WO2023159915A1 - 车辆轨迹的预测方法、装置 - Google Patents

车辆轨迹的预测方法、装置 Download PDF

Info

Publication number
WO2023159915A1
WO2023159915A1 PCT/CN2022/117878 CN2022117878W WO2023159915A1 WO 2023159915 A1 WO2023159915 A1 WO 2023159915A1 CN 2022117878 W CN2022117878 W CN 2022117878W WO 2023159915 A1 WO2023159915 A1 WO 2023159915A1
Authority
WO
WIPO (PCT)
Prior art keywords
vehicle
feature information
information
map
prediction
Prior art date
Application number
PCT/CN2022/117878
Other languages
English (en)
French (fr)
Inventor
李荣华
陈红丽
王宁
卢丽婧
Original Assignee
中国第一汽车股份有限公司
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 中国第一汽车股份有限公司 filed Critical 中国第一汽车股份有限公司
Publication of WO2023159915A1 publication Critical patent/WO2023159915A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
    • G01C21/26—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network
    • G01C21/28—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network with correlation of data from several navigational instruments
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
    • G01C21/26—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network
    • G01C21/28—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network with correlation of data from several navigational instruments
    • G01C21/30—Map- or contour-matching
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/29—Geographical information databases

Definitions

  • the present application relates to the field of intelligent networked automatic driving, in particular, to a method and device for predicting vehicle trajectories.
  • the vehicle trajectory prediction of autonomous vehicles mainly relies on the calculation of historical trajectories, and is only applicable to simple driving situations with a short prediction range.
  • Embodiments of the present application provide a method and device for predicting vehicle trajectories, so as to at least solve the technical problem in the related art that vehicle trajectories cannot be predicted in real time.
  • a method for predicting a vehicle trajectory including: when the vehicle is running, triggering a prediction program for predicting the vehicle trajectory; calling the map feature information of the high-precision map, and loading it into Memory, wherein the map feature information is static data; in response to the prediction command, capture the dynamic perception information during the operation of the vehicle, wherein the dynamic perception information includes at least: the positioning information of the vehicle; based on the positioning information of the vehicle, the vehicle is queried from the memory The map feature information of the current area; input the queried map feature information and dynamic perception information into the prediction model, and predict the predicted trajectory of the vehicle.
  • the method before calling the map feature information of the high-precision map, the method further includes: obtaining at least one version of the high-precision map, and performing offline preprocessing on any one or more high-precision maps; traversing each For each lane in the high-precision map, extract the feature information of the lane center point of the lane, wherein the feature information includes at least one of the following: front and rear neighbor features, left and right neighbor features adjacent to the vehicle; feature of the lane center point of the lane The information is stored to the target disk.
  • call the map feature information of the high-precision map and load it into the memory including: after starting the prediction program, based on the version of the high-precision map, read the corresponding map feature information from the disk storage file of the target disk; The retrieved map feature information is loaded into the memory in the form of a tree data structure.
  • the method before inputting the queried map feature information and dynamic perception information into the prediction model, the method further includes: performing tensor calculation on the map feature information in memory to obtain tensor information required for video memory calculation.
  • the method further includes: acquiring historical driving data and current location information of the vehicle within a historical time; and calculating the historical driving trajectory of the vehicle based on the historical location information and current location information included in the historical driving data.
  • a vehicle trajectory prediction device including: a trigger module, configured to trigger a prediction program for predicting vehicle trajectory when the vehicle is running; a processing module, configured to Take the map feature information of the high-precision map and load it into the memory, wherein the map feature information is static data; the capture module is set to respond to the prediction command and capture the dynamic perception information during the running of the vehicle.
  • the dynamic perception information includes at least: the vehicle's positioning information; the query module is set to be based on the positioning information of the vehicle, and the map feature information of the current area where the vehicle is located is inquired from the memory; the prediction module is set to input the inquired map feature information and dynamic perception information into the prediction model, and predict Get the predicted trajectory of the vehicle.
  • the acquisition module is configured to acquire at least one version of the high-precision map;
  • the preprocessing module is configured to perform offline preprocessing on any one or more high-precision maps;
  • the extraction module is configured to traverse each type of high-precision map For each lane in the map, the feature information of the lane center point of the lane is extracted, and the feature information includes at least one of the following: front and rear neighbor features and left and right neighbor features adjacent to the vehicle;
  • the storage module is set to the lane center point of the lane Characteristic information is stored to the target disk.
  • the device also includes: a reading module, configured to read the corresponding map feature information from the disk storage file of the target disk based on the version of the high-precision map after starting the prediction program;
  • the acquired map feature information is loaded into the memory in the form of a tree data structure.
  • a computer-readable storage medium includes a stored program, wherein, when the program is running, the device where the computer-readable storage medium is located is controlled to execute the above-mentioned embodiment.
  • a processor is also provided, and the processor is used to run a program, wherein, when the program is running, the vehicle trajectory prediction method in the above-mentioned embodiments is executed.
  • an electronic device including: one or more processors; a storage device for storing one or more programs; when one or more programs are used by one or more The processor executes, so that one or more processors execute the above-mentioned method for predicting the vehicle trajectory.
  • the prediction program for predicting the vehicle trajectory when the vehicle is running, the prediction program for predicting the vehicle trajectory is triggered; the map feature information of the high-precision map is called and loaded into the memory, wherein the map feature information is static data; the response prediction commands to capture the dynamic perception information during vehicle operation; based on the positioning information of the vehicle, query the map feature information of the current area where the vehicle is located from the memory; input the queried map feature information and dynamic perception information into the prediction model, and predict The way of predicting the trajectory of the vehicle achieves the purpose of real-time response by deploying the prediction model, thereby realizing the technical effect of real-time early warning of the trajectory of the vehicle, and then solving the technical problem that the vehicle trajectory cannot be predicted in real time in related technologies.
  • Fig. 1 is a schematic flow chart of a method for predicting a vehicle trajectory according to an embodiment of the present application
  • FIG. 2 is a schematic diagram of an optional LaneGCN network dynamic and static data division according to an embodiment of the present application
  • FIG. 3 is a schematic flowchart of an optional offline preprocessing of high-precision map data according to an embodiment of the present application
  • Fig. 4 is a schematic flow chart of an optional torch tensor algorithm according to an embodiment of the present application.
  • Fig. 5 is a schematic structural diagram of a vehicle prediction device according to an embodiment of the present application.
  • an embodiment of a method for predicting a vehicle trajectory is provided. It should be noted that the steps shown in the flow chart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and, Although a logical order is shown in the flowcharts, in some cases the steps shown or described may be performed in an order different from that shown or described herein.
  • Fig. 1 is a schematic flow chart of a method for predicting a vehicle trajectory according to an embodiment of the present application. As shown in Fig. 1, the method includes the following steps:
  • Step S102 when the vehicle is running, trigger a prediction program for predicting the trajectory of the vehicle;
  • the vehicle in the above steps may be a vehicle driven by a user, or may be a self-driving vehicle.
  • an self-driving vehicle is taken as an example for illustration.
  • the vehicle is pre-installed with a navigation system, which at least includes: a navigation device for obtaining high-precision map-related information and vehicle dynamic perception information, and a prediction device for predicting the vehicle's driving trajectory.
  • the navigation device may be a navigator built into the vehicle, or a navigator additionally installed on the vehicle; the predictive device here may be a device running a predictive program, for example, a built-in controller in the vehicle, additionally installed in the vehicle on the electronic terminal, etc.
  • the navigation system can be automatically turned on and always in working condition.
