Abnormality detection method, abnormality detection device, and program

Abstract

The abnormality detection method includes: acquiring position information indicating a position of a moving body (20) and operation information indicating an operation of the moving body (20) associated with the position information (S410); when the position indicated by the position information is a specific position, acquiring a travel model (S420) associated with the specific position and generated on the basis of movement data relating to past movement of at least one of the one or more moving bodies including the moving body (20); if the position indicated by the position information is a specific position, calculating an abnormality degree indicating the degree of abnormality related to the movement of the moving body (20) on the basis of the operation information and the travel model (S425); determining the presence or absence of an abnormality related to the movement of the moving body (20) on the basis of the degree of abnormality (S435); and outputting the result of the determination to the outside (S450).

Description

Technical Field

[0001] This disclosure relates to anomaly detection methods, anomaly detection devices, and procedures for detecting movement-related anomalies of a moving body. Background Technology

[0002] Previously, anomaly detection methods for detecting anomalies in moving objects have been proposed (for example, see Patent Document 1).

[0003] Existing technical documents

[0004] Patent documents

[0005] Patent Document 1: International Publication No. 2021 / 149340 Summary of the Invention

[0006] The problem that the invention aims to solve

[0007] In recent years, the threat of cyberattacks launched by malicious attackers against autonomous vehicles and other mobile entities has increased.

[0008] Therefore, the purpose of this disclosure is to provide an anomaly detection method, etc., that can detect anomalies related to the movement of a mobile body even if the location information of the mobile body has been tampered with by a malicious attacker.

[0009] Technical solutions for solving the problem

[0010] An anomaly detection method disclosed herein includes: an information acquisition step, acquiring first location information representing the position of a moving body and first action information associated with the first location information representing the action of the moving body; a model acquisition step, where, if the position represented by the first location information is a first specific position, acquiring a first driving model associated with the first specific position, generated based on first movement data related to the past movement of at least one of more than one moving bodies including the moving body; an anomaly degree calculation step, where, if the position represented by the first location information is the first specific position, calculating a first anomaly degree representing the degree of anomaly related to the movement of the moving body based on the first action information and the first driving model; a determination step, determining the presence or absence of anomalies related to the movement of the moving body based on the first anomaly degree; and an output step, outputting the determination result to an external source.

[0011] An anomaly detection device disclosed herein includes: an information acquisition unit that acquires first position information representing the position of a moving body and first action information associated with the first position information representing the action of the moving body; a model acquisition unit that, when the position represented by the first position information is a first specific position, acquires a first driving model associated with the first specific position, generated based on first movement data related to the past movement of at least one of more than one moving bodies including the moving body; an anomaly degree calculation unit that, when the position represented by the first position information is the first specific position, calculates a first anomaly degree representing the degree of anomaly related to the movement of the moving body based on the first action information and the first driving model; a determination unit that, based on the first anomaly degree, determines whether an anomaly related to the movement of the moving body exists; and an output unit that outputs the determination result to an external location.

[0012] One technical solution disclosed herein relates to a program for enabling a computer to perform anomaly detection processing, the anomaly detection processing comprising: an information acquisition step, acquiring first location information representing the position of a moving body and first action information associated with the first location information representing the action of the moving body; a model acquisition step, if the position represented by the first location information is a first specific position, acquiring a first driving model associated with the first specific position, generated based on first movement data related to the past movement of at least one of more than one moving bodies including the moving body; an anomaly degree calculation step, if the position represented by the first location information is the first specific position, calculating a first anomaly degree representing the degree of anomaly related to the movement of the moving body based on the first action information and the first driving model; a determination step, determining the presence or absence of an anomaly related to the movement of the moving body based on the first anomaly degree; and an output step, outputting the determination result to an external source.

[0013] Invention Effects

[0014] According to an anomaly detection method or the like involved in one of the technical solutions disclosed herein, even if the location information of a moving body is tampered with by a malicious attacker, anomalies related to the movement of the moving body can still be detected. Attached Figure Description

[0015] Figure 1 This is a block diagram illustrating the configuration of the anomaly detection system involved in the implementation method.

[0016] Figure 2 This is a block diagram illustrating the structure of the model generation unit involved in the implementation method.

[0017] Figure 3This is a schematic diagram illustrating an example of the data structure of vehicle information stored in the information acquisition unit involved in the implementation method.

[0018] Figure 4 This is a schematic diagram illustrating an example of a data structure for the vehicle information stored in the model acquisition unit involved in the implementation method.

[0019] Figure 5 This is a schematic diagram illustrating an example of a data structure for vehicle information stored in the determination unit involved in the implementation method.

[0020] Figure 6 This is a flowchart of the model learning process involved in the implementation method.

[0021] Figure 7 This is a flowchart of the classification process involved in the implementation method.

[0022] Figure 8 This is a schematic diagram illustrating the situation in which the classification department involved in the implementation determines whether the travel location of the mobile body is an intersection.

[0023] Figure 9 This is a schematic diagram illustrating the situation where the classification department of the embodiment determines the driving pattern of the mobile body.

[0024] Figure 10 This is a schematic diagram illustrating the situation where the classification department of the embodiment determines the driving mode of the mobile body.

[0025] Figure 11A This is a flowchart of the first weighting coefficient calculation process involved in the implementation method.

[0026] Figure 11B This is a flowchart of the second weighting coefficient calculation process involved in the implementation method.

[0027] Figure 12 This is a schematic diagram illustrating an example of a scenario in which the model storage unit involved in the implementation stores driving models and weighting coefficients in association.

[0028] Figure 13 This is a flowchart of the anomaly detection process involved in the implementation method.

[0029] Figure 14 This is a schematic diagram illustrating an example of how the anomaly calculation unit in the embodiment calculates anomalies related to the movement of the moving body.

[0030] Figure 15 This is a schematic diagram illustrating an example of an image displayed by a display unit according to an embodiment.

[0031] Figure 16This is a schematic diagram illustrating an example of a method for verifying the validity (legitimacy) of location information at bends in a driving road, as considered by the inventor. Detailed Implementation

[0032] (The reason for obtaining one of the technical solutions disclosed herein)

[0033] In recent years, various measures have been implemented with the goal of making autonomous vehicles practical.

[0034] For example, demonstration experiments are underway on various services utilizing autonomous vehicles for indoor and outdoor delivery of goods and human mobility, as well as various services utilizing autonomous cleaning robots and autonomous security robots.

[0035] In order to provide safe services using autonomous vehicles, communication is required between autonomous vehicles and remote monitoring locations to enable status monitoring of autonomous vehicles and remote operation of autonomous vehicles in emergencies.

[0036] However, with the introduction of communication systems that enable communication between autonomous vehicles and remotely monitored locations, the risk arises that autonomous vehicles become targets of cyberattacks via these systems.

[0037] In the past, in order to reduce this risk, anomaly detection devices for cyberattacks on vehicles have been proposed.

[0038] On the other hand, when introducing such anomaly detection devices, the numerous over-detection (false alarms) of anomalies caused by the diversity of vehicle driving modes become a problem to be solved.

[0039] Therefore, for example, in Patent Document 1, in order to detect vehicle anomalies caused by network attacks while suppressing over-detection of anomalies, a method is proposed that utilizes the characteristics of the location represented by the current location information to detect vehicle anomalies.

[0040] However, in the case of an anomaly detection model that selects to use characteristics suitable for the current location of the vehicle (e.g., whether the location is a regular road or a highway) based on the location information representing the current location of the vehicle, as disclosed in Patent Document 1, it is based on the premise that the current location of the vehicle shown by the location information is correct.

[0041] For example, if the location information is tampered with by a malicious attacker, anomaly detection using an anomaly detection model corresponding to the location shown in the location information may be bypassed (for example, an unauthorized vehicle is actually in an abnormal state of traveling at 100 km / h on a normal road, but by tampering with the location information, the road the vehicle is traveling on is identified as a highway, and thus the anomaly of the vehicle may not be detected).

[0042] Therefore, it is necessary to verify the validity of the location information. To this end, for example, the following method is considered: converting information of different categories from the vehicle's location information (e.g., logs of vehicle control signals, speed information, etc.) into location information, and comparing the converted location information with the location information to verify the validity of the location information.

[0043] However, during this conversion process, unexpected position deviations or discrepancies may occur (for example, when the position shift shown by the position information is estimated based on the information of velocity and angular velocity, the trajectory of the original position information may differ greatly from the actual position trajectory due to the influence of friction, etc).

