Factor analysis server for evaluating driving suitability of autonomous vehicle on road

Abstract

An analysis server for analyzing factors of a dangerous road section by receiving a precise road map and a basic safety message (BSM) provided from an autonomous vehicle, according to the present invention, comprises a communication unit, a storage unit and a control unit, wherein the control unit controls the communication unit and the storage unit so as to accumulate the BSM or the precise road map in the storage unit for a specific period of time, configures a dataset by calculating an independent variable from the BSM and calculating a dependent variable from the precise road map, generates and trains a dangerous road section classification model by applying an XGBoost algorithm to the dataset, calculates importance for each factor of the dangerous road section, and performs factor analysis on the dangerous road section on the basis of the calculated importance for each factor.

Description

Factor Analysis Server for Evaluating Autonomous Vehicle Roadworthiness

[0001] The present invention relates to autonomous vehicles, and more particularly, to a factor analysis server for evaluating the road-driving suitability of autonomous vehicles.

[0002] <National Research and Development Project Supporting This Invention>

[0003] Assignment ID: 1711193465, Assignment Number: 2021-0-01415

[0004] Ministry Name: National Police Agency, Project Management (Specialist) Agency Name: Information and Communications Technology Planning and Evaluation Institute

[0005] Research Project Name: Autonomous Driving Technology Development Innovation Project

[0006] Research Project Title: Test Scenario Creation and Multi-Agent-Based Simulation SW Technology Development for Edge-Linked Urban Autonomous Driving Service Verification

[0007] Project implementing organization: Next Generation Convergence Technology Research Institute / Korea Electronics and Telecommunications Research Institute

[0008] Contribution rate: 1 / 1, Research period: April 1, 2021 - December 31, 2025

[0009] An autonomous vehicle (AV) is a system that has the ability to detect and process external information while driving, recognizes the surrounding environment, determines its own driving path, and drives independently using its own power, without the driver having to control the brakes, steering wheel, or accelerator pedal.

[0010] The market for autonomous vehicles is rapidly growing each year, and commercialization is expected in the near future. Consequently, the number of test areas for autonomous vehicle on-road operation is steadily increasing. In the early stages of autonomous vehicle deployment, mixed traffic with conventional vehicles is inevitable. Therefore, various studies are being conducted to analyze the impact of autonomous vehicle adoption. Furthermore, systems capable of analyzing detailed road sections are required to support the monitoring of road hazards by control systems and to provide data for real-time recognition, judgment, and control of autonomous vehicles.

[0011] The purpose of the present invention is to provide an analysis server and an analysis method thereof that perform factor analysis to evaluate the road driving suitability of an autonomous vehicle that provides verification and service in a real road environment.

[0012] According to an embodiment of the present invention, an analysis server for analyzing factors of a dangerous road section that receives a basic safety message (BSM) provided from a precise road map and an autonomous vehicle includes a communication unit for receiving the basic safety message (BSM) and the precise road map, a storage unit for storing the received basic safety message (BSM) or the precise road map, and a control unit for identifying and analyzing spatial characteristics of a dangerous road section where an event has occurred by utilizing static factors for each road section using the stored basic safety message (BSM) or the precise road map, wherein the control unit controls the communication unit and the storage unit to accumulate the basic safety message (BSM) or the precise road map in the storage unit for a specific period of time, calculates an independent variable from the basic safety message (BSM), calculates a dependent variable from the precise road map to form a dataset, and generates and learns a dangerous road section classification model by applying an XGBoost algorithm to the dataset, calculates an importance for each factor of the dangerous road section, and performs factor analysis of the dangerous road section based on the calculated importance for each factor.

[0013] In this embodiment, the basic safety message (BSM) includes at least one of information about the speed, lateral acceleration, longitudinal acceleration, and rotation rate of the autonomous vehicle.

[0014] In this embodiment, the independent variables include information such as curve radius, slope, number of intersections, number of lanes, length of road section, and whether there is an intersection for each road section.

[0015] In this embodiment, the dependent variable includes risky driving behavior, and the road segment is classified as a road segment where the risky driving behavior occurs by utilizing a hybrid SOM K-means++ model for the risky driving behavior.

[0016] In this embodiment, the importance of each factor is calculated using parameters such as the amount of change in slope, length of road section, slope, and curve radius in the dangerous road section.

