A method, device, equipment and storage medium for automatic live broadcast and directing of sports events

Abstract

The present application discloses a method for automatic live broadcast and directing of sports events, which relates to the technical field of live broadcast data processing of events, and includes the following steps: obtaining a live video stream of the event; performing semantic segmentation on the video stream to obtain key area information of the stadium; obtaining a video frame at a certain moment in the video stream, performing camera target detection on the video frame, and calculating the camera center at the same moment based on the detection result and the key area information; using a key event recognition network to locate key events in the video stream, and performing multi-angle lens switching based on the positioning result. This solution first locates the key area on the stadium, then obtains the position of the players and the ball on the field, and then calculates the real position of the camera center based on this information to realize automatic camera movement, and finally identifies the key events in the event, and uses a close-up camera to achieve the effect of observing the same key event from multiple angles. The entire process does not require human participation, and realizes the automatic live broadcast and broadcast of sports events.

Description

Technical Field

[0001] The present application relates to the technical field of live broadcast data processing of sports events, and in particular to a method, device, equipment and storage medium for automatic live broadcast and directing of sports events. Background Art

[0002] When broadcasting sports events nowadays, a photographer usually sets up the camera, manually rotates the camera's shooting angle and manually zooms in and out to achieve zooming in and out of the picture. The director then produces the videos captured by multiple cameras on the spot according to the director's rules and then broadcasts them. The entire process requires the collaboration of multiple people to complete, which not only consumes a lot of manpower costs, but the quality of the live broadcast will inevitably be affected by manual operations.

[0003] In order to solve the problems caused by manual directing, Patent 201911248535.X discloses an intelligent directing method that can replace manual directing in sports events. It first defines the intelligent directing rules for an event, including the lens switching trigger conditions and the corresponding director signals, and then receives multiple real-time signals on the field, and performs image recognition on the video frames contained in these signals. Finally, it determines whether the recognition results meet the trigger conditions. If they meet the conditions, it switches to the director signal corresponding to the trigger condition for broadcast. Although it realizes automatic directing of the event, it fails to solve the problem that the photographer needs to manually operate the camera, and the live broadcast quality is still affected by manual operation. Summary of the invention

[0004] The present application provides a method for automatic live broadcast and directing of sports events, which aims to utilize a wide panoramic camera set up on a competition field, close-up cameras at multiple angles, and artificial intelligence technology to achieve automatic live broadcast and directing of the entire sports event without the need for human intervention, thus saving labor costs.

[0005] To achieve the above objectives, this application adopts the following technical solutions:

[0006] The present invention provides a method for automatically broadcasting and directing sports events, comprising the following steps:

[0007] Get the live video stream of the event;

[0008] Performing semantic segmentation on the video stream to obtain key area information of the stadium;

[0009] Obtaining a video frame at a certain moment in the video stream, performing camera movement target detection on the video frame, and calculating the camera movement center at the same moment according to the detection result and the key area information;

[0010] A key event recognition network is used to locate key events in the video stream, and multi-angle lens switching is performed according to the positioning results.

[0011] Preferably, the semantic segmentation of the video stream to obtain key area information of the stadium includes: inputting the video stream into a pre-established semantic segmentation model for training to obtain the key area of ​​the stadium KEY_AREA={L b ,R b ,U b ,D b ,H l ,R a ,G}, where L b , R b , U b , D b are the left and right upper and lower boundaries of the stadium, H l For the midfield line, R a is the penalty area and G is the goal area.

[0012] Preferably, the pre-established semantic segmentation model adopts a coarse-separation plus fine-separation network structure, wherein the coarse-separation network is convolution->joint pyramid upsampling->hole space pyramid pooling, and the fine-separation network is an optimization module based on skip connections.

[0013] Preferably, the acquiring of a video frame at a certain moment in the video stream, performing camera movement target detection on the video frame, and calculating the camera movement center at the same moment according to the detection result and the key area information comprises:

[0014] Obtaining the video frame at time t in the video stream;

[0015] The object detection network is used to process the video frame to obtain the position B of the ball on the field at time t t =(x b ,y b ) and the athlete's position P t ={(x1,y1),…,(x k ,y k )}, where k is the number of athletes on the field at time t;

[0016] The camera movement center at the tth moment is calculated according to the positions of the ball and the player and the key area information.

