample data are marked with rectangular detection frames, and then the marked map sample data are classified according to the categories of the sample sound icons they contain. Then, the map sample data with the same category are superimposed together to obtain the detection frame sample data set corresponding to each category of sample sound icons. For easy observation, only the detection frame can be displayed, as shown in Figure 13a As shown in the figure, for a certain game product, after superimposed visualization processing, we can obtain the detection frame sample data set A corresponding to the car sound icon, the detection frame sample data set B corresponding to the footstep sound icon, the detection frame sample data set C corresponding to the gun sound icon, and the detection frame sample data set D corresponding to the silenced gun sound icon. By observing these four detection frame sample data sets, we can find that the position of the detection frame corresponding to each sample sound icon in the map sample data is regular, that is, the display position of each sample sound icon is fixed at 12 positions, and the interval angle between adjacent positions may also be the same.

[0139] Furthermore, a clustering algorithm can be used to obtain the precise position of the sample sound icon. The specific process is: first, the detection frame sample data set is rotationally enhanced to obtain a target detection frame sample data set. For example, 90°, 180°, and 270° rotation enhancements can be used respectively so that the annotation boxes in the detection frame sample data set can cover all 12 angles. Assuming that the target detection frame sample data set corresponding to icon category Y includes at least two detection frames, the computer device can cluster at least two detection frames in the target detection frame sample data set corresponding to icon category Y through a clustering algorithm to obtain N clusters, and then generate N fixed detection frames corresponding to icon category Y based on the center positions of the N clusters, where N is an integer greater than 1. Figure 13aIn the corresponding embodiment, N is equal to 12. Among them, the clustering algorithm is a typical representative of unsupervised learning in machine learning, clustering is essentially a set partition problem, and data is divided into a specified number of categories according to internal similarity without artificial definition of category standards. The algorithm has been applied in many practical problems such as data analysis and pattern recognition.

[0140] The clustering algorithm process provided by the embodiment of the application is as follows:

[0141] a) Initialize the number of clustering categories N, which represents the final convergence of N positions. Randomly initialize N coordinate points on the map sample data as initial clustering centers.

[0142] b) Calculate the initial clustering center closest to the center point of each detection box to obtain the clustering center point to which the detection box belongs. When all detection boxes are calculated, the detection boxes owned by each clustering center point can be obtained.

[0143] c) Calculate the mean of the center points of all detection boxes owned by each clustering center point to obtain N new clustering center points.

[0144] d) Loop the operations of b) and c) until the positions of all clustering center points no longer change.

[0145] e) Take the final N clustering center points as the center points of the detection boxes, and take the coordinate mean of all detection boxes corresponding to each clustering center point as the coordinates of the N detection boxes obtained by clustering. According to the N groups of coordinates, N fixed detection boxes can be generated.

[0146] In Figure 13a-13b In the corresponding embodiment, the number of categories N is set to 12, as shown in Figure 13b After clustering the detection box sample data set A, the detection box sample data set B, the detection box sample data set C, and the detection box sample data set D by the above clustering algorithm, 12 fixed detection boxes corresponding to the car sound icon, the footstep sound icon, the gun sound icon, and the sound-eliminated gun sound icon can be obtained.

[0147] In step S208, a mask picture and a mask identifier are generated based on the icon template, the pixel coordinates of the mask picture in the target map are obtained, the association between the mask picture and the mask identifier is stored in the mask configuration file, and the pixel coordinates of the mask picture in the target map are stored in the mask configuration file.

[0148] Specifically, the computer device can rotate the at least two icon templates to obtain at least two rotated icon sequences, each of which can include N rotated icons corresponding to each icon template, and the at least two icon templates have different icon categories. Further, the at least two rotated icon sequences can be subjected to effective area cropping to obtain at least two cropped rotated icon sequences, where the effective area cropping refers to removing the transparent area at the boundary of the rotated icon caused by the rotation of the icon template. Further, the at least two cropped rotated icon sequences can be subjected to binarization processing to obtain at least two target rotated icon sequences, and then the at least two target rotated icon sequences are subjected to erosion processing and dilation processing. In the at least two target rotated icons, the area obtained after erosion can be marked as a first chroma (such as white), and the area obtained after dilation can be marked as a second chroma (such as black), thereby obtaining at least two marked rotated icon sequences containing the first chroma and the second chroma. Further, in the at least two marked rotated icon sequences, the area that is not marked as the first chroma and not marked as the second chroma can be marked as a third chroma (such as gray), and finally a mask picture corresponding to each rotated icon can be obtained. Further, the computer device needs to generate a mask identifier corresponding to each mask picture according to the icon category and the rotation angle of each rotated icon, and obtain the pixel coordinates of each mask picture in the corresponding target map, and then the association between each mask picture and the mask identifier, and the pixel coordinates of each mask picture in the target map can be stored in a mask configuration file. Subsequently, the mask identifier can be matched and searched in the mask configuration file to obtain the mask picture corresponding to the sound icon, and the foreground pixel coordinates can be obtained by analyzing the mask picture. The above pixel coordinates can include the foreground pixel coordinates, the background pixel coordinates and the pixel coordinates of the foreground and background transition region corresponding to the mask picture. Each mask picture and the angle of one sound icon correspond to each other. The rotation angle, the pixel coordinates of the mask picture in the target map and the mask picture can be configured in a project, and the rotation angle as an index can find the storage path and the pixel coordinates of the mask picture.

