WO2022222647A1 - 车辆意图预测方法、装置、设备及存储介质 - Google Patents

车辆意图预测方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2022222647A1
WO2022222647A1 PCT/CN2022/080971 CN2022080971W WO2022222647A1 WO 2022222647 A1 WO2022222647 A1 WO 2022222647A1 CN 2022080971 W CN2022080971 W CN 2022080971W WO 2022222647 A1 WO2022222647 A1 WO 2022222647A1
Authority
WO
WIPO (PCT)
Prior art keywords
vehicle
image information
model
target
path
Prior art date
Application number
PCT/CN2022/080971
Other languages
English (en)
French (fr)
Inventor
罗文�
韦俏杏
梁远桂
汪业栋
韦华超
卢潇泓
Original Assignee
东风柳州汽车有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 东风柳州汽车有限公司 filed Critical 东风柳州汽车有限公司
Publication of WO2022222647A1 publication Critical patent/WO2022222647A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
    • G05D1/00—Control of position, course, altitude or attitude of land, water, air or space vehicles, e.g. using automatic pilots
    • G05D1/02—Control of position or course in two dimensions
    • G05D1/021—Control of position or course in two dimensions specially adapted to land vehicles
    • G05D1/0231—Control of position or course in two dimensions specially adapted to land vehicles using optical position detecting means
    • G05D1/0246—Control of position or course in two dimensions specially adapted to land vehicles using optical position detecting means using a video camera in combination with image processing means
    • G05D1/0251—Control of position or course in two dimensions specially adapted to land vehicles using optical position detecting means using a video camera in combination with image processing means extracting 3D information from a plurality of images taken from different locations, e.g. stereo vision
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
    • G05D1/00—Control of position, course, altitude or attitude of land, water, air or space vehicles, e.g. using automatic pilots
    • G05D1/02—Control of position or course in two dimensions
    • G05D1/021—Control of position or course in two dimensions specially adapted to land vehicles
    • G05D1/0212—Control of position or course in two dimensions specially adapted to land vehicles with means for defining a desired trajectory
    • G05D1/0214—Control of position or course in two dimensions specially adapted to land vehicles with means for defining a desired trajectory in accordance with safety or protection criteria, e.g. avoiding hazardous areas
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
    • G05D1/00—Control of position, course, altitude or attitude of land, water, air or space vehicles, e.g. using automatic pilots
    • G05D1/02—Control of position or course in two dimensions
    • G05D1/021—Control of position or course in two dimensions specially adapted to land vehicles
    • G05D1/0212—Control of position or course in two dimensions specially adapted to land vehicles with means for defining a desired trajectory
    • G05D1/0221—Control of position or course in two dimensions specially adapted to land vehicles with means for defining a desired trajectory involving a learning process
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
    • G05D1/00—Control of position, course, altitude or attitude of land, water, air or space vehicles, e.g. using automatic pilots
    • G05D1/02—Control of position or course in two dimensions
    • G05D1/021—Control of position or course in two dimensions specially adapted to land vehicles
    • G05D1/0276—Control of position or course in two dimensions specially adapted to land vehicles using signals provided by a source external to the vehicle

