Data selection method, data selection device, and data selection program

Abstract

This data selection device acquires the position of a first sensor which detected an event; acquires the distances from the first sensor of each among a plurality of sensors installed in a vehicle; selects either sensors among the plurality of sensors for which the distance is a threshold value or less, or a determined number of sensors from among the plurality of sensors extracted in order of shortness of the distance, as second sensors; and uses sensor data obtained from the second sensors and sensor data obtained from the first sensor to execute a task regarding the event.

Description

Data Selection Method, Data Selection Device, and Data Selection Program

[0001] The present invention relates to a data selection method, a data selection device, and a data selection program.

[0002] Recently, connected cars equipped with cameras and sensors have become popular. Also, as a technology related to digital twins, a technology that maps the positions of vehicles, people, accidents, etc. to world coordinates in real time based on information obtained from connected cars is expected.

[0003] To realize such a technology, AI inference processing such as image recognition using AI is used. Conventionally, as a method for reducing the processing amount and energy consumption in AI inference processing, event-driven inference has been proposed (see, for example, Non-Patent Document 1). In event-driven inference, for example, normally, a moving image with a low frame rate is the analysis target, and when a specific event (for example, detection of a person shown in an image) occurs, the analysis target is switched to a moving image with a high frame rate.

[0004] Keita Mikami, Xu Shi, Noriyoshi Inoue, Ryosuke Fubayashi, Yoshinori Matsuo, Ikue Yamazaki, NTT Technical Journal, "Software Technology in the Data-Centric Era, A High-Resolution Multi-Camera Analysis Infrastructure to Support Future Smart Cities", [online], [searched on September 9, 2024], Internet (https: / / journal.ntt.co.jp / article / 17188)

[0005] However, the conventional technology has a problem that efficient inference processing may not be executable in some cases.

[0006] For example, when images are obtained from a plurality of cameras, event-driven inference can be realized by performing inference using images obtained from a small number of cameras normally and performing inference using images obtained from many cameras when an event occurs. On the other hand, from the perspective of processing cost, it is desirable that the amount of data to be processed is small even when an event occurs.

[0007] In contrast, Non-Patent Document 1 does not describe a method for selecting the minimum necessary number of cameras from among multiple cameras to acquire images when an event occurs.

[0008] In other words, conventional technology does not consider which camera's images are most likely to accurately accomplish the target task. To put it another way, conventional technology does not consider which camera is most likely to contain the information necessary for performing task inference.

[0009] The present invention has been made in view of the above, and aims to perform efficient inference processing.

[0010] To solve the above-mentioned problems and achieve the objective, the data selection method according to the present invention is characterized in that it obtains the position of a first sensor that detected an event, obtains the distance between each of a plurality of sensors mounted on the vehicle and the first sensor, selects as a second sensor either a sensor from the plurality of sensors whose distance is less than or equal to a threshold, or a predetermined number of sensors extracted from the plurality of sensors in ascending order of distance, and has a computer perform a process to execute a task related to the event using the sensor data obtained from the second sensor and the sensor data obtained from the first sensor.

[0011] According to the present invention, it becomes possible to perform efficient inference processing.

[0012] Figure 1 is a diagram showing an example configuration of an inference system according to the first embodiment. Figure 2 is a diagram showing an example configuration of an information processing terminal. Figure 3 is a diagram showing an example configuration of a server. Figure 4 is a flowchart showing the processing flow of the information processing terminal. Figure 5 is a flowchart showing the processing flow of the server. Figure 6 is a diagram illustrating an embodiment. Figure 7 is a diagram illustrating an embodiment. Figure 8 is a diagram showing an example of a computer that executes a data selection program.

[0013] Hereinafter, one embodiment of the present invention will be described in detail with reference to the drawings. However, the present invention is not limited by this embodiment. Furthermore, in the drawings, the same parts are denoted by the same reference numerals.

[0014] [First Embodiment] The configuration of the inference system will be explained using Figure 1. Figure 1 is a diagram showing an example of the configuration of an inference system according to the first embodiment. As shown in Figure 1, the inference system 1 has an information processing terminal 10_1, an information processing terminal 10_2, an information processing terminal 10_3, an information processing terminal 10_4, and a server 20. The server 20 is an example of a data selection device. Hereafter, information processing terminals 10_1, 10_2, 10_3, and 10_4 may be referred to simply as information processing terminal 10 without distinction. Inference refers to processing performed by a trained model, but in this embodiment, the inference does not need to be a machine learning-based process and may be implemented in a system based on heuristically determined signal processing.