[0044] Therefore, it is necessary to detect anomalies with high accuracy while avoiding over-detection of anomalies that may occur due to such unexpected positional deviations or differences.

[0045] Incidentally, it is believed that services utilizing autonomous vehicles will continue to expand in the future as a solution to labor shortages.

[0046] However, autonomous vehicles equipped with IoT (Internet of Things) are at risk of being hacked. Once hacked, a replay attack can be launched against the autonomous vehicle using simple commands, so it can be said that it is at high risk of becoming a target of cyberattacks.

[0047] As a countermeasure, many anomaly detection methods for remote vehicle anomaly monitoring have been proposed. However, given the large number of vehicles being monitored, how to suppress over-detection of anomalies has become a challenge.

[0048] Therefore, measures are being implemented to modify the anomaly detection model based on the current location of moving objects such as autonomous vehicles, as indicated by location information, to perform anomaly detection that better matches the characteristics of that location, thereby suppressing over-detection of anomalies. However, regarding this measure, if the location information is tampered with by a malicious attacker, as mentioned above, anomaly detection may be evaded.

[0049] Therefore, the inventors considered the need to determine whether the location information itself had been tampered with.

[0050] To detect tampering with location information, it is necessary to verify the validity of the location information by comparing it with other categories of information about the moving body.

[0051] However, through demonstration experiments, the inventors discovered that when the validity of location information is confirmed by comparing it with other types of information of the moving body, not every location can be used as a location for comparing this information. There are locations where the moving body travels in a driving mode suitable for such comparison and locations where the moving body travels in a driving mode unsuitable for such comparison.

[0052] For example, in locations with many pedestrians, vehicles are more likely to use irregular driving patterns, such as meandering or repeatedly moving forward and backward, to avoid pedestrians. Therefore, false confirmations may occur in such locations when verifying the validity of location information. To reiterate, the driving patterns mentioned here refer to classifications related to the vehicle's actions, such as going straight, turning right, turning left, or making a U-turn at a given location.

[0053] Conversely, in locations where it is evident that a vehicle tends to travel in a specific driving pattern, such as where the road is winding, misidentification in confirming the validity of location information is suppressed.

[0054] Figure 16 This is a schematic diagram illustrating an example of a method for verifying the validity of location information at points where a driving road curves, as considered by the inventor.

[0055] exist Figure 16 In the data, (a) represents the change in velocity and angular velocity of the moving body when it moved from the starting point A to the destination B in the past (time series data), and (b) represents the change in velocity and angular velocity of the moving body when it moves from the starting point A to the destination B this time (time series data).

[0056] like Figure 16 As shown, the inventors discovered that if the validity of location information can be confirmed based on the past speed and angular velocity shifts of the mobile body at locations where it can be seen that the mobile body tends to travel in a specific driving pattern (in this case, two turning points to the right at 90 degrees), and the current speed and angular velocity shifts of the mobile body when it travels to that location again, then the occurrence of misidentification in the confirmation of the validity of location information can be suppressed.

[0057] Based on this discovery, the inventors then conducted extensive experiments and research.

[0058] As a result, the inventors came up with the following anomaly detection method, anomaly detection device, and procedure disclosed herein.

[0059] An anomaly detection method disclosed herein includes: an information acquisition step, acquiring first location information representing the position of a moving body and first action information associated with the first location information representing the action of the moving body; a model acquisition step, where, if the position represented by the first location information is a first specific position, acquiring a first driving model associated with the first specific position, generated based on first movement data related to the past movement of at least one of more than one moving bodies including the moving body; an anomaly degree calculation step, where, if the position represented by the first location information is the first specific position, calculating a first anomaly degree representing the degree of anomaly related to the movement of the moving body based on the first action information and the first driving model; a determination step, determining the presence or absence of anomalies related to the movement of the moving body based on the first anomaly degree; and an output step, outputting the determination result to an external source.

[0060] According to the above-described anomaly detection method, based on first action information representing the movement of the mobile body, which is associated with first location information indicating that the mobile body is traveling in a driving mode suitable for confirming the validity of the location information, and a first driving model associated with the first specific location, a first anomaly degree of the mobile body at the first specific location can be calculated. Based on the calculated first anomaly degree, the presence or absence of anomalies related to the movement of the mobile body can be determined. Therefore, according to the above-described anomaly detection method, even if the obtained first location information is tampered with, it can be determined that anomalies related to the movement of the mobile body exist.

[0061] Thus, according to the above anomaly detection method, even if the location information of a moving object is tampered with by a malicious attacker, anomalies related to the movement of that moving object can still be detected.

[0062] Alternatively, in the information acquisition step, second location information representing the position of the moving body and second action information representing the action of the moving body associated with the second location information may also be acquired. In the model acquisition step, if the position represented by the second location information is a second specific position, a second driving model generated based on second movement data related to the past movement of the at least one moving body associated with the second specific position may also be acquired. In the anomaly calculation step, if the position represented by the second location information is the second specific position, a second anomaly degree representing the degree of anomaly related to the movement of the moving body may also be calculated based on the second action information and the second driving model. In the determination step, the determination may be performed based on the first anomaly degree and the second anomaly degree.

[0063] According to the above-mentioned anomaly detection method, it is possible to determine the presence or absence of anomalies related to the movement of a mobile body based on the first degree of anomaly at a first specific location where the mobile body is traveling in a driving mode suitable for confirming the validity of the location information, and the second degree of anomaly at a second specific location where the mobile body is traveling in a driving mode suitable for confirming the validity of the location information.

[0064] Therefore, based on the above-mentioned anomaly detection method, it is possible to detect movement-related anomalies of moving objects with higher accuracy.

[0065] Alternatively, the anomaly detection method may further include a first anomaly adjustment step, in which a first weighted anomaly is calculated based on the first anomaly and a first weighted coefficient that corresponds to the position represented by the first position information; a second weighted anomaly is calculated based on the second anomaly and a second weighted coefficient that corresponds to the position represented by the second position information; and in the determination step, the determination is performed based on the first weighted anomaly and the second weighted anomaly.

[0066] According to the above anomaly detection method, it is possible to determine whether there is an anomaly related to the movement of the moving body based on the first weighted anomaly degree obtained by weighting the first anomaly degree with a corresponding relationship to the first position and the second weighted anomaly degree obtained by weighting the second anomaly degree with a corresponding relationship to the second position.

[0067] Therefore, based on the above-mentioned anomaly detection method, it is possible to detect movement-related anomalies of moving objects with higher accuracy.

[0068] Alternatively, the anomaly detection method may further include a second anomaly adjustment step, in which a third weighted anomaly is calculated based on the first anomaly and a third weighted coefficient that corresponds to the first driving model, and a fourth weighted anomaly is calculated based on the second anomaly and a fourth weighted coefficient that corresponds to the second driving model. In the determination step, the determination is performed based on the third weighted anomaly and the fourth weighted anomaly.

[0069] According to the above anomaly detection method, it is possible to determine whether there is an anomaly related to the movement of the moving body based on the third weighted anomaly obtained by weighting the first anomaly degree with a corresponding relationship to the first driving model, and the fourth weighted anomaly obtained by weighting the second anomaly degree with a corresponding relationship to the second driving model.

[0070] Therefore, based on the above-mentioned anomaly detection method, it is possible to detect movement-related anomalies of moving objects with higher accuracy.

[0071] Alternatively, in the information acquisition step, the method may also acquire first time information associated with the first location information and second time information associated with the second location information. The anomaly detection method further includes a third anomaly adjustment step, in which a fifth weighted anomaly is calculated based on the first anomaly and a fifth weighted coefficient of the first time represented by the first time information, and a sixth weighted anomaly is calculated based on the second anomaly and a sixth weighted coefficient of the second time represented by the second time information. In the determination step, the determination is performed based on the fifth weighted anomaly and the sixth weighted anomaly.

[0072] According to the above anomaly detection method, it is possible to determine whether there is an anomaly related to the movement of the moving body based on the fifth weighted anomaly obtained by weighting the first anomaly based on the attenuation coefficient corresponding to the first time when the first position information is obtained, and the sixth weighted anomaly obtained by weighting the first anomaly based on the attenuation coefficient corresponding to the second time when the second position information is obtained.

[0073] Therefore, based on the above-mentioned anomaly detection method, it is possible to detect movement-related anomalies of moving objects with higher accuracy.