Definitions

  • the present application relates to the technical field of automatic driving, and in particular, to a vehicle intention prediction method, apparatus, device, and storage medium.
  • autonomous driving technology is becoming more and more common, and the frequency of use is gradually increasing.
  • Autonomous vehicles are generally equipped with drones that can obtain image information to expand the perception range of autonomous vehicles and further improve the accuracy of monitoring the surrounding road parts of autonomous vehicles.
  • due to the vision of drones and vehicles Inconsistency there is a large difference in the image perception targets, so that the processing of the perception information by the vehicle processor will be delayed, so that the behavior of the autonomous vehicle cannot be accurately predicted, thus affecting the normal operation of the vehicle.
  • the main purpose of the present application is to provide a vehicle intention prediction method, device, device and storage medium, aiming to solve the technical problem that the behavior of an automatic driving vehicle cannot be accurately predicted in the prior art.
  • the application provides a vehicle intention prediction method, the method comprises the following steps:
  • the predicted behavior result of the target vehicle is obtained according to the vehicle intention model and the target behavior planning path.
  • the method before generating the planned path for the target behavior of the target vehicle by using the path planning model according to the first environment image information and the second environment image information, the method further includes:
  • the target behavior planning path of the target vehicle is generated by a path planning model, including:
  • a target behavior planning path of the target vehicle is generated by a path planning model according to the first environment image information and the second environment image information under the Frenet coordinates.
  • generating the target behavior planning path of the target vehicle through a path planning model according to the first environment image information and the second environment image information under the Frenet coordinates includes:
  • the target path is calculated by a preset prediction model, and a target behavior planning path of the target vehicle is generated.
  • the calculation of the target path through a preset prediction model to generate the target behavior planning path of the target vehicle includes:
  • a corresponding target behavior planning path is generated according to the path planning curve.
  • the method before obtaining the predicted behavior result of the target vehicle according to the vehicle intention model and the target behavior planning path, the method further includes:
  • the initial behavior prediction model is trained by the state feature, the behavior prediction, the insertion area and the insertion time to obtain a behavior prediction model
  • a vehicle intention model is obtained according to the behavior prediction model and the loss function.
  • the obtaining the vehicle intention model according to the behavior prediction model and the loss function includes:
  • a vehicle intent model is generated from the intent model that satisfies the prediction condition and the corresponding predicted probability.
  • the obtaining the predicted behavior result of the target vehicle according to the vehicle intention model and the target behavior planning path includes:
  • prediction results of the first intention model and the second intention model are the same, use the prediction result of the first intention model as the predicted behavior result of the target vehicle;
  • the prediction result of the second intention model is used as the predicted behavior result of the target vehicle.
  • the present application also proposes a vehicle intention prediction device, the vehicle intention prediction device includes:
  • an acquisition module configured to acquire environmental image information within a preset first range of the target vehicle to obtain first environmental image information
  • the acquiring module is further configured to acquire environmental image information within a preset second range of the target vehicle to obtain second environmental image information, wherein the preset second range is greater than the preset first range;
  • a generating module configured to generate a target behavior planning path of the target vehicle through a path planning model according to the first environment image information and the second environment image information;
  • the obtaining module is configured to obtain the predicted behavior result of the target vehicle according to the vehicle intention model and the planned path of the target behavior.
  • the present application also proposes a vehicle intention prediction device, the vehicle intention prediction device comprising: a memory, a processor, and a vehicle intention prediction device stored on the memory and executable on the processor A program configured to implement the vehicle intention prediction method as described above.
  • the present application further provides a storage medium, where a vehicle intention prediction program is stored thereon, and the vehicle intention prediction method as described above is implemented when the vehicle intention prediction program is executed by the processor.
  • the first environmental image information is obtained by acquiring the environmental image information within the preset first range of the target vehicle;
  • the second environmental image information is obtained by acquiring the environmental image information within the preset second range of the target vehicle, wherein the The preset second range is larger than the preset first range;
  • the target behavior planning path of the target vehicle is generated by the path planning model according to the first environment image information and the second environment image information; according to the vehicle intention model and the target behavior planning path to obtain the predicted behavior result of the target vehicle.
  • the behavior of the target vehicle is planned according to the environmental information within the preset range, and the method of integrating the behavior planning of the target vehicle and the vehicle intention is used to obtain the prediction of the vehicle behavior, and then predict the driving danger in advance, and further improve the automatic driving of the target vehicle. time security.
  • FIG. 1 is a schematic structural diagram of a vehicle intention prediction device of a hardware operating environment involved in a solution of an embodiment of the present application
  • FIG. 2 is a schematic flowchart of the first embodiment of the vehicle intention prediction method of the application
  • FIG. 3 is a schematic flowchart of the second embodiment of the vehicle intention prediction method of the application.
  • FIG. 4 is a schematic flowchart of the third embodiment of the vehicle intention prediction method of the application.
  • FIG. 5 is a structural block diagram of the first embodiment of the vehicle intention prediction apparatus of the present application.
  • FIG. 1 is a schematic structural diagram of a vehicle intention prediction device in a hardware operating environment according to an embodiment of the present application.
  • the vehicle intention prediction device may include: a processor 1001 , such as a central processing unit (Central Processing Unit, CPU), a communication bus 1002 , a user interface 1003 , a network interface 1004 , and a memory 1005 .
  • the communication bus 1002 is configured to realize the connection communication between these components.
  • the user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface.
  • the network interface 1004 may include a standard wired interface and a wireless interface (such as a wireless fidelity (WIreless-FIdelity, WI-FI) interface).
  • the memory 1005 may be a high-speed random access memory (Random Access Memory, RAM), or may be a stable non-volatile memory (Non-Volatile Memory, NVM), such as a disk memory.
  • RAM Random Access Memory
  • NVM Non-Volatile Memory
  • the memory 1005 may also be a storage device independent of the aforementioned processor 1001 .
  • FIG. 1 does not constitute a limitation on the vehicle intention prediction device, and may include more or less components than the one shown, or combine some components, or arrange different components.
  • the memory 1005 as a storage medium may include an operating system, a network communication module, a user interface module, and a vehicle intention prediction program.
  • the network interface 1004 is mainly configured to perform data communication with the network server; the user interface 1003 is mainly configured to perform data interaction with the user; the processor 1001 in the vehicle intention prediction device of the present application
  • the memory 1005 may be set in the vehicle intention prediction device, and the vehicle intention prediction device invokes the vehicle intention prediction program stored in the memory 1005 through the processor 1001, and executes the vehicle intention prediction method provided by the embodiment of the present application.
  • FIG. 2 is a schematic flowchart of a first embodiment of a method for predicting vehicle intent in the present application.
  • the vehicle intention prediction method includes the following steps:
  • Step S10 Obtaining the environmental image information within the preset first range of the target vehicle to obtain the first environmental image information.
  • the executive body of this embodiment is a controller of a vehicle that can realize automatic driving, or other devices that can realize the same function, which is not limited in this embodiment.
  • the environmental image information within the preset first range is collected by a camera installed on the target vehicle for capturing the front of the vehicle and the surrounding environment, which includes road conditions around the target vehicle and surrounding vehicles.
  • the preset first range refers to the farthest range that can be captured by the vehicle-mounted camera.
  • Step S20 Obtaining environmental image information within a preset second range of the target vehicle to obtain second environmental image information, wherein the preset second range is larger than the preset first range.
  • the environmental image information within the preset second range is collected by a camera installed on the vehicle-mounted drone for capturing information around the vehicle and the road surface, which includes road information around the target vehicle and more information. Information about surrounding vehicles within a long range.
  • the preset second range refers to the farthest range that can be captured by the camera on the vehicle-mounted drone.
  • the drone camera has a wider field of view, it can perceive image information that the vehicle camera cannot perceive. For example, the lane line information and moving vehicle information farther from the vehicle, when there is a large truck in front of the target vehicle, the large truck blocks the field of view of the on-board camera of the target vehicle, and the camera on the on-board drone can collect more comprehensive information .
  • the vehicle-mounted drone enables the follow-up shooting mode, the position is corrected in real time, moves with the target vehicle, and remains directly above the vehicle. At the same time, the time of the vehicle-mounted drone is synchronized with the time of the target vehicle.
  • Step S30 Generate a target behavior planning path of the target vehicle through a path planning model according to the first environment image information and the second environment image information.
  • the image information collected by the on-board camera and the drone camera is combined, and the combined image The information further generates the optimal path curve of the target vehicle through the path planning model, so as to obtain the target behavior planning path of the target vehicle.
  • Step S40 Obtain the predicted behavior result of the target vehicle according to the vehicle intention model and the target behavior planning path.
  • vehicle intention model refers to a behavior prediction model and prediction probability obtained based on surrounding vehicles.
  • a prediction result corresponding to the vehicle prediction path consistent with the target behavior planning path is found in the vehicle intention model as the predicted behavior result of the target vehicle.
  • the first environmental image information is obtained by acquiring the environmental image information within a preset first range of the target vehicle; the second environmental image information is obtained by acquiring the environmental image information within the preset second range of the target vehicle, wherein, The preset second range is larger than the preset first range; the target behavior planning path of the target vehicle is generated by the path planning model according to the first environment image information and the second environment image information; according to the vehicle intention The model and the target behavior planning path obtain the predicted behavior result of the target vehicle.
  • the behavior of the target vehicle is planned according to the environmental information within the preset range, and the method of integrating the behavior planning of the target vehicle and the vehicle intention is used to obtain the prediction of the vehicle behavior, and then predict the driving danger in advance, and further improve the automatic driving of the target vehicle. time security.
  • FIG. 3 is a schematic flowchart of a second embodiment of a vehicle intention-based prediction method according to the present application.
  • the method for predicting vehicle intention based on the present embodiment further includes:
  • Step S301 Perform time synchronization matching on the first environment image information and the second environment image information to obtain the first environment image information and the second environment image information corresponding to the synchronization time.
  • Step S302 Perform coordinate transformation on the first environment image information and the second environment image information corresponding to the synchronization time to obtain the first environment image information and the second environment image information under Frenet coordinates.
  • the controller needs to convert the world coordinate system in the two images into the Frenet coordinate system to obtain the converted image information.
  • Step S30' Generate the target behavior planning path of the target vehicle through the path planning model according to the first environment image information and the second environment image information under the Frenet coordinates.
  • the controller can obtain the target behavior planning path of the target vehicle by processing the images using the path planning model.
  • generating the target behavior planning path of the target vehicle through a path planning model according to the first environment image information and the second environment image information under the Frenet coordinates includes: according to the first environment image under the Frenet coordinates information and the second environment image information to obtain the surrounding environment information of the target vehicle; based on the surrounding environment information, plan a target path for the target vehicle to travel on a preset road section, wherein the preset road section is based on the target vehicle The preset distance in front of the current position is determined; the target path is calculated by a preset prediction model to generate the target behavior planning path of the target vehicle.
