WO2023231212A1

WO2023231212A1 - Prediction model training method and apparatus, and map prediction method and apparatus

Info

Publication number: WO2023231212A1
Application number: PCT/CN2022/117340
Authority: WO
Inventors: 王己龙
Original assignee: 合众新能源汽车股份有限公司
Priority date: 2022-06-02
Filing date: 2022-09-06
Publication date: 2023-12-07
Also published as: CN115115790A

Abstract

A prediction model training method and apparatus, and a map prediction method and apparatus. The prediction model training method comprises: acquiring a plurality of pieces of sample data collected by a mobile device during a moving process, each piece of the sample data comprising sample positioning information, sample speed information and sample point cloud information; according to the sample speed information, performing motion compensation processing on the sample point cloud information, so as to generate a sample vehicle speed compensation map; inputting the sample vehicle speed compensation map into a preset model, and outputting a prediction map; according to the prediction map and a sample high-definition map, calculating a loss value, the sample high-definition map being generated according to the sample positioning information; and according to the loss value, training the preset model until the preset model meets a preset training condition, so as to obtain a prediction model. Therefore, a map can be accurately predicted.

Description

预测模型的训练方法、地图预测方法及装置Prediction model training method, map prediction method and device

本申请要求在2022年6月2日提交中国专利局、申请号为202210622429.9、发明名称为“预测模型的训练方法、地图预测方法及装置”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。This application claims priority to the Chinese patent application filed with the China Patent Office on June 2, 2022, with application number 202210622429.9 and the invention title "Prediction model training method, map prediction method and device", the entire content of which is incorporated by reference. in this application.

技术领域Technical field

本申请属于信息处理技术领域，尤其涉及一种预测模型的训练方法、地图预测方法及装置。The present application belongs to the field of information processing technology, and in particular relates to a training method for a prediction model, a map prediction method and a device.

背景技术Background technique

目前，高精地图依赖定位信息，但是在隧道等场景下，定位信号较弱，此时无法利用定位信息确定移动装置附近的高精地图。在获取点云信息时，车辆自身姿态、运动状态或路面都可能会发生变化，这种变化会导致基于点云信息生成的点云地图发生畸变。At present, high-precision maps rely on positioning information. However, in tunnels and other scenarios, the positioning signal is weak. At this time, the positioning information cannot be used to determine the high-precision map near the mobile device. When obtaining point cloud information, the vehicle's own posture, motion state, or road surface may change. Such changes will cause distortion in the point cloud map generated based on the point cloud information.

因此，在面对复杂的环境时，缺少一种准确预测地图的方法。Therefore, there is a lack of a method to accurately predict maps when facing complex environments.

发明内容Contents of the invention

本申请实施例提供一种预测模型的训练方法、地图预测方法及装置，能够解决目前地图预测的准确度不高的问题。Embodiments of the present application provide a prediction model training method, a map prediction method and a device, which can solve the current problem of low accuracy of map prediction.

第一方面，本申请实施例提供一种预测模型的训练方法，该方法包括：In a first aspect, embodiments of the present application provide a method for training a prediction model, which method includes:

获取移动装置在移动过程中采集到的多个样本数据，每个所述样本数据包括样本定位信息、样本速度信息和样本点云信息；Obtain multiple sample data collected by the mobile device during movement, each of the sample data includes sample positioning information, sample speed information and sample point cloud information;

根据样本速度信息对所述样本点云信息进行运动补偿处理，生成样本车速补偿地图；Perform motion compensation processing on the sample point cloud information according to the sample speed information to generate a sample vehicle speed compensation map;

将所述样本车速补偿地图输入至预设模型，输出预测地图；Input the sample vehicle speed compensation map into the preset model and output the prediction map;

根据所述预测地图和样本高精地图计算损失值，所述样本高精地图根据所述样本定位信息生成；Calculate the loss value based on the predicted map and the sample high-precision map, and the sample high-precision map is generated based on the sample positioning information;

根据所述损失值训练所述预设模型，直至所述预设模型满足预设训练条件，得到预测模型。The preset model is trained according to the loss value until the preset model meets the preset training conditions, and a prediction model is obtained.

第二方面，本申请实施例提供一种地图预测方法，该方法包括：In a second aspect, embodiments of the present application provide a map prediction method, which method includes:

获取移动装置采集的运动数据，所述运动数据至少包括：速度信息和点云信息；Obtain motion data collected by the mobile device, where the motion data at least includes: speed information and point cloud information;

根据所述速度信息对所述点云信息进行运动补偿处理，生成车速补偿地图；Perform motion compensation processing on the point cloud information according to the speed information to generate a vehicle speed compensation map;

将所述车速补偿地图输入至预测模型，输出目标地图。The vehicle speed compensation map is input to the prediction model and a target map is output.

第三方面，本申请实施例提供一种预测模型的训练装置，该装置包括：In a third aspect, embodiments of the present application provide a training device for a prediction model, which device includes:

第一获取模块，用于获取移动装置在移动过程中采集到的多个样本数据，每个所述样本数据包括样本定位信息、样本速度信息和样本点云信息；The first acquisition module is used to acquire multiple sample data collected by the mobile device during movement. Each of the sample data includes sample positioning information, sample speed information and sample point cloud information;

第一补偿模块，用于根据样本速度信息对所述样本点云信息进行运动补偿处理，生成样本车速补偿地图；The first compensation module is used to perform motion compensation processing on the sample point cloud information according to the sample speed information, and generate a sample vehicle speed compensation map;

第一输入模块，用于将所述样本车速补偿地图输入至预设模型，输出预测地图；The first input module is used to input the sample vehicle speed compensation map into the preset model and output the predicted map;

计算模块，用于根据所述预测地图和样本高精地图计算损失值，所述样本高精地图根据所述样本定位信息生成；A calculation module configured to calculate a loss value based on the prediction map and the sample high-precision map, and the sample high-precision map is generated based on the sample positioning information;

训练模块，用于根据所述损失值训练所述预设模型，直至所述预设模型满足预设训练条件，得到预测模型。A training module is used to train the preset model according to the loss value until the preset model meets the preset training conditions to obtain a prediction model.

第四方面，本申请实施例提供一种地图预测装置，该装置包括：In a fourth aspect, embodiments of the present application provide a map prediction device, which includes:

第二获取模块，用于获取移动装置采集的运动数据，所述运动数据至少包括：速度信息和点云信息；The second acquisition module is used to acquire motion data collected by the mobile device, where the motion data at least includes: speed information and point cloud information;

第二补偿模块，用于根据所述速度信息对所述点云信息进行运动补偿处理，生成车速补偿地图；a second compensation module, configured to perform motion compensation processing on the point cloud information according to the speed information and generate a vehicle speed compensation map;

第二输入模块，用于将所述车速补偿地图输入至预测模型，输出目标地图。The second input module is used to input the vehicle speed compensation map into the prediction model and output the target map.

第五方面，本申请实施例提供了一种电子设备，该设备包括：处理器以及存储有计算机程序指令的存储器；处理器执行计算机程序指令时，实现如第一方面或者第一方面的任一可能实现方式中的方法。In a fifth aspect, embodiments of the present application provide an electronic device, which includes: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, it implements the first aspect or any one of the first aspects Methods in possible implementations.

第六方面，本申请实施例提供了一种可读存储介质，该计算机可读存储介质上存储有计算机程序指令，计算机程序指令被处理器执行时实现如第一方面或者第一方面的任一可能实现方式中的方法。In a sixth aspect, embodiments of the present application provide a readable storage medium. Computer program instructions are stored on the computer-readable storage medium. When the computer program instructions are executed by a processor, the implementation of the first aspect or any one of the first aspects is implemented. Methods in possible implementations.

本申请实施例中，通过获取移动装置在移动过程中采集到的多个样本数据，其中，每个样本数据包括样本定位信息、样本速度信息和样本点云信息。根据样本速度信息对样本点云信息进行运动补偿处理，生成样本车速补偿地图。这里，能够弥补在移动过程中，基于样本点云信息生成的地图中产生的畸变。将样本车速补偿地图输入至预设模型，输出预测地图。根据预测地图和基于样本定位信息生成的样本高精地图计算损失值，根据损失值训练预设模型，能够不断减小预测地图和样本高精地图之间的差异，直至预设模型满足预设训练条件，得到预测模型。In the embodiment of the present application, multiple sample data collected by the mobile device during movement are obtained, where each sample data includes sample positioning information, sample speed information, and sample point cloud information. Motion compensation is performed on the sample point cloud information based on the sample speed information to generate a sample vehicle speed compensation map. Here, the distortion produced in the map generated based on sample point cloud information during movement can be compensated. Input the sample vehicle speed compensation map to the preset model and output the prediction map. Calculate the loss value based on the prediction map and the sample high-precision map generated based on the sample positioning information, and train the preset model based on the loss value, which can continuously reduce the difference between the prediction map and the sample high-precision map until the preset model meets the preset training conditions to get the prediction model.

