JP7300700B2 - Pedestrian Accident Risk Evaluation Method by Point and Pedestrian Accident Risk Judgment System by Point - Google Patents

Description

The present invention relates to a method for objectively evaluating the risk of an accident between a vehicle and a pedestrian, and a system for determining the risk.

In general, most of the accidents that occur while vehicles are running are at intersections, and it is true that many accidents occur between vehicles, but not a few accidents occur between vehicles and pedestrians. Conventionally, in road design, the state of dangerous intersections has been evaluated exclusively from the driver's perspective, in view of the fact that the incidence of accidents depends on whether or not the driver perceives danger. For example, by using a traffic dynamics simulation device and acquiring the degree of danger perceived by the driver in the simulation space, it was used to design new intersections and roads while reflecting microscopic information. (See Patent Document 1).

However, even if a simulation space with various conditions is constructed, various information is collected, and a new intersection etc. is designed reflecting this information, a new intersection etc. will not be constructed based on the design. In that case, a phenomenon different from the simulation space may occur. In other words, even though the design takes safety into consideration, it is possible that changes in traffic volume, changes in pedestrian behavior, unforeseen events, etc., may lead to other dangers due to realistic factors. rice field.

JP-A-2002-140786 JP 2016-218862 A

Therefore, it has been proposed to detect speed information from an actually traveling vehicle (probe car) and evaluate an area with a high accident risk from the causal relationship with the accident (see Patent Document 2). This technology utilizes the cause-and-effect relationship between the number of accidents occurring in areas where there is a large difference between the speed at which ordinary vehicles travel and the speed at which commercial vehicles travel. For this reason, we tried to evaluate the possibility of accidents by running both general and commercial vehicles as probe cars.

However, although it is certain that changes in speed according to vehicle characteristics as described above are one of the causes of the causal relationship with the accident rate, changes in the accident rate are not due solely to changes in speed. In addition, what is generally called an accident includes cases involving vehicles and cases involving pedestrians, and it was attempted to centrally deal with these cases.

By the way, among the accidents caused by the traveling of vehicles, the proportion of accidents involving pedestrians is higher than that involving vehicles in cases causing serious damage such as fatal accidents. This is because, in the case of an accident involving a vehicle, if the vehicle is traveling at a low speed rather than at a high speed, the majority of the cases will be limited to property damage. This is because there is a high frequency of serious damage caused by

However, the evaluation of which points in the road traffic network have what degree of danger has not surfaced until now unless an accident actually occurs, and from the viewpoint of preventing pedestrian accidents , potential hazards should be evaluated even at points where accidents do not actually occur. It is also possible to assess potential hazards based on road environmental conditions at a point on the road and the amount of vehicle traffic, but this focuses exclusively on vehicle traffic. Therefore, the factor of the amount of pedestrians was neglected. However, it should not be simply determined that a point with a large number of pedestrians is dangerous. In other words, empirically, it is clear that the probability of an accident occurring at an intersection where vehicles are slowing down or stopped is low even if the traffic volume and the number of pedestrians are enormous. It is easy to understand from empirical rules that the risk of accidents is high in places where pedestrians and moving vehicles approach each other frequently, even at places where the number of people is small.

The present invention has been made in view of the above points, and its object is to measure the frequency of occurrence of situations approximating the occurrence of an accident without simply measuring the traffic volume of vehicles and the number of pedestrians. It is an object of the present invention to provide a pedestrian accident risk evaluation method and a pedestrian risk determination system for each point while paying attention to the above.

Therefore, the present invention relating to a point-by-point pedestrian accident risk assessment method selects a plurality of intersections in advance, probe vehicle traffic volume information including the traffic volume of probe vehicles equipped with a collision warning device entering the intersection, Pedestrian collision warning information including the number of collision warnings with pedestrians in the vicinity of the intersection centered by the collision warning device mounted on the probe vehicle, and walking including the number of accidents actually encountered by pedestrians and the pedestrian accident information, and calculate the ratio of the number of times the collision warning included in the pedestrian collision warning information is detected to the traffic volume of the probe vehicle included in the probe vehicle traffic information as the pedestrian collision warning occurrence rate. Then, a pedestrian accident number statistical model is created based on the correlation between the pedestrian collision warning occurrence rate and the number of accidents in the pedestrian accident information, and the pedestrian collision warning occurrence rate at the evaluation target point is used as the pedestrian accident number. It is characterized by calculating the probability of occurrence of a pedestrian accident by matching with a statistical model.

According to the above configuration, it is possible to obtain the probability of occurrence of an accident by using the accident occurrence conditions at a plurality of pre-selected intersections as a statistical model and comparing the model with the statistical model. The probability of occurrence of an accident at this time is a point that can determine the degree of danger based on the magnitude of the probability of occurrence among points where an accident actually occurred. If the conditions are similar to the point, it is also possible to determine that an accident may occur later.

In the point-based pedestrian accident risk assessment method of the present invention, in the above configuration, for each intersection, structure information including the number of branches and the high-rise or low-rise buildings or the density of buildings around the intersection; General vehicle traffic information including the traffic amount of general vehicles entering the intersection or both are further acquired, and each element included in the structure information or the general vehicle traffic information or both And for the pedestrian collision warning occurrence rate, create a pedestrian accident number statistical model based on the correlation with the number of accidents in the pedestrian accident information, and use the structure information or the general vehicle traffic information at the evaluation target point or these The probability of occurrence of a pedestrian accident is calculated by collating each element included in both and the pedestrian collision warning occurrence rate with the pedestrian accident number statistical model.