  • the navigation system of the vehicle can be automatically turned on, and when the vehicle is running, it can trigger the prediction program for predicting the trajectory of the vehicle, that is, trigger the above-mentioned prediction device to run the prediction program.
  • Step S104 call the map feature information of the high-precision map, and load it into the memory, wherein the map feature information is static data;
  • the high-precision map in the above steps is an electronic map with higher precision and more data dimensions. Its precision can reach 20cm.
  • the data dimension includes almost all surrounding static information related to traffic except road information, such as: a certain range Buildings, obstacles, etc. within the map, such information may be map feature information in this embodiment.
  • the map feature information of the high-precision map can be automatically obtained through the network through the vehicle's navigation device, and loaded into the memory.
  • Step S106 responding to the prediction instruction, capturing dynamic perception information during the running of the vehicle, wherein the dynamic perception information at least includes: positioning information of the vehicle;
  • the dynamic perception information of the vehicle in the above steps may include information such as left-turn and right-turn attributes of the current lane, whether there are traffic lights in the current lane, the front, rear, left, and right lanes of the current lane, and vehicle positioning information. Since the dynamic perception information of the vehicle will affect the vehicle The timeliness and accuracy of the navigation system can be changed by users according to their needs.
  • the prediction device can be automatically turned on, and after the vehicle starts driving, the prediction instruction is automatically generated and sent to the navigation system.
  • the navigation system can use the navigation device, the camera installed on the vehicle and other devices Comprehensive acquisition of dynamic perception information during vehicle operation, for example, capture vehicle positioning information through the network through navigation equipment; the camera installed on the vehicle can capture images of the front and rear of the vehicle, and then determine the current location of the vehicle through the lane line recognition algorithm Lane, and other lanes in the front, rear, left, and right; through the network to capture information such as left-turn and right-turn attributes of the current lane, whether there are traffic lights in the current lane, etc.
  • Step S108 based on the positioning information of the vehicle, query the map feature information of the area where the vehicle is currently located from the memory;
  • the map feature information of the high-precision map cached in the memory in advance can be queried through the positioning information, and the map of the high-precision map of the area where the vehicle is currently located can be obtained characteristic information.
  • Step S110 input the queried map feature information and dynamic perception information into the prediction model, and predict the predicted trajectory of the vehicle.
  • the prediction model in the above steps can be a set of algorithm models used in the prediction device to predict the vehicle trajectory, and can include algorithms such as LaneGCN network (Lane Graph Convolution Network, lane graph convolution network), torch tensor, but not limited to Therefore, other algorithms are also possible.
  • LaneGCN network Li Graph Convolution Network, lane graph convolution network
  • torch tensor but not limited to Therefore, other algorithms are also possible.
  • the map of the high-precision map of the area where the vehicle is currently located can be The feature information and dynamic perception information during vehicle operation are sent to the prediction device, and the prediction device inputs the received map feature information of the high-precision map of the area where the vehicle is currently located and the dynamic perception information during vehicle operation into the prediction model, and predicts
  • the model can calculate the predicted trajectory of the vehicle through the LaneGCN network, torch tensor algorithm, etc.
  • the method before calling the map feature information of the high-precision map, the method further includes: obtaining at least one version of the high-precision map, and performing offline preprocessing on any one or more high-precision maps; traversing each For each lane in the high-precision map, extract the feature information of the lane center point of the lane, wherein the feature information includes at least one of the following: front and rear neighbor features, left and right neighbor features adjacent to the vehicle; feature of the lane center point of the lane The information is stored to the target disk.
  • the above-mentioned offline pre-processing may be to pre-process any one or more types of map feature information of high-precision maps in an offline state.
  • the target disk can be any kind of disk that can store the map feature information of the high-precision map, and this application does not specifically limit it.
  • the target disk can be a built-in disk in the vehicle, or an additionally installed disk.
  • At least one version of high-precision map can be obtained through the network through the vehicle's navigation device, and offline preprocessing is performed on any one or more high-precision maps, and then by traversing each high-precision map
  • Each lane in the map is refined, and the feature information of the lane center point of the lane is extracted, such as: front and rear neighbor features, left and right neighbor features adjacent to the vehicle, etc., and finally the extracted feature information of the lane center point can be stored To the target disk, it is convenient to directly read the characteristic information in the target disk in the future.
  • call the map feature information of the high-precision map and load it into the memory including: after starting the prediction program, based on the version of the high-precision map, read the corresponding map feature information from the disk storage file of the target disk; The retrieved map feature information is loaded into the memory in the form of a tree data structure.
  • the above-mentioned tree data structure can be selected according to the needs of the user.
  • the data structure of K-d tree is selected for illustration.
  • the navigation system is automatically turned on, and the dynamic perception information during the operation of the vehicle can be comprehensively obtained through the navigation equipment, the camera installed on the vehicle, and other equipment.
  • Different versions of map feature information therefore, based on the version of the currently used high-precision map, the map feature information of the high-precision map corresponding to the version can be read from the target disk, and then combined with the acquired For dynamic perception information, the retrieved map feature information around the vehicle is loaded into the memory in the form of a K-d tree data structure.
  • the method before inputting the queried map feature information and dynamic perception information into the prediction model, the method further includes: performing tensor calculation on the map feature information in memory to obtain tensor information required for video memory calculation.
  • the tensor calculation mentioned above can be selected according to the needs of the user.
  • the torch tensor calculation is chosen as an example for illustration.
  • the torch tensor calculation can be performed on the read vehicle dynamic perception information and map feature information, The tensor information required for video memory calculation is obtained, and then the calculated tensor information can be passed to the prediction device, and the tensor information can be input into the prediction model through the prediction device.
  • the method further includes: acquiring historical driving data and current location information of the vehicle within a historical time; and calculating the historical driving trajectory of the vehicle based on the historical location information and current location information included in the historical driving data.
  • the above-mentioned historical time is a period of time required by the vehicle during historical driving.
  • the driving data of the vehicle can be stored in the database as the historical driving data used in the subsequent navigation process.
  • the historical driving data can be collected from the history composed of positioning information.
  • the position that the vehicle has traveled in the historical driving process can be calculated, and then a historical driving trajectory composed of different positions can be generated.
  • the high-precision map that is, the static data is independent of the vehicle trajectory, that is, the dynamic data.
  • the lanes in the area can be established in advance through the map feature extraction algorithm. feature information.
  • the map feature information is loaded into the memory, the map feature data is queried according to the vehicle's positioning information, and then the map feature data and dynamic perception information are input into the prediction model, and the output of the prediction model is the predicted vehicle Predict the trajectory, so that the extraction of map feature information can be limited to within 1 millisecond.
  • the navigation device is automatically turned on
  • Step 31 the navigation device obtains at least one version of the high-precision map that needs to extract features through the network
  • Step 32 the navigation device traverses the lanes in the high-definition map, and then enters step 34;
  • Step 33 the navigation device extracts the feature information of the center point of the lane, such as: front and rear neighbor features, left and right neighbor features and other information adjacent to the vehicle;
  • Step 34 storing lane information and lane center point feature information
  • Step 35 after the information is stored, the process ends.
  • the torch tensor calculation is also needed to obtain the historical running track of the vehicle during runtime, and organize the feature information in the memory into the tensor information required for the memory calculation. Whether it is a CPU or a GPU, there is a warmup process during the initial calculation, so the initial calculation of the torch tensor will take a long time during runtime. In order to eliminate this part of the performance jitter, when the prediction program is initialized, the calculation of the torch tensor is performed first to ensure the smooth operation of the subsequent prediction process.
