Information processing device, information processing method, and program

Abstract

To provide an information processing device, an information processing method, and a program capable of predicting risks of occurring accidents, even on road sections or intersections with no accident history.SOLUTION: An information processing device 100 includes: a prediction unit 113 that uses a prediction model trained by using road data including a road structure and / or road attributes as input data and accident history data indicating the history of accidents that occurred on the road and including the weather at the time of the accident as correct data, and including multiple weather-specific models corresponding to each of the weather conditions, so as to input road data of a target road, which is not limited to the road with accident history data, to the prediction model to predict the risk of accidents at each of one or more target sections and / or one or more target intersections of the target road; and an output unit 114 that outputs risk information indicating a risk predicted on one or more target sections and / or each of one or more target intersections to a terminal device used by a user.SELECTED DRAWING: Figure 4

Description

Technical Field

[0001] The present invention relates to an information processing apparatus, an information processing method, and a program.

Background Art

[0002] Conventionally, techniques for predicting the risk of accidents at road sections and intersections have been known. For example, Patent Document 1 discloses an information processing method for evaluating the accident risk of a road section based on road section data related to the road section and data related to the accident history of the section.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] Here, there is a need to grasp the accident risk even in sections and intersections where accidents have not yet occurred on newly constructed roads, etc., and also in sections and intersections where accidents have not occurred for a long time. However, in the technique of Patent Document 1, since an accident history is required for the road section to be predicted, there is a problem that the accident risk cannot be predicted in sections and intersections where accidents have not occurred.

[0005] Therefore, an object of the present invention is to provide an information processing apparatus, an information processing method, and a program that can predict the occurrence risk of accidents even in road sections and intersections without an accident history.

[0006] An information processing apparatus according to an aspect of the present invention is a prediction model learned by learning data using, as input data, road data including the structure and / or attributes of a road, and using, as correct data, accident history data indicating the history of accidents that occurred on the road and including the weather at the time of the accident occurrence. The prediction model includes a plurality of weather-specific models corresponding to respective weathers. The information processing apparatus includes a prediction unit that inputs road data of a target road not limited to roads having accident history data into the prediction model and predicts the risk of accidents at one or more target sections and / or one or more target intersections of the target road, and an output unit that outputs risk information indicating the predicted risks at one or more target sections and / or one or more target intersections to a terminal device used by a user.

[0007] An information processing method according to an aspect of the present invention includes a computer that uses, as input data, road data including the structure and / or attributes of a road, and uses, as correct data, accident history data indicating the history of accidents that occurred on the road and including the weather at the time of the accident occurrence. The computer learns a prediction model by learning data. The prediction model includes a plurality of weather-specific models corresponding to respective weathers. The computer inputs road data of a target road not limited to roads having accident history data into the prediction model, predicts the risk of accidents at one or more target sections and / or one or more target intersections of the target road, and outputs risk information indicating the predicted risks at one or more target sections and / or one or more target intersections to a terminal device used by a user.

[0008] A program according to one aspect of the present invention is a prediction model learned by learning data in which road data including the structure and / or attributes of a road is used as input data for a computer, and accident history data indicating the history of accidents that occurred on the road and including the weather at the time of the accident is used as correct data. The program uses a prediction model including a plurality of weather-specific models corresponding to each weather to input the road data of a target road not limited to a road with accident history data into the prediction model, and predicts the risk of accidents in one or more target sections and / or one or more target intersections of the target road. The program also has an output function that causes a terminal device used by a user to output risk information indicating the predicted risk in each of the one or more target sections and / or the one or more target intersections.

Advantages of the Invention

[0009] According to the present invention, it is possible to provide an information processing device, an information processing method, and a program that can predict the risk of accidents even in sections and intersections of roads without an accident history.

Brief Description of the Drawings

[0010]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Embodiments for Carrying Out the Invention

[0011] With reference to the accompanying drawings, a preferred embodiment of the present invention (hereinafter referred to as "this embodiment") will be described. In each figure, those with the same reference numerals have the same or similar configurations.

[0012] <1. System Configuration>

[0013] With reference to FIG. 1, an example of the system configuration of the prediction system 1 according to this embodiment will be described.

[0014] The prediction system 1 is a system for providing a service (hereinafter also referred to as "accident risk prediction service") that predicts the risk of accidents at road sections (links) and intersections (nodes) using machine learning and deep learning technologies and visualizes the prediction results. As shown in FIG. 1, the prediction system 1 includes a server device 100 used by a person providing the accident risk prediction service and a terminal device 200 used by a user. The server device 100 and the terminal device 200 are communicably connected to each other via a network N. Further, the server device 100 may be communicably connected to a device such as a third-party system different from the prediction system 1 (hereinafter also referred to as "external device") via the network N.

[0015] Network N is composed of a wireless network or a wired network. Examples of networks include a mobile phone network, a PHS (Personal Handy-phone System) network, a wireless LAN (Local Area Network, including communication compliant with IEEE802.11 (so-called Wi-Fi (registered trademark))), 3G (3rd Generation), LTE (Long Term Evolution), 4G (4th Generation), 5G (5th Generation), WiMax (registered trademark), infrared communication, visible light communication, Bluetooth (registered trademark), a wired LAN, a telephone line, a power line communication network, and a network compliant with IEEE1394, etc.

[0016] Server device 100 is an information processing device for providing an accident risk prediction service. By executing a predetermined program (hereinafter also referred to as the "server program"), server device 100 realizes a server function that provides information regarding the risk of an accident occurring (hereinafter also referred to as the "accident risk" or "risk of an accident") at a predicted road section or intersection in response to various requests from users. Server device 100 may provide, for example, as a Web server, a Web site (hereinafter also referred to as the "accident risk prediction site") for providing an accident risk prediction service to users. Further, the accident risk may be, for example, the probability of an accident occurring (hereinafter also referred to as the "occurrence probability"). The value of the accident risk (hereinafter also referred to as the "risk value") may be indicated by a numerical value, for example, between 0.0 and 1.0, or may be scaled to 0 to 100 (unit: "%").

