JP4860273B2 - Method and program for determining roadside status classification of links in digital road maps - Google Patents

Description

  The present invention relates to a technique for automatically determining a roadside state classification of a link in a digital road map (Digital Road Map = DRM) using a computer.

  In recent years, car navigation systems have been attached to many automobiles, and VICS that displays traffic jams, construction information, etc. is becoming available in many cities. Digital road maps are the basic data for these systems. Digital road maps in Japan are managed by the Japan Digital Road Map Association as a “National Digital Road Map Database”. Each regional development station, prefecture, designated city, road-related public corporation / public corporation Maintenance is performed every year based on information from road managers nationwide.

  The digital road map database is created using the Geographical Survey Institute 1/25000 topographical map as the base map, basic road data with a width of 5.5 m or more, narrow road data with a size of less than 3 to 5.5 m, background data for rivers, railways, etc. It is divided into four layers of facility data such as public facilities. The road network data is composed of nodes such as intersections and linear links defined by the nodes. (In this specification, a digital road network is also referred to as a network, and digital road map data is also referred to as network data.)

In addition to the car navigation system and the road information display system, the digital road network data is used for prediction of road traffic. When planning urban traffic policy, it is essential to predict the traffic volume of major roads in the city, and when building a new town or large-scale commercial facility, forecast the traffic volume in the surrounding area. It is necessary to take measures such as improving existing roads and constructing new roads.
This kind of road traffic volume prediction uses digital road network data to set inflow / outflow / branch of traffic (automobile) from each node, link flow rate between nodes, average flow velocity, etc. Perform by calculation.

  In order to execute a simulation for predicting road traffic volume, link performance functions must be given for all links in the target network. Commonly used link performance functions include a QV equation and a BPR function (see Non-Patent Document 1). Regardless of which function is used, it is necessary to determine the traffic capacity for each link. It is desirable to determine the traffic capacity according to the road condition classification. For this purpose, it is necessary to determine the road condition classification for each link.

The roadside situation is a parameter for correcting the influence of friction caused by potential interference caused by roadside reasons such as the inflow of vehicles from the alley and the possibility of pedestrians and bicycles jumping out on the traffic capacity. Roadside conditions are generally divided into flat, mountainous, and urban areas.
Although roadside status classification is often used for traffic volume calculation as described above, its use is not limited to traffic volume calculation, and it can be used for various purposes for traffic prediction and management. Conceivable.

  The task of assigning roadside status classification to each link is currently performed manually. That is, for each link in the target network, the corresponding road data is searched from the data of the road traffic census (national road traffic situation survey conducted mainly by the Ministry of Land, Infrastructure, Transport and Tourism, etc.) The description is being transferred. (However, since road traffic census data covers only major roads, it is necessary for the user to determine roadside status classification for other roads.) This is also done by the method of judging the situation category.

  However, it takes a lot of time and labor to manually add roadside status classification to links in this way. For example, the digital road map of the Kinki region is composed of more than 60,000 links, excluding intercity highways and city highways, and the amount of work is huge. In addition, this work is not completed only once, but it must be constantly updated to the latest in order to reflect the road environment that changes due to the opening of new roads, conversion to residential land, and the like.

Japan Society of Civil Engineers Civil Engineering Planning Research Committee, Traffic Demand Prediction Technology Subcommittee, Road Traffic Demand Prediction Theory and Application (Part 1) Issued on August 31, 2003

  The present invention has been made for the purpose of significantly reducing the labor required for the task of assigning roadside status classification to the links of the digital road map as described above. According to the roadside situation classification determining method of the link according to the present invention, the roadside situation classification that had to be given manually can be automatically determined and assigned with high accuracy.

In order to solve the above-mentioned problem, the roadside situation classification determination method of the link according to the present invention is as follows.
A method for determining roadside status classification of links in a digital road map,
a) Created in advance and stored in the storage means "If [Condition A determined based on the link itself or / and the surrounding features of the link], the roadside status classification is [T] or not [T]") The reading means reads a plurality of rule formulas in accordance with the template and a plurality of learning links to which roadside situation classifications are assigned in advance,
Based on b) said reading means read from said memory means a plurality of rule expression and a plurality of learning links, application calculating means, themselves and / or their surroundings of a feature from the said plurality of learning links and, calculating a reliability of the respective rule expression based on the number of learning links a number of false learning link that is true,
c) The judging means applies the above rule formulas in descending order of reliability until the road condition classification is determined to be one for the link of the digital road map to be judged, thereby determining the road condition classification for the link. Steps,
d) an output means for outputting a result obtained based on the determination;
And wherein the Rukoto to have a.

