KR20230039809A - Vehicle condition prediction system and method - Google Patents

Abstract

A vehicle condition prediction system comprises: a first layer generating data including one or more of the collective perception message (CPM) and basic safety message (BSM) of a vehicle; a second layer receiving the data through a network and parsing the same; a third layer generating a model by learning the parsed data before receiving new data after receiving the data, and predicting one or more of the speed, direction, and location of the vehicle by applying a Gaussian process (GP) based on the model; and a fourth layer generating a warning message by linking the result of the predicting with a safety application. Therefore, the performance of vehicle wireless communication technology can be increased.

Description

Vehicle condition prediction system and method {Vehicle condition prediction system and method}

The present invention relates to a vehicle condition prediction system and method.

Connected and Autonomous Vehicles (CAVs) rely on information from sensors and communications to make decisions. In a cooperative vehicle safety (CVS) system, information of a remote vehicle (RV) may be used in a host vehicle (HV) through a wireless network.

The performance of vehicle safety applications is highly dependent on the reliability of vehicle-to-everything (V2X) communication technology. Two popular V2X standards, Dedicated Short-Range Communication (DSRC) and Cellular-V2X (C-V2X), have been proposed in recent years as reliable communication solutions to support latency-sensitive safety systems.

However, both solutions still suffer from scalability issues for real-time applications. Therefore, problems such as information loss due to data collision, especially in congested scenarios, are unavoidable based on current communication technologies.

In order to solve the above-described problems, the present invention is to solve the above-mentioned problems in predicting the vehicle state, to mitigate the detrimental effect of data loss on the performance of safety applications, and to reduce the loss of V2X communication or It is intended to mitigate the effects of performance degradation.

The problems of the present invention are not limited to those mentioned above, and other problems not mentioned will be clearly understood by those skilled in the art from the description below.

A vehicle state prediction system according to one feature of the present invention is a first layer that generates data including at least one of collective perception messages (CPM) and basic safety messages (BSM) of the vehicle. , A second layer that receives and parses the data through the network, generates a model by learning the parsed data until new data is received after receiving the data, and based on the model, a Gaussian process (GP ; Gaussian Process) to predict any one or more of the speed, direction, and location of the vehicle, and a fourth layer to generate a warning message by linking the predicted result with a safety application.

The third layer generates a model by learning based on a tracking process for tracking the history of the vehicle, the most recent n points from the history of the vehicle, and an up-to-date model bank including one or more Gaussian process models and the A DOM array that compares models and predicts the state of the vehicle based on the selected model to generate DOM data, and parses the prediction information included in the DOM data based on the target of the module included in the fourth layer. and a target classifier for classifying and transmitting to the fourth layer.

The fourth layer includes an FCW module generating a forward collision warning message and a BSW module generating a blind spot warning message; and an EEBL module generating an emergency brake electronic detection message.

A method for predicting a vehicle state according to another feature of the present invention includes, in a DOM array, loading the most recent n points of vehicle history, the DOM array learning kernel parameters of a GP based on the n points. generating a first model, comparing the similarity of the first model with a latest model bank including one or more Gaussian process models in the DOM array, the DOM array, the first model bank in the latest model bank selecting the similar model if there is a model similar to the model; selecting the first model if there is no model similar to the first model in the latest model bank; and the DOM array. The arrangement includes predicting at least one of speed and direction of the vehicle using a Gaussian process based on the selected model.

The DOM arrangement may further include adding the first model to the latest model bank when the first model is selected.

In the step of calling the n points, the DOM array, if a new packet is not received after the most recent packet received from the vehicle, the DOM array, through the tracking process, the most recent n points of the history of the vehicle The step of loading the points is further included.

The DOM arrangement further includes updating a history of the vehicle based on the new packet if a new packet is received after the most recent packet received from the vehicle.

The predicting may include predicting a position of the vehicle based on at least one of the predicted speed and direction of the vehicle and an equation of motion.

The present invention can improve the performance of vehicle wireless communication technology. Data loss to safety application performance can be mitigated. The past information can be used to predict the future state of the vehicle, and the robustness of the CVS system can be improved by using a dynamic object map-based architecture to decouple the application and information/recognition subsystems.

The effects of the present invention are not limited to those mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description below.

1 is a diagram for explaining a hierarchy of a vehicle state prediction system according to an exemplary embodiment.
FIG. 2 is a flowchart of a method for predicting a vehicle state of a DOM array of FIG. 1 according to an embodiment.
3 is an exemplary diagram for describing a history of a vehicle according to an exemplary embodiment.
4 is an exemplary diagram for explaining a model bank learned in one embodiment.