[0017] The terminal device 200 is an information processing device used by a user, such as a terminal like a smartphone or a laptop. By executing a predetermined program (hereinafter also referred to as the "user program"), the terminal device 200 transmits and receives information (hereinafter also referred to as the "output information") for outputting on the screen the risk of an accident predicted in cooperation with the server device 100, outputs the accident risk on the screen based on the output information, or receives operation inputs for various requests from the user. In this embodiment, an example in which the user program is a general-purpose web browser will be described, but as another example, it may be an application program dedicated to the prediction system 1 (so-called native app, etc.). The user may, for example, access the accident risk prediction site via a web browser and cause each screen of the accident risk prediction site to be output to the terminal device 200.

[0018] <2. Overview>

[0019] Referring to FIG. 2, the overview of the prediction system 1 will be described.

[0020] (1) As shown in FIG. 2, in the prediction system 1, in the server device 100, a model (hereinafter also referred to as the "prediction model") for predicting the accident risk is learned using accident-related data as input data and learning data with accident history data as correct answer data. In other words, in the prediction system 1, with the accident risk as the target variable, the prediction model is learned using the accident-related data as the explanatory variable for explaining the accident risk. The prediction model may include, for example, a road section model for predicting the accident risk of a target section (hereinafter also referred to as the "target section") of a road and an intersection model for predicting the accident risk of a target intersection (hereinafter also referred to as the "target intersection").

[0021] The prediction model may include, for example, at least any one of (a) a model by accident type, (i) a model by occurrence time zone, (u) a model by age group, and (e) a model by weather. Details of each of the models (a) to (e) will be described later.

[0022] Accident-related data may include, for example, at least any one of (a) road data including the structure of the road (in other words, the shape of the road), road attributes, and / or blind spot information, (b) pedestrian flow data regarding the pedestrian flow on the road, (c) probe data, (d) map data including terrain, etc. Details of each of the data (a) to (d) will be described later. Also, accident history data is data indicating the history of accidents that occurred on the road. Details of the accident history data will also be described later. Also, the "road" referred to here may include, for example, an intersection where two or more roads intersect.

[0023] According to the above configuration, the prediction model can learn the characteristics such as the structure and attributes of the road at a location with a high accident risk.

[0024] (2) Next, the server device 100 inputs the accident-related data of the road to be predicted (hereinafter also referred to as the "target road") into the prediction model according to the user's request received from the terminal device 200, etc., and predicts the accident risk for each of one or more target sections and / or one or more target intersections of the target road. Regarding the target section, for example, it may be specified by dividing the link of the road network at each unit length (for example, 100 m, etc.). Also, the target road is not limited to a road with accident history data. In other words, the target road may be a road without accident history data. A road without accident history data may be, for example, (a) a newly constructed road or the like that has not had an accident so far, (b) a road that has had an accident before, but for which there is no history of the accident that occurred.

[0025] (3) Next, the server device 100, according to the user's request received from the terminal device 200, etc., for each of one or more target sections and one or more target intersections of the target road, color-codes the risk information indicating the accident risk predicted in (2) above according to the risk value and outputs it onto the map on the screen of the terminal device 200.

[0026] Based on the above configuration, the prediction system 1 can predict the risk of accidents in one or more target sections and / or one or more target intersections for target roads that are not limited to roads with accident history data. Therefore, the prediction system 1 can predict the risk of accidents occurring in sections or intersections of roads without an accident history.

[0027] For example, in the identification of dangerous road sections and intersections, only the sections and intersections where accidents or the like have actually occurred have been identified, and accident countermeasures have often been taken after the fact. Based on the above configuration, the prediction system 1 can predict the accident risk for sections and intersections of roads where no accident has actually occurred, and can assist in taking accident countermeasures in advance.

[0028] <3. Functional Configuration>

[0029] Referring to FIG. 3, the functional configuration of the server device 100 according to the present embodiment will be described. As shown in FIG. 3, the server device 100 includes a control unit 110, a communication unit 120, and a storage unit 130.

[0030] [Control Unit]

[0031] The control unit 110 includes a reception unit 111 and a prediction unit 113. Further, the control unit 110 may include, for example, an acquisition unit 112, an output unit 114, and / or a generation unit 115.

[0032] [Reception Unit]

[0033] The reception unit 111 receives various requests from the terminal device 200. The reception unit 111 may receive, for example, a request for predicting the accident risk for a specific area from the terminal device 200. Further, the reception unit 111 may receive, for example, various operation requests including a request for outputting the screens shown in FIGS. 5 to 10 from the terminal device 200.

[0034] [Acquisition Unit]

[0035] The acquisition unit 112 acquires various data such as accident-related data and / or accident history data from an external device.

[0036] Here, with reference to FIG. 4, an example of the data configuration of accident-related data and accident history data will be described.

[0037] As shown in FIG. 4, the accident history data may be, for example, traffic accident statistical information which is open data of the police agency. The accident history data may include, for example, for each accident that occurred during a predetermined period (e.g., the past three years from a specific point in time, etc.), the accident occurrence date and time, the occurrence location (e.g., latitude, longitude, etc.), the accident details (e.g., vehicle-to-vehicle, or vehicle-to-person, etc.), the accident occurrence location (e.g., road, intersection, etc.), the weather at the time of the accident, the age group of the victims, and the like.

[0038] The accident-related data may include, for example, map data, pedestrian flow data, road data, and / or probe data, etc.

[0039] The map data may include, for example, land use data which is open data of the Geospatial Information Authority of Japan, and / or topographic data regarding the terrain. The topographic data may include, for example, the elevation, slope inclination, slope azimuth, etc. of a specific area. Specifically, the topographic data may include, as statistical values for each unit area (e.g., for each 250m mesh) in a specific area, the average elevation, the highest elevation, the lowest elevation, the maximum inclination angle, the maximum inclination direction, the minimum inclination angle, the average inclination angle, and the like. The topographic data may be, for example, elevation mesh data, etc.