上述说明仅是本发明技术方案的概述，为了能够更清楚了解本发明的技术手段，而可依照说明书的内容予以实施，并且为了让本发明的上述和其它目的、特征和优点能够更明显易懂，以下特举本发明的具体实施方式。The above description is only an overview of the technical solution of the present invention. In order to have a clearer understanding of the technical means of the present invention, it can be implemented according to the content of the description, and in order to make the above and other objects, features and advantages of the present invention more obvious and understandable. , the specific embodiments of the present invention are listed below.

附图说明Description of the drawings

为了更清楚地说明本申请实施例的技术方案，下面将对本申请实施例中所需要使用的附图作简单的介绍，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。In order to explain the technical solutions of the embodiments of the present application more clearly, the drawings required to be used in the embodiments of the present application will be briefly introduced below. For those of ordinary skill in the art, without exerting creative efforts, they can also Additional drawings can be obtained from these drawings.

图1是本申请实施例提供的一种预测模型的训练过程和应用过程的示意图；Figure 1 is a schematic diagram of the training process and application process of a prediction model provided by the embodiment of the present application;

图2是本申请实施例提供的一种预测模型的训练方法的流程图；Figure 2 is a flow chart of a training method for a prediction model provided by an embodiment of the present application;

图3是本申请实施例提供的一种模型结构的示意图；Figure 3 is a schematic diagram of a model structure provided by an embodiment of the present application;

图4是本申请实施例提供的一种地图预测方法的流程图；Figure 4 is a flow chart of a map prediction method provided by an embodiment of the present application;

图5是本申请实施例提供的一种预测模型的训练装置的结构示意图；Figure 5 is a schematic structural diagram of a training device for a prediction model provided by an embodiment of the present application;

图6是本申请实施例提供的一种地图预测装置的结构示意图；Figure 6 is a schematic structural diagram of a map prediction device provided by an embodiment of the present application;

图7是本申请实施例提供的一种电子设备的结构示意图。FIG. 7 is a schematic structural diagram of an electronic device provided by an embodiment of the present application.

具体实施例Specific embodiments

下面将详细描述本申请的各个方面的特征和示例性实施例，为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及具体实施例，对本申请进行进一步详细描述。应理解，此处所描述的具体实施例仅被配置为解释本申请，并不被配置为限定本申请。对于本领域技术人员来说，本申请可以在不需要这些具体细节中的一些细节的情况下实施。下面对实施例的描述仅仅是为了通过示出本申请的示例来提供对本申请更好的理解。Features and exemplary embodiments of various aspects of the present application will be described in detail below. In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be described in further detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only configured to explain the present application and are not configured to limit the present application. It will be apparent to one skilled in the art that the present application may be practiced without some of these specific details. The following description of embodiments is merely intended to provide a better understanding of the present application by illustrating examples thereof.

需要说明的是，在本文中，诸如第一和第二等之类的关系术语仅仅用来将一个实体或者操作与另一个实体或操作区分开来，而不一定要求或者暗示这些实体或操作之间存在任何这种实际的关系或者顺序。而且，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、物品或者设备不仅包括那些要素，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、物品或者设备所固有的要素。在没有更多限制的情况下，由语句“包括……”限定的要素，并不排除在包括要素的过程、方法、物品或者设备中还存在另外的相同要素。It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that these entities or operations are mutually exclusive. any such actual relationship or sequence exists between them. Furthermore, the terms "comprises," "comprises," or any other variations thereof are intended to cover a non-exclusive inclusion such that a process, method, article, or apparatus that includes a list of elements includes not only those elements, but also those not expressly listed other elements, or elements inherent to the process, method, article or equipment. Without further limitation, an element defined by the statement "including..." does not exclude the presence of additional identical elements in the process, method, article, or device that includes the element.

首先，对于本申请实施例涉及的技术术语进行介绍。First, the technical terms involved in the embodiments of this application are introduced.

高精地图，是相对于普通地图来说的，它提供了更高精度，内容更为丰富的地图信息，主要服务于自动驾驶。High-precision maps, compared to ordinary maps, provide higher-precision and richer map information, mainly serving autonomous driving.

在逆向工程中通过测量仪器得到的产品外观表面的点数据集合也称之为点云，通常使用三维坐标测量机所得到的点数量比较少，点与点的间距也比较大，叫稀疏点云；而使用三维激光扫描仪或照相式扫描仪得到的点云，点数量比较大并且比较密集，叫密集点云。In reverse engineering, the collection of point data on the appearance surface of a product obtained through measuring instruments is also called a point cloud. Usually, the number of points obtained by using a three-dimensional coordinate measuring machine is relatively small, and the distance between points is relatively large, which is called a sparse point cloud. ; The point cloud obtained by using a 3D laser scanner or a photographic scanner has a relatively large number of points and is relatively dense, which is called a dense point cloud.

点云是一个数据集，数据集中的每个点代表一组X、Y、Z几何坐标和一个强度值，这个强度值根据物体表面反射率记录返回信号的强度。当这些点组合在一起时，就会形成一个点云，即空间中代表三维形状或对象的数据点集合。A point cloud is a data set. Each point in the data set represents a set of X, Y, Z geometric coordinates and an intensity value. This intensity value records the intensity of the returned signal based on the reflectivity of the object surface. When these points are combined together, they form a point cloud, a collection of data points in space that represents a three-dimensional shape or object.

语义图，均属于概率图范畴，是世界的一个简化概率模型，为无人驾驶提供先验知识。语义可以从图像深度学习中获得，通过分类、检测、分割等模型获得。Semantic graphs all belong to the category of probability graphs. They are a simplified probability model of the world and provide prior knowledge for autonomous driving. Semantics can be obtained from image deep learning through classification, detection, segmentation and other models.

梯度提升决策树(Gradient Boosting Decision Tree，GBDT)，是一种基于boosting增强策略的加法模型，训练的时候采用前向分布算法进行贪婪的学习，每次迭代都学习一棵分类回归树(Classification And Regression Tree，CART)来拟合之前t-1棵树的预测结果与训练样本真实值的残差。Gradient Boosting Decision Tree (GBDT) is an additive model based on the boosting enhancement strategy. During training, the forward distribution algorithm is used for greedy learning. Each iteration learns a classification and regression tree (Classification And Regression Tree, CART) to fit the residual between the prediction result of the previous t-1 tree and the true value of the training sample.

XGBoost对GBDT进行了一系列优化，比如损失函数进行了二阶泰勒展开、目标函数加入正则项、支持并行和默认缺失值处理等，在可扩展性和训练速度上有了巨大的提升。XGBoost has made a series of optimizations for GBDT, such as second-order Taylor expansion of the loss function, adding regular terms to the objective function, supporting parallelism and default missing value processing, etc., which has greatly improved the scalability and training speed.

本申请实施例提供的方法至少可以应用于下述应用场景中，下面进行说明。The methods provided by the embodiments of this application can be applied to at least the following application scenarios, which are described below.

在依靠高精地图的辅助驾驶领域，高精地图是利用本车接收到的定位信息，实施解析本车周边道路场景，进行拼接平滑过滤后得到的。In the field of assisted driving that relies on high-precision maps, high-precision maps are obtained by using the positioning information received by the vehicle to analyze the road scenes around the vehicle and perform splicing and smoothing filtering.

所以高精地图依赖定位信息，但是在隧道等场景下，车载定位装置无法接收到定位信号或定位信号较弱，此时无法利用位置信息获取车辆附近高精地图信息。Therefore, high-precision maps rely on positioning information. However, in tunnels and other scenarios, the vehicle-mounted positioning device cannot receive positioning signals or the positioning signals are weak. At this time, location information cannot be used to obtain high-precision map information near the vehicle.

在激光雷达应用中，一帧点云是由激光雷达逐行逐列的扫描获得，在扫描时难免车辆自身姿态、运动状态或路面不发生变化，这种变化会导致激光发射原点发生变化，使得到同一目标的激光往返飞行时间发生变化，进而引起激光点云扫描到的物体发生畸变。In lidar applications, a frame of point cloud is obtained by scanning line by line with lidar. It is inevitable that the vehicle's posture, motion state or road surface will not change during scanning. This change will cause the laser emission origin to change, causing The round-trip flight time of the laser to the same target changes, which in turn causes distortion of the objects scanned by the laser point cloud.

虽然根据车辆自身运动状态的信息可以减小这种畸变，受限于计算量及车辆自身状态信号的精度等因素的影响，这种畸变并不能完全消除，并且会对构建局部语义地图产生影响。Although this distortion can be reduced based on the information about the vehicle's own motion state, this distortion cannot be completely eliminated due to factors such as the amount of calculation and the accuracy of the vehicle's own state signal, and will have an impact on the construction of local semantic maps.

此外，局部语义地图中车道线的构建依赖于车道线和其他地面反射率的差异。当车道线表面被泥土、污渍、雪等物体覆盖，或者当周边出现高反射率的目标(如积水)，或者出现激光雷达间相互串扰时，局部语义地图中车道线的构建就会收到影响。In addition, the construction of lane lines in local semantic maps relies on the difference in reflectivity between lane lines and other ground surfaces. When the surface of the lane lines is covered by objects such as mud, stains, snow, etc., or when there are high-reflectivity targets (such as stagnant water) in the surroundings, or when crosstalk between lidars occurs, the construction of lane lines in the local semantic map will receive Influence.

