KR102233723B1 - Driving pattern management server for autonomous vehicles and driving pattern classifying method thereof - Google Patents

Abstract

The present invention relates to a driving pattern management server for autonomous vehicles and a driving pattern analysis method thereof to correctly determine a dangerous driving pattern of the autonomous vehicle. According to one embodiment of the present invention, a driving pattern management server receiving driving data provided from autonomous vehicles operating on roads comprises: a communication unit receiving the driving data transmitted from the autonomous vehicles; a storage unit storing the received driving data and infrastructure information of the roads; and a control unit extracting the accumulated driving data as a driving pattern for each road section through machine learning and analyzing the driving pattern in connection with the infrastructure information of the roads. The control unit controls the communication unit and the storage unit to accumulate the driving data in the storage unit for a specific period, classifies the accumulated driving data for each section of the roads, forms the divided driving data into a data set in a form suitable for the machine learning, extracts the driving pattern by performing clustering through a self-organizing map (SOM) algorithm and K-means++ algorithm used for the data set, and analyzes driving characteristics of the autonomous vehicle for each section of the roads by linking the driving pattern and the road infrastructure information.

Description

Operation pattern management server of autonomous vehicle and its operation pattern analysis method {DRIVING PATTERN MANAGEMENT SERVER FOR AUTONOMOUS VEHICLES AND DRIVING PATTERN CLASSIFYING METHOD THEREOF}

The present invention relates to a vehicle driving pattern classification system, and more particularly, to a driving pattern management server and a driving pattern analysis method for deriving and analyzing a driving pattern using driving information of an autonomous vehicle on a real road.

Conventional studies on vehicle driving patterns have focused on classifying dangerous driving according to predefined traffic standards by analyzing vehicle driving patterns. For example, it is the degree to which the driving pattern of the vehicle is analyzed based on variables such as speed, acceleration, and rotation, and the result is used to determine whether the driving pattern of the vehicle corresponds to safe driving.

However, there is a problem in that the analysis of the driving pattern does not reflect the actual road environment on which the vehicle is running. That is, in the conventional driving pattern analysis, the actual road characteristics such as facilities, geometry, and environment of the road on which the vehicle is running are not reflected at all. The conventional analysis technique for the driving pattern does not reflect the actual road environment that has a great influence on the safe driving of the vehicle. With this analysis method, there is a limitation in generating autonomous data of level 4 or higher safety driving standards for autonomous vehicles that require high stability.

An object of the present invention is to provide an apparatus and method for generating and analyzing a driving pattern of an autonomous vehicle using driving data acquired in a real road environment.

A driving pattern management server receiving driving data provided from an autonomous driving vehicle according to an embodiment of the present invention includes a communication unit receiving the driving data transmitted from the autonomous driving vehicle, the received driving data, and road infrastructure information. A storage unit for storing, and a control unit for deriving the accumulated driving data as a driving pattern for each road section through machine learning, and analyzing the driving pattern in connection with the infrastructure information, wherein the control unit includes the driving data The communication unit and the storage unit are controlled to accumulate in the storage unit for a specific period, the accumulated driving data is divided by section of the road, and the classified driving data is configured as a data set in a form suitable for machine learning, and the A self-organizing map (SOM) algorithm and a K-means++ algorithm for a data set are used to derive a driving pattern by clustering, and by linking the driving pattern and the road infrastructure information, the autonomous vehicle operates by section of the road. Interpret the characteristics.

In this embodiment, the driving data includes information on the speed, acceleration, rotation rate, and altitude values of the autonomous vehicle, and a wheel brake signal and a turn signal signal of the autonomous vehicle.

In this embodiment, the controller configures the data set by processing the wheel brake signal, the turn signal signal, and the speed, acceleration, rotation rate, and altitude values with a Piecewise Aggregate Approximation (PAA) algorithm.

In this embodiment, the control unit converts the wheel brake signal and the turn signal signal corresponding to the categorical data into numeric data through One-Hot Encoding and then approximates the partial aggregation (Piecewise). Aggregate Approximation: PAA) algorithm.

A method of analyzing a driving pattern of a driving pattern management server for receiving driving data of an autonomous driving vehicle according to an embodiment of the present invention includes the steps of collecting the driving data transmitted from the autonomous driving vehicle for a specific period, and the collected driving Dividing the data for each section of the road, and configuring the divided driving data into a data set for machine learning, applying a self-organizing map (SOM) algorithm and a K-means++ algorithm to the data set to determine a driving pattern And analyzing the driving characteristics of the autonomous vehicle for each section of the road by linking the driving pattern with the infrastructure information of the road.

