WO2023137864A1 - 基于自动驾驶的感知决策模型升级方法、系统及电子装置 - Google Patents

基于自动驾驶的感知决策模型升级方法、系统及电子装置 Download PDF

Info

Publication number
WO2023137864A1
WO2023137864A1 PCT/CN2022/083064 CN2022083064W WO2023137864A1 WO 2023137864 A1 WO2023137864 A1 WO 2023137864A1 CN 2022083064 W CN2022083064 W CN 2022083064W WO 2023137864 A1 WO2023137864 A1 WO 2023137864A1
Authority
WO
WIPO (PCT)
Prior art keywords
scene
information
preset
model
perception
Prior art date
Application number
PCT/CN2022/083064
Other languages
English (en)
French (fr)
Inventor
李国庆
邓文武
宋升弘
卢红喜
金晨
周俊杰
衡阳
Original Assignee
浙江吉利控股集团有限公司
吉利汽车研究院（宁波）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 浙江吉利控股集团有限公司, 吉利汽车研究院（宁波）有限公司 filed Critical 浙江吉利控股集团有限公司
Publication of WO2023137864A1 publication Critical patent/WO2023137864A1/zh

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G07—CHECKING-DEVICES
    • G07C—TIME OR ATTENDANCE REGISTERS; REGISTERING OR INDICATING THE WORKING OF MACHINES; GENERATING RANDOM NUMBERS; VOTING OR LOTTERY APPARATUS; ARRANGEMENTS, SYSTEMS OR APPARATUS FOR CHECKING NOT PROVIDED FOR ELSEWHERE
    • G07C5/00—Registering or indicating the working of vehicles
    • G07C5/08—Registering or indicating performance data other than driving, working, idle, or waiting time, with or without registering driving, working, idle or waiting time
    • G07C5/0808—Diagnosing performance data