[0040] The map data may include, for example, POI data regarding POIs (Points of Interest). The POI data may indicate, for example, that a specific location is a location related to services such as government offices, organizations, welfare, transportation, warehousing, clothing, medicine, health and hygiene, real estate, rental, exhibition venues, schools, hobby classes, libraries, restaurants, sports, hobbies, entertainment, leisure, automobiles, motorcycles, bicycles, drives, sales, wholesales, etc. Also, the POI data may be summarized, for example, for each business type (such as beauty salons, food and beverage, medical institutions, convenience stores, etc.) for these locations.

[0041] The pedestrian flow data may be statistical data such as census data, and may include, for example, the total population, population by gender and / or age, number of households, etc. for each region for each survey year. As another example, the pedestrian flow data may be GPS log data of mobile terminals per unit area (such as a 500m mesh, etc.) per unit period (such as every 1 hour, etc.).

[0042] The road data may be a digital road map and may include, for example, the structure and / or attributes of the road. The structure of the road may be, for example, intersections, areas near intersections, general single roads, level crossings, tunnels, bridges, with / without traffic lights, right curves, left curves, straight lines, etc. Also, the attributes of the road may include, for example, road display type, road route type, road administrator, road width classification, bypass flag (indicating, for example, that a road section is a bypass), motorway flag (indicating, for example, that a road is a motorway), access prohibition type code (indicating, for example, the access prohibition status of a road section), one-way traffic type code (indicating, for example, the one-way traffic status of a road section), number of lanes, zone 30 area (indicating, for example, that a road section is included in the zone 30 area), temporary stop position (indicating, for example, including positions where a temporary stop is required), speed limit information indicating speed limits, etc.

[0043] Probe data is data obtained from a sensor device or the like installed in a vehicle, and is, for example, car navigation data and / or drive recorder data, etc. The probe data may include, for example, the passing time, latitude, longitude, altitude, passing speed, GPS accuracy, temperature, air pressure, and / or radio wave intensity, etc. when the vehicle passes a specific point. The probe data may include, for example, dangerous driving data described later.

[0044] Accident-related data may include, for example, weather information, blind spot information regarding blind spots on the road, traffic jam information indicating the traffic jam situation on the road, road surface condition information indicating the road surface condition of a road section or an intersection, and / or traffic regulation information indicating traffic regulations on the road, etc.

[0045] The blind spot information may be, for example, information indicating visibility in one or more target sections and / or one or more target intersections of the target road. This visibility may be, for example, the ratio of the visible azimuths (specifically, a value obtained by dividing the number of visible azimuths by the number of all azimuths (16)) in a predetermined viewpoint position and / or a predetermined visual field range. The predetermined viewpoint position may be, for example, a value representing the height of a human's line of sight when standing upright (specifically, a value between 1.2 and 1.7 m, etc.). Also, the visual field range may be, for example, a value representing the human visual field range (specifically, a value between 100 and 130 mm, etc.). Note that the blind spot information may be generated by a generation unit 115 described later, for example.

[0046] [Prediction Unit]

[0047] Returning to FIG. 3, the description will be continued. The prediction unit 113 uses a prediction model learned from learning data (teacher data), inputs road data of a target road not limited to a road with accident history data into the prediction model, and predicts the risk of accidents in one or more target sections and / or one or more target intersections of this target road. The timing at which the prediction unit 113 predicts the accident risk may be, for example, predicted each time the screens of the accident risk prediction sites shown in FIGS. 5 to 10 are output, or may be predicted in advance separately from the output of these screens. In the latter case, for example, the prediction unit 113 stores risk information indicating the prediction result in the storage unit 130, and the output unit 114 may refer to the storage unit 130 and output the screen of the accident risk prediction site based on the risk information.

[0048] The learning data may use, for example, road data including the structure and / or attributes of the road as input data, and accident history data indicating the history of accidents that occurred on the road as correct answer data.

[0049] The input data of the learning data (hereinafter, also simply referred to as "input data") may include, for example, pedestrian flow data. The pedestrian flow data is, for example, data indicating the population per unit area in an area including at least a part of the road.

[0050] The prediction unit 113 may predict the risk of accidents in one or more target sections and / or one or more target intersections of the target road based on, for example, the regression of a prediction model with the road data obtained by learning as an explanatory variable and the accident history data as an objective variable. According to such a configuration, using road data including the structure and / or attributes of the road, the accident risk can be predicted even on a road where no accident has occurred. Therefore, for example, the accident risk of a road without an accident history can be predicted based on the accident history of roads with similar structures and / or attributes.

[0051] The input data may include, for example, pedestrian flow data. The prediction unit 113 may, for example, input the pedestrian flow data of the target road into the prediction model to predict the risk of accidents in one or more target sections and / or one or more target intersections of the target road. According to such a configuration, the prediction unit 113 can predict the accident risk based on the flow and movement of people in the target section and target intersection.

[0052] The accident history data (or accident-related data) may include, for example, weather information indicating the weather at the time of the accident. The climate information may indicate, for example, the weather, temperature, humidity, precipitation probability, lightning probability, wind speed, and / or wind direction at the time of the accident. Also, the climate information may be classified by the weather at the time of the accident, such as sunny, cloudy, foggy, rainy, or thundery. In this case, the prediction model may include, for example, a plurality of climate-specific models corresponding to each weather (specifically, weather, etc.). Specifically, the prediction model may include a model for predicting the accident risk when it is sunny and / or a model for predicting the accident risk when it is rainy. Also, the prediction model may be further divided by whether it is a target section or a target intersection. For example, when the occurrence time zone is the morning time zone, the prediction model may include a model for predicting the accident risk in a sunny target section and a model for predicting the accident risk in a sunny target intersection.

[0053] The prediction unit 113 may, for example, use each of the plurality of climate-specific models to predict the accident risk for each climate. Also, the prediction unit 113 may, for example, calculate a statistical value (such as the total value, average value, median value, or mode value, etc. The same applies hereinafter) of the risk values of the accident risks predicted by each of the plurality of climate-specific models. It is conceivable that the tendency of accident occurrence varies by climate in the target section and target intersection. Specifically, it is conceivable that in each of the target section and target intersection, there may be a tendency that accidents occur more frequently when it is raining and less frequently when it is sunny. According to such a configuration, the accident risk can be predicted based on the climate-specific tendency in the target section and target intersection.