[0074] Alternatively, in the model acquisition step, a third driving model can be acquired, which is the first driving model that has established a corresponding relationship with the manufacturer of the moving body. In the anomaly calculation step, a third anomaly is calculated, which is based on the first action information and the first anomaly of the third driving model. In the determination step, the determination is made based on the third anomaly.

[0075] Based on the above anomaly detection method, a third driving model corresponding to the manufacturer of the moving object can be used to determine whether there are any anomalies related to the movement of the moving object.

[0076] Therefore, based on the above-mentioned anomaly detection method, it is possible to detect movement-related anomalies of moving objects with higher accuracy.

[0077] Alternatively, the first action information may include velocity information representing the speed of the moving body and / or angular velocity information representing the angular velocity of the moving body.

[0078] Therefore, based on the above-mentioned anomaly detection method, it is possible to detect movement-related anomalies of a moving body based on its velocity and / or angular velocity.

[0079] Alternatively, in the output step, the output can be performed by generating an output image that overlays an image representing the result of the determination onto a map containing the location represented by the first location information, and displaying the output image on a monitor.

[0080] Therefore, according to the above-described anomaly detection method, when an anomaly related to the movement of a moving body occurs, the user of the above-described anomaly detection method can visually identify the occurrence of the situation.

[0081] Alternatively, the anomaly detection method may also include a model generation step, in which the first driving model is generated based on the first movement data.

[0082] Therefore, based on the above anomaly detection method, the first driving model can be generated.

[0083] Alternatively, in the model generation step, based on multiple mobile data including the first mobile data, one or more driving models including the first driving model can be generated, and each of the multiple mobile data can be classified into any one of multiple category destinations including one or more category destinations corresponding to each of the one or more driving models. Based on the mobile data classified into each of the one or more category destinations, each of the one or more driving models can be generated.

[0084] Therefore, based on the above anomaly detection method, it is possible to generate more than one driving model corresponding to each of the more than one classified destination.

[0085] Alternatively, the anomaly detection method may further include a weighted coefficient calculation step, in which a weighted coefficient representing the reliability of the driving model associated with each of the more than one driving models is calculated.

[0086] Therefore, based on the above anomaly detection method, a weighted coefficient representing reliability can be calculated for each of the multiple generated first-order driving models.

[0087] An anomaly detection device disclosed herein includes: an information acquisition unit that acquires first position information representing the position of a moving body and first action information associated with the first position information representing the action of the moving body; a model acquisition unit that, when the position represented by the first position information is a first specific position, acquires a first driving model associated with the first specific position, generated based on first movement data related to the past movement of at least one of more than one moving bodies including the moving body; an anomaly degree calculation unit that, when the position represented by the first position information is the first specific position, calculates a first anomaly degree representing the degree of anomaly related to the movement of the moving body based on the first action information and the first driving model; a determination unit that, based on the first anomaly degree, determines whether an anomaly related to the movement of the moving body exists; and an output unit that outputs the determination result to an external location.

[0088] According to the above-described anomaly detection device, it is possible to calculate a first degree of anomaly of the mobile body at a first specific location based on first position information associated with a first location indicating that the mobile body is traveling in a driving mode suitable for confirming the validity of the location information, first action information indicating the mobile body's action at a first specific location, and a first driving model associated with the first specific location. Based on the calculated first degree of anomaly, it is possible to determine whether an anomaly related to the movement of the mobile body exists. Therefore, according to the above-described anomaly detection device, it is possible to determine the existence of an anomaly related to the movement of the mobile body even if the obtained first location information has been tampered with.

[0089] Thus, according to the above-mentioned anomaly detection device, even if the location information of the moving object is tampered with by a malicious attacker, anomalies related to the movement of the moving object can still be detected.

[0090] One technical solution disclosed herein relates to a program for enabling a computer to perform anomaly detection processing, the anomaly detection processing comprising: an information acquisition step, acquiring first location information representing the position of a moving body and first action information associated with the first location information representing the action of the moving body; a model acquisition step, if the position represented by the first location information is a first specific position, acquiring a first driving model associated with the first specific position, generated based on first movement data related to the past movement of at least one of more than one moving bodies including the moving body; an anomaly degree calculation step, if the position represented by the first location information is the first specific position, calculating a first anomaly degree representing the degree of anomaly related to the movement of the moving body based on the first action information and the first driving model; a determination step, determining the presence or absence of an anomaly related to the movement of the moving body based on the first anomaly degree; and an output step, outputting the determination result to an external source.

[0091] According to the above procedure, based on the first position information indicating the movement of the mobile body associated with the first position information indicating that the mobile body is traveling in a driving mode suitable for confirming the validity of the position information, the first action information indicating the movement of the mobile body at the first specific position, and the first driving model associated with the first specific position, a first anomaly degree of the mobile body at the first specific position can be calculated, and based on the calculated first anomaly degree, the presence or absence of an anomaly related to the movement of the mobile body can be determined. Therefore, according to the above procedure, even if the obtained first position information is tampered with, it can be determined that an anomaly related to the movement of the mobile body exists.

[0092] Thus, according to the above procedure, even if the location information of a moving object is tampered with by a malicious attacker, anomalies related to the movement of that moving object can still be detected.

[0093] The following description, with reference to the accompanying drawings, illustrates a specific example of an anomaly detection system according to the present disclosure. The embodiments shown herein represent only one specific example of the present disclosure. Therefore, the values, shapes, constituent elements, configurations and connections of constituent elements, as well as steps (processes) and their order shown in the following embodiments are merely examples and are not intended to limit the present disclosure. Furthermore, the figures are schematic diagrams and not necessarily strictly illustrated. In the figures, substantially identical components are labeled with the same reference numerals, and repeated descriptions are omitted or simplified.

[0094] (Implementation Method)

[0095] The anomaly detection system involved in the implementation method will be described below.

[0096] This anomaly detection system is a system for detecting movement-related anomalies of the moving object that is the target of the detection.

[0097] <Composition>

[0098] Figure 1 This is a block diagram illustrating the configuration of the anomaly detection system 1 according to the implementation method.

[0099] like Figure 1 As shown, the anomaly detection system 1 includes an anomaly detection device 10, a moving body 20, a network 30, and a display device 40.

[0100] Network 30 is connected to multiple devices, including mobile body 20 and anomaly detection device 10, to enable communication between the connected devices. Network 30 is, for example, the Internet.

[0101] Display device 40 is connected to anomaly detection device 10 and displays images output from anomaly detection device 10. Display device 40 is, for example, a monitor.

[0102] Here, the display device 40 is described as an external device of the anomaly detection device. However, as another configuration example, the display device 40 may also be included as a component of the anomaly detection device 10.

[0103] In the anomaly detection system 1, the moving body 20 is the moving body that becomes the object of anomaly detection.

[0104] exist Figure 1 The diagram illustrates a single moving body 20, but the anomaly detection system 1 is not limited to a single moving body 20 that is the object of anomaly detection; it may also have two or more moving bodies 20. Furthermore, to avoid making the explanation overly complex, we will describe it using only one moving body 20.

[0105] The mobile body 20 is an autonomous driving device. Examples of mobile bodies 20 include autonomous vehicles used for the delivery of goods, autonomous vehicles used for the movement of people, autonomous cleaning robots, and autonomous security robots.

[0106] The mobile body 20 sends vehicle control signals to the anomaly detection device 10 sequentially via the network 30. The vehicle control signals include position information indicating the position of the mobile body 20 and action information associated with the position information indicating the actions of the mobile body 20.

[0107] Action information, for example, is velocity information representing the speed of the moving body 20 and / or angular velocity information representing the angular velocity of the moving body 20.

[0108] For example, the mobile body 20 may also send vehicle control signals exchanged sequentially within the mobile body network constructed within the mobile body 20 when the mobile body 20 is autonomously driving to the anomaly detection device 10.

[0109] Mobile body 20 can also use VPN (Virtual Private Network) communication to send vehicle control signals to anomaly detection device 10. Thus, mobile body 20 can more securely send vehicle control signals to anomaly detection device 10.

[0110] The anomaly detection device 10 detects movement-related anomalies of the mobile body 20 based on vehicle control signals sent from the mobile body 20.

[0111] The anomaly detection device 10 can be implemented, for example, in a computer device equipped with a processor (e.g., a CPU (Central Processing Unit)) and memory (e.g., ROM (Read Only Memory) and / or RAM (Random Access Memory)), by the processor executing a program stored in the memory. In this case, the anomaly detection device 10 can also be implemented, for example, in one or more computer devices that can communicate with each other in a cloud or locally deployed environment.