[0017] According to the analysis server and its analysis method for evaluating the road driving suitability of an autonomous vehicle according to the embodiment of the present invention described above, the road driving suitability of an autonomous vehicle can be analyzed by calculating the importance of each factor.

[0018] FIG. 1 is a block diagram exemplarily showing an analysis system for analyzing factors for evaluating road driving suitability of an autonomous vehicle according to an embodiment of the present invention.

[0019] Figure 2 is a block diagram exemplarily showing the configuration of a vehicle sensor module loaded or mounted on an autonomous vehicle to generate a basic safety message (BSM).

[0020] Fig. 3 is a block diagram exemplarily showing the configuration of an analysis server for evaluating the road driving suitability of an autonomous vehicle illustrated in Fig. 1.

[0021] Figure 4 is a flowchart showing a method for constructing a risk road section classification model and analyzing factors of a risk road section performed by the analysis server of the present invention.

[0022] Figure 5 is a flowchart showing in more detail the method for constructing the risk road section classification model of Figure 4 and analyzing factors of the risk road section.

[0023] Figure 6 is a diagram exemplarily showing a method for calculating a curve radius in the data set configuration process illustrated in Figure 5.

[0024] Figure 7 is a diagram showing an example of constructing a final dataset for classification of hazardous road sections.

[0025] Figure 8 is a diagram showing an example of the final dataset for classification of hazardous road sections.

[0026] Figure 9 is a diagram exemplifying the XGBoost learning process.

[0027] Figure 10 is a diagram visually showing the results of applying a risk road section classification model.

[0028] Figure 11 is a graph showing the results of factor analysis and importance calculation for deriving risk road section scenarios.

[0029] Figure 12 is a diagram showing the results of deriving scenarios for risky road sections using the selected key factors.

[0030] Figure 13 is a photograph showing visualized data based on the results obtained in the present invention.

[0031] A drawing showing the best mode for carrying out the present invention is Fig. 4.

[0032] Hereinafter, embodiments of the present invention will be described clearly and in detail to the extent that a person having ordinary skill in the art can easily practice the present invention.

[0033] Hereinafter, some embodiments of the present invention will be described in detail with reference to exemplary drawings. When designating components in each drawing, identical components may be given the same reference numerals, as much as possible, even if they appear in different drawings. Furthermore, when describing the present invention, detailed descriptions of known related structures or functions may be omitted if it is determined that such detailed descriptions may obscure the gist of the present invention.

[0034] FIG. 1 is a block diagram exemplarily showing an analysis system for analyzing factors for evaluating the road-driving suitability of an autonomous vehicle according to an embodiment of the present invention. Referring to FIG. 1, an analysis system (10) for acquiring a basic safety message (hereinafter, BSM) and a precise road map of an autonomous vehicle (100) and analyzing key characteristics of a dangerous road section includes an autonomous vehicle (100), a communication network (200), and an analysis server (300).

[0035] The autonomous vehicle (100) may be a vehicle equipped with an autonomous driving system and various types of vehicles capable of driving on actual roads to acquire verification big data. The vehicle sensor module (105) mounted on the autonomous vehicle (100) generates a basic safety message (BSM) of the autonomous vehicle (100) using various sensors. For example, the autonomous vehicle (100) may include a speed sensor, an acceleration sensor, a yaw rate sensor, a position sensor, etc. Then, the autonomous vehicle (100) generates a basic safety message (BSM) based on the sensing data detected from these sensors and transmits it to an analysis server (300) via a communication network (200).

[0036] The communication network (200) provides a communication channel between the autonomous vehicle (100) and the analysis server (300). The communication network (200) refers to a wireless or wired communication structure for exchanging information between each node, such as the autonomous vehicle (100) or the analysis server (300). For example, the communication network (200) may include vehicle-to-everything (V2X) communication for exchanging information between vehicles and objects such as other vehicles, mobile devices, and roads. Alternatively, the communication network (200) may include 3GPP (3rd Generation Partnership Project), LTE (Long Term Evolution), WIMAX (World Interoperability for Microwave Access), Wi-Fi, 3G, 4G, 5G, 6G, etc., but the present invention is not limited thereto.

[0037] The analysis server (300) builds and trains a dangerous road section classification model using the XGBoost algorithm that uses a decision tree as a basic learner, using the basic safety message (BSM) and the detailed road map provided from the autonomous vehicle (100) and the communication network (200). The analysis server (300) can use the basic safety message (BSM) and the detailed road map to identify the spatial characteristics of the dangerous road section where an event occurred by utilizing static factors for each road section.