Definitions

  • the present application relates to the field of vehicle technology, in particular to a method, system and electronic device for upgrading a perception decision model based on automatic driving.
  • the main purpose of this application is to provide an automatic driving-based perception decision-making model upgrade method, system and electronic device, aiming at solving the technical problem of how to improve the reliability of mass-produced vehicle functions and algorithms.
  • the present application provides a method for upgrading a perception decision model based on automatic driving.
  • the method for upgrading a perception decision model based on automatic driving includes:
  • the preset perception decision model is upgraded according to the failure data of the perception decision.
  • the step of obtaining the road scene information to be tested it also includes:
  • a preset perception decision model is constructed according to the preset decision regulation model and the preset perception model.
  • the step of acquiring a plurality of driving manipulation data samples and a plurality of scene environment data samples includes:
  • the step of judging whether a plurality of scene flow identification information satisfies a preset identification condition it further includes:
  • Scene flow labeling is performed on a plurality of scene sample data to be processed, respectively, to obtain a plurality of scene sample data to be trained;
  • the preset scene flow recognition model is upgraded according to the plurality of scene sample data to be trained.
  • the step of determining the perception decision failure data according to the vehicle perception decision information it includes:
  • the step of determining perception decision failure data according to the vehicle perception decision information is executed.
  • the step of judging whether the vehicle perception decision information satisfies a preset failure condition it further includes:
  • the preset perception decision model is upgraded based on a plurality of failure data of the perception decision.
  • the step of upgrading the preset perception decision model based on a plurality of failure data of the perception decision includes:
  • the preset perception decision model is upgraded according to the plurality of perception decision annotation data.
  • the present application also proposes an automatic driving-based perception decision-making model upgrading system, the automatic driving-based perception decision-making model upgrading system including:
  • An acquisition module configured to acquire road scene information to be tested
  • the analysis module is configured to perform failure analysis on the road scene information to be tested based on a preset perception decision model to obtain vehicle perception decision information;
  • a determination module configured to determine perception decision failure data according to the vehicle perception decision information
  • the upgrading module is configured to upgrade the preset perception decision model according to the failure data of the perception decision.
  • the system for upgrading the perception decision model based on automatic driving further includes a building block
  • the building module is configured to acquire a plurality of driving manipulation data samples and a plurality of scene-based environment data samples;
  • the building module is further configured to train an initial decision regulation model based on a plurality of driving manipulation data samples to obtain a preset decision regulation model;
  • the building module is further configured to train an initial perception model based on a plurality of the scene-based environment data samples to obtain a preset perception model;
  • the construction module is further configured to construct a preset perception decision model according to the preset decision regulation model and the preset perception model.
  • the acquiring module is further configured to acquire a plurality of road scene sample information
  • the acquiring module is further configured to input a plurality of road scene sample information into a preset scene flow identification model to obtain a plurality of scene flow identification information;
  • the acquisition module is further configured to judge whether a plurality of the scene flow identification information meets a preset identification condition
  • the acquisition module is further configured to, when the plurality of scene identification flow information meet the preset identification condition, perform data cleaning on the plurality of scene identification information respectively, and obtain a plurality of driving manipulation data samples and a plurality of scene environment data samples.
  • the application first obtains the road scene information to be tested, and then conducts failure analysis on the road scene information to be tested based on the preset perception decision model to obtain the vehicle perception decision information, and then determines the perception decision failure data according to the vehicle perception decision information, and upgrades the preset perception decision model based on the perception decision failure data.
  • the preset perception decision-making model can be upgraded according to the perception decision-making failure data corresponding to the vehicle perception decision-making information, thereby improving the reliability of the functions and algorithms of mass-produced models.
  • FIG. 1 is a schematic flow diagram of the first embodiment of the method for upgrading the perception decision model based on automatic driving in the present application
  • FIG. 2 is a schematic diagram of the scene-based indexing and data cleaning system of the first embodiment of the automatic driving-based perception decision-making model upgrading method of the present application;
  • FIG. 3 is a schematic flow diagram of the second embodiment of the automatic driving-based perception decision-making model upgrading method of the present application
  • FIG. 4 is a schematic diagram of the algorithm data-driven and closed-loop automation system of the second embodiment of the automatic driving-based perception decision-making model upgrading method of the present application;
  • Fig. 5 is a structural block diagram of the first embodiment of the automatic driving-based perception decision-making model upgrading system of the present application.
  • FIG. 1 is a schematic flowchart of a first embodiment of a method for upgrading a perception decision model based on automatic driving in the present application.
  • the method for upgrading the perception decision model based on automatic driving includes the following steps:
  • Step S10 Obtain the road scene information to be tested.
  • the execution subject of this embodiment may be an automatic driving-based perception decision-making model upgrading device with functions such as image processing, data processing, network communication, and program operation.
  • the automatic driving-based perception decision-making model upgrading system includes a scene-based indexing and data cleaning system, an algorithm data-driven and closed-loop automation system, etc., and may also be other computer devices with similar functions, which are not limited in this embodiment.
  • the road scene information to be tested may be road environment information collected when the vehicle is driving. Before the step of obtaining the road scene information to be tested, it is necessary to obtain multiple driving manipulation data samples and multiple scene-based environment data samples, train the initial decision-making regulatory model based on the multiple driving-manipulation data samples, obtain a preset decision-making regulation model, train the initial perception model based on multiple scene-based environmental data samples, obtain a preset perception model, and then construct a preset perception decision-making model based on the preset decision-making regulation model and the preset perception model.
  • the driving control data sample can be understood as the vehicle driving control command generated when unmanned driving is realized
  • the scene environment data sample can be understood as the road environment information and weather information corresponding to the vehicle driving control command.
  • the processing method for obtaining multiple driving manipulation data samples and multiple scene environment data samples can be to obtain multiple road scene sample information, input the multiple road scene sample information into the preset scene flow recognition model, obtain multiple scene flow recognition information, judge whether the multiple scene flow recognition information meets the preset recognition conditions, and respectively perform data cleaning on the multiple scene flow recognition information when the multiple scene recognition flow information meets the preset recognition conditions, and obtain multiple driving manipulation data samples and multiple scene environment data samples.
  • the preset recognition condition can be understood as the absence of unrecognized scene streams and the like.
  • road scene sample information can be understood as road environment information, weather environment information, and vehicle driving manipulation data.
  • Scene flow identification information includes general scene information, extreme scene information, and unrecognized scene information. For example, tunnels, high-speed ramps, rain, snow and fog, and other corner-case data.
  • multiple road scene sample information can be input into the preset scene recognition flow model, and the multiple road scene sample information is subjected to scene recognition and scene indexing, and is divided into regular scene index information, extreme scene index information, and unrecognized scene flow.
  • the preset scene stream recognition model is trained with the scene stream, and the Over the Air Technology (OTA) upgrade of the trained preset scene stream recognition model is completed.
  • OTA Over the Air Technology
  • the multiple road scene sample information there is no unrecognized scene flow in the multiple road scene sample information, and it is determined that the preset recognition conditions are met, and it is necessary to perform regular scene indexing and extreme scene indexing on the multiple road scene sample information, and then structure the data of the indexed road scene sample information, and perform environmental data cleaning on the structured road scene sample information to obtain multiple driving manipulation data samples and multiple scene-based environment data samples, and finally, it is necessary to classify and store the multiple driving manipulation data samples and multiple scene-based environment data samples.
  • Fig. 2 is a schematic diagram of the scene indexing and data cleaning system of the first embodiment of the automatic driving-based perception decision model upgrading method of the present application.
  • the user can collect a plurality of road scene sample information through the sensor installed on the vehicle, wherein the road scene sample information includes vehicle handling data, road environment and weather information, etc., and then input the collected multiple road environment sample information into the preset scene flow recognition model, so that the preset scene flow recognition model can perform scene recognition on the multiple road scene sample information, and classify the multiple road scene sample information according to the scene recognition results, and construct a regular scene index and an extreme scene index. and unrecognized scene flow database.
  • the data corresponding to the regular scene index and the data corresponding to the extreme scene index are data structured, and classified according to the structured data to obtain the driving manipulation data and the scene environment data to be processed. After that, the scene environment data to be processed needs to be cleaned to obtain the scene environment data.
  • the driving control data and the scene environment data are stored separately, and the driving control data is used as a driving control data sample, and the scene environment data is used as a scene environment data sample.
  • the preset scene flow recognition model in this embodiment can provide end-to-end scene data from the perspective of model training, and use scene labels as indexes for efficient access, which greatly improves the data screening efficiency of model training, and it is easier to complete the training of specific algorithm models and the evaluation of inference algorithms by taking scene data on demand.
  • the introduction of scene-based indexing technology in the front-end of data collection greatly improves the efficiency of data screening and scene selection during back-end algorithm model training.
  • Step S20 Perform failure analysis on the road scene information to be tested based on a preset perception decision model to obtain vehicle perception decision information.
  • the preset perception decision-making model is a combined model of the preset decision-making regulation and control model and the preset perception model.
  • the road scene information to be tested can be input into the preset perception decision-making model to obtain vehicle perception decision-making information.
  • vehicle perception decision-making information can be understood as environmental perception information and vehicle automatic driving information output by the preset perception decision-making model according to the road scene information to be tested.
  • Step S30 Determine perception decision failure data according to the vehicle perception decision information.
  • the preset failure condition is that the vehicle perception decision information is inconsistent with the scene environment data information of the driver operation data set corresponding to the road scene information to be tested.
  • Perception decision failure data is the data obtained by the user from failure analysis of vehicle perception decision information.
  • the vehicle perception decision information is A
  • the driver operation data set scene environment data corresponding to the road scene information to be tested is B
  • the vehicle perception decision information A is inconsistent with the driver operation data set scene environment data B corresponding to the road scene information to be tested
  • the road scene information to be tested and the vehicle perception decision information can be used as the perception decision failure data.
  • Step S40 Perform model upgrade on the preset perception decision model according to the failure data of the perception decision.
  • the failure data of the perception decision-making will be marked, so that the failure data after marking can be used for training the preset perception decision-making model, etc.
  • the vehicle perception decision information does not meet the preset failure conditions, it is determined that the vehicle perception decision information is correct. It is necessary to obtain the indicators and function evaluation report of the preset perception decision model based on the vehicle perception decision information, and based on the preset perception decision model indicators and function evaluation report, the preset perception decision model is mass-produced and deployed to obtain multiple vehicle perception decision information.
  • the processing method of upgrading the preset perception decision model based on multiple perception decision failure data may be respectively labeling multiple perception decision failure data, obtaining multiple perception decision annotation data, and upgrading the preset perception decision model according to the multiple perception decision annotation data.
  • the road scene information to be tested is first obtained, and then failure analysis is performed on the road scene information to be tested based on the preset perception decision model to obtain vehicle perception decision information, and then the perception decision failure data is determined according to the vehicle perception decision information, and the preset perception decision model is upgraded according to the perception decision failure data.