[0015] The number of information processing terminals 10 is not limited to those shown in Figure 1. Each information processing terminal 10 has a sensor. In this embodiment, the sensor is assumed to be a camera. The server 20 acquires images captured by the cameras via the information processing terminals 10. The image is an example of sensor data.

[0016] Server 20 performs inference processing such as object detection using the acquired images. The cameras from which images are acquired include not only fixed cameras installed at fixed points, such as surveillance cameras, but also mobile cameras such as in-vehicle cameras.

[0017] The sensor is not limited to a camera; it may also be a microphone mounted on a handset such as a smartphone, or a LiDAR sensor. For example, the server 20 can perform speech recognition using acoustic signals collected from the microphone, and estimate point clouds and depth maps based on depth obtained from the LiDAR sensor.

[0018] The configuration of the information processing terminal will be explained using Figure 2. Figure 2 is a diagram showing an example of the configuration of an information processing terminal. As shown in Figure 2, the information processing terminal 10 receives a camera image and the camera's position and direction as input. The information processing terminal 10 also outputs the camera's position and direction, and the position, velocity, and feature quantities of the detected event. Note that the term "event" may be rephrased as "event".

[0019] The camera image may be a video file or a streaming data format. Furthermore, the camera image, camera position, and orientation are synchronized in time through appropriate interpolation of update rates, etc.

[0020] Furthermore, in the case of a moving camera, the camera's position and orientation are represented by GNSS / IMU, etc., and self-position correction is performed as needed. In the case of a fixed camera, the camera's position and orientation are represented by the pan-tilt-zoom (PTZ) preset number, etc.

[0021] The information processing terminal 10 includes a data input unit 11, a determination unit 12, a detection unit 13, and an estimation unit 14. The information processing terminal 10 also stores an AI model 15, an event detection state 16, and camera hardware information 17.

[0022] The data input unit 11 accepts data input. The determination unit 12 determines whether or not to perform event detection (an example of inference). If the determination unit 12 determines that no event detection is performed, it notifies the data input unit 11 that image input is unnecessary. For example, events include at least one of the following: accidents, damage or malfunction of traffic lights, road collapses, objects on the road, dirt on the road surface, traffic congestion, emergency vehicles, vehicles stopped at dangerous locations such as highways, and slow-moving vehicles.

[0023] The detection unit 13 detects events by inputting the received data into the AI ​​model 15. The detection unit 13 stores the event detection results (movement trajectory, duration, etc.) in the event detection state 16. Event detection is, for example, object detection in an image. The event detection results may also be feature quantities represented by vectors or the like, depending on the processing content of the AI ​​model 15.

[0024] The estimation unit 14 refers to the camera hardware information 17 and estimates the position and / or velocity of the event. The estimation unit 14 may estimate the position and velocity in one shot. The camera hardware information 17 is, for example, the focal length of the camera lens if the camera performs monocular distance measurement. A depth map may be used as the camera hardware information 17, or distances obtained from a stereo camera may be used.

[0025] If no event is detected, or if no event is detected, the information processing terminal 10 may output only the camera's position and direction.

[0026] The server configuration will be explained using Figure 3. Figure 3 is a diagram showing an example of the server configuration. As shown in Figure 3, the server 20 receives data output from the information processing terminal 10 as input. The server 20 accepts input of the camera's position and direction, as well as the position, velocity, and feature quantities of events. The server 20 can accept data input from multiple information processing terminals 10.

[0027] Furthermore, the server 20 issues instructions to the information processing terminal 10 to start and stop event detection. The determination unit 12 of the information processing terminal 10 may determine whether or not to perform event detection in response to the instructions from the server 20.

[0028] The server 20 includes a camera status management unit 21, an event management unit 22, a search unit 23, and an instruction unit 24. It also stores camera status information 25, event information 26, and search condition information 27.

[0029] The camera status management unit 21 adds, updates, deletes, etc., the input camera position and orientation to the camera status information 25. In other words, the camera status information 25 includes the camera position and orientation.

[0030] The event management unit 22 integrates event information such as the event's location, velocity, and features, and adds, updates, and deletes this information from the event information 26. In other words, the event information 26 includes the event's location, velocity, and features. The event management unit 22 also synchronizes the time between camera images as needed.