[0038] The analysis server (300) can precisely assess road driving suitability by considering the real-time nature of the autonomous vehicle's (100) perception / judgment / control, and quantitatively calculate major contributing static factors of dangerous road sections. The analysis server (300) can monitor the current location of the autonomous vehicle (100) on the road section in which it operates, as well as dangerous road sections in real time, thereby enabling an immediate response to dangerous situations.

[0039] Additionally, the analysis server (300) can predict the expected risk areas on the road on which the autonomous vehicle (100) operates based on the characteristics of the risk factors, based on the importance of the derived static factors. The analysis server (300) can monitor the accident risk level for each section of the road on which the autonomous vehicle (100) operates in conjunction with road infrastructure, and use this data to enhance the safety of vehicle operation.

[0040] The analysis server (300) described above can analyze important factors for assessing road suitability based on the Basic Safety Message (BSM) and precision road map provided by autonomous vehicles (100) collected on actual roads. The derived important factors can be used to ensure traffic safety in mixed traffic situations where autonomous vehicles (100) and conventional vehicles coexist.

[0041] FIG. 2 is a block diagram exemplarily showing the configuration of a vehicle sensor module that is loaded or installed in an autonomous vehicle and generates a basic safety message (BSM). Referring to FIG. 2, the vehicle sensor module (105) can generate a basic safety message (BSM) of an autonomous vehicle (100, see FIG. 1) and transmit it to an analysis server (300). The vehicle sensor module (105) can include a sensor unit (110), a sensor hub (130), and a vehicle communication unit (150).

[0042] The sensor unit (110) recognizes movement or the surrounding environment, such as speed, acceleration, position, and yaw rate, detected when the autonomous vehicle (100) is in operation. The sensor unit (110) may include a speed sensor (111), an acceleration sensor (112), a position sensor (113), a yaw rate sensor (114), etc., to sense the movement characteristics of the autonomous vehicle (100). Here, the acceleration sensor (112) may include a longitudinal acceleration sensor that detects longitudinal acceleration of the autonomous vehicle (100) and a lateral acceleration sensor that detects lateral acceleration. In addition, the sensor unit (110) may include a lidar sensor (115) or a radar sensor (116) to recognize objects or situations around the autonomous vehicle.

[0043] 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) can receive sensing data randomly transmitted from the plurality of sensors periodically or aperiodically. The sensing data of each of the plurality of sensors (111 to 116) is collected by the sensor hub (130). The sensor hub (130) can also convert the sensing data transmitted from the sensor unit (110) into a data format for efficient transmission. The sensor hub (130) can be implemented using a processor or various computational cores.

[0044] The vehicle communication unit (150) transmits sensing data provided from the sensor hub (130) through a communication network (200). For example, the vehicle communication unit (150) may include a wired or wireless communication module that supports vehicle-to-everything (V2X) communication.

[0045] It will be readily understood that the types of sensors included in the vehicle sensor module (105) are not limited to those illustrated. That is, the vehicle sensor module (105) may include various sensors for detecting movement or environmental information of the autonomous vehicle (100), including an altitude sensor or a temperature sensor. The vehicle sensor module (105) described above may be mounted integrally or modularly on the autonomous vehicle (100).

[0046] FIG. 3 is a block diagram exemplarily showing the configuration of an analysis server for evaluating the road driving suitability of an autonomous vehicle illustrated in FIG. 1. The analysis server (300, see FIG. 1) processes a basic safety message (BSM) and a precise road map based on real-time road driving transmitted from a vehicle sensor module (105, see FIG. 2) to quantitatively analyze major contributing static factors of a dangerous road section. Referring to FIG. 3, the analysis server (300) may include a communication unit (320), a storage unit (340), and a control unit (360).

[0047] The communication unit (320) includes a receiving unit (321) and a transmitting unit (323). The receiving unit (321) receives a precision road map and a basic safety message (BSM) transmitted from a vehicle sensor module (105). The receiving unit (321) can convert the precision road map and the vehicle sensor module (105) transmitted through a communication network (200, see FIG. 1) into a data format processed by the control unit (360). The receiving unit (321) will transmit the received precision road map and the vehicle sensor module (105) to the control unit (360). The transmitting unit (323) can transmit the results of the analysis of important factors of a dangerous road section derived from the analysis server (300) to a location that requires them.