According to the above-described system and method for generating a driving pattern of an autonomous vehicle according to an exemplary embodiment of the present invention, it is possible to precisely generate a driving pattern of the autonomous driving vehicle according to an actual road. Therefore, by using the driving pattern of the present invention, it is possible to more accurately determine the dangerous driving pattern of the autonomous vehicle, and to prevent accidents and to quickly respond to abnormal driving based on a representative driving pattern.

1 is a block diagram illustrating a system for obtaining and analyzing driving data of an autonomous vehicle according to an exemplary embodiment of the present invention.
2 is a block diagram illustrating an exemplary configuration of a vehicle module of the present invention.
FIG. 3 is a block diagram illustrating a driving pattern management server illustrated in FIG. 1 by way of example.
4 is a flowchart briefly showing a driving pattern analysis method performed by the driving pattern management unit of FIG. 3.
5 is a flow chart showing in more detail the driving pattern analysis method of FIG. 4.
6 is a diagram illustrating a part of a road map for collecting driving data of an autonomous vehicle according to an embodiment of the present invention.
7 is a block diagram illustrating a procedure for collecting and configuring driving data according to an embodiment of the present invention.
8 is a graph briefly showing an example of the partial aggregation approximation (PAA) algorithm of the present invention mentioned in FIG. 7.
9 is a diagram illustrating a method of configuring a data set for categorical data by way of example.
10 is a diagram illustrating a hybrid clustering operation corresponding to the operation pattern derivation step of FIGS. 4 to 5 by way of example.
11 is a diagram illustrating a mapping relationship between road sections on an actual road and a cluster by way of example.
12 is a table showing an example of clustered clusters and analysis results.

Hereinafter, some embodiments of the present invention will be described in detail with reference to exemplary drawings. In adding reference numerals to elements of each drawing, the same elements may have the same numerals as possible even if they are indicated on different drawings. In addition, in describing the present invention, when it is determined that a detailed description of a related known configuration or function may obscure the subject matter of the present invention, the detailed description may be omitted.

1 is a block diagram illustrating a system for obtaining and analyzing driving data of an autonomous vehicle according to an exemplary embodiment of the present invention. Referring to FIG. 1, a system 10 for acquiring driving data of an autonomous vehicle includes a vehicle module 100, a communication network 200, and a driving pattern management server 300.

The vehicle module 100 is installed in an autonomous vehicle. The autonomous driving vehicle may be a vehicle equipped with an autonomous driving system in various types of vehicles capable of driving on real roads for acquiring driving data. The vehicle module 100 mounted on the autonomous vehicle generates driving data of the autonomous vehicle using various sensors.

For example, the vehicle module 100 may include a speed sensor, an acceleration sensor, a yaw rate sensor, and an altitude sensor of an autonomous vehicle. In addition, driving data from these sensors may be transmitted to the driving pattern management server 300 through the communication network 200. In addition, the vehicle module 100 may sample a signal of a speed wheel brake operation or a direction indicator of the autonomous vehicle and transmit it to the driving pattern management server 300 as driving data.

The communication network 200 provides a communication channel between the vehicle module 100 and the driving pattern management server 300. The communication network 200 refers to a wireless or wired communication structure for exchanging information between respective nodes such as the vehicle module 100 or the driving pattern management server 300. For example, the communication network 200 may include vehicle to everything communication (V2X) for exchanging information with objects such as other vehicles, mobile devices, and roads. Alternatively, the communication network 200 may include 3rd Generation Artnership Project (3GPP), Long Term Evolution (LTE), World Interoperability for Microwave Access (WIMAX), Wi-Fi, 3G, 4G, 5G, 6G, etc. However, the present invention is not limited thereto.

The driving pattern management server 300 generates and interprets a driving pattern of an autonomous vehicle, which is a core function of the present invention. That is, the driving pattern management server 300 collects driving data transmitted when driving on a real road from an autonomous vehicle. The driving pattern management server 300 configures the collected driving data into data in a form for applying a machine learning algorithm. The driving pattern management server 300 generates a driving pattern for each road section by using unsupervised learning on the configured data.

In addition, the driving pattern management server 300 performs a function of interpreting the derived driving pattern in connection with a road infrastructure such as a geometric structure or facility of a road section. The driving pattern management server 300 may increase the accuracy of the driving pattern of the autonomous vehicle by using a causal relationship between the driving pattern for each section and the road infrastructure. Thereafter, the driving pattern management server 300 may be used to determine whether the autonomous vehicle operates normally or abnormally by using the analysis result. Alternatively, it may be used to supplement the driving algorithm of the autonomous vehicle by using the analysis result associated with the actual road of the driving pattern.

