JP2009282873A - Accident warning reporting system using bayesian network - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To solve problems with the complexity of accident causes by using a Bayesian network, and calculating accurate accident probabilities on the basis of previous accident data. <P>SOLUTION: A traffic management center has a data unit for retaining previous accident data. A Bayesian network system has a Bayesian network forming unit for forming a probability table for each cause of an accident at the occurrence of an accident, based on the previous accident data about a certain dangerous location and data on the travel of each vehicle passing through the location, and a Bayesian network computing unit for obtaining, when a vehicle driven approaches the dangerous location, information about the vehicle and information about the states of various roads, and comparing these pieces of information with the probability table formed by the Bayesian network forming unit, so as to calculate an accident probability that the vehicle is facing. A driver I/F unit obtains the vehicle information about each cause of an accident and reports an accident warning to the driver. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to an accident warning notification system for notifying a driver of a vehicle when a vehicle driven by the driver approaches a dangerous place where traffic accidents frequently occur.

  As an accident warning notification system, warning information is used for the purpose of estimating a predicted accident from general accident data extracted from past accident data and information on the vehicle the driver is driving, and preventing the predicted accident. Is presented to the driver.

  In addition, accident data is acquired for each road in advance, and then when the driver actually approaches a certain accident occurrence point, the VICS system (GPS, etc.) recognizes the position information and the accident at that position. There is a method of presenting a warning message to the driver, which is considered appropriate based on a comparison between the data and vehicle data driven by the driver.

  Furthermore, there is a method of calculating and evaluating a risk value for an accident occurrence as a sum of accident factors set as a plurality of parameters.

Japanese Patent Laid-Open No. 11-120478 JP 2003-14474 A Japanese Patent Laid-Open No. 11-306458 JP 2006-330833 A

  However, when the above-mentioned vehicle approaches a place where there has been a traffic accident in the past, it compares common points in the comparison between past accident data and the current situation of the vehicle, and whether there is a common point there, In the method of issuing a warning message, warning information corresponding to individual road conditions cannot be presented to the driver.

  In addition, accident data is acquired for each individual road in advance, and then when the driver actually approaches a certain accident occurrence point, the VICS system (GPS, etc.) recognizes the position information and the accident at that position. In the method of presenting warning messages to the driver, which are considered appropriate from the comparison of the data and the vehicle data driven by the driver, or in the comparison between the accident data and the vehicle data driven by the driver, the past accident In the method of comparing the common points in the comparison of data and the current situation of the vehicle and outputting a warning message depending on whether there is a common point there, a state where multiple accident factors are entangled (complex system) ), The true factor cannot be pointed out accurately.

  Furthermore, the accident factor is set as a plurality of parameters, and in the method of calculating and evaluating the risk value for the occurrence of the accident as the sum of them, each accident factor is basically independent in a probabilistic sense. It can not be said that it represents the degree of risk for. For example, when multiple accident factors, such as excessive speed due to rain, occur, the risk is not a simple sum of the risk levels for each accident factor (linear), but an accident when multiple factors overlap. If the probability of occurrence (non-linear), that is, the probability of occurrence of an accident under the condition of excessive speed due to rain, for example, is not considered, the true risk should not be expressed. In other words, in the conventional technology, when a plurality of accident factors are tangled, the degree of risk is not accurately calculated, and therefore, it cannot be said that the danger of the accident can be accurately indicated.

In the present invention, a Bayesian network is used for the self-warning notification system, and in a complex system in which multiple accident factors are entangled, more accurate accident probabilities are calculated from past accident data, and the problems related to the complexity of accident factors are eliminated. It consists of a traffic management center that manages accident data, a driver I / F unit that consists of a vehicle information acquisition unit and a warning notification unit, and an accident warning notification system that consists of a Bayesian network system.
The traffic management center has a data section for storing past accident data such as police, and the Bayesian network system is based on past accident data in a certain dangerous place and traveling data of each vehicle passing through the place. A Bayesian network forming unit that creates a probability table for each accident factor with respect to the occurrence of an accident, and when a traveling vehicle approaches the dangerous place, obtains information on the vehicle and various road conditions, and Comparing with the probability table formed in the network forming unit, it has a Bayesian network calculation unit that calculates the accident probability that the vehicle is facing,
The driver I / F unit is composed of parts for obtaining vehicle information related to each accident factor and notifying the driver of an accident warning, and calculating an actual accident occurrence probability from past accident data in the data unit. Even when a plurality of accident factors are entangled, the Bayesian network system calculates the accident occurrence probability more accurately and notifies the driver I / F unit of the result.