  • the predictive device is automatically turned on
  • Step 41 after the prediction program is started, first find out the corresponding disk storage file according to the currently used high-precision map, and load these map feature information into the memory in the form of a K-d tree;
  • Step 42 load the prediction model into the video memory, in order to reduce the performance loss caused by the first operation of the torch tensor, warmup the torch, call the required tensor operation in advance, and the initialization is completed;
  • Step 43 during the running of the prediction program, receive the automatic driving perception data, and calculate the vehicle's running history trajectory according to the perceived historical data and current value;
  • Step 44 query and obtain high-precision map feature information according to the perceived location information
  • Step 45 using the above-mentioned dynamic information and static information as input, and submitting it to the prediction model for reasoning;
  • Step 46 perform post-processing on the inference output, and transmit the prediction result to the regulation and control module, that is, the control module of the vehicle.
  • This application discloses a space-for-time model deployment scheme.
  • a model deployment scheme of space-for-time is proposed.
  • the solution processes the calculation of the static high-precision map data offline, loads it into the memory when the program is running, and then provides real-time query, combined with the dynamic trajectory data of the vehicle operation, to perform prediction model reasoning.
  • the application also discloses a torch tensor operation warmup strategy. In view of the slow processing speed of the first torch tensor operation, the operation is processed when the prediction program starts to avoid excessive runtime overhead.
  • a vehicle trajectory prediction device which can execute the vehicle trajectory prediction method provided in the above-mentioned embodiment 1.
  • the specific implementation and preferred application scenarios are the same as the above-mentioned embodiment 1, and will not be described here. repeat.
  • FIG. 5 is a schematic structural diagram of a vehicle prediction device according to an embodiment of the present application.
  • the device includes: a trigger module 50 configured to trigger a prediction program for predicting vehicle trajectory when the vehicle is running;
  • the processing module 52 is configured to call the map characteristic information of the high-precision map and load it into the memory, wherein the map characteristic information is static data;
  • the capture module 54 is configured to respond to the prediction instruction and capture the dynamic perception information during the running of the vehicle,
  • the dynamic perception information at least includes: the positioning information of the vehicle;
  • the query module 56 is set to query the map feature information of the area where the vehicle is currently located from the memory based on the positioning information of the vehicle;
  • the prediction module 58 is set to query the map
  • the feature information and dynamic perception information are input to the prediction model, and the predicted trajectory of the vehicle is predicted.
  • the above-mentioned trigger module 50, processing module 52, capture module 54, query module 56, and prediction module 58 can be run in a computer terminal as a part of the device, and the above-mentioned modules can be executed by a processor in the computer terminal
  • the computer terminal can also be a smart phone (such as an Android mobile phone, an IOS mobile phone, etc.), a tablet computer, a handheld computer, and a mobile Internet device (Mobile Internet Devices, MID), PAD and other terminal devices.
  • the device further includes: a first acquisition module, configured to acquire at least one version of high-precision maps; a preprocessing module, configured to perform off-line preprocessing on any one or more types of high-precision maps; an extraction module, It is set to traverse each lane in each kind of high-precision map, and extract the feature information of the lane center point of the lane, wherein the feature information includes at least one of the following: front and rear neighbor features, left and right neighbor features adjacent to the vehicle; storage module, Set to store the feature information of the lane center point of the lane to the target disk.
  • a first acquisition module configured to acquire at least one version of high-precision maps
  • a preprocessing module configured to perform off-line preprocessing on any one or more types of high-precision maps
  • an extraction module It is set to traverse each lane in each kind of high-precision map, and extract the feature information of the lane center point of the lane, wherein the feature information includes at least one of the following: front and rear neighbor features,
  • the processing module also includes: a reading module, configured to read the corresponding map feature information from the disk storage file of the target disk based on the version of the high-precision map after starting the prediction program;
  • the retrieved map feature information is loaded into the memory in the form of a tree data structure.
  • the above-mentioned first acquisition module, preprocessing module, extraction module, storage module, reading module, and loading module can be run in a computer terminal as a part of the device, and can be executed by a processor in the computer terminal
  • the computer terminal can also be terminal devices such as smart phones (such as Android mobile phones, IOS mobile phones, etc.), tablet computers, palmtop computers, and mobile Internet devices (Mobile Internet Devices, MID), PAD.
  • the device further includes: a first calculation module configured to perform tensor calculation on the map feature information in the internal memory to obtain tensor information required for video memory calculation.
  • a first calculation module configured to perform tensor calculation on the map feature information in the internal memory to obtain tensor information required for video memory calculation.
  • the device further includes: a second acquisition module, configured to acquire historical driving data and current location information of the vehicle within a historical time; a second calculation module, configured to obtain historical location information and current location information included in the historical driving data The positioning information is used to calculate the historical driving trajectory of the vehicle.
  • a second acquisition module configured to acquire historical driving data and current location information of the vehicle within a historical time
  • a second calculation module configured to obtain historical location information and current location information included in the historical driving data The positioning information is used to calculate the historical driving trajectory of the vehicle.
  • the terminal can also be a smart phone (such as an Android mobile phone, an IOS mobile phone, etc.), a tablet computer, a handheld computer, a mobile Internet device (Mobile Internet Devices, MID), a PAD and other terminal devices.
  • a computer-readable storage medium includes a stored program, wherein, when the program is running, the device where the computer-readable storage medium is located is controlled to execute the vehicle trajectory in the above-mentioned embodiments method of prediction.
  • a processor is also provided, and the processor is used to run a program, wherein, when the program is running, the vehicle trajectory prediction method in the above-mentioned embodiments is executed.
  • the disclosed technical content can be realized in other ways.
  • the device embodiments described above are only illustrative.
  • the division of the units can be a logical function division.
  • multiple units or components can be combined or can be Integrate into another system, or some features may be ignored, or not implemented.
  • the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of units or modules may be in electrical or other forms.
  • the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple units. Part or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.
  • each functional unit in each embodiment of the present application may be integrated into one processing unit, each unit may exist separately physically, or two or more units may be integrated into one unit.
  • the above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.
  • the integrated unit is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer-readable storage medium.
  • the technical solution of the present application is essentially or part of the contribution to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , including several instructions for enabling a computer device (which may be a personal computer, server or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application.
  • the aforementioned storage media include: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disc, etc., which can store program codes. .
  • the solutions provided in the embodiments of this application can be applied in the field of intelligent networked automatic driving.
  • trigger the prediction program for predicting the trajectory of the vehicle retrieve the map feature information of the high-precision map and load it into the memory, where the map feature information is static data; respond to the prediction command to capture the vehicle running process
  • the dynamic perception information in the vehicle based on the positioning information of the vehicle, query the map feature information of the current area where the vehicle is located from the memory; input the query map feature information and dynamic perception information into the prediction model, and predict the way to obtain the predicted trajectory of the vehicle , by deploying the prediction model, the purpose of real-time response is achieved, thereby realizing the technical effect of real-time early warning of the vehicle's trajectory, and then solving the technical problem that the vehicle trajectory cannot be predicted in real time in related technologies.

