Passage trajectory analysis device, analysis method, and computer program

Abstract

A device according to one aspect of the present disclosure is a device for analyzing a passage trajectory of a passage subject, the device comprising: a memory having a storage region for first detection data and second detection data which are defined below; and a processor for analyzing, on the basis of the first detection data and the second detection data, the passage trajectory. The processor executes integration processing for integrating the first and second detection data under an OR condition, and abnormality determination for determining the presence or absence of abnormal passage by the passage subject by using the integrated data as input data.　First detection data: data including the position and time of a passage subject detected by a first radio wave sensor Second detection data: data including the position and time of a passage subject detected by a second radio wave sensor

Description

Traffic trajectory analysis device, analysis method, and computer program

[0001] This disclosure relates to a traffic trajectory analysis device, an analysis method, and a computer program. This application claims priority to Japanese Application No. 2023-136256, filed on August 24, 2023, and incorporates by reference all of the contents of said Japanese application.

[0002] There is known a technology that senses moving objects around an intersection and generates warning information indicating the location where a pedestrian will cross. For example, in Patent Document 1, when there is a location where a pedestrian will cross at a location different from the location where a crosswalk is formed, warning information about such a location is generated. In addition, information about pedestrians who are behaving abnormally, such as crossing an area of ​​the roadway other than the crosswalk, is notified to the driver of the vehicle.

[0003] Japanese Patent Application Laid-Open No. 2021-026560

[0004] An apparatus according to one aspect of the present disclosure is an apparatus for analyzing the travel trajectory of a passing subject, and includes: a memory having storage areas for first detection data and second detection data defined below; and a processor that analyzes the travel trajectory based on the first detection data and the second detection data, wherein the processor executes an integration process that integrates the first and second detection data using an OR condition; and an abnormality determination that uses the integrated data as input data to determine whether or not the passing subject is traveling abnormally.

[0005] First detection data: Data including the location and time of the passing subject detected by the first radio wave sensor Second detection data: Data including the location and time of the passing subject detected by the second radio wave sensor

[0006] The embodiments of the present disclosure may be realized not only by the above-described device, but also by a system, a method, an integrated circuit, a computer program, or a non-transitory computer-readable recording medium, or any combination thereof. The recording medium may be either volatile or non-volatile. The device may be configured as multiple individual devices. When configured as multiple individual devices, they may be arranged in a single housing or separated into two or more housings.

[0007] FIG. 1 is a system configuration diagram showing an example of a traffic assistance system. FIG. 2 is a block diagram showing an example of the hardware configuration of a master device. FIG. 3 is a block diagram showing an example of the hardware configuration of a slave device. FIG. 4 is an explanatory diagram showing an example of the data structure of detected data. FIG. 5 is a flowchart showing an example of detection data integration processing. FIG. 6 is an explanatory diagram showing an example of the data structure of an autoencoder. FIG. 7 is a flowchart showing an example of abnormal pedestrian traffic determination. FIG. 8 is an explanatory diagram showing variations in abnormal pedestrian traffic. FIG. 9 is a plan view of an intersection showing an example of a display of pedestrian traffic trajectories.

[0008] <Problem to be Solved by the Present Disclosure> In Patent Document 1, the position of a pedestrian is simply identified from an image captured by an on-board camera, so abnormal traffic occurring outside the angle of view of the on-board camera is overlooked. The same applies when an obstacle such as a large vehicle exists between the on-board camera and the pedestrian. In view of the above conventional problems, the present disclosure aims to enable appropriate analysis of whether the traffic trajectory of a passing subject is abnormal.

[0009] Effect of the Present Disclosure According to the present disclosure, it is possible to appropriately analyze whether the travel trajectory of a passing subject is abnormal or not.

[0010] <Outline of Embodiments of the Present Invention> The following is a description of outlines of embodiments of the present invention. (1) A device according to one aspect of the present embodiment is a device for analyzing a travel trajectory of a passing subject, and includes: a memory having storage areas for first detection data and second detection data defined below; and a processor that analyzes the travel trajectory based on the first detection data and the second detection data, wherein the processor executes an integration process that integrates the first and second detection data using an OR condition; and an abnormality determination that determines whether or not the passing subject is traveling abnormally using the integrated data as input data.

[0011] First detection data: Data including the location and time of the passing subject detected by the first radio wave sensor Second detection data: Data including the location and time of the passing subject detected by the second radio wave sensor

[0012] According to the analysis device of this embodiment, the processor executes an integration process that integrates the first and second detection data using an OR condition, and an abnormality determination that determines whether or not the passing subject is traveling abnormally using the integrated data as input data, so that the presence or absence of abnormal traveling can be determined for a wider detection area than when only one type of detection data is used as input data, thereby making it possible to appropriately analyze whether or not the traveling trajectory of the passing subject is abnormal.