According to the present invention, each person can be obtained by calculating a more accurate accident occurrence probability. For example,
(1) For the driver, it is possible to predict the accident more accurately and prevent the accident from occurring.
(2) The police station can be expected to reduce the number of accidents.
(3) For automobile manufacturers or car navigation manufacturers, it is considered that new demand will arise for cars and car navigation systems.
(4) It is considered that a demand for a system related to the present invention is generated for a system construction manufacturer.

  Also, for example, if the lower limit value of the accident occurrence probability that the driver notifies the warning is set in advance, and the accident occurrence probability is low, the warning is not notified, so that the warning notification frequently occurs for the driver described above. It also leads to elimination of annoyance.

  The invention of claim 1 is an accident warning notification system comprising three parts: a data part, a Bayesian network system, and a driver I / F part. The data section means a center that manages accident information, such as the police, and is collectively referred to as a traffic accident management center in the present invention. The Bayesian Network System Department forms a probability table for accidents using past accident data, etc., and calculates the accident probability by applying the currently running vehicle data to the formed Bayesian network. Indicates the site to be performed.

The driver I / F unit is a part that acquires position information during driving, various vehicle data that leads to accident factors, and calculates accident probability by passing these data to the Bayesian network system. If necessary, the driver I / F section issues an accident warning to the driver.
Note that the Bayesian network system is a Bayesian network based on past accident data held by a traffic accident management center and the like, and traveling data of each vehicle passing through the place, for example, speed, weather, daily illuminance, etc. In the forming unit, a probability value table is created for each factor that leads to the accident.

  Further, when the traveling vehicle approaches the dangerous place, for example, information on various vehicles such as speed, weather, and illuminance is acquired, and the probability value table formed in the Bayesian network forming unit By comparison, the accident probability that the vehicle is facing is calculated by the Bayesian network calculation unit. As a result, the probability of occurrence of an actual accident is calculated from past accident data. For example, even when a plurality of accident factors are intertwined, the probability of occurrence of the accident can be calculated more accurately.

  The invention of claim 2 has a function of notifying a driver of an accident warning when the accident occurrence probability calculated by the Bayesian network computing unit is equal to or higher than a preset threshold, The driver can set the threshold by himself / herself. Thus, the driver himself can set the warning frequency. That is, the troublesomeness caused by the frequent occurrence of warning information notification to the driver as pointed out in Patent Document 4 can be eliminated.

  In the invention of claim 3, the traveling vehicle obtains information on its position, performs a calculation by a Bayesian network when the position is approaching a dangerous place, and if it is not approaching, the vehicle again Information about the position is to be acquired. As a result, it is possible to save time and effort to calculate and form a Bayesian network each time position information is acquired, and to quickly calculate and form a Bayesian network when approaching a location that is truly dangerous.

  According to a fourth aspect of the present invention, in the traveling vehicle, after the accident probability is calculated by the Bayesian network calculation unit, a Bayesian network is formed again using various vehicle data used at the time of the Bayesian network calculation. As a result, a Bayesian network is sequentially formed, and a calculation using the latest Bayesian network is always performed for a running vehicle.

  In the invention of claim 5, when notifying the driver of an accident warning, the risk of the accident is notified, and the combination of each accident factor when the accident probability exceeds the threshold is extracted, and each accident factor By giving a warning to the driver regarding the point, it is possible to clarify the matters that the driver should pay attention to. Thus, the driver can clearly recognize what is actually to be noted.

  According to the invention of claim 6, a questionnaire is given to the driver in advance, and registration regarding the precautions specific to each driver is performed. For example, if it is necessary to pay attention to the speed, registration regarding the speed is performed. And when the driver approaches a certain dangerous place, regarding the pre-registered accident factor, the accident factor is actually occurring, for example, when the speed exceeds the speed limit, By lowering the warning notification threshold for the driver, a warning specific to each driver can be notified.

  In other words, compared to the conventional accident warning notification system, when calculating the probability of accident occurrence, the accident occurrence probability is not simply calculated by adding the probability that each accident factor actually leads to an accident (linear) Using Bayesian networks, it is meaningful to analyze each accident factor from actual accident data in a complex manner (non-linear) and calculate the accident occurrence probability based on the analysis.

