Control device and control method

Abstract

This control device comprises: a data acquisition unit (121) that acquires data relating to reflected waves of radio waves transmitted from an antenna (11) of a base station (10) and reflected back from an object; and a prediction unit (123) that uses a model, in which the data acquired by the data acquisition unit and celestial sphere image data obtained by imaging the surroundings of the base station are used as explanatory variables and position information indicating the future position of the object is used as an objective variable, to predict the future position of the object from the data and the celestial sphere image data.

Description

Control device and control method

[0008] ,

[0007] , ,

[0001] The present invention relates to a control device and a control method.

[0002] Conventionally, a technique for estimating the position of a terminal connected to a base station has been known (for example, Patent Document 1). The method described in Patent Document 1 estimates the position of the terminal using the time-series information of the received power of millimeter waves and the time-series information of the shielding relationship between objects estimated from the image information.

[0003] Japanese Patent Application Laid-Open No. 2023-116958

[0004] However, in the technique described in Patent Document 1, it is a prerequisite that the object to be estimated can acquire the received power of millimeter waves, and the estimation target is limited.

[0005] One aspect of the present invention aims to accurately predict the future position of an object existing around a base station.

[0006] To solve the above problems, a control device according to one aspect of the present invention includes a data acquisition unit that acquires data regarding a reflected wave that bounces back when a radio wave transmitted from an antenna of a base station hits an object, and using the data acquired by the data acquisition unit and the omnidirectional image data obtained by photographing the periphery of the base station as explanatory variables and position information indicating the future position of the object as an objective variable, a prediction unit that predicts the future position of the object from the data and the omnidirectional image data.

[0007] To solve the above problems, a control method according to one aspect of the present invention includes a data acquisition step of acquiring data regarding a reflected wave that bounces back when a radio wave transmitted from an antenna of a base station hits an object, and a prediction step of predicting the future position of the object from the data and the omnidirectional image data using a model with the data and the omnidirectional image data obtained by photographing the periphery of the base station as explanatory variables and position information indicating the future position of the object as an objective variable.

[0008] Each aspect of the present invention may be implemented by a computer, in which case a control program that enables the implementation of the control device by a computer by operating the computer as each part (software element) of the control device, and a computer-readable recording medium on which the program is recorded, also fall within the scope of the present invention.

[0009] According to one aspect of the present invention, it is possible to accurately predict the future position of an object located around a base station.

[0010] This diagram illustrates communication and sensing at a base station. This is a graph showing the relationship between azimuth and elevation. This diagram schematically shows a prediction mode according to an embodiment of the present invention. This diagram illustrates an example of an effect. This diagram illustrates an example of the configuration of a wireless communication system according to an embodiment of the present invention. This is a block diagram illustrating an example of the hardware configuration of a server according to an embodiment of the present invention. This is a block diagram illustrating an example of the hardware configuration of a base station according to an embodiment of the present invention. This diagram illustrates the correspondence between sensing data and a 360-degree image. This diagram illustrates the correspondence between sensing data and a 360-degree image. This is a schematic diagram showing an example of the flow of each data. This is a flowchart illustrating an example of processing performed by a base station according to an embodiment of the present invention. This is a block diagram illustrating the configuration of a computer that can be used as a server, base station, etc.

[0011] Embodiments of the present invention will be described in detail below with reference to the drawings. In the drawings, identical or substantially identical components are denoted by the same reference numerals and will not be repeated in the description.

[0012] In the following explanation, "image" also includes the meaning of "data that represents an image (image data)."

[0013] (Application Examples) An application example of the present invention will be described with reference to Figure 1. Figure 1 is a diagram illustrating an embodiment in which communication and sensing are performed simultaneously. In recent years, there has been consideration of utilizing radio waves used in communication with terminal UE (User Equipment) for purposes other than communication. Such technologies are also called ISAC (Integrated Sensing and Communication), JCAS (Joint Communication And Sensing), etc. Conventionally, communication systems and sensing systems were developed and optimized separately. Compared to such conventional methods, by integrating communication and sensing, it becomes possible to share frequency bands, hardware, signal waveforms, etc., and improvements in frequency efficiency, reduction of hardware costs, assistance of communication through sensing, and utilization of sensing results as services can be expected.