[0149] Please see Figure 14 is a flowchart of a mask picture generation process provided by an embodiment of the present application. As shown in Figure 14As shown, taking the car sound icon template as an example, first rotate the car sound icon template according to the rotation angle to obtain a rotated car sound icon, then calculate the effective area of ​​the rotated car sound icon, remove the area with more transparent colors at the boundary caused by the icon rotation, and obtain the cropped rotated car sound icon. Then, through the transparency channel, the effective area is converted into a target rotated car sound icon with white opaque areas and black transparent areas. Further, the target rotated car sound icon is morphologically eroded, and the area obtained after the erosion is marked as white, corresponding to a value of 255; at the same time, the target rotated car sound icon can be morphologically expanded, and the area obtained after the expansion is marked as black, corresponding to a value of 0. Finally, the unmarked middle area (i.e., the area at the junction of the foreground and background that may be controversial and cannot be used as the foreground or background) is marked as gray, corresponding to a value of 128, thereby obtaining a car sound icon mask image divided by the foreground and background areas. The pixel coordinates of the mask image in the target map can also be calculated for associated storage. The process of generating mask images corresponding to other types of sound icons is the same as that of generating the car sound icon mask image, and will not be repeated here.

[0150] Step S209 , obtaining icon sample data, color-labeling the icon sample data, and inputting the labeled icon sample data into an initial color recognition model for training to obtain a color recognition model.

[0151] Specifically, the computer device can extract icon sample data from the map sample data, color-label the icon sample data, and then input the labeled icon sample data into an initial color recognition model. The initial color recognition model then outputs predicted color attributes corresponding to the icon sample data. Furthermore, a target loss function can be generated based on the predicted color attributes. The target loss function can be used to modify model parameters in the initial color recognition model to obtain a color recognition model that can be subsequently used to identify the color attributes of sound icons.

[0152] Please see again Figure 9 ,like Figure 9 As shown, the color recognition model needs to identify the color depth of the foreground area corresponding to each sound icon, which is divided into dark, light and non-icon. Since the foreground pixels of the sound icon need to be extracted, the shape of the icon will have a greater impact on the recognition result. Therefore, the embodiment of the present application selects a convolutional neural network to extract the edge features of the icon; and since the color depth has a greater impact on the recognition result, and color belongs to a shallow feature, a shallow network can be used. Therefore, the embodiment of the present application builds a shallow (2-layer) convolutional neural network as the backbone network, including a 3*3 convolutional layer and a 5*5 convolutional layer, followed by a fully connected layer, and uses logsoftmax as the final activation function for standardization, and finally the color attributes of the sound icon can be output.

[0153] Wherein, 3 types of data need to be labeled when color labeling icon sample data, the specific process can be:

[0154] a) Dark and light icon labeling: Icon detection model is used to obtain icon sample data from a large amount of map sample data, and color recognition is performed using feature threshold method to preliminarily determine the color depth of the icon, and according to the color depth, it is stored in different folders, dark color is marked as 0, light color is marked as 1, and then artificial correction is performed, that is, the labeled icon sample data used for model training can be obtained.

[0155] b) Randomly intercept non-icon areas in map sample data to obtain a batch of non-icon sample data, and mark them as 2 to obtain labeled icon sample data used for model training.

[0156] In addition, during the model training process, the output result of the logsoftmax activation function can be input into the CrossEntropyLoss function (cross entropy loss function, mainly used to determine the closeness of the actual output and the expected output) as the calculation method of the target loss function, and the Adam optimization method (adaptive momentestimation, adaptive moment estimation) can be used for model training. The Adam optimization method can design independent adaptive learning rates for different model parameters by calculating the first and second moment estimates of the gradient.

