Competition monitoring system, image analysis device, image analysis method and program

Abstract

To provide a competition monitoring system, image analysis device, image analysis method, and program capable of assisting in the safe progress of contact sports. [Solution] The competition monitoring system 1 of this disclosure includes a terminal device 10, a plurality of imaging devices 20-1 to 20-4 capable of imaging a plurality of athletes P1 participating in a specific contact sport, and an image analysis device 30 that detects athletes in a specific state from among the athletes captured by the plurality of imaging devices and outputs the detection result to the terminal device 10. The terminal device 10 provides the analysis result of the image analysis device 30 to at least one of the competition personnel P2 to P4 in substantially real time.

Description

Technical Field

[0001] The present disclosure relates to a competition monitoring system, an image analysis device, an image analysis method, and a program.

Background Art

[0002] Contact sports such as rugby, American football, and soccer have a relatively high risk of injury because contact between players frequently occurs. Therefore, ensuring the safety of players during competition in contact sports is extremely important. In particular, actions involving intense collisions between players, such as tackles in rugby and American football, are likely to lead to trauma and injury to the head and neck of players. Trauma to the head and neck of players caused by tackles or the like is likely to result in serious injuries such as concussion, cerebral hemorrhage, and neck fractures, and the risk is extremely high.

[0003] For example, Patent Document 1 below describes a technique for monitoring a user by attaching various sensors to the clothing of the user including sports players.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] Of the injuries mentioned above, concussions in particular are known to require prompt protection after they occur to prevent permanent disability in athletes. On the other hand, most concussions that occur in the context of sports do not involve loss of consciousness, and athletes who suffer a concussion may continue competing, optimistically assessing their condition. Therefore, it is crucial for medical staff, referees, and other sports personnel who support the competition without directly participating to accurately identify and detect concussions among athletes and to implement timely suspension of the competition or provide protection and treatment to athletes in order to ensure safe competition. To accurately implement the aforementioned measures, there is a need for technologies that can assist in the detection and identification of concussions.

[0006] In view of the above-mentioned issues, this disclosure aims to provide a competition monitoring system, an image analysis device, an image analysis method, and a program that can assist in the safe progress of competition in contact sports. [Means for solving the problem]

[0007] To achieve the above objective, a competition monitoring system according to a first aspect of this disclosure includes a terminal device, a plurality of imaging devices capable of imaging a plurality of athletes participating in a particular contact sport, and an image analysis device that detects athletes in a specific state among the athletes captured in the images captured by the plurality of imaging devices and outputs the detection results to the terminal device.

[0008] In such a competition monitoring system, it is possible to automatically detect only those athletes in a specific state during a competition and output the information to a terminal device. This allows users who possess the terminal device, such as those involved in the competition, to immediately recognize the presence of athletes in a specific state, thereby assisting in understanding the athletes' condition.

[0009] A competition monitoring system according to a second aspect of the present disclosure, wherein the competition monitoring system according to the first aspect of the present disclosure includes at least one of a fallen state, a lying-down state, and a kneeling state.

[0010] In such a competition monitoring system, the state an athlete takes when a concussion occurs can be detected preferentially, and the detection results can be used to understand and identify the occurrence of a concussion.

[0011] A third aspect of the present disclosure relates to a competition monitoring system according to the first or second aspect of the present disclosure, wherein the image analysis device includes an inference unit that identifies a competitor in a specific state within an image by inputting the captured image into a trained model that has learned training data including images containing a competitor in a specific state.

[0012] Such a competition monitoring system will be able to automatically estimate which athletes are in a specific state within the captured images.

[0013] A competition monitoring system according to a fourth aspect of the present disclosure, in which the competition monitoring system according to a third aspect of the present disclosure comprises a plurality of pre-trained models, and integrates inference results obtained by inputting the captured images into each of the plurality of pre-trained models to identify a competitor in a specific state within the captured images.

[0014] In such a competition monitoring system, it is possible to perform so-called ensemble learning using multiple pre-trained models, thereby improving the inference accuracy of the inference unit.

[0015] A fifth aspect of the present disclosure relates to a competition monitoring system according to any of the first to fourth aspects of the present disclosure, wherein the image analysis device includes a timer and outputs to the terminal device whether or not there is a competitor whose specific state continues for a predetermined period of time.

[0016] In such a competition monitoring system, information regarding the duration of a specific condition can also be output to the terminal device.