[0014] One example of sensing that utilizes radio waves for purposes other than communication is estimating the position of an object (vehicle 50 in Figure 1) moving around a base station 10, as shown in Figure 1. Radio waves transmitted from the antenna 11 of the base station 10 are reflected back by the vehicle 50. By analyzing the reflected waves at the base station 10, the distance from the base station 10 to the vehicle 50, the azimuth angle of the vehicle 50 relative to the base station 10, and the elevation angle of the vehicle 50 relative to the base station 10 can be obtained. The azimuth angle is the angle that indicates the direction in which the object is located relative to the observation point (base station 10 in Figure 1), and is expressed as a relative angle from a reference direction on the horizontal plane (for example, north) with the observation point as the center. The elevation angle is the angle in the vertical direction from the ground surface, indicating the height at which the object is located relative to the observation point.

[0015] The azimuth angle and elevation angle will be explained with reference to Figure 2. Figure 2 is a graph showing the relationship between the azimuth angle and elevation angle, with the horizontal axis representing the azimuth angle and the vertical axis representing the elevation angle. Regions R1 to R3 are reflection points from an object, indicating the "direction" and "height" at which the object reflected the radio waves. The base station 10 can estimate the position of the vehicle 50 using these parameters. Such estimation is achieved by sensing the azimuth angle and elevation angle. In this embodiment, past sensing data, i.e., past azimuth angles and elevation angles, are used to predict the future position of an object. One method for predicting future positions using past sensing data is the well-known method using linear regression. The method using linear regression has the advantages of being simple and having a low computational load, but its simplicity leads to the problem of low accuracy.

[0016] Therefore, in this embodiment, a machine learning model is constructed to predict the future position of an object from past sensing data. Figure 3 is a schematic diagram showing how the model is used to predict the future position of an object from past sensing data. By inputting past azimuth and elevation angles sensed up to time t (..., time t-2, time t-1, time t) into the model, the positions P1 to P3 of each object at time t+1 are predicted. Here, two patterns are possible for constructing the model: (1) constructing a "general-purpose model" that can be used for any base station, and (2) constructing a "specialized model" that is specific to the target base station.

[0017] Pattern 1 (general-purpose model) allows the use of data collected from numerous different base stations for model training. Furthermore, it has the advantage of allowing a single trained model to be shared across all base stations. However, such a "general-purpose model" may not adequately account for base station-specific features, potentially resulting in lower prediction accuracy. "Base station-specific features" refer to characteristics that indicate the surrounding urban structure, such as surrounding buildings, road structures, street trees, and other features specific to the location where the base station is installed. Because these features vary significantly depending on the location, it is difficult for a single "general-purpose model" to cover the characteristics of all base stations, thus potentially leading to lower prediction accuracy.

[0018] Pattern 2 (specialized models) requires the collection of vast amounts of training data for each base station where the model will be used. Furthermore, a separate model is needed for each base station. With such specialized models, improved prediction accuracy can be expected by appropriately training the "base station-specific features." However, because the number of base stations is very large, building a model suitable for each individual base station is a significant burden in terms of cost and time.

[0019] Therefore, in this embodiment, a "general-purpose model" is constructed, and in addition to past sensing data (azimuth and elevation angles), base station-specific information is also input as input data to this "general-purpose model" in order to improve prediction accuracy. The "base station-specific information" referred to here is a 360-degree spherical image taken around the target base station. Such a 360-degree spherical image is also called an omnidirectional image or omni-surround image. By adding 360-degree spherical image data in addition to past sensing data (azimuth and elevation angles) as input data to the general-purpose model, the urban structure around the base station can be taken into consideration. As a result, the range and direction of movement of objects are limited, and the accuracy of predicting the future position of objects is improved.

[0020] In the following explanation, we will assume that only one general-purpose model is constructed, but this is not limited to this. Multiple general-purpose models may be constructed (for example, two or three).

[0021] In this embodiment, "estimating the position of an object" means "estimating the current and / or past position of an object" by analyzing reflected waves. However, the term "estimation" here may be used interchangeably with terms such as "prediction," "inference," and "calculation."