The vehicle state prediction system 1 according to an embodiment and the vehicle state prediction method according to an embodiment are composed of a new prediction method and system architecture capable of using all prediction methods. The prediction method is based on non-parametric Bayesian inference using the Gaussian process. Using this method's model, past information can be used to predict future vehicle conditions. The activation architecture accompanying the prediction method is called a dynamic object map-based architecture, and this architecture can improve the robustness of the CVS system by separating the application and information/recognition subsystems.

Hereinafter, the embodiments disclosed in this specification will be described in detail with reference to the accompanying drawings, but the same or similar reference numerals are given to the same or similar components, and overlapping descriptions thereof will be omitted. The suffixes "module" and/or "unit" for components used in the following description are given or used interchangeably in consideration of ease of writing the specification, and do not themselves have a meaning or role distinct from each other. In addition, in describing the embodiments disclosed in this specification, if it is determined that a detailed description of a related known technology may obscure the gist of the embodiment disclosed in this specification, the detailed description thereof will be omitted. In addition, the accompanying drawings are only for easy understanding of the embodiments disclosed in this specification, the technical idea disclosed in this specification is not limited by the accompanying drawings, and all changes included in the spirit and technical scope of the present invention , it should be understood to include equivalents or substitutes.

Terms including ordinal numbers, such as first and second, may be used to describe various components, but the components are not limited by the terms. These terms are only used for the purpose of distinguishing one component from another.

In this application, terms such as "comprise" or "have" are intended to designate that there is a feature, number, step, operation, component, part, or combination thereof described in the specification, but one or more other features It should be understood that the presence or addition of numbers, steps, operations, components, parts, or combinations thereof is not precluded.

A program implemented as a set of commands embodying control algorithms necessary for controlling other components may be installed in a component that controls another component under a specific control condition among components according to an embodiment. The control component may generate output data by processing input data and stored data according to an installed program. The control component may include a non-volatile memory for storing programs and a memory for storing data.

A Gaussian Process (GP) is a stochastic process, which is a collection of random variables indexed by time or space. All finite collections of these random variables have a multivariate normal distribution. That is, all finite linear combinations of them follow a normal distribution.

Gaussian Process Regression (GPR) is a relatively new machine learning method. The GPR model is a non-parametric kernel-based probabilistic model. The GPR model can solve the problem of predicting the value of the response variable ynew given a new input vector xnew and training data. Accordingly, an embodiment of predicting a vehicle state using a GPR model will be described below.

1 is a diagram for explaining a hierarchy of a vehicle state prediction system according to an exemplary embodiment.

The vehicle state prediction system 1 may include a first layer 100 , a second layer 200 , a third layer 300 , and a fourth layer 400 .

The first layer 100 may be a data generation layer. The first layer 100 generates Basic Safety Messages (hereinafter referred to as BSM) through the DSRC log 101 and GPS 102 information, and the BSM Emulator 103 generates the information of the host vehicle (HV, 10). Sends Basic Safety Messages to the Network Cloud.

The network cloud also receives one or more BSMs 202 from surrounding vehicles (RVs) 20 . Here, the surrounding vehicles 20 may represent one or more vehicles located within a predetermined distance based on the location of the host vehicle 10 .

The second layer 200 may be a communication layer. In the second layer 200, the transceiver 201 receives BSM through a network cloud. In addition, the transceiver 201 may receive the latest Collective Perception messages (CPM) including information about the surrounding vehicles 20 from the surrounding vehicles 20 through the network cloud. Also, the transceiver 201 may receive an output of a local perception unit from the host vehicle 10 through a network cloud. The transceiver 201 parses the received data and transmits it to the dynamic object map layer 300 .

Here, the transceiver 201 communicates using Dedicated Short-Range Communication (DSRC) or Cellular-V2X (CV2X) scheme.

The third layer 300 may be a dynamic object map layer. In the third layer 300 , a DOM array (Document Object Model Array) 301 receives data from APIs of on-board sensors of the host vehicle 10 and surrounding vehicles 20 and from the transceiver 201 . The DOM array 301 communicates messages with the tracking process 302 .

The DOM array 301 transmits DOM data about surrounding vehicles to the target classifier 303 . DOM data is data generated in the DOM Array 301 and may include prediction or data fusion data.

The DOM array 301 receives data from the transceivers 201 and on-board sensor APIs of the host vehicle 10 and surrounding vehicles 20 in real time. Here, the data received by the DOM array 301 is composed of one or more of the BSM of the surrounding vehicle 20, the CPM of the surrounding vehicle 20, and a local recognition unit of the host vehicle, and may be expressed as a data package. Also, the data reception unit may be expressed as a packet.