[0157] The embodiment of the present application can adopt a method based on AI target detection and image recognition to accurately detect and identify the sound icon on the target map to determine whether there is an enemy around the first virtual object, and use voice to remind the behavior category, distance, and angle. That is to say, the sound icon that was originally only visual can be converted into a corresponding object behavior prompt voice. By playing the object behavior prompt voice, the player (i.e., the target user) can directly obtain the nearby enemy behavior information, i.e., the object behavior category and object position of the second virtual object. Therefore, the player does not have to pay attention to the target map at all times, nor does he have to concentrate on listening to the enemy sound in the game. Compared with the solution without voice reminders, it can reduce the problem of low enemy situation judgment accuracy and low efficiency caused by the small display size of the game minimap, thereby improving the accuracy and efficiency of obtaining enemy behavior information during the game, and enriching the game's expression form and enhancing the game experience. In addition, the embodiment of the present application adopts a data visualization analysis method to generate a fixed detection frame, and the fixed detection frame can be used to perform fixed position filtering and fixed position correction on the detected icon to be detected, thereby obtaining a more accurate detection result. At the same time, it is also possible to use synthetic labeled data for training to obtain a target-based detection model, and use the low-threshold recall method of the target-based detection model for automated data labeling, so that a large amount of completely real labeled data can be obtained. Compared with manual collection and labeling, it can greatly save the cost of manual labeling and improve the quality and efficiency of labeling. Through the generated mask image, the foreground shape and position of the sound icon can be determined in advance. Even when the color of the sound icon is lighter, the foreground area can be better identified and the foreground color can be extracted. In addition, before training the color recognition model, the feature threshold method is used for color labeling, which saves labeling costs and can accurately identify the color attributes of the sound icon, thereby achieving accurate distance recognition. Moreover, the color recognition model is very small, has a fast inference speed, occupies very little storage and computing resources, and is easier to deploy and use on relevant clients.

[0158] See Figure 15 , is a flowchart of a game data processing method provided by an embodiment of the present application. The embodiment of the present application adopts a method for identifying enemy information based on AI target detection and an enemy reminder scheme based on voice reminders, which can remind the game application client of enemy information (including object behavior categories and object positions) in the game through voice. In the embodiment of the present application, it is necessary to identify the content of the game screen (including the target map) in real time to determine whether there is an enemy near the player. The identified targets include people walking, shooting, and driving. In addition, it is necessary to identify the distance and angle of the location where the enemy occurs relative to the player. Figure 15As shown, in the technical application scheme of the enemy alert application, the process of how the scheme provided by the embodiment of the application identifies the enemy information (i.e., the object behavior category and object position of the second virtual object) through the game screen is explained from the technical application level, including the detection and type identification of the sound icon on the target map, and the use of fixed position post-processing technology (i.e., fixed position filtering) to optimize the detection effect, the extraction of the color of the sound icon to obtain the distance of the enemy (i.e., the estimated object distance between the first virtual object and the second virtual object), the acquisition of the angle of the enemy (i.e., the icon angle) through the icon coordinates, and the acquisition of the reminder content of each event through the reminder interval time setting and the reminder priority setting, and the surrounding enemy reminder to the player through voice in the game application client.

[0159] The four modules of the lower layer respectively represent a data analysis method for obtaining icon fixed position information (i.e., coordinate information of the fixed detection frame), an automatic data labeling and icon detection model training process for icon detection, a method for calculating distance nearness through color and identifying icon angle through icon coordinates, and a reminder scheduling method based on reminder interval time setting and reminder priority setting. Among them, the data analysis method uses the rectangular labeling frame superposition method to visualize about 200 manually labeled map sample pictures to obtain the conclusion that the icon position is fixed and the interval angle is 30 degrees, and the clustering algorithm based on the rectangular labeling frame (also called detection frame) can obtain accurate clustering position to generate a fixed detection frame. The automatic data labeling uses synthesized labeling data to train a basic detection model to obtain a target basic detection model, and uses the target basic detection model to label real data (i.e., map sample data), and the cleaned data is used to train an initial icon detection model to obtain an icon detection model with very high accuracy. The scheme for estimating object distance through color can use feature threshold method, mask perspective method and image recognition method to perform color recognition on the sound icon respectively, and map the color value (i.e., color attribute) to distance value. The reminder scheduling method uses the configured global reminder interval (i.e., the reminder interval time length between different categories of sound icons), the same reminder interval (i.e., the reminder interval time length between the same category of sound icons) and the reminder priority order to determine whether the enemy information needs to be reminded each time and obtain the content that needs to be reminded this time.

[0160] The specific process of all the methods mentioned here can be referred to the above Figure 3 and Figure 6 corresponding embodiments, which will not be repeated here.

[0161] It can be known from the above that the embodiment of the application can adopt an AI target detection and image recognition based method to accurately detect and identify a sound icon on a target map to determine whether there is an enemy situation around the first virtual object, and use voice to remind the behavior category, distance, and angle, that is, the originally visualized sound icon can be converted into a corresponding object behavior prompt voice, and the object behavior prompt voice can be played to enable a player (i.e., a target user) to directly obtain the enemy behavior information around, i.e., the object behavior category and object position of the second virtual object, so that the player does not have to pay attention to the target map at all times or concentrate on listening to the enemy sound in the game. Compared with a scheme without voice prompting, the problem of low accuracy and low efficiency in enemy situation judgment caused by the small display size of the game mini-map can be reduced, and the accuracy and efficiency of obtaining enemy behavior information in the game process can be improved, and the performance form of the game can be enriched, and the game experience can be improved.