[0022] Furthermore, in this embodiment, "predicting the position of an object" means "predicting the future position of an object" using a machine learning model. However, the term "prediction" here may be used interchangeably with terms such as "estimation," "inference," and "computation."

[0023] Referring to Figure 4, an example of the effect obtained by predicting the future position of an object around the base station 10 will be explained. In the scene shown in Figure 4, vehicles 50 and 51 are moving objects moving around the base station 10, and vehicle 51 is assumed to be in a blind spot from the perspective of vehicle 50. Although not particularly limited, it is assumed that vehicles 50 and 51 are vehicles with autonomous driving capabilities and are operating autonomously. In the scene shown in Figure 4, the base station 10 predicts the future positions of vehicles 50 and 51 using a machine learning model based on past sensing data and 360-degree images. The base station 10 notifies vehicles 50 and 51 of the prediction results. This notification allows vehicle 50 to understand that "vehicle 51 will come out from around the corner," that is, that vehicle 51 will come out from a blind spot. To avoid dangers such as collisions, vehicle 50 can take measures such as slowing down. In this way, it is possible to prevent dangers by predicting future positions. In this explanation, we have used vehicles as an example of moving objects targeted for location prediction. However, moving objects are not limited to vehicles; any moving object present in the vicinity of base station 10, such as pedestrians, bicycles, and motorcycles, may also be included. Furthermore, flying objects such as drones may also be included.

[0024] (Example of the configuration of the wireless communication system 100) Figure 5 is a diagram illustrating an example of the configuration of the wireless communication system 100 according to this embodiment. As shown in Figure 5, the wireless communication system 100 has a configuration that includes base stations 10, 10-1, 10-2, ... and servers 20 that are communicatively connected to each of the base stations 10, 10-1, 10-2, ... The wireless communication system 100 is intended to be applied to a 5G network, but is not limited thereto. For example, the wireless communication system 100 may be applied to a future network following a 5G network, a 4G network, or a legacy network prior to a 4G network.

[0025] Communication between the server 20 and the base stations 10, 10-1, 10-2, ... is not particularly limited, but a virtualized virtual network using NFV (Network Function Virtualization) may be used, or a physical core network may be used. Alternatively, data may be processed locally using edge computing. In the following description, if it is not necessary to distinguish between the base stations 10, 10-1, 10-2, ..., base station 10 will be described as a representative.

[0026] The base station 10 communicates with a terminal UE located inside the cell, which is the wireless communication area (see Figure 1). In Figure 1, the terminal UE is illustrated as a smartphone that can be carried by a user, but it is not limited to this. The terminal UE may be a tablet, smartwatch, mobile phone, etc., in addition to a smartphone. Furthermore, the terminal UE may be an electronic device mounted on a vehicle, etc.

[0027] Server 20 is a server capable of controlling each of the base stations 10, 10-1, 10-2, ... and includes a controller 220. Such a controller 220 may be configured as a RIC (RAN Intelligent Controller). A RIC is a controller that manages and controls nodes such as RU (Radio Unit), DU (Distributed Unit), and CU (Central Unit) that constitute a 5G RAN.

[0028] (Hardware configuration of server 20) Figure 6 is a block diagram showing an example of the hardware configuration of server 20 according to this embodiment. As shown in Figure 6, server 20 includes a communication unit 210, a controller 220, a memory 230, and a storage unit 240.

[0029] The communication unit 210 is configured as a communication interface for communicating with base stations 10, 10-1, 10-2, ...

[0030] The controller 220 is composed of, for example, one or more processors. The processors are not particularly limited, but can be implemented as, for example, a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), an MPU (Micro Processor Unit), or an FPGA (Field-Programmable Gate Array). The controller 220 reads a program from the storage unit 240, expands it into the memory 230, and executes processing according to the program. Such a program is one that causes a computer to execute at least a part of the functions described below. The program may also perform its functions in combination with other programs already stored in the storage unit 240, or in combination with other programs implemented in other devices. The program may also be distributed to the server 20 by wireless communication. In this case, the controller 220 will expand the distributed program into the memory 230 and execute processing. In other words, the program does not necessarily have to be stored in the storage unit 240.