[0048] The storage unit (340) may include a vehicle information DB (341), a basic safety message DB (343), a precision road map DB (345), and a factor analysis result DB (347). The storage unit (340) may be configured as storages for storing data managed by the analysis server (300).

[0049] The vehicle information DB (341) stores information about autonomous vehicles (100). For example, if there are two or more autonomous vehicles transmitting basic safety messages (BSM), information for identifying each vehicle may be stored in the vehicle information DB (341). In addition, vehicle characteristic information, such as the type, weight, and size of the autonomous vehicle (100), may also be stored in the vehicle information DB (341).

[0050] The basic safety message DB (343) stores the basic safety message (BSM) transmitted from the vehicle sensor module (105) under the control of the data collection unit (363). The basic safety message (BSM) may be a large amount of big data transmitted from the vehicle sensor module (105) over a long period of time to derive a driving pattern. The basic safety message DB (343) may store the driving information of the autonomous vehicle (100) in the form of raw data.

[0051] The precision road map DB (345) may store information on road infrastructure and road section identifiers (IDs) for generating basic safety messages (BSMs) of autonomous vehicles (100). For example, road infrastructure may include road facilities or information that affect the operation of autonomous vehicles (100), such as lane centerlines, crosswalks, guide lines, overpasses, intersections, roundabouts, and lane increase or decrease. In addition, the precision road map DB (345) may store precision road maps for each road section and provide them to the control unit (360). The factor analysis result DB (347) stores information on major contributing static factors of dangerous road sections generated by the control unit (360).

[0052] The control unit (360) collects the basic safety message (BSM) and the precision road map transmitted through the communication unit (320), and configures the collected data into a processable dangerous road section classification dataset through preprocessing. The control unit (360) uses the XGBoost (Gradient Tree Boost) algorithm that utilizes a decision tree to build and train a dangerous road section classification model using the generated dataset. The analysis server (300) can identify the spatial characteristics of the dangerous road section where an event occurred by utilizing static factors for each road section using the basic safety message (BSM) and the precision road map. After training the dangerous road section classification model, the analysis server (300) checks the classification result and calculates the importance of each factor. The analysis server (300) can identify the static factor characteristics of the dangerous road section and generate a scenario based on the importance of each factor for each dangerous road section.

[0053] To this end, the control unit (360) includes a processor (361), a data collection unit (363), a dataset configuration unit (365), a classification model learning unit (367), and a classification result factor analysis unit (369). Here, preferably, the processor (361) may be configured as hardware, and the data collection unit (363), the dataset configuration unit (365), the classification model learning unit (367), and the classification result factor analysis unit (369) may be provided as software.

[0054] The processor (361) can control the overall operation of the analysis server (300). The processor (361) can patch the basic safety message (BSM) or precision road map received through the communication unit (320) or transmit a control signal. The processor (361) can access the vehicle information DB (341), the basic safety message DB (343), the precision road map DB (345), and the factor analysis result DB (347) of the storage unit (340). The processor (361) can execute an algorithm or program command that constitutes a data collection unit (363), a dataset configuration unit (365), a classification model learning unit (367), and a classification result factor analysis unit (369). The processor (361) can be implemented in the form of a system-on-chip (SoC), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), etc.

[0055] The data collection unit (363) stores the basic safety message (BSM) transmitted from the vehicle sensor module (105, see FIG. 2) of the autonomous vehicle (100) and the precise road map in the storage unit (340). The basic safety message (BSM) is a type of message set transmitted in real time through sensing of safety or emergency situations of the autonomous vehicle (100). For example, if the brakes of the autonomous vehicle (100) are suddenly applied on a specific road section, the vehicle sensor module (105) recognizes this situation as an emergency and urgently transmits the basic safety message (BSM) to the communication network (200) or surrounding vehicles. Therefore, during the period for collecting the basic safety message (BSM), the data collection unit (363) can accumulate the basic safety message (BSM) by period and road identifier (ID). The basic safety message (BSM) can be managed in the form of big data accumulated for a specific meaningful period.

[0056] The dataset composition unit (365) reconstructs the collected basic safety messages (BSM) or precision road maps into a form that is easy for machine learning. The dataset composition unit (365) processes the basic safety messages (BSM) and precision road maps into a form for learning by a deep learning-based clustering algorithm. The dataset composition unit (365) analyzes the composition of the basic safety messages (BSM) and selects variables that affect the driving safety of an autonomous driving road. The dataset composition unit (365) performs preprocessing using, in particular, information on speed, longitudinal acceleration, lateral acceleration, and yaw rate.