[0162] Please refer to Figure 16 FIG. 1 is a structural schematic diagram of a game data processing device provided by an embodiment of the application. The game data processing device can be a computer program (including program code) running on a computer device, for example, the game data processing device is an application software; the device can be used to execute the corresponding steps in the game data processing method provided by the embodiment of the application. As shown in FIG. 1, the game data processing device 1 can include a display module 11, a voice acquisition module 12, and a voice playing module 13. Figure 16

[0163] The display module 11 is configured to display a target map in a game picture used to control a first virtual object; the first virtual object belongs to a first camp; the target map is used to display a sound icon representing an object behavior category and an object position of a second virtual object; the second virtual object belongs to a second camp;

[0164] The display module 11 is specifically configured to display the game picture used to control the first virtual object in response to a start operation of a game application; display the target map centered on the first virtual object in the game picture, acquire the object behavior category and the object position of the second virtual object in a region range contained by the target map, generate the sound icon according to the object behavior category and the object position, and display the sound icon in the target map;

[0165] The voice acquisition module 12 is configured to acquire an object behavior prompt voice corresponding to the sound icon when the target map has the sound icon; the object behavior prompt voice is used to voice prompt the object behavior category and the object position of the second virtual object;

[0166] The voice playing module 13 is configured to play the object behavior prompt voice.

[0167] ​The specific function implementation of the display module 11 can be seen from the above Figure 3 The specific function implementation of the voice acquisition module 12 in step S101 of the corresponding embodiment can be seen from the above Figure 3 The specific function implementation of the voice acquisition module 12 in step S102 of the corresponding embodiment can be seen from the above Figure 6 The specific function implementation of the voice playing module 13 in steps S201-S205 of the corresponding embodiment can be seen from the above Figure 3 The specific function implementation of the voice playing module 13 in step S103 of the corresponding embodiment will not be repeated here.

[0168] Please see Figure 16 The game data processing apparatus 1 can further include a mask generation module 14.

[0169] The mask generation module 14 is configured to rotate at least two icon templates to obtain at least two rotated icon sequences; each of the at least two rotated icon sequences includes N rotated icons corresponding to each icon template; the at least two icon templates have different icon categories; perform effective area cropping on the at least two rotated icon sequences to obtain at least two cropped rotated icon sequences; perform binarization processing on the at least two cropped rotated icon sequences to obtain at least two target rotated icon sequences; perform erosion processing and inflation processing on the at least two target rotated icon sequences, mark the area obtained after erosion as a first chroma in the at least two target rotated icons, mark the area obtained after inflation as a second chroma, and obtain at least two marked rotated icon sequences including the first chroma and the second chroma; mark the area that is not marked as the first chroma and not marked as the second chroma as a third chroma in the at least two marked rotated icon sequences to obtain a mask picture corresponding to each rotated icon; generate a mask identifier corresponding to each mask picture according to the icon category and the rotation angle of each rotated icon, store the pixel coordinates of each mask picture in a target map, the association relationship between each mask picture and the mask identifier, and the pixel coordinates of each mask picture in the target map to a mask configuration file.

[0170] The specific function implementation of the mask generation module 14 can be seen from the above Figure 6 The specific function implementation of the mask generation module 14 in step S208 of the corresponding embodiment will not be repeated here.

[0171] Please see Figure 16 The game data processing apparatus 1 can further include a detection frame generation module 15.

[0172] The detection frame generation module 15 is configured to perform detection frame labeling on the sample sound icons in the map sample data, perform superimposed visual processing on the map sample data with the same icon category and the labeled detection frame, obtain a detection frame sample data set corresponding to each icon category, perform rotation enhancement on the detection frame sample data set, and obtain a target detection frame sample data set. The target detection frame sample data set corresponding to the icon category Y includes at least two detection frames. The at least two detection frames in the target detection frame sample data set corresponding to the icon category Y are clustered to obtain N clustering clusters, and N fixed detection frames corresponding to the icon category Y are generated according to the center positions of the N clustering clusters.

[0173] The specific function implementation of the detection frame generation module 15 can be referred to the above Figure 6 The step S207 in the corresponding embodiment, which will not be repeated here.

[0174] Please refer to Figure 16 The voice acquisition module 12 can include an icon identification unit 121, a filtering unit 122, a category determination unit 123, a position determination unit 124, and a voice generation unit 125.

[0175] The icon identification unit 121 is configured to perform icon identification on a target map to obtain an icon category and icon coordinates corresponding to one or more to-be-detected icons.

[0176] The icon identification unit 121 is specifically configured to input the target map into an icon detection model, perform feature extraction through a feature extraction network in the icon detection model to obtain a picture feature matrix, and perform non-maximum suppression processing on the picture feature matrix to obtain an icon category and icon coordinates corresponding to one or more to-be-detected icons.