[0017] An image analysis device according to a sixth aspect of this disclosure includes an image acquisition unit that acquires captured images of multiple athletes performing a particular contact sport, a detection unit that detects athletes in a particular state among the athletes in the acquired captured images, and an output unit that outputs the detection results from the detection unit.

[0018] Such image analysis devices can automatically detect and output athletes in specific states within captured images, allowing for immediate identification of athletes in specific states based on the output results.

[0019] The seventh aspect of the present disclosure is an image analysis method that uses a computer to acquire captured images of multiple athletes performing a specific contact sport, detects athletes in a specific state from among the athletes in the acquired captured images, and outputs the detection result.

[0020] In this type of image analysis method, athletes in a specific state can be automatically detected and output within the captured image, allowing for immediate identification of athletes in a particular state from the output results.

[0021] A program according to the eighth aspect of this disclosure causes a computer processor to execute a process that acquires captured images of multiple athletes performing a particular contact sport, detects athletes in a specific state among the acquired captured images, and outputs the detection result.

[0022] Such programs can automatically detect and output athletes in specific states within captured images, allowing for immediate identification of athletes in particular states from the output results. [Effects of the Invention]

[0023] The competition monitoring system, image analysis device, image analysis method, and program described herein can assist in ensuring safe competition in contact sports. [Brief explanation of the drawing]

[0024] [Figure 1] This is a schematic explanatory diagram showing an example of a competition monitoring system according to an embodiment of the present disclosure. [Figure 2] This is a block diagram showing an example of the hardware configuration of an image analysis device according to an embodiment of the present disclosure. [Figure 3] This is a functional block diagram of the image analysis device shown in FIG. 2. [Figure 4] This is an example of a captured image captured by a camera. [Figure 5] This shows an example of the input data and output data of the inference unit. [Figure 6] This is an explanatory diagram schematically showing an example of an inference process in the inference unit. [Figure 7] This shows another example of the input data and output data of the inference unit. [Figure 8] This is a flowchart showing an example of an image analysis method according to an embodiment of the present disclosure.

Embodiments for Carrying Out the Invention

[0025] Hereinafter, each embodiment for implementing the present disclosure will be described with reference to the drawings. In the following, the scope necessary for the description for achieving the object of the present disclosure is schematically shown, and the scope necessary for the description of the corresponding part of the present disclosure will be mainly described, and the parts where the description is omitted are assumed to be based on known techniques. In addition, the same or corresponding members in the drawings are denoted by the same or similar reference numerals, and duplicate descriptions are omitted. Further, when a plurality of the same or corresponding members are included in the drawings, in order to make the drawings easy to view, in some cases, only some of them are denoted by reference numerals.

[0026] In the competition monitoring system, image analysis device, image analysis method, and program according to an embodiment described below, rugby is exemplified as a specific contact sport performed by athletes. Note that the contact sports to which the present disclosure is applicable are not limited to rugby.

[0027] Figure 1 is a schematic diagram illustrating an example of a competition monitoring system according to one embodiment of the present disclosure. The competition monitoring system 1 according to this embodiment may be a system for monitoring a rugby game played by multiple players P1 in a competition space 2 called a ground or pitch, as shown in Figure 1. This competition monitoring system 1 includes at least a terminal device 10, cameras 20-1 to 20-4 as an example of a plurality of imaging devices, and an image analysis device 30.

[0028] When a rugby match is played within playing space 2, in addition to the multiple players P1 belonging to each participating team, other personnel involved in the match may also be present, such as medical staff P2 as members of the team to which players P1 belong, a doctor in charge of the match P3, and referees P4. Of these, approximately two medical staff members P2 and one to two doctors in charge of the match P3 are expected per team, and they usually wait around playing space 2. The referees P4 may consist of one head referee inside playing space 2 and one to three assistant referees either inside or outside playing space 2. The aforementioned personnel P2-P4 ensure the safe progress of the match by supporting players P1 and refereeing. In Figure 1, to distinguish between players P1 and personnel P2-P4 who are not players, players P1 in play are shown with circles, and personnel P2-P4 are shown with squares.

[0029] The terminal device 10 enables the owner to receive the analysis results of the image analysis device 30 in substantially real time. Here, the owner may be at least one of the competition officials P2 to P4, for example, the competition doctor P3. The terminal device 10 is not particularly limited as long as it is a device capable of notifying the analysis results of the image analysis device 30, and may be a device with a display such as a tablet terminal or smartphone, or it may be a headset equipped with a microphone and speaker that constitute an intercommunication system. In Figure 1, the terminal device 10 is shown as an example when it is a tablet terminal. Furthermore, it is preferable that multiple members of the competition officials P2 to P4 possess the terminal device 10.