[0057] The dataset composition unit (365) extracts data that meets the criteria by considering the autonomous driving mode, sensor type, and presence or absence of communication failure from the detailed road map and the basic safety message (BSM). In addition, the dataset composition unit (365) calculates independent variable items based on the detailed road map. In addition, the dataset composition unit (365) calculates dependent variable items from the basic safety message (BSM). The independent variable items include information such as the curve radius, slope, number of intersections, number of lanes, length of the road section, and presence or absence of an intersection for each road section. The dependent variable is calculated by checking whether or not there is a dangerous driving behavior for each road section based on the vehicle's speed, acceleration, and rotation rate included in the basic safety message (BSM).

[0058] The classification model learning unit (367) builds and learns a dangerous road section classification model using the XGBoost algorithm, which is a machine learning algorithm, for the dataset constructed by the dataset construction unit (365). The XGBoost algorithm is a decision tree algorithm based on boosting, and is configured to repeatedly learn the same tree model focusing on misclassified parts and update the weights. When the XGBoost algorithm is used, high classification performance is provided even for a small dataset by utilizing parallel processing of data and a regularizer. When the XGBoost algorithm is used, various importance levels for each factor of the dangerous road section can be calculated. Through this, the classification model learning unit (367) can quantitatively analyze the contribution of the corresponding factor based on the calculated importance levels.

[0059] The classification result factor analysis unit (369) checks the result of applying the classification model to the dangerous road section. Then, the classification result factor analysis unit (369) performs factor analysis of the dangerous road section based on the calculated importance. The classification result factor analysis unit (369) calculates the importance for each factor using parameters such as the amount of change in slope, road section length, slope, and curve radius in the dangerous road section. The classification result factor analysis unit (369) can configure a scenario by utilizing the selected major factors. That is, various scenarios can be produced according to the size of the selected major factors. For example, scenarios can be generated when the length of the road section is less than 50 m, when the curve radius is large, and when it is not a downhill road.

[0060] Figure 4 is a flowchart illustrating a method for constructing a risky road section classification model and analyzing factors of risky road sections, performed by the analysis server of the present invention. Referring to Figure 4, the analysis server (300) can accumulate basic safety messages (BSM) and detailed road maps provided by autonomous vehicles (100), and perform major factor analysis of risky road sections based on the accumulated data.

[0061] At step S110, the data collection unit (363) stores the basic safety message (BSM) and the precision road map in the basic safety message DB (343) and the precision road map DB (345) of the storage unit (340). The basic safety message (BSM) can be accumulated by vehicle, period, and road section and stored in the basic safety message DB (343).

[0062] In step S120, the dataset composition unit (365) extracts data that meets the criteria by considering the autonomous driving mode, sensor type, and presence or absence of communication failure from the detailed road map and the basic safety message (BSM). In addition, the dataset composition unit (365) calculates independent variable items based on the detailed road map. In addition, the dataset composition unit (365) calculates dependent variable items from the basic safety message (BSM). The independent variable items include information such as the curve radius, slope, number of intersections, number of lanes, length of the road section, and presence or absence of an intersection for each road section. The dependent variable is calculated by checking whether or not there is a dangerous driving behavior for each road section based on the vehicle's speed, acceleration, and rotation rate included in the basic safety message (BSM).

[0063] In step S130, the classification model learning unit (367) uses the XGBoost algorithm, which is a machine learning algorithm, for the dataset configured by the dataset configuration unit (365) to build and train a dangerous road section classification model. The XGBoost algorithm is a decision tree algorithm based on boosting, and is configured to repeatedly train the same tree model focusing on misclassified parts and update the weights. When the XGBoost algorithm is used, high classification performance is provided even for a small dataset by utilizing parallel processing of data and a regularizer. When the XGBoost algorithm is used, various importance levels for each factor of the dangerous road section can be calculated.