[0017] Preferably, the step of calculating the camera movement center at the tth moment according to the positions of the ball and the player in combination with the key area information comprises:

[0018] Calculate the initial camera center C of the frame at time t t , the calculation formula is:

[0019] C t =σ*B t +(1-σ)*∑P t / k

[0020] Where σ is the weight coefficient of the ball, and the camera center C t 、Mirroring window width and height W t , H t satisfy:

[0021]

[0022] Among them C t (x), C t (y) are C t The x and y coordinates of

[0023] In the time dimension, the exponentially weighted average is used to calculate C t Smoothing is performed to filter out noise, and the calculation formula is:

[0024]

[0025] in are the smoothed camera movement centers at time t and time t-1 respectively, and β is the exponential smoothing coefficient.

[0026] Preferably, the key event recognition network is used to locate the key events in the video stream, and multi-angle lens switching is performed according to the location result, including:

[0027] Using a key event recognition network to identify the type and occurrence time of key events in the video stream;

[0028] Obtaining the event subject corresponding to the key event;

[0029] According to the type, occurrence time and event subject of the key event, video streams of the key event from multiple angles are obtained and switched for playback.

[0030] Preferably, the process of forming the key event identification network includes:

[0031] Collect videos of the same type of competition from different angles;

[0032] The key events of the video are annotated to generate labels {(t s ,t e ,type)}, where t s is the starting time of the key event, t e is the end time of the key event, and type is the type of the key event;

[0033] Extract training samples from the video, where the positive samples and the annotated time region t s ~t e If the overlap exceeds the preset threshold, the rest are negative samples;

[0034] The training samples are trained using a recognition network to obtain a key event recognition network.

[0035] An automatic live broadcast and directing device for sports events, comprising:

[0036] The acquisition module is used to obtain the live video stream of the event;

[0037] A positioning module, used to perform semantic segmentation on the video stream to obtain key area information of the stadium;

[0038] A camera movement module, used to obtain a video frame at a certain moment in the video stream, perform camera movement target detection on the video frame, and calculate the camera movement center at the same moment according to the detection result and the key area information;

[0039] The director module is used to locate key events in the video stream using a key event recognition network, and to switch between multi-angle shots according to the location results.

[0040] An electronic device comprises a memory and a processor, wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement a method for automatic live broadcast and directing of sports events as described in any one of the above.

[0041] A computer-readable storage medium storing a computer program, wherein the computer program, when executed by a computer, implements a method for automatic live broadcast and directing of sports events as described in any one of the above.

[0042] The present invention has the following beneficial effects:

[0043] This solution can locate key areas on the field by performing semantic segmentation on the live video stream of the event, and then use the target detection network to track the positions of athletes and the ball on the field. Then, the real position of the camera center is calculated based on this information to achieve automatic conversion of the camera center. Finally, the key events in the event are identified based on the key event recognition network, and related clips of the key events are switched to play from multiple angles. The entire process does not require human intervention, realizing the automatic live broadcast and rebroadcast of sports events. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0045] Figure 1This is a first flow chart of a method for automatically broadcasting and directing sports events according to Embodiment 1 of the present application;

[0046] Figure 2 This is a structural diagram of the semantic segmentation model of Example 2 of the present application;

[0047] Among them, D0 is the original image, D1 is the image after downsampling the original image, P1 is the coarse prediction result, and P0 is the refined result;

[0048] Figure 3 This is a second flow chart of a method for automatically broadcasting and directing sports events according to Embodiment 3 of the present application;

[0049] Figure 4 This is the third flow chart of a method for automatically broadcasting and directing sports events according to Embodiment 4 of the present application;

[0050] Figure 5 This is a schematic diagram of a device for automatically broadcasting and directing sports events according to Embodiment 5 of the present application;

[0051] Figure 6 It is a schematic diagram of an electronic device implementing a method for automatic live broadcast and directing of sports events according to Example 6 of the present application. DETAILED DESCRIPTION

[0052] The following will be combined with the accompanying drawings to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0053] The terms "first", "second", etc. in the claims and specification of the present application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way are interchangeable under appropriate circumstances. This is merely a way of distinguishing objects with the same properties in the embodiments of the present application. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, so that a process, method, system, product or apparatus that includes a series of units is not necessarily limited to those units, but may include other units not expressly listed or inherent to these processes, methods, products or apparatus.

[0054] Example 1

[0055] like Figure 1 As shown, a method for automatic live broadcast and directing of sports events includes the following steps:

[0056] S110, obtaining a live video stream of the event;

[0057] S120, performing semantic segmentation on the video stream to obtain key area information of the stadium;

[0058] S130, obtaining a video frame at a certain moment in the video stream, performing camera movement target detection on the video frame, and calculating the camera movement center at the same moment according to the detection result and the key area information;

[0059] S140, using a key event recognition network to locate key events in the video stream, and performing multi-angle lens switching according to the location results.