Also ,
e) It is preferable that the providing unit further includes a step of assigning the roadside status classification determined by the determining unit to the link.

  By using the link road condition classification determination method according to the present invention, it is possible to manually assign the road condition classification to a certain number of learning links, and all of them included in the target digital road map data. It is possible to automatically determine and assign the roadside status classification of the link. Therefore, the labor of the work that had to be done manually for all links in the prior art is greatly simplified.

  Further, according to the present invention, it is possible to determine and assign roadside status classifications using only the data of the digital road map itself, so that other data such as road traffic census that has been used for reference is unnecessary. Therefore, work efficiency is improved.

  In addition, since the data can be easily updated in this way, there is also an advantage that it is easy to keep the network data up-to-date, which has conventionally had to be updated at a low frequency.

  Furthermore, since the labor of the operation is simplified in this way, there is an advantage that it becomes easier to obtain a road condition classification of a finer road network than before.

  FIG. 1 is a block diagram showing a main configuration of a computer system 1 according to the present embodiment. The computer system 1 includes a CPU 2, a memory 3, a display unit 4 including a CRT (Cathode Ray Tube) display and a liquid crystal display, a keyboard having character / number input keys and various function keys, and a mouse as a pointing device. The input unit 5 and the storage unit 6 are connected to each other. The storage unit 6 includes a storage medium including a magnetic recording medium, an optical recording medium, a semiconductor memory, and the like in which programs and data are stored in advance. This storage medium may be fixed to the storage unit 6 or may be detachable. In addition, the storage unit 6 includes an OS (Operating System) 6a, a roadside state classification determination program 6b, rule data 6c (described later), learning link data 6d (described later), and the like.

In the present invention, the digital road network data preferably uses the above-mentioned national digital road map database. In the national digital road map database, information such as link length and characteristics is already embedded for each link.
In the following embodiments, roadside status classifications of links of digital road maps are divided into three categories of flat land, mountainous area, and urban area according to the custom, but the classification method of roadside status classification is arbitrary.

FIG. 3 is a diagram showing a network in each region corresponding to the three categories of roadside conditions. As is clear from FIG. 3, the roadside conditions in the flat, mountainous, and urban areas have distinct visual characteristics.
Flatland: The network is sparse. The link is long and straight.
Yamaji: The network is sparse. The link is long and twists and turns.
Urban area ... The network is dense. The link is short and straight.

In the present invention, in order to allow the computer to grasp the visual (morphological) characteristics as described above with respect to the link to be determined and assigned to the roadside situation classification, the human can naturally determine the roadside situation classification appropriately. The machine learning method which is an effective method for realizing the same function as the learning ability being performed on the computer is adopted.
In the machine learning method, first, a certain number of sample data sets are analyzed, and useful rules are extracted from a number of rules created in advance. Various methods can be considered as a method for extracting a useful rule. In the present invention, a method for obtaining a rule with high reliability by paying attention to data features is used.
Next, the roadside situation is automatically determined and assigned by applying the rules thus obtained to the target link in descending order of reliability.

<Preparation stage: creation of regular expressions>
Hereinafter, the creation of a rule formula performed by the user as a pre-stage for executing the roadside road status classification determination program will be described. The user
“If [Condition A is determined based on the link itself or / and the surrounding features of the link], the roadside status classification is [T].”
Create multiple rule expressions according to the template of the structure.

Only [Condition 1] may be applied to [Condition A determined based on the link itself or / and the surrounding features of the link], or multiple conditions such as [Condition 1] and [Condition 2]. May be combined.
The roadside status category “T” applies to flat, mountainous or urban areas. At this stage, the reliability of the created rule expression is not particularly limited.
Specific template expressions and rule expressions will be described later.

《Link features》
The feature of the link will be described. An arbitrary item may be set for the feature of the link, but it is desirable to select an item that is considered effective for classifying the roadside situation, and the feature of the link itself or the feature around the link may be used. The features can be divided into quantitative features that can be expressed numerically and qualitative features that express characteristics. The qualitative features include “road type” having a value such as a general national road or a general city road of a designated city.
Features that are difficult to express appropriately, such as “bending condition” described later, can be quantitatively expressed by quantifying as necessary.

  For example, items such as “link extension (length)”, “bending condition”, “road type”, and “number of lanes” can be used as the features of the link itself. Each of these features has characteristics that are considered to be effective in classifying roadside conditions as described below.