[0064] In step S140, the classification model learning unit (367) can quantitatively analyze the contribution of the corresponding factor based on the calculated importance. The classification result factor analysis unit (369) checks the result of applying the classification model to the dangerous road section and performs factor analysis of the dangerous road section based on the calculated importance. The classification result factor analysis unit (369) calculates the importance of each factor using parameters such as the amount of change in slope, road section length, slope, and curve radius in the dangerous road section. The classification result factor analysis unit (369) can configure a scenario using the selected major factors. That is, various scenarios can be produced according to the size of the selected major factors. For example, scenarios can be generated when the length of the road section is less than 50 m, when the curve radius is large, and when it is not a downhill road.

[0065] In the above, the data collection and dataset configuration of the analysis server (300) according to an embodiment of the present invention, the learning of the risk road section classification model, and the risk road section classification result factor analysis procedure have been briefly described.

[0066] Figure 5 is a flowchart illustrating in more detail the construction of the risk road section classification model of Figure 4 and the method for factor analysis of the risk road section. Referring to Figure 5, the analysis server (300) can perform major factor analysis of the risk road section based on the basic safety message (BSM) and the detailed road map.

[0067] In step S210, the analysis server (300) collects the basic safety message (BSM) of the autonomous vehicle (100) and a detailed road map. More specifically, the data collection unit (363) stores the basic safety message (BSM) transmitted from the autonomous vehicle (100) in the basic safety message database (343). The basic safety message (BSM) is collected over a predetermined period of time and accumulated in the basic safety message database (343). The detailed road map may include status information for each road section and each lane of each road section.

[0068] In steps S220 to S228, the dataset configuration unit (365) configures the collected basic safety messages (BSM) and precision road maps into a dataset (Data-set) suitable for generating and learning a hazardous road section classification model. First, in step S220, the dataset configuration unit (365) removes data containing errors from the collected data and extracts only normal data.

[0069] In steps S221 to S223, the dataset composition unit (365) preprocesses the extracted precision road map data. In step S221, the dataset composition unit (365) calculates raw data of independent variable items based on the precision road map. In step S223, the dataset composition unit (365) calculates information such as curve radius, slope, number of intersections, number of lanes, length of road section, and presence or absence of intersection for each road section from the raw data.

[0070] In steps S222, S224, and S226, the dataset construction unit (365) calculates dependent variable items from the basic safety message (BSM). In step S222, the dataset construction unit (365) extracts raw data of dependent variable items based on the basic safety message (BSM). In step S224, the dataset construction unit (365) maps the road section where the basic safety message (BSM) occurs. In step S226, the dataset construction unit (365) determines whether there is a risky driving behavior for each road section based on the speed, acceleration, and rotation rate of the vehicle included in the basic safety message (BSM) and calculates dependent variables based on the results. Therefore, the dependent variables may include risky driving behavior items.

[0071] In step S228, the final dataset is constructed by combining the independent and dependent variables produced.

[0072] In steps S230 to S232, the classification model learning unit (367) builds and learns a dangerous road section classification model using the XGBoost algorithm from the configured dataset. In step S230, the classification model learning unit (367) builds and learns a dangerous road section classification model using the XGBoost algorithm, which is a machine learning algorithm, for the dataset configured by the dataset construction unit (365). The XGBoost algorithm is a boosting-based decision tree algorithm, and is configured in a way that iteratively learns the same tree model focusing on misclassified parts and updates the weights. When the XGBoost algorithm is used, high classification performance is provided even for a small dataset by utilizing parallel processing of data and a regularizer. In step S232, the classification model learning unit (367) can classify the sections of the detailed road map into a normal section group and a dangerous section group according to the learning result.

[0073] In step S240, the classification result factor analysis unit (369) verifies the learning results and classification results and checks the results of applying the classification model to the dangerous road section. Then, the classification result factor analysis unit (369) performs static factor analysis of the dangerous road section based on the calculated importance. The classification result factor analysis unit (369) calculates the importance for each factor using parameters such as the amount of change in slope, road section length, slope, and curve radius in the dangerous road section.

[0074] In step S250, the classification result factor analysis unit (369) constructs a scenario using the major factors of the selected risky road section. That is, various scenarios can be generated depending on the magnitude of the selected major factors. For example, scenarios can be generated for road sections less than 50 m long, with a large curve radius and no downhill slope. Alternatively, scenarios can be generated by combining major factors such as road sections exceeding 50 m long, with a downhill slope, or with a large curve radius. Additionally, a visualization process can be performed to monitor the road sections for which analysis has been completed.

