Area condition estimation method and learning method thereof, and area condition estimation device and learning device thereof
By employing machine learning to analyze probe car data, the method addresses the challenge of estimating roadside strip congestion, providing accurate navigation guidance for short-term parking.
Patent Information
- Application Number
- JP2020038176
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2020-03-05
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2040-03-05
AI Technical Summary
Existing navigation systems lack the ability to accurately estimate congestion situations on roadside strips, which are typically used for short-term parking, as methods designed for long-term parking lots are inadequate due to frequent vehicle movements and changing conditions.
A method utilizing machine learning to analyze the driving history patterns of probe cars in a model area, distinguishing between congested and non-congested roadside strip situations by collecting and processing vehicle data, including position, speed, and direction changes, to estimate the area situation.
Enables accurate estimation of roadside strip congestion, allowing navigation systems to provide real-time information on parking availability and ease of stopping near destinations, enhancing user experience.
Smart Images

Figure 0007577451000002 
Figure 0007577451000003 
Figure 0007577451000004
Abstract
Description
[Technical field]
[0001] The present invention relates to an area situation estimation method, a learning method thereof, and an area situation estimation device and learning device thereof. The area situation estimation method to which the present invention relates relates to a method of estimating the situation of a vehicle present in a desired area on a map applied to, for example, a navigation device for a vehicle, using AI technology. [Background technology]
[0002] A useful function applied to navigation devices is to estimate the status of vehicles in an area where a destination is located. One example of such area status estimation is estimating the vacancy status of a parking lot. The vacancy status of a parking lot is estimated from sensors installed in the parking lot, as well as from waiting time for parking and the length of the queue of vehicles waiting to park, and is reflected in the navigation device. On the other hand, for example, in front of a station, a driver may wish to park the vehicle on the side of a road near the station for a short period of time without having to park the vehicle in a parking lot. In the case of parking lots, it would be nice if navigation devices were equipped with a function to estimate the vacancy status (congestion status) of roadside strips in a given area. Please refer to Patent Documents 1 to 3 as prior art documents disclosing techniques related to the present invention. [Prior art documents] [Patent documents]
[0003] [Patent Document 1] JP 2016-095663 A [Patent Document 2] JP 2013-068521 A [Patent Document 3] JP 2007-149054 A DISCLOSURE OF THEINVENTION [Problem to be solved by the invention]
[0004] Since vehicles are only allowed to stop on the shoulder of the road for a short time, there is a lot of movement between vehicles trying to stop and vehicles trying to leave. Therefore, methods for estimating the congestion situation in parking lots that assume long periods of stopping (parking) cannot be applied as is. According to the inventor's investigations, no prior art documents were found that disclose or suggest a technique for estimating the congestion status of a roadside strip. [Means for solving the problem]
[0005] The inventors have conducted intensive research to solve the above-mentioned problem, and have come to the conclusion that in a model area, there may be a difference in the patterns of driving history leading up to the stopping of vehicles (probe cars) that stop at the same time on a congested roadside and vehicles that stop at the same time on a non-congested roadside. For example, during peak times, other vehicles are often parked at the desired shoulder location, and as a result, the vehicle exhibits a characteristic driving history pattern before stopping at the desired shoulder location. One characteristic driving history pattern is circling around the desired shoulder location. Here, circling refers to passing the desired shoulder location once, making a left or right turn, a U-turn, and / or reversing, and then approaching the desired shoulder location again. There are also cases where the vehicle is unable to stop at the desired shoulder location (destination) and is forced to stop the vehicle at a location away from the desired location.
[0006] On the other hand, in a non-congested situation, the driver of the vehicle can stop the vehicle at the desired shoulder position without any stress. In other words, the driving history pattern up to the stop does not cause the above-mentioned round trip, and the vehicle does not stop at the shoulder far from the destination. Based on the above findings, the present inventors have arrived at the present invention described below. In the above description, the congestion status of the roadside of a road in an area is taken as an example of the status of an area, but the present invention can be generalized as a method for estimating the status of an area. The estimation method of the present invention is suitable for cases where the status of vehicles in a specific area changes frequently. Examples of such cases include the estimation of the congestion status of the roadside as described above, as well as the movement of vehicles in an area hit by heavy rain.