[0031] The event information 26 may be transmitted to an external APP (application) that utilizes a digital twin. For example, the external APP may use the event information 26 to analyze traffic congestion, predict collisions, detect abnormal vehicles, etc., and notify the user of the results. The external APP may also notify the user of information such as the distance and direction between the abnormal vehicle and the user, and the route to take to avoid it, based on the traffic congestion information, collision prediction, and abnormal vehicle information.

[0032] The search unit 23 searches for a camera to which it will instruct the start or stop of event detection based on the search condition information 27. The instruction unit 24 determines whether or not to issue an instruction to the searched camera based on the search condition information 27, and transmits an instruction to start or stop event detection according to the determination result. The search condition information 27 includes conditions for search and determination. The conditions are, for example, "a camera within a certain distance on a map coordinate system centered on the location where the event occurred," "a camera that is capturing an image of the direction of the camera that detected the event," and "a camera that is capturing an image from the road closest to the location of the camera that detected the event." The search unit 23 may search for a camera that satisfies one or some of the conditions listed here, or it may search for a camera that satisfies all of the conditions.

[0033] Furthermore, the search unit 23 may further narrow down the selection to cameras with high performance, such as resolution, from among the cameras that meet the criteria. In addition, the search unit 23 may exclude cameras that are too close to other selected cameras, and cameras whose field of view overlaps too much (the volume or area of ​​the overlapping region is greater than or equal to a threshold), from among the cameras that meet the criteria.

[0034] The processing flow of the information processing terminal 10 will be explained using Figure 4. Figure 4 is a flowchart showing the processing flow of the information processing terminal. As shown in Figure 4, first the data input unit 11 receives input of the camera image, camera position, and camera direction (step S101).

[0035] The determination unit 12 determines whether or not to perform event detection based on the received information (step S102). If it is determined that event detection should be performed (step S103; Yes), the detection unit 13 performs event detection (step S104). The information processing terminal 10 then transmits the location of the event estimated by the estimation unit 14, the speed of the event, and the feature quantities obtained by the event detection, along with the camera's position and camera's direction, to the server 20 (step S105). The information transmitted along with the feature quantities may also include the camera's position and the direction in which the camera is capturing images, the direction of the target camera relative to a predetermined position (a predetermined camera, etc.), etc.

[0036] If it is determined that event detection should not be performed (step S103; No), the information processing terminal 10 terminates the process shown in Figure 4. The information processing terminal 10 continues to repeat the process shown in Figure 4 semi-permanently at a predetermined frequency (for example, 30 times per second).

[0037] Furthermore, if it is determined that event detection should not be performed (step S103; No), the information processing terminal 10 may transmit the camera's position and orientation to the server 20. In this case, the information processing terminal 10 may transmit the information less frequently, rather than transmitting it every time it is determined that event detection should not be performed.

[0038] Figure 5 will be used to explain the processing flow of the server 20. Figure 5 is a flowchart showing the processing flow of the server. The server 20 receives inputs for the location of an event, the velocity of an event, the feature quantities of an event, the camera's position, and the camera's direction (step S201). The camera state management unit 21 acquires the camera's state (position and direction) from the received information (step S202). The event management unit 22 performs integration processing of the received events (step S203). In other words, the event management unit 22 integrates common event information acquired from multiple information processing terminals 10.

[0039] The search unit 23 searches for the camera to be processed based on the camera status and the integrated event information (step S204). The instruction unit 24 then determines whether or not event detection is necessary for the searched camera (step S205), and instructs the searched camera to start or stop event detection based on the determination result (step S206).

[0040] An embodiment implemented by the inference system 1 will be described using Figures 6 and 7. The embodiment is implemented by the server 20 issuing instructions to each information processing terminal 10. The camera is integrated with the information processing terminal 10. For example, when the camera performs inference processing, it means that the information processing terminal 10 performs inference processing. For example, the server 20 can control which camera it issues what instructions to by setting the search condition information 27.

[0041] [Example 1] Fig. 6 is a diagram for explaining an example. The fixed cameras 10_11, 10_12, 10_13, and 10_14 in Fig. 6 are installed on urban roads and perform continuous observation. The cameras for continuous observation may be made more labor-saving, such as changing the frame rate according to the time zone, or may change the PTZ and perform preset tours.

[0042] An example of the process when the camera for continuous observation is the fixed camera 10_12 will be described. The fixed camera 10_12 is integrated with the information processing terminal 10. However, the function of the information processing terminal 10 may be configured as part of the server 20. In that case, the information processing terminal 10 may be software included in the server 20.