[0112] like Figure 1 As shown, the anomaly detection device 10 includes an information acquisition unit 11, a model acquisition unit 12, an anomaly degree calculation unit 13, an anomaly degree adjustment unit 14, a judgment unit 15, an output unit 16, a model storage unit 17, and a model generation unit 18.

[0113] The information acquisition unit 11 receives vehicle control signals transmitted sequentially from the mobile body 20, and acquires the position information and motion information contained in the received vehicle control signals sequentially. More specifically, the information acquisition unit 11 generates and stores vehicle information, including the position information and motion information contained in the vehicle control signals, sequentially based on the received vehicle control signals.

[0114] The model storage unit 17 stores, for each of the more than one predetermined specific locations, one or more driving models generated based on movement data related to the past movement of at least one of the more than one moving bodies including the moving body 20, associated with that specific location.

[0115] One or more driving models, each for example, are the velocity and / or angular velocity shifts (time series data) of at least one moving body under a specific driving mode at a specific location.

[0116] Here, it is described that the model storage unit 17 stores one or more driving models generated by the model generation unit 18 (described later). However, as another configuration example, the model storage unit 17 may also store one or more driving models generated by an external device of the anomaly detection device 10. In this case, the anomaly detection device 10 is configured without the model generation unit 18.

[0117] When the location indicated by the location information obtained by the information acquisition unit 11 is a specific location among one or more predetermined specific locations, the model acquisition unit 12 retrieves the driving model associated with that specific location from the model storage unit 17.

[0118] As described above, the information acquisition unit 11 acquires position information and motion information sequentially. Therefore, whenever the position represented by the position information acquired by the information acquisition unit 11 becomes a specific position among more than one predetermined specific position, the model acquisition unit 12 acquires the driving model associated with that specific position.

[0119] When the location represented by the location information obtained by the information acquisition unit 11 is a specific location among one of more than one predetermined specific locations, the anomaly calculation unit 13 calculates an anomaly degree representing the degree of anomaly related to the movement of the moving body 20 based on (1) the motion information associated with the location information obtained by the information acquisition unit 11 and (2) the driving model associated with the specific location obtained by the model acquisition unit 12.

[0120] As described above, the information acquisition unit 11 acquires position information and motion information sequentially. Therefore, the anomaly calculation unit 13 calculates the anomaly degree each time the position represented by the position information acquired by the information acquisition unit 11 becomes a specific position among one or more predetermined specific positions.

[0121] The anomaly adjustment unit 14 adjusts the anomaly level calculated by the anomaly calculation unit 13. Here, it is explained that the anomaly adjustment unit 14 calculates a weighted anomaly level by weighting each anomaly level based on a weighting coefficient corresponding to that anomaly level. More specifically, for weighting coefficients that are 0 or higher and 1 or lower, the anomaly adjustment unit 14 calculates the weighted anomaly level by multiplying each anomaly level by its corresponding weighting coefficient.

[0122] The weighting coefficient can be, for example, a weighting coefficient that has established a correspondence with one or more specific locations, a weighting coefficient that has established a correspondence with one or more driving models, or a weighting coefficient based on the time when the location information is obtained by the information acquisition unit 11.

[0123] As mentioned above, the anomaly calculation unit 13 calculates the anomaly degree sequentially. Therefore, the anomaly adjustment unit 14 calculates the weighted anomaly degree each time the anomaly calculation unit 13 calculates the anomaly degree.

[0124] The determination unit 15 determines whether there is an anomaly related to the movement of the moving body 20 based on the weighted anomaly degree calculated by the anomaly degree adjustment unit 14.

[0125] As mentioned above, the anomaly adjustment unit 14 calculates the weighted anomaly degree sequentially. Therefore, each time the anomaly adjustment unit 14 calculates the weighted anomaly degree, the determination unit 15 determines whether there is an anomaly related to the anomaly of the moving body 20 based on one or more weighted anomalies calculated by the anomaly adjustment unit 14 up to that point in time.

[0126] The output unit 16 outputs the determination result of the determination unit 15 to an external device. Here, as an example including but not limited to, the output unit 16 generates an image representing the determination result and outputs the generated image to the display device 40, which is an external device. As another example, the output unit 16 may also generate voice representing the determination result and output the generated voice to a speaker, which is also an external device.

[0127] The model generation unit 18 generates one or more driving models stored in the model storage unit 17.

[0128] Here, we will explain the situation as follows: The model generation unit 18 uses movement data related to the past movement of at least one of the more than one moving bodies including the moving body 20, which is obtained during a predetermined model learning period, to enable the machine learning model corresponding to each of the more than one driving model to learn and generate more than one driving model.

[0129] Furthermore, since the motion data related to the past motion of at least one mobile body is updated according to the daily use of at least one mobile body, the model generation unit 18 can also periodically generate and update more than one driving model.

[0130] Figure 2 This is a block diagram showing the structure of the model generation unit 18.

[0131] like Figure 2 As shown, the model generation unit 18 includes a mobile data storage unit 51, an information acquisition unit 52, a classification unit 53, a driving model learning unit 54, and a weighted coefficient calculation unit 55.

[0132] The mobile data storage unit 51 stores mobile data related to the past movement of at least one of the more than one mobile bodies including the mobile body 20.

[0133] Here, the movement data is defined as a vehicle control signal that includes position information indicating the position of at least one moving body and motion information associated with the position information indicating the action of at least one moving body.

[0134] The information acquisition unit 52 acquires the vehicle control signals acquired during the predetermined model learning period from the mobile data storage unit 51.

[0135] Based on predetermined setting information, the classification unit 53 classifies the vehicle control signals acquired by the information acquisition unit 52 into any one of multiple classification destinations, including one or more classification destinations corresponding to one or more driving models, according to the driving position and / or driving mode of at least one moving body.

[0136] The driving model learning unit 54 generates one or more driving models based on vehicle control signals for each of the more than one classified destination. Here, it is explained that the driving model learning unit 54 generates one or more driving models by using vehicle control signals for each of the more than one classified destination to learn the machine learning models corresponding to each of the more than one driving models.

[0137] Furthermore, the driving model learning unit 54 stores one or more generated driving models in the model storage unit 17.

[0138] The weighting coefficient calculation unit 55 calculates a weighting coefficient representing the reliability of each of the more than one driving models generated by the driving model learning unit 54.

[0139] Furthermore, the weighting coefficient calculation unit 55 stores the calculated weighting coefficients in association with the corresponding driving model in the model storage unit 17.

[0140] <Data Structures>

[0141] Next, the structure of the data processed by the anomaly detection device 10 will be explained.

[0142] Figure 3 This is a schematic diagram illustrating an example of a data structure generated and stored by the information acquisition unit 11 based on a vehicle control signal transmitted from the mobile body 20. The data structure includes position information indicating the position of the mobile body 20 and motion information indicating the movement of the mobile body 20 associated with the position information.

[0143] The information acquisition unit 11 acquires vehicle control signals sent from the mobile body 20 in the form of protocols such as CAN (Controller Area Network), FlexRay, or ROS (Robot Operating System), and parses the acquired vehicle control signals to generate vehicle information.

[0144] like Figure 3 As shown, the vehicle information stored in the information acquisition unit 11 is composed of a timestamp indicating the time when the vehicle control signal is generated, a vehicle ID that serves as an identifier for the mobile body 20, motion information indicating the action of the mobile body 20, and position information indicating the location of the mobile body 20.

[0145] Here, the motion information is defined as the velocity and angular velocity of the moving body 20.

[0146] Vehicle information may also include information related to the operation of equipment while the mobile body 20 is in motion (e.g., information indicating the status of turn signals, hazard warning lights, etc.).

[0147] Figure 4 The model acquisition unit 12 adds vehicle information to the vehicle information stored in the information acquisition unit 11 for obtaining driving model information from the model storage unit 17 and storing it.

[0148] like Figure 4 As shown, the vehicle information stored in the model acquisition unit 12 includes, in addition to timestamps, vehicle IDs, action information, and location information, also includes intermediate points that represent the intermediate points that the mobile body 20 passes through during its movement, the starting point of the mobile body 20's movement, the destination of the mobile body 20's movement, and driving data classifications that represent the vehicle information being classified.

[0149] The model acquisition unit 12 generates intermediate points, starting points, and destinations by parsing the position and motion information of a moving body 20 throughout a movement.