基于上述应用场景，下面对本申请实施例提供的预测模型的训练方法和地图预测方法进行详细说明。Based on the above application scenarios, the training method of the prediction model and the map prediction method provided by the embodiments of the present application are described in detail below.

下面首先对本申请实施例提供的预测模型进行整体性说明。The following first provides an overall description of the prediction model provided by the embodiment of the present application.

图1为本申请实施例提供的一种预测模型的训练过程和应用过程的示意图，如图1所示，分为训练过程110和应用过程120。Figure 1 is a schematic diagram of the training process and application process of a prediction model provided by the embodiment of the present application. As shown in Figure 1, it is divided into a training process 110 and an application process 120.

在训练过程110中，获取移动装置在移动过程中采集到的多个样本数据111，其中，每个样本数据111包括样本定位信息、样本速度信息和样本点云信息。根据样本速度信息对样本点云信息进行运动补偿处理，生成样本车速补偿地图。这里，能够弥补在移动过程中，基于样本点云信息生成的地图中产生的畸变。将样本车速补偿地图输入至预设模型112，输出预测地图113。根据预测地图和基于样本定位信息生成的样本高精地图114计算损失值115，根据损失值115训练预设模型，能够不断减小预测地图和样本高精地图之间的差异，直至预设模型满足预设训练条件，得到预测模型122。In the training process 110, multiple sample data 111 collected by the mobile device during movement are obtained, where each sample data 111 includes sample positioning information, sample speed information, and sample point cloud information. Motion compensation is performed on the sample point cloud information based on the sample speed information to generate a sample vehicle speed compensation map. Here, the distortion produced in the map generated based on sample point cloud information during movement can be compensated. The sample vehicle speed compensation map is input to the preset model 112 and the prediction map 113 is output. Calculate the loss value 115 based on the predicted map and the sample high-precision map 114 generated based on the sample positioning information, and train the preset model based on the loss value 115, which can continuously reduce the difference between the predicted map and the sample high-precision map until the preset model meets The training conditions are preset to obtain the prediction model 122.

在应用过程120中，获取移动装置采集的运动数据121，运动数据至少包括：速度信息和点云信息。根据速度信息对点云信息进行运动补偿处理，生成车速补偿地图。将车速补偿地图输入至预测模型122，输出目标地图123。In the application process 120, motion data 121 collected by the mobile device is obtained. The motion data at least includes: speed information and point cloud information. Motion compensation is performed on the point cloud information based on the speed information to generate a vehicle speed compensation map. The vehicle speed compensation map is input to the prediction model 122 and the target map 123 is output.

由于训练好的预测模型，能够基于车速补偿地图快速准确地预测出，与高精地图非常接近的目标地图。根据速度信息对点云信息进行运动补偿处理，能够弥补在移动过程中，基于点云信息生成的地图中产生的畸变。由此，在检测不到定位信息的情况下，也能通过车速补偿地图输入至训练好的预测模型，快速准确地输出接近于高精地图的目标地图。Due to the well-trained prediction model, a target map that is very close to the high-precision map can be quickly and accurately predicted based on the vehicle speed compensation map. Motion compensation processing of point cloud information based on speed information can compensate for the distortion produced in the map generated based on point cloud information during movement. As a result, even when positioning information cannot be detected, the vehicle speed compensation map can be input to the trained prediction model, and a target map close to a high-precision map can be quickly and accurately output.

下面结合附图，通过具体的实施例及其应用场景对本申请实施例提供的预测模型的训练方法以及地图预测方法分别进行详细地说明。The training method of the prediction model and the map prediction method provided by the embodiments of the present application will be described in detail below with reference to the accompanying drawings through specific embodiments and application scenarios.

下面先对预测模型的训练方法进行说明。Next, the training method of the prediction model will be explained first.

图2为本申请实施例提供的一种预测模型的训练方法的流程图。FIG. 2 is a flow chart of a prediction model training method provided by an embodiment of the present application.

如图2所示，该预测模型的训练方法可以包括步骤210-步骤250，具体如下所示：As shown in Figure 2, the training method of the prediction model may include steps 210 to 250, specifically as follows:

步骤210，获取移动装置在移动过程中采集到的多个样本数据，每个样本数据包括样本定位信息、样本速度信息和样本点云信息。Step 210: Obtain multiple sample data collected by the mobile device during movement. Each sample data includes sample positioning information, sample speed information, and sample point cloud information.

步骤220，根据样本速度信息对样本点云信息进行运动补偿处理，生成样本车速补偿地图。Step 220: Perform motion compensation processing on the sample point cloud information according to the sample speed information to generate a sample vehicle speed compensation map.

步骤230，将样本车速补偿地图输入至预设模型，输出预测地图。Step 230: Input the sample vehicle speed compensation map to the default model and output the prediction map.

步骤240，根据预测地图和样本高精地图计算损失值，样本高精地图根据样本定位信息生成。Step 240: Calculate the loss value based on the prediction map and the sample high-precision map, and the sample high-precision map is generated based on the sample positioning information.

步骤250，根据损失值训练预设模型，直至预设模型满足预设训练条件，得到预测模型。Step 250: Train the preset model according to the loss value until the preset model meets the preset training conditions to obtain a prediction model.

下面，对步骤210-步骤250的内容分别进行描述：Below, the contents of steps 210 to 250 are described respectively:

涉及步骤210。Step 210 is involved.

获取移动装置在移动过程中采集到的多个样本数据，每个样本数据包括样本定位信息、样本速度信息和样本点云信息。Acquire multiple sample data collected by the mobile device during movement. Each sample data includes sample positioning information, sample speed information, and sample point cloud information.

其中，移动装置可以为车辆、飞行器或机器人等，可以移动的装置。The mobile device may be a vehicle, an aircraft, a robot, etc., and may be a mobile device.

其中，样本定位信息可以包括：全球定位***(Global Positioning System，GPS)信息，和全球卫星导航***(Global Navigation Satellite System，GNSS)信息。Among them, the sample positioning information may include: Global Positioning System (GPS) information and Global Navigation Satellite System (GNSS) information.

样本速度信息可以包括：底盘轮速信息和车速信息。Sample speed information may include: chassis wheel speed information and vehicle speed information.

轮速信息可以由移动装置中设置的轮速传感器采集得到。车速信息可以由移动装置中设置的速度传感器采集得到。Wheel speed information can be collected by a wheel speed sensor installed in the mobile device. The vehicle speed information can be collected by a speed sensor installed in the mobile device.

样本点云信息可以为：激光雷达点云信息。The sample point cloud information can be: lidar point cloud information.

激光雷达点云信息，是由三维激光雷达设备扫描得到的空间点的数据集，每一个点都包含了三维坐标信息，也是X、Y、Z三个元素，有的还包含颜色信息、反射强度信息、回波次数信息等。激光点云信息，由激光扫描***向周围发射激光信号，然后收集反射的激光信号得来的，再通过外业数据采集、组合导航、点云解算，便可以计算出这些点的准确空间信息。Lidar point cloud information is a data set of spatial points scanned by a three-dimensional lidar device. Each point contains three-dimensional coordinate information, which is also the three elements of X, Y, and Z. Some also contain color information and reflection intensity. information, echo number information, etc. Laser point cloud information is obtained by the laser scanning system emitting laser signals to the surroundings, and then collecting the reflected laser signals. Then through field data collection, integrated navigation, and point cloud calculation, the accurate spatial information of these points can be calculated .

涉及步骤220。Step 220 is involved.

根据样本速度信息对样本点云信息进行运动补偿处理，生成样本车速补偿地图。Motion compensation is performed on the sample point cloud information based on the sample speed information to generate a sample vehicle speed compensation map.

根据样本速度信息对每一帧的样本点云信息进行运动补偿，获得车道线和路沿方程，即样本车速补偿地图。Motion compensation is performed on the sample point cloud information of each frame according to the sample speed information, and the lane line and curb equations are obtained, that is, the sample speed compensation map.

车行道边缘线是用来指示机动车道的边缘或用来划分机动车与非机动车道分界的线。车行道边缘线分为实线边缘线和虚线边缘线两种，其颜色为白色。The roadway edge line is a line used to indicate the edge of a motor vehicle lane or to delineate the boundary between motor vehicle and non-motor vehicle lanes. There are two types of roadway edge lines: solid edge lines and dotted edge lines. Their color is white.

路沿，是道路边缘的标示，用于提醒路面障碍及宽度。Curbs are markings on the edge of the road to remind you of road obstacles and width.

涉及步骤230。Step 230 is involved.

将样本车速补偿地图输入至预设模型，输出预测地图。Input the sample vehicle speed compensation map to the preset model and output the prediction map.

涉及步骤240。Step 240 is involved.

根据预测地图和样本高精地图计算损失值，样本高精地图根据样本定位信息生成。The loss value is calculated based on the prediction map and the sample high-precision map, and the sample high-precision map is generated based on the sample positioning information.

其中，在步骤240之前，还可以包括以下步骤：Before step 240, the following steps may also be included:

根据样本定位信息确定样本高精地图。Determine the sample high-precision map based on the sample positioning information.

根据模型输出的预测地图和样本高精地图计算损失值，样本高精地图根据样本定位信息生成。The loss value is calculated based on the prediction map and sample high-precision map output by the model, and the sample high-precision map is generated based on the sample positioning information.