[0043] In addition to object detection such as bicycles and emergency vehicles, license plate detection, speed measurement of vehicles registered in the blacklist, object tracking, and determination of stopped, low-speed, and speeding vehicles by speed estimation, the fixed camera 10_12 may also perform processing with a small amount of computation, such as inference processing for anomaly detection based on a model learned by unsupervised learning such as GAN (Generative adversarial networks) or autoencoder. When the fixed camera 10_12 performs inference processing by GAN, the result of inputting an image into the learned discriminator may be used as a determination of whether it is abnormal. When the fixed camera 10_12 performs inference processing by autoencoder, it may determine whether it is abnormal based on the difference between the reconstructed image obtained using an autoencoder learned only with normal images and the normal image. Also, the fixed camera 10_12 may perform anomaly detection by inputting an image into a multimodal LLM and asking whether there is an anomaly. The fixed camera 10_12 may perform the conventional event-driven inference processing.

[0044] When an event to be vigilant about is detected in the inference processing by the fixed camera 10_12, the server 20 searches for a moving camera in the vicinity of the fixed camera 10_12 and instructs to acquire an image (video or stream).

[0045] Here, the server 20 searches for cameras mounted on vehicles that are within a certain threshold distance from the location where the event to be monitored occurred (for example, the location 400 where the accident occurred in Figure 6). The server 20 may also perform the search considering the vehicle's attitude and direction of travel. The server 20 may also consider the direction or area being captured by the camera mounted on the vehicle. The search conditions here may be the same as the search conditions shown in the description of the search unit 23 above. That is, the server 20 may search for cameras that satisfy one or some of the following conditions: "a camera within a certain distance range on a map coordinate system centered on the location where the event occurred," "a camera that is capturing images of the camera that detected the event," "a camera capturing images in the direction of the camera that detected the event," or "a camera capturing images from the road closest to the location of the camera that detected the event." Alternatively, the server 20 may search for cameras that satisfy all the conditions. Furthermore, the server 20 may further narrow down the search from the cameras that meet the conditions to cameras with high performance such as resolution. Furthermore, the server 20 may exclude cameras that are too close to other selected cameras, and cameras whose field of view overlaps too much (the volume or area of ​​the overlapping region is greater than or equal to a threshold), from among the cameras that meet the criteria.

[0046] Server 20 performs inference processing using the searched mobile cameras. The mobile cameras 10_22 may perform object detection in the same way as the fixed cameras 10_12. However, the functions of the information processing terminal 10, which is the main processing unit for object detection, may be included in Server 20.

[0047] Furthermore, when the information processing terminal 10 performs unsupervised anomaly detection using the fixed camera 10_12, the inference processing of the fixed camera 10_12 may be deepened by applying a supervised algorithm, which has learned object detection and accident type using the mobile camera 10_22, to the image at point 400.

[0048] Further, the mobile camera 10_22 may output data obtained by trimming the thumbnail of the image at the location 400. Further, when the information processing terminal 10 uses the mobile camera 10_22 to detect an event different from that using the fixed camera 10_12, the information processing terminal 10 may relay the detection result of the event to other cameras or the like.

[0049] Note that the inference process using the mobile camera 10_22 may have a larger amount of computation than the inference process using the fixed camera 10_12. Further, while the inference process using the fixed camera 10_12 is a process capable of ensuring the estimation accuracy with poor information, the inference process using the mobile camera 10_22 may be such that the estimation accuracy decreases unless the amount of information is rich.

[0050] The server 20 integrates (merges) the event detected by the fixed camera 10_12 and the event detected by the mobile camera 10_22.

[0051] When there is a difference between the event detected by the fixed camera 10_12 and the event detected by the mobile camera 10_22, the server 20 may hold the detection result of the camera closer to the location 400 and discard the other, or may identify the detection results of each other (for example, the movement trajectory of an object). For example, the server 20 is not limited to the proximity to the location 400, and may be based on the presence or absence of overlap at the time of object detection (since the position estimation accuracy of an object with an obstacle in front is likely to drop and is difficult to hold), the confidence level of object detection, the stability or stay time of the movement trajectory, the speed or acceleration (however, abnormal values that cannot occur on the road are discarded), and the specifications of the video source (for example, hold the 4K camera rather than the SD camera) to determine whether to hold the detection result.

[0052] When the mobile camera 10_22 does not detect an event for a certain period of time, or when it can be determined from the integration result that the detected event is not an event to be vigilant, the server 20 instructs the mobile camera 10_22 to stop detecting the event.