Landscapes

  • Engineering & Computer Science (AREA)
  • Remote Sensing (AREA)
  • Radar, Positioning & Navigation (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Automation & Control Theory (AREA)
  • Computational Linguistics (AREA)
  • Navigation (AREA)
  • Traffic Control Systems (AREA)

Abstract

本申请公开了一种车辆轨迹的预测方法、装置。其中，该方法包括：在车辆运行时，触发用于预测车辆行驶轨迹的预测程序；调取高精地图的地图特征信息，并加载至内存，其中，地图特征信息为静态数据；响应预测指令，捕获车辆运行过程中的动态感知信息；基于车辆的定位信息，从内存中查询得到车辆当前所在区域的地图特征信息；将查询到的地图特征信息和动态感知信息输入至预测模型，预测得到车辆的预测轨迹。本申请解决了相关技术中无法实时预测车辆轨迹的技术问题。

Description

车辆轨迹的预测方法、装置
本申请要求于2022年02月23日提交中国专利局、优先权号为202210171001.7、发明名称为“车辆轨迹的预测方法、装置”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及智能网联自动驾驶领域，具体而言，涉及一种车辆轨迹的预测方法、装置。
背景技术
自动驾驶车辆在行驶时，需要提前对所行驶的道路及周边环境进行预测，避免发生安全事故。而目前自动驾驶车的车辆轨迹预测，主要依赖于对历史轨迹的计算,且仅适用于较短预测范围的简单驾驶情况。
针对上述的问题，目前尚未提出有效的解决方案。
发明内容
本申请实施例提供了一种车辆轨迹的预测方法、装置，以至少解决相关技术中无法实时预测车辆轨迹的技术问题。
根据本申请实施例的一个方面，提供了一种车辆轨迹的预测方法，包括:在车辆运行时，触发用于预测车辆行驶轨迹的预测程序；调取高精地图的地图特征信息，并加载至内存，其中，地图特征信息为静态数据；响应预测指令，捕获车辆运行过程中的动态感知信息，其中，动态感知信息至少包括：车辆的定位信息；基于车辆的定位信息，从内存中查询得到车辆当前所在区域的地图特征信息；将查询到的地图特征信息和动态感知信息输入至预测模型，预测得到车辆的预测轨迹。
可选地，在调取高精地图的地图特征信息之前，该方法还包括：获取至少一种版本的高精地图，并对任意一种或多种高精地图进行离线预处理；遍历每种高精地图中的每条车道，提取车道的车道中心点的特征信息，其中，特征信息包括如下至少之一：与车辆相邻的前后邻特征、左右邻特征；将车道的车道中心点的特征信息存储至目标磁盘。
可选地，调取高精地图的地图特征信息，并加载至内存，包括：启动预测程序后，基于高精地图的版本，从目标磁盘的磁盘存储文件中读取对应的地图特征信息；将调取到的地图特征信息以树状数据结构的形式加载至内存。
可选地，在将查询到的地图特征信息和动态感知信息输入至预测模型之前，该方法还包括：将内存中的地图特征信息进行张量计算，得到显存计算需要的张量信息。
可选地，该方法还包括：获取车辆在历史时间内的历史行车数据和当前定位信息；基于历史行车数据中包括的历史定位信息和当前定位信息，计算得到车辆的历史行车轨迹。
根据本申请实施例的另一方面，还提供了一种车辆轨迹的预测装置，包括：触发模块，设置为在车辆运行时，触发用于预测车辆行驶轨迹的预测程序；处理模块，设置为调取高精地图的地图特征信息，并加载至内存，其中，地图特征信息为静态数据；捕获模块，设置为响应预测指令，捕获车辆运行过程中的动态感知信息，动态感知信息至少包括：车辆的定位信息；查询模块，设置为基于车辆的定位信息，从内存中查询得到车辆当前所在区域的地图特征信息；预测模块，设置为将查询到的地图特征信息和动态感知信息输入至预测模型，预测得到车辆的预测轨迹。
可选地，获取模块，设置为获取至少一种版本的高精地图；预处理模块，设置为对任意一种或多种高精地图进行离线预处理；提取模块，设置为遍历每种高精地图中的每条车道，提取车道的车道中心点的特征信息，特征信息包括如下至少之一：与车辆相邻的前后邻特征、左右邻特征；存储模块，设置为将车道的车道中心点的特征信息存储至目标磁盘。
可选地，该装置还包括：读取模块，设置为启动预测程序后，基于高精地图的版本，从目标磁盘的磁盘存储文件中读取对应的地图特征信息；加载模块，设置为将调取到的地图特征信息以树状数据结构的形式加载至内存。
根据本申请实施例的另一方面，还提供了一种计算机可读存储介质，计算机可读存储介质包括存储的程序，其中，在程序运行时控制计算机可读存储介质所在设备执行上述实施例中的车辆轨迹的预测方法。
根据本申请实施例的另一方面，还提供了一种处理器，该处理器用于运行程序，其中，程序运行时执行上述实施例中的车辆轨迹的预测方法。
根据本申请实施例的另一方面，还提供了一种电子设备，包括：一个或多个处理器；存储装置，用于存储一个或多个程序；当一个或多个程序被一个或多个处理器执行，使得一个或多个处理器执行上述的车辆轨迹的预测方法。
在本申请实施例中，采用在车辆运行时，触发用于预测车辆行驶轨迹的预测程序；调取高精地图的地图特征信息，并加载至内存，其中，地图特征信息为静态数据；响应预测指令，捕获车辆运行过程中的动态感知信息；基于车辆的定位信息，从内存中查询得到车辆当前所在区域的地图特征信息；将查询到的地图特征信息和动态感知信 息输入至预测模型，预测得到车辆的预测轨迹的方式，通过部署预测模型，达到了实时响应的目的，从而实现了对车辆的运动轨迹实时性预警的技术效果，进而解决了相关技术中无法实时预测车辆轨迹的技术问题。
附图说明
此处所说明的附图用来提供对本申请的进一步理解，构成本申请的一部分，本申请的示意性实施例及其说明用于解释本申请，并不构成对本申请的不当限定。在附图中：
图1是根据本申请实施例的一种车辆轨迹的预测方法的流程示意图；
图2是根据本申请实施例的一种可选的LaneGCN网络动静态数据划分的示意图；
图3是根据本申请实施例的一种可选的对高精地图数据进行离线预处理的流程示意图；
图4是根据本申请实施例的一种可选的torch张量算法的流程示意图；
图5是根据本申请实施例的一种车辆预测装置的结构示意图。
具体实施方式
为了使本技术领域的人员更好地理解本申请方案，下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本申请一部分的实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都应当属于本申请保护的范围。
需要说明的是，本申请的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换，以便这里描述的本申请的实施例能够以除了在这里图示或描述的那些以外的顺序实施。此外，术语“包括”和“具有”以及他们的任何变形，意图在于覆盖不排他的包含，例如，包含了一系列步骤或单元的过程、方法、系统、产品或设备不必限于清楚地列出的那些步骤或单元，而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
实施例1
根据本申请实施例，提供了一种车辆轨迹的预测方法实施例，需要说明的是，在附图的流程图示出的步骤可以在诸如一组计算机可执行指令的计算机系统中执行，并且，虽然在流程图中示出了逻辑顺序，但是在某些情况下，可以以不同于此处的顺序执行所示出或描述的步骤。
图1是根据本申请实施例的一种车辆轨迹的预测方法的流程示意图，如图1所示， 该方法包括如下步骤：
步骤S102，在车辆运行时，触发用于预测车辆行驶轨迹的预测程序；