  • the path planning model is obtained by training based on a path planning algorithm, and the target path that the target target vehicle travels on the preset road section refers to the target path within a preset distance.
  • the target behavior planning path is generated according to the target path within the preset distance, so that the target vehicle can travel more smoothly and safely.
  • calculating the target path by using a preset prediction model to generate the target behavior planning path of the target vehicle includes: obtaining discrete path points according to the target path; The discrete path points are fitted to obtain the local behavior planning path of the target vehicle; the local behavior planning path is calculated to obtain the cost function value of the local behavior planning path; when the cost function value tends to a preset value
  • a path planning curve corresponding to a preset value of the cost function value is obtained; and a corresponding target behavior planning path is generated according to the path planning curve.
  • the target path refers to the planned reference path within a preset distance.
  • the target path is divided into several sections, and the discrete scatter points, that is, discrete path points, are obtained based on the section points.
  • the partial path is obtained by fitting.
  • the cost function value of the local behavior planning path can be obtained, and when the cost function value approaches the preset value, the target behavior planning path of the target vehicle can be obtained.
  • the target behavior planning path obtained by the fitting calculation in this embodiment is more accurate.
  • the target behavior planning path intercept 50 meters ahead of the target vehicle as the reference path L b of the current target vehicle, divide the reference path L b into 5 equal segments, and at each segment point the corresponding Frenet coordinate system
  • P total (s) is the cost function value of the local path curve s
  • P i (s) is the factor that affects the path curve.
  • the factors that affect the local path curve in this embodiment include the overall smoothness of the target vehicle, and the obstacles. The distance to the target, the deviation from the lane line, and the deviation from the slope of the trajectory origin.
  • the local path cost function is the minimum value
  • the optimal path curve s of the driverless vehicle is obtained, thereby generating the corresponding target behavior planning path.
  • the first environment image information and the second environment image information corresponding to the synchronization time are obtained by performing time synchronization matching on the first environment image information and the second environment image information; Coordinate transformation is performed between the first environment image information and the second environment image information to obtain the first environment image information and the second environment image information under Frenet coordinates; according to the first environment image information and the second environment image information, through the path planning
  • the model generating the target behavior planning path of the target vehicle includes: generating the target behavior planning path of the target vehicle through a path planning model according to the first environment image information and the second environment image information under the Frenet coordinates.
  • FIG. 4 is a schematic flowchart of a third embodiment of a vehicle intention-based prediction method according to the present application.
  • the method for predicting vehicle intention based on the present embodiment further includes:
  • Step S401 Identify moving vehicles and road information within a preset second range of the target vehicle according to the second environment image information.
  • a vehicle identification algorithm is used to identify vehicles within a certain distance in front of the target vehicle, for example, vehicles within 150 meters in front of the target vehicle are identified. , to obtain information about moving vehicles and vehicles within a preset second range of the target vehicle.
  • the lane line recognition algorithm is used to identify the lane lines of the two lanes around the target vehicle, and obtain the lane line information around the target vehicle.
  • Step S402 Obtain the state feature and behavior prediction of the moving vehicle.
  • the state feature of the moving vehicle around the target vehicle that is, the input state feature x
  • the behavior prediction of the moving vehicle can be obtained.
  • Step S403 Acquire an insertion area and a corresponding insertion time of the moving vehicle based on the road surface information.
  • the insertion area and insertion position of the moving vehicle in front of the target vehicle are acquired based on the road surface information in the second environment information.
  • Step S404 Train an initial behavior prediction model by using the state feature, the behavior prediction, the insertion area, and the insertion time to obtain a behavior prediction model.
  • the initial behavior prediction model is trained after the state feature, behavior prediction, insertion area and insertion time are obtained to obtain the behavior prediction model.
  • the initial behavior prediction model is in represents the behavior prediction of the k-th moving vehicle around the target vehicle
  • x represents the input state characteristics of the moving vehicle
  • C s represents the insertion area s of the k-th moving vehicle
  • C t represents the insertion time t of the k-th moving vehicle
  • ⁇ k (x) represents the Gaussian coefficient, represents the mean value of the insertion area and insertion time of the k-th moving vehicle
  • Step S405 Obtain a loss function corresponding to the behavior prediction model based on the behavior prediction model.
  • W 1 and W 2 are parameters, the parameter values are adjusted according to the actual input, S is the total number of inserted regions, represents the truth value of the insertion area s of the current k-th moving vehicle.
  • Step S406 Obtain a vehicle intention model according to the behavior prediction model and the loss function.
  • the surrounding vehicle intention model can be obtained.
  • the obtaining the vehicle intention model according to the behavior prediction model and the loss function includes: obtaining the intention model and corresponding prediction probability that meet the prediction conditions according to the behavior prediction model and the loss function; The intent model that satisfies the prediction condition and the corresponding predicted probability are used to generate the vehicle intent model.
  • the image information collected by the vehicle-mounted UAV camera is limited, and the vehicle intent model needs to be trained in real time based on different surrounding environments and moving vehicles. At the same time, because there are many moving vehicles, different moving vehicles have different behaviors, and the predicted behaviors are also different, so there are also multiple intent models.
  • the loss function can make the real inserted area have the largest weight, and at the same time, perform probability prediction on the time and position information of the moving vehicle inserted into the area, output multiple intent models and the corresponding prediction probability C k,p , and also The highest predicted probability and the corresponding intent model C k,max can be obtained.
  • the vehicle intention model is generated from the intention model and the corresponding prediction probability within the predetermined range of the image that can be collected by the vehicle-mounted UAV camera that meets the prediction conditions.
  • the vehicle intention model is more accurate, and the corresponding probability can be obtained.
  • obtaining the predicted behavior result of the target vehicle according to the vehicle intention model and the target behavior planning path includes: obtaining a first intention model corresponding to the largest predicted probability in the vehicle intention model; obtaining the vehicle intention A path intent model whose predicted path in the model is consistent with the target behavior planning path; obtains a second intent model corresponding to the largest predicted probability in the path intent model; if the predictions of the first intent model and the second intent model If the results are the same, the prediction result of the first intention model is used as the predicted behavior result of the target vehicle; if the prediction results of the first intention model and the second intention model are different, the second intention model is used as the prediction result.
  • the predicted result is taken as the predicted behavior result of the target vehicle.
  • the predicted behavior with the largest predicted behavior probability in the vehicle intention model is output as the predicted behavior result. If the predicted path with the largest predicted behavior probability in the vehicle intention model is inconsistent with the local behavior planning path in the target vehicle target behavior planning path, the output corresponding predicted path in the intention model in the vehicle intention model and the target vehicle target behavior planning path
  • the prediction result corresponding to the intention model with the largest prediction probability among the intention models that are consistent with the local behavior planning path is the predicted behavior result of the target vehicle.
  • the moving vehicle and road surface information within a preset second range are identified by the target vehicle according to the second environment image information; state characteristics and behavior prediction of the moving vehicle are obtained; and the moving vehicle is obtained based on the road surface information
  • the vehicle intention model is generated by the behavior prediction algorithm, which solves the problem of decision-making delay caused by the transmission time of data and the calculation time of the controller.
  • an embodiment of the present application further proposes a vehicle intention prediction device, where the vehicle intention prediction device includes:
  • the acquiring module 10 is configured to acquire the environmental image information within the preset first range of the target vehicle to obtain the first environmental image information;
  • the acquiring module 10 is further configured to acquire environmental image information within a preset second range of the target vehicle, so as to obtain second environmental image information, wherein the preset second range is greater than the preset first range;
  • the generating module 20 is configured to generate a target behavior planning path of the target vehicle through a path planning model according to the first environment image information and the second environment image information;
  • the obtaining module 30 is configured to obtain the predicted behavior result of the target vehicle according to the vehicle intention model and the planned path of the target behavior.
  • the first environmental image information is obtained by acquiring the environmental image information within a preset first range of the target vehicle; the second environmental image information is obtained by acquiring the environmental image information within the preset second range of the target vehicle, wherein, The preset second range is larger than the preset first range; the target behavior planning path of the target vehicle is generated by the path planning model according to the first environment image information and the second environment image information; according to the vehicle intention The model and the target behavior planning path obtain the predicted behavior result of the target vehicle.
  • the behavior of the target vehicle is planned according to the environmental information within the preset range, and the method of integrating the behavior planning of the target vehicle and the vehicle intention is used to obtain the prediction of the vehicle behavior, and then predict the driving danger in advance, and further improve the automatic driving of the target vehicle. time security.
  • the generating module 20 is further configured to perform time synchronization matching on the first environment image information and the second environment image information to obtain the first environment image information and the second environment image information corresponding to the synchronization time.
  • environmental image information ;
  • the first environment image information and the second environment image information corresponding to the synchronization time are carried out coordinate conversion, and the first environment image information and the second environment image information under the Frenet coordinates are obtained;
  • the target behavior planning path of the target vehicle is generated by a path planning model, including:
  • a target behavior planning path of the target vehicle is generated by a path planning model according to the first environment image information and the second environment image information under the Frenet coordinates.
  • the generating module 20 is further configured to obtain the surrounding environment information of the target vehicle according to the first environment image information and the second environment image information under the Frenet coordinates;
  • the target path is calculated by a preset prediction model, and a target behavior planning path of the target vehicle is generated.
  • the generating module 20 is further configured to obtain discrete path points according to the target path;
  • a corresponding target behavior planning path is generated according to the path planning curve.
  • the obtaining module 30 is further configured to identify moving vehicles and road information within a preset second range of the target vehicle according to the second environment image information;
  • the initial behavior prediction model is trained by the state feature, the behavior prediction, the insertion area and the insertion time to obtain a behavior prediction model
  • a vehicle intention model is obtained according to the behavior prediction model and the loss function.
  • the obtaining module 30 is further configured to obtain, according to the behavior prediction model and the loss function, an intention model that satisfies the prediction condition and a corresponding prediction probability;
  • a vehicle intent model is generated from the intent model that satisfies the prediction condition and the corresponding predicted probability.
  • the obtaining module 30 is further configured to obtain a first intention model corresponding to the largest predicted probability in the vehicle intention model
  • prediction results of the first intention model and the second intention model are the same, use the prediction result of the first intention model as the predicted behavior result of the target vehicle;
  • the prediction result of the second intention model is used as the predicted behavior result of the target vehicle.
  • an embodiment of the present application also provides a storage medium, where a vehicle intention prediction program is stored thereon, and when the vehicle intention prediction program is executed by a processor, the steps of the vehicle intention prediction method as described above are implemented.
  • the storage medium adopts all the technical solutions of all the above-mentioned embodiments, it has at least all the functions brought by the technical solutions of the above-mentioned embodiments, and will not be repeated here.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Aviation & Aerospace Engineering (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Remote Sensing (AREA)
  • General Physics & Mathematics (AREA)
  • Automation & Control Theory (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Multimedia (AREA)
  • Electromagnetism (AREA)
  • Traffic Control Systems (AREA)