[0054] The accident history data may include, for example, the time period when the accident occurred. This time period when the accident occurred may be classified by time period, for example, the morning time period (for example, between 7:00 and 10:00 (expressed in 24-hour notation. The same applies hereinafter)), the noon time period (for example, between 10:00 and 16:00), the evening time period (for example, between 16:00 and 20:00), and the night time period (for example, between 20:00 and 7:00). In this case, the prediction model may include, for example, a plurality of models for each time period when the accident occurred. Specifically, the prediction model may include a model for predicting the accident risk during the morning time period, a model for predicting the accident risk during the noon time period, a model for predicting the accident risk during the evening time period, and / or a model for predicting the accident risk during the night time period. Further, the prediction model may be further divided by the target section or the target intersection. For example, when the time period when the accident occurred is the morning time period, the prediction model may include a model for predicting the accident risk in the target section during the morning time period and a model for predicting the accident risk at the target intersection during the morning time period.

[0055] The prediction unit 113 may predict, for example, the accident risk for each time period when the accident occurred by using each of a plurality of models for each time period when the accident occurred. Further, the prediction unit 113 may calculate, for example, a statistical value of the risk values of the accident risks predicted by each of a plurality of models for each time period when the accident occurred. It is conceivable that the tendency of the occurrence of accidents varies by time period in the target section and the target intersection. Specifically, it is conceivable that in each of the target section and the target intersection, there may be a tendency that the number of accidents occurring during the morning time period is large, and conversely, the number of accidents occurring during the noon time period is small. According to such a configuration, the accident risk can be predicted based on the tendency by time period in the target section and the target intersection.

[0056] The accident history data may include, for example, the age groups of each of one or more parties involved in the accident. The age groups may classify the age of the parties, for example, into a young age group (e.g., under 20 years old), an adult age group (20 to less than 70 years old), and an elderly age group (70 years old or older). In this case, the prediction model may include a plurality of age-group-specific models corresponding to each of the age groups of the parties. Specifically, the prediction model may include a model for predicting the accident risk of the young age group, a model for predicting the accident risk of the adult age group, and / or a model for predicting the accident risk of the elderly age group. Also, similar to the above predictions by time zone of occurrence, the prediction model may be further divided by the target section or the target intersection.

[0057] The prediction unit 113 may, for example, predict the accident risk by age group using each of the above plurality of age-group-specific models. Also, the prediction unit 113 may, for example, calculate a statistical value of the risk values of the accident risks predicted by each of the plurality of age-group-specific models. For example, it is conceivable that the tendency of accident occurrence varies by age group in the target section or the target intersection. Specifically, in each of the target section and the target intersection, it is conceivable that there may be a tendency such as a high occurrence of accidents in the young age group and a low occurrence of accidents in the elderly age group. According to such a configuration, the accident risk can be predicted based on the tendency by age group in the target section or the target intersection.

[0058] The prediction model for predicting by the age group of the above parties may be further divided, for example, by whether each of the one or more parties is a pedestrian or a driver operating a vehicle. Specifically, when predicting the accident risks of an elderly pedestrian and an adult driver, the prediction model may include a model for predicting the accident risk of an elderly pedestrian and a model for predicting the accident risk of an adult driver.

[0059] Accident history data may include, for example, accident types. The accident type may indicate, for example, whether the accident is caused by (1) vehicle - to - person, (2) vehicle - to - vehicle, or (3) vehicle alone. The accident type may be further classified by vehicle type. Specifically, the accident type may be further classified according to whether the vehicle is a light vehicle such as a bicycle. More specifically, the accident type may indicate whether the accident is caused by (1) a vehicle other than a light vehicle to a person, (2) between vehicles other than light vehicles, (3) a vehicle other than a light vehicle alone, (4) a light vehicle to a person, (5) between light vehicles, (6) a light vehicle alone, or (7) a light vehicle to a vehicle other than a light vehicle. In this case, the prediction model may include a plurality of accident - type - specific models corresponding to each accident type. Specifically, the prediction model may include a model for predicting the risk of vehicle - to - person accidents, a model for predicting the risk of vehicle - to - vehicle accidents, and / or a model for predicting the risk of vehicle - alone accidents. Further, each of these models may be divided, for example, according to whether the vehicle is a light vehicle or not. Also, the prediction model may be further divided by target section or target intersection, similar to the above - mentioned prediction by time zone.

[0060] The prediction unit 113 may, for example, predict the accident risk for each accident type using each of the above-described multiple models for different accident types. Further, the prediction unit 113 may, for example, calculate a statistical value of the risk values of the accident risks predicted by each of the multiple models for different accident types. For example, it is conceivable that the tendency of the occurrence of accidents differs for each accident type in the target section or target intersection. Specifically, in each of the target section and target intersection, it is conceivable that there may be a tendency such as a large number of vehicle-to-vehicle accidents and, conversely, a small number of vehicle-to-person accidents. According to such a configuration, the accident risk can be predicted based on the tendency for each accident type in the target section or target intersection. Further, for example, in a specific intersection or section, it is conceivable that there may be a tendency such as a large number of accidents when the vehicle is a light vehicle such as a bicycle and a small number of accidents for vehicles other than light vehicles. In another intersection, the opposite tendency may also be conceivable. According to such a configuration, for each target section and target intersection, the accident risk can be predicted based on the tendency for each vehicle type.

[0061] The prediction unit 113 may, for example, predict the accident risk by combining and using multiple types of prediction models. Specifically, the prediction unit 113 may combine (for example, add) the risk values of the accident risks calculated by each of the multiple models for different occurrence time zones and the multiple models for different weather conditions to calculate a comprehensive risk value such as the risk value in the morning time zone and when the weather is clear.

[0062] The input data may include, for example, dangerous driving data. The dangerous driving data is data related to dangerous driving of a vehicle (hereinafter also referred to as "dangerous driving"), and indicates, for example, the position of the detected dangerous driving and the type of dangerous driving for the vehicle traveling on the road. In this case, the prediction unit 113 may, for example, input the dangerous driving data of the target road into the prediction model to predict the accident risk at each of one or more target sections and / or one or more target intersections of the target road. The types of dangerous driving may include, for example, sudden acceleration, sudden deceleration, sudden steering (left and right), wobbling, forward collision warning, lane departure warning (left and right), and / or start delay warning.