[0013] (2) In the analysis device of (1) above, the integration process may include a process of using representative data determined by a predetermined decision logic as input data for the abnormality determination when the first detection data and the second detection data for the same time are identical. The reason for this is that when the above-mentioned identity is found, it is considered that the first radio wave sensor and the second radio wave sensor detected the same passing subject at the same time, and therefore the presence or absence of abnormal passage should be determined for the passing subject as a single passing subject.

[0014] (3) In the analysis device of (1) or (2) above, the integration process may include a process of using both the first detection data and the second detection data as input data for the abnormality determination when there is no identity between the first detection data and the second detection data for the same time. The reason for this is that when there is no identity, it is considered that the first radio wave sensor and the second radio wave sensor have detected different passing entities simultaneously, and therefore the presence or absence of ground traffic should be determined for each passing entity.

[0015] (4) In the analysis device according to any one of (1) to (3) above, the abnormality determination may include a process of determining whether or not abnormal traffic has occurred based on output data output by a learning device using the input data as source data. In this way, the presence or absence of abnormal traffic can be easily determined using the output data from the learning device.

[0016] (5) In the analysis device of (4) above, if the learning device is an autoencoder, the abnormality determination includes a process of determining whether or not there is abnormal traffic based on a comparison result between the input data and the output data of the autoencoder. In this way, the presence or absence of abnormal traffic can be easily determined simply by comparing the input data with the output data output by the autoencoder.

[0017] (6) In the analysis device of (4) above, if the learning device is an RNN learning device, the abnormality determination includes a process of determining whether or not abnormal traffic exists based on a comparison result between the actual input data input to the trained RNN learning device and the predicted data output by the RNN learning device. In this way, the presence or absence of abnormal traffic can be easily determined simply by comparing the input data with the predicted data output by the RNN learning device.

[0018] (7) In the analysis device of (4) above, if the learning device is an SVM learning device, the abnormality determination includes a process of determining whether or not abnormal traffic exists based on a determination result output by the SVM learning device for the input data, the determination result being whether or not the input data conforms to a normal spatial vector distribution. In this way, the presence or absence of abnormal traffic can be easily determined from the determination result output by the SVM learning device.

[0019] (8) The analysis device according to any one of (1) to (7) above may further include a display unit that displays a planar structure of the road, and the display unit may display the travel trajectory in the absence of the abnormal traffic so as to be superimposed on the planar structure of the road. In this way, it becomes possible to consider measures for improving the planar structure of the road based on the travel trajectory in the absence of the abnormal traffic.

[0020] (9) A method according to one aspect of this embodiment is an analysis method executed by the above-described analysis device (1) to (8). Therefore, the analysis method of this embodiment has the same effects as the above-described analysis device (1) to (8).

[0021] (10) A computer program according to one aspect of this embodiment causes a computer to function as the analytical device described above in (1) to (8). Thus, the computer program of this embodiment provides the same effects as the analytical devices described above in (1) to (8).

[0022] <Details of the Embodiments of the Present Invention> Hereinafter, the details of the embodiments of the present invention will be described with reference to the drawings. Note that at least some of the embodiments described below may be combined in any manner.

[0023] [Overall System Configuration] Fig. 1 is a system configuration diagram showing an example of a traffic assistance system 200. As shown in Fig. 1, the traffic assistance system (hereinafter also referred to as "assistance system") 100 includes a radio wave sensor 1 installed near an intersection J, a notification device 2, a roadside communication device 3, an in-vehicle communication device 5 possessed by a vehicle 4, and a management device 6 installed in a traffic management center or the like.

[0024] The traffic assistance system 200 of this embodiment is a system that determines whether or not there is abnormal traffic based on the detection results of the radio wave sensor 1 (for example, the position and speed of a passing subject), and if abnormal traffic is determined to be present, provides the following notification services. The type of subject (passing subject) to be monitored for abnormal traffic is not particularly limited, but in this embodiment, a pedestrian 7 is assumed. Notification 1: The occurrence of abnormal traffic is notified from the alarm device 2 to the local road. Notification 2: The occurrence of abnormal traffic is transmitted to the in-vehicle communication device 5.

[0025] Intersection J is, for example, a crossroads intersection where a north-south road and an east-west road intersect. However, the road directions are merely examples and are not limited to north-south or east-west. At intersection J, vehicle lights 8 that indicate whether vehicles 4 have the right of way and pedestrian lights 9 that indicate whether pedestrians 7 have the right of way are installed. At intersection J, a sensor pole 10 is installed. In addition to the radio wave sensor 1, a notification device 2, a roadside communication device 3, etc. may also be installed on sensor pole 10.

[0026] The radio wave sensor 1 is a radio wave radar for traffic monitoring that detects objects (e.g., pedestrians 7) on the road, and is specifically a millimeter-wave radar. Therefore, the frequency of the modulated wave may be, for example, 24 GHz, 60 GHz, 76 GHz, or 79 GHz. The radio wave sensor 1 emits modulated waves (millimeter waves) in a predetermined direction from a height of several meters above the ground, and detects the position and speed of objects on the road based on an intermediate frequency signal (IF signal) that is a combination of the modulated wave and the reflected wave.