[0075] The above describes the data collection and dataset configuration, classification model training, and static factor and scenario generation procedures for risky road sections based on the importance of each factor by the analysis server (300) according to an embodiment of the present invention. Additionally, a visualization process may be implemented to monitor road sections for which analysis has been completed. Data generated through the visualization process can be used by traffic control centers to monitor risky road sections in real time and establish road safety measures.

[0076] Figure 6 is a diagram illustrating the method for calculating curve radii during the dataset construction process illustrated in Figure 5. Using the collected data, a dataset can be constructed by classifying hazardous road sections. The dataset construction process largely involves an independent variable dataset containing static factors for each road section and a dependent variable dataset containing the presence or absence of risky driving behavior.

[0077] The items that make up an independent variable dataset can be broadly categorized into continuous and discrete. Continuous items can include curve radius, change in curve radius, slope, change in slope, and road section length. Discrete items can include intersections, number of lanes, and more. Independent variable datasets can include modifications to existing factors and other static factors.

[0078] Referring to Figure 6, the change in curve radius and the change in slope can be defined as the difference between the values ​​of the previous road section and the values ​​of the subsequent road section. The curve radius can be calculated using Equation 1.

[0079]

[0080] Figure 7 is a diagram illustrating an example of constructing a final dataset for classifying hazardous road segments. Figure 7 exemplifies a method for classifying road segments where hazardous driving behaviors occur using a hybrid SOM K-means++ model.

[0081] The self-organizing map (SOM) algorithm is first applied to the input data. The SOM algorithm causes the input vectors of the data set to be adjacent to each other as two-dimensional nodes. The Euclidean distance between the nodes assigned in the data space by the SOM algorithm is calculated. Then, unsupervised learning, i.e., clustering, is performed on the data set based on the calculated Euclidean distance. When the data sets are initially clustered by the SOM algorithm, multiple clusters (410, 411, 412, …) can be generated.

[0082] The K-means++ algorithm is applied to multiple clusters 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 achieved than when applying the single clustering method of the Self-Organizing Map (SOM) algorithm. By applying the K-means++ algorithm, multiple clusters (410, 411, 412, …) will be clustered into new unit clusters (420, 421, 422, …).

[0083] The hybrid SOM K-means++ model can be used to classify autonomous vehicle driving patterns by leveraging the vehicle's speed, acceleration, and yaw rate (Yaw) contained in the Basic Safety Message (BSM), which is the dependent variable data for risky driving behavior. Using the hybrid SOM K-means++ model, road sections within risky driving behavior clusters can be classified as road sections where risky driving behavior occurs.

[0084] Figure 8 is a diagram illustrating the final dataset for classification of hazardous road sections. Referring to Figure 8, the road section IDs can be distinguished as Road_11, Road_21, Road_31, and Road_41. For each road section ID, the data shows numerical data for curve radius, curve radius change, slope, light-carriage change, road section length, intersections, number of lanes, and dangerous driving.

[0085] The curve radius in the road section (Road_11) is 560 m, the change in curve radius is 100 m, the gradient is 2.1%, the gradient change is 1.5%, the road section length is 30 m, the number of intersections is 0, the number of lanes is 2, and the number of dangerous driving occurrences is 1.

[0086] In the road section (Road_21), the curve radius is 800 m, the curve radius change is 80 m, the gradient is 0.5%, the gradient change is 1.1%, the road section length is 28 m, there are 2 intersections, the number of lanes is 3, and the number of dangerous driving occurrences is 0.

[0087] In the road section (Road_31), the curve radius is 770 m, the curve radius change is 300 m, the gradient is 1.3%, the gradient change is 0.5%, the road section length is 10 m, there is 1 intersection, the number of lanes is 5, and the number of dangerous driving occurrences is 1.

[0088] In the road section (Road_41), the curve radius is 600 m, the curve radius change is 150 m, the gradient is 2.1%, the gradient change is 0.4%, the road section length is 15 m, the number of intersections is 0, the number of lanes is 0, and the number of dangerous driving occurrences is 0.

[0089] The model training stage utilizes the final dataset to build a machine learning algorithm, XGBoost, and perform classification. XGBoost is a boosting-based decision tree algorithm that can be configured to repeatedly train the same tree model, focusing on misclassified areas, and update the weights.

[0090] Figure 9 is a diagram illustrating the XGBoost learning process. As illustrated in Figure 9, the learning process utilizes data parallel processing and a regularizer to achieve high classification performance even on small datasets. Various importance values ​​can be calculated for each factor, and the contribution of each factor can be quantitatively analyzed using these values.