[0053] In this way, the server 20 obtains the position of the fixed camera 10_12 that detected the event, and obtains the distance between each of the multiple mobile cameras and the fixed camera 10_12. A mobile camera is an example of a sensor mounted on a vehicle. The server 20 may also select a mobile camera (for example, mobile camera 10_22) from among the multiple mobile cameras whose distance is below a threshold. The server 20 uses the images obtained from the selected mobile camera and the images obtained from the fixed camera to perform a task related to the event (for example, integrating the detection results).

[0054] The server 20 may also select a predetermined number of mobile cameras from among multiple mobile cameras, sorted by the smallest distance between them and the fixed cameras 10_12.

[0055] Furthermore, the server 20 can instruct the fixed camera and each of the multiple mobile cameras to start and stop event detection. The server 20 also performs a task of integrating the event detection results based on images obtained from the selected mobile camera with the event detection results based on images obtained from the fixed camera. This allows the server 20 to centrally control the cameras and process the results. Integration is intended to determine whether the detection results from the fixed camera are correct, and to perform detailed detection of results that were coarsely detected by the fixed camera. Coarse detection is, for example, the detection of anomalies on a road. Detailed detection is, for example, the detection of a collision between two cars on a road.

[0056] [Example 2] Using Figure 7, an example will be described that corresponds to a case in which a fixed camera for continuous observation is not used, or a case in which the event occurs outside the field of view of the fixed camera. Figure 7 is a diagram illustrating the example.

[0057] First, Server 20 divides a real-world map into a grid. The grid cells may overlap. Alternatively, Server 20 may divide the map into circular areas instead of a grid. Furthermore, Server 20 may make the divisions variable for each location to match the field of view of a fixed camera, or to vary according to changes in time of day, such as day and night.

[0058] Server 20 selects (extracts) a camera from among multiple cameras to correspond to each cell in the grid. The selection criteria may be based on the size of the overlap area between the camera's field of view and the cell (for example, the largest overlap), or it may be random.

[0059] Furthermore, server 20 may prioritize selecting cameras with fewer blind spots, or it may prioritize selecting mobile cameras from vehicles (connected cars) that have high hardware performance. Server 20 may also select a number of cameras proportional to the number of cameras present in the cell. Additionally, server 20 may use a generation AI to select cameras.

[0060] In the example shown in Figure 7, a grid forms a first rectangular cell with vertices at (x10, y20), (x20, y20), (x10, y30), and (x20, y30). Server 20 picks a fixed camera 10_12 for the first cell.

[0061] The selected fixed camera 10_12 detects an event at point 400. At this time, the server 20 adds a new second cell (rectangular area) designated as point 400. Then, it selects mobile cameras 10_21 and 10_22 for the second cell. Then, mobile cameras 10_21 and 10_22 perform event detection.

[0062] The inference processing of the cameras picked for the second cell provides more detailed inference results for events at location 400.

[0063] Furthermore, the inference method used may be switched between the fixed camera 10_12 picked for the first cell and the mobile cameras 10_21 and 10_22 picked for the second cell. For example, the fixed camera 10_12 may discover the location of the falling object using an anomaly detection method such as GAN, while the mobile cameras 10_21 and 10_22 may identify the type of falling object using object recognition.

[0064] In this way, the server 20 sets a region centered on the location of an event detected by a fixed camera, and extracts a camera corresponding to the region from among the fixed camera and multiple mobile cameras according to the amount by which the sensing range (e.g., the camera's field of view) overlaps with the region, and instructs the extracted camera to perform event detection in the region. This allows the server 20 to select the optimal camera for detection in each region.

[0065] [Program] In one embodiment, the data selection device (server 20) can be implemented by installing a data selection program that performs the above processing as packaged software or online software on a desired computer. For example, by having the above data selection program run on an information processing device, the information processing device can be made to function as a data selection device. The information processing device referred to here includes desktop or notebook personal computers. In addition, the information processing device also includes smartphones, tablet terminals, etc.

[0066] Figure 8 shows an example of a computer running a data selection program. Computer 1000 has, for example, memory 1010 and CPU 1020. Computer 1000 also has a hard disk drive interface 1030, a disk drive interface 1040, a serial port interface 1050, a video adapter 1060, and a network interface 1070. These components are connected by a bus 1080.