According to the driving pattern management server 300 described above, a driving pattern of an autonomous vehicle may be derived based on driving data generated when driving on an actual road. In addition, it can be used for road safety evaluation or rating research by analyzing the correlation between the actual road infrastructure and the driving pattern by processing the derived driving pattern. In addition, the analyzed driving pattern can be used for calculating insurance premiums for autonomous vehicles, or for controlling road infrastructure to minimize accident damage in the event of an abnormal operation of the autonomous vehicle.

2 is a block diagram illustrating an exemplary configuration of a vehicle module that can be mounted on an autonomous vehicle. Referring to FIG. 2, the vehicle module 100 may be mounted on an autonomous vehicle to generate driving data and transmit it to the driving pattern management server 300. The vehicle module 100 may include a sensor unit 110, a sensor hub 130, and a vehicle communication unit 150.

The sensor unit 110 senses information such as speed, acceleration, altitude, and rotation rate detected when the autonomous vehicle is operating. In addition, the sensor unit 110 may sense whether the wheel brake is operated and whether the direction indicator flashes. In order to sense the motion characteristics of the autonomous vehicle, the sensor unit 110 may include a speed sensor 111, an acceleration sensor 112, an altitude sensor 113, and a rotation rate sensor 114.

In addition, the sensor unit 110 may include a wheel brake motion sensor 115 and a direction indicator motion sensor 116. The wheel brake motion sensor 115 senses whether the wheel brake is operated for deceleration of the autonomous vehicle and generates an electric signal. In addition, the direction indicator motion sensor 116 may detect information on the blinking of the direction indicator and provide an electric signal. The sensor unit 110 may further include a radar or lidar sensor for recognizing objects or situations around the autonomous vehicle.

The sensor hub 130 receives and processes sensing data provided from a plurality of sensors 111 to 116 included in the sensor unit 110. The sensor hub 130 may periodically or aperiodically receive sensing data randomly transmitted from a plurality of sensors. The sensing data of each of the plurality of sensors 111 to 116 is collected by the sensor hub 130. The sensor hub 130 may convert sensing data transmitted from the sensor unit 110 into a data format for efficient transmission. The sensor hub 130 may be implemented using a processor or various computing cores.

The vehicle communication unit 150 transmits sensing data provided from the sensor hub 130 through the communication network 200. For example, the vehicle communication unit 150 may include a wired/wireless communication module supporting vehicle object communication (V2X).

The vehicle module 100 described above may be mounted integrally or modularly in an autonomous vehicle.

FIG. 3 is a block diagram illustrating a driving pattern management server illustrated in FIG. 1 by way of example. The driving pattern management server 300 (see FIG. 1) may derive a driving pattern of an autonomous vehicle by processing real-road driving-based driving data transmitted from the vehicle module 100 (see FIG. 1). In addition, the driving pattern management server 300 may analyze a causal relationship between the derived driving pattern and the road infrastructure. Referring to FIG. 3, the driving pattern management server 300 may include a communication unit 320, a storage unit 340, and a control unit 360.

The communication unit 320 includes a receiving unit 321 and a transmitting unit 323. The receiving unit 321 receives driving data transmitted from the vehicle module 100. The receiving unit 321 may change the driving data transmitted through the communication network 200 (see FIG. 1) into a data format processed by the controller 360. The receiving unit 321 will transmit the received driving data to the control unit 360. The transmission unit 323 may output a driving pattern derived from the driving pattern management server 300 or a causal relationship with the actual road infrastructure generated through analysis of the driving pattern.

The storage unit 340 may include a vehicle information DB 341, a driving data DB 343, a driving pattern DB 345, and a road information DB 347. The storage unit 340 may be composed of storages for storing data managed by the driving pattern management server 300.

The vehicle information DB 341 stores information on an autonomous vehicle. For example, when there are two or more autonomous vehicles transmitting driving data, information for identifying each vehicle may be stored in the vehicle information DB 341. In addition, vehicle characteristic information such as vehicle type, weight, and size of the autonomous vehicle may be stored in the vehicle information DB 341.

The driving data DB 343 stores driving data transmitted from the vehicle module 100 under the control of the driving data collection unit 363. The driving data may be a vast amount of big data transmitted over a long period of time from the vehicle module 100 to derive a driving pattern. The driving data DB 343 stores driving information of an autonomous vehicle in the form of raw data.

The driving pattern DB 345 stores the driving pattern generated by the driving pattern generating unit 367 of the control unit 360 from driving data and related information. The driving pattern DB 345 may store driving patterns for each section of a target road or for each identifier (ID). In addition, information generated through analysis of the driving pattern may be stored in the driving pattern DB 345.