上述步骤中的车辆可以是用户驾驶的车辆，也可以是自动驾驶车辆，在本申请实施例中以自动驾驶车辆为例进行说明。为了实现预测车辆行驶轨迹的目的，该车辆上预先安装有导航系统，该导航系统至少包括：获取高精地图相关信息及车辆动态感知信息的导航设备、预测车辆行驶轨迹的预测设备，此处的导航设备可以是内置于车辆中的导航仪，也可以是额外安装在车辆上的导航仪；此处的预测设备可以是运行有预测程序的设备，例如，车辆内置的控制器，额外安装在车辆上的电子终端等。
可选地，为了确保车辆轨迹预测的准确性和及时性，在当前车辆启动之后，导航系统可以自动开启，并一直处于工作状态。
在一种可选的实施例中，车辆启动之后，车辆的导航系统可以自动开启，在车辆运行时，就可以触发用于预测车辆行驶轨迹的预测程序，也即，触发上述的预测设备运行预测程序。
步骤S104，调取高精地图的地图特征信息，并加载至内存，其中，地图特征信息为静态数据；
上述步骤中的高精地图是精度更高、数据维度更多的电子地图，其精度可达到20cm，数据维度包含了除道路信息之外的几乎所有与交通相关的周围静态信息，如：一定范围内的建筑物、障碍物等，这些信息在本实施例中可以是地图特征信息。
在一种可选的实施例中，当车辆启动之后，可以通过车辆的导航设备自动通过网络获取高精地图的地图特征信息，并加载至内存。
步骤S106，响应预测指令，捕获车辆运行过程中的动态感知信息，其中，动态感知信息至少包括：车辆的定位信息；
上述步骤中车辆的动态感知信息可以包括当前车道的左转、右转属性、当前车道是否有交通信号灯、当前车道的前后左右车道以及车辆的定位信息等信息，由于车辆的动态感知信息会影响车辆导航系统的及时性与准确性，用户可以根据需求进行更改。
在一种可选的实施例中，车辆启动后，预测设备可以自动开启，在车辆开始行驶之后，自动生成预测指令并发送给导航系统，导航系统可以通过导航设备、车辆上安装的摄像头等设备综合获取车辆运行过程中的动态感知信息，例如，通过导航设备通过网络捕获车辆的定位信息；通过车辆安装的摄像头可以拍摄到车辆前后方的图像，然后通过车道线识别算法确定该车辆所在的当前车道，以及前后左右其他车道；通过导航设备通过网络捕获当前车道的左转、右转属性、当前车道是否有交通信号灯等信 息。
步骤S108，基于车辆的定位信息，从内存中查询得到车辆当前所在区域的地图特征信息；
在一种可选的实施例中，在得到车辆的定位信息后，可以通过定位信息，查询提前缓存至内存的高精地图的地图特征信息，就可以得到车辆当前所在区域的高精地图的地图特征信息。
步骤S110，将查询到的地图特征信息和动态感知信息输入至预测模型，预测得到车辆的预测轨迹。
上述步骤中的预测模型可以是预测设备中的一组用于预测车辆行驶轨迹的算法模型，可以包括LaneGCN网络(Lane Graph Convolution Network，车道图卷积网络)、torch张量等算法，但不仅限于此，也可以是其他算法。
在一种可选的实施例中，在导航系统获取到车辆当前所在区域的高精地图的地图特征信息以及车辆运行过程中的动态感知信息后，可以将车辆当前所在区域的高精地图的地图特征信息和车辆运行过程中的动态感知信息传送给预测设备，预测设备将接收到的车辆当前所在区域的高精地图的地图特征信息和车辆运行过程中的动态感知信息输入至预测模型中，预测模型通过LaneGCN网络、torch张量算法等，可以计算得到车辆的预测轨迹。
通过本申请上述实施例，采用在车辆运行时，触发用于预测车辆行驶轨迹的预测程序；调取高精地图的地图特征信息，并加载至内存，其中，地图特征信息为静态数据；响应预测指令，捕获车辆运行过程中的动态感知信息；基于车辆的定位信息，从内存中查询得到车辆当前所在区域的地图特征信息；将查询到的地图特征信息和动态感知信息输入至预测模型，预测得到车辆的预测轨迹的方式，通过部署预测模型，并基于实时捕获的动态感知信息预测车辆的预测轨迹，达到了保证预测设备实时响应的目的，从而实现了对车辆的运动轨迹实时性的检测、预警的技术效果，进而解决了相关技术中无法实时预测车辆轨迹的技术问题。
可选地，在调取高精地图的地图特征信息之前，该方法还包括：获取至少一种版本的高精地图，并对任意一种或多种高精地图进行离线预处理；遍历每种高精地图中的每条车道，提取车道的车道中心点的特征信息，其中，特征信息包括如下至少之一：与车辆相邻的前后邻特征、左右邻特征；将车道的车道中心点的特征信息存储至目标磁盘。
上述的离线预处理可以是在离线状态下，对任意一种或多种高精地图的地图特征信息提前进行处理。目标磁盘可以是任何一种能够存储高精地图的地图特征信息的磁 盘，本申请对此不做具体限定，该目标磁盘可以是车辆内置的磁盘，也可以是额外安装的磁盘。
在一种可选的实施例中，可以通过车辆的导航设备通过网络获取至少一种版本的高精地图，并对任意一种或多种高精地图进行离线预处理，然后通过遍历每种高精地图中的每条车道，并提取到车道的车道中心点的特征信息，如：与车辆相邻的前后邻特征、左右邻特征等信息，最后可以将提取到的车道中心点的特征信息存储至目标磁盘，方便后续直接读取目标磁盘中的特征信息。
可选地，调取高精地图的地图特征信息，并加载至内存，包括：启动预测程序后，基于高精地图的版本，从目标磁盘的磁盘存储文件中读取对应的地图特征信息；将调取到的地图特征信息以树状数据结构的形式加载至内存。
上述的树状数据结构可以根据用户的需求选择，在本实施例中，选择以K-d树的数据结构形式进行说明。
在一种可选的实施例中，车辆启动后，导航系统自动开启，可以通过导航设备、车辆上安装的摄像头等设备综合获取车辆运行过程中的动态感知信息，同时，由于目标磁盘中存储有不同版本的地图特征信息，因此，可以基于当前使用的高精地图的版本，从目标磁盘中读取到该版本对应的的高精地图的地图特征信息，然后结合获取到的车辆运行过程中的动态感知信息，将调取到的车辆周围的地图特征信息以K-d树的数据结构形式加载至内存。
可选地，在将查询到的地图特征信息和动态感知信息输入至预测模型之前，该方法还包括：将内存中的地图特征信息进行张量计算，得到显存计算需要的张量信息。
上述中的张量计算可以根据用户的需求选择，在本实施例中，选择以torch张量计算为例进行说明。
在一种可选的实施例中，在从内存中读取到车辆动态感知信息和车辆周围的地图特征信息之后，可以对读取到的车辆动态感知信息和地图特征信息进行torch张量计算，得到显存计算需要的张量信息，然后可以将计算得到的张量信息传递给预测设备，通过预测设备可以将张量信息输入至预测模型中。
可选地，该方法还包括：获取车辆在历史时间内的历史行车数据和当前定位信息；基于历史行车数据中包括的历史定位信息和当前定位信息，计算得到车辆的历史行车轨迹。
上述的历史时间是车辆在历史行车中所需要的一段时间。
在一种可选的实施例中，车辆每一次行驶之后，可以将车辆的行车数据存储在数 据库中，作为后续导航过程中使用的历史行车数据，历史行车数据可以由形成过程中采集到的历史定位信息所构成。在每一次导航过程中，可以基于历史定位信息和当前定位信息，计算该车辆在历史行车过程中所行驶过的位置，进而生成由不同位置构成的历史行车轨迹。
下面结合图2至图4，对本申请一种可选的自动驾驶轨迹预测的方法进行详细说明。
由图2可以看出，高精地图，即静态数据独立于车辆运行轨迹，即动态数据，在确定了一个区域的高精地图后，就可以提前通过地图特征提取算法来建立好该区域内车道的特征信息。在预测程序运行时，将地图特征信息加载入内存，根据车辆的定位信息来查询地图特征数据，然后将地图特征数据和动态感知信息输入至预测模型，该预测模型的输出即为预测到的车辆预测轨迹，这样提取地图特征信息可以限制在1毫秒内。
由图3可知，对高精地图数据进行离线预处理的步骤如下：
首先，车辆启动后，导航设备自动开启；
步骤31，导航设备通过网络获取至少一种版本的需要提取特征的高精地图；
步骤32，导航设备遍历高精地图中的车道，然后进入到步骤34；