[0060] According to Example 1, this solution uses a wide panoramic camera and multiple close-up cameras. First, the wide camera is used to shoot the whole field to generate a video stream S with a width and height of W and H respectively, and a field of view covering the whole field. The video stream S is parsed using a pre-established semantic segmentation model to obtain key area information on the field, including the left and right upper and lower boundaries of the field, the midfield line, the penalty area and the goal area, etc. Then, a video frame at a certain moment is obtained, and the position of the ball and the athlete on the field is located using a target detection network, that is, the target position of the camera movement. Combined with the key area information obtained above, the camera movement center calculation formula is used to calculate the real camera movement center at that moment, thereby realizing the automatic conversion of the camera movement center. Next, the key event recognition network is used to identify the key events contained in the video stream, and the type and occurrence time of the key events are obtained. When key events such as fouls, scrambles, and goals occur, the event subject is first located, and the event subject is strengthened by using strategies such as zooming in and tracking the lens, and the timestamp t of the key event is recorded. Then, the timestamp t is located on the video streams of other close-up cameras, and the key event fragment E is intercepted with it as the starting point. m}, where m is the number of close-up cameras. By switching between different clips of E, the same key event can be observed from different perspectives. This solution uses wide panoramic cameras set up on the field, close-up cameras at multiple angles, and deep learning technology to achieve automatic live broadcast and directing of the entire sports game. It does not require manual participation, saves labor costs, and avoids errors caused by manual operation negligence. At the same time, it realizes automatic switching of lenses, which can describe the same key event from multiple angles and help the audience better understand the event.

[0061] Example 2

[0062] like Figure 2 As shown, a method for automatic live broadcast and directing of sports events includes:

[0063] S210, obtaining a live video stream of the event;

[0064] S220, inputting the video stream into a pre-established semantic segmentation model for training to obtain key area information of the stadium;

[0065] S230, obtaining a video frame at a certain moment in the video stream, performing camera movement target detection on the video frame, and calculating the camera movement center at the same moment according to the detection result and the key area information;

[0066] S240: locating key events in the video stream using a key event recognition network, and performing multi-angle lens switching according to the locating result.

[0067] According to Example 2, after acquiring the video shot by the wide-angle camera, this solution will use the trained semantic segmentation model to analyze the video to identify the key area KEY_AREA={L b ,R b ,U b ,D b ,H l ,R a ,G}, where L b , R b , U b , D b They are the left and right upper and lower boundaries of the court, H l For the midfield line, R a is the restricted area, G is the goal area, where {L b ,R b ,U b ,D b} They are expanded outwards at a certain ratio T to serve as the boundaries of subsequent automatic camera movements. The midfield line, penalty area, and goal are the key coverage areas of subsequent camera movements.

[0068] The computational complexity of the existing semantic segmentation model is related to the pixel size. Since the wide-format camera used in this solution has a high imaging resolution (above 4k), directly inputting high-resolution images into the semantic segmentation model will lead to problems such as excessive video memory usage and difficulty in network training. At the same time, labeling high-resolution images also increases the difficulty of labeling. Therefore, Figure 2 The semantic segmentation model trained in this application uses a coarse-separation and fine-separation network structure, in which the coarse-separation network is convolution->joint pyramid upsampling->hole space pyramid pooling, and the fine-separation network is an optimization module based on skip connection. The specific process is as follows:

[0069] The original image D0 is downsampled to obtain image D1. conv1-conv5 are all 3*3 convolution + relu + pooling structures. The feature scale of the convi layer is half of conv(i-1). Then the feature maps feat3-feat5 corresponding to conv3-conv5 are convolutionally upsampled to the same size, and then the channels are merged to obtain C1, which is used to fuse features from different scales. Then, C1 is subjected to feature extraction and merging at multiple scales using multiple stride convolutions with a step size of s (s=1, 2, 4, 8). Feature layers of different scales are combined to obtain C2, and multi-layer joint upsampling is completed. After passing through the atrous spatial convolution pooling pyramid (ASPP), the coarse prediction result P1 is obtained.

[0070] The refinement module uses the original image D0 and the coarse segmentation result P1 to generate a more refined segmentation: first, P1 is bilinearly upsampled to the same size as D0, and then merged with D0 and input into the network. After pyramid pooling, multi-level upsampling and jump connection, it can capture image content and boundary information at different levels, fuse multi-level granularity segmentation features, and finally obtain the refined result P0.