  • the preset perception decision-making model can be upgraded according to the perception decision-making failure data corresponding to the vehicle perception decision-making information, thereby improving the reliability of the functions and algorithms of mass-produced models, and further improving the user experience.
  • FIG. 3 is a schematic flowchart of a second embodiment of an automatic driving-based perception decision model upgrading method of the present application.
  • step S30 further includes:
  • Step S301 judging whether the vehicle perception decision information satisfies a preset invalidation condition.
  • the preset failure condition is that the vehicle perception decision information is inconsistent with the scene environment data information of the driver operation data set corresponding to the road scene information to be tested.
  • the vehicle perception decision information does not meet the preset failure conditions, it is determined that the vehicle perception decision information is correct. It is necessary to obtain the indicators and function evaluation report of the preset perception decision model based on the vehicle perception decision information, and based on the preset perception decision model indicators and function evaluation report, the preset perception decision model is mass-produced and deployed to obtain multiple vehicle perception decision information.
  • the processing method of upgrading the preset perception decision model based on multiple perception decision failure data may be respectively labeling multiple perception decision failure data, obtaining multiple perception decision annotation data, and upgrading the preset perception decision model according to the multiple perception decision annotation data.
  • Step S302 When the vehicle perception decision information satisfies the preset invalidation condition, determine perception decision failure data according to the vehicle perception decision information.
  • Perception decision failure data is the data obtained by the user from failure analysis of vehicle perception decision information.
  • the vehicle perception decision information is A
  • the driver operation data set scene environment data corresponding to the road scene information to be tested is B
  • the vehicle perception decision information A is inconsistent with the driver operation data set scene environment data B corresponding to the road scene information to be tested
  • the road scene information to be tested and the vehicle perception decision information can be used as the perception decision failure data.
  • the vehicle perception decision information is incorrect, it is determined that the vehicle perception decision information satisfies the preset failure conditions, and the failure data of the perception decision is determined according to the vehicle perception decision information, and the failure data of the perception decision is marked, so that the failure data after marking can be used for training the preset perception decision model, etc.
  • Fig. 4 is a schematic diagram of the algorithm data-driven and closed-loop automation system of the second embodiment of the automatic driving-based perception decision-making model upgrading method of the present application.
  • the user can input the road scene information to be tested into the preset perception decision-making model to obtain corresponding vehicle perception decision-making information, and then perform failure analysis on the vehicle perception decision-making information.
  • the vehicle perception decision information is incorrect, it is determined that the vehicle perception decision information satisfies the preset failure conditions, and the failure data of the perception decision is determined according to the vehicle perception decision information, and the failure data of the perception decision is marked, so that the marked failure data can be used for training the preset perception decision model, etc.
  • the closed-loop design system in the algorithm data-driven and closed-loop automation schematic diagram supports data collection and algorithm model upgrades for mass-produced models. It can collect the data of the deployed algorithm model or function in the reasoning failure of the mass-produced model during the customer's use (within one minute before and after the failure time), and conduct failure mode analysis on the recorded "mode mismatch" data, and carry out targeted customized training, and upgrade the software algorithm through OTA. It can greatly improve the reliability and user experience of the functions and algorithms of mass-produced models.
  • the vehicle perception decision information it is judged whether the vehicle perception decision information satisfies the preset failure condition.
  • the failure data of the perception decision is determined according to the vehicle perception decision information.
  • the obtained vehicle perception decision information is directly used without further verification of the vehicle perception decision information.
  • an embodiment of the present application also proposes a storage medium, on which an automatic driving-based perception decision-making model upgrading program is stored, and when the automatic driving-based perception decision-making model upgrading program is executed by a processor, the steps of the automatic driving-based perception decision-making model upgrading method as described above are implemented.
  • an embodiment of the present application also proposes an electronic device, which may be, for example, a mobile terminal or a fixed terminal.
  • the electronic device includes a processor and a memory, and the memory stores an automatic driving-based perception decision-making model upgrading program, and the processor is used to run the automatic driving-based perception decision-making model upgrading program to implement the steps of the automatic driving-based perception decision-making model upgrading method described above.
  • FIG. 5 is a structural block diagram of the first embodiment of the automatic driving-based perception decision model upgrading system of the present application.
  • the automatic driving-based perception decision-making model upgrade system proposed in the embodiment of the present application includes:
  • the obtaining module 5001 is configured to obtain the road scene information to be tested.
  • the road scene information to be tested may be road environment information collected when the vehicle is driving. Before the step of obtaining the road scene information to be tested, it is necessary to obtain multiple driving manipulation data samples and multiple scene-based environment data samples, train the initial decision-making regulatory model based on the multiple driving-manipulation data samples, obtain a preset decision-making regulation model, train the initial perception model based on multiple scene-based environmental data samples, obtain a preset perception model, and then construct a preset perception decision-making model based on the preset decision-making regulation model and the preset perception model.
  • the driving control data sample can be understood as the vehicle driving control command generated when unmanned driving is realized
  • the scene environment data sample can be understood as the road environment information and weather information corresponding to the vehicle driving control command.
  • the processing method for obtaining multiple driving manipulation data samples and multiple scene environment data samples can be to obtain multiple road scene sample information, input the multiple road scene sample information into the preset scene flow recognition model, obtain multiple scene flow recognition information, judge whether the multiple scene flow recognition information meets the preset recognition conditions, and respectively perform data cleaning on the multiple scene flow recognition information when the multiple scene recognition flow information meets the preset recognition conditions, and obtain multiple driving manipulation data samples and multiple scene environment data samples.
  • the preset recognition condition can be understood as the absence of unrecognized scene streams and the like.
  • road scene sample information can be understood as road environment information, weather environment information, and vehicle driving manipulation data.
  • Scene flow identification information includes general scene information, extreme scene information, and unrecognized scene information. For example, tunnels, high-speed ramps, rain, snow and fog, and other corner-case data.
  • multiple road scene sample information can be input into the preset scene recognition flow model, and the multiple road scene sample information is subjected to scene recognition and scene indexing, and is divided into regular scene index information, extreme scene index information, and unrecognized scene flow.
  • the scene flow of the preset scene flow recognition model is trained, and the OTA upgrade of the trained preset scene flow recognition model is completed.
  • the multiple road scene sample information there is no unrecognized scene flow in the multiple road scene sample information, and it is determined that the preset recognition conditions are met, and it is necessary to perform regular scene indexing and extreme scene indexing on the multiple road scene sample information, and then structure the data of the indexed road scene sample information, and perform environmental data cleaning on the structured road scene sample information to obtain multiple driving manipulation data samples and multiple scene-based environment data samples, and finally, it is necessary to classify and store the multiple driving manipulation data samples and multiple scene-based environment data samples.
  • Fig. 2 is a schematic diagram of the scenario-based indexing and data cleaning system of the first embodiment of the automatic driving-based perception decision-making model upgrading method of the present application.
  • the user can collect a plurality of road scene sample information through the sensor installed on the vehicle, wherein the road scene sample information includes vehicle handling data, road environment and weather information, etc., and then input the collected multiple road environment sample information into the preset scene flow recognition model, so that the preset scene flow recognition model can perform scene recognition on multiple road scene sample information, and classify the multiple road scene sample information according to the scene recognition results and construct a conventional scene index, an extreme scene index and Unrecognized scene flow database, in the specific implementation, the data corresponding to the regular scene index and the data corresponding to the extreme scene index are structured, and classified according to the structured data to obtain the driving control data and the scene environment data to be processed, and then need to clean the scene environment data to be processed to obtain the scene environment data, and finally store the driving control data and the scene environment data separately, and use the driving control
  • the preset scene flow recognition model in this embodiment can provide end-to-end scene data from the perspective of model training, and use scene labels as indexes for efficient access, which greatly improves the data screening efficiency of model training, and it is easier to complete the training of specific algorithm models and the evaluation of inference algorithms by taking scene data on demand.
  • the introduction of scene-based indexing technology in the front-end of data collection greatly improves the efficiency of data screening and scene selection during back-end algorithm model training.
  • the analysis module 5002 is configured to perform failure analysis on the road scene information to be tested based on a preset perception decision model to obtain vehicle perception decision information.
  • the preset perception decision-making model is a combined model of the preset decision-making regulation and control model and the preset perception model.
  • the road scene information to be tested can be input into the preset perception decision-making model to obtain vehicle perception decision-making information.
  • vehicle perception decision-making information can be understood as environmental perception information and vehicle automatic driving information output by the preset perception decision-making model according to the road scene information to be tested.
  • the determination module 5003 is configured to determine perception decision failure data according to the vehicle perception decision information.
  • the preset failure condition is that the vehicle perception decision information is inconsistent with the scene environment data information of the driver operation data set corresponding to the road scene information to be tested.
  • Perception decision failure data is the data obtained by the user from failure analysis of vehicle perception decision information.
  • the vehicle perception decision information is A
  • the driver operation data set scene environment data corresponding to the road scene information to be tested is B
  • the vehicle perception decision information A is inconsistent with the driver operation data set scene environment data B corresponding to the road scene information to be tested
  • the road scene information to be tested and the vehicle perception decision information can be used as the perception decision failure data.
  • the upgrading module 5004 is configured to upgrade the preset perception decision model according to the failure data of the perception decision.
  • the failure data of the perception decision is marked, so that the failure data after marking can be used for training the preset perception decision model, etc.
  • the vehicle perception decision information does not meet the preset failure conditions, it is determined that the vehicle perception decision information is correct. It is necessary to obtain the indicators and function evaluation report of the preset perception decision model based on the vehicle perception decision information, and based on the preset perception decision model indicators and function evaluation report, the preset perception decision model is mass-produced and deployed to obtain multiple vehicle perception decision information.
  • the processing method of upgrading the preset perception decision model based on multiple perception decision failure data may be respectively labeling multiple perception decision failure data, obtaining multiple perception decision annotation data, and upgrading the preset perception decision model according to the multiple perception decision annotation data.
  • the road scene information to be tested is first obtained, and then failure analysis is performed on the road scene information to be tested based on the preset perception decision model to obtain vehicle perception decision information, and then the perception decision failure data is determined according to the vehicle perception decision information, and the preset perception decision model is upgraded according to the perception decision failure data.
  • the preset perception decision-making model can be upgraded according to the perception decision-making failure data corresponding to the vehicle perception decision-making information, thereby improving the reliability of the functions and algorithms of mass-produced models, and further improving the user experience.
  • the methods of the above embodiments can be implemented by means of software plus a necessary general-purpose hardware platform, and of course also by hardware, but in many cases the former is a better implementation.
  • the technical solution of the present application can be embodied in the form of a software product in essence or the part that contributes to the prior art.
  • the computer software product is stored in a storage medium (such as read-only memory/random access memory, magnetic disk, optical disk), and includes several instructions to make a terminal device (which can be a mobile phone, computer, server, or network device, etc.) execute the method described in each embodiment of the application.