According to the above configuration, the pedestrian accident number statistical model will reflect various information such as structure information and general vehicle traffic information, and not only traffic concentration but also the situation around the intersection. It is possible to evaluate the degree of danger of potential accident occurrence. Here, as the structure information, information such as high-rise or low-rise building information is given, but the information is not limited to these, and information focusing on other conditions may be added. This information can include information such as whether the road is flat, uphill, or downhill as conditions of the road itself, and can include information such as planting conditions such as trees in the vicinity of intersections. be able to. However, if the amount of information to be added is enormous, it may not have a direct relationship with the pedestrian accident. preferably.

Further, the present invention relating to a point-by-point pedestrian accident risk assessment method is configured according to any one of the above-described inventions. Based on the ratio of the number of collision warnings detected included in the pedestrian collision warning information to the amount, a collision warning statistical model showing an average detection tendency of collision warnings is created, and the above at each point of the intersection A model deviation residual is calculated from the collision warning detection tendency indicated by the collision warning statistical model and the actual collision warning detection tendency at each point, and the pedestrian warning occurrence rate is calculated with the model deviation residual. It is characterized by being modified.

In the case of such a configuration, as a collision warning statistical model, the average relationship between the traffic volume of the probe vehicle and the number of collision warning detections is shown, and the tendency of detection of collision warnings by this statistical model and the individual By calculating the model deviation residual with the actual collision warning detection trend at the intersection, the pedestrian warning occurrence rate can be corrected. Since this modification can also be categorized with other elements (e.g. elements included in the structure information), a detailed pedestrian accident frequency statistical model can be created, and the accuracy of the accident probability can be evaluated. can be improved. When calculating the probability of accident occurrence based on the average relationship between the traffic volume of probe vehicles and the number of collision warning detections, the increase in the traffic volume of probe vehicles and the increase in the number of collision warning detections should change linearly. However, since it becomes non-linear when using the model deviance residual, the fitness of the probability calculation is improved.

Further, in the invention having each of the above configurations, the prediction of the occurrence of a pedestrian accident is performed based on each element included in the structure information and the vehicle traffic volume information at the evaluation target point and the pedestrian collision warning occurrence rate. Deriving a pre-predicted value by matching with a statistical model, and deriving a post-predicted value by correcting the pre-predicted value based on the presence or absence and number of accidents encountered by pedestrians at the evaluation target point. Of the post-prediction values, the one with the highest accident occurrence probability can be used as the final prediction value.

In the case of the above configuration, it is possible to reflect the post-prediction value that reflects the actual number of accidents in the pre-prediction value based on the statistical model of the number of pedestrian accidents. Accurate evaluation is possible. In this case, the pre-predicted value is the predicted value when compared with the pedestrian accident number statistical model, and the post-predicted value is the predicted value updated by the actual number of accidents for each evaluation target point. . Therefore, ex post facto includes the actual number of accidents in the past, and further includes the case of an increase in the number of accidents due to the occurrence of accidents after the calculation of the predicted value. If the number of accidents after calculation of the predicted value is used for updating, it is possible to cope with ever-changing traffic conditions.

Further, in the invention having each of the above configurations, the collision warning device includes image processing means capable of recognizing a person, speed recognition means, and determination for determining whether a vehicle traveling at a predetermined speed or more collides with a person within a predetermined time. and an output means for outputting a warning when a collision is determined by the determination processing means. It may be configured to be based on the number of alarms output by.

In the case of the above configuration, the collision warning device can cause the person (pedestrian) recognized by the image processing to determine the possibility of a collision from the relationship between the distance to the vehicle and the running speed of the vehicle. In addition, it is also possible to determine the possibility of collision only when the vehicle travels at a speed higher than the walking speed (for example, 7 km/h or higher). Judgment of the possibility of a collision, for example, on the premise that the speed per hour of the vehicle is maintained, within a predetermined time (for example, within 2 seconds) from the relationship with the distance to the recognized person (pedestrian) There is a method of determining that there is a possibility of collision when it is determined that the In addition, the output means for outputting the alarm includes visual and auditory warning means for the vehicle driver, and at the same time, it is possible to separately provide notification means by radio or the like to a server or the like, and the alarm is output by this notification means. Pedestrian collision warning information can be obtained by measuring the location of the collision and the number of outputs.

On the other hand, the present invention relating to the point-by-point pedestrian accident risk assessment system selects a plurality of intersections in advance and stores probe vehicle traffic volume information including the traffic volume of probe vehicles equipped with collision warning devices entering the intersections. and a collision warning device mounted on the probe vehicle stores pedestrian collision warning information including the number of collision warnings with pedestrians around the intersection. A collision warning information database, a pedestrian accident information database that stores pedestrian accident information including the number of accidents that pedestrians have actually encountered, and a pedestrian accident prediction that processes the information stored in each database. The processing means calculates the ratio of the number of collision warnings included in the pedestrian collision warning information to the traffic volume of the probe vehicles included in the probe vehicle traffic volume information, and calculates the pedestrian collision. Create a pedestrian accident number statistical model based on the correlation between the pedestrian collision warning incidence rate and the number of accidents in the pedestrian accident information, and calculate the pedestrian collision warning incidence rate at the evaluation target point. The pedestrian accident occurrence probability is calculated by matching with the pedestrian accident number statistical model.

According to the above configuration, probe vehicle traffic information, pedestrian collision warning information, and pedestrian accident information for each of a plurality of intersections are stored in respective databases, and using the information stored in these databases, the processing means can be processed by The processing means ultimately creates a pedestrian accident number statistical model, and each element included in the structure information and vehicle traffic information at the evaluation target point and specific information on the pedestrian collision warning occurrence rate is matched against the pedestrian accident frequency statistical model, it is possible to identify locations with potentially relatively high accident rates. Structure information, vehicle traffic information, and the like may be stored in the database, and the relationship between this information and the number of accidents in the pedestrian accident information can be reflected in the pedestrian accident number statistical model.