[0007] The first aspect of the present invention is defined as follows: A method for estimating a condition of an area, comprising: Collecting a set of vehicle data for each of a plurality of probe cars present in a model area and a situation in the model area, and generating teacher data including the set of vehicle data and situation data defining the situation in the model area; inputting the teacher data into a situation estimation unit having a learning function, and causing the situation estimation unit to learn a relationship between the set of vehicle data and the situation data; A method for estimating an area situation, comprising inputting a set of vehicle data in an estimation target area in which the situation is unknown to a trained situation estimation unit, and estimating the situation of the estimation target area.
[0008] According to the estimation method of the first aspect thus defined, when estimating the situation of an area, vehicle data of multiple probe cars present in the area is used. Here, vehicle data refers to all data related to the vehicle stored or generated by the probe car, such as probe data and vehicle inner data. Probe data is composed exclusively of position information (x, y) indicating the driving history of the probe car and its acquisition time t. Vehicle inner data refers to data related to various conditions of the vehicle output from sensors, control units, and other devices equipped in the vehicle, and refers to data that can be transmitted over a network installed in the vehicle via various communication protocols including CAN (Controller Area Network).
[0009] In this invention, a model area is defined, and the vehicle data of each of the multiple probe cars present in the model area and the situation data defining the situation in the model area are used as teacher data. The teacher data is collected and input to a situation estimation unit. This situation estimation unit has a learning function (machine learning function such as deep learning). In other words, the learning mode of AI processing is executed. This learning mode allows learning of the relationship between the set of vehicle data and the situation data that specifies the situation of the model area. For example, a set of travel trajectories of a probe car parked on a side strip in the model area during an arbitrarily specified time period can be used as the set of vehicle data. "Congestion data" and "non-congestion data" can be used as situation data that specifies the situation of the model area, for example, whether the side strip of a road in the model area is congested or not congested. In this case, whether the situation is congested or not is judged by a person based on an inspection of the current situation of the model area during a specified time period or on a photographed image thereof. Also, the photographed image can be processed and the judgment can be made by a computer device.
[0010] In the operation mode, a set of vehicle data in the estimation target area, whose situation is unknown, is input to the learned situation estimation unit. The situation estimation unit outputs situation data that specifies the situation of the estimation target area based on the input set of vehicle data. The situation of the estimation target area is estimated from this situation data. At this time, multiple probe cars are selected in the estimation target area using the same rules as those for selecting multiple probe cars in the model area, and the vehicle data of these are input to the learned situation estimation unit.
[0011] The second aspect of the present invention is defined as follows: Collecting a set of vehicle data for each of a plurality of probe cars present in a model area and a situation in the model area, and generating teacher data including the set of vehicle data and situation data defining the situation in the model area; A learning method for estimating an area situation, comprising inputting the teacher data to a situation estimation unit having a learning function, and causing the unit to learn the relationship between the set of vehicle data and the situation data. A second aspect of the present invention defines a learning mode in the first aspect. By executing the learning method defined in the second aspect, a learned situation estimation unit is obtained.
[0012] The third aspect of the present invention is defined as follows: preparing a trained situation estimation unit which is input with training data consisting of a set of vehicle data of each of a plurality of probe cars present in a model area and situation data defining a situation in the model area, and which is made to train a relationship between the set of vehicle data and the situation data; A method for estimating an area situation, comprising inputting a set of vehicle data in an estimation target area in which the situation is unknown to the learned situation estimation unit, and estimating the situation of the estimation target area. By using the trained situation estimation unit defined in the second aspect of the present invention, it becomes possible to estimate the situation of an estimation target area where the situation is unknown.