[0150] In addition, the model acquisition unit 12, for example, parses the position information and motion information of one or more series of movements included in a movement of a moving body 20, and classifies a series of vehicle control signals corresponding to the position information and motion information of the series of movements into one of multiple classification destinations.

[0151] Here, the model acquisition unit 12 may also classify the first series of vehicle control signals corresponding to the first series of position information and motion information contained in a movement into one of the multiple classification destinations, and classify the second series of vehicle control signals corresponding to the second series of position information and motion information contained in a movement into another of the multiple classification destinations.

[0152] Multiple classification destinations include, for example, classification destination A01-P01 indicating that the mobile body 20 is in region A01 of one or more predetermined specific locations and is in driving mode P01 as shown in driving mode P01; and classification destination A02-P02 indicating that the mobile body 20 is in region A02 of one or more predetermined specific locations and is in driving mode P02 as shown in driving mode P02.

[0153] Figure 5 The determination unit 15 adds information indicating the presence or absence of anomalies related to the movement of the moving body 20 to the vehicle information stored in the model acquisition unit 12 and stores the vehicle information.

[0154] like Figure 5 As shown, the vehicle information stored in the determination unit 15, in addition to timestamps, vehicle IDs, action information, location information, intermediate points passed, origin, destination, and driving data classification, also consists of a series of vehicle information data sequences that are interconnected, including the anomaly degree calculated by the anomaly degree calculation unit 13, the anomaly determination of the presence or absence of anomalies related to the movement of the moving body 20 determined by the determination unit 15, and the data that identify the objects of the determination of the presence or absence of anomalies related to the movement of the moving body 20 by the determination unit 15.

[0155] <Work>

[0156] The anomaly detection device 10 performs model learning processing to generate a driving model and anomaly detection processing to detect anomalies related to the movement of the moving body 20.

[0157] The model learning process and anomaly detection process performed by the anomaly detection device 10 will be described below with reference to the accompanying drawings.

[0158] Figure 6 This is a flowchart of the model learning process performed by the anomaly detection device 10.

[0159] like Figure 6 As shown, after the model learning process begins, the information acquisition unit 52 acquires the mobile data (here, vehicle control signal) acquired during the predetermined model learning period from the mobile data storage unit 51 (step S10).

[0160] After acquiring the vehicle control signal, the classification unit 53, based on predetermined setting information, classifies the vehicle control signal acquired by the information acquisition unit 52 into any one of a plurality of classification destinations, including one or more classification destinations corresponding to one or more driving models, according to the driving position and / or driving mode of at least one moving body (step S20).

[0161] Here, it is explained that the classification unit 53 classifies the vehicle control signal into any one of a plurality of classification destinations, including one or more classification destinations corresponding to one or more driving models, based on the driving position and driving mode of at least one moving body.

[0162] Examples of specific classification methods used by classification department 53 will be provided later. Figure 7 The flowchart is used for illustration.

[0163] After classifying the vehicle control signals, the driving model learning unit 54 uses the vehicle control signals classified to one or more classification destinations to enable the machine learning models corresponding to one or more driving models to learn and generate each of the one or more driving models (step S30).

[0164] After generating one or more driving models, the weighting coefficient calculation unit 55 calculates a weighting coefficient representing the reliability of each driving model in the one or more driving models (step S40).

[0165] A specific example of weighting coefficient calculation for the weighting coefficient calculation unit 55 will be provided later. Figure 11A , Figure 11B The flowchart is used for illustration.

[0166] After calculating the weighting coefficients of one or more driving models, the model storage unit 17 stores the driving models generated by the driving model learning unit 54 and the weighting coefficients generated by the weighting coefficient calculation unit 55 in a mutually related manner (step S50).

[0167] After step S50 is completed, the model learning process ends.

[0168] Figure 7 This is a flowchart illustrating a specific example of the classification process performed by the classification unit 53 in step S20.

[0169] like Figure 7 As shown, after the classification process begins, the classification unit 53 reads the predetermined setting information related to the driving model (step S110).

[0170] The configuration information is used to categorize the data used according to the characteristics of the driving model to be generated.

[0171] For example, if the driving model to be generated is a driving model used to detect the difference between the estimated movement path based on the shift of speed and angular velocity in driving modes such as right turn, left turn, straight, and U-turn at intersections and the shift of position shown by the position information at the intersections, in order to capture the situation where the position information has been tampered with, the setting information includes information indicating that the position to be classified is an intersection and that the driving mode to be classified is right turn, left turn, straight, and U-turn.

[0172] In this embodiment, the driving model to be generated is described as a driving model used to detect the difference between the estimated movement path based on the shift of speed and angular velocity in driving modes such as right turn, left turn, straight, and U-turn at intersections and the shift of position shown by the position information at the intersections, etc., in order to capture the situation where the position information has been tampered with.

[0173] After reading the setting information, the classification unit 53 determines the travel location of the moving body whose location is represented by the location information based on the location information contained in the vehicle control signal and using information related to the path and / or location set in advance using map information, etc. (step S120).

[0174] Here, determining the travel location of a mobile body whose location information represents its position refers to the act of judging whether the travel location of the mobile body 20, whose location information represents its position, matches a path and / or location set in advance using map information, etc. In other words, it refers to the act of judging whether the travel location of the mobile body 20, whose location information represents its position, is a specific location.

[0175] The so-called path and / or location set in advance using map information, for example, is information used to determine whether the travel location of the moving body 20, whose location is indicated by location information, is an intersection, a one-way street, or a pedestrian-only road.

[0176] In this embodiment, information related to a path and / or location pre-defined using map information or the like is described as information used to determine whether it is an intersection.

[0177] Figure 8 This is a schematic diagram showing whether the location of the moving body 20, whose location information is determined by the classification unit 53, is an intersection.

[0178] As a method for determining whether the travel location of the moving body 20, whose location information indicates its location, is an intersection, for example, Figure 8 As shown, a method can be considered to determine whether the travel location of the moving body 20, whose location is indicated by location information, is within a predetermined range from the center of the intersection.

[0179] Return to Figure 7 , and then we will continue to explain the classification process.

[0180] After determining the travel location of the mobile body whose location information indicates the location, the classification unit 53 determines the travel mode of the mobile body 20 at that location based on the shift of the location indicated by the location information of the mobile body 20 at that location (step S130).

[0181] In this embodiment, the driving mode to be determined is assumed to be any one of going straight, turning right, turning left, and making a U-turn.

[0182] Figure 9 , Figure 10 This is a schematic diagram illustrating the driving mode of the moving body 20 whose position is determined by the classification unit 53.

[0183] As a method for determining the driving mode of the moving body 20, whose position is represented by position information, based on the movement of the moving body 20, for example, such as Figure 8 As shown, the driving mode of the mobile body 20 can be determined by calculating the difference between the azimuth angle of the mobile body 20 when it enters the area within a predetermined range from the center of the intersection when it enters the area indicated by the location information and the azimuth angle when it leaves the area, and by applying a threshold judgment to the calculated difference.

[0184] Alternatively, as another method, for example, such as Figure 9 As shown, a method called Shaplet, which processes time-series waveform data, can also be used to learn the trajectories of multiple moving bodies in each driving mode in advance, determine which driving mode the trajectory of the moving body 20 whose position information represents is similar to, and thus determine the driving mode of the moving body 20 whose position information represents is.

[0185] Alternatively, as another method, for example, such as Figure 10 As shown, the velocity and / or angular velocity shifts of multiple moving bodies in each driving mode can also be clustered in advance to determine which driving mode the velocity and / or angular velocity shift of the moving body 20 whose position is represented by the position information is similar to, thereby determining the driving mode of the moving body 20 whose position is represented by the position information.

[0186] Return to Figure 7 , and then we will continue to explain the classification process.

[0187] After determining the travel location and travel mode of the mobile body 20 whose location information represents the location, the classification unit 53 classifies the vehicle control signal containing the location information into any one of multiple classification destinations based on the determined travel location and travel mode (step S140).

[0188] After step S140 is completed, the classification process ends.

[0189] Figure 11A , Figure 11B The flowcharts are respectively an example of the weighted coefficient calculation performed by the weighted coefficient calculation unit 55 in step S40, showing the first weighted coefficient calculation process and the second weighted coefficient calculation process.

[0190] Here, the first weighting coefficient calculation process is a process in which the weighting coefficient calculation unit 55 calculates the weighting coefficients based on the amount of data of the vehicle control signal, which is used as learning data for the machine learning model to learn. The second weighting coefficient calculation process is a process in which the accuracy of the learned machine learning model is verified and the weighting coefficients are calculated based on the verification result.