In addition, as another invention related to the point-based pedestrian accident risk assessment system, a plurality of intersections are selected in advance, and probe vehicle traffic volume information including the traffic volume of probe vehicles equipped with a collision warning device entering the intersection and a pedestrian collision warning information including the number of collision warnings with pedestrians around the intersection by a collision warning device mounted on the probe vehicle. A pedestrian collision warning information database, a pedestrian accident information database that stores pedestrian accident information including the number of accidents actually encountered by pedestrians, and the occurrence of pedestrian accidents by processing the information stored in each database and a processing means for calculating a prediction, wherein the processing means calculates the traffic volume of probe vehicles included in the probe vehicle traffic volume information at all preselected intersections, and the collision warning volume included in the pedestrian collision warning information. creating a collision warning statistical model showing an average collision warning detection tendency based on the ratio of the number of times that is detected, the collision warning detection tendency indicated by the collision warning statistical model for each intersection point, and , calculating the model deviation residual from the detection tendency of the actual collision warning at each point, and creating a pedestrian accident number statistical model based on the correlation with the number of accidents in the pedestrian accident information for the model deviation residual. and calculating the pedestrian accident occurrence probability by collating the model deviation residual at the evaluation target point with the pedestrian accident number statistical model.

According to the above configuration, it is possible to obtain the model deviation residual error for each intersection under individual conditions. can create a pedestrian accident number statistical model. In the present invention as well, structure information, vehicle traffic information, etc. can be stored in the database and reflected in the pedestrian accident number statistical model, as described above.

The invention relating to the point-based pedestrian accident risk assessment system may be configured to include an input unit for inputting individual information and an output unit for outputting the calculated pedestrian accident occurrence probability. In this case, the output section may display road information (map data) on the monitor and then display the information while changing the color according to the probability of pedestrian accidents.

According to the point-based pedestrian accident risk assessment method of the present invention, the risk is evaluated by focusing on the occurrence frequency of situations that are similar to the actual occurrence of an accident, rather than on the amount of vehicle traffic. can do. In addition, in the case of an invention configured to update the pre-predicted value to the post-predicted value, the predicted value can be obtained with the highest probability, and even after some accident prevention measures have been taken, the effect of the measures can be verified. can do. In this verification, the quantitative relationship between the frequency at which pedestrian collision warnings are detected and the actual number of accidents involving pedestrians is calculated, enabling quantitative comparison after the fact. is.

Furthermore, by widely disclosing the results evaluated by the above-described evaluation method, it is possible to call attention to areas with a high degree of danger, which can be used to prevent accidents from occurring. In particular, as a result of being able to make local residents aware of it through publication by administrative agencies, it is possible to arouse the attention of residents who use it as a community road and have a high traffic frequency.

On the other hand, according to the point-based pedestrian accident risk assessment system of the present invention, various types of information can be processed based on the evaluation method described above, and processing using data such as pedestrian collision warning information can be realized. can be In addition, it enables processing including a huge amount of data such as structure information and vehicle traffic information. In addition, individual information included in structure information, vehicle traffic information, and pedestrian collision warning information includes primary data obtained from external information agencies, as well as independently processed secondary data. As for processing, if a processing means is used, it becomes easy to process these huge amounts of data.

BRIEF DESCRIPTION OF THE DRAWINGS It is explanatory drawing which shows embodiment of the pedestrian accident risk determination system according to a point. FIG. 4 is an explanatory diagram showing an example of an output state in an output section (monitor, etc.); BRIEF DESCRIPTION OF THE DRAWINGS It is explanatory drawing which shows 1st Embodiment of the pedestrian accident risk determination method according to a point. It is explanatory drawing which shows 2nd Embodiment of the pedestrian accident risk determination method classified by point.

BEST MODE FOR CARRYING OUT THE INVENTION An embodiment of the present invention will be described below with reference to the drawings. For convenience of explanation, the point-based pedestrian accident risk determination system will be described, and then the point-based pedestrian accident risk determination method will be described.

<Pedestrian Accident Risk Judgment System by Point>
FIG. 1 is a schematic diagram showing an embodiment of a point-based pedestrian accident risk determination system. A point-by-point pedestrian accident risk determination system (hereinafter sometimes simply referred to as system) 1 of the present embodiment comprises various databases 11 to 15 and a processing device (processing means) 16 . Details of the various databases 11 to 15 include a structure information database 11 that stores structure information, a vehicle traffic information database 12 that stores vehicle traffic information, and traffic volume information of probe vehicles equipped with collision alarm devices (probe A probe car traffic information database 13 that stores car traffic information), a pedestrian collision warning information database 14 that stores pedestrian collision warning information, and a pedestrian accident information database 15 that stores pedestrian accident information. The information stored in these databases 11 to 15 is obtained from the external information 2, and includes primary data of the external information 2 as well as processed secondary data.

As the external information 2, digital road network data (for example, road network data provided by Zenrin Co., Ltd.) 21, national land numerical information urban area fine mesh data provided by the Ministry of Land, Infrastructure, Transport and Tourism (land use fine mesh data and Omitted) 22, traffic record data (for example, probe data provided by Pioneer Corporation) 23, probe car traffic data 24 and alarm data 25 by probe cars equipped with collision warning devices, and vehicles and pedestrians for each point data (the number of accidents) 26 that occurred between

The structure information stored in the structure information database 11 is, for example, referring to the digital road network data 21, the number of branches at intersections, the digital road network data 21 and the national land numerical information urban area land use subdivisions. While referring to the mesh data 22, types such as a high-rise building land use type mesh, a low-rise building land use type mesh, and a low-rise building (dense land) land use type mesh can be included.