“Link extension”: short… urban, long… flat or mountainous.
“Road type”: General city roads in designated cities… Urban areas can be expected.
“Number of lanes”: More than 4 lanes… Not a mountainous area.
For these items, data embedded in each link of the digital road map database from the beginning can be used as it is.

“Bending”: For quantification, the following equation can be used as an example.
“Bend” = (Sum of angles at the interpolation point -180) / Link length (However, if the numerator value is negative, “Bend” = 0)
An interpolation point is a point given to a link in order to express a bend. Here, by subtracting 180 from the sum of angles, a link that bends at a right angle around a residential land in an urban area can be regarded as a straight line.
Large bends… mountains, small… flats or urban areas can be expected.

  As the feature around the link, “the number of nodes within a predetermined range from the link” or “the number of interpolation points per link within the predetermined range from the link” can be used. As the target range, for example, values such as 500 m, 1000 m, and 1500 m can be set as appropriate. A plurality of range values may be set. The center point of the predetermined range may be an intermediate point between both ends of the link.

  “Interpolation points per link within a specified range from a link” is counted as a single link if the number of links included in the specified range is only partially included. The interpolation points count the number included in the range. By dividing the number of interpolation points by the number of links and using the value of the number of interpolation points per link, it is possible to distinguish urban areas and mountainous areas.

"template"
In this embodiment, the following two templates are used.
・ "If [Condition 1] is satisfied, the roadside situation is" T ". "
Example: “If [number of interpolation points per link within 500m] ≧ [18], the roadside situation is [flat ground].”
・ "If [Condition 1] and [Condition 2] are satisfied, the roadside condition is" T ".""
Example: “If [number of interpolation points per link within 500m] <[1] and [extension] <[393], the roadside situation is [mountainland].”
Here, when [Condition 1] and [Condition 2] are quantitative features, an expression satisfying [Feature] ≧ [Value] or [Feature] <[Value] is entered. However, in [Value], several values are appropriately set between the maximum value and the minimum value of the data.
When the condition is a qualitative feature, an expression satisfying [feature] = [attribute value] is entered.
Example: “If [road type] = [general city street of the designated city], the roadside status is [urban area].”

<Rules>
By using the features and templates set as described above, a plurality of rule expressions as shown in FIG. 4 are created. However, only a part of the equations is displayed in FIG. The plurality of rule expressions obtained in this way are stored in the storage unit 6 as rule expression data 6c.
At this stage, many rules whose reliability is not yet known are extracted. Some of the rules are useful for determining roadside status categories, but others are not.

  Hereinafter, the roadside situation classification determination process of the link executed by the CPU 2 executing the link roadside situation classification determination program according to the present invention will be described with reference to the flowchart shown in FIG.

<Calculation of reliability of regular expression>
First, when the user inputs a predetermined instruction command through the input unit 5, the CPU 2 calculates the reliability of each rule formula based on the rule formula data 6c and the learning link data 6d stored in the storage unit 6 ( Step S1).

The learning link is a link to which roadside status classification is given in advance in a digital road map. The learning link is preferably created based on the link included in the network data to be determined for the roadside situation classification, but is not particularly limited.
In this embodiment, 750 out of the links constituting the digital road map of the Kinki region are extracted at random, and roadside status classification is assigned to each link manually. In the present invention, this is referred to as “learning link data”. It can be said that the learning link data is a digital road map including links to which roadside status classifications are assigned and map information for determining the features around each link.

Any method may be used as the method of calculating the reliability, but in this example, the likelihood obtained by the following equation is used.
Reliability = log (([Number of data to be true] + a) / ([Number of data to be false] + a))
(A is a parameter that prevents division by 0 and increases the reliability as the number of data increases. A is determined empirically, but it is usually appropriate to set a value of about 0.1 to 0.5. .)

However, when the reliability is positive, the rule expression predicate is “is”, and when the reliability is negative, the absolute value is taken, and the rule expression predicate is “not”.
For example, there is a rule expression "If [road type] = [general city road in the designated city], the roadside state is [urban area].", And the road type is actually a general city road in the designated city. If there are 100 links, 98 of which are urban and the other 2 are not urban, if a = 0.25,
Reliability = log ((98 + 0.25) / (2 + 0.25)) = 3.77659
It becomes. Conversely, if two are urban and the remaining 98 are not urban, the confidence is
Reliability = log ((2 + 0.25) / (98 + 0.25)) =-3.77659
It becomes. In other words, if the absolute value of the obtained reliability is taken and the rule expression predicate “is” is “not”, the rule expression with reliability 3.76759 “if [road type] = [general city road of the designated city] ] Then the roadside situation is not [urban].
Is obtained.