Abstract

本申请属于自动驾驶技术领域，公开了一种车辆意图预测方法、装置、设备及存储介质。该方法包括：获取目标车辆预设第一范围内的环境图像信息，以得到第一环境图像信息；获取目标车辆预设第二范围内的环境图像信息，以得到第二环境图像信息，其中，所述预设第二范围大于所述预设第一范围；根据所述第一环境图像信息和所述第二环境图像信息通过路径规划模型生成所述目标车辆的目标行为规划路径；根据车辆意图模型和所述目标行为规划路径得到所述目标车辆的预测行为结果。

Description

车辆意图预测方法、装置、设备及存储介质
相关申请
本申请要求于2021年4月21号申请的、申请号为202110429099.7的中国专利申请的优先权，其全部内容通过引用结合于此。
技术领域
本申请涉及自动驾驶技术领域，尤其涉及一种车辆意图预测方法、装置、设备及存储介质。
背景技术
随着科技的发展，自动驾驶技术越来越普遍，使用频率逐步提高。自动驾驶车上一般会带有可获取图像信息的无人机，以扩大自动驾驶车辆的感知范围，进一步提高了对自动驾驶车周边道路部分监测的准确性，但是由于无人机和车辆的视野不一致，图像感知目标存在较大差异，从而使车辆处理器对于感知信息的处理会有延迟，导致自动驾驶车辆的行为不能够被准确预测，从而影响了车辆的正常运行。
上述内容仅用于辅助理解本申请的技术方案，并不代表承认上述内容是现有技术。
申请内容
本申请的主要目的在于提供一种车辆意图预测方法、装置、设备及存储介质，旨在解决现有技术自动驾驶车辆行为不能被准确预测的技术问题。
为实现上述目的，本申请提供了一种车辆意图预测方法，所述方法包括以下步骤:
获取目标车辆预设第一范围内的环境图像信息，以得到第一环境图像信息；
获取目标车辆预设第二范围内的环境图像信息，以得到第二环境图像信息，其中，所述预设第二范围大于所述预设第一范围；
根据所述第一环境图像信息和所述第二环境图像信息通过路径规划模型 生成所述目标车辆的目标行为规划路径；
根据车辆意图模型和所述目标行为规划路径得到所述目标车辆的预测行为结果。
在一实施方式中，所述根据所述第一环境图像信息和所述第二环境图像信息通过路径规划模型生成所述目标车辆的目标行为规划路径之前，还包括：
将所述第一环境图像信息和所述第二环境图像信息进行时间同步匹配，得到同步时间对应的第一环境图像信息和第二环境图像信息；
将所述同步时间对应的第一环境图像信息和第二环境图像信息进行坐标转换，得到Frenet坐标下的第一环境图像信息和第二环境图像信息；
根据所述第一环境图像信息和所述第二环境图像信息通过路径规划模型生成所述目标车辆的目标行为规划路径，包括：
根据所述Frenet坐标下的第一环境图像信息和第二环境图像信息通过路径规划模型生成所述目标车辆的目标行为规划路径。
在一实施方式中，所述根据所述Frenet坐标下的第一环境图像信息和第二环境图像信息通过路径规划模型生成所述目标车辆的目标行为规划路径，包括：
根据所述Frenet坐标下第一环境图像信息和所述第二环境图像信息得到目标车辆的周边环境信息；
基于所述周边环境信息规划所述目标车辆在预设路段行驶的目标路径，其中，所述预设路段为根据所述目标车辆当前所处位置前方预设距离确定的；
对所述目标路径通过预设预测模型进行计算，生成所述目标车辆的目标行为规划路径。
在一实施方式中，所述对所述目标路径通过预设预测模型进行计算，生成所述目标车辆的目标行为规划路径，包括：
根据所述目标路径得到离散路径点；
通过预设拟合模型对所述离散路径点进行拟合，以得到所述目标车辆的局部行为规划路径；
对所述局部行为规划路径进行计算，得到局部行为规划路径的代价函数值；
当所述代价函数值趋于预设值时，得到代价函数值为预设值对应的路径 规划曲线；
根据所述路径规划曲线生成对应的目标行为规划路径。
在一实施方式中，所述根据车辆意图模型和所述目标行为规划路径得到所述目标车辆的预测行为结果之前，还包括：
根据所述第二环境图像信息识别目标车辆预设第二范围内的移动车辆及路面信息；
获取所述移动车辆的状态特征及行为预测；
基于所述路面信息获取所述移动车辆的插入区域及对应的插入时间；
通过所述状态特征、所述行为预测、所述插入区域及所述插入时间对初始行为预测模型进行训练，以得到行为预测模型；
基于所述行为预测模型得到所述行为预测模型对应的损失函数；
根据所述行为预测模型和所述损失函数得到车辆意图模型。
在一实施方式中，所述根据所述行为预测模型和所述损失函数得到车辆意图模型，包括：
根据所述行为预测模型和所述损失函数，得到满足预测条件的意图模型和对应的预测概率；
将所述满足预测条件的的意图模型和对应的预测概率生成车辆意图模型。
在一实施方式中，所述根据车辆意图模型和所述目标行为规划路径得到所述目标车辆的预测行为结果，包括：
获取所述车辆意图模型中预测概率最大对应的第一意图模型；
获取所述车辆意图模型中预测路径和所述目标行为规划路径一致的路径意图模型；
获取所述路径意图模型中预测概率最大对应的第二意图模型；
若所述第一意图模型和所述第二意图模型的预测结果相同，则将所述第一意图模型的预测结果作为目标车辆的预测行为结果；
若所述第一意图模型和所述第二意图模型的预测结果不相同，则将所述第二意图模型的预测结果作为目标车辆的预测行为结果。
此外，为实现上述目的，本申请还提出一种车辆意图预测装置，所述车 辆意图预测装置包括：
获取模块，被配置为获取目标车辆预设第一范围内的环境图像信息，以得到第一环境图像信息；
所述获取模块，还被配置为获取目标车辆预设第二范围内的环境图像信息，以得到第二环境图像信息，其中，所述预设第二范围大于所述预设第一范围；
生成模块，被配置为根据所述第一环境图像信息和所述第二环境图像信息通过路径规划模型生成所述目标车辆的目标行为规划路径；
得到模块，被配置为根据车辆意图模型和所述目标行为规划路径得到所述目标车辆的预测行为结果。
此外，为实现上述目的，本申请还提出一种车辆意图预测设备，所述车辆意图预测设备包括：存储器、处理器及存储在所述存储器上并可在所述处理器上运行的车辆意图预测程序，所述车辆意图预测程序配置为实现如上文所述的车辆意图预测方法。
此外，为实现上述目的，本申请还提出一种存储介质，所述存储介质上存储有车辆意图预测程序，所述车辆意图预测程序被处理器执行时实现如上文所述的车辆意图预测方法。
本申请通过获取目标车辆预设第一范围内的环境图像信息，以得到第一环境图像信息；获取目标车辆预设第二范围内的环境图像信息，以得到第二环境图像信息，其中，所述预设第二范围大于所述预设第一范围；根据所述第一环境图像信息和所述第二环境图像信息通过路径规划模型生成所述目标车辆的目标行为规划路径；根据车辆意图模型和所述目标行为规划路径得到所述目标车辆的预测行为结果。通过上述方式，根据预设范围内的环境信息对目标车辆进行行为规划，采用目标车辆行为规划和车辆意图融合的方法，得到车辆行为的预测，进而提前预测行车危险，进一步提高目标车辆在自动驾驶时的安全性。
附图说明
图1是本申请实施例方案涉及的硬件运行环境的车辆意图预测设备的结构示意图；
图2为本申请车辆意图预测方法第一实施例的流程示意图；
图3为本申请车辆意图预测方法第二实施例的流程示意图；
图4为本申请车辆意图预测方法第三实施例的流程示意图；
图5为本申请车辆意图预测装置第一实施例的结构框图。
本申请目的的实现、功能特点及优点将结合实施例，参照附图做进一步说明。
具体实施方式
应当理解，此处所描述的具体实施例仅用以解释本申请，并不用于限定本申请。
参照图1，图1为本申请实施例方案涉及的硬件运行环境的车辆意图预测设备的结构示意图。
如图1所示，该车辆意图预测设备可以包括：处理器1001，例如中央处理器(Central Processing Unit，CPU)，通信总线1002、用户接口1003，网络接口1004，存储器1005。其中，通信总线1002被配置为实现这些组件之间的连接通信。用户接口1003可以包括显示屏(Display)、输入单元比如键盘(Keyboard)，可选用户接口1003还可以包括标准的有线接口、无线接口。网络接口1004可选的可以包括标准的有线接口、无线接口(如无线保真(WIreless-FIdelity，WI-FI)接口)。存储器1005可以是高速的随机存取存储器(Random Access Memory，RAM)，也可以是稳定的非易失性存储器(Non-Volatile Memory，NVM)，例如磁盘存储器。存储器1005可选的还可以是独立于前述处理器1001的存储装置。
本领域技术人员可以理解，图1中示出的结构并不构成对车辆意图预测设备的限定，可以包括比图示更多或更少的部件，或者组合某些部件，或者不同的部件布置。
如图1所示，作为一种存储介质的存储器1005中可以包括操作系统、网 络通信模块、用户接口模块以及车辆意图预测程序。
在图1所示的车辆意图预测设备中，网络接口1004主要被配置为与网络服务器进行数据通信；用户接口1003主要被配置为与用户进行数据交互；本申请车辆意图预测设备中的处理器1001、存储器1005可以设置在车辆意图预测设备中，所述车辆意图预测设备通过处理器1001调用存储器1005中存储的车辆意图预测程序，并执行本申请实施例提供的车辆意图预测方法。
本申请实施例提供了一种车辆意图预测方法，参照图2，图2为本申请一种车辆意图预测方法第一实施例的流程示意图。
本实施例中，所述车辆意图预测方法包括以下步骤：
步骤S10：获取目标车辆预设第一范围内的环境图像信息，以得到第一环境图像信息。
需要理解的是，本实施例的执行主体为可实现自动驾驶的车辆的控制器，或其他能够实现相同功能的设备，本实施例对此不加以限制。