步骤33，导航设备提取车道中心点特征信息，如：与车辆相邻的前后邻特征、左右邻特征等信息；
步骤34，存储车道信息及车道中心点特征信息；
步骤35，信息存储完毕后，流程结束。
同时，运行时还需要使用torch张量计算，来获取车辆的历史运行轨迹，以及将内存中的特征信息组织成显存计算需要的张量信息。无论是CPU还是GPU，在初始计算的时候，都有一个warmup的过程，所以运行时torch张量初始计算会占用较长时间。为了消除这部分的性能抖动，在预测程序初始化时，先进行一下torch张量的计算，保证后续预测进程的流畅运行。
由图4可知，torch张量算法的处理步骤如下：
首先，车辆启动后，预测设备自动开启；
步骤41，预测程序启动后，首先根据当前所用高精地图，找出其对应的磁盘存储文件，将这些地图特征信息以K-d树的形式加载入内存；
步骤42，将预测模型加载入显存，为了减少torch张量第一次操作带来的性能损耗，对torch进行warmup，提前调用一下需要的张量运算，初始化完成；
步骤43，预测程序运行过程中，接收自动驾驶感知数据，并且可以根据感知的历史数据及当前值，计算车辆的运行历史轨迹；
步骤44，根据感知的位置信息查询获取高精地图特征信息；
步骤45，将上述动态信息及静态信息作为输入，交给预测模型进行推理；
步骤46，对推理输出进行后处理，传递预测结果给规控模块，即车辆的控制模块。
本申请公开了一个空间换时间的模型部署方案。针对预测模型预处理时间开销过高，无法满足实时预测的情况，提出一个空间换时间的模型部署方案。该方案针对预测模型的输入特点，将静态高精地图数据的计算进行离线处理，程序运行时加载进内存，后续提供实时查询，结合车辆运行的动态轨迹数据，进行预测模型推理。同时，本申请还公开了一种torch张量操作warmup策略。针对第一次torch张量操作处理速度慢的特点，将该操作在预测程序启动时处理，避免运行时时间开销过大。
实施例2
根据本申请实施例，提供了一种车辆轨迹的预测装置，该装置可以执行上述实施例1中提供的车辆轨迹预测方法，具体实现方式和优选应用场景与上述实施例1相同，在此不做赘述。
图5是根据本申请实施例的一种车辆预测装置的结构示意图，如图5所示，该装置包括：触发模块50，设置为在车辆运行时，触发用于预测车辆行驶轨迹的预测程序；处理模块52，设置为调取高精地图的地图特征信息，并加载至内存，其中，地图特征信息为静态数据；捕获模块54，设置为响应预测指令，捕获车辆运行过程中的动态感知信息，其中，动态感知信息至少包括：车辆的定位信息；查询模块56，设置为基于车辆的定位信息，从内存中查询得到车辆当前所在区域的地图特征信息；预测模块58，设置为将查询到的地图特征信息和动态感知信息输入至预测模型，预测得到车辆的预测轨迹。
此处需要说明的是，上述触发模块50、处理模块52、捕获模块54、查询模块56、预测模块58可以作为装置的一部分运行在计算机终端中，可以通过计算机终端中的处理器来执行上述模块实现的功能，计算机终端也可以是智能手机(如Android手机、IOS手机等)、平板电脑、掌上电脑以及移动互联网设备(Mobile Internet Devices，MID)、PAD等终端设备。
可选地，该装置还包括：第一获取模块，设置为获取至少一种版本的高精地图；预处理模块，设置为对任意一种或多种高精地图进行离线预处理；提取模块，设置为遍历每种高精地图中的每条车道，提取车道的车道中心点的特征信息，其中，特征信息包括如下至少之一：与车辆相邻的前后邻特征、左右邻特征；存储模块，设置为将车道的车道中心点的特征信息存储至目标磁盘。
可选地，该处理模块还包括：读取模块，设置为启动预测程序后，基于高精地图的版本，从目标磁盘的磁盘存储文件中读取对应的地图特征信息；加载模块，设置为将调取到的地图特征信息以树状数据结构的形式加载至内存。
此处需要说明的是，上述第一获取模块、预处理模块、提取模块、存储模块、读取模块、加载模块可以作为装置的一部分运行在计算机终端中，可以通过计算机终端中的处理器来执行上述模块实现的功能，计算机终端也可以是智能手机(如Android手机、IOS手机等)、平板电脑、掌上电脑以及移动互联网设备(Mobile Internet Devices，MID)、PAD等终端设备。
可选地，该装置还包括：第一计算模块，设置为将内存中的地图特征信息进行张量计算，得到显存计算需要的张量信息。
可选地，该装置还包括：第二获取模块，设置为获取车辆在历史时间内的历史行车数据和当前定位信息；第二计算模块，设置为基于历史行车数据中包括的历史定位信息和当前定位信息，计算得到车辆的历史行车轨迹。
此处需要说明的是，上述第一计算模块、第二获取模块、第二计算模块可以作为装置的一部分运行在计算机终端中，可以通过计算机终端中的处理器来执行上述模块实现的功能，计算机终端也可以是智能手机(如Android手机、IOS手机等)、平板电脑、掌上电脑以及移动互联网设备(Mobile Internet Devices，MID)、PAD等终端设备。
实施例3
根据本申请实施例，还提供了一种计算机可读存储介质，计算机可读存储介质包括存储的程序，其中，在程序运行时控制计算机可读存储介质所在设备执行上述实施例中的车辆轨迹的预测方法。
实施例4
根据本申请实施例，还提供了一种处理器，处理器用于运行程序，其中，程序运行时执行上述实施例中的车辆轨迹的预测方法。
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。
在本申请的上述实施例中，对各个实施例的描述都各有侧重，某个实施例中没有详述的部分，可以参见其他实施例的相关描述。
在本申请所提供的几个实施例中，应该理解到，所揭露的技术内容，可通过其它的方式实现。其中，以上所描述的装置实施例仅仅是示意性的，例如所述单元的划分，可以为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口，单元或模块的间接耦合或通信连接，可以是电性或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外，在本申请各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。
所述集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。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或者该技术方案的全部或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中，包括若干指令用以使得一台计算机设备(可为个人计算机、服务器或者网络设备等)执行本申请各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、移动硬盘、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述仅是本申请的优选实施方式，应当指出，对于本技术领域的普通技术人员来说，在不脱离本申请原理的前提下，还可以做出若干改进和润饰，这些改进和润饰也应视为本申请的保护范围。
工业实用性
本申请实施例中提供的方案，可应用于智能网联自动驾驶领域中。通过采用在车辆运行时，触发用于预测车辆行驶轨迹的预测程序；调取高精地图的地图特征信息，并加载至内存，其中，地图特征信息为静态数据；响应预测指令，捕获车辆运行过程中的动态感知信息；基于车辆的定位信息，从内存中查询得到车辆当前所在区域的地图特征信息；将查询到的地图特征信息和动态感知信息输入至预测模型，预测得到车 辆的预测轨迹的方式，通过部署预测模型，达到了实时响应的目的，从而实现了对车辆的运动轨迹实时性预警的技术效果，进而解决了相关技术中无法实时预测车辆轨迹的技术问题。