[0191] In the first and second weighting coefficient calculation processes, the weighting coefficients to be calculated are values ​​greater than or equal to 0 and less than or equal to 1. Furthermore, the weighting coefficients calculated through these processes are used to calculate the weighted anomaly degree by the anomaly degree adjustment unit 14.

[0192] Figure 11A This is a flowchart of the first weighted calculation process performed by the weighted coefficient calculation unit 55.

[0193] After the first weighting coefficient calculation process begins, the weighting coefficient calculation unit 55 selects an unselected classification destination from among more than one classification destinations (step S210). Here, an unselected classification destination refers to a classification destination that has not yet been selected in the loop process formed from the process of step S210 to the "yes" process of step S280 described later.

[0194] After selecting an unselected category destination, the weighted coefficient calculation unit 55 determines whether the number of vehicle control signals contained in the selected category destination, i.e. the number of learning data, is above the first threshold (step S220).

[0195] In the processing of step S220, if the number of data in the learning data is greater than or equal to the first threshold (step S220: Yes), the weighting coefficient calculation unit 55 calculates the weighting coefficient to 1 (step S230).

[0196] In the process of step S220, if the number of data in the learning data is not greater than or equal to the first threshold (step S220: No), the weighting coefficient calculation unit 55 further determines whether the number of data in the learning data is greater than or equal to the second threshold, which is smaller than the first threshold (step S240).

[0197] In step S240, if the number of data points in the learning data is greater than or equal to the second threshold (step S240: Yes), the weighting coefficient calculation unit 55 calculates the weighting coefficients to be values ​​greater than 0 and less than 1. At this time, the weighting coefficient calculation unit 55 calculates the weighting coefficients based on the number of data points in the learning data so that the larger the number of data points, the larger the value of the weighting coefficients (step S250).

[0198] In the processing of step S240, if the number of data in the learning data is not greater than the second threshold (step S240: No), the weighting coefficient calculation unit 55 calculates the weighting coefficient as 0 (step S260).

[0199] When the processing in step S230, step S250, and step S260 is completed, the weighting coefficient calculation unit 55 associates the calculated weighting coefficient with the corresponding driving model and stores it in the model storage unit 17 (step S270).

[0200] Figure 12 This is a schematic diagram illustrating an example of how the model storage unit 17 stores driving models and weighting coefficients in association.

[0201] like Figure 12 As shown, the model storage unit 17 stores the driving model and weighting coefficients in an interconnected manner.

[0202] Return to Figure 11A Let's continue with the explanation of the first weighted calculation process.

[0203] After the calculated weighting coefficients are stored in the model storage unit 17, the weighting coefficient calculation unit 55 determines whether there is an unselected classification destination among the more than one classification destination (step S280).

[0204] If there is an unselected classification destination in the process of step S280 (step S280: Yes), the first weighted calculation process will re-enter the process of step S210.

[0205] In step S280, if there is no unselected classification destination (step S280: No), the first weighted calculation process ends.

[0206] Figure 11B This is a flowchart of the second weighted calculation process performed by the weighted coefficient calculation unit 55.

[0207] After the second weighting coefficient calculation process begins, the weighting coefficient calculation unit 55 selects an unselected classification destination from among more than one classification destinations (step S310). Here, an unselected classification destination refers to a classification destination that has not yet been selected in the loop process formed from the process of step S310 to the "yes" process of step S380 described later.

[0208] After selecting an unselected classification destination, the weighted coefficient calculation unit 55 divides the vehicle control signals contained in the selected classification destination into learning data for the machine learning model to learn and verification data for using the machine learning model learned from the learning data to verify the accuracy of the data classification in the selected classification destination (step S312).

[0209] Then, the driving model learning unit 54 uses the learning data separated by the processing in step S312 to train the machine learning model, and uses the machine learning model trained by the learning data and the validation data to verify the accuracy of the data classification in the classification destination (step S314).

[0210] After verifying the accuracy of the data classification, the weighted coefficient calculation unit 55 determines whether the accuracy of the verified data classification is above the first threshold (step S320).

[0211] In the processing of step S320, if the accuracy of the verified data classification is above the first threshold (step S320: Yes), the weighting coefficient calculation unit 55 calculates the weighting coefficient to 1 (step S330).

[0212] In step S320, if the accuracy of the verified data classification is not greater than or equal to the first threshold (step S320: No), the weighting coefficient calculation unit 55 further determines whether the accuracy of the verified data classification is greater than or equal to the second threshold, which is smaller than the first threshold (step S340).

[0213] In step S340, if the accuracy of the verified data classification is above the second threshold (step S340: Yes), the weighting coefficient calculation unit 55 calculates the weighting coefficient to a value greater than 0 and less than 1. At this time, the weighting coefficient calculation unit 55 calculates the weighting coefficient based on the accuracy of the verified data classification so that the higher the accuracy, the larger the value of the weighting coefficient (step S350).

[0214] In step S340, if the accuracy of the verified data classification is not above the second threshold (step S340: No), the weighting coefficient calculation unit 55 calculates the weighting coefficient as 0 (step S360).

[0215] When the processing in step S330, step S350, and step S360 is completed, the weighting coefficient calculation unit 55 associates the calculated weighting coefficient with the corresponding driving model and stores it in the model storage unit 17 (step S370).

[0216] After the calculated weighting coefficients are stored in the model storage unit 17, the weighting coefficient calculation unit 55 determines whether there is an unselected classification destination among the more than one classification destination (step S380).

[0217] If there is an unselected classification destination in the process of step S380 (step S380: Yes), the second weighted calculation process will re-enter the process of step S310.

[0218] In the process of step S380, if there is an unselected classification destination (step S380: No), the second weighted calculation process ends.

[0219] In addition, the weighting coefficient calculation unit 55 can also calculate the weighting coefficients by performing processes other than the first weighting calculation process or the second weighting calculation process described above.

[0220] For example, the distribution of vectors representing the velocity and / or angular velocity shifts of multiple moving bodies in various driving modes in the feature space is as follows: Figure 10 In the case shown by the classification results, the weighting coefficient calculation unit 55 can calculate the weighting coefficients in the following manner: for driving models of driving patterns corresponding to classes (in this case, classes (1) and (3)) with smaller distance differences from vectors in the same class in the feature space, the weighting coefficients are larger; for driving models of driving patterns corresponding to classes (in this case, classes (2) and (4)) with larger distance differences from vectors in the same class in the feature space, the weighting coefficients are smaller.

[0221] In this case, the weighting coefficient calculation unit 55 can, for example, calculate larger weighting coefficients (0.9 and 0.9 in this case) for driving models of driving patterns corresponding to class (1) and class (3) with smaller distance differences from vectors of the same class in the feature space, and smaller weighting coefficients (0.3 and 0.1 in this case) for driving models of driving patterns corresponding to class (4) and class (2) with larger distance differences from vectors of the same class in the feature space.

[0222] Figure 13 This is a flowchart of the anomaly detection process performed by the anomaly detection device 10.

[0223] like Figure 13As shown, after the anomaly detection process begins, the information acquisition unit 11 waits until a new vehicle control signal containing location information and motion information is sent from the mobile body 20 (repeated step S405: No). When a vehicle control signal is sent from the mobile body 20 (step S405: Yes), the sent vehicle control signal is received, and the location information and motion information contained in the vehicle control signal are obtained according to the received vehicle control signal (step S410).

[0224] After obtaining the location information and motion information, the model acquisition unit 12 determines whether the location represented by the obtained location information is a specific location among one or more predetermined specific locations (step S415).

[0225] In step S415, if the location indicated by the acquired location information is one of more than one predetermined specific locations (step S415: Yes), the model acquisition unit 12 acquires the driving model associated with the location indicated by the acquired location information from the model storage unit 17 (step S420). At this time, the model acquisition unit 12 determines the driving mode of the mobile body 20 at that location based on the shift in position indicated by the location information of the mobile body 20 at that location, and acquires a driving model with a driving mode consistent with the determined driving mode. That is, the model acquisition unit 12 acquires from the model storage unit 17 the driving model with a driving mode consistent with the determined driving mode from one of more than one driving models associated with the location indicated by the acquired location information.

[0226] After obtaining the driving model, the anomaly calculation unit 13 calculates the anomaly degree, which represents the degree of anomaly related to the movement of the moving body 20, based on the motion information obtained in step S410 and the driving model obtained in step S420 (step S425).