可以理解的是，所述预设第一范围内的环境图像信息是通过安装于目标车辆上的用于摄取车辆前方及周边环境的摄像头采集的，它包括了目标车辆周边的路况及周边车辆。所述预设第一范围指的是车载摄像头所能够采集到的最远范围。
步骤S20：获取目标车辆预设第二范围内的环境图像信息，以得到第二环境图像信息，其中，所述预设第二范围大于所述预设第一范围。
需要说明的是，所述预设第二范围内的环境图像信息是通过安装于车载无人机上的用于摄取车辆周边及路面信息的摄像头采集的，它包括了目标车辆周边的路面信息和更远范围内的周边车辆信息。所述预设第二范围指的是车载无人机上的摄像头所能够采集到的最远范围。
可以理解的是，因为无人机摄像头的视野范围更广，能够感知车载摄像头感知不到的图像信息。例如车辆更远处的车道线信息和移动车辆信息，当目标车辆前方有一辆大货车时，大货车挡住了目标车辆车载摄像头的视野，而车载无人机上的摄像头能够采集到更为全面的信息。
在具体实现中，车载无人机启用跟随拍摄模式，位置实时校正，跟随目标车辆移动，保持位于车辆正上方，同时，车载无人机自带的时间和目标车辆的时间同步。
步骤S30：根据所述第一环境图像信息和所述第二环境图像信息通过路径规划模型生成所述目标车辆的目标行为规划路径。
需要说明的是，在目标车辆的车载摄像头和目标车辆的车载无人机上的摄像头采集到周边的环境信息后，将车载摄像头和无人机摄像头采集到的图像信息进行结合，根据结合后的图像信息通过路径规划模型进一步生成目标车辆的最优路径曲线，从而得到目标车辆的目标行为规划路径。
步骤S40：根据车辆意图模型和所述目标行为规划路径得到所述目标车辆的预测行为结果。
需要说明的是，所述车辆意图模型指的是基于周边车辆所得到的行为预测模型和预测概率。
在具体实现中，在得到目标车辆的目标行为规划路径后，在车辆意图模型中寻找与目标行为规划路径一致的车辆预测路径对应的预测结果作为目标车辆的预测行为结果。
本实施例通过获取目标车辆预设第一范围内的环境图像信息，以得到第一环境图像信息；获取目标车辆预设第二范围内的环境图像信息，以得到第二环境图像信息，其中，所述预设第二范围大于所述预设第一范围；根据所述第一环境图像信息和所述第二环境图像信息通过路径规划模型生成所述目标车辆的目标行为规划路径；根据车辆意图模型和所述目标行为规划路径得到所述目标车辆的预测行为结果。通过上述方式，根据预设范围内的环境信息对目标车辆进行行为规划，采用目标车辆行为规划和车辆意图融合的方法，得到车辆行为的预测，进而提前预测行车危险，进一步提高目标车辆在自动驾驶时的安全性。
参照图3，图3为本申请一种基于车辆意图预测方法第二实施例的流程示意图。
基于上述第一实施例，本实施例基于车辆意图预测方法在所述步骤S30之前，还包括：
步骤S301：将所述第一环境图像信息和所述第二环境图像信息进行时间同步匹配，得到同步时间对应的第一环境图像信息和第二环境图像信息。
需要说明的是，在图像信息采集时，需要将目标车辆车载摄像头采集到 的第一环境图像信息和目标车辆车载无人机摄像头采集到的第二环境图像信息的时间进行同步传输匹配。
步骤S302：将所述同步时间对应的第一环境图像信息和第二环境图像信息进行坐标转换，得到Frenet坐标下的第一环境图像信息和第二环境图像信息。
可以理解的是，在将第一环境图像信息和第二环境图像信息同步传输到控制器后，控制器需要将二者图像中的世界坐标系转换为Frenet坐标系，得到转换后的图像信息。
在具体实现中，将世界坐标系转换为摄像头坐标系转换为摄像头坐标系有
Figure PCTCN2022080971-appb-000001
其中，有世界坐标系(X W,Y W,Z W)，摄像头坐标系(X C,Y C,Z C),R为3×3正交单位矩阵，T为三维平移向量，摄像头坐标系转换为图像坐标系有
Figure PCTCN2022080971-appb-000002
其中，有图像坐标系(X,Y),f为摄像头焦距，在图像坐标系中设有一元三次方程拟合路径方程Y＝aX 3+bX 2+cX+d，图像坐标系转换为Frenet坐标系有：
Figure PCTCN2022080971-appb-000003
步骤S30’：根据所述Frenet坐标下的第一环境图像信息和第二环境图像信息通过路径规划模型生成所述目标车辆的目标行为规划路径。
需要说明的是，在得到Frenet坐标系下的第一环境图像信息和第二环境图像信息，控制器通过利用路径规划模型对图像进行处理，可以得到目标车辆的目标行为规划路径。
进一步地，所述根据所述Frenet坐标下的第一环境图像信息和第二环境图像信息通过路径规划模型生成所述目标车辆的目标行为规划路径，包括：根据所述Frenet坐标下第一环境图像信息和所述第二环境图像信息得到目标 车辆的周边环境信息；基于所述周边环境信息规划所述目标车辆在预设路段行驶的目标路径，其中，所述预设路段为根据所述目标车辆当前所处位置前方预设距离确定的；对所述目标路径通过预设预测模型进行计算，生成所述目标车辆的目标行为规划路径。
需要说明的是，所述路径规划模型是基于路径规划算法训练得到的，所述目标目标车辆在预设路段行驶的目标路径指的是在预设一段距离内的目标路径。在本实施例中，根据预设距离内的目标路径生成目标行为规划路径，目标车辆到的行驶更为平稳和安全。
进一步地，所述对所述目标路径通过预设预测模型进行计算，生成所述目标车辆的目标行为规划路径，包括：根据所述目标路径得到离散路径点；通过预设拟合模型对所述离散路径点进行拟合，以得到所述目标车辆的局部行为规划路径；对所述局部行为规划路径进行计算，得到局部行为规划路径的代价函数值；当所述代价函数值趋于预设值时，得到代价函数值为预设值对应的路径规划曲线；根据所述路径规划曲线生成对应的目标行为规划路径。
可以理解的是，目标路径指的是在预设一段距离内的规划参考路径，将目标路径等分为几段，基于分段点得到分离散点，即离散路径点，最后通过拟合得到局部行为规划路径，利用预设的计算模型，可以得到局部行为规划路径的代价函数值，在代价函数值趋近于预设值时，得到目标车辆的目标行为规划路径。本实施例通过拟合计算得到的目标行为规划路径更为准确。
例如，在目标行为规划路径的结果上截取目标车辆前方50米作为当前目标车辆的参考路径L b，将参考路径L b等分为5段，在每个分段点处将对应Frenet坐标系上的坐标点沿纵坐标轴生成等5个分离散点，构成离散路径点集合Q＝{P 1i,P 2i,P 3i,P 4i,P 5i}，其中 i表示等分离散点的第 i个，使用一元三次方程式对离散路径点集合进行拟合，得到Frenet坐标系上的局部路径曲线s。有局部路径代价函数方程
Figure PCTCN2022080971-appb-000004
其中P total(s)为局部路径曲线s的代价函数值，P i(s)为影响路径曲线的因素，其中本实施例设定影响局部路径曲线的因素有目标车辆整体平顺性、与障碍物目标的距离、与车道线的偏离度、与轨迹原点斜率的偏离量。当局部路径代价函数最小值时，得到无人驾驶车最优路径曲线s，从而生成对应的目标行为规划路径。
本实施例通过将所述第一环境图像信息和所述第二环境图像信息进行时 间同步匹配，得到同步时间对应的第一环境图像信息和第二环境图像信息；将所述同步时间对应的第一环境图像信息和第二环境图像信息进行坐标转换，得到Frenet坐标下的第一环境图像信息和第二环境图像信息；根据所述第一环境图像信息和所述第二环境图像信息通过路径规划模型生成所述目标车辆的目标行为规划路径，包括：根据所述Frenet坐标下的第一环境图像信息和第二环境图像信息通过路径规划模型生成所述目标车辆的目标行为规划路径。通过车载摄像头和车载无人机摄像头的图像结合，解决了目标车辆视野差，无人机视野广但实时性差的问题，同时二者摄像头传感器均是感知当前时刻的信息，处理器通过感知信息实时做出判断，使后续的决策更为及时准确。
参照图4，图4为本申请一种基于车辆意图预测方法第三实施例的流程示意图。
基于上述第一实施例，本实施例基于车辆意图预测方法在所述步骤S40之前，还包括：
步骤S401：根据所述第二环境图像信息识别目标车辆预设第二范围内的移动车辆及路面信息。
需要说明的是，基于车载无人机的摄像头采集到的第二环境图像信息采用车辆识别算法对目标车辆前方一段距离范围内的车辆进行识别，例如对目标车辆前方150米范围内的车辆进行识别，得到目标车辆预设第二范围内的移动车辆及车辆的信息。
可以理解的是，基于车载无人机的摄像头采集到的第二环境图像信息采用车道线识别算法对目标车辆周边两条车道的车道线进行识别，得到目标车辆周边的车道线信息。
步骤S402：获取所述移动车辆的状态特征及行为预测。
需要说明的是，根据第二环境图像信息可以获取目标车辆周边移动车辆的状态特征，即输入状态特征x，并获取移动车辆的行为预测。
步骤S403：基于所述路面信息获取所述移动车辆的插入区域及对应的插入时间。
可以理解的是，基于第二环境信息中的路面信息获取移动车辆的在目标 车辆前方的插入区域及插入位置。
步骤S404：通过所述状态特征、所述行为预测、所述插入区域及所述插入时间对初始行为预测模型进行训练，以得到行为预测模型。
需要说明的是，在得到状态特征、行为预测、插入区域及插入时间对初始行为预测模型进行训练，以得到行为预测模型。
在具体实现中，所述初始行为预测模型为
Figure PCTCN2022080971-appb-000005
其中
Figure PCTCN2022080971-appb-000006
表示目标车辆周边第k台移动车辆的行为预测，x表示移动车辆的输入状态特征，C s表示第k台移动车辆的插入区域s，C t表示第k台移动车辆的插入时间t；Π k(x)表示高斯系数，
Figure PCTCN2022080971-appb-000007
表示第k台移动车辆插入区域及插入时间的均值；
Figure PCTCN2022080971-appb-000008