  Examples of the present invention will be described below. FIG. 1 is a configuration example of an accident warning notification system showing an example of the present invention. In the figure, an accident warning notification system 0101 passes a function of forming a Bayesian network based on past accident data and the like and a function of forming data of various vehicles through a Bayesian network formed by the function of forming the Bayesian network. Thus, the system has a function of calculating the accident probability and issuing warning information to the driver according to the accident probability.

1 is an accident warning notification system 0101 composed of a traffic accident management center 0111, a Bayesian network system 0102, and a driver I / F unit 0103. In each system unit, for example, the Bayesian network system 0102, a Bayesian network Network formation and computation are performed, and the driver I / F unit 0103 is located at a contact point with an actual driver, and has functions such as acquisition of data regarding the vehicle and notification of warning information to the driver. The traffic accident management center 0111 means a center that holds data 0110 related to the cause of an accident when an accident occurs, and specifically corresponds to the police. These data are used by the Bayesian network forming unit 0114 to form a Bayesian network.
Next, actual processing related to FIG. 1 will be described. First, the accident warning notification system 0101 can be mainly divided into the following two according to the function. One is a function for forming a Bayesian network, and the other is a function for calculating the actual accident occurrence probability by passing the traveling vehicle data through the Bayesian network. For each function, the main execution parts are a Bayesian network forming unit 0114 that forms a Bayesian network and a Bayesian network calculating unit 0115 that performs operations of the Bayesian network.

  Hereinafter, these parts will be described separately. First, the Bayesian network forming unit 0114 holds the past accident data 0110 held by the traffic accident management center 0111 and the vehicle data from the traveling vehicle as vehicle data 0113 when the Bayesian network is formed. Based on the above, it indicates the part that actually forms a Bayesian network. In addition, the Bayesian network formed in this way is held again as vehicle data 0113, and when the Bayesian network is formed next time, by adding new vehicle data to the vehicle data 0113, the Bayesian network is again formed. Reshape the Bayesian network.

  A specific formation method will be described in Example 4 described later. The Bayesian network calculation unit 0115 acquires the vehicle information of the current driver from the vehicle information acquisition unit 0105 in the driver I / F unit 0103, and the vehicle information is created by the Bayesian network formation unit 0114. By passing through the Bayesian network, the part having the function of calculating the accident probability for the driver is indicated. A specific calculation method is described in Example 4. If the accident probability is higher than the threshold value 0116, the warning content is transmitted to the warning notification unit 0119 of the driver I / F unit 0103 through the warning content transmission unit 0118, and a warning is given to the driver through a speaker or the like. In this case, regardless of the threshold value 0116, the driver's vehicle information is used for forming a new Bayesian network through the vehicle data acquisition unit 0117.

  In these processes, the accident data 0110 of the traffic accident management center 0111 or the data related to the place 0106 designated as the special danger area by the traffic accident management center 0111 is held as the dangerous area data 0107, and these are held. The dangerous area data 0107 is obtained by receiving the position information 0109 acquired from the vehicle information acquisition unit 0105 from the position information acquisition unit 0104 in the traveling vehicle and comparing it with the dangerous area data 0107. Is generated or designated as a special danger area, 0108 is to calculate and form an actual Bayesian network.

  If no accident has occurred in the past or the area is not a special danger area, position information acquisition 0109 is performed again, and comparison with the danger area data 0107 is repeated. Details will be described in Example 2.

  FIG. 2 is a processing flowchart of the accident warning notification system. In this flow, first, in the traveling vehicle, information on the position of the traveling vehicle is acquired (acquisition of position information (step 0201)). For example, it is assumed that the information on the position is acquired through a car navigation system (with a GPS function) possessed by the driver.

Subsequently, it is determined whether or not it is actually necessary to form or calculate a Bayesian network at the position of the acquired position information. This is because forming and calculating a Bayesian network each time position information is acquired takes time for the processing, and it is practically difficult to realize it. In addition, when making such a determination, there may have been no accident in the past at that location, or a traffic accident management center (the police, etc.) may issue a special danger area (a special warning about the accident by the police, etc.) Judgment whether it is not specified in the place specified in (5), etc., and decide on the necessity of such accident prediction.
In this embodiment, vehicle information is acquired (step 0203) only when it is determined that an accident has actually occurred in the past (step 0202) or a special danger zone (step 0209). Through the actual Bayesian network, the probability of occurrence of an accident that the vehicle faces is calculated (step 0204). A specific Bayesian network calculation method will be described in the fourth embodiment.