Abstract

本申请公开了一种基于自动驾驶的感知决策模型升级方法、系统及电子装置，所述方法包括：获取待测试道路场景信息；基于预设感知决策模型对待测试道路场景信息进行失效分析，获得车辆感知决策信息；根据车辆感知决策信息确定感知决策失效数据；根据感知决策失效数据对预设感知决策模型进行模型升级。

Description

基于自动驾驶的感知决策模型升级方法、系统及电子装置
本申请要求于2022年1月19号申请的、申请号为202210057249.0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
本申请涉及车辆技术领域，尤其涉及一种基于自动驾驶的感知决策模型升级方法、系统及电子装置。
 背景技术
随着人工智能技术的发展，汽车电动化、智能化技术也将衍生全新的变化。L3级及以上的自动驾驶功能对车辆智能化程度及功能备份有着更高的要求，目前各大主机厂及新造车势力均投入了大量的资源进行算法及功能的验证。尽管如此，对不同的新量产车型而言，其可量产的自动驾驶模型训练仍需频频投入不菲的研发、测试资源。
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。
 技术问题
本申请的主要目的在于提供了一种基于自动驾驶的感知决策模型升级方法、系统及电子装置，旨在解决如何提升量产车型功能和算法的可靠性的技术问题。
 技术解决方案
为实现上述目的，本申请提供了一种基于自动驾驶的感知决策模型升级方法，所述基于自动驾驶的感知决策模型升级方法包括：
获取待测试道路场景信息；
基于预设感知决策模型对所述待测试道路场景信息进行失效分析，获得车辆感知决策信息；
根据所述车辆感知决策信息确定感知决策失效数据；
根据所述感知决策失效数据对所述预设感知决策模型进行模型升级。
在一实施方式中，所述获取待测试道路场景信息的步骤之前，还包括：
获取多个驾驶操控数据样本及多个场景化环境数据样本；
基于多个所述驾驶操控数据样本对初始决策规控模型进行训练，获得预设决策规控模型；
基于多个所述场景化环境数据样本对初始感知模型进行训练，获得预设感知模型；
根据所述预设决策规控模型及所述预设感知模型构建预设感知决策模型。
在一实施方式中，所述获取多个驾驶操控数据样本及多个场景化环境数据样本的步骤，包括：
获取多个道路场景样本信息；
将多个所述道路场景样本信息输入至预设场景流识别模型中，获得多个场景流识别信息；
判断多个所述场景流识别信息是否满足预设识别条件；
在多个所述场景识别流信息满足所述预设识别条件时，分别对多个所述场景流识别信息进行数据清洗，获得多个驾驶操控数据样本及多个场景化环境数据样本。
在一实施方式中，所述判断多个所述场景流识别信息是否满足预设识别条件的步骤之后，还包括：
在多个所述场景识别流信息不满足所述预设识别条件时，对多个所述道路场景样本信息进行数据清洗，获得多个待处理场景样本数据；
分别对多个所述待处理场景样本数据进行场景流标注，获得多个待训练场景样本数据；
根据多个所述待训练场景样本数据对预设场景流识别模型进行模型升级。
在一实施方式中，所述根据所述车辆感知决策信息确定感知决策失效数据的步骤之前，包括：
判断所述车辆感知决策信息是否满足预设失效条件；
在所述车辆感知决策信息满足所述预设失效条件时，执行所述根据所述车辆感知决策信息确定感知决策失效数据的步骤。
在一实施方式中，所述判断所述车辆感知决策信息是否满足预设失效条件的步骤之后，还包括：
在所述车辆感知决策信息不满足所述预设失效条件时，根据所述车辆感知决策信息生成模型功能评估报告；
根据所述模型功能评估报告对所述预设感知决策模型进行量产部署，以获得多个车辆感知决策信息；
对多个所述车辆感知决策信息进行失效分析，获得多个感知决策失效数据；
基于多个所述感知决策失效数据对所述预设感知决策模型进行升级。
在一实施方式中，所述基于多个所述感知决策失效数据对所述预设感知决策模型进行升级的步骤，包括：
分别对多个所述感知决策失效数据进行标注，获得多个感知决策标注数据；
根据多个所述感知决策标注数据对所述预设感知决策模型进行升级。
此外，为实现上述目的，本申请还提出一种基于自动驾驶的感知决策模型升级系统，所述基于自动驾驶的感知决策模型升级系统包括：
获取模块，被配置为获取待测试道路场景信息；
分析模块，被配置为基于预设感知决策模型对所述待测试道路场景信息进行失效分析，获得车辆感知决策信息；
确定模块，被配置为根据所述车辆感知决策信息确定感知决策失效数据；
升级模块，被配置为根据所述感知决策失效数据对所述预设感知决策模型进行模型升级。
在一实施方式中，所述基于自动驾驶的感知决策模型升级系统还包括构建模块；
所述构建模块，被配置为获取多个驾驶操控数据样本及多个场景化环境数据样本；
所述构建模块，还被配置为基于多个所述驾驶操控数据样本对初始决策规控模型进行训练，获得预设决策规控模型；
所述构建模块，还被配置为基于多个所述场景化环境数据样本对初始感知模型进行训练，获得预设感知模型；
所述构建模块，还被配置为根据所述预设决策规控模型及所述预设感知模型构建预设感知决策模型。
在一实施方式中，所述获取模块，还被配置为获取多个道路场景样本信息；
所述获取模块，还被配置为将多个所述道路场景样本信息输入至预设场景流识别模型中，获得多个场景流识别信息；
所述获取模块，还被配置为判断多个所述场景流识别信息是否满足预设识别条件；
所述获取模块，还被配置为在多个所述场景识别流信息满足所述预设识别条件时，分别对多个所述场景流识别信息进行数据清洗，获得多个驾驶操控数据样本及多个场景化环境数据样本。
 有益效果
本申请首先获取待测试道路场景信息，然后基于预设感知决策模型对待测试道路场景信息进行失效分析，获得车辆感知决策信息，之后根据车辆感知决策信息确定感知决策失效数据，并根据感知决策失效数据对预设感知决策模型进行模型升级。相较于现有技术中需要针对性的对不同量产的车型进行模型训练，而本申请中通过自动驾驶系统数据的闭环设计，可以根据车辆感知决策信息对应的感知决策失效数据对预设感知决策模型进行模型升级，从而提升了量产车型功能和算法的可靠性。
 附图说明
图1为本申请基于自动驾驶的感知决策模型升级方法第一实施例的流程示意图；
图2为本申请基于自动驾驶的感知决策模型升级方法第一实施例的场景化索引与数据清洗系统原理图；
图3为本申请基于自动驾驶的感知决策模型升级方法第二实施例的流程示意图；
图4为本申请基于自动驾驶的感知决策模型升级方法第二实施例的算法数据驱动与闭环自动化系统原理图；
图5为本申请基于自动驾驶的感知决策模型升级系统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。
 本发明的实施方式
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。
本申请实施例提供了一种基于自动驾驶的感知决策模型升级方法，参照图1，图1为本申请基于自动驾驶的感知决策模型升级方法第一实施例的流程示意图。
本实施例中，所述基于自动驾驶的感知决策模型升级方法包括以下步骤：
步骤S10：获取待测试道路场景信息。
易于理解的是，本实施例的执行主体可以是具有图像处理、数据处理、网络通讯和程序运行等功能的基于自动驾驶的感知决策模型升级设备，其中，基于自动驾驶的感知决策模型升级系统中包括场景化索引与数据清洗系统和算法数据驱动与闭环自动化系统等，也可以为其他具有相似功能的计算机设备等，本实施例并不加以限制。
待测试道路场景信息可以为车辆在行驶时采集的道路环境信息等。获取待测试道路场景信息的步骤之前，还需要获取多个驾驶操控数据样本及多个场景化环境数据样本，基于多个驾驶操控数据样本对初始决策规控模型进行训练，获得预设决策规控模型，基于多个场景化环境数据样本对初始感知模型进行训练，获得预设感知模型，之后根据预设决策规控模型及预设感知模型构建预设感知决策模型。其中驾驶操控数据样本可以理解为实现无人驾驶时所产生的车辆行驶操控指令，场景化环境数据样本可以为理解为车辆行驶操控指令对应的道路环境信息及天气信息等。
获取多个驾驶操控数据样本及多个场景化环境数据样本的处理方式可以为获取多个道路场景样本信息，将多个道路场景样本信息输入至预设场景流识别模型中，获得多个场景流识别信息，判断多个场景流识别信息是否满足预设识别条件，在多个场景识别流信息满足预设识别条件时，分别对多个场景流识别信息进行数据清洗，获得多个驾驶操控数据样本及多个场景化环境数据样本。预设识别条件可以理解为不存在未识别场景流等。
在具体实现中，在多个场景识别流信息不满足预设识别条件时，对多个道路场景样本信息进行数据清洗，获得多个待处理场景样本数据，分别对多个待处理场景样本数据进行场景流标注，获得多个待训练场景样本数据，根据多个待训练场景样本数据对预设场景流识别模型进行模型升级。
需要说明的是，道路场景样本信息可以理解为道路环境信息、天气环境信息及车辆驾驶操控数据等。场景流识别信息包括常规场景信息、极端场景信息及未识别场景信息等。例如隧道、高速匝道、雨雪雾天等，以及其他边界化难题（Corner-case）数据等。