[0013] The fourth aspect of the present invention is defined as follows: In a first aspect, the collection of the model area conditions is performed when the vehicle data is collected. According to the estimation method of the fourth aspect defined as above, the situation data for identifying the situation in the model area and the vehicle data are obtained in the same time period, so that the relationship between the two is close, and therefore, the situation can be estimated with high accuracy.
[0014] The fifth aspect of the present invention is defined as follows: In a first aspect, collection of the model area conditions is performed after collection of the vehicle data. According to the estimation method of the fifth aspect thus defined, it is possible to estimate the situation of the area that will occur thereafter based on the vehicle data. For example, by executing the estimation method of the present invention immediately before arriving at the destination, it is possible to estimate the situation of the destination at the time of arrival. Here, the time period during which vehicle data is collected at the destination is a predetermined time period going back from a predetermined time immediately before arriving at the destination (for example, the time when the situation estimation is started). Then, the situation of the destination at the time of arrival at the destination is estimated based on the vehicle data obtained during this time period.
[0015] The sixth aspect of the present invention is defined as follows: In a first aspect, the situation data is congestion data defining a congestion state of a side strip of a road in an area or non-congestion data defining a non-congestion state of the side strip of a road in an area, The vehicle data includes a driving history pattern of the probe car before the vehicle is stopped, and / or a distance between a destination and a stopping position. According to the estimation method defined in the sixth aspect thus defined, it is possible to estimate the congestion state of the shoulders of roads in an area.
[0016] The seventh aspect of the present invention is defined as follows: In the estimation method defined in a sixth aspect, the pattern of the travel history includes at least one of a speed change, a coordinate change, and a direction change. According to the estimation method defined in the seventh aspect thus defined, driving condition data that is easy to acquire is selected as the vehicle data, and therefore the method becomes easy to implement.
[0017] The eighth aspect of the present invention is defined as follows: An estimation device for estimating an area situation, comprising: a learning unit; and an operation unit, The learning unit is a training data generating unit that collects a set of vehicle data of each of a plurality of probe cars present in a model area and a situation of the model area, and generates training data including the set of vehicle data and situation data that defines the situation of the model area; a situation estimation unit having a learning function, the situation estimation unit receiving the teacher data and learning a relationship between the set of vehicle data and the situation data; The operation unit includes: An area situation estimation device that inputs a set of vehicle data in an estimation target area in which a situation is unknown to the learned situation estimation unit, and estimates the situation of the estimation target area. According to the device of the eighth aspect thus defined, the same effects as those of the first aspect can be obtained.
[0018] The ninth aspect of the present invention is defined as follows: a training data generating unit that collects a set of vehicle data of each of a plurality of probe cars present in a model area and a situation of the model area, and generates training data including the set of vehicle data and situation data that defines the situation of the model area; A learning device for an area situation estimation method, the learning device including a situation estimation unit having a learning function, receiving the teacher data and learning the relationship between the set of vehicle data and the situation data. According to the device of the ninth aspect thus defined, the same effects as those of the second aspect can be obtained.
[0019] The tenth aspect of the present invention is defined as follows: a trained situation estimation unit that receives training data including a set of vehicle data of each of a plurality of probe cars present in a model area and situation data that defines a situation in the model area, and trains the trained situation estimation unit to learn a relationship between the set of vehicle data and the situation data; An area situation estimation device that inputs a set of vehicle data in an estimation target area in which the situation is unknown to the learned situation estimation unit, and estimates the situation of the estimation target area. According to the device of the tenth aspect thus defined, the same effects as those of the third aspect can be obtained.
[0020] The eleventh aspect of the present invention is defined as follows: The device according to any one of the eighth to tenth aspects, wherein the collection of the model area conditions is executed when the vehicle data is collected. According to the device of the eleventh aspect thus defined, the same effects as those of the fourth aspect can be obtained.
[0021] The twelfth aspect of the present invention is defined as follows: The device according to any one of the eighth to tenth aspects, wherein the collection of the model area conditions is executed after the collection of the vehicle data. According to the device of the twelfth aspect thus defined, the same effects as those of the fifth aspect can be obtained.