[0177] The filtering unit 122 is configured to perform fixed position filtering on one or more to-be-detected icons according to an icon category and icon coordinates corresponding to the one or more to-be-detected icons.

[0178] The one or more to-be-detected icons include a to-be-detected icon X.

[0179] The filtering unit 122 is specifically configured to obtain, based on the icon category corresponding to the icon to be detected X, N fixed detection frames that match the icon category corresponding to the icon to be detected X; the N fixed detection frames are respectively located at fixed positions in the target map; N is an integer greater than 1; and is specifically configured to obtain, based on the icon coordinates corresponding to the icon to be detected X, the distances between the icon to be detected X and the N fixed detection frames, and obtain the closest distance from the N distances; if the closest distance is less than a distance threshold, retaining the icon to be detected X; if the closest distance is greater than the distance threshold, filtering out the icon to be detected X; when one or more icons to be detected have completed fixed position filtering, determining the retained icon to be detected as a sound icon;

[0180] a category determining unit 123 for determining, when a sound icon is obtained after performing fixed position filtering on one or more icons to be detected, the icon category corresponding to the sound icon as the object behavior category of the second virtual object;

[0181] a position determining unit 124 for obtaining an icon angle of the sound icon in the target map according to the icon category and icon coordinates corresponding to the sound icon, and determining the icon angle as the object position of the second virtual object;

[0182] The position determination unit 124 is specifically configured to determine, among the N fixed detection frames that match the icon category corresponding to the sound icon, the fixed detection frame closest to the sound icon as the target detection frame based on the icon coordinates corresponding to the sound icon; obtain a default azimuth angle corresponding to the target detection frame, and determine the default azimuth angle as the icon angle of the sound icon in the target map;

[0183] The voice generating unit 125 is configured to generate an object behavior prompt voice corresponding to the sound icon according to the object behavior category and the object position.

[0184] The specific function implementation of the icon recognition unit 121 can be found in the above Figure 6 The specific functional implementation of step S201, the filtering unit 122, and the category determination unit 123 in the corresponding embodiment can be referred to above. Figure 6 The specific functional implementation of the position determination unit 124 in step S202 of the corresponding embodiment can be found in the above Figure 6 The specific function implementation of the speech generation unit 125 in step S203 of the corresponding embodiment can be referred to above. Figure 6 Step S205 in the corresponding embodiment will not be described in detail here.

[0185] In one embodiment, the number of the above-mentioned sound icons is one or more, and one sound icon corresponds to one object behavior prompt voice;

[0186] Please also see Figure 16 The voice playing module 13 can include a voice selecting unit 131 and a voice playing unit 132.

[0187] The voice selecting unit 131 is configured to obtain a reminding interval duration and a reminding priority, extract an object behavior prompt voice meeting the reminding interval duration from one or more object behavior prompt voices as a to-be-played object behavior prompt voice, and determine a to-be-played object behavior prompt voice with the highest reminding priority in the to-be-played object behavior prompt voice as a target object behavior prompt voice according to the reminding priority.

[0188] The voice playing unit 132 plays the target object behavior prompt voice.

[0189] The specific function implementation of the voice selecting unit 131 and the voice playing unit 132 can refer to the step S103 in the above-mentioned Figure 3 embodiment, which will not be repeated here.

[0190] In an embodiment, the object position further includes an estimated object distance between the first virtual object and the second virtual object.

[0191] Please also refer to Figure 16 The voice generating unit 125 can include a distance determining sub-unit 1251 and a voice generating sub-unit 1252.

[0192] The distance determining sub-unit 1251 is configured to obtain a color attribute of the sound icon, and determine the estimated object distance between the first virtual object and the second virtual object according to the color attribute.

[0193] The distance determining sub-unit 1251 is specifically configured to perform color space conversion on an initial color matrix of the sound icon to obtain a target color matrix, extract foreground pixels in a color filtering interval from the target color matrix, generate a purity mean value of the foreground pixels, determine the color attribute of the sound icon according to the purity mean value, and map the color attribute to the estimated object distance between the first virtual object and the second virtual object; or

[0194] The distance determining sub-unit 1251 is specifically configured to perform normalization processing on an initial color matrix of the sound icon to obtain a target color matrix, perform feature extraction on the target color matrix through a convolution layer in the color recognition model to obtain an icon edge feature corresponding to the sound icon, perform feature integration on the icon edge feature through a full connection layer in the color recognition model to obtain a color feature, output a color attribute corresponding to the color feature through an output layer in the color recognition model, and map the color attribute to the estimated object distance between the first virtual object and the second virtual object.