[0027] The radio wave sensor 1 of the traffic assistance system 200 includes at least the following two radio wave sensors 1A and 1B. First radio wave sensor 1A: A radio wave sensor installed in the northwest corner of intersection J. Second radio wave sensor 1B: A radio wave sensor installed in the southeast corner of intersection J. Hereinafter, when describing the configuration and functions common to the first radio wave sensor 1A and the second radio wave sensor 1B, they will be referred to as "radio wave sensor 1."

[0028] In this embodiment, the first radio wave sensor 1A is a "master device" that functions as a device for analyzing travel trajectories, including determining whether or not there is abnormal travel, and the second radio wave sensor 1B is a "slave device" that provides the master device with detection data necessary for analysis. Therefore, hereinafter, the first radio wave sensor 1A may be referred to as the "master device 1A," and the second radio wave sensor 1B may be referred to as the "slave device 1B."

[0029] Detection areas A1 and A2 are set for the first radio wave sensor 1A and the second radio wave sensor 1B. The detection areas A1 and A2 are ranges that include at least the crosswalks 11 and that can obtain reflected waves of sufficient strength to detect objects. In the example of Fig. 1, each of the detection areas A1 and A2 overlaps and covers four crosswalks 11. However, the detection areas may be set separately, such that the detection area A1 covers two crosswalks 11 and the detection area A2 covers the remaining two crosswalks 11.

[0030] The alarm device 2 is a device that alerts passersby to the risk of an accident, and may be, for example, a speaker or an electronic bulletin board. The alarm device 2 may also be a device that alerts passersby both audibly and visually. The alarm device 2 is connected to the master device 1A via a communication line (not shown) of a specified standard. When the master device 1A detects abnormal passage by a pedestrian 7, it transmits a control signal to the alarm device 2 instructing it to perform a specified alarm operation. Upon receiving the control signal, the alarm device 2 performs the specified alarm operation (at least one of audio output and text display indicating abnormal passage).

[0031] The roadside communication device 3 is, for example, an ITS (Intelligent Transport Systems) radio device or a cellular radio device capable of wireless communication via a radio base station 12. However, the roadside communication device 3 may have the functions of both an ITS radio device and a cellular radio device. The roadside communication device 3 includes the following communication devices: First communication device 3A: A communication device connected to the master device 1A via a communication line of a predetermined standard Second communication device 3B: A communication device connected to the slave device 1B via a communication line of a predetermined standard

[0032] When the master unit 1A detects abnormal passage by a pedestrian 7, it transmits notification information including the location where the abnormal passage occurred to the first communication device 3A. The first communication device 3A generates a communication frame including the notification information and transmits the generated communication frame to the in-vehicle communication device 5. The slave unit 1B transmits its own detection data (the position, speed, and time of occurrence of the pedestrian 7) to the second communication device 3B. The second communication device 3B transfers the detection data to the first communication device 3A. The transferred detection data is used by the master unit 1A to determine whether or not abnormal passage has occurred.

[0033] [Hardware Configuration of Master Device] Fig. 2 is a block diagram showing an example of the hardware configuration of the master device 1 A. As shown in Fig. 2, the master device 1 A includes a processor 101, a non-volatile memory 102, a volatile memory 103, a transmitter / receiver 104, and a communication interface (I / F) 107.

[0034] The processor 101 is, for example, a central processing unit (CPU). The processor 101 is not limited to a CPU, but may also be a programmable logic device such as an application specific integrated circuit (ASIC), a gate array, or a field programmable gate array (FPGA). In this case, the ASIC or programmable logic device is configured to be able to execute the same processing as the control program 110A.

[0035] The non-volatile memory 102 is, for example, a flash memory, a hard disk, or a ROM (Read Only Memory). A control program 110A for causing the processor 101 to execute predetermined information processing is stored in the non-volatile memory 102. The volatile memory 103 is, for example, a semiconductor memory such as an SRAM (Static Random Access Memory) or a DRAM (Dynamic Random Access Memory). The volatile memory 103 temporarily reads out the control program 110AA and data necessary for information processing.

[0036] The transmitter / receiver 104 includes a transmitter circuit 105 and a receiver circuit 106. The transmitter circuit 105 includes a transmitter antenna 105a. While FIG. 2 illustrates an example in which only one transmitter antenna 105a is provided, multiple transmitter antennas 105a may be provided. The transmitter circuit 105 generates a modulated wave (radio wave) of a predetermined frequency and transmits the generated modulated wave from the transmitter antenna 105a. When the transmitted modulated wave hits an object (e.g., a pedestrian, a bicycle, or a vehicle), a portion of the modulated wave becomes a reflected wave.