[0091] Figure 10 is a diagram visually illustrating the results of applying a risky road section classification model. The factor analysis step verifies the results of applying the risky road section classification model and conducts factor analysis of the risky road section based on the calculated importance ratings. The model application results, as illustrated in Figure 10, confirm the factors and criteria used to classify risky road sections.

[0092] The classified groups are nodes marked with "Leaf." When the value is negative, the road segment is defined as a normal road segment group, and when it is positive, it is defined as a risky road segment group. This allows for the identification of the static factors and numerical criteria for determining the risky road segment groups.

[0093] Figure 11 is a graph showing the results of factor analysis and importance calculations for deriving risky road section scenarios. Referring to Figure 11, the graph indicates the number of features used to classify data in the model, with slope change and road section length being the most frequently utilized.

[0094] Since then, slope and curve radius have been widely utilized, and it can be confirmed that four factors have been utilized in Fig. 11, in the case of an autonomous vehicle (100), a scenario can be derived by selecting the main static factors as slope change amount, road section length, slope, and curve radius.

[0095] Figure 12 is a diagram illustrating the results of scenario derivation for risky road sections using selected key factors. Referring to Figure 12, scenarios include road sections less than 50 m in length, with a large curve radius and no downhill slope (Scenario 1). Road sections longer than 50 m include road sections that are downhill (Scenario 3) or have large curve radii (Scenarios 4 and 5).

[0096] Figure 13 is a photograph showing visualized data based on the results of the present invention. A visualization process can be performed to monitor the analyzed dangerous road sections. For precise analysis of the road sections, the autonomous driving infrastructure of the control center can be utilized to identify key characteristics.

[0097] From a control center perspective, the present invention can utilize Basic Safety Message (BSM) data transmitted via V2X terminals from all autonomous vehicles, including the Fanta G Bus, in autonomous vehicle pilot operation zones. The present invention can be used as reference material for analyzing the roadworthiness of vehicles providing various autonomous vehicle services, such as freight and passenger transport.

[0098] In addition, when the present invention develops and commercializes an autonomous driving navigation application, it can proactively provide information on dangerous road sections and serve as guidance for autonomous vehicles.

[0099] The above-described embodiments are specific examples for practicing the present invention. The present invention will encompass not only the embodiments described above, but also embodiments that can be easily modified or modified. Furthermore, the present invention will encompass techniques that can be easily modified and implemented using the embodiments described above. Therefore, the scope of the present invention should not be limited to the above-described embodiments, but should be defined not only by the claims set forth below, but also by equivalents of the claims of the present invention.

Claims

1. In an analysis server that analyzes factors of a dangerous road section that receives a precision road map and a basic safety message (BSM) provided by an autonomous vehicle: A communication unit receiving the above basic safety message (BSM) and the above precision road map; A storage unit for storing the received basic safety message (BSM) or the precision road map; Including a control unit that uses the stored basic safety message (BSM) or the precision road map to identify and analyze the spatial characteristics of the dangerous road section where the event occurred by utilizing static factors for each road section, The above control unit: Controlling the communication unit and the storage unit to accumulate the basic safety message (BSM) or the precision road map in the storage unit for a specific period of time; A dataset is constructed by calculating independent variables from the above basic safety message (BSM) and dependent variables from the above detailed road map; By applying the XGBoost algorithm to the above dataset, a risk road section classification model is created and learned, and the importance of each factor of the risk road section is calculated; and An analysis server that performs factor analysis of the risk road section based on the importance of each factor produced above.

2. In paragraph 1, The above basic safety message (BSM) is an analysis server that includes at least one of information about the speed, lateral acceleration, longitudinal acceleration, and rotation rate of the autonomous vehicle.

3. In paragraph 1, The above independent variables are analysis servers that contain information such as curve radius, slope, number of intersections, number of lanes, length of road section, and whether there is an intersection for each road section.

4. In paragraph 3, The above dependent variable includes risky driving behavior, and an analysis server classifies the road section as a road section where risky driving behavior occurs by utilizing a hybrid SOM K-means++ model for the above risky driving behavior.

5. In paragraph 1, An analysis server that calculates the importance of each factor by using parameters such as gradient change amount, road section length, gradient, and curve radius in the above-mentioned dangerous road section.

Images