[0195] The distance determination subunit 1251 is specifically used to obtain a target mask identifier having a mapping relationship between an icon category and an icon angle corresponding to the sound icon; perform a matching search in a mask configuration file based on the target mask identifier, and determine a mask image in the mask configuration file that matches the target mask identifier as the target mask image; the mask configuration file is used to store the association relationship between the mask identifier and the mask image, as well as the pixel coordinates of the mask image in the target map; parse the target mask image to obtain foreground pixel coordinates; determine the foreground pixels corresponding to the sound icon based on the foreground pixel coordinates, and generate a purity mean of the foreground pixels; determine the color attribute of the sound icon based on the purity mean; and map the color attribute to an estimated object distance between the first virtual object and the second virtual object.

[0196] The voice generation subunit 1252 is used to generate the object behavior prompt voice corresponding to the sound icon according to the icon category, icon angle and estimated object distance corresponding to the sound icon.

[0197] The specific functional implementation of the distance determination subunit 1251 can be found in the above Figure 6 The specific function implementation of the speech generation subunit 1252 in the corresponding embodiment of step S204 can be found in the above Figure 6 Step S205 in the corresponding embodiment will not be described in detail here.

[0198] In an embodiment of the present application, a target map can be displayed on a game screen used to control a first virtual object. When a sound icon representing the object behavior category and object location of a second virtual object appears on the target map, an object behavior prompt voice corresponding to the sound icon can be further obtained and then played. As can be seen from the above, when a target user controls the first virtual object to play a game, when a second virtual object belonging to another faction appears near the first virtual object, a corresponding sound icon can be displayed on the target map in the current game screen to indicate the object behavior category and object location of the second virtual object. In an embodiment of the present application, the sound icon can be detected and recognized, thereby converting the sound icon, which was originally only visible, into a corresponding object behavior prompt voice. By playing the object behavior prompt voice, the target user can directly obtain nearby enemy behavior information, namely, the object behavior category and object location of the second virtual object. Therefore, the player does not need to pay attention to the target map at all times, nor does he need to concentrate on listening to enemy sounds in the game. Compared with a solution without voice prompts, the problem of low enemy situation judgment accuracy and low efficiency caused by the small display size of the game mini-map can be reduced, thereby improving the accuracy and efficiency of obtaining enemy behavior information during the game and enriching the game's presentation.

[0199] See Figure 17, is a schematic diagram of the structure of a computer device provided in an embodiment of the present application. Figure 17 As shown, the computer device 1000 may include: a processor 1001, a network interface 1004 and a memory 1005. In addition, the above-mentioned computer device 1000 may also include: a user interface 1003, and at least one communication bus 1002. The communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), a keyboard (Keyboard), and the user interface 1003 may optionally include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1004 may be a high-speed RAM memory, or a non-volatile memory (non-volatile memory), such as at least one disk memory. The memory 1005 may optionally be at least one storage device located away from the aforementioned processor 1001. As Figure 17 As shown, the memory 1005 as a computer-readable storage medium may include an operating system, a network communication module, a user interface module, and a device control application.

[0200] In such Figure 17 In the computer device 1000 shown, the network interface 1004 can provide network communication functions; the user interface 1003 is mainly used to provide an interface for user input; and the processor 1001 can be used to call the device control application stored in the memory 1005 to achieve:

[0201] Displaying a target map in a game screen for controlling a first virtual object; the first virtual object belongs to a first camp; the target map is used to display a sound icon representing an object behavior category and an object position of a second virtual object; the second virtual object belongs to a second camp;

[0202] When a sound icon exists on the target map, obtaining an object behavior prompt voice corresponding to the sound icon; the object behavior prompt voice is used to voice prompt the object behavior category and object position of the second virtual object;

[0203] Play the object's behavior prompt voice.

[0204] It should be understood that the computer device 1000 described in the embodiment of the present application can execute the above Figure 3 、 Figure 6 The description of the game data processing method in any corresponding embodiment will not be repeated here. In addition, the description of the beneficial effects of adopting the same method will not be repeated here either.

[0205] In addition, it should be pointed out that the embodiment of the present application further provides a computer-readable storage medium, and the computer-readable storage medium stores a computer program executed by the aforementioned game data processing device 1, and the aforementioned computer program includes program instructions. When the aforementioned processor executes the aforementioned program instructions, the aforementioned game data processing device 1 can be executed. Figure 3 、 Figure 6 The description of the game data processing method described above in any corresponding embodiment will not be repeated here. In addition, the description of the beneficial effects of using the same method will not be repeated here. For technical details not disclosed in the computer-readable storage medium embodiment involved in this application, please refer to the description of the method embodiment of this application.

[0206] The computer-readable storage medium may be the game data processing device provided in any of the aforementioned embodiments or the internal storage unit of the computer device, such as the hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the computer device. Furthermore, the computer-readable storage medium may also include both the internal storage unit of the computer device and an external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by the computer device. The computer-readable storage medium may also be used to temporarily store data that has been output or is about to be output.

[0207] In addition, it should be noted that the present application also provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs the above-mentioned Figure 3 、 Figure 6 The method provided by any corresponding embodiment.