[0022] The thirteenth aspect of the present invention is defined as follows: The situation data is congestion data defining a congestion state of a side strip of a road in an area or non-congestion data defining a non-congestion state of the side strip of a road in the area, The device according to any one of the eighth to tenth aspects, wherein the vehicle data includes a pattern of a driving history of the probe car before the probe car is stopped and / or a distance between a destination and a stopping position. The device according to the thirteenth aspect defined in this manner can provide the same effects as those of the sixth aspect. The device according to the eighth aspect defined in this manner can provide the same effects as those of the sixth aspect.
[0023] The third aspect of the present invention is defined as follows: The device according to a thirteenth aspect, wherein the driving history pattern includes at least one of a speed change, a coordinate change, and a direction change. According to the device of the fourteenth aspect thus defined, the same effects as those of the seventh aspect can be obtained. [Brief description of the drawings]
[0024] [Figure 1] FIG. 1 is a block diagram showing a teacher data creation device according to the present invention. [Diagram 2] FIG. 2 is a schematic diagram showing the state (congestion) of the sidewalk in the model area. [Diagram 3]FIG. 3 is a schematic diagram showing another state (non-congested state) of the side strip in the model area. [Figure 4] FIG. 4 is a schematic diagram showing another state (non-congested state) of the side strip in the model area. [Diagram 5] FIG. 5 is a block diagram of a computer device functioning as a situation estimation device (learning mode) of the present invention. [Figure 6] FIG. 6 is a block diagram showing the computer system when it is functioning as the situation estimation device (operation mode) of the present invention. [Figure 7] FIG. 7 is a flowchart showing the operation of the teacher data creation device according to the embodiment. [Figure 8] FIG. 8 is a flow chart showing the details of step 5 in the flow chart of FIG. [Figure 9] FIG. 9 is a flow chart showing the details of step 6 in the flow chart of FIG. [Figure 10] FIG. 10 is a flowchart showing a method for acquiring vehicle data in the estimation target area. [Figure 11] FIG. 11 is a flow chart showing the details of step 105 in the flow chart of FIG. [Figure 12] FIG. 12 is a flow chart showing the details of step 106 in the flow chart of FIG. [Figure 13] FIG. 13 shows the hardware configuration when a computer device is used as the teacher data creation device of FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0025] FIG. 1 is a block diagram showing a teacher data creation device 1 according to an embodiment of the present invention. This teacher data creation device 1 is composed of a vehicle data storage unit 2, a probe car extraction unit 3, a vehicle data set creation unit 4, an area condition input unit 5, a teacher data creation unit 6, an output unit 7 and a teacher data storage unit 8. Vehicle data collected from probe cars is stored in the vehicle data storage unit 2. The vehicle data includes probe data and vehicle inner data.
[0026] The probe car extraction unit 3 extracts probe cars parked in the shoulder strip of the road included in the model area MA shown by the dashed line in FIG. 2 (hereinafter, sometimes referred to as the "model road") during a specified time period. The model area MA can be selected arbitrarily, but for example, as shown in FIG. 2, a specified area centered on a station can be used as the model area MA. The time period is, for example, 10 minutes, and probe cars parked in the shoulder strip of the model road at any time T0 are extracted. When the number of extracted probe cars, i.e., the number of probe cars parked in the shoulder strip of the model road at time T0, is less than 10, the time is traced back in time up to a total of 10 times every minute, and probe cars parked other than the probe cars already extracted are extracted. When the total number of extracted probe cars reaches 10 while tracing back in time, extraction of further probe cars is stopped. Even if the total number of extracted probe cars is less than 10, the time traced back in time 10 minutes from time T0 (T 0-10 ) the extraction is stopped.