  It should be noted that, as in this embodiment, considering an accident in the past (step 0202) or a special danger area (step 0209) as an area that requires an accident prediction is merely an example, and in actuality, flexibly The area will be designated. Subsequently, when the accident occurrence probability obtained in this way is equal to or higher than a threshold value predetermined by the driver (step 0205), warning information is notified to the driver (step 0206).

  In addition, such a threshold value can be set by the driver, for example. In other words, if the driver sets this threshold value high, warning information will be issued only in situations where the danger is really imminent, and conversely, if this threshold value is set low, the driver will detect any dangerous situation. be able to.

  Subsequently, after warning information is notified to the driver in this way, registration in the Bayesian network is performed (step 0207). Such registration to the Bayesian network is performed even if the previous warning information notification (step 0206) is not performed.

  In this way, a Bayesian network based on new vehicle data is formed one by one. Thereafter, acquisition of position information (step 0201) is performed again, and warning information is notified as necessary.

  FIG. 3 describes the processing sequence between the Bayesian network system 0102 and the driver I / F unit 0103 in FIG. 1 regarding the mutual processing between the driver I / F unit and the Bayesian network calculation unit.

  First, the location information acquired by the GPS or the like in the location information acquisition unit 0301 is taken into the Bayesian network system 0102 by the location information acquisition unit 0104 in the Bayesian network system 0102, and based on the location information, past information is acquired. The presence / absence of accident data and the forced acquisition area are determined (step 0108). If it is determined that a Bayesian network needs to be formed or calculated, then vehicle information acquisition (step 0303) is performed. Specifically, traffic information and accident information using the VICS system is acquired from the traffic center (step 0302), and data such as speed, daily illuminance, and driver attributes are acquired from the in-vehicle detection device. Step 0304). Subsequently, after obtaining the vehicle information, the data is passed through the Bayesian network calculation unit 0115 to calculate the actual accident probability.

  If the accident probability is equal to or greater than the threshold (step 0116), the warning content is transmitted (step 0118) and the information is sent to the warning notification unit 0119. Further, registration with the Bayesian network (step 0305) is performed in order to use the data for the next accident data.

  Note that the above-mentioned driver attributes include, for example, a questionnaire for each driver in advance, and registration regarding precautions peculiar to the driver as each accident factor such as rain, curve, and daily illuminance.

  When calculating the actual accident probability, for example, if it is specified in advance by a driver's questionnaire that more attention is required regarding the curve, the threshold for the accident probability when the curve is approached Measures such as lowering are taken. By doing in this way, notification of warning information tailored to the driver can be performed.

  In the present embodiment, an actual example of the formation and calculation of the Bayesian network will be described. In the description of this section, only the weather and speed are taken as examples of the accident factors for the sake of simplicity. However, in reality, as shown in Fig. 3, it is assumed that traffic jam information, accident information, speed, daily illuminance, driver attributes, etc. are also included as accident factors. In this regard, a Bayesian network is formed. Therefore, even if multiple accident factors are included, the basic concept of Bayesian network formation and calculation will not change, so it will be omitted in this section (details are given at the end of this section). . Based on the above, first, only the formation of a Bayesian network will be described.

  First, there is Bayesian theory as a fundamental principle in forming and calculating such a Bayesian network. Bayesian theory is a theorem that obtains the probability of the next event when a certain result (data) is obtained. The following formula is given as a formula based on this concept.

P (A | B) = P (B | A) x P (A) / P (B) (1)
In the above equation, the conditional probability P (A | B) that A occurs when B occurs is the conditional probability P (B | A) that B occurs when A occurs and the probability P (A and B each occur) A) and P (B). When this is applied to this accident warning notification system as A: Weather, B: Speed, C: Accident,
P (C | A, B) = P (A, B | C) x P (C) / P (A, B) (2)
It becomes.

  However, each factor (for example, weather, speed, etc.) that is related to the occurrence of the accident is assumed to be independent as described in FIG.

  In other words, A: weather, B: under the speed C: the conditional probability P (C | A, B) of the accident is C: the conditional probability P (A, B | C) under the accident C: Accident probability P (C), A: Weather and B: Speed probability P (A, B). Therefore, with respect to the above equation (2), a table (FIG. 5A, B, C) regarding the probability value of the right term is created.