[0030] Cameras 20-1 to 20-4 should be positioned in multiple locations around the competition space 2 to reliably image multiple competitors during the competition, thereby capturing the entire competition space 2. In this embodiment, as shown in Figure 1, four cameras 20-1 to 20-4 are shown as an example, positioned adjacent to the four corners of the competition space 2. Two-dimensional cameras using image sensors such as CCD (Charge Coupled Device) sensors or CMOS (Complementary Metal Oxide Semiconductor) sensors can be used as cameras 20-1 to 20-4. However, cameras 20-1 to 20-4 are not limited to the two-dimensional cameras described above. For example, some of cameras 20-1 to 20-4, or in addition to cameras 20-1 to 20-4, can be three-dimensional cameras such as stereo cameras or structured lights. Furthermore, the arrangement and number of cameras 20-1 to 20-4 are not particularly limited, and the optimal arrangement and number for capturing the entire competition space 2 can be adopted.

[0031] Figure 2 is a block diagram showing an example of the hardware configuration of an image analysis device according to one embodiment of the present disclosure. The image analysis device 30 is a device capable of detecting the state (posture) of athletes in images captured by cameras 20-1 to 20-4. As shown in Figure 2, this image analysis device 30 can be configured with a well-known computer such as a server. The computer may include, for example, a processor 31, ROM (Read Only Memory) 32 and RAM (Random Access Memory) 33 as an example of memory, storage 34, a communication interface 35, and an input / output interface 36. Furthermore, each of the above-mentioned components may be connected to each other so as to be able to communicate with one another via an internal bus 37.

[0032] The processor 31 is a central processing unit capable of executing various programs and performing arbitrary arithmetic operations. This processor 31 can be configured, for example, as a CPU (Central Processing Unit). The processor 31 may be capable of reading various programs stored in the ROM 32 and / or storage 34 and executing those programs using the RAM 33 as a working area.

[0033] ROM32 may be capable of storing various programs and data. RAM33 may also be capable of temporarily storing programs or data as a working area.

[0034] The storage 34 can consist of recording media such as an HDD (Hard Disk Drive), SSD (Solid State Drive), or flash memory, and may store various programs including an operating system, as well as various data necessary to operate the image analysis device 30. In this embodiment, the ROM 32 or storage 34 may store programs and various data used to detect a specific state of the athlete P1.

[0035] The communication interface (I / F) 35 may be an interface capable of wireless communication for communicating various data with cameras 20-1 to 20-4 and terminal devices 10. Communication standards such as CAN (Controller Area Network), Ethernet (registered trademark), LTE (Long Term Evolution), FDDI (Fiber Distributed Data Interface), or Wi-Fi (registered trademark) may be used in this communication interface 35.

[0036] The input / output interface (I / F) 36 may be an interface that connects to an operating device 38, which may include, for example, a keyboard, touch panel, or pointing device, and a display 39, which is composed of a liquid crystal display (LCD) or an organic electro-luminescent display (OELD), to realize the input and output of various types of data. The operating device 38 and the display 39 are mainly used by the administrator of the image analysis device 30 when controlling the image analysis device 30.

[0037] Incidentally, when a concussion occurs in a rugby match, as mentioned above, in most cases it does not involve loss of consciousness, but the player almost always falls or kneels immediately after the concussion. In addition, when a concussion occurs, the player tends to be slow to recover from the fall and may be unsteady when standing up. Therefore, for the sports officials P2 to P4 to understand that a player has suffered a concussion, information that the player is in a specific state (specific posture) that is highly associated with the occurrence of a concussion, such as falling, lying on their side, or kneeling on the ground, is useful. Taking these points into consideration, the image analysis device 30 according to this embodiment, which has the configuration described above, assists the sports officials P2 to P4 in understanding the occurrence of a concussion by detecting the specific states of the player, particularly those mentioned above.

[0038] Figure 3 is a functional block diagram of the image analysis device shown in Figure 2. In order to assist in the detection of concussions by the aforementioned sports personnel P2 to P4, the image analysis device 30 according to this embodiment functions to include at least an image acquisition unit 41, a detection unit 42 for detecting athlete P1A in a specific state (see Figures 5 and 7), and an output unit 43, as shown in Figure 3. Here, the specific state preferably includes at least one of a fallen state, a lying-down state, and a kneeling state, but is not limited to these and may also include other states related to concussions. Note that Figure 3 mainly illustrates only the functions related to image analysis for identifying athletes in a specific state, and other functions not directly related to this function are omitted from the illustration.