The road information DB 347 may store an identifier (ID) for each section of a road in which driving data of an autonomous vehicle is generated or information on a road infrastructure. For example, road infrastructure may include road facilities or information that affect driving patterns such as lane centerlines, crosswalks, taxiways, overpasses, intersections, roundabouts, and increasing or decreasing lanes. The driving data or driving pattern will be managed based on the precision road map provided by the road information DB 347.

The control unit 360 collects the driving data transmitted through the communication unit 320, configures the collected driving data in a form that can be processed, generates a driving pattern, and interprets the generated driving pattern in connection with the road infrastructure. . To this end, the control unit 360 includes a processor 361, a driving data collection unit 363, a driving data configuration unit 365, a driving pattern generating unit 367, and a driving pattern analyzing unit 369. Here, preferably, the processor 361 may be configured with hardware, and the driving data collection unit 363, the driving data configuration unit 365, the driving pattern generation unit 367, and the driving pattern analysis unit 369 are software Can be provided as

The processor 361 may control the overall operation of the driving pattern management server 300. The processor 361 may patch driving data received through the communication unit 320 or transmit a control signal. The processor 361 may access the vehicle information DB 341, the driving pattern DB 343, and the road information DB 345 of the storage unit 340. The processor 361 may execute an algorithm or program command constituting the operation data collection unit 363, the operation data construction unit 365, the operation pattern generation unit 367, and the operation pattern analysis unit 369. The processor 361 may be implemented in the form of a system-on-chip (SoC), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or the like.

The driving data collection unit 363 stores driving data transmitted from the vehicle module 100 (refer to FIG. 2) of the autonomous vehicle in the storage unit 340. The driving data is data generated by driving the autonomous vehicle multiple times during a predetermined period for a target road section. Accordingly, during the period for collecting the driving data, the driving data may be accumulated in the driving data collection unit 363 for each period and for each road identifier (ID). Driving data can be managed in the form of big data over a specific period in order to derive meaningful driving patterns.

The driving data configuration unit 365 reconstructs the collected driving data in a form that is easy for machine learning. The driving data configuration unit 365 may analyze the configuration of the accumulated driving data and select variables that affect classification of the driving pattern. For example, the driving data configuration unit 365 may select only driving data related to speed, acceleration, rotation rate, altitude values, wheel brakes, and direction indicators that affect the driving pattern. The driving data configuration unit 365 classifies data for each vehicle of the autonomous vehicle and classifies it for each road identifier (ID). In addition, the driving data configuration unit 365 may delete data insufficient for pattern extraction. In particular, the operation data configuration unit 365 divides the data type into a numeric type and a categorical type and configures the data type in a form capable of machine learning. For example, the driving data configuration unit 365 converts categorical data into numeric data by using one-hot encoding in the case of categorical driving data. In addition, the driving data configuration unit 365 may configure the driving data using a Piecewise Aggregate Approximation (hereinafter, referred to as PAA) algorithm.

The driving pattern generation unit 367 generates a driving pattern for each road section by using unsupervised learning on the configured data. The driving pattern generation unit 367 clusters the driving data set generated through partial aggregation approximation (PAA) using a hybrid-type machine learning algorithm. For example, the driving pattern generation unit 367 processes the driving data set with a self-organizing map (SOM) algorithm, and additionally clusters the result values using the K-means++ algorithm. I can. This technique will be described in detail through the drawings to be described later.

The driving pattern analysis unit 369 analyzes the driving pattern for each cluster derived through machine learning. The driving pattern analysis unit 369 may select factors affecting the driving pattern from the geometry or facilities of road sections belonging to each cluster, the number of lanes, section length, road signs, traffic control equipment, and various facilities. That is, the driving pattern analysis unit 369 may estimate a causal relationship for each cluster and the infrastructure of the target road provided from the road information DB 347 of the storage unit 340.

The control unit 360 may refer to the detection of normal and abnormal driving of the autonomous vehicle and subsequent control operations by using the analysis result of the driving pattern provided from the driving pattern analysis unit 369. In addition, the controller 360 may update the driving algorithm of the autonomous vehicle using the analysis result. Accordingly, the driving pattern management server 300 of the present invention can support the driving of an autonomous vehicle with high reliability and stability assured on a real road.

4 is a flowchart briefly showing a driving pattern analysis method performed by the driving pattern management unit of FIG. 3. Referring to FIG. 4, the driving pattern management server 300 may accumulate driving data provided from an autonomous vehicle, and may derive and interpret a driving pattern from the accumulated driving data.