[0071] Example 3

[0072] like Figure 3 As shown, a method for automatic live broadcast and directing of sports events includes:

[0073] S310, obtaining a live video stream of the event;

[0074] S320, performing semantic segmentation on the video stream to obtain key area information of the stadium;

[0075] S330, obtaining a video frame at time t in the video stream;

[0076] S340, using a target detection network to process the video frame to obtain the position of the ball and the position of the player on the court at time t;

[0077] S350, calculating the camera movement center at the tth moment according to the positions of the ball and the player and the key area information;

[0078] S360: locating key events in the video stream using a key event recognition network, and performing multi-angle lens switching according to the locating result.

[0079] According to Example 3, after identifying the key area of ​​the court, this solution will obtain the video frame at time t in the video stream, first use the target detection network to locate the position B of the ball on the court t =(x b ,y b ), and the athlete's position Pt ={(x1,y1),…,(x k ,y k )}, where k is the number of players on the field at time t. Different from the traditional target detection network, this scheme adopts a multi-frame + time series input strategy to maintain the accuracy of ball tracking, making the input a multi-frame stacked IP t = {I t-d ,…,I t ,…,I t+d}, where IP t is the input of the detection network at time t, I t-d ,I t ,I t+d They are images at time td, t, and t+d, respectively, where d is the time span before and after the stack. Then, the multi-layer LSTM is input into the network feature extraction layer to regress the position and category of the current detection frame. In this way, the time series can be implicitly modeled when the ball and the athlete are occluded, and the position of the current frame can be estimated based on the previous frame. At the same time, the misdetection of other balls on the field can be eliminated.

[0080] Next, we calculate the real center of the camera movement. First, we calculate the initial camera movement center C of the frame at time t. The calculation formula is:

[0081] C t =σ*B t +(1-σ)*∑P t / k

[0082] Where σ is the weight coefficient of the ball, and the camera center C t 、Mirroring window width and height W t , H t Should meet:

[0083]

[0084] Among them C t (x), C t (y) are C t The x and y coordinates of .

[0085] In the time dimension, the exponentially weighted average is used to calculate C t Smoothing is performed to filter out noise, and the calculation formula is:

[0086]

[0087] in are the smoothed camera movement centers at time t and time t-1 respectively, and β is the exponential smoothing coefficient.

[0088] Let the real camera center at time t be R t, the camera speed is V t , then V t =γV t-1 +(1-γ)(R t -R t-1 ), where γ is the speed keeping coefficient, and the mirror movement start threshold ST is set. t ≤S T And V t-1 =0, V t =0, that is, when the camera movement speed is less than the start threshold and the previous frame image is a still image, the current frame is also a still image, and the camera movement center remains unchanged.

[0089] Example 4

[0090] like Figure 4 As shown, a method for automatic live broadcast and directing of sports events includes:

[0091] S410, obtaining a live video stream of the event;

[0092] S420, performing semantic segmentation on the video stream to obtain key area information of the stadium;

[0093] S430, obtaining a video frame at a certain moment in the video stream, performing camera movement target detection on the video frame, and calculating the camera movement center at the same moment according to the detection result and the key area information;

[0094] S440, using a key event recognition network to identify the type and occurrence time of key events in the video stream;

[0095] S450, obtaining the event subject corresponding to the key event;

[0096] S460: According to the type, occurrence time and event subject of the key event, obtain video streams of the key event from multiple angles, and switch to play.

[0097] According to Example 4, this solution uses a key event recognition network to recognize events in a video stream, and switches and plays the recognized key events from multiple angles to achieve the effect of showing the details of the same event from different angles. Specifically, when recognizing key events such as fouls, scrambles, and goals, first locate the event subject, use strategies such as zooming in and tracking to strengthen the event subject, and record the timestamp t of the event. Then, locate the timestamp t on the video stream of other close-up cameras, and use it as the starting point to intercept the recognized key event segment E={E1,...,E m}, where m is the number of close-up cameras. Finally, by switching between different clips of E, the effect of observing the same event from different perspectives is achieved.

[0098] Among them, the training process of the key event recognition network used in this solution is:

[0099] (1) Collect videos of the same type of competition from different angles;

[0100] (2) Annotate the key events of the collected videos and generate labels {(t s ,t e ,type)}, where t s is the starting time of the key event, t e is the end time of the key event, and type is the type of the key event, including foul, scramble, goal, etc.;

[0101] (3) Extract training samples from the collected videos, where the positive samples and the annotated time region t s ~t e If the overlap exceeds a certain threshold, the rest are negative samples;

[0102] (4) Use recognition networks such as X3D and TPN to train the extracted training samples.