[0039] Figure 4 shows an example of an image captured by a camera. The image acquisition unit 41 can be implemented mainly by the communication interface 35 and acquires images from multiple cameras 20-1 to 20-4 that include multiple athletes participating in a competition (e.g., during a match). For example, the image captured by camera 20-1 may be one that includes multiple athletes competing in the competition space 2, as shown in Figure 4.

[0040] The detection unit 42 can be mainly implemented by the processor 31 and detects a specific athlete P1A (see Figures 5 and 7) from among the athletes in the captured image acquired by the image acquisition unit 41. The specific method for detecting athlete P1A in a specific state is not particularly limited, but in this embodiment, an example is provided which includes an inference unit 44 that detects athlete P1A in a specific state using an inference model based on machine learning.

[0041] The inference unit 44 uses deep learning with a multi-layered hierarchical neural network to recognize a desired object from a specific image. When an image captured by cameras 20-1 to 20-4 is input to this inference unit 44, it may include a trained model that estimates a specific state among the athletes included in the image, for example, an athlete in a fallen state, and outputs it as a detection result. The format of the output information output here is not particularly limited, but for example, it may output an image with a mark (for example, a frame 50 described later) placed on the athlete in question in the image. Furthermore, the inference unit 44 or the trained model within the inference unit 44 may be provided separately for each camera 20-1 to 20-4, or the images captured by each camera 20-1 to 20-4 may be input to a single inference unit 44 or the trained model within the inference unit 44.

[0042] Figure 5 shows an example of input and output data for the inference unit, where Figure 5(A) is an example of input data and Figure 5(B) is an example of output data. When the inference unit 44 receives an image, for example, as shown in Figure 5(A), it may estimate a specific state within the image, specifically a fallen athlete P1A, and output an image with a frame 50 added surrounding the fallen athlete P1A, as shown in Figure 5(B).

[0043] The trained model included in the inference unit 44 may be, for example, a neural network model trained by supervised learning. The training data used for this supervised learning (also called the training dataset) can be generated, for example, by performing annotation work on images captured by any of the cameras 20-1 to 20-4, labeling the areas in the images where athletes are shown in a fallen state. Such training data can be divided into training data and validation data, and the training data can be used to train the model before training, and the validation data can be used to verify the inference accuracy of the trained model.

[0044] Furthermore, in order to reduce the time required to train the pre-trained model included in the inference unit 44 and the amount of training data, it is preferable to utilize transfer learning. Specifically, for example, pre-trained models used in well-known object detection systems, such as YOLO (You Look Only Once) or other image recognition systems, can be used.

[0045] Figure 6 is a schematic diagram illustrating an example of the inference process in the inference unit. In this embodiment, ensemble learning is employed to improve the inference accuracy in the inference unit 44. Specifically, an integrated (majority-voting) deep learning network using bagging can be employed. Briefly explaining this network, as shown in Figure 6, first, multiple trained models (such trained models are sometimes called weak learners) 45-1 to 45-N (N=2 or greater) trained on different training data are prepared. Then, a single captured image is input as input data to each of these multiple trained models 45-1 to 45-N, and the obtained inference results are collected in the inference result identification unit 46, integrated, specifically by majority vote or summation, and output as a detection result. By utilizing the ensemble learning described above, a significant improvement in the detection accuracy of athlete P1A in a specific state can be expected. While there is no particular limit to the number of pre-trained models 45-1 to 45-N, using around 3 or 5 models resulted in better detection results compared to using a single pre-trained model.

[0046] Figure 7 shows another example of input and output data for the inference unit, where Figure 7(A) is an example of input data and Figure 7(B) is an example of output data. Incidentally, the images captured by cameras 20-1 to 20-4 are captured from a fixed position at a specific angle of view, and during a rugby match, there are many contact plays and scrums between players. As a result, in the images captured by cameras 20-1 to 20-4, as shown in Figure 7(A), it is often the case that other players are positioned between a player P1A in a specific state in the captured image and the camera. In general, when detecting objects in an image, if the image of a specific object in the image (e.g., a player in a specific state) is hidden by an object in front of the specific object (e.g., another player) (hereinafter referred to as "occlusion"), the detection accuracy of the specific object decreases significantly. Therefore, if occlusion occurs, the detection accuracy of player P1A in a specific state may decrease.