In step S110, the driving data collection unit 363 stores driving data transmitted from the vehicle module 100 (refer to FIG. 2) of the autonomous vehicle in the driving data DB 343 of the storage unit 340. The driving data may be accumulated for each vehicle, period, and road section and stored in the driving data DB 343.

In step S120, the driving data configuration unit 365 configures the collected driving data stored on the driving data DB 343 in a form for machine learning. The driving data configuration unit 365 classifies and processes the driving data according to the data type.

Driving data can be divided into numerical and categorical types. Numerical data can be processed with a partial aggregate approximation (PAA) algorithm, but categorical data requires conversion to a numeric type. Accordingly, the driving data configuration unit 365 converts categorical data such as wheel brakes or turn signal signals into numeric data using a One-Hot encoding technique.

In addition, the driving data configuration unit 365 may configure a data set for applying machine learning using a partial aggregation approximation (PAA) algorithm. The driving data configuration unit 365 applies Min-Max Normalization to all driving data. That is, the driving data may be normalized to values having a minimum of '0' and a maximum of '1'.

In step S130, the driving pattern generation unit 367 generates a driving pattern for each road section by using unsupervised learning on the configured data set. The driving pattern generation unit 367 clusters the data set generated through partial aggregation approximation (PAA) using a hybrid-type machine learning algorithm. The driving pattern generation unit 367 may process the data set with a self-organizing map (SOM) algorithm, and apply a K-means++ algorithm to the result value.

In step S140, the driving pattern analysis unit 369 analyzes the driving pattern for each cluster derived through machine learning. The driving pattern analysis unit 369 may select factors affecting the driving pattern from the geometry or facilities of road sections belonging to each cluster, the number of lanes, section length, road signs, traffic control equipment, and various facilities. That is, the driving pattern analysis unit 369 may estimate a causal relationship for each cluster and the infrastructure of the target road provided from the road information DB 347 of the storage unit 340.

In the above, the procedure of collecting and deriving and interpreting the driving pattern of the driving pattern management server 300 according to the exemplary embodiment of the present invention has been briefly described. Here, a technique for using the analysis result of the driving pattern by the driving pattern management server 300 has not been described in detail. However, it will be well understood that the analysis result of the driving pattern can be used as the basis data for the detection of abnormal driving of an autonomous vehicle, an update of an autonomous driving algorithm, and automobile insurance premiums.

5 is a flow chart showing in more detail the driving pattern analysis method of FIG. 4. Referring to FIG. 5, the driving pattern management server 300 generates a driving pattern using real road driving data provided from an autonomous vehicle, and performs an analysis such as a causal relationship between the driving pattern and the infrastructure of the road.

In step S210, the driving pattern management server 300 collects driving data of the autonomous vehicle. More specifically, the driving data collection unit 363 stores driving data transmitted from the autonomous vehicle in the driving data DB 343. The driving data is collected for a predetermined period and accumulated in the driving data DB 343.

In steps S220 to S228, the driving data configuration unit 365 configures the collected driving data stored in the driving data DB 343 into a data set in a form suitable for machine learning. In step S220, the driving data configuration unit 365 identifies whether the driving data selected for configuration is a numeric type or a categorical type.

If the selected driving data is categorical, the procedure moves to step S222. However, if the selected driving data is a numeric type, the procedure moves to step S224. In step S222, the driving data configuration unit 365 converts the driving data corresponding to the categorical type into a numerical value by applying One-Hot Encoding. In step S224, the operation data configuration unit 365 processes numerical data or numerical categorical data by applying a partial aggregation approximation (PAA) algorithm.

In step S226, the driving data construction unit 365 checks whether the data processed according to the partial aggregation approximation (PAA) algorithm in step S224 is the last data among the collected driving data. If the data processed in step S224 is the last driving data (in the direction of'Yes'), the procedure moves to step S228. However, if the data processed in step S224 is not the last driving data (in the direction of'No'), the procedure moves to step S227.

In step S227, the driving data construction unit 365 will select the remaining data to which the partial aggregation approximation (PAA) algorithm is applied. Thereafter, the process returns to step S220 to repeat the process according to the data type. In step S228, the operation data construction unit 365 configures the operation data rearranged into time series data by a partial aggregation approximation (PAA) algorithm as a data set for applying the operation data to the machine learning algorithm.

In steps S230 to S234, the driving pattern generation unit 367 generates a driving pattern using machine learning on the configured data set. In step S230, the driving pattern generation unit 367 processes the data set generated through the partial aggregation approximation (PAA) algorithm with an artificial neural network-based self-organizing map (SOM) algorithm.