在具体实现中，可以将多个道路场景样本信息输入至预设场景识别流模型中，并将多个道路场景样本信息进行场景识别和场景索引，分别划分为常规场景索引信息、极端场景索引信息及未识别场景流，若多个道路场景样本信息中存在未识别场景流，判定不满足预设识别条件，并需要对未识别场景流进行环境数据清洗（例如将拍摄图片中的人脸或车牌号进行模糊处理），之后将处理后的场景流进行存储，并对场景流进行标注，最后根据标注后的场景流对预设场景流识别模型进行训练，将训练后的预设场景流识别模型完成空中下载技术（Over the Air Technology，OTA）升级等。
在本实施例中在多个道路场景样本信息中不存在未识别场景流，判定满足预设识别条件，并需要对多个道路场景样本信息进行常规场景索引及极端场景索引，之后对索引后的道路场景样本信息的数据结构化，并将结构化的道路场景样本信息进行环境数据清洗，分别获得多个驾驶操控数据样本及多个场景化环境数据样本，最后还需要将多个驾驶操控数据样本及多个场景化环境数据样本进行分类存储等。
参考图2，图2为本申请基于自动驾驶的感知决策模型升级方法第一实施例的场景化索引与数据清洗系统原理图，用户可以通过车辆上安装的传感器采集多个道路场景样本信息，其中道路场景样本信息包括车辆操控数据、道路环境及天气信息等，之后可以将采集的多个道路环境样本信息输入至预设场景流识别模型，以使预设场景流识别模型对多个道路场景样本信息进行场景识别，并根据场景识别结果对多个道路场景样本信息进行分类，并构建常规场景索引、极端场景索引及未识别场景流数据库。在具体实现中，将常规场景索引对应的数据及极端场景索引对应的数据进行数据结构化，并根据结构化后的数据进行分类，获得驾驶操控数据及待处理场景化环境数据，之后需要将待处理场景化环境数据进行环境数据清洗，以获得场景化环境数据，最后将驾驶操控数据和场景化环境数据分别进行存储，并将驾驶操控数据作为驾驶操控数据样本，将场景化环境数据作为场景化环境数据样本等。
应理解的是，还可以将未识别场景流进行环境数据清洗，将清洗后的数据流进行存储，并将清洗后的数据流进行场景标注，将标注后的数据流对预设场景流识别模型进行训练，将训练后的预设场景流识别模型完成OTA升级等。
还需要说明的是，现有的数据采集系统，几乎都是对视频流或点云帧数据流按照时间片段（比如按一分钟时长进行切分）强行切分，虽然存储快捷简单，但取用极为不便。主要体现在两个方面：1、无法对数据进行标签化，不利于快速获取期望的场景数据；2、模型训练所需要的特定场景数据流，极有可能会被分段存储于两个切片文件中，因此一般需要特定的可视化软件进行手工甄选、拼接，极大增加了数据筛选的时间和成本。而本实施例中的预设场景流识别模型能够从模型训练的视角，提供端到端的场景化数据，以场景标签作为索引高效存取，极大提升模型训练的数据筛选效率，且按需取场景数据更容易完成特定算法模型的训练和推理算法的评估。在数据采集前端引入场景化索引技术，大大提升后端算法模型训练时的数据筛选与场景选择效率等。
步骤S20：基于预设感知决策模型对所述待测试道路场景信息进行失效分析，获得车辆感知决策信息。
需要说明的是，预设感知决策模型为预设决策规控模型及预设感知模型的组合模型。
在具体实现中需要基于多个驾驶操控数据样本对初始决策规控模型进行训练，获得预设决策规控模型，基于多个场景化环境数据样本对初始感知模型进行训练，获得预设感知模型，之后根据预设决策规控模型及预设感知模型构建预设感知决策模型等。
在本实施例中可以将待测试道路场景信息输入至预设感知决策模型中，获得车辆感知决策信息，车辆感知决策信息可以理解为预设感知决策模型根据待测试道路场景信息输出的环境感知信息及车辆自动驾驶信息等。
步骤S30：根据所述车辆感知决策信息确定感知决策失效数据。
应理解的是，需要判断车辆感知决策信息是否满足预设失效条件，在车辆感知决策信息满足预设失效条件时，根据车辆感知决策信息确定感知决策失效数据。预设失效条件为车辆感知决策信息与待测试道路场景信息对应的驾驶员操作数据集场景化环境数据信息不一致等。
感知决策失效数据为用户对车辆感知决策信息进行失效分析所得到的数据等。例如车辆感知决策信息为A，待测试道路场景信息对应的驾驶员操作数据集场景化环境数据为B，则车辆感知决策信息A与待测试道路场景信息对应的驾驶员操作数据集场景化环境数据B对应不一致，可以将待测试道路场景信息及车辆感知决策信息作为感知决策失效数据等。
步骤S40：根据所述感知决策失效数据对所述预设感知决策模型进行模型升级。
在具体实现中，若车辆感知决策信息不正确，则对感知决策失效数据进行标注，使标注后的失效数据对预设感知决策模型进行训练等。
还需要说明的是，在车辆感知决策信息不满足预设失效条件时，确定车辆感知决策信息是正确，需要根据车辆感知决策信息获取预设感知决策模型的指标及功能评估报告，并基于预设感知决策模型的指标及功能评估报告对预设感知决策模型进行量产部署，以获得多个车辆感知决策信息，之后对多个车辆感知决策信息进行失效分析，获得多个感知决策失效数据，基于多个感知决策失效数据对预设感知决策模型进行升级。
基于多个感知决策失效数据对预设感知决策模型进行升级的处理方式可以为分别对多个感知决策失效数据进行标注，获得多个感知决策标注数据，根据多个感知决策标注数据对预设感知决策模型进行升级等。
在本实施例中，首先获取待测试道路场景信息，然后基于预设感知决策模型对待测试道路场景信息进行失效分析，获得车辆感知决策信息，之后根据车辆感知决策信息确定感知决策失效数据，并根据感知决策失效数据对预设感知决策模型进行模型升级。相较于现有技术中需要针对性的对不同量产的车型进行模型训练，而本实施例中通过自动驾驶系统数据的闭环设计，可以根据车辆感知决策信息对应的感知决策失效数据对预设感知决策模型进行模型升级，从而提升了量产车型功能和算法的可靠性，进而提高了用户体验感。
参考图3，图3为本申请基于自动驾驶的感知决策模型升级方法第二实施例的流程示意图。
基于上述第一实施例，在本实施例中，所述步骤S30，还包括：
步骤S301：判断所述车辆感知决策信息是否满足预设失效条件。
预设失效条件为车辆感知决策信息与待测试道路场景信息对应的驾驶员操作数据集场景化环境数据信息不一致等。
在车辆感知决策信息不满足预设失效条件时，确定车辆感知决策信息是正确，需要根据车辆感知决策信息获取预设感知决策模型的指标及功能评估报告，并基于预设感知决策模型的指标及功能评估报告对预设感知决策模型进行量产部署，以获得多个车辆感知决策信息，之后对多个车辆感知决策信息进行失效分析，获得多个感知决策失效数据，基于多个感知决策失效数据对预设感知决策模型进行升级。
基于多个感知决策失效数据对预设感知决策模型进行升级的处理方式可以为分别对多个感知决策失效数据进行标注，获得多个感知决策标注数据，根据多个感知决策标注数据对预设感知决策模型进行升级等。
步骤S302：在所述车辆感知决策信息满足所述预设失效条件时，根据所述车辆感知决策信息确定感知决策失效数据。
感知决策失效数据为用户对车辆感知决策信息进行失效分析所得到的数据等。例如车辆感知决策信息为A，待测试道路场景信息对应的驾驶员操作数据集场景化环境数据为B，则车辆感知决策信息A与待测试道路场景信息对应的驾驶员操作数据集场景化环境数据B对应不一致，可以将待测试道路场景信息及车辆感知决策信息作为感知决策失效数据等。
在具体实现中，若车辆感知决策信息不正确，则判定车辆感知决策信息满足预设失效条件，并根据车辆感知决策信息确定感知决策失效数据，对感知决策失效数据进行标注，使标注后的失效数据对预设感知决策模型进行训练等。
参考图4，图4为本申请基于自动驾驶的感知决策模型升级方法第二实施例的算法数据驱动与闭环自动化系统原理图，用户可以将待测试道路场景信息输入至预设感知决策模型中以获得对应的车辆感知决策信息，之后对车辆感知决策信息进行失效分析，在车辆感知决策信息不满足预设失效条件时，确定车辆感知决策信息是正确，需要根据车辆感知决策信息获取预设感知决策模型的指标及功能评估报告，并基于预设感知决策模型的指标及功能评估报告对预设感知决策模型进行量产部署，以获得多个车辆感知决策信息，之后对多个车辆感知决策信息进行失效分析，获得多个感知决策失效数据，分别对多个感知决策失效数据进行失效数据标注，并根据标注后的失效数据对预设感知决策模型进行定制化训练，以实现OTA升级。若车辆感知决策信息不正确，则判定车辆感知决策信息满足预设失效条件，并根据车辆感知决策信息确定感知决策失效数据，对感知决策失效数据进行标注，使标注后的失效数据对预设感知决策模型进行训练等。
还需要说明的是，算法数据驱动与闭环自动化原理图中的闭环设计系统支持量产车型的数据采集与算法模型升级。能够收集量产车型在客户使用过程中，部署的算法模型或功能在推理失效情况下的数据（失效时刻前后一分钟内），并对记录的“模式失配”数据进行失效模式分析，并进行针对性的定制化训练，通过OTA方式对软件算法进行升级。能够大大提升量产车型功能和算法的可靠性和用户体验。