The number of branches at an intersection is a condition based on the number of road connections, and is three for a Saegusa intersection and four for a Yotsugi intersection. In addition, high-rise building land use type mesh is a mesh for land that consists of commercial / business buildings or condominiums with 4 stories or more in areas where buildings are densely packed in residential areas, urban areas, etc., and low-rise building land The use type mesh is a mesh of land on which residential buildings of three stories or less are collectively distributed. The low-rise building (densely populated land) land use type mesh is a mesh of land where residential buildings of three stories or less are densely packed.

The vehicle traffic volume information stored in the vehicle traffic volume information database 12 can be the total value of the traffic volume of general vehicles flowing into each intersection for one year. For example, it can be processed by map-matching the information of the traffic record data 23 to the digital road network data 21 . The traffic record data 23 is, for example, a collection of car navigation system data sold by Pioneer Corporation, and can exclusively reflect the actual running of ordinary vehicles.

The probe car traffic information stored in the probe car traffic information database 13 is uniquely collected from the amount of traffic flowing into each intersection by probe vehicles (probe cars) equipped with collision warning devices. It measures the volume of traffic to intersections through unconscious passage by (a large number of) probe cars. The amount of traffic can be the total value during the period when a plurality (a large number) of probe cars travel for a predetermined period. As long as the probe car continues running, the total amount can be updated. The traffic volume at each intersection is processed by acquiring the movement state based on the position information from the collision warning device, for example, and map-matching the aggregated information (probe car traffic data 24) with the digital road network data 21. be able to.

The pedestrian collision warning information stored in the pedestrian collision warning information database 14 is particularly collected in order to obtain the number of collision warning detections by the collision warning device mounted on the probe car. Information of the probe car warning data 25 obtained by summing up the detection positions and detection times of the pedestrian collision warning in the probe car while the car is running is used. By map-matching the digital road network data 21 with the position where the collision warning is detected, it is possible to integrate the number of warning detections in the surrounding area based on the position of the intersection. The peripheral area can be, for example, within a radius of 30 m from the center of the intersection.

The pedestrian accident information stored in the pedestrian accident information database 15 utilizes the information in the accident data 26, and includes the locations of pedestrian accidents (intersections and their surroundings) that have occurred in the past several years (for example, seven years). ) and the number of cases. The periphery of the intersection can be defined as within a radius of 15 m from the center of the intersection. If this area is enlarged, it will not be sufficiently distinguished from other nearby intersections, and if it is made smaller, it will be excluded from accidents around the intersection.

As described above, various types of information are stored in the individual databases 11 to 15 as processed secondary data in addition to the primary data. It is possible to calculate the incidence rate. Specifically, for each point, the pedestrian collision warning occurrence rate is calculated by dividing the traffic volume information (traffic volume) from the probe vehicle (probe car) by the pedestrian collision warning information (number of warning detections). This is classified according to the correlation with structure information and vehicle traffic information. A statistical model of the number of pedestrian accidents is created by summarizing the classified multiple types of correlation states. Since this statistical model has multiple types of correlation states, if the above information (categorized) at other points can be obtained, it is possible to infer the probability of pedestrian accidents by comparing it with the statistical model. It becomes.

In addition, the above pedestrian accident number statistical model can apply a negative binomial regression model using structure information, general vehicle traffic information, and pedestrian collision warning information as predictive variables and pedestrian accident information as outcome variables. can. Furthermore, by using the parameters estimated by the negative binomial regression model as prior parameters and the number of pedestrian accidents included in the pedestrian accident information as new data, Bayesian update is performed to determine the risk of pedestrian accidents. (statistically expected number of pedestrian accidents), and its maximum posterior probability (MAP) estimate is used as an indicator of potential pedestrian accident risk (final pedestrian accident statistics). expected value).

In this case, the prior distribution is a gamma distribution (Poisson conjugate prior distribution), and the Shape parameter (α: φ in the negative binomial distribution) and the Rate parameter (β: φλ ⁻¹ in the negative binomial distribution) are negative By calculating using the binomial regression model of , the final statistical expected value can be obtained by performing Bayesian update with an updated model formula such as the following formula using these as prior parameters.

As mentioned above, the posterior distribution of the statistical expected value of the number of pedestrian accidents by Bayesian update is the maximum posterior probability The (MAP) estimate is used to calculate a potential pedestrian accident hazard index (statistical expected number of final pedestrian accidents). Apart from this, when checking the improvement status of pedestrian accident prevention measures (preventive construction, etc.), it is also possible to verify the effect of the measures by calculating the posterior distribution using the same Bayesian update. .

The individual and specific information stored in each of the databases 11 to 15 can be stored as secondary data by processing the various data 21 to 26 of the external information 2 by the calculation unit 17 of the processing device 16. can. The processing unit 16 is also provided with a storage unit 18 for temporarily storing data read out from the databases 11 to 15 when the calculation unit 17 performs calculations.

By connecting an output unit (monitor) 3 to this system 1, it is possible to display information stored after calculation results or processing. The output unit 3 can display specific points (intersections) in the information (map information) of the digital road network data 21, and display them by color-coding in descending order of pedestrian accident probability after processing. good. Further, the system 1 may be configured so that the input section 4 can be connected, and by inputting a command from the input section 4, the information displayed on the output section 3 can be changed as appropriate.

As an output state in the output unit (monitor, etc.) 3, there may be graphic information mapped to map information as shown in FIG. The circles shown in this figure are supposed to be color-coded when they are actually displayed, with red indicating the highest degree of danger and gradually changing from yellow to green to indicate the degree of danger in different colors. I am displaying. It should be noted that the size of the circle can be changed in addition to the color or along with the color. In addition, a small white circle “・” is also displayed inside the circle to indicate that it is an intersection where a pedestrian accident actually occurred. It can also be categorized to indicate that the number of occurrences is large.