在一种可能的实施例中，在步骤240之前，还可以包括以下步骤：In a possible embodiment, before step 240, the following steps may also be included:

根据样本定位信息生成初始样本高精地图；Generate an initial sample high-precision map based on sample positioning information;

根据样本定位信息对样本点云信息进行运动补偿，生成定位补偿地图；Perform motion compensation on the sample point cloud information based on the sample positioning information and generate a positioning compensation map;

根据定位补偿地图，从初始样本高精地图中筛选出预设扫描范围内的样本高精地图。According to the positioning compensation map, the sample high-precision map within the preset scanning range is selected from the initial sample high-precision map.

由于一帧样本点云信息是由激光雷达逐行逐列的扫描获得，在扫描时周边环境可能会发生变化，这种变化会导致激光发射原点发生变化，使得到同一目标的激光往返飞行时间发生变化，进而引起激光点云扫描到的物体发生畸变。这时为了避免或减小这种畸变，可以根据样本定位信息对样本点云信息进行运动补偿，生成定位补偿地图。Since a frame of sample point cloud information is obtained by scanning the laser radar row by row, the surrounding environment may change during scanning. This change will cause the laser emission origin to change, causing the laser round-trip flight time to the same target to change. changes, thereby causing distortion of the objects scanned by the laser point cloud. In order to avoid or reduce this distortion, the sample point cloud information can be motion compensated based on the sample positioning information to generate a positioning compensation map.

根据样本定位信息对样本点云信息进行运动补偿，可建立局部语义地图，并获得车道线和路沿方程，即生成定位补偿地图。By performing motion compensation on the sample point cloud information based on the sample positioning information, a local semantic map can be established, and the lane line and curb equations can be obtained, that is, a positioning compensation map can be generated.

比较定位补偿地图和从初始样本高精地图，利用定位补偿地图筛选，从初始样本高精地图中筛选出预设扫描范围内的样本高精地图，也就是激光雷达实际扫描范围内的样本高精地图。Compare the positioning compensation map and the high-precision map from the initial sample. Use the positioning compensation map to filter out the sample high-precision map within the preset scanning range from the initial sample high-precision map, that is, the sample high-precision map within the actual scanning range of the lidar. map.

其中，上述涉及到的根据定位补偿地图，从初始样本高精地图中筛选出预设扫描范围内的样本高精地图的步骤中，具体可以包括以下步骤：Among them, the above-mentioned step of screening out the sample high-precision map within the preset scanning range from the initial sample high-precision map based on the positioning compensation map may specifically include the following steps:

确定样本点云信息激光雷达采集样本点云信息时的预设扫描范围；Determine the preset scanning range of the sample point cloud information lidar when collecting sample point cloud information;

确定预设扫描范围内的定位补偿地图；Determine the positioning compensation map within the preset scanning range;

从初始样本高精地图中，筛选与预设扫描范围内的定位补偿地图相匹配的样本高精地图。From the initial sample high-precision map, select sample high-precision maps that match the positioning compensation map within the preset scanning range.

确定样本点云信息激光雷达采集样本点云信息时的预设扫描范围，比如，以移动装置中的预设点为圆心，以50米的范围是预设扫描范围。然后，确定该预设扫描范围内的定位补偿地图。初始样本高精地图可以是以移动装置中的预设点为圆心，以200米的范围采集位置信号得到的。最后，从初始样本高精地图中，筛选与预设扫描范围内的定位补偿地图相匹配的样本高精地图。Determine the preset scanning range when the sample point cloud information lidar collects sample point cloud information. For example, the preset point in the mobile device is the center of the circle, and the range of 50 meters is the preset scanning range. Then, determine the positioning compensation map within the preset scanning range. The initial sample high-precision map can be obtained by collecting location signals in a range of 200 meters with the preset point in the mobile device as the center of the circle. Finally, from the initial sample high-precision map, sample high-precision maps that match the positioning compensation map within the preset scanning range are screened.

在一种可能的实施例中，步骤240中，具体可以包括以下步骤：In a possible embodiment, step 240 may specifically include the following steps:

从样本车速补偿地图中提取第一特征向量，第一特征向量用于表征移动装置在移动过程中所处环境的道路特征；道路特征包括车道线特征，和/或，路沿特征；Extract a first feature vector from the sample speed compensation map. The first feature vector is used to characterize the road features of the environment in which the mobile device is moving during movement; the road features include lane line features and/or curb features;

从样本高精地图出提取第二特征向量，第二特征向量用于表征道路特征；Extract the second feature vector from the sample high-precision map, and the second feature vector is used to characterize the road characteristics;

根据第一特征向量和第二特征向量计算损失值。The loss value is calculated based on the first eigenvector and the second eigenvector.

从样本车速补偿地图中提取第一特征向量，第一特征向量可以用于表征车道线特征，和/或，路沿特征；从样本高精地图出提取第二特征向量，第二特征向量也用于表征车道线特征，和/或，路沿特征。Extract the first feature vector from the sample speed compensation map. The first feature vector can be used to characterize lane line features and/or curb features; extract the second feature vector from the sample high-precision map. The second feature vector is also used Used to characterize lane line characteristics and/or curb characteristics.

具体地，当样本车速补偿地图和样本高精地图为车道线和路沿方程时，第一特征向量可以为样本车速补偿地图对应的车道线和路沿方程的系数，第二特征向量可以为样本高精地图对应的车道线和路沿方程的系数。Specifically, when the sample speed compensation map and the sample high-precision map are lane lines and curb equations, the first feature vector can be the coefficients of the lane lines and curb equations corresponding to the sample speed compensation map, and the second feature vector can be the sample Coefficients of lane lines and curb equations corresponding to high-precision maps.

根据第一特征向量和第二特征向量计算损失值，根据损失值训练预设模型，训练目标是使第一特征向量靠近第二特征向量。The loss value is calculated based on the first feature vector and the second feature vector, and the preset model is trained based on the loss value. The training goal is to make the first feature vector close to the second feature vector.

涉及步骤250。Step 250 is involved.

根据损失值训练预设模型，直至预设模型满足预设训练条件，得到预测模型。Train the preset model based on the loss value until the preset model meets the preset training conditions and obtain the prediction model.

具体可以基于Xgboost算法，根据损失值训练预设模型，直至预设模型满足预设训练条件，确定模型参数。将模型参数带入预设模型，得到预测模型。Specifically, the preset model can be trained based on the Xgboost algorithm based on the loss value until the preset model meets the preset training conditions and the model parameters are determined. Bring the model parameters into the preset model to obtain the prediction model.

其中，步骤250中，Xgboost将若干弱学习器的预测结果组合成强学习器，对损失函数进行二阶泰勒展开。二阶泰勒展开主要是为了解决非线性优化问题，其收敛速度比梯度下降速度更快。其需要解决的问题可以描述为：对于目标函数f(x)，在无约束条件的情况下求它的最小值。Among them, in step 250, Xgboost combines the prediction results of several weak learners into a strong learner, and performs a second-order Taylor expansion of the loss function. The second-order Taylor expansion is mainly used to solve nonlinear optimization problems, and its convergence speed is faster than gradient descent. The problem that needs to be solved can be described as: for the objective function f(x), find its minimum value without constraints.

二阶泰勒展开要得到的结果是，在现有的极小值估计值的附近对f(x)做二阶泰勒展开，进而找到极小点的下一个估计值，反复迭代直到函数的一阶导数小于某个接近0的阀值。The result of the second-order Taylor expansion is to perform a second-order Taylor expansion of f(x) near the existing minimum estimated value, and then find the next estimated value of the minimum point, and iterate repeatedly until the first-order function The derivative is less than a certain threshold close to 0.

并且，步骤250中，还将预测项和正则化项结合起来，在优化过程中加入损失函数的二阶导数信息，简化函数，以实现计算资源优化，采用弱分类器集成算法选取合适的参数。正则化项一般是模型复杂程度的单调递增函数，因此可以使用模型参数向量的范数来计算。Moreover, in step 250, the prediction term and the regularization term are also combined, and the second-order derivative information of the loss function is added in the optimization process to simplify the function to achieve computing resource optimization and use a weak classifier integration algorithm to select appropriate parameters. The regularization term is generally a monotonically increasing function of the model complexity and therefore can be calculated using the norm of the model parameter vector.

由于，在机器学习算法中如果只使用经验风险最小化去优化损失函数则很可能造成过拟合的问题，通常在损失函数中加入一些描述模型复杂程度的正则化项，使得模型在拥有较好的预测能力的同时不会因为模型过于复杂而产生过拟合现象，即结构风险最小化。Since in machine learning algorithms, if you only use empirical risk minimization to optimize the loss function, it is likely to cause over-fitting problems. Some regularization terms that describe the complexity of the model are usually added to the loss function to make the model have better performance. While improving the prediction ability, it will not cause over-fitting due to the model being too complex, that is, the structural risk will be minimized.