The tracking process 302 keeps track of the history of surrounding vehicles 20 through the data received in the DOM array 301 . Here, the history of surrounding vehicles 20 may be a data package that the DOM array 301 most recently received.

If the next data package is not received after the latest data package, the DOM array 301 can predict the state of the surrounding vehicles 20 through the history of the surrounding vehicles 20 tracked in the tracking process 302 .

The DOM arrangement 301 can visualize the host vehicle 10 and the surrounding vehicles 20 .

The fourth layer may be a safety application layer. The fourth layer 400 includes a Forward Collision Warning (FCW) module 401, a Blind Spot Warning (BSW) module 402, Emergency Electronic Brake Lights, Hereinafter, a module for executing an application program related to vehicle safety, such as an EEBL module 403, is included.

The target classifier 303 may parse prediction information included in the received DOM data. The target classifier 303 allows parsing of data based on the output of the DOM array 301 . The target classifier 303 can trigger each of the modules 401 - 403 included in the safety application layer 400 based on a prediction instead of receiving new information about the state of the surrounding vehicle 20 .

The target classifier 303 maps the received DOM data to targets for each application program related to safety. The target classifier 303 transmits DOM data corresponding to each target to each module 401 to 403 included in the safety application layer 400 . For example, the target classifier 303 transmits DOM data corresponding to the FCW target to the FCW module 401 .

Each of the modules 401 to 403 included in the fourth layer 400 generates a message about a warning target of each module. The fourth layer 400 warns the host vehicle 10 using a message through CAN communication.

The vehicle state prediction system 1 may include the most recent BSM received from the surrounding vehicle 20, the most recent CPM containing information of the surrounding vehicle 20, or the most recent local recognition unit of the host vehicle 10. Based on the output, non-parametric Bayesian inference about the predicted state of the surrounding vehicle 20 may be performed.

FIG. 2 is a flowchart of a method for predicting a vehicle state of a DOM array of FIG. 1 according to an embodiment. 3 is an exemplary diagram for describing a history of a vehicle according to an exemplary embodiment. 4 is an exemplary diagram for explaining a model bank learned in one embodiment.

Referring to FIG. 2 , an embodiment in which a DOM array predicts the state of surrounding vehicles 20 through the history of the surrounding vehicles 20 tracked in the tracking process 302 is illustrated.

If the packet is not received, the DOM array 301 retrieves the most recent n points of the history of the surrounding vehicles 20 tracked from the tracking process 302 (S1). Here, n is an integer greater than or equal to 1. For example, n may be 5. Here, the point may indicate information received by the DOM array 301 .

The history of surrounding vehicles 20 tracked from the tracking process 302 may be represented similarly to the table of FIG. 3 . In the example of FIG. 3 , when n is 5, the most recent n points are points when t = t1, t3, t4, t5, and t8. At this time, s(t1), s(t3), s(t4), s(t5), and s(t8) are the speed data of the surrounding vehicles 20 when t = t1, t3, t4, t5, and t8 am. At this time, h(t1), h(t3), h(t4), h(t5), h(t8) are direction data of surrounding vehicles 20 when t = t1, t3, t4, t5, t8 am. The history of the surrounding vehicles 20 tracked in the tracking process 302 may include speed data and direction data of the surrounding vehicles 20 .

In the DOM array 301, a new model is created by inputting the most recent n points in an on-the-fly learning mode (S2). In the immediate learning mode, where n points, a new GP model is fit using a predefined kernel and learning window. In the immediate learning mode, n points are used to learn the kernel parameters of the GP.

Here, the new model may be a GPR model generated using GPR analysis.

In the example of FIG. 3 , the speed and direction of the surrounding vehicles 20 may be regarded as independent time series and modeled.

The DOM array 301 compares the new model with the latest model bank (S3). Here, the latest model bank may be a model bank learned by the immediate learning mode.

In the example of FIG. 4, the latest model bank may include velocity models (FIG. 4(a)) and direction models (FIG. 4(b)). The up-to-date model bank may contain other models for variables needed to predict the position of surrounding vehicles 20 . The models included in the latest model bank may include one or more GP models. Referring to FIG. 4 , 16 GP models such as gp ₁ , gp ₂ , gp ₃ , ... , and gp ₁₆ are included in the velocity models (FIG. 4(a)). In addition, direction models (FIG. 4(b)) include 16 GP models such as gp ₁ , gp ₂ , gp ₃ , ... , and gp ₁₆ .

The DOM array 301 compares the new model with the latest model bank according to the similarity criterion, and determines whether there is a similar model in the latest model bank.