[0031] The memory 230 is a computer-readable recording medium and consists of at least one of the following: RAM (Random Access Memory), ROM (Read Only Memory), EPROM (Erasable Programmable ROM), EEPROM® (Electrically Erasable Programmable ROM), etc. Such memory 230 may also be called a register, cache, main memory, etc.

[0032] The storage unit 240 is a computer-readable recording medium, and is composed of, for example, an HDD (Hard Disk Drive), an SSD (Solid State Drive), etc. Alternatively, the storage unit 240 may be a portable recording medium such as a flexible disk, optical disk, compact disk, or Blu-ray® disk. The storage unit 240 is also sometimes referred to as an auxiliary storage device. The storage unit 240 stores information that is read, written, and referenced by the controller 220. In the example shown in Figure 6, the storage unit 240 stores a training dataset 241 and an inference model 242.

[0033] The training dataset 241 and the inference model 242 are described below. The training dataset 241 is a dataset containing training data (labeled data used in machine learning). Machine learning is a technique that learns the features contained in input data and generates a "model" that predicts the result corresponding to newly input data.

[0034] The inference model 242 is an AI (Artificial Intelligence) model generated by machine learning using the training dataset 241. In this embodiment, the inference model 242 takes "past sensing data" and "spherical image data" as input data and predicts the "future position of an object" based on this input data. Therefore, the output data of the inference model 242 is "position information indicating the future position of an object".

[0035] Here, we will explain the use of 360-degree images in machine learning. Unprocessed 360-degree images, used as input data for machine learning models, contain a large amount of information, making it difficult to extract desired features. To accurately predict the future position of an object, it is necessary to enable the model to recognize the urban structure surrounding the base station. To achieve this recognition, in this embodiment, the 360-degree image is preprocessed before being used in training. The preprocessing applied to the 360-degree image is semantic segmentation.

[0036] Semantic segmentation is a well-known image processing technique that classifies pixels in an image into different classes. By extracting urban structures through semantic segmentation, models can more easily recognize them. When semantic segmentation is performed on a 360-degree image, vehicles are classified into the "vehicle" class, buildings into the "building" class, roads into the "road" class, and vegetation into the "vegetation" class, on a pixel-by-pixel basis. In semantic segmentation, the same object is assigned the same label. For example, if a 360-degree image contains multiple vehicles, these vehicles are not distinguished, and all vehicles are classified into the "vehicle" class. By using a 360-degree image that has undergone semantic segmentation as input data, necessary features can be extracted, and the model can be trained efficiently.

[0037] The training data included in the training dataset 241 is a representation of the input data that is fed into the inference model 242, with the values ​​that the inference model 242 should output being considered as the ground truth data. Such input data can also be referred to as explanatory variables. The ground truth data can also be referred to as the target variable.

[0038] The inference model 242 predicts the future position of an object from past sensing data and spherical image data. Therefore, the training data included in the training dataset 241 is data that associates the position information indicating the future position of an object, obtained by applying past sensing data and spherical image data to input data representing past sensing data and spherical image data, with the ground truth data.

[0039] The training data described above is generated by associating the input data with the correct answer data and stored in the storage unit 240 as the training dataset 241.

[0040] Sensing data and 360-degree image data may be transmitted from each base station (base stations 10, 10-1, 10-2, ...) to the server 20 and stored in the storage unit 240.

[0041] (Example of Controller 220 Function) The controller 220 functions as a training data acquisition unit 221 and a learning unit 222 by reading a program from the storage unit 240 and executing the program using the memory 230 as a working area.

[0042] The training data acquisition unit 221, when generating the inference model 242, refers to the storage unit 240 to acquire the training dataset 241. The training data acquisition unit 221 outputs the acquired training dataset 241 to the learning unit 222.

[0043] The learning unit 222 generates an inference model 242 for predicting the future position of an object from sensing data and 360-degree image data by machine learning using the training dataset 241 acquired from the training data acquisition unit 221. As described above, the inference model 242 is a computational model that uses sensing data and 360-degree image data as explanatory variables and position information indicating the future position of an object as the objective variable. The machine learning algorithm is not particularly limited, but for example, a convolutional neural network (CNN) may be used. The inference model 242 generated by such machine learning is stored in the storage unit 240.