[0037] The receiving circuit 106 includes multiple receiving antennas 106a. That is, multiple (four in the figure) receiving antennas 106a are provided to detect the azimuth angle of an object. The receiving circuit 106 performs signal processing on the received reflected wave to generate reflected wave data. The generated reflected wave data is provided to the processor 101.

[0038] The control program 110A includes a computer program for causing the processor 101 to execute the following information processing: Detection processing: A process for generating detection data including the position, speed, and occurrence time of the pedestrian 7 based on an intermediate frequency signal obtained by combining a modulated wave and a reflected wave from the master device 1A itself.

[0039] Integration process (FIG. 5): A process for integrating the detection data of the master device 1A and the detection data of the slave device 1B under an OR condition. Abnormality determination (FIG. 7): A process for determining whether or not there is abnormal passage of a pedestrian 7 based on the data obtained by the integration process. Details of the integration process (FIG. 5) and abnormality determination (FIG. 7) will be described later.

[0040] The communication interface 107 is a communication module connected to the first communication device 3A. The communication processing performed by the communication interface 107 includes the following data transfers. Data transfer 1: The detection data (position, speed, and time of occurrence of the pedestrian 7) of the second radio wave sensor 1B received from the first communication device 3A is transferred to the processor 101. Data transfer 2: The detection data (position, speed, and time of occurrence of the pedestrian 7) after the integrated processing input from the processor 101 is transferred to the first communication device 3A. The integrated detection data is sent to the management device 6.

[0041] Data transfer 3: A control signal for the notification operation input from the processor 101 is transferred to the notification device 2. Data transfer 4: Notification information input from the processor 101, including the location of the abnormal traffic occurrence, is transferred to the first communication device 3A. The notification information is then transmitted to the in-vehicle communication device 5.

[0042] [Hardware Configuration of Slave Device] Fig. 3 is a block diagram showing an example of the hardware configuration of slave device 1B. As shown in Fig. 3, slave device 1B includes a processor 101, nonvolatile memory 102, volatile memory 103, a transmitter / receiver 104, and a communication interface (I / F) 107, and its hardware configuration is similar to that of master device 1A (Fig. 2). Therefore, the following describes the differences in function from master device 1A (Fig. 2).

[0043] The non-volatile memory 102 stores a control program 110B for causing the processor 101 to execute predetermined information processing. The control program 110B includes computer programs for causing the processor 101 to execute the following information processing: Detection processing: A process for generating detection data including the position, speed, and occurrence time of the pedestrian 7 based on an intermediate frequency signal obtained by combining a modulated wave and a reflected wave from the own device (slave unit 1B).

[0044] Transfer process: A process in which the detection data of the own device (slave device 1B) is output to the communication interface 107 and transferred to the master device 1A. Therefore, the communication interface 107 transmits the detection data input from the processor 101 to the second communication device 3B, and the second communication device 3B transfers the received detection data to the master device 1A.

[0045] [Data Structure of Sensing Data] Fig. 4 is an explanatory diagram showing an example of the data structure of the sensing data D1 and D2. "D1" in Fig. 4 is sensing data (hereinafter also referred to as "first sensing data") generated by the sensing process of the master device 1A. "D2" in Fig. 4 is sensing data (hereinafter also referred to as "second sensing data") generated by the sensing process of the slave device 1B.

[0046] The first detection data D1 is a vector whose elements are the position (X1, Y1) of the detected object (maximum point of the reflected wave) defined in a "first coordinate system," the velocity V1 of the detected object at that position, and the time of occurrence of that position (e.g., the time of detection of the position) t1. The first coordinate system is an orthogonal coordinate system in which the origin O1 is the ground surface position of the radio wave irradiation point of the master device 1A, and the position of the detected object is defined by two orthogonal axes on the ground surface. The Y1 axis is, for example, the intersection line between the ground and a vertical plane including the radio wave irradiation axis of the master device 1A (the normal direction of the radio wave irradiation surface).

[0047] The velocity V1 is the velocity in the direction connecting the origin O1 and the position (X1, Y1) of the detected object, and for example, the direction away from the origin O1 is positive and the direction toward the origin O1 is negative. In the detection process, the processor 101 of the master device 1A determines whether the detected object is a pedestrian 7, for example, based on the magnitude of the velocity V1. If the detected object is a pedestrian 7, the processor 101 of the master device 1A calculates the position (X1, Y1) of the pedestrian 7 based on the distance R1 from the origin O1 to the pedestrian 7.

[0048] The second detection data D2 is a vector defined in a "second coordinate system" and has as its elements the position (X2, Y2) of the detected object (the maximum point of the reflected wave), the velocity V2 of the detected object at that position, and the time of occurrence of that position (e.g., the time of detection of the position) t2. The second coordinate system is an orthogonal coordinate system in which the origin O2 is the ground surface position of the radio wave irradiation point of the slave unit 1B, and the position of the detected object is defined by two orthogonal axes on the ground surface. The Y2 axis is, for example, the intersection line between the ground and a vertical plane including the radio wave irradiation axis of the slave unit 1B (the normal direction of the radio wave irradiation surface). In this embodiment, the Y2 axis is an axis in the opposite direction to the Y1 axis and included in the same vertical plane.