其中，预设模型包括N个首尾相连的子模型，N为大于1的整数，步骤250中，具体可以包括以下步骤：The preset model includes N sub-models connected end to end, where N is an integer greater than 1. In step 250, the following steps may be included:

将第N-1子模型对应的损失值输入至第N子模型；Input the loss value corresponding to the N-1 sub-model into the N-th sub-model;

根据第N-1子模型对应的损失值和预设阈值，训练预设模型，直至预设模型满足预设训练条件，得到预测模型。According to the loss value and the preset threshold corresponding to the N-1th sub-model, the preset model is trained until the preset model meets the preset training conditions, and a prediction model is obtained.

Xgboost模型底层使用CART，也称CART回归树，有助于算法的高效优化，提升运行速度。The bottom layer of the Xgboost model uses CART, also called CART regression tree, which helps to efficiently optimize the algorithm and improve the running speed.

其中，N个首尾相连的子模型为CART回归树，是一种以二叉树为逻辑结构的，用于完成线性回归任务的决策树。它采用一种二分递归分割的技术，分割方法采用基于最小距离的基尼指数估计函数，将当前的样本集分为两个子样本集，使得生成的的每个非叶子节点都有两个分支。因此，CART算法生成的决策树是结构简洁的二叉树。CART在每一个节点上都采用二分法，即每个节点都只能有两个子节点，最后构成的是二叉树。Among them, N sub-models connected end to end are CART regression trees, which are decision trees with a binary tree as a logical structure and are used to complete linear regression tasks. It uses a bisection recursive segmentation technology. The segmentation method uses the Gini index estimation function based on the minimum distance to divide the current sample set into two sub-sample sets, so that each generated non-leaf node has two branches. Therefore, the decision tree generated by the CART algorithm is a binary tree with a simple structure. CART adopts the dichotomy method on each node, that is, each node can only have two child nodes, and the final structure is a binary tree.

CART回归树，首先是决策树的生成：基于训练数据生成决策树，生成的决策树要尽量大。然后是决策树的剪枝：用验证数据集对以生成的树进行剪枝并选择最优子树。这时用损失函数最小作为剪枝标准。CART regression tree, first is the generation of decision tree: generate decision tree based on training data, and the generated decision tree should be as large as possible. Then there is the pruning of the decision tree: use the verification data set to prune the generated tree and select the optimal subtree. At this time, the minimum loss function is used as the pruning criterion.

其中，Xgboost模型回归树切割点可以采用近似值算法，枚举类算法提高了运行速度。Among them, the Xgboost model regression tree cutting point can use approximation algorithm, and the enumeration algorithm improves the running speed.

其中，近似算法是一种处理优化问题完全性的方式，它无法确保最优解。近似算法的目标是在多项式时间内尽可能地接近最优值。它虽然无法给出精确最优解，但可以将问题收敛到最终解的近似值。Among them, approximation algorithm is a way to deal with the completeness of optimization problems, which cannot ensure the optimal solution. The goal of approximation algorithms is to get as close as possible to the optimal value in polynomial time. Although it cannot give an exact optimal solution, it can converge the problem to an approximation of the final solution.

枚举算法是日常中使用到的最多的一个算法，其核心思想就是枚举所有的可能。枚举法的本质就是从所有候选答案中去搜索正确的解，使用该算法需要满足两个条件：可预先确定候选答案的数量；候选答案的范围在求解之前有一个确定的集合。The enumeration algorithm is the most commonly used algorithm in daily life. Its core idea is to enumerate all possibilities. The essence of the enumeration method is to search for the correct solution from all candidate answers. Using this algorithm needs to meet two conditions: the number of candidate answers can be determined in advance; the range of candidate answers has a determined set before solving.

枚举类算法比如：切割比例可以分别枚举为：1，9；2，8；9，1，等，……。Enumeration algorithms such as: cutting ratio can be enumerated as: 1, 9; 2, 8; 9, 1, etc., etc.

其中，CART回归树的具体训练流程可以包括：Among them, the specific training process of CART regression tree can include:

首先，确定n棵树，每一棵树从全部特征向量中随机有放回的选取若干特征；然后，每一棵树从它所拥有的特征中最小平方误差确定最佳***点，并根据树的深度，叶子节点数量确定是否提前停止***；接着，保存每一棵树的最佳***点；将样本点云信息作为第一棵回归树训练输入，利用L2正则损失函数避免过拟合。这里，为防止模型过拟合，提高模型的泛化能力，通常会在损失函数的后面添加一个正则化项。L1正则化和L2正则化可以看做是损失函数的惩罚项。所谓“惩罚”是指对损失函数中的某些参数做一些限制。First, n trees are determined, and each tree randomly selects several features with replacement from all feature vectors; then, each tree determines the best split point from the minimum square error of the features it possesses, and based on the tree The depth and number of leaf nodes determine whether to stop splitting in advance; then, the best split point of each tree is saved; the sample point cloud information is used as the training input of the first regression tree, and the L2 regular loss function is used to avoid overfitting. Here, in order to prevent the model from overfitting and improve the generalization ability of the model, a regularization term is usually added after the loss function. L1 regularization and L2 regularization can be regarded as the penalty term of the loss function. The so-called "penalty" refers to placing some restrictions on certain parameters in the loss function.

然后，如图3所示，将样本点云信息，以及第一棵回归树输出的预测地图和样本高精地图之间的损失值作为第二棵树的输入；训练目标是使损失值无限接近于0。以此类推，将第N-1子模型对应的损失值输入至第N子模型；根据第N-1子模型对应的损失值和预设阈值，训练预设模型，直至预设模型满足预设训练条件，得到预测模型。Then, as shown in Figure 3, the sample point cloud information and the loss value between the prediction map output by the first regression tree and the sample high-precision map are used as the input of the second tree; the training goal is to make the loss value infinitely close at 0. By analogy, the loss value corresponding to the N-1 sub-model is input to the N-th sub-model; based on the loss value corresponding to the N-1 sub-model and the preset threshold, the preset model is trained until the preset model meets the preset training conditions to obtain the prediction model.

也就是说，对于第N棵树，前N-1棵树作为一个整体，其预测输出与目标值残差作为训练输入。每次迭代都学习一棵CART树来拟合之前N-1棵树的预测结果与训练样本真实值的残差。That is to say, for the Nth tree, the first N-1 trees are taken as a whole, and their predicted output and target value residuals are used as training input. Each iteration learns a CART tree to fit the residual between the prediction results of the previous N-1 trees and the true values of the training samples.

图4为本申请实施例提供的一种地图预测方法的流程图。Figure 4 is a flow chart of a map prediction method provided by an embodiment of the present application.

如图4所示，该地图预测方法可以包括步骤410-步骤430，该方法应用于地图预测装置，具体如下所示：As shown in Figure 4, the map prediction method may include steps 410 to 430. The method is applied to the map prediction device, specifically as follows:

步骤410，获取移动装置采集的运动数据，运动数据至少包括：速度信息和点云信息。Step 410: Obtain motion data collected by the mobile device. The motion data at least includes: speed information and point cloud information.

其中，速度信息可以包括：底盘轮速信息和车速信息。轮速信息可以由移动装置中设置的轮速传感器采集得到。车速信息可以由移动装置中设置的速度传感器采集得到。Among them, the speed information may include: chassis wheel speed information and vehicle speed information. Wheel speed information can be collected by a wheel speed sensor installed in the mobile device. The vehicle speed information can be collected by a speed sensor installed in the mobile device.

点云信息可以为：激光雷达点云信息。Point cloud information can be: lidar point cloud information.

步骤420，根据速度信息对点云信息进行运动补偿处理，生成车速补偿地图。Step 420: Perform motion compensation processing on the point cloud information according to the speed information to generate a vehicle speed compensation map.

根据速度信息对点云信息进行运动补偿处理，能够弥补在移动过程中，基于点云信息生成的地图中产生的畸变。Motion compensation processing of point cloud information based on speed information can compensate for the distortion produced in the map generated based on point cloud information during movement.

步骤430，将车速补偿地图输入至预测模型，输出目标地图。Step 430: Input the vehicle speed compensation map to the prediction model and output the target map.

在一种可能的实施例中，步骤430中，具体可以包括以下步骤：In a possible embodiment, step 430 may specifically include the following steps:

在预设时间段内，将车速补偿地图输入至预测模型，得到第一目标地图；预测模型中设置有多个与时间标识信息相对应的参数；Within a preset time period, the vehicle speed compensation map is input into the prediction model to obtain the first target map; the prediction model is set with multiple parameters corresponding to the time stamp information;

确定运动数据被采集时对应的时间标识信息；Determine the time stamp information corresponding to when the motion data is collected;

根据时间标识信息对应的参数，对第一目标地图进行调整，得到目标地图。According to the parameters corresponding to the time stamp information, the first target map is adjusted to obtain the target map.

这里，在预设时间段内，将车速补偿地图输入至预测模型，得到第一目标地图；预设时间段内可以为定位信息减弱或消失后的预设时间段内，比如，预设时间段可以为2分钟。然后，确定运动数据被采集时对应的时间标识信息。Here, within a preset time period, the vehicle speed compensation map is input into the prediction model to obtain the first target map; the preset time period can be a preset time period after the positioning information weakens or disappears, for example, the preset time period Can be 2 minutes. Then, determine the time stamp information corresponding to when the motion data was collected.