  After calculating the reliability of each rule expression using the learning link itself or / and the surrounding features of the link as described above, the CPU 2 arranges the rule expressions in descending order of reliability (step S2). As a result, a regular expression sequence as shown in FIG. 5 was obtained. FIG. 5 shows only four regular expressions.

<Linkage roadside status classification, grant>
Next, the CPU 2 applies the rule formulas for which the reliability is calculated in descending order of reliability to the road status classification of the target link by applying it to the link of the target digital road map until the road status classification is determined to be one. Is determined (step S3). That is, a rule formula having no effect on the determination of the roadside situation classification is ignored, and a rule formula having the next highest reliability is applied. When the roadside situation classification is finally narrowed down to one, the determination process is terminated.
In addition, if a plurality of rule expressions contradicting each other are applied in this process, only the rule expression with the higher reliability is applied, and the rule expression with the lower reliability is ignored.

For example, let us consider a case where the feature of a certain link for which roadside situation classification is to be determined is as follows.
Number of interpolation points per link within 1000m ... 1
Number of interpolation points per link within 1500m ... 1
Number of nodes within 1500m… 95
Number of lanes… 2

The rules are applied to the links having the above features in descending order of reliability.
Rule 1: “If [number of interpolation points per link within 1000m] <[2], it is not [mountain area]”
This rule eliminates the possibility of mountainous areas, and the possibility of roadside situation classification is flat or urban.
Rule 2: “If [number of interpolation points per link within 1500m] <[2], it is not [mountain area]”
This regular expression corresponds to the feature of the link, but has no effect because it has already been determined that it is not a mountainous area. Consider the following rule:
Rule 3: "If [number of nodes within 1500m] ≥ [521], then [urban area]"
Since this rule does not correspond to the feature of the link, it is ignored.
Rule 4: “If [number of nodes within 1500m] <[96] and [number of lanes] = [2], it is not [urban area]”
This rule formula eliminates the possibility of urban areas, and finally determines the roadside status classification of this link as flat.

  It should be noted that the above-described rule expressions 1 to 4 are taken for explanation, and in fact, until the rule expression 4 in which the roadside situation classification is finally determined is considered (as a result, it is processed as invalid. Multiple rule expressions are considered).

  Next, the CPU 2 assigns a roadside situation category to the link based on the result of the roadside situation category judgment performed in step S3 (step S4). Here, the CPU 2 displays the result of roadside situation classification determination on the display unit 4 without assigning the roadside situation classification, and the user who sees the determination result gives the roadside situation classification to the link using the input unit 5. It may be configured to input a grant command that is an instruction of whether or not.

Further, the CPU 2 repeats the processes of steps S 3 and S 4 until roadside status classification is assigned to all links included in the target network (step S5).

<Accuracy verification>
In order to verify the accuracy of the method for determining the roadside situation classification of the link according to the present invention, a test set consisting of 250 links randomly selected from the links constituting the digital road map of the Kinki region is used, and determined by the method of the present invention. The roadside situation classification was compared with the roadside situation classification judged manually. The case where both results matched was regarded as success, and the case where they did not match was regarded as failure. The results are as follows, and it was found that the roadside situation classification can be determined with considerably high accuracy by the link roadside situation classification judging method of the present invention.
Total: 250 (100.0%)
Success: 208 (83.2%)
Failure: 27 (10.8%)
Within tolerance: 15 (6.0%)
Note that “within tolerance” means that the output of this method does not coincide with manual work, but it is difficult to clearly distinguish roadside conditions in regions, so it cannot be asserted that the output of this method is incorrect. Refers to the case.

In the above test, the causes determined to be within the allowable range or failure were as follows.
・ Links are located at the borders such as in the corners of cities ・ The villages are not so big that they cannot be called cities ・ Locations are mountainous but linear

The method for determining the roadside state classification of the link in the digital road map according to the present invention has been described above. However, the above is only an example, and it is obvious that modifications and improvements may be appropriately made within the spirit of the present invention. .
Hereinafter, some modifications will be described.

<Modification 1>
If the reliability of the rule expression has already been calculated, the result of step S1 can be omitted by storing the result in the storage unit 6. In this case, when executing the roadside situation classification determination program, the CPU 2 performs the processing after step S2 based on the user inputting an instruction to use the reliability data already calculated. This can simplify the work when updating network data having the same geographical location.