[0227] Figure 14 This is a schematic diagram illustrating an example of how the anomaly calculation unit 13 calculates anomalies related to the movement of the moving body 20, specifically anomalies related to the position information of the moving body 20.

[0228] exist Figure 14 In the diagram, the dashed arrows represent the trajectory of the mobile body 20 as shown by the location information sent by the mobile body 20, while the solid arrows represent the actual trajectory of the mobile body 20.

[0229] Figure 14 The example shown illustrates a situation where the location information sent by mobile body 20 is maliciously altered by an attacker, resulting in the location information being rewritten as if mobile body 20 is going straight, even though it is actually turning right.

[0230] like Figure 14As shown, if the location information sent from the mobile body 20 is tampered with, the location of the mobile body 20 shown in the location information sent from the mobile body 20 will differ from the actual location of the mobile body 20.

[0231] Here, as a method for detecting tampering of location information, one could consider using information such as velocity and angular velocity, which are considered difficult for attackers to tamper with, to determine the anomaly (tampering) of location information.

[0232] One approach is to use odometry, a method that estimates position based on velocity and angular velocity, and identifies anomalies when the difference between the estimated position and the position shown in the position information exceeds a certain threshold. However, this simple method can lead to significant errors in the estimated position based on velocity and angular velocity due to factors such as friction, potentially resulting in over-detection of anomalies.

[0233] As an example of a countermeasure, such as Figure 14 As shown, by learning the speed and angular velocity shifts (time series data) of a specific driving mode at a specific location and confirming the difference between the learned speed and angular velocity shifts (time series data) and the actual speed and angular velocity shifts (time series data), it is possible to suppress the occurrence of abnormal over-detection.

[0234] exist Figure 14 In the example shown, the model acquisition unit 12 determines the driving mode of the mobile body 20 at that location based on the position shift shown by the position information of the mobile body 20, and determines a driving model of the driving mode that is consistent with the determined driving mode. The anomaly calculation unit 13 calculates the anomaly related to the position information of the mobile body by comparing the speed and angular velocity shift (time series data) shown by the acquired driving model with the speed and angular velocity shift (time series data) sent from the mobile body 20.

[0235] In this case, for example, the difference between time series data can be calculated using a known method such as Dynamic Time Warping (DTW), and the calculated difference can be thresholded to determine the anomaly.

[0236] Return to Figure 13 Next, we will explain the abnormal detection and handling process.

[0237] After calculating the anomaly degree, the anomaly degree adjustment unit 14 adjusts the calculated anomaly degree. That is, the anomaly degree adjustment unit 14 calculates the weighted anomaly degree by multiplying the calculated anomaly degree by the weighting coefficient corresponding to the anomaly degree (step S430).

[0238] At this time, the anomaly adjustment unit 14 can calculate the weighted anomaly degree, for example, by multiplying the calculated anomaly degree by a weighting coefficient that establishes a correspondence with the position represented by the position information obtained in the processing of step S410, or by multiplying the calculated anomaly degree by a weighting coefficient that establishes a correspondence with the driving model obtained in the processing of step S420.

[0239] Furthermore, for example, if the vehicle control signal sent from the mobile body 20 includes time information associated with the location information, the information acquisition unit 11 may also acquire the time information associated with the location information in the processing of step S410, and the anomaly adjustment unit 14 may calculate the weighted anomaly degree by multiplying the calculated anomaly degree by a weighting coefficient based on the time represented by the time information acquired in the processing of step S410.

[0240] After calculating the weighted anomaly degree, the determination unit 15 determines whether there is an anomaly related to the movement of the moving body 20 based on the calculated weighted anomaly degree (step S435). Here, if more than one weighted anomaly degree is calculated in the loop process formed from the "yes" process in step S405 to the "no" process in step S455 described later, the determination unit 15 determines whether there is an anomaly related to the movement of the moving body 20 based on the more than one weighted anomaly degree.

[0241] The determination unit 15 may, for example, calculate the sum of more than one weighted anomalies, and if the calculated sum is above a threshold, determine that there is an anomaly related to the movement of the moving body 20.

[0242] Alternatively, the determination unit 15 may, for example, calculate the cumulative anomaly Z based on one or more weighted anomalies in order to make the influence of earlier (older) weighted anomalies smaller and the influence of newer (most recent) weighted anomalies larger, using the attenuation coefficient β(t), and determine that there is an anomaly related to the movement of the moving body 20 if the calculated cumulative anomaly Z is above a threshold.

[0243] Here, the attenuation coefficient β(t) is set in such a way that the larger the difference from the current time becomes, the smaller its value becomes.

[0244] In this case, if the weighted anomaly at time t is set as z(t), the cumulative anomaly Z can be calculated, for example, by Z=β(t1)×z(t1)+ β(t2)×z(t2)+ β(t3)×z(t3)+…

[0245] If, after processing in step S435, an anomaly related to the movement of the moving body 20 is determined (step S440: Yes), the output unit 16 generates an image representing the determination result (step S445). Furthermore, the output unit 16 displays a map of the generated image on the display device 40 (step S450).

[0246] At this time, for example, in order to notify the user using the anomaly detection device 10 (e.g., the monitor monitoring the moving body 20) of the judgment result in a form that is easy to understand as an anomaly, the output unit 16 may also generate an image that visualizes various information.

[0247] As an example of various types of information, it could be information used to display the shift of the position of the moving body 20 estimated based on its speed and angular velocity and the shift of the position of the moving body 20 indicated by the position information sent from the moving body 20 on a corresponding map. For example, it could be information used to display the shift of the speed and angular velocity of the moving body 20 during its movement and the shift of the speed and angular velocity of the moving body 20 in the past side by side. For example, if the vehicle control signal sent from the moving body 20 contains camera images captured by the moving body 20, it could also be information used to display an image obtained by cropping the corresponding part of the camera image.

[0248] Figure 15 This is a schematic diagram showing an example of an image displayed by the display device 40 caused by the output unit 16.

[0249] like Figure 15 As shown, the output unit 16 causes the image displayed by the display device 40 to include the travel trajectory of the mobile body 20, which is indicated by the position information sent from the mobile body 20 by the dashed arrow, and the travel trajectory of the mobile body 20, which is indicated by the solid arrow, which is indicated by the travel model of the travel mode that is consistent with the travel mode determined according to the speed and angular velocity sent from the mobile body 20.

[0250] In addition, the output unit 16 displays the image on the display device 40, showing the speed and angular velocity of the moving body 20 during its movement, and the speed and angular velocity of the moving body 20 in the past, that is, the speed / angular velocity of the moving body 20 shown in the driving model.

[0251] Additionally, the output unit 16 displays a button embedded in the image shown on the display device 40, which points to a camera image captured by the moving body 20 at the corresponding location.

[0252] By displaying such an image on the display device 40, for example, a monitor who is monitoring the movement of the moving body 20 can identify abnormalities in the movement of the moving body 20, prompting the monitor to begin an analysis related to the movement of the moving body 20.

[0253] Return to Figure 13 Next, we will explain the abnormal detection and handling process.

[0254] In the process of step S415, if the location indicated by the obtained location information is not one of the more than one predetermined specific locations (step S415: No), and if no abnormality related to the movement of the moving body 20 is determined through the process of step S435 (step S440: No), the information acquisition unit 11 determines whether the movement of the moving body 20 has ended (step S455).

[0255] In the process of step S455, if it is determined that the movement of the moving body 20 has not ended (step S455: No), the abnormality detection process re-enters the process of step S405.

[0256] If the process in step S450 ends, and if in step S455 it is determined that the movement of the moving body 20 has ended (step S455: Yes), the abnormality detection process ends.

[0257] <Consideration>

[0258] According to the anomaly detection device 10 with the above structure, it can calculate the anomaly degree of the mobile body 20 at a specific location based on the location information of the mobile body 20 indicating that the mobile body 20 is traveling in a driving mode suitable for confirming the validity of the location information, the motion information indicating the movement of the mobile body 20 at a specific location, and the driving model associated with that specific location. Based on the calculated anomaly degree, it can determine whether there is an anomaly related to the movement of the mobile body 20. Therefore, according to the anomaly detection device 10, it is possible to determine whether there is an anomaly related to the movement of the mobile body 20 when the obtained location information has been tampered with.