[0027] For each of the probe cars extracted in this manner, the vehicle data before the probe car stopped is read from the vehicle data storage unit 2. As this vehicle data, a driving history pattern that indicates whether or not the vehicle was circulating around the stop position before stopping can be adopted. As a driving history pattern, for example, a driving trajectory (coordinate changes) for five minutes before stopping is formed from probe data. If the driving trajectory overlaps with the stop position, the probe car is considered to have been circulating. In addition, for example, the frequency of speed changes and direction changes in the five minutes before stopping can also be data that suggests the probe car was circulating. For speed changes and direction changes, vehicle inner data other than probe data can also be used.
[0028] Situation data (congested data, non-congested data) that specifies whether the side strips of roads in the model area are congested or not congested during the time period when the probe car is extracted is manually input by a person via the area situation input unit 5. The judgment of congestedness or not can be made on-site, or it can be made from images taken of the street (road) in front of the station during a specified time period. In another embodiment of the present invention, the condition of the roadside strip in the time period following the time period in which the probe car was extracted is used. The teacher data creation unit 7 combines the congestion data or non-congestion data input via the area condition input unit 5 with a set of vehicle data of each probe car issued during the acquisition time period, to generate teacher data.
[0029] An example of training data is shown in Table 1. [Table 1]
[0030] In Table 1, when a probe car is stopped on the shoulder in the model area MA, a "1" is added to the "Stopped in area" column of the vehicle data. When such a probe car does not exist, a "0" is added to the corresponding column of the probe car after the probe car with the "1". In addition, other vehicle data of the probe car with the "0" added will also be added with a "0". To determine whether or not the probe car has made a detour, if the driving trajectory overlaps with the stopping position, a "1" is entered in the "Double Driving Trajectory" column, and if there is no overlap, a "0" is entered. Similarly, if the driving trajectory has more than five turns, a "1" is entered in the "Turns > 5" column, and if the number of turns is five or less, a "0" is entered.
[0031] In order to determine whether a road is congested or not based on the distance from the stop position to the station, if the probe car's navigation device sets the station as the destination, a "1" is entered in the "Station is destination" column, and if the station is not set as the destination, a "0" is entered. In addition, if the distance from the stop position to the station is 50m or more, a "1" is entered in the "More than 50m" column, and if it is less than 50m, a "0" is entered. If the distance from the stop position to the station is 50m or more even though the station is set as the destination, the road in front of the station is estimated to be congested.
[0032] The situation shown in Figure 2 is noted in Table 1 as "Crowded 1." The situation shown in Figure 3 is recorded in Table 1 as "non-congested 1." In this example, only three probe cars were parked during the specified time period, and it can be seen that the shoulder of the road in front of the station was always empty. Note that the data for the fourth and subsequent vehicles that were not extracted is 0. The situation shown in Figure 4 is shown in Table 1 as "non-congestion 2." Here, the vehicle data of probe cars 1 to 5 is the same for congestion 1 (Figure 2) and non-congestion 2 (Figure 4). However, probe cars 6 to 10 in non-congestion 2 are able to stop without making any rounds. This means that an empty space was found immediately after the previous stopped vehicle left the shoulder. In other words, there are always stopped vehicles on the road in front of the station, and at first glance it looks like a congested situation, but in reality vehicles are frequently entering and exiting, so vehicles arriving later can stop without any stress, which means that it is a non-congested situation.
[0033] For the model area shown in Figs. 2 to 4, a data set (teaching data) shown in Table 1 is acquired over a long period of time. When the number of training data reaches a number suitable for AI processing, this training data is input to a learning device 100 shown in FIG. This learning device 100 includes a teacher data storage unit 8 and a situation estimation unit 15. The teacher data storage unit 8 includes a vehicle data set storage area 8A and an area situation storage area 8B. When the teacher data stored in the teacher data storage unit 8 is input to the situation estimation unit 15, the situation estimation unit 15 performs machine learning based on an AI program, and as a result, the vehicle data set and the situation data are associated in a predetermined manner.