Subsequently, in step S232, the driving pattern generation unit 367 processes the result value of the self-organizing map (SOM) algorithm using the K-means++ algorithm. The data set may be clustered through hybrid machine learning that combines the SOM algorithm in step S230 and the K-means++ algorithm in step S323. In step S234, the driving data is clustered into a plurality of patterns as a result of the machine learning performed in steps S230 and S232. For example, the driving data may be clustered into a first driving pattern to an n-th driving pattern.

In step S240, the driving pattern analysis unit 369 analyzes the first driving pattern to the n-th driving pattern derived through machine learning. The driving pattern analysis unit 369 may select factors affecting the driving pattern from the geometry or facilities of road sections belonging to each cluster, the number of lanes, section length, road signs, traffic control equipment, and various facilities. That is, the driving pattern analysis unit 369 may estimate a causal relationship for each cluster and the infrastructure of the target road provided from the road information DB 347 of the storage unit 340.

In the above, a procedure for collecting and deriving and interpreting a driving pattern of driving data of the driving pattern management server 300 according to an embodiment of the present invention has been described. It will be well understood that the above-described method of deriving and interpreting the driving pattern is merely an embodiment of the present invention, and various changes are possible.

6 is a diagram illustrating a part of a road map for collecting driving data of an autonomous vehicle according to an embodiment of the present invention. Referring to FIG. 6, the road map 400 may be divided into a plurality of road sections.

The road on which the autonomous vehicle travels may be divided into a plurality of road sections 410, 412, 414, 416, …. Each road section may be divided in consideration of various road infrastructures such as increasing or decreasing lanes, intersections, and crosswalks. In addition, a unique road section identifier (ID) may be assigned to the divided road sections to map the driving data pattern of the autonomous vehicle. Driving data may be collected in units of divided road sections.

7 is a block diagram illustrating a procedure for collecting and configuring driving data according to an embodiment of the present invention. Referring to FIG. 7, the driving data 510 provided from the vehicle module 100 (refer to FIG. 1) may be divided into numeric data and categorical data according to data types. The driving data 510 may be used as input data for generating a driving pattern by applying the one-hot encoding 520 or directly applying the partial aggregation approximation (PAA) algorithm 530 according to the data type.

The driving data provided from the vehicle module 100 may include speed, acceleration, rotation rate, altitude value, wheel brake signal, turn signal signal, and the like. It will be well understood that the driving data may further include various driving detection signals detected from a lidar sensor or an image sensor in addition to these. Among the driving data 510, values of speed, acceleration, rotation rate, and altitude may be provided as numerical data measured according to the movement of the autonomous vehicle. On the other hand, a wheel brake signal or a turn signal signal is categorical data that only indicates whether it is on or off, not a numerical value.

The driving data configuration unit 365 (refer to FIG. 3) may separate the collected driving data 510 for each vehicle. In addition, the driving data configuration unit 365 may select only data that has been driven in the autonomous driving mode from the driving data 510 and data that can be identified by a road section identifier (ID). Also, for easy analysis, the driving data 510 including the missing value may be deleted.

In addition, the driving data configuration unit 365 may calculate the section length by using the average and speed of the length of the driving road section in order to set the minimum recognizable section of the driving pattern. In addition, the driving data configuration unit 365 will exclude from the data configuration a road section shorter than the section length corresponding to the minimum section among the collected driving data 510.

In the case of numerical data, the driving data construction unit 365 immediately applies a partial aggregation approximation (PAA) algorithm 530 for unifying the section length. That is, the speed, acceleration, rotation rate, and altitude values may be unified through the partial aggregate approximation (PAA) algorithm 530 without any additional processing. On the other hand, for categorical data such as wheel brake or turn signal signals, conversion to a numeric type is required.

Accordingly, the driving data configuration unit 365 converts categorical driving data such as wheel brakes or turn signal signals into numeric data using the one-hot encoding 520. In addition, the driving data configuration unit 365 may process data converted into numeric data using a partial aggregate approximation (PAA) algorithm 530. The driving data 510 processed by the partial aggregation approximation (PAA) algorithm is provided as input data for deriving a driving pattern.

8 is a graph briefly showing an example of the partial aggregation approximation (PAA) algorithm of the present invention mentioned in FIG. 7. Referring to FIG. 8, a method of applying a partial aggregation approximation (PAA) algorithm for constructing a data set for speed among variables of driving data is illustrated as an example.