在本实施例中，判断车辆感知决策信息是否满足预设失效条件，在车辆感知决策信息满足预设失效条件时，根据车辆感知决策信息确定感知决策失效数据，相较于现有技术中，直接利用得到的车辆感知决策信息，并不会对车辆感知决策信息进一步验证，而本实施例中需要对车辆感知决策信息进行失效分析，从而提升量产车型功能和算法的可靠性，进而保证了预设感知决策模型的准确性。
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基于自动驾驶的感知决策模型升级程序，所述基于自动驾驶的感知决策模型升级程序被处理器执行时实现如上文所述的基于自动驾驶的感知决策模型升级方法的步骤。
此外，本申请实施例还提出一种电子装置，所述电子装置可以是例如移动终端、固定终端，所述电子装置包括处理器与存储器，所述存储器存储有基于自动驾驶的感知决策模型升级程序，所述处理器用于运行所述基于自动驾驶的感知决策模型升级程序以实现如上文所述的基于自动驾驶的感知决策模型升级方法的步骤。
参照图5，图5为本申请基于自动驾驶的感知决策模型升级系统第一实施例的结构框图。
如图5所示，本申请实施例提出的基于自动驾驶的感知决策模型升级系统包括：
获取模块5001，被配置为获取待测试道路场景信息。
待测试道路场景信息可以为车辆在行驶时采集的道路环境信息等。获取待测试道路场景信息的步骤之前，还需要获取多个驾驶操控数据样本及多个场景化环境数据样本，基于多个驾驶操控数据样本对初始决策规控模型进行训练，获得预设决策规控模型，基于多个场景化环境数据样本对初始感知模型进行训练，获得预设感知模型，之后根据预设决策规控模型及预设感知模型构建预设感知决策模型。其中驾驶操控数据样本可以理解为实现无人驾驶时所产生的车辆行驶操控指令，场景化环境数据样本可以为理解为车辆行驶操控指令对应的道路环境信息及天气信息等。
获取多个驾驶操控数据样本及多个场景化环境数据样本的处理方式可以为获取多个道路场景样本信息，将多个道路场景样本信息输入至预设场景流识别模型中，获得多个场景流识别信息，判断多个场景流识别信息是否满足预设识别条件，在多个场景识别流信息满足预设识别条件时，分别对多个场景流识别信息进行数据清洗，获得多个驾驶操控数据样本及多个场景化环境数据样本。预设识别条件可以理解为不存在未识别场景流等。
在具体实现中，在多个场景识别流信息不满足预设识别条件时，对多个道路场景样本信息进行数据清洗，获得多个待处理场景样本数据，分别对多个待处理场景样本数据进行场景流标注，获得多个待训练场景样本数据，根据多个待训练场景样本数据对预设场景流识别模型进行模型升级。
需要说明的是，道路场景样本信息可以理解为道路环境信息、天气环境信息及车辆驾驶操控数据等。场景流识别信息包括常规场景信息、极端场景信息及未识别场景信息等。例如隧道、高速匝道、雨雪雾天等，以及其他边界化难题（Corner-case）数据等。
在具体实现中，可以将多个道路场景样本信息输入至预设场景识别流模型中，并将多个道路场景样本信息进行场景识别和场景索引，分别划分为常规场景索引信息、极端场景索引信息及未识别场景流，若多个道路场景样本信息中存在未识别场景流，判定不满足预设识别条件，并需要对未识别场景流进行环境数据清洗（例如将拍摄图片中的人脸或车牌号进行模糊处理），之后将处理后的场景流进行存储，并对场景流进行标注，最后根据标注后的场景流对预设场景流识别模型进行训练，将训练后的预设场景流识别模型完成OTA升级等。
在本实施例中在多个道路场景样本信息中不存在未识别场景流，判定满足预设识别条件，并需要对多个道路场景样本信息进行常规场景索引及极端场景索引，之后对索引后的道路场景样本信息的数据结构化，并将结构化的道路场景样本信息进行环境数据清洗，分别获得多个驾驶操控数据样本及多个场景化环境数据样本，最后还需要将多个驾驶操控数据样本及多个场景化环境数据样本进行分类存储等。
参考图2，图2为本申请基于自动驾驶的感知决策模型升级方法第一实施例的场景化索引与数据清洗系统原理图，用户可以通过车辆上安装的传感器采集多个道路场景样本信息，其中道路场景样本信息包括车辆操控数据、道路环境及天气信息等，之后可以将采集的多个道路环境样本信息输入至预设场景流识别模型，以使预设场景流识别模型对多个道路场景样本信息进行场景识别，并根据场景识别结果对多个道路场景样本信息进行分类并构建常规场景索引、极端场景索引及未识别场景流数据库，在具体实现中，将常规场景索引对应的数据及极端场景索引对应的数据进行数据结构化，并根据结构化后的数据进行分类，获得驾驶操控数据及待处理场景化环境数据，之后需要将待处理场景化环境数据进行环境数据清洗，以获得场景化环境数据，最后将驾驶操控数据和场景化环境数据分别进行存储，并将驾驶操控数据作为驾驶操控数据样本，将场景化环境数据作为场景化环境数据样本等。
应理解的是，还可以将未识别场景流进行环境数据清洗，并将清洗后的数据流进行存储，并将清洗后的数据流进行场景标注，将标注后的数据流对预设场景流识别模型进行训练，将训练后的预设场景流识别模型完成OTA升级等。
还需要说明的是，现有的数据采集系统，几乎都是对视频流或点云帧数据流按照时间片段（比如按一分钟时长进行切分）强行切分，虽然存储快捷简单，但取用极为不便。主要体现在两个方面：1、无法对数据进行标签化，不利于快速获取期望的场景数据；2、模型训练所需要的特定场景数据流，极有可能会被分段存储于两个切片文件中，因此一般需要特定的可视化软件进行手工甄选、拼接，极大增加了数据筛选的时间和成本。而本实施例中的预设场景流识别模型能够从模型训练的视角，提供端到端的场景化数据，以场景标签作为索引高效存取，极大提升模型训练的数据筛选效率，且按需取场景数据更容易完成特定算法模型的训练和推理算法的评估。在数据采集前端引入场景化索引技术，大大提升后端算法模型训练时的数据筛选与场景选择效率等。
分析模块5002，被配置为基于预设感知决策模型对所述待测试道路场景信息进行失效分析，获得车辆感知决策信息。
需要说明的是，预设感知决策模型为预设决策规控模型及预设感知模型的组合模型。
在具体实现中需要基于多个驾驶操控数据样本对初始决策规控模型进行训练，获得预设决策规控模型，基于多个场景化环境数据样本对初始感知模型进行训练，获得预设感知模型，之后根据预设决策规控模型及预设感知模型构建预设感知决策模型等。
在本实施例中可以将待测试道路场景信息输入至预设感知决策模型中，获得车辆感知决策信息，车辆感知决策信息可以理解为预设感知决策模型根据待测试道路场景信息输出的环境感知信息及车辆自动驾驶信息等。
确定模块5003，被配置为根据所述车辆感知决策信息确定感知决策失效数据。
应理解的是，需要判断车辆感知决策信息是否满足预设失效条件，在车辆感知决策信息满足预设失效条件时，根据车辆感知决策信息确定感知决策失效数据。预设失效条件为车辆感知决策信息与待测试道路场景信息对应的驾驶员操作数据集场景化环境数据信息不一致等。
感知决策失效数据为用户对车辆感知决策信息进行失效分析所得到的数据等。例如车辆感知决策信息为A，待测试道路场景信息对应的驾驶员操作数据集场景化环境数据为B，则车辆感知决策信息A与待测试道路场景信息对应的驾驶员操作数据集场景化环境数据B对应不一致，可以将待测试道路场景信息及车辆感知决策信息作为感知决策失效数据等。
升级模块5004，被配置为根据所述感知决策失效数据对所述预设感知决策模型进行模型升级。
在具体实现中，若车辆感知决策信息不正确，则对感知决策失效数据进行标注，使标注后的失效数据对预设感知决策模型进行训练等。
还需要说明的是，在车辆感知决策信息不满足预设失效条件时，确定车辆感知决策信息是正确，需要根据车辆感知决策信息获取预设感知决策模型的指标及功能评估报告，并基于预设感知决策模型的指标及功能评估报告对预设感知决策模型进行量产部署，以获得多个车辆感知决策信息，之后对多个车辆感知决策信息进行失效分析，获得多个感知决策失效数据，基于多个感知决策失效数据对预设感知决策模型进行升级。
基于多个感知决策失效数据对预设感知决策模型进行升级的处理方式可以为分别对多个感知决策失效数据进行标注，获得多个感知决策标注数据，根据多个感知决策标注数据对预设感知决策模型进行升级等。
在本实施例中，首先获取待测试道路场景信息，然后基于预设感知决策模型对待测试道路场景信息进行失效分析，获得车辆感知决策信息，之后根据车辆感知决策信息确定感知决策失效数据，并根据感知决策失效数据对预设感知决策模型进行模型升级。相较于现有技术中需要针对性的对不同量产的车型进行模型训练，而本实施例中通过自动驾驶系统数据的闭环设计，可以根据车辆感知决策信息对应的感知决策失效数据对预设感知决策模型进行模型升级，从而提升了量产车型功能和算法的可靠性，进而提高了用户体验感。
本申请基于自动驾驶的感知决策模型升级系统的其他实施例或具体实现方式可参照上述各方法实施例，此处不再赘述。
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/随机存取存储器、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 (11)