[0044] Note that, although the server 20 has been described here as the entity that generates the inference model 242, the present invention is not limited to this. The entity that generates the inference model 242 may be any device.

[0045] (Hardware Configuration of Base Station 10) FIG. 7 is a block diagram showing an example of the hardware configuration of the base station 10 according to the present embodiment. Although the illustration of the base stations 10-1, 10-2,... is omitted, the base stations 10-1, 10-2,... are also configured with the same hardware as the base station 10.

[0046] The base station 10 includes an antenna 11 that transmits radio waves, a communication unit 110, a controller 120, a memory 130, a storage unit 140, and a camera 150. Since the hardware configurations of the communication unit 110, the controller 120, the memory 130, and the storage unit 140 are the same as those of the communication unit 210, the controller 220, the memory 230, and the storage unit 240 of the server 20 described above, the description of the overlapping parts will be omitted.

[0047] The controller 120 reads a program from the storage unit 140 and executes the program using the memory 130 as a working area, thereby functioning as a data acquisition unit 121, an image processing unit 122, a prediction unit 123, and a notification unit 124.

[0048] The radio waves transmitted from the antenna 11 are reflected by an object existing around the base station 10 and bounce back. The data acquisition unit 121 analyzes the bounced-back reflected waves to acquire sensing data such as the distance from the base station 10 to the object, the azimuth angle of the object with respect to the base station 10, and the elevation angle of the object with respect to the base station 10. Such sensing data is stored in the storage unit 140.

[0049] The camera 150 has a wide-angle lens with a wide angle of view, a fisheye lens, etc., and is a camera capable of capturing an omnidirectional image. The omnidirectional image 70 captured by the camera 150 is stored in the storage unit 140.

[0050] ​

[0051] The prediction unit 123 predicts the future position of an object using the inference model 242 stored in the storage unit 140, based on the sensing data acquired by the data acquisition unit 121 and the 360-degree image from which semantic segmentation has been performed. The prediction unit 123 outputs the prediction result to the notification unit 124. Although this description assumes that the inference model 242 is stored in the storage unit 140, the inference model 242 does not necessarily have to be stored in the storage unit 140. If the inference model 242 is not stored in the storage unit 140, the inference model 242 can be used by referring to the storage unit 240 of the server 20. The inference model 242 may also be stored in any device other than the server 20, as long as the prediction unit 123 can access it.

[0052] The notification unit 124 notifies each object, for example, the vehicles 50 and 51 shown in Figure 4, of the prediction results. This notification makes it possible to prevent dangers such as collisions.

[0053] Figure 8 illustrates the correspondence between sensing data and a 360-degree image. As shown in Figure 8, the coordinate axes of the sensing data 60 and the coordinate axes of the 360-degree image 70 coincide or nearly coincide. More specifically, the pixel value 60a at coordinate (m, n) in the sensing data 60 coincides or nearly coincides with the object (vehicle 50) at the pixel at coordinate (m, n) in the 360-degree image 70. In contrast, the coordinate axes of a normal image, which is not a 360-degree image, do not coincide with the coordinate axes of the sensing data 60.

[0054] The advantages arising from the coordinate axes coinciding or nearly coinciding will be explained with reference to Figure 9. As mentioned above, when a CNN is adopted as the machine learning algorithm, it becomes easier to associate the sensing data 60-62 with the city structure in the 360-degree image 70 during the convolution operation, enabling effective learning. Note that the symbol 80 in Figure 9 indicates a convolution filter.

[0055] (Data flow during prediction) Figure 10 is a schematic diagram showing an example of the data flow when predicting the future position of an object.

[0056] As shown in Figure 10, the input data to the inference model 242 consists of the 360-degree image data after semantic segmentation shown in block 250 and the past sensing data (..., time t-2, time t-1, time t) shown in block 251.

[0057] The inference model 242 predicts the future position of an object from the 360-degree image data after semantic segmentation and past sensing data. The prediction result is output as position information indicating the future position of the object, as shown in block 253. The 360-degree image data after semantic segmentation and past sensing data are explanatory variables in the inference model 242, and the position information indicating the future position of the object is the dependent variable in the inference model 242.

[0058] (Processing Flow) Next, with reference to Figure 11, the processing flow performed by the base station 10 will be explained. Figure 11 is a flowchart showing an example of the processing performed by the base station 10.