[0034] Vehicle data of probe cars is collected for the estimation target area where the situation is unknown, and is stored in the vehicle data set storage area 23A of the memory unit 23 of the estimation device 200. When such a set of vehicle data is input to the trained situation estimation unit 15, situation data corresponding to the set of vehicle data is output from the output unit 17 together with, for example, its probability. When the situation data is congestion data, it is estimated that the side strip of the road in the estimation target area is in a congested situation.
[0035] Next, the operation of the estimation device of the present invention will be described with reference to the flowcharts in FIG. <Learning mode> 7, a model area MA is specified on map data used by the navigation device in step 1. In step 2, roads (model roads) included in the model area MA where a vehicle can stop are extracted. In step 3, an arbitrary time T0 is specified. This allows the time period 10 minutes prior to time T0 (T -10~T0 ) is specified. In step 4, the situation of the model area MA during the time period specified in step 3 (whether it was congested or not) is identified, and situation data (congested data or not congested data) is created. The situation of the model area is identified by data input from the area situation input unit 5, and is converted into situation data by the teacher data creation unit. It is also possible to identify the situation in the model area MA in the time period following the time period specified in step 3 and create situation data based on that.
[0036] In step 5, probe cars parked on the side strip of the model road during the specified time period are extracted. More specifically, the operation shown in FIG. 8 is executed. That is, in step 51, the probe cars parked on the shoulder of the model road at the specified time T0 are extracted in order of proximity to the station. When the total number of extracted probe cars reaches 10 (step 52 = YES), the process proceeds to step 57 and the extracted probe cars are stored. When the total number of extracted probe cars is less than 10 (step 52 = NO), the process proceeds to step 53.
[0037] In step 53, probe cars that are parked on the shoulder of the model road one minute before the specified time T0 and have not yet been extracted are extracted in order of proximity to the station. When the total number of probe cars extracted in step 51 and step 53 reaches 10 (step 54 = YES), proceed to step 57. When the total number is less than 10, proceed to step 55, and thereafter, the same operation is repeated every minute up to 10 minutes (step 55 = NO). Even if the total number extracted does not reach 10 even after going back 10 minutes, proceed to step 57 (step 55 = YES).
[0038] 7, in step 6, the traveling history pattern of the probe car stored in step 57 is identified to identify the vehicle data. More specifically, the operation shown in FIG. 9 is executed. That is, the stop position S of the probe car No. 1 that was first extracted in step 61 X1,Y1 and stop time T S1 Identify the stop time T S1 The time period varies depending on the probe car. -10~T0 ) may be earlier than the time. In step 62, the data stored in the vehicle data storage unit 2 for the probe car No. 1 is referred to, and its stop time T S1 Create a driving track going back 5 minutes from the current time. The travel trajectory created in this way is the stopping position S X1,Y1 If the vehicle has passed through the intersection (step 63 = YES), a "1" is entered in the [overlapping driving trajectories] column of Table 1 as the vehicle data (step 631), and if not, a "0" is entered in the same column (step 632). In step 64, the created driving trajectory is referenced, and if the number of direction turns exceeds five (step 64 = YES), a "1" is entered in the [Direction Turns > 5] column of Table 1 as vehicle data (step 641), and if not, a "0" is entered in the same column (step 642).
[0039] In step 65, it is confirmed whether the navigation device of probe car No. 1 has set the station as the destination. If the station has been set as the destination (step 65 = YES), a "1" is entered in the [Station is destination] column of Table 1 as vehicle data (step 651); if not, a "0" is entered in the same column (step 652). In step 66, the probe car No. 1 stops at position S X1,Y1 The distance from to the station (central point) is calculated, and if the distance is 50 m or more (step 66 = YES), a "1" is entered in the [50 m or more] column in Table 1 as vehicle data (step 661), and if not, a "0" is entered in the same column (step 662). This process is carried out for all the extracted probe cars.