其中，上述涉及到的参数用于调整模型输出的第一目标地图进行调整，起到衰减因子的作用。Among them, the above-mentioned parameters are used to adjust the first target map output by the model and function as attenuation factors.

其中，预测模型中设置有多个与时间标识信息相对应的参数；时间标识信息可以包括第一时间标识信息、第二时间标识信息，……，第N时间标识信息。与时间标识信息相对应的参数可以分别为：1、0.8、0.6，……。The prediction model is provided with multiple parameters corresponding to the time stamp information; the time stamp information may include first time stamp information, second time stamp information, ..., Nth time stamp information. The parameters corresponding to the time stamp information can be: 1, 0.8, 0.6, ....

例如，第一时间标识信息、第二时间标识信息，第N时间标识信息可以为：第一个10秒内、第二个10秒内，第N个10秒内。或者，第五个10秒内、第十个10秒内，第N个10秒内。在此不做限定。For example, the first time identification information, the second time identification information, and the Nth time identification information can be: within the first 10 seconds, within the second 10 seconds, and within the Nth 10 seconds. Or, within the fifth 10 seconds, within the tenth 10 seconds, and within the Nth 10 seconds. No limitation is made here.

根据时间标识信息对应的参数，对第一目标地图进行调整，得到目标地图。即可以使第一目标地图与参数相乘，得到目标地图，以实现使目标地图逐渐地平稳过渡到车速补偿地图。According to the parameters corresponding to the time stamp information, the first target map is adjusted to obtain the target map. That is, the first target map can be multiplied by the parameters to obtain the target map, so as to gradually and smoothly transition the target map to the vehicle speed compensation map.

示例性地，以第一目标地图中的一个车道线y为例，假设与时间标识信息相对应的参数分别为：1、0.8、0.6，……。在第一个10秒内，根据时间标识信息对应的参数，对第一目标地图进行调整，得到的目标地图为y，修正量为Δy；在个10秒内，根据时间标识信息对应的参数，对第一目标地图进行调整，修正量为0.8*Δy，，以此类推。For example, taking a lane line y in the first target map as an example, assume that the parameters corresponding to the time identification information are: 1, 0.8, 0.6, .... Within the first 10 seconds, the first target map is adjusted according to the parameters corresponding to the time stamp information. The obtained target map is y and the correction amount is Δy; within the first 10 seconds, according to the parameters corresponding to the time stamp information, Adjust the first target map with a correction amount of 0.8*Δy, and so on.

由此，根据时间标识信息对应的参数，对第一目标地图进行调整，得到目标地图，让目标地图逐渐地接近于车速补偿地图，最终，让地图完全切换成车速补偿地图。Therefore, according to the parameters corresponding to the time stamp information, the first target map is adjusted to obtain the target map, so that the target map gradually approaches the vehicle speed compensation map, and finally, the map is completely switched to the vehicle speed compensation map.

在未检测到第一时间段对应的定位信息的情况下，将车速补偿地图输入至预测模型，输出目标地图；When the positioning information corresponding to the first time period is not detected, the vehicle speed compensation map is input into the prediction model and the target map is output;

将高精地图，切换显示为预测地图，高精地图根据第二时间段对应的定位信息确定，第二时间段先于第一时间段。Switch the display of the high-precision map to a predicted map. The high-precision map is determined based on the positioning information corresponding to the second time period, and the second time period precedes the first time period.

其中，若检测到第一时间段对应的定位信息，则说明此时无法生成高精地图，所以需要将车速补偿地图输入至预测模型，输出目标地图，以避免由于定位信息消失导致的移动装置的控制部分剧烈变化，影响用户使用。Among them, if the positioning information corresponding to the first time period is detected, it means that the high-precision map cannot be generated at this time, so the vehicle speed compensation map needs to be input into the prediction model and the target map is output to avoid the loss of the mobile device due to the disappearance of the positioning information. The control part changes drastically, affecting the user's use.

这里，由于训练好的预测模型，能够基于车速补偿地图快速准确地预测出与高精地图非常接近的目标地图，所以将高精地图切换显示为预测地图，可以保证由高精地图到目标地图之间的平滑过渡，提升用户体验。Here, since the trained prediction model can quickly and accurately predict the target map that is very close to the high-precision map based on the vehicle speed compensation map, switching the display of the high-precision map to the prediction map can ensure the transition from the high-precision map to the target map. Smooth transition between images to enhance user experience.

本申请实施例中，根据速度信息对点云信息进行运动补偿处理，能够弥补在移动过程中，基于点云信息生成的地图中产生的畸变。由于训练好的预测模型，能够基于车速补偿地图快速准确地预测出，与高精地图非常接近的目标地图。由此，在检测不到定位信息的情况下，也能通过将车速补偿地图输入至训练好的预测模型，快速准确地输出接近高精地图的目标地图。In the embodiment of the present application, motion compensation processing is performed on the point cloud information based on the speed information, which can compensate for the distortion generated in the map generated based on the point cloud information during the movement. Due to the well-trained prediction model, a target map that is very close to the high-precision map can be quickly and accurately predicted based on the vehicle speed compensation map. As a result, even when positioning information cannot be detected, the vehicle speed compensation map can be input into the trained prediction model to quickly and accurately output a target map that is close to a high-precision map.

基于上述图2所示的预测模型的训练方法，本申请实施例还提供一种预测模型的训练装置，如图5所示，该装置500可以包括：Based on the training method of the prediction model shown in Figure 2, embodiments of the present application also provide a training device for the prediction model. As shown in Figure 5, the device 500 may include:

第一获取模块510，用于获取移动装置在移动过程中采集到的多个样本数据，每个样本数据包括样本定位信息、样本速度信息和样本点云信息。The first acquisition module 510 is used to acquire multiple sample data collected by the mobile device during movement. Each sample data includes sample positioning information, sample speed information, and sample point cloud information.

第一补偿模块520，用于根据样本速度信息对样本点云信息进行运动补偿处理，生成样本车速补偿地图。The first compensation module 520 is used to perform motion compensation processing on the sample point cloud information according to the sample speed information, and generate a sample vehicle speed compensation map.

第一输入模块530，用于将样本车速补偿地图输入至预设模型，输出预测地图。The first input module 530 is used to input the sample vehicle speed compensation map into the preset model and output the predicted map.

计算模块540，用于根据预测地图和样本高精地图计算损失值，样本高精地图根据样本定位信息生成。The calculation module 540 is used to calculate the loss value based on the prediction map and the sample high-precision map, and the sample high-precision map is generated based on the sample positioning information.

训练模块550，用于根据损失值训练预设模型，直至预设模型满足预设训练条件，得到预测模型。The training module 550 is used to train the preset model according to the loss value until the preset model meets the preset training conditions to obtain the prediction model.

在一种可能的实现方式中，该装置500还可以包括：In a possible implementation, the device 500 may also include:

生成模块，用于根据样本定位信息生成初始样本高精地图。The generation module is used to generate an initial sample high-precision map based on the sample positioning information.

补偿模块，用于根据样本定位信息对样本点云信息进行运动补偿，生成定位补偿地图。The compensation module is used to perform motion compensation on the sample point cloud information based on the sample positioning information and generate a positioning compensation map.

筛选模块，用于根据定位补偿地图，从初始样本高精地图中筛选出预设扫描范围内的样本高精地图。The filtering module is used to filter out sample high-precision maps within the preset scanning range from the initial sample high-precision map based on the positioning compensation map.

在一种可能的实现方式中，筛选模块，具体用于：In a possible implementation, the screening module is specifically used for:

在一种可能的实现方式中，计算模块540，具体用于：In a possible implementation, the calculation module 540 is specifically used to:

在一种可能的实现方式中，预设模型包括N个首尾相连的子模型，N为大于1的整数，训练模块550，具体用于：In a possible implementation, the preset model includes N sub-models connected end to end, N is an integer greater than 1, and the training module 550 is specifically used for:

基于上述图4所示的地图预测方法，本申请实施例还提供一种地图预测装置，如图6所示，该装置600可以包括：Based on the above-mentioned map prediction method shown in Figure 4, an embodiment of the present application also provides a map prediction device. As shown in Figure 6, the device 600 may include:

第二获取模块610，用于获取移动装置采集的运动数据，运动数据至少包括：速度信息和点云信息。The second acquisition module 610 is used to acquire motion data collected by the mobile device. The motion data at least includes: speed information and point cloud information.

第二补偿模块620，用于根据速度信息对点云信息进行运动补偿处理，生成车速补偿地图。The second compensation module 620 is used to perform motion compensation processing on the point cloud information according to the speed information and generate a vehicle speed compensation map.

第二输入模块630，用于将车速补偿地图输入至预测模型，输出目标地图。The second input module 630 is used to input the vehicle speed compensation map into the prediction model and output the target map.