[0049] The velocity V2 is the velocity in the direction connecting the origin O2 and the position (X2, Y2) of the detected object, and for example, the direction away from the origin O2 is positive and the direction toward the origin O2 is negative. In the detection process, the processor 101 of the slave unit 1B determines whether the detected object is a pedestrian 7, for example, based on the magnitude of the velocity V2. If the detected object is a pedestrian 7, the processor 101 of the slave unit 1B calculates the position (X1, Y1) of the pedestrian 7 based on the distance R2 from the origin O2 to the pedestrian 7.

[0050] 4, for example, let (Sa, Sb) be the coordinate values ​​in the first coordinate system of the installation position of the slave unit 1B (limitation O2 of the second coordinate system). In this case, if the master unit 1A and the slave unit 1B detect the same pedestrian 7, the following relationship holds between each element of the first detection data D1 and each element of the second detection data D2. X position: X1 = -X2 (1) Y position: Y1 = -Y2 + Sb (2) Speed: V1 = V2 × cos(θ1 - θ2) (3) where θ1 is the azimuth angle in the clockwise direction from the Y1 axis, and θ2 is the azimuth angle in the clockwise direction from the Y2 axis.

[0051] The coordinate values ​​(Sa, Sb) of the installation position of the slave unit 1B are measured in advance by a traffic manager using a management terminal such as a laptop PC connected to the master unit 1A, and are recorded in the non-volatile memory 102 of the master unit 1A. The coordinate values ​​(Sa, Sb) can be measured, for example, by attaching a corner reflector (not shown) to the installation position of the slave unit 1B and having the master unit 1A detect the position of the reflector.

[0052] [Detection Data Integration Process] Fig. 5 is a flowchart showing an example of detection data integration process executed by the master device 1A. The integration process in Fig. 5 is executed by the processor 101 of the master device 1A at predetermined control intervals (e.g., 100 ms). The meanings of the variables in Fig. 5 are as follows:

[0053] M1: storage area in the volatile memory 103 for the first sensed data D1; M2: storage area in the volatile memory 103 for the converted second sensed data tD2; tD2: data obtained by converting each element of the second sensed data D2 into the first coordinate system; rD: data representing the first sensed data D1 and the converted second sensed data tD2.

[0054] The processor 101 of the master device 1A buffers the sensed data D1 generated by its own sensing process in a storage area M1. Furthermore, upon receiving the second sensed data D2 from the slave device 1B, the processor 101 converts each element of the second sensed data D2 into the first coordinate system using the above-described relational expressions (1) to (3), and stores the converted second sensed data tD2 in the storage area M2. Hereinafter, the converted second sensed data tD2 will also be referred to as "converted data tD2."

[0055] Furthermore, any of the following may be adopted as the logic for determining the representative data rD: Logic 1: Each element of the representative data rD is set to each element of the first sensed data D1; Logic 2: Each element of the representative data rD is set to each element of the converted data tD2; Logic 3: Each element of the representative data rD is set to the average or median value of each element of the first sensed data D1 and each element of the converted data tD2.

[0056] 5, the processor 101 determines whether or not the sensed data D1 is present in the region M1 (step ST11). If the determination result in step ST11 is positive, the processor 101 determines whether or not the converted data tD2 is present in the region M2 (step ST12).

[0057] If the determination result of step ST12 is positive, the processor 101 determines whether the first sensed data D1 and the converted data tD2 are identical (step ST13). The identity determination is performed, for example, by determining whether the difference values ​​(ΔX, ΔY, ΔV) of each element of both data D1 and tD2, which have the same occurrence times t1 and t2, are within a predetermined threshold value (step ST14). In the illustrated example, the difference values ​​(ΔX, ΔY, ΔV) being within the threshold value is described as "D1 ≒ tD2."

[0058] If the determination result in step ST14 is positive, the processor 101 determines that the representative data rD is the target of abnormality determination (step ST15). The establishment of "D1 ≒ tD2" (step ST14) is considered to be due to the master device 1A and the slave device 1B simultaneously detecting the same pedestrian 7.

[0059] If the determination result in step ST14 is negative, the processor 101 determines that both the first sensed data D1 and the converted data tD2 are to be subjected to abnormality determination (step ST17). If "D1 ≒ tD2" (step ST14) does not hold, this is because the master device 1A and the slave device 1B have simultaneously detected different pedestrians 7.

[0060] If the determination result in step ST12 is negative, the processor 101 determines that only the first sensed data D1 is the target of abnormality determination (slave ST18). This is because the situation in which the first sensed data D1 is present in the region M1 (yes in step ST11) and the converted data tD2 is absent in the region M2 (no in step ST12) is estimated to be a situation in which only the master device 1A is detecting a pedestrian 7.