Although the intersections based on the digital road network data are displayed in the map information on the output unit (monitor, etc.) 3, no particular display is made for intersections where the probability of pedestrian accidents is not particularly high. I am assuming. This is to provide a warning map for displaying dangerous intersections, and emphasizes that attention is paid to intersections with a high degree of danger rather than intersections with a low degree of danger. If the local government publishes the map information of the entire specific local government area by such a display method, it can contribute to arousing the attention of residents regarding pedestrian accidents.

<Method for Determining Pedestrian Accident Risk Degree by Point (First Embodiment)>
Next, a first embodiment of the point-based pedestrian accident risk determination method will be described. FIG. 3(a) conceptually shows a point-by-point pedestrian accident risk determination method (hereinafter sometimes simply referred to as a determination method) according to the first embodiment. This determination method can be roughly classified into three blocks. Quadrilaterals drawn by dashed dotted lines in the drawing indicate individual blocks. That is, a pedestrian accident number statistical model estimation (creation) block, a pedestrian collision warning occurrence rate calculation block, and a pedestrian accident occurrence probability calculation (evaluation) block.

The pedestrian accident number statistical model estimation block is a block that creates a pedestrian accident number statistical model based on the traffic volume of general vehicles, structure information, the number of pedestrian accidents, and the pedestrian collision warning occurrence rate for each point. As shown in the figure, general vehicle traffic volume uses general vehicle probe data (traffic data 23 described above) and the digital road network (digital road network data 21 described above), and structure information is The information obtained by classifying the national land numerical information urban area land use subdivided mesh data 22 is used. The number of pedestrian accidents uses the accident data 26 described above.

Pedestrian collision warnings are detected even at intersections where no pedestrian accidents have occurred, so the incidence of pedestrian collision warnings is assumed to have a significant impact on the statistical model of the number of pedestrian accidents. In other words, the pedestrian collision warning occurrence rate is calculated from the traffic volume at each location by probe cars equipped with collision warning devices and the number of warning detections at each location. information, number of accidents). Specifically, a negative binomial regression model is applied using general vehicle traffic, structure information, and pedestrian collision warning occurrence rates as predictors and the number of pedestrian accidents as response variables. The distribution obtained by applying this negative binomial regression model makes it possible to obtain the probability of a pedestrian accident occurring at a point under the same conditions.

As shown in Fig. 3(b), the probability of occurrence of pedestrian accidents by the negative binomial regression model is used as the prior distribution of the predicted values, and A distribution may be obtained. The posterior distribution can be, for example, by Bayesian updating by means of Bayesian statistics. In this case, the maximum posterior probability (MAP) estimated value from the posterior distribution can be used as the final predicted value of the number of pedestrian accidents at the determination target point.

In this case, the prior distribution can be a predicted value based on the pedestrian accident number statistical model, and the posterior distribution can be a predicted value updated by the actual number of accidents for each evaluation target point. In addition, if the posterior distribution is updated with the number of new accidents after the previously calculated predicted value has been determined, it is possible to obtain a predicted value that corresponds to ever-changing traffic conditions, and to take measures to prevent accidents. In some cases, it is also possible to verify the effect of the countermeasures.

<Method for Determining Pedestrian Accident Risk Degree by Point (Second Embodiment)>
Next, a second embodiment of the point-by-point pedestrian accident risk determination method will be described. FIG. 4A conceptually shows the determination method of the second embodiment. This determination method is also the same as in the first embodiment, and there are three blocks: a pedestrian accident number statistical model estimation (creation) block, a pedestrian collision warning occurrence rate calculation block, and a pedestrian accident occurrence probability calculation (evaluation) block. divided into blocks.

A portion different from the first embodiment is a pedestrian collision warning occurrence rate calculation block. As mentioned above, the pedestrian collision warning occurrence rate is assumed to have a large impact on the pedestrian accident number statistical model. This is what I decided to do.

That is, a first statistical model (collision warning statistical model) is created using the traffic volume by probe cars and the number of pedestrian collision warning detections at each point, and the collision indicated by the collision warning statistical model at each intersection point A model deviation residual is calculated from the warning detection tendency and the actual collision warning detection tendency at each point, and this is used instead of the pedestrian collision warning occurrence rate in the first embodiment. . Therefore, the model deviation residual is reflected in the pedestrian accident number statistical model, and the pedestrian accident number statistical model is created based on the correlation between the model deviation residual and the number of accidents in the pedestrian accident information. is.

When calculating the probability of accident occurrence based on the average relationship between the traffic volume and the number of collision warning detections, the increase in traffic volume and the increase in the number of collision warning detections will change linearly. Since it becomes non-linear when using residuals, it also has the effect of improving the fitness of probability calculation.

Also in the present embodiment, the final predicted value of the pedestrian accident rate may be calculated by Bayesian updating the prior distribution as shown in FIG. 4B to the posterior distribution.

A specific example of calculating the predicted value of the pedestrian accident incidence rate according to the embodiment described above will be described below.

<Example 1>
Pedestrian accident risk at different points was evaluated using real-world information. The evaluation target is 5462 community road intersections in Toyohashi City, Aichi Prefecture. The digital road network data uses the road network data provided by Zenrin Co., Ltd. The land use fine mesh data is provided by the Ministry of Land, Infrastructure, Transport and Tourism. The national land numerical information urban area subdivided mesh data provided was used, and the traffic performance data of general vehicles used the probe data provided by Pioneer Corporation.

The collision warning system uses Mobileye (a registered trademark of Mobileye Vision Technologies Limited), and when the vehicle is traveling at a speed of approximately 7 km/h or more, it will collide with a person recognized within 2.0 seconds. A collision warning shall be issued when judged. 39 vehicles of business operators (taxi companies) are used as probe vehicles (probe cars), and 39 of them are equipped with the above-mentioned collision warning device. used the information. In addition, since the position information of the probe car can be grasped from the position information of the collision warning device, as described later, based on the position information of the collision warning device, map matching is performed with the digital road network data. It is possible to calculate the traffic volume of cars.