在一种可能的实现方式中，第二输入模块630，具体用于：In a possible implementation, the second input module 630 is specifically used to:

综上，本申请实施例中，根据速度信息对点云信息进行运动补偿处理，能够弥补在移动过程中，基于点云信息生成的地图中产生的畸变。由于训练好的预测模型，能够基于车速补偿地图快速准确地预测出，与高精地图非常接近的目标地图。由此，在检测不到定位信息的情况下，也能通过将车速补偿地图输入至训练好的预测模型，快速准确地输出接近高精地图的目标地图。In summary, in the embodiments of the present application, motion compensation processing is performed on point cloud information based on speed information, which can compensate for the distortion generated in the map generated based on point cloud information during movement. Due to the well-trained prediction model, a target map that is very close to the high-precision map can be quickly and accurately predicted based on the vehicle speed compensation map. As a result, even when positioning information cannot be detected, the vehicle speed compensation map can be input into the trained prediction model to quickly and accurately output a target map close to a high-precision map.

图7示出了本申请实施例提供的一种电子设备的结构示意图。FIG. 7 shows a schematic structural diagram of an electronic device provided by an embodiment of the present application.

在电子设备可以包括处理器701以及存储有计算机程序指令的存储器702。The electronic device may include a processor 701 and a memory 702 storing computer program instructions.

具体地，上述处理器701可以包括中央处理器(CPU)，或者特定集成电路(Application Specific Integrated Circuit，ASIC)，或者可以被配置成实施本申请实施例的一个或多个集成电路。Specifically, the above-mentioned processor 701 may include a central processing unit (CPU), or an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), or may be configured to implement one or more integrated circuits according to the embodiments of the present application.

存储器702可以包括用于数据或指令的大容量存储器。举例来说而非限制，存储器702可包括硬盘驱动器(Hard Disk Drive，HDD)、软盘驱动器、闪存、光盘、磁光盘、磁带或通用串行总线(Universal Serial Bus，USB)驱动器或者两个或更多个以上这些的组合。在合适的情况下，存储器702可包括可移除或不可移除(或固定)的介质。在合适的情况下，存储器702可在综合网关容灾设备的内部或外部。在特定实施例中，存储器702是非易失性固态存储器。在特定实施例中，存储器702包括只读存储器(ROM)。在合适的情况下，该ROM可以是掩模编程的ROM、可编程ROM(PROM)、可擦除PROM(EPROM)、电可擦除PROM(EEPROM)、电可改写ROM(EAROM)或闪存或者两个或更多个以上这些的组合。 Memory 702 may include bulk storage for data or instructions. By way of example, and not limitation, memory 702 may include a hard disk drive (HDD), a floppy disk drive, flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a Universal Serial Bus (USB) drive or two or more A combination of many of the above. Memory 702 may include removable or non-removable (or fixed) media, where appropriate. Where appropriate, the memory 702 may be internal or external to the integrated gateway disaster recovery device. In certain embodiments, memory 702 is non-volatile solid-state memory. In certain embodiments, memory 702 includes read-only memory (ROM). Where appropriate, the ROM may be a mask-programmed ROM, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), electrically rewritable ROM (EAROM) or flash memory or A combination of two or more of these.

处理器701通过读取并执行存储器702中存储的计算机程序指令，以实现图所示实施例中的任意一种方法。The processor 701 reads and executes the computer program instructions stored in the memory 702 to implement any method in the embodiment shown in the figure.

在一个示例中，电子设备还可包括通信接口703和总线710。其中，如图7所示，处理器701、存储器702、通信接口703通过总线710连接并完成相互间的通信。In one example, the electronic device may also include a communication interface 703 and a bus 710 . Among them, as shown in Figure 7, the processor 701, the memory 702, and the communication interface 703 are connected through the bus 710 and complete communication with each other.

通信接口703，主要用于实现本申请实施例中各模块、装置、单元和/或设备之间的通信。The communication interface 703 is mainly used to implement communication between modules, devices, units and/or equipment in the embodiments of this application.

总线710包括硬件、软件或两者，将电子设备的部件彼此耦接在一起。举例来说而非限制，总线可包括加速图形端口(AGP)或其他图形总线、增强工业标准架构(EISA)总线、前端总线(FSB)、超传输(HT)互连、工业标准架构(ISA)总线、无限带宽互连、低引脚数(LPC)总线、存储器总线、微信道架构(MCA)总线、***组件互连(PCI)总线、PCI-Express (PCI-X)总线、串行高级技术附件(SATA)总线、视频电子标准协会局部(VLB)总线或其他合适的总线或者两个或更多个以上这些的组合。在合适的情况下，总线710可包括一个或多个总线。尽管本申请实施例描述和示出了特定的总线，但本申请考虑任何合适的总线或互连。 Bus 710 includes hardware, software, or both, coupling the components of the electronic device to each other. By way of example, and not limitation, the bus may include Accelerated Graphics Port (AGP) or other graphics bus, Enhanced Industry Standard Architecture (EISA) bus, Front Side Bus (FSB), HyperTransport (HT) interconnect, Industry Standard Architecture (ISA) Buses, Infinite Bandwidth Interconnect, Low Pin Count (LPC) Bus, Memory Bus, Micro Channel Architecture (MCA) Bus, Peripheral Component Interconnect (PCI) Bus, PCI-Express (PCI-X) Bus, Serial Advanced Technology Attachment (SATA) bus, Video Electronics Standards Association Local (VLB) bus or other suitable bus or a combination of two or more of these. Where appropriate, bus 710 may include one or more buses. Although the embodiments of this application describe and illustrate a specific bus, this application contemplates any suitable bus or interconnection.

该电子设备可以执行本申请实施例中的方法，从而实现结合图1至图4描述的方法。The electronic device can execute the method in the embodiment of the present application, thereby realizing the method described in conjunction with FIGS. 1 to 4 .

另外，结合上述实施例中的方法，本申请实施例可提供一种计算机可读存储介质来实现。该计算机可读存储介质上存储有计算机程序指令；该计算机程序指令被处理器执行时实现图1至图4中的方法。In addition, combined with the methods in the above embodiments, embodiments of the present application can provide a computer-readable storage medium for implementation. Computer program instructions are stored on the computer-readable storage medium; when the computer program instructions are executed by the processor, the methods in Figures 1 to 4 are implemented.

需要明确的是，本申请并不局限于上文所描述并在图中示出的特定配置和处理。为了简明起见，这里省略了对已知方法的详细描述。在上述实施例中，描述和示出了若干具体的步骤作为示例。但是，本申请的方法过程并不限于所描述和示出的具体步骤，本领域的技术人员可以在领会本申请的精神后，作出各种改变、修改和添加，或者改变步骤之间的顺序。To be clear, this application is not limited to the specific configurations and processes described above and illustrated in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications and additions, or change the order between steps after understanding the spirit of the present application.

以上所述的结构框图中所示的功能块可以实现为硬件、软件、固件或者它们的组合。当以硬件方式实现时，其可以例如是电子电路、专用集成电路(ASIC)、适当的固件、插件、功能卡等等。当以软件方式实现时，本申请的元素是被用于执行所需任务的程序或者代码段。程序或者代码段可以存储在机器可读介质中，或者通过载波中携带的数据信号在传输介质或者通信链路上传送。“机器可读介质”可以包括能够存储或传输信息的任何介质。机器可读介质的例子包括电子电路、半导体存储器设备、ROM、闪存、可擦除ROM(EROM)、软盘、CD-ROM、光盘、硬盘、光纤介质、射频(RF)链路，等等。代码段可以经由诸如因特网、内联网等的计算机网络被下载。The functional blocks shown in the above structural block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it may be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, or the like. When implemented in software, elements of the application are programs or code segments that are used to perform the required tasks. The program or code segments may be stored in a machine-readable medium or transmitted over a transmission medium or communications link via a data signal carried in a carrier wave. "Machine-readable medium" may include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, and the like. Code segments may be downloaded via computer networks such as the Internet, intranets, and the like.

还需要说明的是，本申请中提及的示例性实施例，基于一系列的步骤或者装置描述一些方法或***。但是，本申请不局限于上述步骤的顺序，也就是说，可以按照实施例中提及的顺序执行步骤，也可以不同于实施例中的顺序，或者若干步骤同时执行。It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or devices. However, the present application is not limited to the order of the above steps. That is to say, the steps may be performed in the order mentioned in the embodiment, or may be different from the order in the embodiment, or several steps may be performed simultaneously.

以上所述，仅为本申请的具体实施方式，所属领域的技术人员可以清楚地了解到，为了描述的方便和简洁，上述描述的***、模块和单元的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。应理解，本申请的保护范围并不局限于此，任何熟悉本技术领域的技术人员在本申请揭露的技术范围内，可轻易想到各种等效的修改或替换，这些修改或替换都应涵盖在本申请的保护范围之内。The above are only specific implementation modes of the present application. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the above-described systems, modules and units can be referred to the foregoing method embodiments. The corresponding process will not be described again here. It should be understood that the protection scope of the application is not limited to this. Any person familiar with the technical field can easily think of various equivalent modifications or substitutions within the technical scope disclosed in the application, and these modifications or substitutions should be covered. within the protection scope of this application.