[0063] For example, it is conceivable that the tendency of accident occurrence varies depending on the type of dangerous driving on the target road. Specifically, in each of the target section and the target intersection, it is conceivable that there may be a tendency such as more accidents occurring in the target section where more sudden accelerations are detected. According to the above configuration, the accident risk can be predicted based on the tendency of the type of dangerous driving in the target section and the target intersection.

[0064] The prediction unit 113 may update, for example, the risk value calculated by the prediction model based on the blind spot information. Specifically, the prediction unit 113 may recalculate the risk value including the visibility of the target intersection as an explanatory variable of the prediction model. The prediction unit 113 may update with this recalculated risk value. For example, if the risk value calculated without including visibility is "0.23" and the risk value recalculated including visibility is "0.74", the risk value may be updated to the recalculated "0.74". Visibility is an aspect of the blind spot information and indicates the degree of visibility in the target section or the target intersection.

[0065] [Learning unit]

[0066] The prediction unit 113 may include, for example, a learning unit 113a. The learning unit 113a learns a prediction model using learning data. The prediction model is, for example, a model learned by a machine learning or deep learning method. The prediction model may be, for example, at least any one of a gradient boosting decision tree, a random forest, a logistic regression, a support vector machine, and a neural network. In other words, the prediction model may be any one of these models, or a combination of any two or more of these models.

[0067] [Output unit]

[0068] The output unit 114 causes the terminal device 200 to output risk information indicating the predicted risks for each of one or more target sections and / or one or more target intersections. For example, the output unit 114 generates output information including the risk information and for outputting the screens shown in FIGS. 5 to 10. The output unit 114 transmits this generated output information to the terminal device 200 to cause the terminal device 200 to output these screens.

[0069] According to the above configuration, the output unit 114 can visualize the prediction of the accident risk for each of one or more target sections and / or one or more target intersections for target roads not limited to roads with accident history data. Therefore, even for sections and intersections of roads without an accident history, the user can visually recognize the accident risk, so that it is possible to assist the user in taking prior accident prevention measures.

[0070] For example, the output unit 114 may output risk information by weather based on the accident risks predicted by each of a plurality of weather-based models in response to a user request from the terminal device 200. According to such a configuration, the accident risk by weather can be visualized. For this reason, in the target section and target intersection, it is possible to easily visually recognize in what kind of weather the accident risk is high or low, etc., and the user can be made to grasp the tendency of the accident risk by weather.

[0071] For example, the output unit 114 may output risk information by occurrence time zone based on the accident risks predicted by each of a plurality of occurrence time zone-based models in response to a user request from the terminal device 200. According to such a configuration, the accident risk by occurrence time zone can be visualized. For this reason, in the target section and target intersection, it is possible to easily visually recognize in which time zone the accident risk is high or low, etc., and the user can be made to grasp the tendency of the accident risk by time zone.

[0072] The output unit 114 may output risk information for each age group based on the accident risks for each age group predicted by each of the plurality of age-group models, for example, in response to a user request from the terminal device 200. According to such a configuration, the accident risks for each age group can be visualized. Therefore, in the target section or intersection, it is possible to easily visually recognize which age group has a high accident risk and which has a low accident risk, etc., and the user can be made to understand the trend of the accident risks for each age group.

[0073] The output unit 114 may output risk information for each accident type based on the accident risks for each accident type predicted by each of the plurality of accident-type models, for example, in response to a user request from the terminal device 200. According to such a configuration, the accident risks for each accident type can be visualized. Therefore, in the target section or intersection, it is possible to easily visually recognize which accident type has a high accident risk and which has a low accident risk, etc., and the user can be made to understand the trend of the accident risks for each accident type.

[0074] [Generation unit] The generation unit 115 generates various types of information related to the prediction of accident risks. The generation unit 115 may generate, for example, blind spot information. Specifically, the generation unit 115 (1) Causes the acquisition unit 112 to acquire elevation data and building data respectively from an external device or the storage unit 130. (2) Generates surface data indicating a surface model in a specific area including the target road based on the elevation data and building data acquired in (1) above. (3) Analyze the visibility at each of the target section and the target intersection based on the surface data generated in (2) above, the value of a predetermined viewpoint position, and the value of a predetermined viewing range. Regarding the technique for analyzing such visibility, for example, it is disclosed in "Advanced viewshed analysis: a Quantum GIS plug-in for the analysis of visual landscapes", Zoran Cuckovic, Journal of Open Source Software, 1(4), 32, doi:10.21105 / joss.00032. Specifically, in the analysis of visibility at each of the target section and the target intersection, the generation unit 115 may use the plug-in function "Advanced viewshed analysis for QGIS" of QGIS, which is open-source software of GIS (Geographic Information System). The generation unit 115 may input the surface data, the value of a predetermined viewpoint position, and the value of a predetermined viewing range into this function to analyze the visibility. (4) Generate information indicating the result analyzed in (3) above as blind spot information. The generation unit 115 may use, for example, the "visibility index" (a form of visibility) output from the plug-in function of the above QGIS as the blind spot information.

[0075] Elevation data is data in which the elevation of each point in a specific area including the target road is recorded. The elevation data may be, for example, mesh data representing the position of cells (e.g., for each 1m grid) into which the ground surface is divided and the representative elevation of the ground surface within the cell. The elevation data may be, for example, a Digital Elevation Model (DEM).

[0076] Building data is data indicating the shape and height of each of one or more buildings existing in a specific area including the target road. The building data may be, for example, data of a 3D city model provided by the Ministry of Land, Infrastructure, Transport and Tourism in the leading PLATEAU project, or three-dimensional data at the time of building design. [Communication unit]

[0077] The communication unit 120 transmits and receives various information including output information to and from the terminal device 200 etc. via the network N, and also transmits and receives various data including accident history data and accident-related data to and from an external device.