The structure information, vehicle traffic information, probe car information, and pedestrian collision warning information are classified or specified as follows. As the structure information, first, information on the number of road connections was specified from the road network data. Concretely, it is information that there are 3 lines at the Saegusa intersection and 4 lines at the Yotsugi intersection. Second, based on the numerical national land information city area land use subdivision mesh data, the surrounding environment was identified as a high-rise building land use type mesh, a low-rise building land use type mesh, and a low-rise building (dense land) land use type mesh.

The vehicle traffic information was based on the probe data (general vehicle data) provided by Pioneer Corporation, and was a value obtained by totaling the traffic volume of general vehicles flowing into each intersection for one year. Specifically, position data (point data) indicating the position of each vehicle every 3 seconds is map-matched with the digital road network data. More specifically, by dividing the vehicle position data (point data) by trip and calculating the nearest road link (neighboring link) from the position data (point data), the link cost of the neighboring link can be reduced to 1/10. was reduced to In addition, the shortest route search is performed by the Dijkstra method from the first departure point to the last arrival point of each trip divided by trip, and neighboring links limited to road links on the extracted shortest route are recalculated. bottom. A link is a road segment between nodes, and a node is a node such as an intersection (including other points on the road network representation).

The probe car information is data on the traffic volume of the probe car (probe vehicle), and this traffic volume data is based on the position information of the collision warning device, and uses the longitude and latitude data of the probe car about every 10 seconds. The information obtained for one year (September 1, 2017 to August 31, 2018) was calculated by map matching with the digital road network data, and obtained by totaling them. The traffic volume in this case is calculated based on the position data (point data) obtained by the probe car, using the same method as the above-described method for calculating the traffic volume of general vehicles.

Pedestrian collision warning information shall use the information that the collision warning device mounted on the probe car (probe vehicle) has been activated, that is, the mobile eye (registered trademark of the company), and at a speed of about 7 km / h or more This system utilizes a collision warning that is issued under a situation where it is determined that a person recognized when the vehicle is running will collide within 2.0 seconds. The location where the collision warning system has been activated can be grasped from the position information of the collision warning system. Therefore, we counted the number of times that this collision warning system was activated and used it as the number of pedestrian collision warnings for each intersection. In addition, in order to count as collision warnings related to intersections, we decided to count the number of cases where the alarm was activated within a radius of 30 m from the center of the intersection as the number of cases at the intersection.

Here, the pedestrian collision warning occurrence rate was calculated based on the information shown above. In other words, the pedestrian collision warning occurrence rate is the number of pedestrian collision warning occurrences per probe car traffic volume, and can be calculated by dividing the number of pedestrian collision warning occurrences at each intersection by the probe car traffic volume. It is possible.

Pedestrian accident information was calculated based on data on pedestrian accidents that actually occurred in the past. Specifically, in order to limit the number of pedestrian accidents related to intersections, we aggregated data for seven years from 2008 to 2015 (excluding 2012) that occurred within a radius of 15m from the center of each intersection, and collected data for each intersection. number of pedestrian accidents per year.

Using the above information, we created a statistical model of the number of pedestrian accidents. For the pedestrian accident number statistical model, a negative binomial regression model was applied using structure information, general vehicle traffic information, and pedestrian collision warning information as predictive variables, and pedestrian accident information as outcome variables. Coefficient parameters and variance parameters of predictor variables estimated by the negative binomial regression model are shown in Table 1.

Estimated values such as the coefficient parameter and the variance parameter of the predictor variables are shown in Table 1. As an evaluation of this calculation result, the pedestrian collision warning occurrence rate is 3.31, so the negative binomial According to the characteristics of the regression model, if the pedestrian collision warning occurrence rate decreases by 10%, the statistically expected value of the number of accidents (accident risk) is 1.0-exp ^{(-1 x 3.31)} = A 28.18 (approximately 30)% reduction is shown.

Here, based on the above statistical model, the conditions belonging to the top 1% of the distribution (meaning the highest pedestrian collision warning occurrence rate) are the highest risk, and the conditions belonging to the range of 1% to 2% are the conditions. High risk, conditions in the range of 2% to 5% are medium risk, conditions in the range of 5% to 10% are low risk, conditions in the range of 10% to 20% are general risk, and so on. Others were classified as no risk. By comparing each intersection with each condition, the degree of danger for each intersection is evaluated (determined) and mapped on the map data to visualize the points and degree of danger as illustrated in FIG. It became possible to grasp the

Furthermore, in this embodiment, in order to grasp the potential pedestrian accident hazard at each intersection, the posterior distribution is calculated by the Bayesian statistical method (Bayes update) to obtain the maximum posterior probability (MAP) estimate. I decided to Specifically, the distribution of the probability of occurrence of pedestrian accidents (statistical expected value) according to the negative binomial regression model is defined as a prior distribution, and this prior distribution is defined as a gamma distribution (a conjugate prior distribution of Poisson distribution). The a priori Shape parameter (α: φ in negative binomial distribution) and a priori Rate parameter (β: φλ ⁻¹ in negative binomial distribution) corresponding to the intersection were calculated using a negative binomial regression model, respectively. Furthermore, using these prior parameters and the actual number of pedestrian accidents at each intersection as new data, Bayesian update is performed using the updated model formula described in Equation 1 above, and the posterior distribution of the statistically expected number of pedestrian accidents is calculated. We then used the maximum a posteriori probability (MAP) estimate as an indicator of potential pedestrian accident risk (the statistical expectation of the eventual number of pedestrian accidents). 1% or less, 1% to 2%, 2% to 5%, 5% to 10%, 10% to 20% of the top (meaning the highest pedestrian collision warning occurrence rate) in the posterior distribution in this case , and other ranges were used to evaluate the degree of danger of each intersection, and the state as illustrated in FIG. 2 was able to be visually grasped.