The DOM array 301 selects a model with similarity if there is a model with similarity in the latest model bank, and if there is no model with similarity, selects a new model and adds the new model to the latest model bank ( S4). A similarity may be a similarity higher than a certain level.

The DOM array 301 predicts the RV location using the selected model (S5). The DOM array 301 also tracks the location of surrounding vehicles 20 using the selected model while no packets are being received.

The DOM array 301 may predict the direction and speed after the latest data packet of the surrounding vehicle 20 is received using the GP of the model selected in step S4. The DOM array 301 may indirectly estimate the state of the surrounding vehicle 20, such as the location of the surrounding vehicle 20, using an equation of motion. The state of the surrounding vehicle 20 may be estimated using predicted values of the direction, speed, and motion equation of the surrounding vehicle 20 .

In addition, the vehicle state prediction method of the DOM array 301 according to an embodiment may model and predict acceleration and yaw rate, and estimate the position, direction, and speed of the vehicle using vehicle dynamics. can do.

When a new packet is received, the DOM array 301 updates the history (S6). Here, the history may be the history of surrounding vehicles 20 . The DOM array 301 repeats steps S1-S5 while not receiving a packet following the new packet after step S6.

As such, one embodiment of the present invention is a system and method capable of predicting the position, speed, etc. of a vehicle after the latest data packet is received using a data bank of kernel parameters derived using GPR. A design can be made that uses the GPR method to design a method that learns the vehicle's behavior, trajectory, etc. for a finite period of time, and then determines how long a particular learned model remains valid before a new model is trained. Here, a valid state may mean given adjustable parameters.

Although the embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements made by those skilled in the art in the field to which the present invention belongs are also the rights of the present invention. belong to the range

1: Vehicle condition prediction system
10: host vehicle
20: surrounding vehicles
100: first layer
101: DSRC log
102: GPS
103 : BSM Emulator
200: second layer
201: transceiver
202: BSM
300: 3rd layer
301 : DOM Array
302: tracking process
303: target classifier
400: 4th layer
401: FCW module
402: BSW module
403: EEBL module

Claims (8)

A first layer generating data including any one or more of collective perception messages (CPM) and basic safety messages (BSM) of the vehicle;
a second layer receiving and parsing the data through a network;
After receiving the data, until new data is received, a model is created by learning the parsed data, and based on the model, a Gaussian Process (GP) is applied to determine the speed, direction, and location of the vehicle. a third layer predicting one or more; and
A vehicle condition prediction system comprising a fourth layer for generating a warning message by linking the predicted result with a safety application. According to claim 1,
The third layer,
a tracking process for tracking the history of the vehicle;
Creating a model by learning based on the most recent n points from the history of the vehicle, comparing the model with an up-to-date model bank containing one or more Gaussian process models, and predicting the state of the vehicle based on the selected model. DOM array to create DOM data; and
and a target classifier for parsing prediction information included in the DOM data, classifying it based on a target of a module included in the fourth layer, and transmitting the result to the fourth layer. According to claim 1,
The fourth layer,
An FCW module generating a forward collision warning message;
BSW module generating a blind spot warning message; and
A vehicle condition prediction system comprising at least one of an EEBL module generating an emergency brake electronic detection message. The DOM array includes the steps of calling the most recent n points of the vehicle's history;
In the DOM arrangement, generating a first model by learning kernel parameters of the GP based on the n points;
comparing the similarity of the first model with an up-to-date model bank including one or more Gaussian process models;
In the DOM arrangement, if there is a model similar to the first model in the latest model bank, selecting a model similar to the first model;
In the DOM arrangement, if there is no model similar to the first model in the latest model bank, selecting the first model; and
The vehicle state prediction method of claim 1 , wherein the DOM arrangement comprises predicting at least one of speed and direction of the vehicle using a Gaussian process based on the selected model. According to claim 4,
The vehicle state prediction method of claim 1 , wherein the DOM arrangement further comprises adding the first model to the latest model bank when the first model is selected. According to claim 4,
In the step of loading the n points,
In the DOM array, if a new packet is not received after the most recently received packet from the vehicle, the DOM array further includes the step of calling the most recent n points in the history of the vehicle through a tracking process, How to predict vehicle condition. According to claim 6,
The vehicle state prediction method of claim 1 , wherein the DOM arrangement further comprises updating a history of the vehicle based on the new packet if a new packet is received after the most recent packet received from the vehicle. According to claim 4,
The predicting step is
And predicting the position of the vehicle based on at least one of the predicted speed and direction of the vehicle and an equation of motion.

Images