[0040] Returning to Fig. 7, it can be seen that a set of vehicle data (10 items) was created in step 6 above, and situation data (congested data, non-congested data) was created in step 4. The set of vehicle data and the situation data are combined to create the teacher data shown in Table 1 (step 7). This teacher data is temporarily stored in the teacher data storage unit 8 of the learning device 100 shown in Fig. 5. General-purpose pre-processing (e.g., padding the data or adding dummy data) is performed on the stored teacher data, and the data is input to the situation estimation unit 15 incorporating an AI program. The situation estimation unit 15 executes machine learning to learn the relationship between the set of vehicle data and the situation data.
[0041] <Operation mode> Next, a description will be given of an operation mode that uses the trained situation estimation unit 15. Here, the trained situation estimation unit 15 may be the one that has executed the learning, or may be a physically different situation estimation unit 15 in which a copy program of the trained program contents is installed. When driving a car to take a family member to a station, there is a case where the driver wants to know whether the car can easily stop on the side of the road in front of the station. In this example operation mode, the driver is asked whether the car can easily stop on the road near the nearest station when driving to the nearest station or when estimating the congestion status of the road. When the navigation device has the estimation function and the function is turned on with the nearest station as the destination, the congestion status of the road in front of the nearest station when the driver arrives at the nearest station is estimated and displayed on the navigation device.
[0042] In the flowchart of Fig. 10, when the destination is estimated in step 101, the estimation device 200 (see Fig. 6) sets an estimation target area around the destination. The setting method is not particularly limited, but may be a range of a predetermined distance (e.g., 100 m) from the destination centered on the destination, or a range of a road (primary intersecting road) that intersects with a road adjacent to the destination, or a road (secondary intersecting road) that further intersects with the intersecting road (primary intersecting road) that is a predetermined distance (e.g., 100 m) from the destination. In step 102, roads within the estimated target area where cars can stop are extracted.
[0043] In step 103, a time period up to 10 minutes before the current time when the estimation function was turned on is specified. In step 105, the data stored in the vehicle data storage unit 2 is referred to, and the current time T m From 10 minutes before (T -10~m ) to extract probe cars that were parked in the estimation target area. The rules for extraction are shown in the flowchart of Figure 11. The processing according to this flowchart is substantially the same as that according to the flowchart of Figure 8, so a description thereof will be omitted. In step 106, vehicle data is identified from the extracted driving history pattern of the probe car. The method of identification is shown in the flowchart of Fig. 12. The process according to this flowchart (the creation rules of Table 1) is substantially the same as that shown in the flowchart of Fig. 9, so a description thereof will be omitted.
[0044] A set of vehicle data is created in step 106, which is input to the learned situation estimation unit 15. Based on the input set of vehicle data, the learned situation estimation unit 15 outputs situation data, i.e., congestion data or non-congestion data, together with its probability. Based on the congestion data or non-congestion data, the driver can determine whether it will be easy or difficult to stop on the road in front of the nearest station, which is the destination. In this example, the situation at the destination at the time when the estimation function was turned on is estimated. Since the situation does not change suddenly whether it is congested or not, it is considered that the estimation result at a time before the scheduled arrival time at the destination can be effectively used. Of course, if the situation in the model area in the time period after the time period when the probe car was extracted is stored as situation data in the teacher data, an estimate can be made of the situation in the time period after the time when the estimation function was turned on, i.e., the time period when the vehicle arrives at the destination. When estimating the situation in a time period after the time period in which the probe car was extracted, it is preferable to set a time sequence for the vehicle data obtained from the probe car. In the example of Table 1, the probe car is re-extracted every minute, and the probe car extracted later is assigned the number (No.) after the probe car extracted earlier.
[0045] FIG. 13 shows the hardware configuration of the area condition estimation device. The calculation unit 300 includes a CPU 301, a ROM 303, and a RAM 305, and controls the entire system. It also functions as a probe car extraction unit 3, a vehicle data set creation unit 4, a teacher data creation unit 6, a situation estimation unit 15, and an estimation target identification unit 201. The ROM 303 is a non-volatile memory in which a control program for controlling the calculation unit 300 and the like are stored. The RAM 305 readably stores various setting values preset by a user via an input device 330 such as a keyboard, and provides a working area for the CPU 301. The input device 330 functions as an area situation input unit 5. The control program for controlling the calculation unit 300 may be stored not only in the ROM 303, but also in the RAM 305 or the first and second storage devices 340 and 350.