Claims (18)

  1. 一种车辆轨迹的预测方法，包括：
    在车辆运行时，触发用于预测车辆行驶轨迹的预测程序；
    调取高精地图的地图特征信息，并加载至内存，其中，所述地图特征信息为静态数据；
    响应预测指令，捕获所述车辆运行过程中的动态感知信息，其中，所述动态感知信息至少包括：所述车辆的定位信息；
    基于所述车辆的定位信息，从所述内存中查询得到所述车辆当前所在区域的地图特征信息；
    将查询到的所述地图特征信息和所述动态感知信息输入至预测模型，预测得到所述车辆的预测轨迹。
  2. 根据权利要求1所述的方法，其中，在调取高精地图的地图特征信息之前，所述方法还包括：
    获取至少一种版本的高精地图，并对任意一种或多种高精地图进行离线预处理；
    遍历每种高精地图中的每条车道，提取所述车道的车道中心点的特征信息，其中，所述特征信息包括如下至少之一：与所述车辆相邻的前后邻特征、左右邻特征；
    将所述车道的车道中心点的特征信息存储至目标磁盘。
  3. 根据权利要求2所述的方法，其中，调取高精地图的地图特征信息，并加载至内存，包括：
    启动所述预测程序后，基于所述高精地图的版本，从所述目标磁盘的磁盘存储文件中读取对应的地图特征信息；
    将调取到的所述地图特征信息以树状数据结构的形式加载至所述内存。
  4. 根据权利要求3所述的方法，其中，在将查询到的所述地图特征信息和所述动态感知信息输入至预测模型之前，所述方法还包括：
    将所述内存中的所述地图特征信息进行张量计算，得到显存计算需要的张量信息。
  5. 根据权利要求1所述的方法，其中，所述方法还包括：获取所述车辆在历史时间内的历史行车数据和当前定位信息；基于所述历史行车数据中包括的历史定位信息和所述当前定位信息，计算得到所述车辆的所述历史行车轨迹。
  6. 一种车辆轨迹的预测装置，包括：
    触发模块，设置为在车辆运行时，触发用于预测车辆行驶轨迹的预测程序；
    处理模块，设置为调取高精地图的地图特征信息，并加载至内存，其中，所述地图特征信息为静态数据；
    捕获模块，设置为响应预测指令，捕获所述车辆运行过程中的动态感知信息，其中，所述动态感知信息至少包括：所述车辆的定位信息；
    查询模块，设置为基于所述车辆的定位信息，从所述内存中查询得到所述车辆当前所在区域的地图特征信息；
    预测模块，设置为将查询到的所述地图特征信息和所述动态感知信息输入至预测模型，预测得到所述车辆的预测轨迹。
  7. 根据权利要求6所述的装置，其中，所述装置还包括：
    获取模块，设置为获取至少一种版本的高精地图；
    预处理模块，设置为对任意一种或多种高精地图进行离线预处理；
    提取模块，设置为遍历每种高精地图中的每条车道，提取所述车道的车道中心点的特征信息，其中，所述特征信息包括如下至少之一：与所述车辆相邻的前后邻特征、左右邻特征；
    存储模块，设置为将所述车道的车道中心点的特征信息存储至目标磁盘。
  8. 根据权利要求7所述的装置，其中，所述处理模块包括：
    读取模块，设置为启动所述预测程序后，基于所述高精地图的版本，从所述目标磁盘的磁盘存储文件中读取对应的地图特征信息；
    加载模块，设置为将调取到的所述地图特征信息以树状数据结构的形式加载至所述内存。
  9. 一种计算机可读存储介质，所述计算机可读存储介质包括存储的程序，其中，在所述程序运行时控制所述计算机可读存储介质所在设备执行如下方法：
    在车辆运行时，触发用于预测车辆行驶轨迹的预测程序；
    调取高精地图的地图特征信息，并加载至内存，其中，所述地图特征信息为静态数据；
    响应预测指令，捕获所述车辆运行过程中的动态感知信息，其中，所述动态感知信息至少包括：所述车辆的定位信息；
    基于所述车辆的定位信息，从所述内存中查询得到所述车辆当前所在区域的地图特征信息；
    将查询到的所述地图特征信息和所述动态感知信息输入至预测模型，预测得到所述车辆的预测轨迹。
  10. 如权利要求9所述的存储介质，在所述程序运行时控制所述计算机可读存储介质所在设备执行如下方法：
    获取至少一种版本的高精地图，并对任意一种或多种高精地图进行离线预处理；
    遍历每种高精地图中的每条车道，提取所述车道的车道中心点的特征信息，其中，所述特征信息包括如下至少之一：与所述车辆相邻的前后邻特征、左右邻特征；
    将所述车道的车道中心点的特征信息存储至目标磁盘。
  11. 如权利要求10所述的存储介质，在所述程序运行时控制所述计算机可读存储介质所在设备执行如下方法：
    启动所述预测程序后，基于所述高精地图的版本，从所述目标磁盘的磁盘存储文件中读取对应的地图特征信息；
    将调取到的所述地图特征信息以树状数据结构的形式加载至所述内存。
  12. 如权利要求11所述的存储介质，在所述程序运行时控制所述计算机可读存储介质所在设备执行如下方法：
    将所述内存中的所述地图特征信息进行张量计算，得到显存计算需要的张量信息。
  13. 如权利要求9所述的存储介质，在所述程序运行时控制所述计算机可读存储介质所在设备执行如下方法：
    获取所述车辆在历史时间内的历史行车数据和当前定位信息；基于所述历史行车数据中包括的历史定位信息和所述当前定位信息，计算得到所述车辆的所述历史行车轨迹。
  14. 一种处理器，所述处理器用于运行程序，其中，所述程序运行时执行如下方法：
    在车辆运行时，触发用于预测车辆行驶轨迹的预测程序；
    调取高精地图的地图特征信息，并加载至内存，其中，所述地图特征信息为静态数据；
    响应预测指令，捕获所述车辆运行过程中的动态感知信息，其中，所述动态感知信息至少包括：所述车辆的定位信息；
    基于所述车辆的定位信息，从所述内存中查询得到所述车辆当前所在区域的地图特征信息；
    将查询到的所述地图特征信息和所述动态感知信息输入至预测模型，预测得到所述车辆的预测轨迹。
  15. 根据权利要求14所述的处理器，其中，所述程序运行时执行如下方法：
    获取至少一种版本的高精地图，并对任意一种或多种高精地图进行离线预处理；
    遍历每种高精地图中的每条车道，提取所述车道的车道中心点的特征信息，其中，所述特征信息包括如下至少之一：与所述车辆相邻的前后邻特征、左右邻特征；
    将所述车道的车道中心点的特征信息存储至目标磁盘。
  16. 根据权利要求15所述的处理器，其中，所述程序运行时执行如下方法：
    启动所述预测程序后，基于所述高精地图的版本，从所述目标磁盘的磁盘存储文件中读取对应的地图特征信息；
    将调取到的所述地图特征信息以树状数据结构的形式加载至所述内存。
  17. 根据权利要求16所述的处理器，其中，所述程序运行时执行如下方法：
    将所述内存中的所述地图特征信息进行张量计算，得到显存计算需要的张量信息。
  18. 根据权利要求14所述的处理器，其中，所述程序运行时执行如下方法：
    获取所述车辆在历史时间内的历史行车数据和当前定位信息；基于所述历史行车数据中包括的历史定位信息和所述当前定位信息，计算得到所述车辆的所述历史行车轨迹。
PCT/CN2022/117878 2022-02-23 2022-09-08 车辆轨迹的预测方法、装置 WO2023159915A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210171001.7 2022-02-23
CN202210171001.7A CN114637770A (zh) 2022-02-23 2022-02-23 车辆轨迹的预测方法、装置