[0061] If the result of the determination in step ST11 is negative, the processor 101 determines again whether or not the transformation data tD2 exists in the region M2 (step ST16).

[0062] If the determination result in step ST16 is positive, processor 101 determines only converted data tD2 as the target for abnormality determination (slave ST18). This is because a situation in which no detected data D1 is in region M1 (negative in step ST11) and converted data tD2 is in region M2 (positive in step ST16) is presumed to be a situation in which only slave unit 1B is detecting a pedestrian 7.

[0063] If the determination result in step ST16 is negative, the processor 101 skips the identification of the determination target (steps ST15, ST17, ST18, and ST19) and ends the process. When the first detection data D1 is not present in the region M1 (negative in step ST11) and the converted data tD2 is not present in the region M2 (positive in step ST16), it is estimated that neither the master device 1A nor the slave device 1B has detected the pedestrian 7.

[0064] 1, the management device 6 is a computer connected to the public communication network 13. The management device 6 includes a control unit 61, a storage unit 62, and a communication unit 63 housed in a housing, and a sensor database 64 constructed in the storage unit 62. A user of the management device 6 is, for example, an operations manager of a traffic control system that manages traffic signal control at an intersection J.

[0065] The sensor database 64 is electronic data constructed in a predetermined data array in the storage unit 62. Note that a part or all of the sensor database 64 may be constructed in an external storage device (not shown) connected to the housing of the management device 6. An operation unit 66 and a display unit 67 are connected to the housing of the management device 6. Although FIG. 1 illustrates a desktop computer as an example of the management device 6, the management device 6 may also be a mobile computer such as a notebook computer or a tablet computer.

[0066] The control unit 61 is an arithmetic processing device including a CPU, RAM, etc. The control unit 61 may include integrated circuits such as a GPU (Graphics Processing Unit) and an FPGA. The storage unit 62 is an auxiliary storage device including non-volatile memory such as an HDD (Hard Disk Drive) and an SSD (Solid State Drive). The storage unit 62 may include a flash ROM (Read Only Memory), a USB (Universal Serial Bus) memory, an SD card, etc.

[0067] The communication unit 63 is a communication interface that is connected to the public communication network 13 in accordance with a predetermined communication standard. Therefore, the communication unit 63 can receive a communication frame including detection data of the pedestrian 7 from the master device 1A via the wireless base station 12. The received detection data of the pedestrian 7 is any one of the above-mentioned detection data D1, converted data tD2, and representative data rD.

[0068] The communication unit 63 transfers the received communication frame to the control unit 61. The control unit 61 extracts the detection data of the pedestrian 7 from the communication frame transferred from the communication unit 63 and stores the extracted detection data in the sensor database 64. In this way, the sensor database 64 of the management device 6 accumulates the detection data of the pedestrian 7 sensed by the first and second radio wave sensors 1A, 1B, respectively.

[0069] The operation unit 66 is an input device such as a keyboard and a pointing device. A user of the management device 6 can send a predetermined instruction to the control unit 61 of the management device 6 by inputting a command or clicking on the operation unit 66. The display unit 67 is a display device such as a liquid crystal monitor or an organic EL panel. The display unit 67 can display a GUI (Graphical User Interface) screen or the like for receiving operation input on the operation unit 66.

[0070] The storage unit 62 stores a computer program 65 that causes the control unit 61 to execute predetermined information processing, and an autoencoder 68 used for abnormality determination. The autoencoder 68 is used as a learning device that uses, for example, detection data of a large number of pedestrians 7 stored in the sensor database 64 as input information and pre-learns normal travel trajectories of a large number of pedestrians 7.

[0071] [Autoencoder] Fig. 6 is an explanatory diagram showing an example of the data structure of the autoencoder 68. As shown in Fig. 6, the autoencoder 68 is configured as a neural network including an input layer including a predetermined number of input nodes x, an output layer including a predetermined number of output nodes y, and a hidden layer z located between the input nodes x and the output nodes y. The position (X1, Y1) of the detection data that progresses every second, for example, is input to the input node x.

[0072] The computer program 65 of the management device 6 includes a program for executing pre-learning (training) on ​​the autoencoder 68. In the pre-learning, the position (X1, Y1) of the detection data of the pedestrian 7 for a predetermined period (e.g., three months) is input to the input node x, and when the difference between the position (X1, Y1) of the input data x and the position (X1, Y1) of the output data y converges to within a predetermined value, it is determined that "learning has been completed."

[0073] The control unit (e.g., GPU, CPU, etc.) 61 of the management device 6 can also determine whether the movement trajectory of the pedestrian 7 is abnormal based on whether the trained autoencoder 68 can correctly restore the movement trajectory. Therefore, the management device 6 functions as an analyzer for the movement trajectory of the pedestrian 7. The processor 101 of the master device 1A downloads the trained autoencoder 68 from the management device 6 and stores it in the non-volatile memory 102. The autoencoder 68 is used to determine whether there is an abnormality in the pedestrian 7 ( FIG. 7 ), which will be described later.