[0208] The terms "first", "second", etc. in the description, claims, and drawings of the embodiments of the present application are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "comprising" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, apparatus, product, or device comprising a series of steps or units is not limited to the listed steps or modules, but may optionally include steps or modules not listed, or may optionally include other step units inherent to these processes, methods, apparatuses, products, or devices.

[0209] Those skilled in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been described in the above description in a general manner. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0210] The above disclosure is only the preferred embodiments of the present application, and of course cannot be used to limit the scope of the rights of the present application, so equivalent changes made in accordance with the claims of the present application are still within the scope of the present application.

Claims

1. A method for processing game data, characterized in that: include: displaying a target map in a game screen for controlling a first virtual object; The first virtual object belongs to a first camp; the target map is used to display a sound icon representing an object behavior category and an object position of a second virtual object; The second virtual object belongs to the second camp; When the sound icon exists in the target map, an object behavior prompt voice corresponding to the sound icon is obtained; the object behavior prompt voice is used to voice prompt the object behavior category and object position of the second virtual object; the object position includes the icon angle of the sound icon in the target map and the estimated object distance between the first virtual object and the second virtual object, the estimated object distance is obtained by color attribute mapping determined based on the purity mean of foreground pixels, and the foreground pixels are determined according to the foreground pixel coordinates obtained by parsing the target mask image; the target mask image refers to the mask image that matches the target mask identifier in the mask configuration file, and the mask configuration file is used to store the association relationship between the mask identifier and the mask image, as well as the pixel coordinates of the mask image in the target map; The target mask identifier refers to a mask identifier having a mapping relationship between the icon category and the icon angle corresponding to the sound icon; Play the object behavior prompt voice.

2. The method according to claim 1, characterized in that The displaying of the target map in the game screen for controlling the first virtual object includes: In response to a start operation for the game application, displaying a game screen for controlling the first virtual object; A target map centered on the first virtual object is displayed in the game screen, an object behavior category and an object position of a second virtual object within an area included in the target map are obtained, a sound icon is generated according to the object behavior category and the object position, and the sound icon is displayed in the target map.

3. The method according to claim 1, characterized in that When the sound icon exists on the target map, obtaining the object behavior prompt voice corresponding to the sound icon includes: Performing icon recognition on the target map to obtain icon categories and icon coordinates corresponding to one or more icons to be detected; Performing fixed position filtering on the one or more icons to be detected according to the icon categories and icon coordinates respectively corresponding to the one or more icons to be detected; When the sound icon is obtained after performing fixed position filtering on the one or more icons to be detected, determining the icon category corresponding to the sound icon as the object behavior category of the second virtual object; acquiring an icon angle of the sound icon in the target map according to the icon category and icon coordinates corresponding to the sound icon, and determining the icon angle as the object position of the second virtual object; The object behavior prompt voice corresponding to the sound icon is generated according to the object behavior category and the object position.

4. The method according to claim 3, characterized in that The performing icon recognition on the target map to obtain icon categories and icon coordinates corresponding to one or more icons to be detected includes: Inputting the target map into an icon detection model, performing feature extraction through a feature extraction network in the icon detection model, and obtaining an image feature matrix; the image feature matrix is ​​used to represent features of different scales corresponding to one or more icons to be detected identified by the icon detection model; Non-maximum suppression processing is performed on the image feature matrix to obtain icon categories and icon coordinates corresponding to the one or more icons to be detected.

5. The method according to claim 3, characterized in that The one or more icons to be detected include an icon to be detected X; The performing fixed position filtering on the one or more icons to be detected according to the icon categories and icon coordinates respectively corresponding to the one or more icons to be detected includes: According to the icon category corresponding to the icon to be detected X, N fixed detection frames matching the icon category corresponding to the icon to be detected X are obtained; the N fixed detection frames are respectively located at fixed positions in the target map; N is an integer greater than 1; According to the icon coordinates corresponding to the icon X to be detected, the distances between the icon X to be detected and the N fixed detection frames are obtained, and the closest distance is obtained from the N distances; If the closest distance is less than the distance threshold, retain the icon X to be detected; If the closest distance is greater than the distance threshold, filtering out the icon X to be detected; When the one or more icons to be detected have completed the fixed position filtering, the remaining icon to be detected is determined as the sound icon.

6. The method according to claim 5, characterized in that The acquiring, according to the icon category and icon coordinates corresponding to the sound icon, the icon angle of the sound icon in the target map includes: Among the N fixed detection frames that match the icon category corresponding to the sound icon, determine the fixed detection frame closest to the sound icon as the target detection frame according to the icon coordinates corresponding to the sound icon; A default azimuth angle corresponding to the target detection frame is obtained, and the default azimuth angle is determined as the icon angle of the sound icon in the target map.

7. The method according to claim 3, characterized in that Generating the object behavior prompt voice corresponding to the sound icon according to the object behavior category and the object position includes: acquiring a color attribute of the sound icon, and determining an estimated object distance between the first virtual object and the second virtual object according to the color attribute; Generate an object behavior prompt voice corresponding to the sound icon according to the icon category, icon angle and the estimated object distance corresponding to the sound icon.