The speed of the autonomous vehicle varies for each road section or according to time, and thus the input speed may be provided in different numbers according to a unit time or section. In order to construct a data set with such a speed, a partial aggregation approximation (PAA) algorithm is applied as a means for separating the time series into the same time interval and using the average value in each interval as a representative value.

For example, as shown, in a time series section for 12 seconds, the speed value may be provided as 12 values. However, in the case of unifying into 7 segments (Segments=7) to compose the data set, the time interval of 12 seconds is divided into 7 unit intervals, and the average values of the speed in each unit interval are divided into 7 representative values. Is mapped.

In this way, the speed, acceleration, turnover, and altitude values are processed into input data (data set) to derive the driving pattern. In addition, the partial aggregation approximation (PAA) algorithm can be applied to categorical data quantified by one-hot encoding. By applying a partial aggregate approximation (PAA) algorithm to all variables constituting the driving data, it can be processed into a data set for generating a driving pattern.

9 is a diagram illustrating a method of configuring a data set for categorical data by way of example. Referring to FIG. 9, application of a one-hot encoding and partial aggregation approximation (PAA) algorithm to a turn signal signal, which is one of categorical data, is illustrated.

The turn signal signal can be classified into Off, Left Signal (LTS), Right Signal (RTS), and Hazard Light. However, since Off and Hazard Light states can be expressed as a left signal (LTS) and a right signal (RTS), after one-hot encoding, only these variables can be simplified.

Each of the left signal LTS and the right signal RTS may be mapped to numerical data '0' or '1' according to on-off by one-hot encoding. For example, it is assumed that 8 left signals (LTS) and right signals (RTS) are provided in section 1, respectively. Then, the one-hot encoded result of the left signal LTS is '0, 0, 1, 0, 0, 1, 1, 1', and the one-hot encoded result of the right signal RTS is '1, It can be mapped to 0, 0, 0, 0, 0, 0, 0'.

However, the turn signal signal must be unified into 7 segments (Segments=7) in order to compose a data set for deriving a driving pattern. Therefore, by applying the partial aggregation approximation (PAA) algorithm, the left signal (LTS) is '0, 0.3, 0.6, 0, 0.6, 1, 1', and the right signal (RTS) is '0.8, 0, 0, 0, It may be composed of 0, 0, 0'.

In the above, the procedure of one-hot encoding and partial aggregation approximation (PAA) algorithm for constructing a data set of categorical data has been exemplarily described. It will be appreciated that the method of constructing this data set can be applied in a similar manner to not only the turn signal but also the wheel brake signal.

10 is a diagram illustrating a hybrid clustering operation corresponding to the operation pattern derivation step of FIGS. 4 to 5 by way of example. Referring to FIG. 10, the driving pattern input data constructed through the partial aggregation approximation (PAA) algorithm is generated as a driving pattern by a self-organizing map (SOM) and a hybrid clustering operation using the K-means++ algorithm.

The driving pattern generation unit 367 (refer to FIG. 3) preferentially applies a self-organizing map (SOM) algorithm to the driving pattern input data configured through a partial aggregation approximation (PAA) algorithm. By the self-organizing map (SOM) algorithm, the input vectors of the data set are adjacent to nodes in two dimensions. The Euclidean distances of nodes allocated in the data space will be calculated by the self-organizing map (SOM) algorithm. And, based on the calculated Euclidean distance, unsupervised learning, that is, clustering will be performed on the data set. Data sets may be clustered into a plurality of clusters 610, 611, 612, 613, 614, ...) as shown by the self-organizing map (SOM) algorithm.

The K-means++ algorithm is applied to a plurality of clusters 610, 611, 612, 613, 614, ...) provided as a result of the self-organizing map (SOM) algorithm. The K-means++ algorithm is an algorithm that enhances the initial value selection function compared to the K-means algorithm. By additionally applying the K-means++ algorithm, higher clustering performance can be provided than when the single clustering method of the self-organizing map (SOM) algorithm is applied.

According to the application of the K-means++ algorithm, a plurality of clusters 610, 611, 612, 613, 614,… will be clustered into new unit clusters 630, 631, 632, 633, 634, 635. The six clusters show that the driving patterns of autonomous vehicles can be largely clustered into six.

11 and 12 are diagrams showing a method of interpreting the derived driving pattern in connection with infrastructure on an actual road. 11 exemplarily shows a mapping relationship between road sections of an actual road and a cluster. 12 is a table summarizing the infrastructure or characteristics of each clustered cluster and a road that is a factor thereof.

Referring to FIGS. 11 and 12, it can be seen that driving data in road sections of an autonomous vehicle are derived as driving patterns corresponding to a plurality of clusters (Cluster 1 to Cluster 6) through the hybrid clustering operation of the present invention. I can.