<Modification 2>
After rearranging the rule expressions in descending order of reliability, a process for eliminating redundant rule expressions may be performed. This makes it possible to increase the efficiency of the determination process. Redundant rules are related to the magnitude of values and those related to logical products.
(Redundancy with respect to magnitude)
For example, if the rule expressions are arranged in the following order, the second rule expression is redundant and can be deleted.
1. If [the number of nodes within 500m] ≧ [128], the roadside situation is [urban area].
2. If [number of nodes within 500m] ≧ [129], the roadside situation is [urban area].
(Redundancy related to logical AND)
For example, if the rule expressions are arranged in the following order, the second rule expression is redundant and can be deleted.
1. If [Condition x], the roadside situation is [Plane].
2. If [Condition x] and [Condition y], the roadside situation is [Flatland].

<Modification 3>
When the roadside situation classification determination of the link is performed, the application may be terminated with a rule expression up to a certain reliability. In this case, the link for which the roadside situation classification cannot be determined is determined manually by the user. By doing so, since it is possible to manually perform links that are difficult to make an automatic determination clearly, it is possible to improve efficiency while suppressing deterioration in accuracy of the determination results.

<Modification 4>
In order to improve the accuracy of roadside classification judgment, add some classifications to the roadside situation classification that you want to judge finally, perform judgment processing once, and then classify it into the final roadside situation classification You can also.
For example, in addition to “plain”, “mountain”, and “city”, “mountain village” and “urban coast” are added, and then “mountain” is changed to “mountain”. “Urban coast” can be converted back to “Urban”.

The block diagram which shows the principal part structure of the computer system by one Embodiment of this invention. The flowchart which shows the flow of roadside condition classification provision of the link of this invention. The figure which shows the network of each area corresponding to three divisions of a roadside situation. An example of a plurality of regular expressions created using a template. An example of a rule expression arranged in the descending order of reliability.

Explanation of symbols

1 ... computer system 2 ... CPU
3 ... Memory 4 ... Display unit 5 ... Input unit 6 ... Storage unit

Claims (5)

A method for determining roadside status classification of links in a digital road map,
a) Created in advance and stored in the storage means "If [Condition A determined based on the link itself or / and the surrounding features of the link], the roadside status classification is [T] or not [T]") The reading means reads a plurality of rule formulas in accordance with the template and a plurality of learning links to which roadside situation classifications are assigned in advance,
b) Based on a plurality of rule expressions and a plurality of learning links read by the reading means from the storage means, the calculating means applies themselves or / and their surrounding features from among the plurality of learning links. And calculating the reliability of each rule expression based on the number of learning links that are true and the number of learning links that are false.
c) The judging means applies the above rule formulas in descending order of reliability until the road condition classification is determined to be one for the link of the digital road map to be judged, thereby determining the road condition classification for the link. Steps,
d) an output means for outputting a result obtained based on the determination;
A roadside situation classification determination method for a link in a digital road map. e) applying means, roadside situation category determination process of the link in the digital road map according to claim 1, a roadside situation segment is more determined in the determination means, characterized by further comprising the step of applying to said link. As the feature of the link itself stored in the storage means, including all or part of the length of the link itself, the degree of bending, the road type, the number of lanes,
3. The roadside situation classification of the link in the digital road map according to claim 1, wherein the feature around the link includes the number of nodes within a predetermined range from the link or / and the number of interpolation points per link. Judgment method. A program for causing a computer to perform processing for determining a roadside situation classification of a link in a digital road map, the computer comprising:
a) A plurality of pre-created templates based on the precondition that “the condition of roadside status is [T] or not [T] if [condition A determined based on the link itself or / and the surrounding features of the link]” A storage means for storing each of the regular formulas and a plurality of learning links to which roadside situation classifications are assigned in advance,
b) Reading the regular expression and the learning link from the storage means, applying themselves or / and their surrounding features from among a plurality of learning links, and the number of learning links that are true and false Calculating means for calculating the reliability of each rule expression based on the number of learning links
c) Judgment means for determining the roadside situation classification for the link by applying the above rule formulas in descending order of reliability until the roadside situation classification is determined to be one for the link of the digital road map to be judged.
d) output means for outputting the result obtained based on the determination;
A program for determining roadside state classification of links in a digital road map, characterized by functioning as 5. The roadside situation classification determination program for a link in a digital road map according to claim 4, wherein the computer further gives a roadside situation classification determined by the determination means to the link, or the roadside determined by the determination means. A program for determining roadside road status classification in a digital road map, which functions as a granting unit for granting the roadside status classification based on an assignment command given by a user based on the result of the situation classification.

Images