表示第k台移动车辆插入区域及插入时间的协方差。
步骤S405：基于所述行为预测模型得到所述行为预测模型对应的损失函数。
可以理解的是，针对行为预测模型，有损失函数
Figure PCTCN2022080971-appb-000009
其中，W 1及W 2为参数，根据实际输入调整参数值，S为插入区域的总数，
Figure PCTCN2022080971-appb-000010
表示当前第k台移动车辆的插入区域s的真值。
步骤S406：根据所述行为预测模型和所述损失函数得到车辆意图模型。
可以理解的是，在得到行为预测模型和所述损失函数可以得到周边的车辆意图模型。
进一步地，所述根据所述行为预测模型和所述损失函数得到车辆意图模型，包括：根据所述行为预测模型和所述损失函数，得到满足预测条件的意图模型和对应的预测概率；将所述满足预测条件的的意图模型和对应的预测概率生成车辆意图模型。
需要说明的是，车载无人机摄像头采集到的图像信息有限，需要实时基于不同的周边环境及移动车辆进行训练得到车辆意图模型。同时因为移动车辆有很多，不同的移动车辆行为不同，预测行为也不相同，所以得到的意图模型也有多个。
在具体实现中，损失函数可以使真正插入的区域拥有最大的权重，同时 对移动车辆插入该区域内的时间和位置信息进行概率预测，输出多个意图模型及对应预测概率C k,p，还可以得到最高的预测概率及对应的意图模型C k,max。将满足预测条件即车载无人机摄像头所能采集图像的既定范围内的意图模型和对应的预测概率生成车辆意图模型。
可以理解的是，本实施例通过将预测概率和意图模型对应结合，使车辆意图模型更为精准，能够得到相应的概率。
进一步地，所述根据车辆意图模型和所述目标行为规划路径得到所述目标车辆的预测行为结果，包括：获取所述车辆意图模型中预测概率最大对应的第一意图模型；获取所述车辆意图模型中预测路径和所述目标行为规划路径一致的路径意图模型；获取所述路径意图模型中预测概率最大对应的第二意图模型；若所述第一意图模型和所述第二意图模型的预测结果相同，则将所述第一意图模型的预测结果作为目标车辆的预测行为结果；若所述第一意图模型和所述第二意图模型的预测结果不相同，则将所述第二意图模型的预测结果作为目标车辆的预测行为结果。
需要说明的是，通过结合目标车辆的目标行为规划路径和车辆意图模型中对应的概率，有：
Figure PCTCN2022080971-appb-000011
即若车辆意图模型中意图模型预测行为概率最大的预测路径和目标车辆目标行为规划路径中的局部行为规划路径一致，则输出车辆意图模型中意图模型预测行为概率最大的预测行为为预测行为结果，若车辆意图模型中意图模型预测行为概率最大的预测路径和目标车辆目标行为规划路径中的局部行为规划路径不一致，则输出车辆意图模型中意图模型中对应的预测路径与目标车辆目标行为规划路径中的局部行为规划路径一致的意图模型中预测概率最大的意图模型对应的预测结果为目标车辆的预测行为结果。
本实施例通过根据所述第二环境图像信息识别目标车辆预设第二范围内的移动车辆及路面信息；获取所述移动车辆的状态特征及行为预测；基于所述路面信息获取所述移动车辆的插入区域及对应的插入时间；通过所述状态特征、所述行为预测、所述插入区域及所述插入时间对初始行为预测模型进行训练，以得到行为预测模型；基于所述行为预测模型得到所述行为预测模型对应的损失函数；根据所述行为预测模型和所述损失函数得到车辆意图模型。通过周边移动车辆和路面信息进行车辆意图模型的训练，最终得到较为 准确的车辆意图模型，对目标车辆的预测行为结果也能更为准确，进一步提高目标车辆在自动驾驶过程中的安全性，同时利用行为预测算法生成车辆意图模型，解决了因为数据的传输时间、控制器的运算时间等导致决策延迟的问题。
此外，参照图5，本申请实施例还提出一种车辆意图预测装置，所述车辆意图预测装置包括：
获取模块10，被配置为获取目标车辆预设第一范围内的环境图像信息，以得到第一环境图像信息；
所述获取模块10，还被配置为获取目标车辆预设第二范围内的环境图像信息，以得到第二环境图像信息，其中，所述预设第二范围大于所述预设第一范围；
生成模块20，被配置为根据所述第一环境图像信息和所述第二环境图像信息通过路径规划模型生成所述目标车辆的目标行为规划路径；
得到模块30，被配置为根据车辆意图模型和所述目标行为规划路径得到所述目标车辆的预测行为结果。
本实施例通过获取目标车辆预设第一范围内的环境图像信息，以得到第一环境图像信息；获取目标车辆预设第二范围内的环境图像信息，以得到第二环境图像信息，其中，所述预设第二范围大于所述预设第一范围；根据所述第一环境图像信息和所述第二环境图像信息通过路径规划模型生成所述目标车辆的目标行为规划路径；根据车辆意图模型和所述目标行为规划路径得到所述目标车辆的预测行为结果。通过上述方式，根据预设范围内的环境信息对目标车辆进行行为规划，采用目标车辆行为规划和车辆意图融合的方法，得到车辆行为的预测，进而提前预测行车危险，进一步提高目标车辆在自动驾驶时的安全性。
在一实施例中，所述生成模块20，还被配置为将所述第一环境图像信息和所述第二环境图像信息进行时间同步匹配，得到同步时间对应的第一环境图像信息和第二环境图像信息；
将所述同步时间对应的第一环境图像信息和第二环境图像信息进行坐标 转换，得到Frenet坐标下的第一环境图像信息和第二环境图像信息；
根据所述第一环境图像信息和所述第二环境图像信息通过路径规划模型生成所述目标车辆的目标行为规划路径，包括：
根据所述Frenet坐标下的第一环境图像信息和第二环境图像信息通过路径规划模型生成所述目标车辆的目标行为规划路径。
在一实施例中，所述生成模块20，还被配置为根据所述Frenet坐标下第一环境图像信息和所述第二环境图像信息得到目标车辆的周边环境信息；
基于所述周边环境信息规划所述目标车辆在预设路段行驶的目标路径，其中，所述预设路段为根据所述目标车辆当前所处位置前方预设距离确定的；
对所述目标路径通过预设预测模型进行计算，生成所述目标车辆的目标行为规划路径。
在一实施例中，所述生成模块20，还被配置为根据所述目标路径得到离散路径点；
通过预设拟合模型对所述离散路径点进行拟合，以得到所述目标车辆的局部行为规划路径；
对所述局部行为规划路径进行计算，得到局部行为规划路径的代价函数值；
当所述代价函数值趋于预设值时，得到代价函数值为预设值对应的路径规划曲线；
根据所述路径规划曲线生成对应的目标行为规划路径。
在一实施例中，所述得到模块30，还被配置为根据所述第二环境图像信息识别目标车辆预设第二范围内的移动车辆及路面信息；
获取所述移动车辆的状态特征及行为预测；
基于所述路面信息获取所述移动车辆的插入区域及对应的插入时间；
通过所述状态特征、所述行为预测、所述插入区域及所述插入时间对初始行为预测模型进行训练，以得到行为预测模型；
基于所述行为预测模型得到所述行为预测模型对应的损失函数；
根据所述行为预测模型和所述损失函数得到车辆意图模型。
在一实施例中，所述得到模块30，还被配置为根据所述行为预测模型和所述损失函数，得到满足预测条件的意图模型和对应的预测概率；
将所述满足预测条件的的意图模型和对应的预测概率生成车辆意图模型。
在一实施例中，所述得到模块30，还被配置为获取所述车辆意图模型中预测概率最大对应的第一意图模型；
获取所述车辆意图模型中预测路径和所述目标行为规划路径一致的路径意图模型；
获取所述路径意图模型中预测概率最大对应的第二意图模型；
若所述第一意图模型和所述第二意图模型的预测结果相同，则将所述第一意图模型的预测结果作为目标车辆的预测行为结果；
若所述第一意图模型和所述第二意图模型的预测结果不相同，则将所述第二意图模型的预测结果作为目标车辆的预测行为结果。
此外，本申请实施例还提出一种存储介质，所述存储介质上存储有车辆意图预测程序，所述车辆意图预测程序被处理器执行时实现如上文所述的车辆意图预测方法的步骤。
由于本存储介质采用了上述所有实施例的全部技术方案，因此至少具有上述实施例的技术方案所带来的所有功能，在此不再一一赘述。
需要说明的是，以上所描述的工作流程仅仅是示意性的，并不对本申请的保护范围构成限定，在实际应用中，本领域的技术人员可以根据实际的需要选择其中的部分或者全部来实现本实施例方案的目的，此处不做限制。
另外，未在本实施例中详尽描述的技术细节，可参见本申请任意实施例所提供的车辆意图预测方法，此处不再赘述。
此外，需要说明的是，在本文中，术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、物品或者系统不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、物品或者系统所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括该要素的过程、方法、物品或者系统中还存在另外的相同要素。
上述本申请实施例序号仅仅为了描述，不代表实施例的优劣。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件，但很多情况下前者是更佳的实施方式。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质(如只读存储器(Read Only Memory，ROM)/RAM、磁碟、光盘)中，包括若干指令用以使得一台终端设备(可以是手机，计算机，服务器，或者网络设备等)执行本申请各个实施例所述的方法。
以上仅为本申请的可选实施例，并非因此限制本申请的专利范围，凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本申请的专利保护范围内。