[0259] Thus, according to the above-mentioned anomaly detection device 10, even if the location information of the moving body 20 is tampered with by a malicious attacker, anomalies related to the movement of the moving body 20 can be detected.

[0260] (Replenish)

[0261] As described above, examples of the technology disclosed in this application have been presented based on implementation methods. However, this disclosure is not limited to these implementation methods. Various modifications to these implementation methods that can be conceived by those skilled in the art, as well as combinations of different implementation methods or elements of modifications, may be included within the scope of one or more technical solutions of this disclosure, as long as they do not depart from the spirit of this disclosure.

[0262] (1) One or more mobile bodies including mobile body 20 may sometimes use different autonomous driving algorithms depending on the manufacturer. Therefore, sometimes the driving modes of one or more mobile bodies may differ between different manufacturers.

[0263] Therefore, the classification unit 53 can classify the vehicle control signals into different classification destinations according to the manufacturer, the driving model learning unit 54 can generate driving models according to the manufacturer, and the model storage unit 17 can store the driving models according to the manufacturer.

[0264] In this case, the model storage unit 17 stores the driving model and establishes a correspondence with the manufacturer, and the model acquisition unit 12 retrieves the driving model that has established a correspondence with the manufacturer of the mobile body 20 from the model storage unit 17.

[0265] In addition, in this case, the weighting coefficient calculation unit 55 can also calculate the weighting coefficient independently for each driving mode, according to the manufacturer.

[0266] Furthermore, since the classification unit 53 categorizes vehicle control signals into different manufacturer-specific destinations, the number of vehicle control signals at each destination can sometimes be small. In such cases, the classification unit 53 may also merge vehicle control signals from multiple locations with similar driving modes into the same destination. That is, for example, the classification unit 53 may also categorize vehicle control signals for driving modes P01 and P02 in regions A01 and A02 into the same destination. Moreover, in this case, the classification unit 53 may also employ hierarchical clustering analysis from machine learning to classify the vehicle control signals.

[0267] (2) The overall or specific technical solutions of this disclosure can also be implemented by a system, apparatus, method, integrated circuit, program, or a non-transitory recording medium such as a computer-readable CD-ROM. In addition, they can be implemented by any combination of systems, apparatus, methods, integrated circuits, programs, and non-transitory recording media. For example, this disclosure can also be implemented as a program for causing a computer device to perform the processing performed by the anomaly detection device 10.

[0268] Industrial availability

[0269] This disclosure can be widely used in anomaly detection devices for detecting vehicle anomalies, etc.

[0270] Explanation of reference numerals in the attached figures

[0271] 1 Anomaly detection system; 10 Anomaly detection device; 11, 52 Information acquisition unit; 12 Model acquisition unit; 13 Anomaly degree calculation unit; 14 Anomaly degree adjustment unit; 15 Judgment unit; 16 Output unit; 17 Model storage unit; 18 Model generation unit; 20 Moving body; 30 Network; 40 Display device; 51 Moving data storage unit; 53 Classification unit; 54 Driving model learning unit; 55 Weighted coefficient calculation unit.

Claims

1. An anomaly detection method, comprising: The information acquisition step involves acquiring first position information representing the position of the moving body and first action information associated with the first position information representing the action of the moving body; The model acquisition step involves obtaining a first driving model associated with the first specific location, where the location represented by the first location information is a first specific location. This model is generated based on first movement data related to the past movement of at least one of the more than one moving bodies that includes the moving body. The anomaly calculation step involves calculating a first anomaly degree, representing the degree of anomaly related to the movement of the moving body, based on the first action information and the first driving model, when the position represented by the first position information is the first specific position. The determination step involves determining, based on the first degree of abnormality, whether or not there is an abnormality related to the movement of the moving body. and The output step outputs the result of the determination to an external source.

2. The anomaly detection method according to claim 1, In the information acquisition step, second position information representing the position of the moving body and second action information representing the action of the moving body associated with the second position information are also acquired. In the model acquisition step, if the position represented by the second position information is a second specific position, a second driving model generated based on second movement data related to the past movement of the at least one moving body and associated with that second specific position is also acquired. In the anomaly calculation step, if the position represented by the second position information is the second specific position, a second anomaly degree, representing the degree of anomaly related to the movement of the moving body, is calculated based on the second action information and the second driving model. In the determination step, the determination is made based on the first anomaly degree and the second anomaly degree.

3. The anomaly detection method according to claim 2, The method also includes a first anomaly adjustment step, in which a first weighted anomaly is calculated based on the first anomaly and a first weighted coefficient that establishes a correspondence with the position represented by the first location information; and a second weighted anomaly is calculated based on the second anomaly and a second weighted coefficient that establishes a correspondence with the position represented by the second location information. In the determination step, the determination is made based on the first weighted anomaly degree and the second weighted anomaly degree.

4. The anomaly detection method according to claim 2, The system also includes a second anomaly adjustment step, in which a third weighted anomaly is calculated based on the first anomaly and a third weighted coefficient that corresponds to the first driving model; and a fourth weighted anomaly is calculated based on the second anomaly and a fourth weighted coefficient that corresponds to the second driving model. In the determination step, the determination is made based on the third weighted anomaly degree and the fourth weighted anomaly degree.

5. The anomaly detection method according to claim 2, In the information acquisition step, first time information representing the time associated with the first location information and second time information representing the time associated with the second location information are also acquired. The anomaly detection method further includes a third anomaly adjustment step, in which a fifth weighted anomaly is calculated based on the first anomaly and a fifth weighted coefficient representing the first time step using the information from the first time step; and a sixth weighted anomaly is calculated based on the second anomaly and a sixth weighted coefficient representing the second time step using the information from the second time step. In the determination step, the determination is made based on the fifth weighted anomaly degree and the sixth weighted anomaly degree.

6. The anomaly detection method according to claim 1, In the model acquisition step, a third driving model is acquired, which is the first driving model that has established a correspondence with the manufacturer of the mobile body. In the anomaly calculation step, a third anomaly is calculated, which is based on the first action information and the first anomaly of the third driving model. In the determination step, the determination is made based on the third anomaly degree.

7. The anomaly detection method according to any one of claims 1 to 6, The first action information includes velocity information representing the speed of the moving body and / or angular velocity information representing the angular velocity of the moving body.

8. The anomaly detection method according to any one of claims 1 to 6, In the output step, the output is performed by generating an output image that overlays an image representing the result of the determination onto a map containing the location represented by the first location information, and displaying the output image on a monitor.

9. The anomaly detection method according to claim 1, It also includes a model generation step, in which the first driving model is generated based on the first movement data.

10. The anomaly detection method according to claim 9, In the model generation step, Based on multiple movement data including the first movement data, one or more driving models including the first driving model are generated. Each of the multiple mobile data is classified into one of multiple classification destinations, including one or more classification destinations corresponding to each of the one or more driving models. Based on the mobile data classified into each of the one or more classification destinations, each of the one or more driving models is generated.

11. The anomaly detection method according to claim 10, It also includes a weighted coefficient calculation step, in which a weighted coefficient representing the reliability of the driving model is calculated for each of the more than one driving models.

12. An anomaly detection device, comprising: The information acquisition unit acquires first position information indicating the position of the moving body and first action information associated with the first position information indicating the action of the moving body; The model acquisition unit, when the position represented by the first position information is a first specific position, acquires a first driving model associated with the first specific position, which is generated based on first movement data related to the past movement of at least one of the more than one moving bodies including the moving body. An anomaly calculation unit, when the position represented by the first position information is the first specific position, calculates a first anomaly degree based on the first action information and the first driving model, representing the degree of anomaly related to the movement of the moving body; The determination unit determines, based on the first anomaly degree, whether or not an anomaly related to the movement of the moving body exists; as well as The output unit outputs the result of the determination to the outside.

13. A program used to cause a computer to perform anomaly detection processing. The anomaly detection process includes: The information acquisition step involves acquiring first position information representing the position of the moving body and first action information associated with the first position information representing the action of the moving body; The model acquisition step involves obtaining a first driving model associated with the first specific location, where the location represented by the first location information is a first specific location. This model is generated based on first movement data related to the past movement of at least one of the more than one moving bodies that includes the moving body. The anomaly calculation step involves calculating a first anomaly degree, representing the degree of anomaly related to the movement of the moving body, based on the first action information and the first driving model, when the position represented by the first position information is the first specific position. The determination step involves determining, based on the first degree of abnormality, whether or not there is an abnormality related to the movement of the moving body. and The output step outputs the result of the determination to an external source.

Images