[0103] After the training is completed, the video stream of the real-time game is input into the key event recognition network to locate the key events in the game.

[0104] Example 5

[0105] like Figure 5 As shown, a sports event automatic live broadcast and directing device comprises:

[0106] The acquisition module 10 is used to acquire the live video stream of the event;

[0107] A positioning module 20, used for performing semantic segmentation on the video stream to obtain key area information of the stadium;

[0108] A camera movement module 30 is used to obtain a video frame at a certain moment in the video stream, perform camera movement target detection on the video frame, and calculate the camera movement center at the same moment according to the detection result and the key area information;

[0109] The director module 40 is used to locate key events in the video stream using a key event recognition network, and to perform multi-angle lens switching according to the location results.

[0110] One implementation method of the above-mentioned device may be: the acquisition module 10 acquires the live video stream of the event; the positioning module 20 performs semantic segmentation on the video stream to obtain key area information of the stadium; the camera movement module 30 acquires the video frame at a certain moment in the video stream, performs camera movement target detection on the video frame, and calculates the camera movement center at the same moment based on the detection result and the key area information; the director module 40 uses the key event recognition network to locate the key events in the video stream, and performs multi-angle lens switching based on the positioning result.

[0111] Example 6

[0112] like Figure 6 As shown, an electronic device includes a memory 601 and a processor 602, wherein the memory 601 is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor 602 to implement the above-mentioned method for automatic live broadcast and directing of sports events.

[0113] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the electronic device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0114] A computer-readable storage medium storing a computer program, wherein the computer program enables a computer to implement the above-mentioned method for automatic live broadcast and directing of sports events when executed.

[0115] Exemplarily, the computer program may be divided into one or more modules / units, one or more modules / units are stored in the memory 601 and executed by the processor 602, and the I / O interface transmission of data is completed by the input interface 605 and the output interface 606 to complete the present invention. The one or more modules / units may be a series of computer program instruction segments that can complete specific functions, and the instruction segments are used to describe the execution process of the computer program in the computer device.

[0116] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may include, but is not limited to, a memory 601 and a processor 602. Those skilled in the art may understand that this embodiment is only an example of a computer device and does not constitute a limitation on the computer device. The computer device may include more or fewer components, or a combination of certain components, or different components. For example, the computer device may also include an input device 607, a network access device, a bus, etc.

[0117] The processor 602 may be a central processing unit (CPU), or other general-purpose processors 602, digital signal processors 602 (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor 602 may be a microprocessor 602 or any conventional processor 602, etc.

[0118] The memory 601 may be an internal storage unit of a computer device, such as a hard disk or memory of the computer device. The memory 601 may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (SmartMediaCard, SMC), a secure digital (SecureDigital, SD) card, a flash card (FlashCard), etc. equipped on the computer device. Further, the memory 601 may also include both an internal storage unit of the computer device and an external storage device. The memory 601 is used to store computer programs and other programs and data required by the computer device. The memory 601 may also be used to temporarily store in the output device 608, and the aforementioned storage media include various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory ROM603, a random access memory RAM604, a disk or an optical disk.

[0119] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for automatic live broadcast and directing of sports events, characterized in that: The following steps are involved: Get the live video stream of the event; The video stream is semantically segmented to obtain key area information of the stadium, where the key area of ​​the stadium is KEY_AREA={L b ,R b ,U b ,D b ,H l ,R a ,G}, where L b , R b , U b , D b are the left and right upper and lower boundaries of the stadium, H l For the midfield line, R a is the penalty area, G is the goal area; Obtaining a video frame at a certain moment in the video stream, performing camera movement target detection on the video frame, and calculating the camera movement center at the same moment according to the detection result and the key area information; Using a key event recognition network to locate key events in the video stream, and switching between multiple angles of shots based on the location results; The acquiring of a video frame at a certain moment in the video stream, performing camera movement target detection on the video frame, and calculating the camera movement center at the same moment according to the detection result and the key area information includes: Obtaining the video frame at time t in the video stream; The object detection network is used to process the video frame to obtain the position B of the ball on the field at time t t =(x b ,y b ) and the athlete's position P t ={(x1,y1),…,(x k ,y k )}, where k is the number of athletes on the field at time t; Calculate the camera movement center at the tth moment according to the positions of the ball and the player and the key area information; The step of calculating the camera movement center at the tth moment according to the positions of the ball and the player and the key area information includes: Calculate the initial camera center C of the frame at time t t , the calculation formula is: C t =σ*B t +(1-σ)*∑P t / k Where σ is the weight coefficient of the ball, and the camera center C t 、Mirroring window width and height W t , H t satisfy: Among them C t (x), C t (y) are C t x, y coordinates, T is the scale factor, indicating {L b ,R b ,U b ,D b The extent of outward expansion; In the time dimension, the exponentially weighted average is used to calculate C t Smoothing is performed to filter out noise, and the calculation formula is: in are the smoothed camera movement centers at time t and time t-1 respectively, and β is the exponential smoothing coefficient.