  1. 一种基于自动驾驶的感知决策模型升级方法，其中，所述基于自动驾驶的感知决策模型升级方法包括以下步骤：
    获取待测试道路场景信息；
    基于预设感知决策模型对所述待测试道路场景信息进行失效分析，获得车辆感知决策信息；
    根据所述车辆感知决策信息确定感知决策失效数据；
    根据所述感知决策失效数据对所述预设感知决策模型进行模型升级。
  2. 如权利要求1所述的方法，其中，所述获取待测试道路场景信息的步骤之前，还包括：
    获取多个驾驶操控数据样本及多个场景化环境数据样本；
    基于多个所述驾驶操控数据样本对初始决策规控模型进行训练，获得预设决策规控模型；
    基于多个所述场景化环境数据样本对初始感知模型进行训练，获得预设感知模型；
    根据所述预设决策规控模型及所述预设感知模型构建预设感知决策模型。
  3. 如权利要求1所述的方法，其中，所述获取多个驾驶操控数据样本及多个场景化环境数据样本的步骤，包括：
    获取多个道路场景样本信息；
    将多个所述道路场景样本信息输入至预设场景流识别模型中，获得多个场景流识别信息；
    判断多个所述场景流识别信息是否满足预设识别条件；
    在多个所述场景识别流信息满足所述预设识别条件时，分别对多个所述场景流识别信息进行数据清洗，获得多个驾驶操控数据样本及多个场景化环境数据样本。
  4. 如权利要求3所述的方法，其中，所述判断多个所述场景流识别信息是否满足预设识别条件的步骤之后，还包括：
    在多个所述场景识别流信息不满足所述预设识别条件时，对多个所述道路场景样本信息进行数据清洗，获得多个待处理场景样本数据；
    分别对多个所述待处理场景样本数据进行场景流标注，获得多个待训练场景样本数据；
    根据多个所述待训练场景样本数据对预设场景流识别模型进行模型升级。
  5. 如权利要求1-4任一项所述的方法，其中，所述根据所述车辆感知决策信息确定感知决策失效数据的步骤之前，包括：
    判断所述车辆感知决策信息是否满足预设失效条件；
    在所述车辆感知决策信息满足所述预设失效条件时，执行所述根据所述车辆感知决策信息确定感知决策失效数据的步骤。
  6. 如权利要求5所述的方法，其中，所述判断所述车辆感知决策信息是否满足预设失效条件的步骤之后，还包括：
    在所述车辆感知决策信息不满足所述预设失效条件时，根据所述车辆感知决策信息生成模型功能评估报告；
    根据所述模型功能评估报告对所述预设感知决策模型进行量产部署，以获得多个车辆感知决策信息；
    对多个所述车辆感知决策信息进行失效分析，获得多个感知决策失效数据；
    基于多个所述感知决策失效数据对所述预设感知决策模型进行升级。
  7. 如权利要求6所述的方法，其中，所述基于多个所述感知决策失效数据对所述预设感知决策模型进行升级的步骤，包括：
    分别对多个所述感知决策失效数据进行标注，获得多个感知决策标注数据；
    根据多个所述感知决策标注数据对所述预设感知决策模型进行升级。
  8. 一种基于自动驾驶的感知决策模型升级系统，其中，所述基于自动驾驶的感知决策模型升级系统包括：
    获取模块，被配置为获取待测试道路场景信息；
    分析模块，被配置为基于预设感知决策模型对所述待测试道路场景信息进行失效分析，获得车辆感知决策信息；
    确定模块，被配置为根据所述车辆感知决策信息确定感知决策失效数据；
    升级模块，被配置为根据所述感知决策失效数据对所述预设感知决策模型进行模型升级。
  9. 如权利要求8所述的系统，其中，所述基于自动驾驶的感知决策模型升级系统还包括构建模块；
    所述构建模块，被配置为获取多个驾驶操控数据样本及多个场景化环境数据样本；
    所述构建模块，还被配置为基于多个所述驾驶操控数据样本对初始决策规控模型进行训练，获得预设决策规控模型；
    所述构建模块，还被配置为基于多个所述场景化环境数据样本对初始感知模型进行训练，获得预设感知模型；
    所述构建模块，还被配置为根据所述预设决策规控模型及所述预设感知模型构建预设感知决策模型。
  10. 如权利要求8所述的系统，其中，所述获取模块，还被配置为获取多个道路场景样本信息；
    所述获取模块，还被配置为将多个所述道路场景样本信息输入至预设场景流识别模型中，获得多个场景流识别信息；
    所述获取模块，还被配置为判断多个所述场景流识别信息是否满足预设识别条件；
    所述获取模块，还被配置为在多个所述场景识别流信息满足所述预设识别条件时，分别对多个所述场景流识别信息进行数据清洗，获得多个驾驶操控数据样本及多个场景化环境数据样本。
  11. 一种电子装置，所述电子装置包括处理器与存储器，所述存储器存储有基于自动驾驶的感知决策模型升级程序，所述处理器被配置为运行所述基于自动驾驶的感知决策模型升级程序以实现如权利要求1-7任一项所述的基于自动驾驶的感知决策模型升级方法的步骤。
PCT/CN2022/083064 2022-01-19 2022-03-25 基于自动驾驶的感知决策模型升级方法、系统及电子装置 WO2023137864A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210057249.0 2022-01-19
CN202210057249.0A CN114078242A (zh) 2022-01-19 2022-01-19 基于自动驾驶的感知决策模型升级方法及系统