Claims (10)

  1. 一种车辆意图预测方法，其中，所述车辆意图预测方法包括：
    获取目标车辆预设第一范围内的环境图像信息，以得到第一环境图像信息；
    获取目标车辆预设第二范围内的环境图像信息，以得到第二环境图像信息，其中，所述预设第二范围大于所述预设第一范围；
    根据所述第一环境图像信息和所述第二环境图像信息通过路径规划模型生成所述目标车辆的目标行为规划路径；
    根据车辆意图模型和所述目标行为规划路径得到所述目标车辆的预测行为结果。
  2. 如权利要求1所述的车辆意图预测方法，其中，所述根据所述第一环境图像信息和所述第二环境图像信息通过路径规划模型生成所述目标车辆的目标行为规划路径之前，还包括：
    将所述第一环境图像信息和所述第二环境图像信息进行时间同步匹配，得到同步时间对应的第一环境图像信息和第二环境图像信息；
    将所述同步时间对应的第一环境图像信息和第二环境图像信息进行坐标转换，得到Frenet坐标下的第一环境图像信息和第二环境图像信息；
    根据所述第一环境图像信息和所述第二环境图像信息通过路径规划模型生成所述目标车辆的目标行为规划路径，包括：
    根据所述Frenet坐标下的第一环境图像信息和第二环境图像信息通过路径规划模型生成所述目标车辆的目标行为规划路径。
  3. 如权利要求2所述的车辆意图预测方法，其中，所述根据所述Frenet坐标下的第一环境图像信息和第二环境图像信息通过路径规划模型生成所述目标车辆的目标行为规划路径，包括：
    根据所述Frenet坐标下第一环境图像信息和所述第二环境图像信息得到目标车辆的周边环境信息；
    基于所述周边环境信息规划所述目标车辆在预设路段行驶的目标路径， 其中，所述预设路段为根据所述目标车辆当前所处位置前方预设距离确定的；
    对所述目标路径通过预设预测模型进行计算，生成所述目标车辆的目标行为规划路径。
  4. 如权利要求3所述的车辆意图预测方法，其中，所述对所述目标路径通过预设预测模型进行计算，生成所述目标车辆的目标行为规划路径，包括：
    根据所述目标路径得到离散路径点；
    通过预设拟合模型对所述离散路径点进行拟合，以得到所述目标车辆的局部行为规划路径；
    对所述局部行为规划路径进行计算，得到局部行为规划路径的代价函数值；
    当所述代价函数值趋于预设值时，得到代价函数值为预设值对应的路径规划曲线；
    根据所述路径规划曲线生成对应的目标行为规划路径。
  5. 如权利要求1所述的车辆意图预测方法，其中，所述根据车辆意图模型和所述目标行为规划路径得到所述目标车辆的预测行为结果之前，还包括：
    根据所述第二环境图像信息识别目标车辆预设第二范围内的移动车辆及路面信息；
    获取所述移动车辆的状态特征及行为预测；
    基于所述路面信息获取所述移动车辆的插入区域及对应的插入时间；
    通过所述状态特征、所述行为预测、所述插入区域及所述插入时间对初始行为预测模型进行训练，以得到行为预测模型；
    基于所述行为预测模型得到所述行为预测模型对应的损失函数；
    根据所述行为预测模型和所述损失函数得到车辆意图模型。
  6. 如权利要求5所述的车辆意图预测方法，其中，所述根据所述行为预测模型和所述损失函数得到车辆意图模型，包括：
    根据所述行为预测模型和所述损失函数，得到满足预测条件的意图模型和对应的预测概率；
    将所述满足预测条件的的意图模型和对应的预测概率生成车辆意图模型。
  7. 如权利要求1-6中任一项所述的车辆意图预测方法，其中，所述根据车辆意图模型和所述目标行为规划路径得到所述目标车辆的预测行为结果，包括：
    获取所述车辆意图模型中预测概率最大对应的第一意图模型；
    获取所述车辆意图模型中预测路径和所述目标行为规划路径一致的路径意图模型；
    获取所述路径意图模型中预测概率最大对应的第二意图模型；
    若所述第一意图模型和所述第二意图模型的预测结果相同，则将所述第一意图模型的预测结果作为目标车辆的预测行为结果；
    若所述第一意图模型和所述第二意图模型的预测结果不相同，则将所述第二意图模型的预测结果作为目标车辆的预测行为结果。
  8. 一种车辆意图预测装置，其中，所述车辆意图预测装置包括：
    获取模块，被配置为获取目标车辆预设第一范围内的环境图像信息，以得到第一环境图像信息；
    所述获取模块，还被配置为获取目标车辆预设第二范围内的环境图像信息，以得到第二环境图像信息，其中，所述预设第二范围大于所述预设第一范围；
    生成模块，被配置为根据所述第一环境图像信息和所述第二环境图像信息通过路径规划模型生成所述目标车辆的目标行为规划路径；
    得到模块，被配置为根据车辆意图模型和所述目标行为规划路径得到所述目标车辆的预测行为结果。
  9. 一种车辆意图预测设备，其中，所述设备包括：存储器、处理器及存储在所述存储器上并可在所述处理器上运行的车辆意图预测程序，所述车辆意图预测程序配置为实现如权利要求1至7中任一项所述的车辆意图预测方法。
  10. 一种存储介质，其中，所述存储介质上存储有车辆意图预测程序，所述车辆意图预测程序被处理器执行时实现如权利要求1至7任一项所述的车辆意图预测方法。
PCT/CN2022/080971 2021-04-21 2022-03-15 车辆意图预测方法、装置、设备及存储介质 WO2022222647A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110429099.7 2021-04-21
CN202110429099.7A CN113189989B (zh) 2021-04-21 2021-04-21 车辆意图预测方法、装置、设备及存储介质