Although the creation of the prior distribution is not essential, it is possible to reflect the actual occurrence of accidents in the statistical values by Bayesian updating the actual number of pedestrian accidents as new data. Secondly, among the intersections under the same conditions, the degree of risk of places where pedestrian accidents have actually occurred is updated to a high evaluation. It is possible to assess the potential accident hazards in places where there is no safety.

<Example 2>
In addition, as a second example, the degree of accident risk of pedestrians at each point was evaluated using actual information in the same manner as described above. The evaluation target is 5463 community road intersections in Toyohashi City, Aichi Prefecture (one more point than in Example 1). , is the same as in Example 1. In addition, pedestrian warning devices were also selected in the same way, and 39 probe cars (probe vehicles) were also selected. Various types of information used for processing (structure information, vehicle traffic information, probe car information, pedestrian collision warning information, and pedestrian accident information) are the same as in the first embodiment.

In this embodiment, instead of using the probability of occurrence of a pedestrian collision warning as an index, a statistical model of collision warning is created and model deviation residual is used as the index. Specifically, the traffic volume of probe cars at each intersection was used as a predictor variable, and the number of pedestrian collision warnings at each intersection was used as a response variable to create a statistical model applied to the coefficient data. More specifically, a negative binomial regression model was applied to the prediction and response coefficients above to estimate the coefficient and variance parameters of the predictor variables. Then, we calculated the model deviation residual from the estimated negative binomial regression model for the number of pedestrian collision warnings at each intersection.

Including the model deviation residual calculated in this way, a pedestrian accident frequency statistical model was created. For the pedestrian accident number statistical model, a negative binomial regression model was applied using structure information, general vehicle traffic information, and the above-mentioned model deviation residual as predictor variables, and pedestrian accident information as an outcome variable. Coefficient parameters and variance parameters of predictor variables estimated by the negative binomial regression model are shown in Table 2.

Estimated values such as coefficient parameters and variance parameters of the predictor variables are shown in Table 2. As an evaluation of the calculation results, as far as comparing with the AIC (Akaike's Information Criterion) of Example 1, Since the AIC of this embodiment is smaller, it can be said that the degree of matching is favorable. This AIC is an index showing the suitability of the statistical model, and the smaller the number, the better the model. Such a difference is that the probability of occurrence of a pedestrian collision warning in Example 1 is the average value (linear) of the number of pedestrian collision warnings with respect to the traffic volume of probe cars, whereas by using the model deviation residual, This is probably because the relationship between the traffic volume of probe cars and the number of pedestrian collision warnings becomes non-linear.

Points with high model deviation residuals mean points where many pedestrian collision warnings are occurring in relation to the average relationship between the traffic volume of probe cars and the number of pedestrian collision warnings. Since it is thought that there are a large number of pedestrians who are in a situation close to an accident, it is possible to improve the compatibility by using the model deviation residual and reflecting it in the pedestrian accident number statistical model. Become.

The evaluation after creating the pedestrian accident number statistical model is the same as in Example 1. 2% to 2%, 2% to 5%, 5% to 10%, 10% to 20%, and other ranges are used to evaluate the degree of danger at each intersection, and visually grasp the situation as illustrated in Fig. 2. I was able to make it possible. Bayesian update also enabled the evaluation of the risk based on the posterior distribution.

Although the embodiments and examples according to the present invention are as described above, they are merely examples of the present invention and are not meant to limit the present invention to the above-described embodiments and the like. Therefore, it is also possible to change each requirement according to the above embodiment as appropriate or add other elements. For example, collision warning devices from other manufacturers may be used, and the number of samples may also be increased with more probe cars. In addition, when outputting to the output unit, the highest risk to general risk from the top 1% or less, 1% to 2%, 2% to 5%, 5% to 10%, and 10% to 20% of the distribution , and others are classified as having no risk, but this range may be changed as appropriate and appropriate ranges may be set for each region.

1 Pedestrian Accident Risk Judgment System by Location 2 External Information 3 Output Unit 4 Input Unit 11 Structure Information Database 12 Vehicle Traffic Information Database 13 Probe Car Traffic Information Database 14 Pedestrian Collision Warning Information Database 15 Pedestrian Accident Information Database 16 processing device (processing means)
17 calculation unit 18 storage unit 21 digital road network data 22 land use subdivided mesh data 23 traffic record data 24 probe car traffic data 25 probe car warning data 26 accident data

Claims (8)