[0059] In step S101, the camera 150 captures a 360-degree image of the area around the base station 10. The process then proceeds to step S102, where the controller 120 performs semantic segmentation on the 360-degree image acquired in step S101. This makes it easier for the inference model 242 to recognize the urban structure around the base station 10, improving prediction accuracy.

[0060] In step S103, the controller 120 analyzes the reflected waves that bounce back after the radio waves transmitted from the antenna 11 are reflected off the object, thereby acquiring sensing data such as the distance from the base station 10 to the object, the azimuth angle of the object relative to the base station 10, and the elevation angle of the object relative to the base station 10.

[0061] The process proceeds to step S104, where the controller 120 inputs the 360-degree image data after semantic segmentation acquired in step S102 and the sensing data acquired in step S103 into the inference model 242. The inference model 242 predicts the future position of objects from this data (step S105). The process proceeds to step S106, where the controller 120 notifies each object, for example, the vehicles 50 and 51 shown in Figure 4, of the prediction results. This notification makes it possible to prevent dangers such as collisions.

[0062] As described above, the control method according to this embodiment includes a data acquisition step (S103) of acquiring data (sensing data) relating to reflected waves (sensing data) that are reflected back when radio waves transmitted from the antenna 11 of the base station 10 hit an object, and a prediction step (S105) of predicting the future position of an object from the sensing data and the spherical image data, using a model (inference model 242) in which the sensing data and spherical image data of the area around the base station 10 are explanatory variables and position information indicating the future position of the object is the objective variable.

[0063] Note that the processing flow shown in the flowchart in Figure 11 is just one example, and steps may be deleted, new steps added, or the processing order rearranged as long as it does not deviate from the main point.

[0064] (Effects) As described above, the following effects can be obtained according to this embodiment.

[0065] The control device (for example, controller 120) includes a data acquisition unit 121 that acquires data (sensing data) relating to reflected waves that bounce back when radio waves transmitted from the antenna 11 of the base station 10 hit an object, and a prediction unit 123 that uses the data acquired by the data acquisition unit 121 and 360-degree spherical image data taken around the base station 10 as explanatory variables, and position information indicating the future position of the object as the objective variable, to predict the future position of the object from the data and 360-degree spherical image data.

[0066] With the above configuration, by adding 360-degree spherical image data in addition to past sensing data (azimuth and elevation) as input data to the inference model 242, the urban structure around the base station 10 can be taken into consideration. This makes it easier for the inference model 242 to recognize the urban structure around the base station 10, and since the range and direction of movement of objects are limited, the accuracy of predicting the future position of objects is improved.

[0067] Furthermore, the data relating to reflected waves (sensing data) may include at least the azimuth angle of the object relative to the base station 10, and the elevation angle of the object relative to the base station 10.

[0068] According to the above configuration, the radio waves used for communication with the terminal UE can be utilized for purposes other than communication, which is expected to improve frequency efficiency, reduce hardware costs, and enable new services.

[0069] Furthermore, the data related to reflected waves (sensing data) may also include the distance from the base station 10 to the object. In addition, the distance from the base station 10 to the object may be used as training data when generating the model.

[0070] The object is a moving object that moves around the base station 10. The prediction unit 123 may perform semantic segmentation on the 360-degree image data and then input the 360-degree image data to the inference model 242.

[0071] With the above configuration, the inference model 242 receives spherical image data from which semantic segmentation has been performed, making it easier for the inference model 242 to recognize the urban structure around the base station 10, and improving the accuracy of predicting the future position of objects.

[0072] The controller 120 may also further include a notification unit 124 that notifies the object of the prediction result predicted by the prediction unit 123 (see Figure 7).

[0073] With the above configuration, the prediction results are notified to each object, for example, the vehicles 50 and 51 shown in Figure 4, so that the vehicles 50 and 51 can understand each other's future positions. This makes it possible to prevent dangers such as collisions.