[0078] [Memory unit]

[0079] The memory unit 130 stores a prediction model, learning data, risk information, output information, etc. The memory unit 130 may store each piece of information using a database management system (DBMS), or may store each piece of information using a file system. When using a DBMS, a table may be provided for each of the above information, and the tables may be associated with each other to manage each piece of information.

[0080] The memory unit 130 may include, for example, a model memory unit 131. The model memory unit 131 stores a prediction model. Also, the memory unit 130 may include, for example, a learning data memory unit 132. The learning data memory unit 132 stores learning data.

[0081] <4. Screen example>

[0082] With reference to FIGS. 5 to 10, a screen example of the prediction system 1 will be described. FIGS. 5 to 10 are diagrams showing examples of a screen or a part of a screen output on the accident risk prediction site of the prediction system 1.

[0083] FIG. 5 is a diagram showing an example of the top screen of the accident risk prediction site. As shown in FIG. 5, the top screen A1 includes an accident history selection area a11, a display data selection area a12, a map display area a13, and a risk statistical value output area a14.

[0084] The accident history selection area a11 is an area for selecting the accident history (denoted as "accident record" in Fig. 5) to be output to the map display area a13 when "accident history data" is selected in the display data selection area a12. The accident history selection area a11 can select, for example, the accident history to be output by time period (denoted as "time" in Fig. 5), age group (denoted as "age" in Fig. 5), accident type (denoted as "accident type" in Fig. 5), accident severity, and weather (denoted as "weather" in Fig. 5) (in other words, it can be filtered). The output unit 114 can output the accident history to the map in the map display area a13 based on the content selected by the user in the accident history selection area a11 and the accident history data.

[0085] The display data selection area a12 is an area for selecting the data to be output to the map display area a13. The display data selection area a12 can select, for example, at least one of the following data: (a) the accident risk of the target section (denoted as "road accident prediction" in Fig. 5), (b) the accident risk of the target intersection (denoted as "road accident prediction" in Fig. 5), (c) accident history data (denoted as "accident record" in Fig. 5), (d) the accident risk of the target section and the target intersection in the school zone (denoted as "school zone" in Fig. 5), and (e) pedestrian flow data (denoted as "population and pedestrian flow" in Fig. 5), and output it on the map in the map display area a13. The output unit 114 can output risk information indicating the corresponding accident risk to the map in the map display area a13 based on, for example, the content selected by the user in the display data selection area a12 and the result of the prediction by the prediction unit 113.

[0086] The map display area a13 is an area that outputs a map of a specified area (in this example, the area around "〇〇 Prefecture") specified by the user. The output unit 114 superimposes and outputs on the map display area a13 line segments (target sections) and dots (target intersections) colored according to their respective risk values (0.0 to 1.0) as risk information of the accident risks at the target section and the target intersection, respectively. The map display area a13 may output a legend for this color-coding in the upper left corner of the map (in this example, the accident risk of the target section is denoted as "road accident risk", and the accident risk of the target intersection is denoted as "intersection accident risk").

[0087] The risk statistic value output area a14 is an area that outputs the statistical value (in this example, the average value) of the risk values of the accident risks of all one or more target sections and one or more target intersections within the specified area.

[0088] FIG. 6 is a map display area a13b output when "(A) Accident risk of target section" is selected in the display data selection area a12. The map display area a13a may output the risk information of the accident risk predicted using a prediction model for predicting the accident risk of the target section. On the map of the map display area a13b, the accident risk of the target section is output as a line segment color-coded in black and white shades according to the magnitude of its risk value (the lighter the color as the risk value is smaller, and the darker the color as the risk value is larger).

[0089] FIG. 7 is a map display area a13a output when "Accident risk of target intersection" is selected in the display data selection area a12. The map display area a13a may output the risk information of the accident risk predicted using a prediction model for predicting the accident risk of the target intersection. On the map of the map display area a13a, similar to the target intersection in FIG. 6, the accident risk of the target intersection is output as a dot color-coded in shades according to the magnitude of its risk value.

[0090] FIG. 8 is a display data selection dialog a12a that can be output from the top screen A1. The display data selection dialog a12a is a screen for selecting data to be output to the map display area a13, similar to the display data selection area a12 in FIG. 6. In the display data selection dialog a12a, the accident risks to be output on the map can be selected by accident type, time zone, age group, and weather. For example, when the user selects the accident type, time zone, age group, and / or weather of the accident risk to be output in the display data selection dialog a12a and presses the OK button, the output unit 114 outputs the risk information of the accident risk corresponding to the selected accident type, time zone, and / or age group on the map in the map display area a13.

[0091] FIG. 9 is an example of cutting out a part of the map in the map display area a13 that outputs the accident risks by age group of the victims in the target section. FIG. 9(a) is an example of outputting the accident risks of all age groups (hereinafter also referred to as "all accidents"). FIG. 9(b) is an example of outputting the accident risks of the young age group. FIG. 9(c) is an example of outputting the accident risks of the elderly.

[0092] In the areas surrounded by the dashed lines in FIGS. 9(a) to 9(c), compared with the output of all accidents in FIG. 9(a), there are areas with lighter colors, that is, areas with lower risk values, in the output of the young age group in FIG. 9(b). That is, for this section, the user can confirm that the risk of accidents in which children as pedestrians become victims is low, and it is a relatively safe section for children. Also, in the areas surrounded by this dashed line, compared with the output of all accidents in FIG. 9(a), there are areas with lighter colors, that is, areas with lower risk values, in the output of the elderly in FIG. 9(b). That is, for this section, the user can confirm that the risk of accidents in which the elderly as pedestrians become victims is low, and it is a relatively safe section for the elderly.

[0093] FIG. 10 is an example of cutting out a part of the map of the map display area a13 that outputs the accident risk by time zone in the target section. FIG. 10(a) is an example of outputting the accident risk for all time zones (hereinafter also referred to as "all accidents"). FIG. 10(b) is an example of outputting the accident risk for the morning time zone. FIG. 10(c) is an example of outputting the accident risk for the evening time zone.