[0046] The first storage device 340 functions as the vehicle data storage unit 2. As the vehicle data storage unit 2, a server at a center that controls the probe cars can be used. The second storage device 350 functions as the memory unit 32 of the teacher data storage unit 8 . It is preferable that the first and second storage devices utilize a partial area of a memory device of the server system, such as a hard memory or a flash memory.
[0047] A portion of the RAM in the calculation unit can be used as a buffer memory for temporarily storing data. The output device 320 is a display and an audio output device, and the input device 330 is an audio input section, a touch panel keyboard, a mouse, etc. that are arranged over the display. Communication with the probe car is possible via the communication interface 360. The devices constituting the computer are connected via a system bus 370 .
[0048] The present invention is not limited to the above-mentioned embodiments, examples, and modified examples. Various modified aspects that do not deviate from the scope of the claims and that can be easily conceived by a person skilled in the art are also included in the present invention. [Explanation of symbols]
[0049] 1. Teacher data creation device 15 Situation Estimation Department 100 Learning Device 200 Estimation device MA Model Area
Claims
1. A method for estimating a condition of an area, comprising: Collect a set of vehicle data including probe data or vehicle inner data of each of a plurality of probe cars present in a model area and a situation of the model area, and generate teacher data including the set of vehicle data and situation data that defines the situation of the model area; inputting the teacher data into a situation estimation unit having a learning function, and causing the situation estimation unit to learn a relationship between the set of vehicle data and the situation data; A method for estimating an area situation, comprising: inputting a set of vehicle data in an arbitrary estimation target area, the arbitrary estimation target area being different from the model area, the situation of which is unknown, to a trained situation estimation unit; and estimating a situation of the estimation target area at the time of the input, The situation data is congestion data defining a congestion state of a side strip of a road in an area or non-congestion data defining a non-congestion state of the side strip of a road in the area, A method for estimating area conditions, wherein the vehicle data includes a driving history pattern of the probe car before stopping, and / or a distance between a destination and a stopping position.
2. The method of claim 1 , wherein the collection of the model area conditions is performed at the time of collection of the vehicle data.
3. The method of claim 1 , wherein the collection of the model area conditions is performed after the collection of the vehicle data.
4. The method of claim 1 , wherein the driving history pattern includes at least one of a speed change, a coordinate change, and a direction change.
5. An estimation device for estimating an area situation, comprising: a learning unit; and an operation unit, The learning unit is a training data generating unit that collects a set of vehicle data including probe data or vehicle inner data of each of a plurality of probe cars present in a model area and a situation of the model area, and generates training data including the set of vehicle data and situation data that defines the situation of the model area; a situation estimation unit having a learning function, the situation estimation unit receiving the teacher data and learning a relationship between the set of vehicle data and the situation data; The operation unit includes:
1. An area situation estimation device, comprising: an input unit that has already learned a set of vehicle data in an arbitrary estimation target area different from the model area, the situation of which is unknown; and an area situation estimation unit that estimates a situation of the estimation target area at the time of the input; The situation data is congestion data defining a congestion state of a side strip of a road in an area or non-congestion data defining a non-congestion state of the side strip of a road in the area, An area condition estimation device, wherein the vehicle data includes the driving history pattern of the probe car before stopping, and / or the distance between the destination and the stopping position.
Citation Information
Patent Citations
Method of estimating traffic situation using probe information, and system for estimating and providing traffic situation
JP2002251698A
Vehicular drive support system
JP2007149054A
System for estimating traffic jam status
JP2008250493A
Traffic information estimation system, estimation method, provision system and computer program
JP2010191614A
Device, computer program and method for estimating traffic information
JP2011113117A