[0047] In this embodiment, taking the above points into consideration, images containing occlusion are actively used as training data for training the trained model in the inference unit 44, thereby maintaining the detection accuracy of athlete P1A in a specific state in captured images containing occlusion. As a result, even when an captured image containing occlusion, as shown in Figure 7(A), is input, it is possible to output captured image output information with an additional frame 50 surrounding the fallen athlete P1A, as shown in Figure 7(B).

[0048] By the way, in order to obtain a trained model that can accurately detect a fallen athlete P1A even in captured images that include occlusion, it is necessary to prepare multiple captured images as training data in which other athletes are positioned in front of the fallen athlete P1A. However, collecting multiple such images is not practical because it is time-consuming and expensive. Therefore, in this embodiment, it is preferable to generate the aforementioned captured images using data augmentation.

[0049] Data augmentation can be achieved, for example, by the following method: First, an image containing a fallen athlete P1A is prepared. Next, images of one or more athletes P1 are placed on the aforementioned image so as to overlap with the fallen athlete P1A, and stored as input images. Then, the above process is repeated multiple times, slightly changing the placement of the images of one or more athletes P1, thereby expanding the amount of data. This type of data augmentation is particularly effective in reducing the cost of collecting training data.

[0050] By using the detection unit 42, which includes the inference unit 44 described above, it is possible to automatically and in real time identify athletes in a specific state within the captured images taken by cameras 20-1 to 20-4.

[0051] Returning to the explanation of Figure 3, the output unit 43 can be mainly implemented by the communication interface 35, and outputs the detection results detected by the detection unit 42, specifically information about a competitor in a specific state, as output information to the terminal device 10.

[0052] Furthermore, the output unit 43 can also add additional information to the output information output to the terminal device 10. For example, as mentioned above, athletes who suffer a concussion tend to maintain a specific state for a relatively long period of time. In other words, considering the points mentioned above, changing the output information according to the time the specific state is maintained is likely to make it easier for the situation of athletes in a specific state to be communicated to the competition officials P2 to P4. Considering this point, the image analysis device 30 according to this embodiment may also include a timer 47 in addition to the configuration described above. The output unit 43 should then output to the terminal device whether or not there are athletes whose specific state has been maintained for a predetermined time. It should be noted that there are no particular limitations on how the competition officials P2 to P4 are notified about athletes whose specific state has been maintained for a predetermined time. To give a specific example, the color of the frame 50 included in the image as output information, as shown in Figure 5(B) and Figure 7(B), can be changed according to the time that a fallen athlete P1A has maintained the fallen state, or pre-set text data can be added according to the time that the fallen state has been maintained.

[0053] As described above, the competition monitoring system 1 and image analysis device 30 according to this embodiment can automatically identify athletes in a specific state from images captured by cameras 20-1 to 20-4 and output information about the athletes in that specific state to the terminal device 10. As a result, competition personnel P2 to P4, etc., who possess the terminal device 10 can accurately and in real time know that an athlete is in a specific state, and can reliably grasp and identify the occurrence of a concussion. Therefore, the competition monitoring system 1 and image analysis device 30 can assist in the safe progress of the competition.

[0054] Next, the image analysis method according to this embodiment will be described. In the following description, the image analysis method according to this embodiment will be explained using an example where it is performed using the image analysis device 30 that constitutes the competition monitoring system 1 according to this embodiment described above. Furthermore, the following description shows the image analysis method when the competition monitoring system 1 is operating during a competition.

[0055] The image analysis method according to this embodiment can be implemented mainly by the image analysis device 30, and more specifically by the processor 31 within the image analysis device 30. In this regard, the image analysis method according to this embodiment can be provided in the form of a program or program product for causing the processor 31 to perform a predetermined operation, or in the form of a non-temporary computer-readable recording medium storing the program. Furthermore, the effects described below also serve to explain the effects achieved by the competition monitoring system 1 and image analysis device 30 of this embodiment described above.

[0056] Figure 8 is a flowchart showing an example of an image analysis method according to one embodiment of the present disclosure. The image analysis method according to this embodiment acquires images of multiple athletes participating in a specific contact sport (corresponding to step S1 described later), detects athletes in a specific state from among the athletes in the acquired images (corresponding to step S2 described later), and outputs the detection result (corresponding to step S4 described later). A detailed explanation follows below.