[0094] In the area surrounded by the dashed line in FIGS. 10(a) to 10(c), compared with the output of all accidents in FIG. 10(a), there are areas with darker colors, that is, areas with larger risk values, in the output of the morning time zone in FIG. 10(b). That is, for this area, the user can confirm that it is a relatively dangerous area with a high accident risk in the morning time zone. Also, in the area surrounded by this dashed line, compared with the output of all accidents in FIG. 10(a), there are areas with darker colors in the output of the evening time zone in FIG. 10(c). That is, for this area, the user can confirm that it is a relatively dangerous area with a high accident risk in the evening time zone.

[0095] FIG. 11 is an example of cutting out a part of the map of the map display area a13 that outputs the accident risk by time zone and weather in the target section. FIG. 11(a) is an example of outputting the accident risk for the morning time zone and all weather conditions (hereinafter also referred to as "morning accidents"). FIG. 11(b) is an example of outputting the accident risk for the morning time zone and when the weather is sunny. FIG. 11(c) is an example of outputting the accident risk for the morning time zone and when it is raining.

[0096] In the areas enclosed by the dashed lines in FIGS. 11(a) to 11(c), compared with the outputs for all weather conditions in FIG. 11(a), there are areas where the color is darker, i.e., areas where the risk value is larger, in the output for sunny weather in FIG. 11(b). That is, for this area, the user can confirm that it is a relatively dangerous area with a high risk of accidents when it is sunny. Also, in the area enclosed by this dashed line, compared with the outputs for all weather conditions in FIG. 11(a), there are areas where the color is darker in the output for rainy weather in FIG. 11(c). That is, for this area, the user can confirm that it is a relatively dangerous area with a high risk of accidents when it is rainy.

[0097] <5. Operation Example>

[0098] Referring to FIG. 12, an operation example of the prediction system 1 according to the present embodiment will be described. FIG. 11(a) is a flowchart showing the flow of processing during the learning of the prediction model. FIG. 11(b) is a flowchart showing the flow of processing when predicting and outputting the accident risk using the learned prediction model. Note that the order of processing in the flowcharts shown below is an example and may be changed as appropriate.

[0099] As shown in FIG. 12(a), the learning unit 113a of the server device 100 learns a prediction model using the learning data stored in the learning data storage unit 132 (S10). The learning unit 113a stores the learned prediction model in the model storage unit 131 (S11).

[0100] As shown in FIG. 12(b), the reception unit 111 of the server device 100 receives a request for the output of the top screen A1 from the terminal device 200 (S20). The prediction unit 113 uses the learned prediction model described above to input accident-related data of a target road that is not limited to roads with accident history data into the prediction model, and predicts the risk of accidents in one or more target sections and / or one or more target intersections of the target road (S21). The terminal device 200 is caused to output risk information indicating the predicted risks in one or more target sections and / or one or more target intersections respectively (S22).

[0101] <6. Hardware Configuration>

[0102] Referring to FIG. 13, an example of the hardware configuration when the server device 100 described above is realized by a computer 800 will be described. Note that the functions of each device can also be realized by dividing them among a plurality of devices.

[0103] As shown in FIG. 13, the computer 800 includes a processor 801, a memory 803, a storage device 805, an input I / F unit 807, a data I / F unit 809, a communication I / F unit 811, and a display device 813.

[0104] The processor 801 controls various processes in the computer 800 by executing programs stored in the memory 803. For example, each functional unit included in the control unit 110 of the server device 100 can be realized by the processor 801 executing a program (for example, a server program) temporarily stored in the memory 803.

[0105] The memory 803 is a storage medium such as a RAM (Random Access Memory). The memory 803 temporarily stores the program code of the program executed by the processor 801 and the data required during the execution of the program.

[0106] The storage device 805 is a non-volatile storage medium such as a hard disk drive (HDD) or a flash memory. The storage device 805 stores an operating system, various models and various programs for realizing the above-described configurations. In addition, the storage device 805 can also store a table for registering learning data and the like, and a DB for managing the table. Such programs and data are referred to by the processor 801 by being loaded into the memory 803 as needed.

[0107] The input I / F unit 807 is a device for receiving input from a user. Specific examples of the input I / F unit 807 include a keyboard, a mouse, a touch panel, various sensors, a wearable device, and the like. The input I / F unit 807 may be connected to the computer 800 via an interface such as USB (Universal Serial Bus).

[0108] The data I / F unit 809 is a device for inputting data from outside the computer 800. Specific examples of the data I / F unit 809 include a drive device for reading data stored in various storage media. The data I / F unit 809 may be provided outside the computer 800. In that case, the data I / F unit 809 is connected to the computer 800 via an interface such as USB.

[0109] The communication I / F unit 811 is a device for performing data communication via the Internet N, either wired or wirelessly, with a device outside the computer 800. The communication I / F unit 811 may be provided outside the computer 800. In that case, the communication I / F unit 811 is connected to the computer 800 via an interface such as USB.

[0110] The display device 813 is a device for displaying various information. Specific examples of the display device 813 include, for example, a liquid crystal display, an organic EL (Electro-Luminescence) display, a display of a wearable device, and the like. The display device 813 may be provided outside the computer 800. In that case, the display device 813 is connected to the computer 800 via, for example, a display cable. Also, when a touch panel is adopted as the input I / F unit 807, the display device 813 can be configured integrally with the input I / F unit 807.

[0111] Note that the above embodiments are examples for explaining the present invention, and the present invention is not intended to be limited only to those embodiments. Further, the present invention can be variously modified without departing from its gist. Furthermore, those skilled in the art can adopt embodiments in which each element described above is replaced with an equivalent one, and such embodiments are also included in the scope of the present invention.

[0112] [Modification Example]

[0113] Although the present invention has been described based on the above embodiments, the following cases are also included in the present invention.

[0114] [Modification Example 1]

[0115] At least a part of each configuration in the server device 100 according to the above embodiment may be provided in the terminal device 200. For example, the functions of the reception unit 111 and the output unit 114 of the control unit 110 in the server device 100 may be implemented in the terminal device 200. Specifically, the terminal device 200 may install, for example, an application program dedicated to the prediction system 1 (hereinafter also referred to as a "prediction app") and execute this prediction app to realize these configurations. The terminal device 200 acquires risk information from the server device 100 and stores it in advance in its own storage unit, and the prediction app may refer to this storage unit and output a screen similar to the accident risk prediction site in response to a user's request based on the risk information.