2. A method for automatic live broadcast and directing of sports events according to claim 1, characterized in that: The performing semantic segmentation on the video stream to obtain key area information of the playing field includes: inputting the video stream into a pre-established semantic segmentation model for training to obtain the key area KEY_AREA of the playing field.

3. A method for automatic live broadcast and directing of sports events according to claim 2, characterized in that: The pre-established semantic segmentation model adopts a coarse-separation plus fine-separation network structure, wherein the coarse-separation network is convolution->joint pyramid upsampling->hole space pyramid pooling, and the fine-separation network is an optimization module based on skip connections.

4. The method for automatic live broadcast and directing of sports events according to claim 1, characterized in that: The key event recognition network is used to locate the key events in the video stream, and multi-angle lens switching is performed according to the location result, including: Using a key event recognition network to identify the type and occurrence time of key events in the video stream; Obtaining the event subject corresponding to the key event; According to the type, occurrence time and event subject of the key event, video streams of the key event from multiple angles are obtained and switched for playback.

5. A method for automatic live broadcast and directing of sports events according to claim 4, characterized in that: The formation process of the key event identification network includes: Collect videos of the same type of competition from different angles; The key events of the video are annotated to generate labels {(t s ,t e ,type)}, where t s is the starting time of the key event, t e is the end time of the key event, and type is the type of the key event; Extract training samples from the video, where the positive samples and the annotated time region t s ~t e If the overlap exceeds the preset threshold, the rest are negative samples; The training samples are trained using a recognition network to obtain a key event recognition network.

6. An automatic live broadcast and directing device for sports events, characterized in that: include: The acquisition module is used to obtain the live video stream of the event; The positioning module is used to perform semantic segmentation on the video stream to obtain key area information of the stadium. The key area of ​​the stadium is KEY_AREA={L b ,R b ,U b ,D b ,H l ,R a ,G}, where L b , R b , U b , D b are the left and right upper and lower boundaries of the stadium, H l For the midfield line, R a is the penalty area, G is the goal area; A camera movement module, used to obtain a video frame at a certain moment in the video stream, perform camera movement target detection on the video frame, and calculate the camera movement center at the same moment according to the detection result and the key area information; A director module, used for locating key events in the video stream using a key event recognition network, and performing multi-angle lens switching according to the positioning results; The acquiring of a video frame at a certain moment in the video stream, performing camera movement target detection on the video frame, and calculating the camera movement center at the same moment according to the detection result and the key area information includes: Obtaining the video frame at time t in the video stream; The object detection network is used to process the video frame to obtain the position B of the ball on the field at time t t =(x b ,y b ) and the athlete's position P t ={(x1,y1),…,(x k ,y k )}, where k is the number of athletes on the field at time t; Calculate the camera movement center at the tth moment according to the positions of the ball and the player and the key area information; The step of calculating the camera movement center at the tth moment according to the positions of the ball and the player and the key area information includes: Calculate the initial camera center C of the frame at time t t , the calculation formula is: C t =σ*B t +(1-σ)*∑P t / k Where σ is the weight coefficient of the ball, and the camera center C t 、Mirroring window width and height W t , H t satisfy: Among them C t (x), C t (y) are C t x, y coordinates, T is the scale factor, indicating {L b ,R b ,U b ,D b The extent of outward expansion; In the time dimension, the exponentially weighted average is used to calculate C t Smoothing is performed to filter out noise, and the calculation formula is: in are the smoothed camera movement centers at time t and time t-1 respectively, and β is the exponential smoothing coefficient.

7. An electronic device, characterized in that: It comprises a memory and a processor, wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement a method for automatic live broadcast and directing of sports events as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that: The computer program enables a computer to implement a method for automatic live broadcast and directing of sports events as claimed in any one of claims 1 to 5 when executed.

Images