Claims

一种预测模型的训练方法，其中，所述方法包括：A method for training a prediction model, wherein the method includes:

获取移动装置在移动过程中采集到的多个样本数据，每个所述样本数据包括样本定位信息、样本速度信息和样本点云信息；Obtain multiple sample data collected by the mobile device during movement, each of the sample data includes sample positioning information, sample speed information and sample point cloud information;

根据样本速度信息对所述样本点云信息进行运动补偿处理，生成样本车速补偿地图；Perform motion compensation processing on the sample point cloud information according to the sample speed information to generate a sample vehicle speed compensation map;

将所述样本车速补偿地图输入至预设模型，输出预测地图；Input the sample vehicle speed compensation map into the preset model and output the prediction map;

根据所述预测地图和样本高精地图计算损失值，所述样本高精地图根据所述样本定位信息生成；Calculate the loss value based on the predicted map and the sample high-precision map, and the sample high-precision map is generated based on the sample positioning information;

根据所述损失值训练所述预设模型，直至所述预设模型满足预设训练条件，得到预测模型。The preset model is trained according to the loss value until the preset model meets the preset training conditions, and a prediction model is obtained.
根据权利要求1所述的方法，其中，在所述根据所述预测地图和样本高精地图计算损失值之前，所述方法还包括：The method according to claim 1, wherein before calculating the loss value based on the prediction map and the sample high-precision map, the method further includes:

根据所述样本定位信息生成初始样本高精地图；Generate an initial sample high-precision map based on the sample positioning information;

根据所述样本定位信息对所述样本点云信息进行运动补偿，生成定位补偿地图；Perform motion compensation on the sample point cloud information according to the sample positioning information, and generate a positioning compensation map;

根据所述定位补偿地图，从所述初始样本高精地图中筛选出预设扫描范围内的所述样本高精地图。According to the positioning compensation map, the sample high-precision map within the preset scanning range is selected from the initial sample high-precision map.
根据权利要求2所述的方法，其中，所述根据所述定位补偿地图，从所述初始样本高精地图中筛选出预设扫描范围内的所述样本高精地图，包括：The method according to claim 2, wherein filtering out the sample high-precision map within a preset scanning range from the initial sample high-precision map according to the positioning compensation map includes:

确定样本点云信息激光雷达采集所述样本点云信息时的所述预设扫描范围；Determine the preset scanning range when the sample point cloud information lidar collects the sample point cloud information;

确定所述预设扫描范围内的所述定位补偿地图；Determine the positioning compensation map within the preset scanning range;

从所述初始样本高精地图中，筛选与所述所述预设扫描范围内的所述定位补偿地图相匹配的所述样本高精地图。Select the sample high-precision map that matches the positioning compensation map within the preset scanning range from the initial sample high-precision map.
根据权利要求1所述的方法，其中，所述根据所述预测地图和样本高精地图计算损失值，包括：The method according to claim 1, wherein calculating the loss value based on the prediction map and the sample high-precision map includes:

从所述样本车速补偿地图中提取第一特征向量，所述第一特征向量用于表征所述移动装置在移动过程中所处环境的道路特征；所述道路特征包括车道线特征，和/或，路沿特征；A first feature vector is extracted from the sample speed compensation map. The first feature vector is used to characterize the road features of the environment in which the mobile device is moving; the road features include lane line features, and/or , curb characteristics;

从所述样本高精地图出提取第二特征向量，所述第二特征向量用于表征所述道路特征；Extract a second feature vector from the sample high-precision map, where the second feature vector is used to characterize the road characteristics;

根据所述第一特征向量和所述第二特征向量计算所述损失值。The loss value is calculated based on the first feature vector and the second feature vector.
根据权利要求1或2所述的方法，其中，所述预设模型包括N个首尾相连的子模型，N为大于1的整数，所述根据所述损失值训练所述预设模型，直至所述预设模型满足预设训练条件，得到预测模型，包括：The method according to claim 1 or 2, wherein the preset model includes N sub-models connected end to end, N is an integer greater than 1, and the preset model is trained according to the loss value until the The above preset model meets the preset training conditions and the prediction model is obtained, including:

将第N-1子模型对应的损失值输入至第N子模型；Input the loss value corresponding to the N-1 sub-model into the N-th sub-model;

根据所述第N-1子模型对应的损失值和预设阈值，训练所述预设模型，直至所述预设模型满足预设训练条件，得到所述预测模型。According to the loss value and the preset threshold corresponding to the N-1th sub-model, the preset model is trained until the preset model meets the preset training conditions, and the prediction model is obtained.
一种地图预测方法，其中，所述方法包括：A map prediction method, wherein the method includes:

获取移动装置采集的运动数据，所述运动数据至少包括：速度信息和点云信息；Obtain motion data collected by the mobile device, where the motion data at least includes: speed information and point cloud information;

根据所述速度信息对所述点云信息进行运动补偿处理，生成车速补偿地图；Perform motion compensation processing on the point cloud information according to the speed information to generate a vehicle speed compensation map;

将所述车速补偿地图输入至预测模型，输出目标地图。The vehicle speed compensation map is input to the prediction model and a target map is output.
根据权利要求6所述的方法，其中，所述将所述车速补偿地图输入至预测模型，输出预测地图，包括：The method according to claim 6, wherein said inputting the vehicle speed compensation map to a prediction model and outputting the prediction map includes:

在预设时间段内，将所述车速补偿地图输入至所述预测模型，得到第一目标地图；所述预测模型中设置有多个与时间标识信息相对应的参数；Within a preset time period, the vehicle speed compensation map is input into the prediction model to obtain a first target map; a plurality of parameters corresponding to the time stamp information are set in the prediction model;

确定所述运动数据被采集时对应的时间标识信息；Determine the time stamp information corresponding to when the motion data is collected;

根据所述时间标识信息对应的参数，对所述第一目标地图进行调整，得到所述目标地图。According to the parameters corresponding to the time stamp information, the first target map is adjusted to obtain the target map.
根据权利要求6所述的方法，其中，所述将所述车速补偿地图输入至预测模型，输出预测地图，包括：The method according to claim 6, wherein said inputting the vehicle speed compensation map to a prediction model and outputting the prediction map includes:

在未检测到第一时间段对应的定位信息的情况下，将所述车速补偿地图输入至所述预测模型，输出所述目标地图；If the positioning information corresponding to the first time period is not detected, input the vehicle speed compensation map to the prediction model and output the target map;

将高精地图，切换显示为所述预测地图，所述高精地图根据第二时间段对应的定位信息确定，所述第二时间段先于所述第一时间段。Switch and display the high-precision map as the predicted map. The high-precision map is determined based on the positioning information corresponding to the second time period, and the second time period precedes the first time period.
一种预测模型的训练装置，其中，所述装置包括：A training device for a prediction model, wherein the device includes:

第一获取模块，用于获取移动装置在移动过程中采集到的多个样本数据，每个所述样本数据包括样本定位信息、样本速度信息和样本点云信息；The first acquisition module is used to acquire multiple sample data collected by the mobile device during movement. Each of the sample data includes sample positioning information, sample speed information and sample point cloud information;

第一补偿模块，用于根据样本速度信息对所述样本点云信息进行运动补偿处理，生成样本车速补偿地图；The first compensation module is used to perform motion compensation processing on the sample point cloud information according to the sample speed information, and generate a sample vehicle speed compensation map;

第一输入模块，用于将所述样本车速补偿地图输入至预设模型，输出预测地图；The first input module is used to input the sample vehicle speed compensation map into the preset model and output the predicted map;

计算模块，用于根据所述预测地图和样本高精地图计算损失值，所述样本高精地图根据所述样本定位信息生成；A calculation module configured to calculate a loss value based on the prediction map and the sample high-precision map, and the sample high-precision map is generated based on the sample positioning information;

训练模块，用于根据所述损失值训练所述预设模型，直至所述预设模型满足预设训练条件，得到预测模型。A training module is used to train the preset model according to the loss value until the preset model meets the preset training conditions to obtain a prediction model.
一种地图预测装置，其中，所述装置包括：A map prediction device, wherein the device includes:

第二获取模块，用于获取移动装置采集的运动数据，所述运动数据至少包括：速度信息和点云信息；The second acquisition module is used to acquire motion data collected by the mobile device, where the motion data at least includes: speed information and point cloud information;

第二补偿模块，用于根据所述速度信息对所述点云信息进行运动补偿处理，生成车速补偿地图；a second compensation module, configured to perform motion compensation processing on the point cloud information according to the speed information and generate a vehicle speed compensation map;

第二输入模块，用于将所述车速补偿地图输入至预测模型，输出目标地图。The second input module is used to input the vehicle speed compensation map into the prediction model and output the target map.
一种电子设备，其中，所述设备包括：处理器以及存储有计算机程序指令的存储器；所述处理器执行所述计算机程序指令时实现如权利要求1-8任意一项所述的方法。An electronic device, wherein the device includes: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the method according to any one of claims 1-8 is implemented.
一种可读存储介质，其中，所述计算机可读存储介质上存储有计算机程序指令，所述计算机程序指令被处理器执行时实现如权利要求1-8任意一项所述的方法。A readable storage medium, wherein computer program instructions are stored on the computer-readable storage medium, and when the computer program instructions are executed by a processor, the method according to any one of claims 1-8 is implemented.
一种计算机程序，包括计算机可读代码，当所述计算机可读代码在计算处理设备上运行时，导致所述计算处理设备执行根据权利要求1-8中任一项所述的方法。A computer program comprising computer readable code which, when run on a computing processing device, causes the computing processing device to perform a method according to any one of claims 1-8.