Publications (1)

Publication Number Publication Date
WO2023159915A1 true WO2023159915A1 (zh) 2023-08-31

Family

ID=81948530

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/117878 WO2023159915A1 (zh) 2022-02-23 2022-09-08 车辆轨迹的预测方法、装置

Country Status (2)

Country Link
CN (1) CN114637770A (zh)
WO (1) WO2023159915A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114637770A (zh) * 2022-02-23 2022-06-17 中国第一汽车股份有限公司 车辆轨迹的预测方法、装置
CN115390103A (zh) * 2022-08-29 2022-11-25 智道网联科技(北京)有限公司 卫星定位信号的异常检测方法、装置及电子设备

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110400490A (zh) * 2019-08-08 2019-11-01 腾讯科技（深圳）有限公司 轨迹预测方法和装置
CN112212874A (zh) * 2020-11-09 2021-01-12 福建牧月科技有限公司 车辆轨迹预测方法、装置、电子设备及计算机可读介质
WO2021062596A1 (en) * 2019-09-30 2021-04-08 Beijing Voyager Technology Co., Ltd. Systems and methods for predicting a vehicle trajectory
US20210174668A1 (en) * 2019-12-10 2021-06-10 Samsung Electronics Co., Ltd. Systems and methods for trajectory prediction
CN113753038A (zh) * 2021-03-16 2021-12-07 京东鲲鹏(江苏)科技有限公司 一种轨迹预测方法、装置、电子设备和存储介质
CN114637770A (zh) * 2022-02-23 2022-06-17 中国第一汽车股份有限公司 车辆轨迹的预测方法、装置

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110400490A (zh) * 2019-08-08 2019-11-01 腾讯科技（深圳）有限公司 轨迹预测方法和装置
WO2021062596A1 (en) * 2019-09-30 2021-04-08 Beijing Voyager Technology Co., Ltd. Systems and methods for predicting a vehicle trajectory
US20210174668A1 (en) * 2019-12-10 2021-06-10 Samsung Electronics Co., Ltd. Systems and methods for trajectory prediction
CN112212874A (zh) * 2020-11-09 2021-01-12 福建牧月科技有限公司 车辆轨迹预测方法、装置、电子设备及计算机可读介质
CN113753038A (zh) * 2021-03-16 2021-12-07 京东鲲鹏(江苏)科技有限公司 一种轨迹预测方法、装置、电子设备和存储介质
CN114637770A (zh) * 2022-02-23 2022-06-17 中国第一汽车股份有限公司 车辆轨迹的预测方法、装置

Also Published As

Publication number Publication date
CN114637770A (zh) 2022-06-17

Similar Documents

Publication Publication Date Title
WO2023159915A1 (zh) 车辆轨迹的预测方法、装置
WO2022156520A1 (zh) 一种云路协同的自动驾驶模型训练、调取方法及系统
US11281227B2 (en) Method of pedestrian activity recognition using limited data and meta-learning
AU2019200934A1 (en) Autonomous vehicle operated with guide assistance
US8880282B2 (en) Method and system for risk prediction for a support actuation system
EP3893148A1 (en) Method and device for controlling vehicle, and vehicle
WO2020052277A1 (zh) 用于控制车辆和自主驾驶车辆的方法和装置
KR102143034B1 (ko) 객체의 미래 움직임 예측을 통한 동영상에서의 객체 추적을 위한 방법 및 시스템
US11250279B2 (en) Generative adversarial network models for small roadway object detection
CN112824182A (zh) 自动泊车方法、装置、计算机设备和存储介质
CN114511632A (zh) 车位地图的构建方法及装置
CN112602319A (zh) 一种对焦装置、方法及相关设备
US20240092385A1 (en) Driving Policy Determining Method and Apparatus, Device, and Vehicle
CN110971826A (zh) 视频前端监控装置及方法
CN116681739A (zh) 目标运动轨迹生成方法、装置及电子设备
JP7450754B2 (ja) 画像解析から得られたフィンガープリントを用いた、画像フレーム全体に亘る脆弱な道路利用者の追跡
KR102143031B1 (ko) 정지 영상에서 객체의 미래 움직임을 예측하는 방법 및 시스템
CN112989194A (zh) 一种车联网用户请求与服务偏好融合的推荐方法及系统
CN112215042A (zh) 一种车位限位器识别方法及其系统、计算机设备
CN111104937A (zh) 车门信息检测方法、装置、计算机设备和存储介质
JP7309817B2 (ja) 自動車の車体の運動を検出するための方法、システム及びコンピュータプログラム
CN110276322B (zh) 一种结合车机闲散资源的图像处理方法及装置
CN115797412B (zh) 动态对象异常值并行检测方法、装置、系统、设备及介质
CN111959517B (zh) 一种距离的提示方法、装置、计算机设备和存储介质
CN114238681A (zh) 图像检索方法、计算机程序产品、存储介质及电子设备

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: 22928181

Country of ref document: EP

Kind code of ref document: A1