  First, the probability regarding (1) P (A, B | C = 1) ”in FIG. 5 is obtained from past accident data (0110). For example, if the probability that the weather is rainy (A = 1) and the speed is above the speed limit (B = 1) when an accident has occurred (C = 1) in the past, P (A = 1, B = 1 | C = 1) is obtained. Similarly, if the probability that an accident has occurred in the past (C = 1) and the weather is rain (A = 1) and the speed is below the speed limit (B = 0) is obtained, P (A = 1 , B = 0 | C = 1). The same applies to P (A = 0, B-1 | C = 1) and P (A = 0, B = 0 | C = 1). Further, the probabilities (2) and (3) in FIG. 5 are obtained from the vehicle data (0117). That is, every time a traveling vehicle, for example, “has an accident in the past” or approaches a special danger area, the Bayesian network forming unit 0114 recalculates the probabilities of P (A, B) and P (C). It will be. A Bayesian network is formed based on the probabilities acquired as described above.

  In the specific example of FIG. 5, a specific probability value is applied as an example, but there is no basis for the numerical value. Actually, the probability values in FIG. 5 are calculated from the police and the traveling vehicle. The formation of the Bayesian network as described above is performed by the Bayesian network forming unit 0114 in FIG. 1, and the table information is held once in the vehicle data (0113). Thereafter, the Bayesian network forming unit 0114 re-creates the Bayesian network every time new accident data or travel data is received.

  Next, using the Bayesian network formed as described above, a method for calculating the probability of occurrence of an accident when the vehicle has actually caused an accident in the past or approaches a special danger area will be described. With respect to the table of FIG. 5 created in the formation of the Bayesian network, by comparing with the above equation (2), it is possible to calculate a probability value in which each factor leads to an actual accident. A specific calculation formula for calculating the probability value is shown below.

(1) P (C = 1 | A = 1, B = 1) = P (A = 1, B = 1 | C = 1) × P (C = 1) / P (A = 1, B = 1) = 0.1 × 1/5 = 0.02%
(2) P (C = 1 | A = 1, B = 0) = P (A = 1, B = 0 | C = 1) × P (C = 1) / P (A = 1, B = 0) = 0.1 × 1/10 = 0.01%
(3) P (C = 1 | A = 0, B = 1) = P (A = 0, B = 1 | C = 1) × P (C = 1) / P (A = 0, B = 1) = 0.7 × 1/50 = 0.014%
(4) P (C = 1 | A = 0, B = 0) = P (A = 0, B = 0 | C = 0) × P (C = 0) / P (A = 0, B = 0) = 0.1 × 1/35 = 0.003%
In order to obtain the left term in the above equation, the probabilities in FIG. 5 obtained above may be substituted. The above formula shows the result of actually substituting the probability values shown in FIG. Here, for example, if the accident occurrence threshold is set to 0.02% or more, among the above formulas, when rain (A = 1) and speed are above the speed limit (B = 1) Only when the accident probability exceeds the threshold, warning information for the driver is notified.

  As described above, when a Bayesian network is formed, the calculation is performed according to the above-described calculation formula. Regarding the calculation and formation of an actual Bayesian network, for example, a Bayesian network construction support system (Baysian Network Construction) sold by Mathematical Systems Inc. System: BayoNet) (http://www.msi.co.jp/BAYONET) and HUGIN Explorer (http://www.hugin.com/Products_Services/Products/Commercial/Explorer) sold by Hugin Explorer You can also.

  Further, in this embodiment, for the sake of simplification, each accident factor has been described by focusing on two items, weather (whether it is rain) and speed (whether it exceeds the speed limit). However, actually, for example, various items such as snow, heavy rain, thunder, etc. can be set for the above-mentioned weather, and the probability for each item can be obtained.

  In addition to the weather and speed, it is also possible to arrange items such as traffic jam information, daily illuminance, etc. in FIG. 3 and calculate the accident occurrence probability for each item. In particular, regarding driver attributes, as described in the third embodiment, if a questionnaire for each driver is executed in advance and it is necessary to pay particular attention to precautions specific to each driver, for example, weather. In FIG. 6, when the weather flag is “1”, the accident probability threshold is set to 0.01%, for example. By so doing, it is possible to notify the driver of warning information even when (1) rain (A = 1) and the speed is less than the speed limit (B = 0).

  Finally, specific warning contents when the warning is notified to the driver based on the prediction of the accident occurrence probability will be described. First, as described above, when the vehicle has had an accident in the past or approaches a special danger area, the probability of occurrence of the accident is calculated. If the value exceeds the threshold, a warning is notified. However, only in this case, the driver does not know what to pay attention to.