Publications (1)

Publication Number Publication Date
WO2022222647A1 true WO2022222647A1 (zh) 2022-10-27

Family

ID=76977812

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/080971 WO2022222647A1 (zh) 2021-04-21 2022-03-15 车辆意图预测方法、装置、设备及存储介质

Country Status (2)

Country Link
CN (1) CN113189989B (zh)
WO (1) WO2022222647A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113189989B (zh) * 2021-04-21 2022-07-01 东风柳州汽车有限公司 车辆意图预测方法、装置、设备及存储介质
CN116125953A (zh) * 2023-02-22 2023-05-16 吉林大学 基于飞行器的车辆监控系统及方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108352112A (zh) * 2015-11-04 2018-07-31 大众汽车有限公司 用于确定车辆的驾驶意图的方法和车辆通信系统
CN111998864A (zh) * 2020-08-11 2020-11-27 东风柳州汽车有限公司 无人车局部路径规划方法、装置、设备及存储介质
CN112133089A (zh) * 2020-07-21 2020-12-25 西安交通大学 一种基于周围环境与行为意图的车辆轨迹预测方法、系统及装置
CN112558608A (zh) * 2020-12-11 2021-03-26 重庆邮电大学 一种基于无人机辅助的车机协同控制及路径优化方法
CN113189989A (zh) * 2021-04-21 2021-07-30 东风柳州汽车有限公司 车辆意图预测方法、装置、设备及存储介质

Family Cites Families (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2040137B1 (en) * 2007-09-21 2012-05-16 The Boeing Company Predicting aircraft trajectory
EP2654028B1 (en) * 2012-04-20 2018-09-26 Honda Research Institute Europe GmbH Orientation sensitive traffic collision warning system
CN104699102B (zh) * 2015-02-06 2017-07-18 东北大学 一种无人机与智能车协同导航与侦查监控系统及方法
CN105318888B (zh) * 2015-12-07 2018-08-14 北京航空航天大学 基于无人机感知的无人驾驶车辆路径规划方法
CN106530694B (zh) * 2016-11-07 2019-01-11 深圳大学 一种基于交通拥堵传播模型的交通拥堵预测方法及系统
ES2943122T3 (es) * 2017-06-29 2023-06-09 Boeing Co Método y sistema para operar de forma autónoma una aeronave
CN108305275B (zh) * 2017-08-25 2021-02-12 深圳市腾讯计算机系统有限公司 主动跟踪方法、装置及系统
CN110126730A (zh) * 2018-02-02 2019-08-16 上海博泰悦臻电子设备制造有限公司 车辆变道提醒方法及系统
CN109032173A (zh) * 2018-07-09 2018-12-18 上海交通大学 一种基于空地交互的车载无人机系统
CN110007675B (zh) * 2019-04-12 2021-01-15 北京航空航天大学 一种基于行车态势图的车辆自动驾驶决策系统及基于无人机的训练集制备方法
CN115016531A (zh) * 2019-05-09 2022-09-06 深圳市速腾聚创科技有限公司 车辆自动驾驶方法、装置、计算机设备和存储介质
CN112519765A (zh) * 2019-09-03 2021-03-19 北京百度网讯科技有限公司 车辆控制方法、装置、设备和介质
US11345342B2 (en) * 2019-09-27 2022-05-31 Intel Corporation Potential collision warning system based on road user intent prediction
CN212243156U (zh) * 2020-05-14 2020-12-29 吴韵驰 一种大型车辆动态监控预警系统
CN111580522A (zh) * 2020-05-15 2020-08-25 东风柳州汽车有限公司 无人驾驶汽车的控制方法、汽车和存储介质
CN112015843B (zh) * 2020-09-02 2023-10-24 中国科学技术大学 基于多车意图交互结果的行车风险态势评估方法及系统
CN112347993B (zh) * 2020-11-30 2023-03-17 吉林大学 一种基于车辆-无人机协同的高速公路车辆行为和轨迹预测方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108352112A (zh) * 2015-11-04 2018-07-31 大众汽车有限公司 用于确定车辆的驾驶意图的方法和车辆通信系统
CN112133089A (zh) * 2020-07-21 2020-12-25 西安交通大学 一种基于周围环境与行为意图的车辆轨迹预测方法、系统及装置
CN111998864A (zh) * 2020-08-11 2020-11-27 东风柳州汽车有限公司 无人车局部路径规划方法、装置、设备及存储介质
CN112558608A (zh) * 2020-12-11 2021-03-26 重庆邮电大学 一种基于无人机辅助的车机协同控制及路径优化方法
CN113189989A (zh) * 2021-04-21 2021-07-30 东风柳州汽车有限公司 车辆意图预测方法、装置、设备及存储介质

Also Published As

Publication number Publication date
CN113189989B (zh) 2022-07-01
CN113189989A (zh) 2021-07-30

Similar Documents

Publication Publication Date Title
US11783590B2 (en) Method, apparatus, device and medium for classifying driving scenario data
EP3568334B1 (en) System, method and non-transitory computer readable storage medium for parking vehicle
EP4152204A1 (en) Lane line detection method, and related apparatus
CN111123933B (zh) 车辆轨迹规划的方法、装置、智能驾驶域控制器和智能车
EP3693944A1 (en) Method and device for short-term path planning of autonomous driving through information fusion by using v2x communication and image processing
WO2022222647A1 (zh) 车辆意图预测方法、装置、设备及存储介质
EP3822852B1 (en) Method, apparatus, computer storage medium and program for training a trajectory planning model
US20220410939A1 (en) Collision detection method, electronic device, and medium
CN109109863B (zh) 智能设备及其控制方法、装置
CN112740268B (zh) 目标检测方法和装置
WO2021217420A1 (zh) 车道线跟踪方法和装置
US20210287387A1 (en) Lidar point selection using image segmentation
CN110969064B (zh) 一种基于单目视觉的图像检测方法、装置及存储设备
WO2020215254A1 (zh) 车道线地图的维护方法、电子设备与存储介质
JP2023508705A (ja) データ伝送方法および装置
CN113240734B (zh) 一种基于鸟瞰图的车辆跨位判断方法、装置、设备及介质
CN111652072A (zh) 轨迹获取方法、轨迹获取装置、存储介质和电子设备
US20230300196A1 (en) Data sharing method and apparatus applied between vehicles, medium, and electronic device
CN115339453B (zh) 车辆换道决策信息生成方法、装置、设备和计算机介质
CN110276801B (zh) 一种物体定位方法、装置及存储介质
JP2022164640A (ja) マルチモーダル自動ラベル付けと能動的学習のためのデータセットとモデル管理のためのシステムと方法
CN113119999B (zh) 自动驾驶特征的确定方法、装置、设备、介质及程序产品
CN112712608B (zh) 用于通过车辆收集性能数据的系统和方法
WO2023155580A1 (zh) 一种对象识别方法和装置
US11657268B1 (en) Training neural networks to assign scores

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22790756

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 22790756

Country of ref document: EP

Kind code of ref document: A1