[0067] Memory 1010 includes ROM (Read Only Memory) 1011 and RAM (Random Access Memory) 1012. ROM 1011 stores, for example, a boot program such as BIOS (Basic Input Output System). The hard disk drive interface 1030 is connected to the hard disk drive 1090. The disk drive interface 1040 is connected to the disk drive 1100. For example, a removable storage medium such as a magnetic disk or optical disk is inserted into the disk drive 1100. The serial port interface 1050 is connected to, for example, a mouse 1110 and a keyboard 1120. The video adapter 1060 is connected to, for example, a display 1130.

[0068] The hard disk drive 1090 stores, for example, an OS 1091, an application program 1092, a program module 1093, and program data 1094. That is, the program that defines each process of the data selection device is implemented as a program module 1093 in which executable code for a computer is written. The program module 1093 is stored, for example, in the hard disk drive 1090. For example, a program module 1093 for performing the same processes as the functional configuration in the data selection device is stored in the hard disk drive 1090. Note that the hard disk drive 1090 may be replaced by an SSD.

[0069] Furthermore, the configuration data used in the processing of the above-described embodiment is stored as program data 1094 in, for example, memory 1010 or hard disk drive 1090. The CPU 1020 then reads the program module 1093 and program data 1094 stored in memory 1010 or hard disk drive 1090 into RAM 1012 as needed and executes the processing of the above-described embodiment.

[0070] Furthermore, the program module 1093 and program data 1094 are not limited to being stored in the hard disk drive 1090; for example, they may be stored in a removable storage medium and read by the CPU 1020 via a disk drive 1100 or the like. Alternatively, the program module 1093 and program data 1094 may be stored in another computer connected via a network (LAN (Local Area Network), WAN (Wide Area Network), etc.). The program module 1093 and program data 1094 may then be read by the CPU 1020 from the other computer via a network interface 1070.

[0071] Although embodiments applying the invention made by the present inventors have been described above, the present invention is not limited by the descriptions and drawings that constitute part of the disclosure of the present invention in this embodiment. That is, all other embodiments, examples, and operational techniques made by those skilled in the art based on this embodiment are included in the scope of the present invention.

[0072] 1. Inference System 10, 10_1, 10_2, 10_3, 10_4 Information Processing Terminal 10_11, 10_12, 10_13, 10_14 Fixed Camera 10_21, 10_22, 10_23, 10_24 Mobile Camera 11. Data Input Unit 12. Judgment Unit 13. Detection Unit 14. Estimation Unit 15. AI Model 16. Event Detection State 17. Camera Hardware Information 20. Server 21. Camera State Management Unit 22. Event Management Unit 23. Search Unit 24. Instruction Unit 25. Camera State Information 26. Event Information 27. Search Condition Information 400 Locations

Claims

1. A data selection method characterized by: obtaining the position of a first sensor that detected an event; obtaining the distance between each of a plurality of sensors mounted on the vehicle and the first sensor; selecting either a sensor from the plurality of sensors whose distance is less than or equal to a threshold, or a predetermined number of sensors extracted from the plurality of sensors in ascending order of distance, as a second sensor; and having a computer perform a process to execute a task related to the event using the sensor data obtained from the second sensor and the sensor data obtained from the first sensor.

2. The data selection method according to claim 1, characterized in that the first sensor and each of the plurality of sensors are instructed to start detecting the event and to stop detecting the event.

3. The data selection method according to claim 1, characterized in that the task is to perform the integration of the event detection result based on sensor data obtained from the second sensor and the event detection result based on sensor data obtained from the first sensor.

4. The data selection method according to claim 1, characterized in that a region is set centered on the location of the event detected by the first sensor, sensors corresponding to the region are extracted from the first sensor and the plurality of sensors according to the amount by which the sensing range overlaps with the region, and the extracted sensors are made to perform detection of the event in the region.

5. A data selection device characterized by acquiring the position of a first sensor that detected an event, acquiring the distance between each of a plurality of sensors mounted on the vehicle and the first sensor, selecting either a sensor from the plurality of sensors whose distance is less than or equal to a threshold, or a predetermined number of sensors extracted from the plurality of sensors in ascending order of distance, as a second sensor, and performing a task related to the event using the sensor data obtained from the second sensor and the sensor data obtained from the first sensor.

6. A data selection program characterized by obtaining the position of a first sensor that detected an event, obtaining the distance between each of a plurality of sensors mounted on the vehicle and the first sensor, selecting either a sensor from the plurality of sensors whose distance is less than or equal to a threshold, or a predetermined number of sensors extracted from the plurality of sensors in ascending order of distance, as a second sensor, and causing a computer to execute a process to perform a task related to the event using the sensor data obtained from the second sensor and the sensor data obtained from the first sensor.

Images