[0116] [Modification Example 2]

[0117] In the above-described embodiment, an example in which the model storage unit 131 is provided in the storage unit 130 of the server device 100 as a place to store the prediction model has been described. However, the place where the prediction model is stored is not limited to this. The prediction model may be stored, for example, in the storage unit of an external device. The prediction unit 113 of the server device 100 may, for example, instruct an API provided by this external device and used to utilize the function of the prediction model to predict the accident risk using the prediction model. As a response to this instruction, the prediction unit 113 may acquire, from this external device through the acquisition unit 112, risk information indicating the predicted risks at one or more target intervals and / or one or more target intersections as the result of the prediction.

[0118] [Modification Example 3]

[0119] In the above-described embodiment, an example in which the control unit 110 of the server device 100 includes a learning unit 113a that learns the prediction model has been described. However, the learning unit according to the present invention is not limited to this. The learning unit according to the present invention may be provided, for example, in a device different from a device including a prediction unit that predicts the accident risk using the prediction model. This different device may be, for example, an external device in a third-party system. For example, the acquisition unit 112 included in the control unit 110 of the server device 100 may acquire the learned prediction model from an external device and store the acquired prediction model in the model storage unit 131.

Explanation of Reference Numerals

[0120] 1... Prediction system, 100... Server device, 110... Control unit, 111... Reception unit, 112... Acquisition unit, 113... Prediction unit, 113a... Learning unit, 114... Output unit, 115... Generation unit, 120... Communication unit, 130... Storage unit, 200... Terminal device, 800... Computer, 801... Processor, 803... Memory, 805... Storage device, 807... Input I / F unit, 809... Data I / F unit, 811... Communication I / F unit, 813... Display device.

Claims

1. A prediction model trained with learning data that uses road data including the structure and / or attributes of a road as input data and accident history data indicating the history of accidents that occurred on the road and including the weather at the time of the accident as correct answer data. The prediction model includes a plurality of weather-specific models corresponding to each of the weathers. The road data of a target road not limited to the road with the accident history data is input into the prediction model, and a prediction unit that predicts the risk of accidents in one or more target sections and / or one or more target intersections of the target road, An output unit that outputs risk information indicating the predicted risk in each of the one or more target sections and / or one or more target intersections to a terminal device used by a user. An information processing apparatus comprising: Information processing apparatus.

2. The prediction unit predicts the risk for each weather using each of the plurality of weather-specific models. The output unit outputs the risk information for each weather in response to a request from the terminal device by the user. The information processing apparatus according to claim 1.

3. The prediction unit predicts the risk of accidents in one or more target sections and / or one or more target intersections of the target road based on the regression of the prediction model using the road data obtained by the learning as an explanatory variable and the accident history data as a target variable. The information processing apparatus according to claim 1.

4. The input data further includes flow data indicating the population per unit area in an area including at least a part of the road. The prediction unit inputs the flow data of the target road into the prediction model and predicts the risk of accidents in one or more target sections and / or one or more target intersections of the target road. The information processing apparatus according to claim 1 or 2.

5. The accident history data includes the time zone when the accident occurred. The prediction model includes a plurality of models for each occurrence time zone corresponding to each of the occurrence time zones. The information processing apparatus according to claim 1 or 2.

6. The prediction unit predicts the risk for each occurrence time zone using each of the plurality of models for each occurrence time zone. The output unit causes the risk information to be output for each occurrence time zone in response to the request of the user from the terminal device. The information processing apparatus according to claim 5.

7. The accident history data includes the age group of each of one or more parties to the accident. The prediction model includes a plurality of models for each age group corresponding to each of the age groups of the parties. The information processing apparatus according to claim 1 or 2.

8. The prediction unit predicts the risk for each age group using each of the plurality of models for each age group. The output unit causes the risk information to be output for each age group in response to the request of the user from the terminal device. The information processing apparatus according to claim 7.

9. The accident history data includes an accident type indicating whether the accident is caused by vehicle-to-person, vehicle-to-vehicle, or vehicle-alone. The prediction model includes a plurality of models for each accident type corresponding to each of the accident types. The information processing apparatus according to claim 1 or 2.

10. The accident type is further classified into whether the vehicle is a light vehicle or not. The information processing apparatus according to claim 9.

11. The prediction unit predicts the risk for each accident type using each of the plurality of models for each accident type. The output unit outputs the risk information for each accident type in response to the request of the user from the terminal device. The information processing apparatus according to claim 9.

12. The input data further includes dangerous driving data indicating the position and type of dangerous driving detected for a vehicle traveling on a road. The prediction unit inputs the dangerous driving data of the target road into the prediction model and predicts the risk of accidents in one or more target sections and / or one or more target intersections of the target road. The information processing apparatus according to claim 1 or 2.

13. The road data further includes dead angle information regarding dead angles of the road. The information processing apparatus according to claim 1 or 2.

14. A computer Using a prediction model learned by learning data with road data including the structure and / or attributes of a road as input data and accident history data indicating the history of accidents that occurred on the road and including the weather at the time of the accident as correct data, the prediction model including a plurality of weather-specific models corresponding to each of the weathers, inputs the road data of a target road not limited to the road with the accident history data into the prediction model, predicts the risk of accidents in one or more target sections and / or one or more target intersections of the target road, Causes a terminal device used by a user to output risk information indicating the predicted risks in each of the one or more target sections and / or one or more target intersections. An information processing method.

15. To a computer A prediction model learned from learning data that uses road data including the structure of a road and / or the attributes of a road as input data and accident history data indicating the history of accidents that occurred on the road and including the weather at the time of the accident as correct data, the prediction model including a plurality of weather-specific models corresponding to each of the weathers, inputs the road data of a target road not limited to the road with the accident history data into the prediction model, and has a prediction function for predicting the risk of accidents at one or more target sections and / or one or more target intersections of the target road, and an output function for causing a terminal device used by a user to output risk information indicating the predicted risk at each of the one or more target sections and / or the one or more target intersections. Program.

Images