[0057] In the competition space 2 equipped with the aforementioned competition monitoring system 1, when a competition, such as a rugby match, begins, cameras 20-1 to 20-4 are activated and begin imaging the competition space 2. At least one of the competition officials P2 to P4 is provided with a terminal device 10 in advance.

[0058] When imaging by cameras 20-1 to 20-4 begins, the image analysis method according to this embodiment begins. First, the image acquisition unit 41 begins acquiring the images captured by cameras 20-1 to 20-4 (step S1). The images acquired by the image acquisition unit 41 are sent to the detection unit 42, where the detection unit 42 begins detecting athletes in a specific state (step S2). This detection of athletes in a specific state can be performed by inputting the captured images into a trained model in the inference unit 44.

[0059] Next, the detection unit 42 measures the time a competitor maintains a specific state based on the detection result, and updates the output information output from the output unit 43 according to the duration (step S3). Specifically, the output information is updated by adding information about the time a competitor maintains a specific state to the detection result of the detection unit 42. Note that the update in step S3 can be omitted. Then, the obtained output information is transmitted to the terminal device 10 (step S4).

[0060] The terminal device 10 automatically and virtually in real time displays information regarding the presence or absence of athletes in a specific condition and the duration of that condition. Competition officials P2-P4, who possess the terminal device 10, identify athletes who require attention based on the information displayed on the terminal device 10 and determine whether or not those athletes have suffered a concussion. This allows competition officials P2-P4 to take necessary actions such as interrupting the competition or protecting and treating specific athletes.

[0061] As described above, the image analysis method according to this embodiment can also automatically and in real time detect whether or not there is a competitor in a specific state within the captured images of cameras 20-1 to 20-4, and output the detection result to the terminal device 10.

[0062] In each of the above embodiments, the term "processor" refers to a processor in a broad sense, and includes general-purpose processors (e.g., CPUs) and dedicated processors (e.g., GPUs: Graphics Processing Units, ASICs: Application Specific Integrated Circuits, FPGAs: Field Programmable Gate Arrays, programmable logic devices, etc.).

[0063] Furthermore, the processor operations in each of the above embodiments may not be performed by a single processor, but may also be performed by multiple processors located in physically separate locations working together. Also, the order of the processor operations is not limited to the order described in each of the above embodiments, but may be changed as appropriate.

[0064] This disclosure is not limited to the embodiments described above, and can be implemented with various modifications without departing from the spirit of this disclosure. All such modifications are included in the technical concept of this disclosure. [Explanation of Symbols]

[0065] 1. Competition monitoring system 2. Competition space 10 Terminal devices 20-1~20-4 Camera 30 Image analysis device 41 Image acquisition unit 42 Detection unit 43 Output section 44 Reasoning part 47 Timer 50 frames P1 Competitor P1A: Athletes in a specific condition (falling state) P2-P4 Competition officials

Claims

1. Terminal device and Multiple imaging devices capable of imaging multiple athletes participating in a specific contact sport, The system includes an image analysis device that detects athletes in a specific state among the athletes captured by the plurality of imaging devices and outputs the detection result to the terminal device. Competition monitoring system.

2. The aforementioned specific state includes at least one of the following: falling, lying on one's side, and kneeling on the ground. The competition monitoring system according to claim 1.

3. The image analysis device includes an inference unit that identifies the athlete in the specific state within the captured image when the captured image is input to a trained model that has learned training data including images containing the athlete in the specific state. The competition monitoring system according to claim 1.

4. The inference unit comprises multiple pre-trained models, and integrates the inference results obtained by inputting the captured image into each of the multiple pre-trained models to identify the athlete in the specific state within the captured image. The competition monitoring system according to claim 3.

5. The image analysis device includes a timer and outputs to the terminal device whether or not there is a competitor whose specific state has continued for a predetermined period of time. The competition monitoring system according to claim 1.

6. An image acquisition unit that acquires images of multiple athletes participating in a specific contact sport, A detection unit for detecting athletes in a specific state among the athletes in the acquired captured image, The system includes an output unit that outputs the detection result from the detection unit. Image analysis device.

7. Using a computer, We obtained images of multiple athletes participating in a specific contact sport. Among the athletes in the acquired image, athletes in a specific state are detected. Output the detection results. Image analysis methods.

8. In a computer processor, We obtained images of multiple athletes participating in a specific contact sport. Among the athletes in the acquired image, athletes in a specific state are detected. Output the detection results, execute the process, program.

Images