[0074] [Determining Abnormalities in Pedestrian Traffic] Fig. 7 is a flowchart showing an example of determining abnormalities in a traffic trajectory, which is executed by the processor 101 of the master device 1A. As shown in Fig. 7, the processor 101 of the master device 1A determines whether or not there is integrated data from a predetermined time T1 ago from the present time (step ST31).

[0075] The predetermined time T1 going back in time can be set to, for example, 3 to 4 seconds. The "integrated data" refers to the detection data of the pedestrian 7 obtained by the integration process ( FIG. 5 ). Specifically, the integrated data is any one of the first detection data D1, the converted second detection data (converted data) tD2, and the representative data rD. If the determination result in step ST31 is negative, the processor 101 waits until the determination result becomes positive.

[0076] If the determination result of step ST31 is positive, the processor 101 generates input data x from the integrated data obtained during the predetermined time T1 and inputs the generated input data x to the autoencoder 68 (step ST32). Next, the processor 101 determines whether or not there is abnormal passage using the autoencoder 68 of the pedestrian 7 (step ST33). This determination can be made, for example, based on the degree of deviation between the output data y and the input data x.

[0077] If the determination result of step ST33 is positive, the processor 101 executes a notification process (step ST34). The notification process is, for example, at least one of a process of transmitting a control signal for an annunciation operation to the annunciation device 2 and a process of transmitting a communication frame including annunciation information to the in-vehicle communication device 5.

[0078] Next, the processor 101 waits for a predetermined waiting time T2 (e.g., 1 second) to elapse (step ST35), and then returns the process to before step ST31. If the determination result of step ST33 is negative, the processor 101 updates the feature amount (e.g., weight or learning parameter) of the autoencoder (step ST36), and then determines whether or not the notification process is being executed (step ST37).

[0079] If the determination result in step ST37 is positive, the processor 101 stops the notification process (step ST38) and then shifts the process to before step ST35. If the determination result in step ST37 is negative, the processor 101 skips step ST38 and shifts the process to before step ST35.

[0080] [Variations in Abnormal Passing] Figure 8 is an explanatory diagram showing variations in abnormal passing by a pedestrian 7. As shown in Figure 8, when viewed from the second radio wave sensor 1B, the pedestrian 7 is located behind the bus 5, for example, and therefore cannot be detected, but can be detected from the first radio wave sensor 1A. Furthermore, the detection areas A1 and A2 of the two radio wave sensors 1A and 1B cover almost the entire area of ​​the intersection J (see Figure 1).

[0081] Therefore, by performing the above-mentioned abnormality determination (FIG. 7) using the data obtained by performing the above-mentioned integration process (FIG. 5) on the detection data D1, D2 from the two radio wave sensors 1A, 1B, it becomes possible to accurately determine abnormal traffic behavior such as "stopping on the sidewalk," "crossing diagonally," "snaking," and "abnormal speed" over a wide area around the intersection J.

[0082] [Display example of pedestrian movement trajectory] Fig. 9 is a plan view of a road showing an example of a display of a movement trajectory of a pedestrian 7 that can be displayed on the display unit 67 of the management device 6. In Fig. 9, the solid arrows indicate the movement trajectory of the pedestrian 7 that has been determined to be normal by the autoencoder 68 installed in the management device 6. The shaded areas indicate "sidewalks."

[0083] In this case, the travel trajectory of pedestrian 7 determined to be normal crosses the roadway along a route that deviates from crosswalk 11 and heads toward sidewalk R2 on the east side. This is thought to be because sidewalk R2, which extends east from intersection J, is located away from roadway R1 and to the north. In this way, by superimposing a travel trajectory determined to be normal by autoencoder 68 on the planar structure of the road, it becomes possible to consider measures to improve the planar structure of the road based on the route of the travel trajectory.

[0084] [Modifications of the Learning Device] In the above-described embodiment, the learning device 68 that performs anomaly determination using the integrated data as source data may be, in addition to an autoencoder, the following learning devices 68, for example: 1) A learning device that employs a convolutional neural network (CNN) (CNN learning device) 2) A learning device that employs a Transformer

[0085] 3) Learning device using a recurrent neural network (RNN) (RNN learning device) An RNN learning device determines whether or not there is abnormal traffic based on the comparison results (e.g., distance) between the actual trajectory (input data) input and the trajectory (output data) predicted by the learning device. Examples of RNN learning devices include a Grated Recurrent Unit (GRU) and a Long Short Term Memory (LSTM).

[0086] 4) Learning Device Using Support Vector Machine (SVM) (SVM Learning Device) An SVM learning device determines whether input data conforms to a normal spatial vector distribution. Because an SVM learning device is an unsupervised learning device, it has the advantage of not needing to collect normal data and being easy to operate. Examples of SVM learning devices include one-class neural networks (OC-NN), deep support vector data description (Deep SVDD), isolation forest, and local outlier factor (LOF).