  In this regard, as shown in FIG. 6 above, regarding the item corresponding to the accident occurrence probability, a warning can be issued to the driver regarding the item whose flag of the corresponding accident factor item is “1”. There is a way to do so. That is, for example, in FIG. 6, if the warning is issued when the weather is rain (A = 1) and the speed is equal to or higher than the speed limit (B = 1), the weather is rain (A = 1) and the speed is the speed limit. The above two items (B = 1) each have an accident factor item flag “1”, and a warning is issued to the driver regarding these two items.

Accident warning notification system diagram Accident warning notification system processing flow chart Driver I / F part-Bayesian network part processing sequence diagram Illustration of the independence of each accident factor Diagram showing an example of forming a Bayesian network Diagram showing a Bayesian network calculation example

Explanation of symbols

0101: Accident warning notification system.
0103: Driver interface unit assuming car navigation and the like.
0106: Designated as a special danger area by the police or the like.
0107: Holds information related to areas where traffic accidents are likely to occur frequently.
0108: In this example, the probability of an accident is calculated when “accident in the past” or “special alert area” is designated.
0109: Information on the position of the traveling vehicle is acquired by a GPS device or the like.
0110: Data on past traffic accidents is stored by location.
0111: Traffic accident management center handling accident data such as police.
0113: Holds past accident data and traveling vehicle data.
0116: A threshold set by the user himself.
0119: A warning for an accident is issued through a speaker or the like.
0202: Accident data stored by the traffic accident management center (the police, etc.).
[0203] Acquisition of vehicle information related to accident factors such as speed and weather.
0205: A threshold set by the user himself / herself.
0209: Special alert area designated by the police.
0302: Acquisition of accident information and traffic jam information by the VICS system.
0305: Reconstruct the Bayesian network and register its table information.

Claims (6)

  It consists of a traffic management center that manages accident data, a driver I / F unit that consists of a vehicle information acquisition unit and a warning notification unit, and an accident warning notification system that consists of a Bayesian network system. A Bayesian network system is a probability table for each accident factor for the occurrence of an accident based on past accident data in a certain dangerous place and the traveling data of each vehicle passing through that place. The Bayesian network forming unit that creates the information, and when the traveling vehicle approaches the dangerous place, acquires information on the vehicle and various road conditions, and the probability that the Bayesian network forming unit has formed By comparing with the table, the basis for calculating the probability of accident that the vehicle is facing is calculated. The driver I / F unit is composed of a part for obtaining vehicle information on each accident factor and notifying the driver of an accident warning, and from the past accident data in the data unit. The actual accident occurrence probability is calculated, and even when a plurality of accident factors are intertwined, the Bayesian network system more accurately calculates the accident occurrence probability, and the result is calculated by the driver I / I / F unit. Accident warning notification system characterized by notifying   The Bayesian network system has a function of notifying the driver of an accident warning when the accident occurrence probability calculated by the Bayesian network computing unit is equal to or higher than a preset threshold, and the threshold The accident warning notification system is characterized in that the driver can set himself / herself and thereby the driver can set the warning frequency.   The Bayesian network system obtains information about the position of a running vehicle, performs a calculation by the Bayesian network when the position is approaching a dangerous place, and if the position is not approaching, the position again. An accident warning notification system characterized in that it is possible to obtain information on the situation and determine whether it is approaching a dangerous place, and by this determination, it is possible to save the time and effort of formation and calculation by the Bayesian network more than necessary.   In the traveling vehicle, the Bayesian network calculation unit again forms a Bayesian network using the various vehicle data used in the Bayesian network calculation after the accident probability is calculated by the Bayesian network calculation unit. An accident warning notification system characterized by the formation of a new Bayesian network.   The Bayesian network system, when notifying the driver of an accident warning, notifies about the risk of the accident, and extracts the combination of each accident factor when the accident probability exceeds a threshold, and relates to each accident factor. An accident warning notification system having a function that makes it possible to clarify matters to be noticed by the driver by giving a warning to the driver.   The Bayesian network system conducts a questionnaire for the driver in advance, registers the precautions specific to each driver, and when the driver approaches a certain dangerous place, the pre-registered accident Accident warning notification, which has a function to lower the threshold of warning notification to the driver under the situation where the accident factor actually occurs, and can notify a warning peculiar to each driver. system.

Images