8. The method according to claim 7, characterized in that The acquiring the color attribute of the sound icon and determining the estimated object distance between the first virtual object and the second virtual object according to the color attribute includes: performing a color space conversion on an initial color matrix of the sound icon to obtain a target color matrix, extracting foreground pixels within a color filter interval from the target color matrix, generating a purity mean of the foreground pixels, determining a color attribute of the sound icon based on the purity mean, and mapping the color attribute to an estimated object distance between the first virtual object and the second virtual object; or The initial color matrix of the sound icon is normalized to obtain a target color matrix, and features of the target color matrix are extracted through the convolutional layer in the color recognition model to obtain icon edge features corresponding to the sound icon, and features of the icon edge features are integrated through the fully connected layer in the color recognition model to obtain color features, and color attributes corresponding to the color features are output through the output layer in the color recognition model, and the color attributes are mapped to the estimated object distance between the first virtual object and the second virtual object.

9. The method according to claim 1, characterized in that Also includes: Rotating at least two icon templates to obtain at least two rotated icon sequences; the at least two rotated icon sequences respectively include N rotated icons corresponding to each icon template; The at least two icon templates have different icon categories; performing effective area cropping on the at least two rotating icon sequences to obtain at least two cropped rotating icon sequences; performing binarization processing on the at least two cropped rotating icon sequences to obtain at least two target rotating icon sequences; Performing erosion and dilation on the at least two target rotating icon sequences, marking, in the at least two target rotating icons, regions obtained after erosion as a first chromaticity and regions obtained after dilation as a second chromaticity, to obtain at least two marked rotating icon sequences including the first chromaticity and the second chromaticity; In at least two marked rotating icon sequences, marking areas that are not marked as the first chromaticity and not marked as the second chromaticity as a third chromaticity, to obtain a mask image corresponding to each rotating icon; Generate a mask identifier corresponding to each mask image according to the icon category and rotation angle corresponding to each rotating icon, obtain the pixel coordinates of each mask image in the target map, and store the association relationship between each mask image and the mask identifier, as well as the pixel coordinates of each mask image in the target map in the mask configuration file.

10. The method according to claim 1, characterized in that The number of the sound icons is one or more, and one sound icon corresponds to one object behavior prompt voice; and playing the object behavior prompt voice includes: Get reminder interval duration and reminder priority; extracting, from one or more object behavior prompt voices, an object behavior prompt voice that meets the reminder interval length as the object behavior prompt voice to be played, and determining, based on the reminder priority, the object behavior prompt voice to be played with the highest reminder priority among the object behavior prompt voices to be played as the target object behavior prompt voice; Play the target object behavior prompt voice.

11. The method according to claim 5, characterized in that Also includes: The sample sound icons in the map sample data are labeled with detection frames, and the labeled map sample data with the same icon category are overlaid and visualized to obtain a detection frame sample dataset corresponding to each icon category. The detection frame sample dataset is rotated and enhanced to obtain a target detection frame sample dataset; The target detection frame sample dataset corresponding to the icon category Y includes at least two detection frames; Clustering is performed on at least two detection frames in the target detection frame sample data set corresponding to the icon category Y to obtain N clusters, and N fixed detection frames corresponding to the icon category Y are generated according to the center positions of the N clusters.

12. A game data processing device, characterized in that: include: A display module, configured to display a target map in a game screen used to control the first virtual object; The first virtual object belongs to the first camp; The target map is used to display a sound icon representing an object behavior category and an object position of a second virtual object; The second virtual object belongs to the second camp; A voice acquisition module is used to obtain the object behavior prompt voice corresponding to the sound icon when there is a sound icon in the target map; the object behavior prompt voice is used to voice prompt the object behavior category and object position of the second virtual object; the object position includes the icon angle of the sound icon in the target map and the estimated object distance between the first virtual object and the second virtual object, the estimated object distance is obtained by color attribute mapping determined based on the purity mean of the foreground pixels, and the foreground pixels are determined according to the foreground pixel coordinates obtained by parsing the target mask image; the target mask image refers to the mask image that matches the target mask identifier in the mask configuration file, and the mask configuration file is used to store the association relationship between the mask identifier and the mask image, as well as the pixel coordinates of the mask image in the target map; the target mask identifier refers to the mask identifier that has a mapping relationship between the icon category and icon angle corresponding to the sound icon; The voice playback module is used to play the object behavior prompt voice.

13. A computer device, characterized in that: include: processor, memory, and network interface; The processor is connected to the memory and the network interface, wherein the network interface is used to provide a data communication function, the memory is used to store program code, and the processor is used to call the program code to execute the method described in any one of claims 1-11.

14. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor and executing the method according to any one of claims 1 to 11.

Images