Publications (1)

Publication Number Publication Date
WO2023137864A1 true WO2023137864A1 (zh) 2023-07-27

Family

ID=80284519

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/083064 WO2023137864A1 (zh) 2022-01-19 2022-03-25 基于自动驾驶的感知决策模型升级方法、系统及电子装置

Country Status (2)

Country Link
CN (1) CN114078242A (zh)
WO (1) WO2023137864A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114078242A (zh) * 2022-01-19 2022-02-22 浙江吉利控股集团有限公司 基于自动驾驶的感知决策模型升级方法及系统
CN115311761B (zh) * 2022-07-15 2023-11-03 襄阳达安汽车检测中心有限公司 非实时车载感知系统测评方法及相关设备

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190228118A1 (en) * 2018-01-24 2019-07-25 Toyota Research Institute, Inc. Systems and methods for identifying human-based perception techniques
CN111325230A (zh) * 2018-12-17 2020-06-23 上海汽车集团股份有限公司 车辆换道决策模型的在线学习方法和在线学习装置
CN112131964A (zh) * 2020-08-31 2020-12-25 南京汽车集团有限公司 一种道路作业车辆视觉感知系统及其使用方法
CN112180921A (zh) * 2020-09-22 2021-01-05 安徽江淮汽车集团股份有限公司 自动驾驶算法训练系统及方法
CN113835421A (zh) * 2020-06-06 2021-12-24 华为技术有限公司 训练驾驶行为决策模型的方法及装置
CN114078242A (zh) * 2022-01-19 2022-02-22 浙江吉利控股集团有限公司 基于自动驾驶的感知决策模型升级方法及系统

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190228118A1 (en) * 2018-01-24 2019-07-25 Toyota Research Institute, Inc. Systems and methods for identifying human-based perception techniques
CN111325230A (zh) * 2018-12-17 2020-06-23 上海汽车集团股份有限公司 车辆换道决策模型的在线学习方法和在线学习装置
CN113835421A (zh) * 2020-06-06 2021-12-24 华为技术有限公司 训练驾驶行为决策模型的方法及装置
CN112131964A (zh) * 2020-08-31 2020-12-25 南京汽车集团有限公司 一种道路作业车辆视觉感知系统及其使用方法
CN112180921A (zh) * 2020-09-22 2021-01-05 安徽江淮汽车集团股份有限公司 自动驾驶算法训练系统及方法
CN114078242A (zh) * 2022-01-19 2022-02-22 浙江吉利控股集团有限公司 基于自动驾驶的感知决策模型升级方法及系统

Also Published As

Publication number Publication date
CN114078242A (zh) 2022-02-22

Similar Documents

Publication Publication Date Title
WO2023137864A1 (zh) 基于自动驾驶的感知决策模型升级方法、系统及电子装置
US20210383209A1 (en) Collecting and processing data from vehicles
CN112396093B (zh) 驾驶场景分类方法、装置、设备及可读存储介质
CN113884961B (zh) Soc校准方法、建模方法、建模装置、计算机设备及介质
CN110175519B (zh) 一种变电站的分合标识仪表识别方法、装置与存储介质
CN114861836B (zh) 一种基于人工智能平台的模型部署方法及相关设备
CN111680753A (zh) 一种数据标注方法、装置、电子设备及存储介质
CN113269042B (zh) 基于行驶车辆违章识别的智能交通管理方法及系统
CN112434178A (zh) 图像分类方法、装置、电子设备和存储介质
CN115830399A (zh) 分类模型训练方法、装置、设备、存储介质和程序产品
CN116258178A (zh) 模型转换方法、装置、电子设备和可读存储介质
CN114818707A (zh) 一种基于知识图谱的自动驾驶决策方法和系统
CN114724140A (zh) 一种基于yolo v3的草莓成熟度检测方法、装置
CN110728229B (zh) 图像处理方法、装置、设备和存储介质
CN114550129B (zh) 一种基于数据集的机器学习模型处理方法和系统
CN116894317A (zh) 数据处理方法、装置、电子设备和介质
CN115543809A (zh) 自动驾驶功能的测试场景库构建方法及装置
US11580666B2 (en) Localization and mapping method and moving apparatus
CN111143423B (zh) 动态场景标注数据挖掘方法、装置以及终端
CN114299295A (zh) 一种数据处理方法及相关装置
CN112729319A (zh) 一种自动化数据采集分析系统及方法
WO2024051248A1 (zh) 一种自动驾驶车辆的数据标记方法及标记装置
CN116840693B (zh) 基于人工智能的充放电测试控制方法及系统
CN114563014B (zh) 一种基于仿真图像的OpenDrive地图自动化检测方法
CN116142201A (zh) 车辆的控制方法、计算机设备和存储介质

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: 22921298

Country of ref document: EP

Kind code of ref document: A1