A method for evaluating pedestrian accident risk for each point using a system comprising a database storing various types of information and a processing device for processing stored in the database, comprising:
By the processing device,
At a plurality of preselected intersections , probe vehicle traffic information including the amount of traffic that probe vehicles equipped with collision warning devices enter into the intersections, and collision warning devices mounted on the probe vehicles. Acquire pedestrian collision warning information including the number of collision warnings with pedestrians detected in the surrounding area and pedestrian accident information including the number of accidents actually encountered by pedestrians,
Calculating the ratio of the number of times the collision warning included in the pedestrian collision warning information is detected to the traffic volume of the probe vehicle included in the probe vehicle traffic volume information as a pedestrian collision warning occurrence rate,
Create a pedestrian accident number statistical model based on the correlation between the pedestrian collision warning occurrence rate and the number of accidents in the pedestrian accident information,
The probability of occurrence of a pedestrian accident is calculated by collating the pedestrian collision warning occurrence rate at the evaluation target point with the pedestrian accident number statistical model.
A point-by-point pedestrian accident risk assessment method characterized by: In the point-by-point pedestrian accident risk assessment method according to claim 1,
By the processing device,
For each intersection, structure information including the number of branches and high-rise or low-rise buildings or density of buildings around the intersection, and general vehicle traffic information including the traffic volume of general vehicles entering the intersection. further obtain either or both,
A pedestrian accident number statistical model is created based on the correlation with the number of accidents in the pedestrian accident information for each element included in the structure information or the general vehicle traffic information or both and the pedestrian collision warning occurrence rate. death,
Occurrence of pedestrian accidents by collating each element included in the structure information or the general vehicle traffic information or both of them at the evaluation target point and the pedestrian collision warning incidence rate with the pedestrian accident number statistical model calculate the probability
A point-by-point pedestrian accident risk assessment method characterized by: The prediction of the occurrence of the pedestrian accident is
Each element included in the structure information and vehicle traffic information at the evaluation target point and the pedestrian collision warning occurrence rate are collated with the pedestrian accident number statistical model to derive a pre-predicted value,
Deriving a post-prediction value by correcting the pre-prediction value based on the presence or absence and number of accidents that pedestrians subsequently encountered at the evaluation point,
3. The point-by-point pedestrian accident risk assessment method according to claim 2, wherein, of the posterior predicted values, the one with the highest accident occurrence probability is used as the final predicted value. In the point-by-point pedestrian accident risk assessment method according to any one of claims 1 to 3 ,
By the processing device,
Based on the ratio of the number of times the collision warning contained in the pedestrian collision warning information is detected to the traffic volume of probe vehicles contained in the probe vehicle traffic volume information at all preselected intersections, an average collision Create a collision warning statistical model that shows the detection tendency of the warning,
calculating a model deviation residual from the collision warning detection tendency indicated by the collision warning statistical model at each point of the intersection and the actual collision warning detection tendency at each point;
A pedestrian accident risk evaluation method for each location, wherein the pedestrian collision warning occurrence rate is corrected using the model deviation residual. The collision warning device includes image processing means capable of recognizing a person, speed recognition means, determination processing means for determining that a vehicle traveling at a speed equal to or higher than a predetermined speed will collide with a person within a predetermined time, and the determination processing means and output means for outputting an alarm when it is determined that a collision will occur,
The point-by-point pedestrian accident risk assessment according to any one of claims 1 to 4, wherein the detection count of collision warnings included in said pedestrian collision warning information is based on the number of warnings output by said output means. Method. A plurality of intersections are selected in advance, and a probe vehicle traffic volume information database storing probe vehicle traffic volume information including the traffic volume of probe vehicles equipped with collision warning devices entering the intersections; and a collision mounted on the probe vehicle. A pedestrian collision warning information database that stores pedestrian collision warning information including the number of collision warnings with pedestrians around the intersection centered on the intersection by the warning device, and a pedestrian collision warning information database that contains the number of accidents that pedestrians have actually encountered Pedestrian accident information database storing pedestrian accident information, information stored in each of the probe vehicle traffic information database, the pedestrian collision warning information database, and the pedestrian accident information database is processed to detect a pedestrian accident. and a processing means for calculating the predicted occurrence of
The processing means
Calculating the ratio of the number of times the collision warning included in the pedestrian collision warning information is detected to the traffic volume of the probe vehicle included in the probe vehicle traffic volume information as a pedestrian collision warning occurrence rate,
Create a pedestrian accident number statistical model based on the correlation between the pedestrian collision warning occurrence rate and the number of accidents in the pedestrian accident information,
A point-by-point pedestrian accident risk assessment system, wherein the probability of occurrence of a pedestrian accident is calculated by collating the pedestrian collision warning occurrence rate at an evaluation target point with the pedestrian accident number statistical model. . A plurality of intersections are selected in advance, and a probe vehicle traffic volume information database storing probe vehicle traffic volume information including the traffic volume of probe vehicles equipped with collision warning devices entering the intersections; and a collision mounted on the probe vehicle. A pedestrian collision warning information database that stores pedestrian collision warning information including the number of collision warnings with pedestrians around the intersection centered on the intersection by the warning device, and a pedestrian collision warning information database that contains the number of accidents that pedestrians have actually encountered Pedestrian accident information database storing pedestrian accident information, information stored in each of the probe vehicle traffic information database, the pedestrian collision warning information database, and the pedestrian accident information database is processed to detect a pedestrian accident. and a processing means for calculating the predicted occurrence of
The processing means
Based on the ratio of the number of times the collision warning contained in the pedestrian collision warning information is detected to the traffic volume of probe vehicles contained in the probe vehicle traffic volume information at all preselected intersections, an average collision Create a collision warning statistical model that shows the detection tendency of the warning,
calculating a model deviation residual from the collision warning detection tendency indicated by the collision warning statistical model at each point of the intersection and the actual collision warning detection tendency at each point;
creating a pedestrian accident number statistical model based on the correlation between the model deviation residual and the number of accidents in the pedestrian accident information;
A point-by-point pedestrian accident risk determination system, wherein the probability of occurrence of a pedestrian accident is calculated by collating the model deviation residual at an evaluation target point with the pedestrian accident number statistical model. For each intersection, a structure information database storing structure information including the number of branches and high-rise or low-rise buildings or density of buildings around the intersection ; Either one or both of a vehicle traffic information database that stores vehicle traffic information,
The pedestrian accident number statistical model in the processing means is created based on the correlation between each element included in the structure information, the general vehicle traffic information, or both, and the number of accidents in the pedestrian accident information. and
The probability of occurrence of pedestrian accidents is calculated by collating each element included in the structure information, the general vehicle traffic information, or both at the evaluation target point with the pedestrian accident number statistical model. The point-based pedestrian accident risk determination system according to claim 6 or 7.

Images