For example, the driving pattern of the autonomous vehicle may be derived as the first cluster 1 by the driving data acquired in the road sections 750, 751, 752, 753, 754, and 755. Driving patterns in this section include speed (9.64 km/h), acceleration (0.05 km/h/s), rotation rate (-3.17 deg/s), altitude change (0 m/s), wheel brake (0.85), It leads to the left traffic light (0.98) and the right traffic light (0).

In the case of the first cluster (Cluster 1), it can be seen that a number of sections turning left at the intersection are included. When interpreted in connection with the road sections 750, 751, 752, 753, 754, 755, it can be interpreted that crosswalks and taxiways have influenced the driving pattern. In other words, it can be seen that there is a tendency to decelerate in front of a crosswalk and accelerate after a crosswalk. In addition, it can be seen that there is a tendency to use the wheel brake in the guide line during rotation.

The driving pattern of the autonomous vehicle in the road sections 730, 731, and 732 may be derived as a third cluster 3. The driving pattern in this section is speed (19.24 km/h), acceleration (0.10 km/h/s), rotation rate (-0.05 deg/s), altitude change (0.17 m/s), wheel brake (0.11), It is derived from the left traffic light (0.04) and the right traffic light (0). In the case of the third cluster (Cluster 3), considering the change in the altitude value, it can be seen that a number of uphill sections are included. When analyzed in connection with the infrastructure of the road sections 730, 731, and 732, it can be seen that the slope exists and the lane is reduced.

In the above, the process of interpreting by linking road sections and driving patterns has been exemplarily described. Clusters that are not described (Cluster 2, Cluster 4 to Cluster 6) can also be interpreted in the same manner as described above in connection with a road section. In the above-described manner, road sections can be derived as driving patterns corresponding to clusters (Cluster 1 to Cluster 6), and analyzed in consideration of road infrastructure and environment. Road sections for which a driving pattern is not derived are cases where the collection time of driving data is less than the standard value or the amount of collected driving data is small.

Through a technology for deriving and analyzing a driving pattern based on driving data obtained from the autonomous driving vehicle on the real road of the above-described method, it can be applied to the driving algorithm of the autonomous driving vehicle. Since the derived driving pattern is obtained from a level 4 level autonomous vehicle, the analysis result of the driving pattern can be applied to the operation algorithm of a high level autonomous vehicle. In addition, it is possible to establish standards for dangerous driving of autonomous vehicles, and thus can be used for calculating insurance premiums for autonomous vehicles. In addition, it can be used in the control center for accident prevention and post-accident treatment that occur when driving autonomous vehicles.

The contents described above are specific examples for carrying out the present invention. The present invention will include not only the above-described embodiments, but also embodiments that can be simply changed or easily changed. In addition, the present invention will also include techniques that can be easily modified and implemented using the embodiments. Therefore, the scope of the present invention should not be defined by being limited to the above-described embodiments, and should be defined by the claims and equivalents of the present invention as well as the claims to be described later.

Claims (5)

In the driving pattern management server receiving driving data provided from an autonomous vehicle:
A communication unit for receiving the driving data transmitted from the autonomous vehicle;
A storage unit for storing the received driving data and road infrastructure information;
A control unit for deriving the accumulated driving data as a driving pattern for each road section through machine learning, and analyzing the driving pattern in connection with the infrastructure information,
The control unit:
Controlling the communication unit and the storage unit to accumulate the driving data in the storage unit for a specific period;
The accumulated driving data is divided for each section of the road, and the divided driving data is configured as a data set for machine learning,
The data set is clustered using a self-organizing map (SOM) algorithm and a K-means++ algorithm to derive a driving pattern, and
Analyzing the driving characteristics of the autonomous vehicle for each section of the road by linking the driving pattern and the infrastructure information,
The driving data includes information on speed, acceleration, rotation rate, and altitude values of the autonomous vehicle, and a wheel brake signal and a turn signal signal of the autonomous vehicle,
The control unit may divide each of the wheel brake signal and the turn signal signal, the speed, acceleration, rotation rate, and altitude values into an equal number of segments for a specific time series, and map a representative value to a piecewise aggregate approximation ) To construct the data set by processing with an algorithm,
The control unit converts the wheel brake signal and the turn signal signal corresponding to the categorical data into numeric data through One-Hot Encoding, and then the Piecewise Aggregate Approximation (PAA) Operation pattern management server that processes according to the algorithm. delete delete delete The method of claim 1,
The driving data is a driving pattern management server received from the autonomous vehicle through Vehicle to Everything (V2X).

Images