[0087] [Other Modifications] The above-described embodiment is illustrative in all respects and is not limiting. The scope of the present invention includes all modifications within the scope of equivalents to the configurations described in the claims. For example, in the above-described embodiment, the passing subject whose travel path is monitored may be not only a pedestrian 7 but also a bicycle, etc.

[0088] In the above-described embodiment, the first radio wave sensor 1A functions as an analysis device that performs integration processing (FIG. 5) and abnormality determination (FIG. 7), but the function as an analysis device may also be implemented in another communication node such as the roadside communication device 3.

[0089] 1 Radio wave sensor 1A First radio wave sensor (master device, analysis device) 1B Second radio wave sensor (slave device) 2 Notification device 3 Roadside communication device 3A First communication device 3B Second communication device 4 Vehicle 5 In-vehicle communication device 6 Management device (analysis device) 7 Pedestrian 8 Vehicle light 9 Pedestrian light 10 Support pole 11 Crosswalk 12 Wireless base station 13 Public communication network 61 Control unit 62 Memory unit 63 Communication unit 64 Sensor database 65 Computer program 66 Operation unit 67 Display unit 68 Autoencoder (learning device) 101 Processor 102 Non-volatile memory 103 Volatile memory 104 Transmitting / receiving unit 105 Transmitting circuit 105a Transmitting antenna 106 Receiving circuit 106a Receiving antenna 107 Communication interface 110A Control program 110B Control program 200 Traffic assistance system A1 Detection area of ​​first radio wave sensor A2 Detection area of ​​second radio wave sensor D1 First detection data D2 Second detection data tD2 Second detection data after conversion (conversion data) rD Representative data

Claims

1. A device for analyzing a travel trajectory of a passing subject, comprising: a memory having a storage area for first detection data and second detection data defined below; and a processor for analyzing the travel trajectory based on the first detection data and the second detection data, wherein the processor executes an integration process for integrating the first and second detection data under an OR condition; and an abnormality determination for determining the presence or absence of abnormal passage by the passing subject using the integrated data as input data. First detection data: Data including the position and time of the passing subject detected by a first radio wave sensor Second detection data: Data including the position and time of the passing subject detected by a second radio wave sensor 2. The travel trajectory analysis device of claim 1, wherein the integration process includes a process of using representative data determined by a predetermined decision logic as input data for the abnormality judgment when there is identity between the first detection data and the second detection data for the same time.

3. A travel trajectory analysis device as described in claim 1 or claim 2, wherein the integration process includes a process in which, if there is no identity between the first detection data and the second detection data for the same time, both the first detection data and the second detection data are used as input data for the abnormality determination.

4. A traffic trajectory analysis device as described in any one of claims 1 to 3, wherein the abnormality determination includes a process of determining whether or not there is abnormal traffic based on output data output by a learning device using the input data as original data.

5. The traffic trajectory analysis device of claim 4, wherein the learning device is an autoencoder, and the abnormality determination includes a process of determining whether or not abnormal traffic exists based on a comparison result between the input data and output data of the autoencoder.

6. The traffic trajectory analysis device of claim 4, wherein the learning device is an RNN learning device, and the abnormality determination includes a process of determining whether or not abnormal traffic has occurred based on a comparison result between the actual input data input to the trained RNN learning device and the predicted data output by the RNN learning device.

7. The traffic trajectory analysis device of claim 4, wherein the learning device is an SVM learning device, and the abnormality determination includes a process of determining whether or not abnormal traffic exists based on a determination result output by the SVM learning device for the input data as to whether or not the input data conforms to a distribution of normal spatial vectors.

8. A traffic trajectory analysis device as described in any one of claims 1 to 7, further comprising a display unit that displays the planar shape of a road to be monitored, said display unit displaying the traffic trajectory in the absence of said abnormal traffic by superimposing it on the planar structure of said road.

9. A method for analyzing a travel trajectory of a passing subject executed by a computer, comprising: a first step of storing first detection data and second detection data defined below; and a second step of analyzing the travel trajectory based on the first detection data and the second detection data, the second step including: an integration process of integrating the first and second detection data under an OR condition; and an abnormality determination process of determining the presence or absence of abnormal passage by the passing subject using the integrated data as input data. First detection data: Data including the position and time of the passing subject detected by a first radio wave sensor Second detection data: Data including the position and time of the passing subject detected by a second radio wave sensor 10. A computer program that causes a computer to function as a device for analyzing the movement trajectory of a passing subject, the computer being configured to function as: a memory having a storage area for first detection data and second detection data defined below; and a processor that analyzes the movement trajectory based on the first detection data and the second detection data, the processor executing an integration process that integrates the first and second detection data under an OR condition; and an abnormality determination that determines whether or not the passing subject is moving abnormally using the integrated data as input data. First detection data: data including the position and time of a passing subject detected by a first radio wave sensor Second detection data: data including the position and time of a passing subject detected by a second radio wave sensor

Images