[0074] [Other Embodiments] As described above, the camera 150 acquires a 360-degree image when predicting the future position of an object using the inference model 242, but this is not the only possible embodiment. Since the urban structure around the base station 10 is not expected to change frequently, one 360-degree image per base station is sufficient. Therefore, for example, one 360-degree image may be acquired when the base station 10 is installed, and this single 360-degree image may be used. With this type of operation, it becomes unnecessary to install the camera 150 at the base station 10. In this case, the processes of steps S101 to S102 described in Figure 11 can also be omitted.

[0075] (Examples of hardware configuration and software implementation) The control blocks of server 20 and base stations 10, 10-1, 10-2, ... (especially the controller 220 and the various parts included in controller 120, etc.) may be implemented by logic circuits (hardware) formed on integrated circuits (IC chips), etc., or by software using a CPU (Central Processing Unit). In the latter case, server 20 and base stations 10, 10-1, 10-2, ... may be configured using a computer (electronic computer). For the sake of explanation, server 20 and base stations 10, 10-1, 10-2, ... will be collectively referred to as "server 20, etc." below.

[0076] Figure 12 is a block diagram illustrating the configuration of a computer 910 that can be used as a server 20, etc. The computer 910 comprises an arithmetic unit 912, a main memory 913, an auxiliary storage device 914, and an input / output interface 915, all connected to each other via a bus 911. The arithmetic unit 912, the main memory 913, and the auxiliary storage device 914 may be, for example, a CPU, RAM (random access memory), a solid-state drive, or a hard disk drive, respectively. An input device 920 for a user to input various information into the computer 910, and an output device 930 for the computer 910 to output various information to the user are connected to the input / output interface 915. The input device 920 and the output device 930 may be built into the computer 910 or connected to the computer 910 (externally). For example, the input device 920 may be a button, keyboard, mouse, touch sensor, etc., and the output device 930 may be a lamp, display, printer, speaker, etc. Alternatively, a device having both input device 920 and output device 930 functions, such as a touch panel that integrates a touch sensor and a display, may be used. The communication interface 916 is an interface for the computer 910 to communicate with an external device.

[0077] The auxiliary storage device 914 stores an information processing program that causes the computer 910 to operate as a server 20, etc. The arithmetic unit 912 then loads the information processing program stored in the auxiliary storage device 914 onto the main memory 913 and executes the instructions contained in the information processing program, thereby causing the computer 910 to function as various parts of the server 20, etc. The recording medium used by the auxiliary storage device 914 to record information such as the information processing program can be any "non-temporary tangible medium" that is readable by a computer, such as tape, disk, card, semiconductor memory, or programmable logic circuit.

[0078] Furthermore, the present invention may employ a configuration in which the computer 910 functions using a program recorded on an external recording medium, or a program supplied to the computer 910 via any transmission medium (such as a communication network or broadcast waves). The present invention can also be realized in the form of a data signal embedded in a carrier wave, in which the above program is embodied by electronic transmission.

[0079] The present invention is not limited to the embodiments described above, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the present invention.

[0080] 100 Wireless communication system 10 Base station 11 Antenna 20 Server 60 Sensing data 70 Spherical image 120 Controller 121 Data acquisition unit 122 Image processing unit 123 Prediction unit 124 Notification unit 242 Inference model

Claims

1. A control device comprising: a data acquisition unit that acquires data relating to reflected waves that bounce back when radio waves transmitted from a base station antenna strike an object; and a prediction unit that predicts the future position of an object from the data and the spherical image data, using a model in which the data acquired by the data acquisition unit and spherical image data of the area around the base station are used as explanatory variables and position information indicating the future position of the object is used as the objective variable.

2. The control device according to claim 1, wherein the data includes at least the azimuth angle of the object with respect to the base station and the elevation angle of the object with respect to the base station.

3. The control device according to claim 1, wherein the object is a moving object that moves around the base station, and the prediction unit performs semantic segmentation on the 360-degree image data and then inputs the 360-degree image data to the model.

4. The control device according to any one of claims 1 to 3, further comprising a notification unit for notifying the object of the prediction result predicted by the prediction unit.

5. A control method comprising: a data acquisition step of acquiring data relating to reflected waves that bounce back when radio waves transmitted from a base station antenna strike an object; and a prediction step of predicting the future position of an object from the data and the spherical image data, using a model in which the data and spherical image data taken around the base station are explanatory variables and position information indicating the future position of the object is the objective variable.

Images