A noise determination method, apparatus, computer device, and storage medium

Abstract

This disclosure provides a noise determination method, apparatus, computer device, and storage medium. The method includes: acquiring measurement data collected by a preset device and a pre-constructed decision tree model; the decision tree model includes multiple internal nodes, optimal segmentation information corresponding to each internal node, multiple leaf nodes, and noise parameter information corresponding to each leaf node; determining the target leaf node corresponding to the measurement data based on the optimal segmentation information corresponding to each internal node; and determining the target noise information of the measurement data based on the noise parameter information corresponding to the target leaf node.

Description

Technical Field

[0001] This disclosure relates to the field of multi-sensor fusion technology, and more specifically, to a noise determination method, apparatus, computer device, and storage medium. Background Technology

[0002] Multi-sensor fusion is an essential component for robust perception in complex scenarios for autonomous vehicles. However, sensor measurement data often contains noise, which can lead to anomalies. Typically, noise can be used to identify these anomalies. Therefore, accurately estimating noise can pinpoint these anomalies, thereby improving the results of multi-sensor fusion.

[0003] Existing noise estimation methods are mainly divided into two categories: online noise estimation and offline noise estimation. Online noise estimation can utilize the adaptive Kalman filter algorithm to estimate the noise of measurement data online. Online estimation of the noise of the current measurement data requires historical measurement data; for example, the noise of the current measurement data might be the average of historical measurement error data. However, the noise calculated for the current measurement data in this way has a time delay and is not accurate enough. Therefore, it cannot accurately detect abnormal data in the measurement data, thus reducing the effectiveness of sensor fusion.

[0004] Offline noise estimation can store sample measurement data and the corresponding measurement error data offline in a k-dimensional tree (kd-tree). When the algorithm runs online, for each sensor's measurement data, it searches for the k nearest sample measurement data and the corresponding sample error data in the kd-tree, and uses the sample error data to estimate the noise. However, such algorithms often require a large amount of memory space to store a large amount of offline sample measurement data and the corresponding measurement error data, which limits the algorithm's running speed and reduces the efficiency of noise estimation. Summary of the Invention

[0005] This disclosure provides at least one noise determination method, apparatus, computer device, and storage medium.

[0006] In a first aspect, embodiments of this disclosure provide a noise determination method, including:

[0007] Acquire measurement data collected by a preset device, and a pre-constructed decision tree model corresponding to the preset device; the decision tree model includes multiple internal nodes, optimal splitting information corresponding to each internal node, multiple leaf nodes, and noise parameter information corresponding to each leaf node;

[0008] Based on the optimal segmentation information corresponding to each internal node, the target leaf node corresponding to the measurement data is determined;

[0009] Based on the noise parameter information corresponding to the target leaf node, the target noise information of the measurement data is determined.

[0010] In this aspect, a decision tree model is pre-constructed. The leaf nodes in this model represent noise parameter information of the measurement data collected by the preset device. Once constructed, it can repeatedly estimate the noise of all data collected by the preset device. The decision tree model only stores the optimal split information and noise parameter information, resulting in a smaller data volume. This improves the algorithm's running efficiency when the decision tree model stores less data. In addition, the noise parameter information stored in the decision tree model is based on sample measurement data from the preset device, which can comprehensively characterize the noise information of the measurement data. In actual noise estimation, there is no need to use historical measurement data, thus improving the efficiency of noise estimation.

[0011] In one optional implementation, the step of constructing the decision tree model includes:

[0012] Obtain the sample dataset collected by the preset device, and based on the sample dataset, determine multiple internal nodes, the optimal segmentation information corresponding to each internal node, and multiple leaf nodes;

[0013] Determine the noise parameter information corresponding to each leaf node;

[0014] The decision tree model is constructed based on the determined multiple internal nodes, multiple leaf nodes, the optimal splitting information corresponding to each internal node, and the noise parameter information corresponding to each leaf node.

[0015] In this implementation, the optimal segmentation information corresponding to the internal nodes can indicate the noise distribution of the measurement data (which cannot accurately represent the actual noise of the measurement data), and the noise parameter information corresponding to the leaf nodes can indicate a more accurate noise distribution of the measurement data. By using multiple internal nodes, multiple leaf nodes, the optimal segmentation information corresponding to each internal node, and the noise parameter information corresponding to each leaf node, a decision tree model with low time complexity can be constructed, which can improve the efficiency of noise estimation in subsequent applications.

[0016] In one optional implementation, determining multiple internal nodes, the optimal splitting information corresponding to each internal node, and multiple leaf nodes based on the sample dataset includes:

[0017] The following steps are performed iteratively to determine multiple internal nodes, the optimal splitting information corresponding to each internal node, and multiple leaf nodes, wherein each iteration generates either one internal node or one leaf node:

[0018] Based on the sample dataset, multiple preset segmentation information is determined;

[0019] For each preset segmentation information, the sample dataset is divided into multiple data subsets based on the preset segmentation information, and noise sub-information corresponding to each data subset is determined respectively;

[0020] Based on the noise sub-information corresponding to each data subset, the noise distribution information of the sample dataset corresponding to the preset segmentation information is determined; wherein, the noise distribution information is used to characterize the noise similarity of the sample dataset corresponding to the preset segmentation information;

[0021] Based on the noise distribution information corresponding to each preset segmentation information, an internal node or a leaf node is generated, and in the case of generating the internal node, the optimal segmentation information corresponding to the internal node is determined.

[0022] This implementation method utilizes preset segmentation information to divide the sample dataset into multiple data subsets. The noise distribution of the measurement data of each sample in the same data subset is similar. By judging whether the noise information (such as noise parameters, i.e., noise mean or noise variance, or noise distribution) corresponding to multiple data subsets is similar, the internal nodes and leaf nodes can be accurately generated using the judgment result, i.e., noise similarity.

[0023] In one optional implementation, the sample dataset includes multiple sample measurement data and measurement error data corresponding to each sample measurement data.

[0024] Determining the noise sub-information corresponding to the data subset includes:

[0025] Based on the measurement error data corresponding to each sample measurement data in the data subset, the noise sub-information corresponding to the data subset is determined.

[0026] This implementation method can accurately determine the noise sub-information of the sample measurement data in the data subset by utilizing the measurement error data corresponding to each sample measurement data.

[0027] In one optional implementation, an internal node is generated based on the noise distribution information corresponding to each preset segmentation information, and the optimal segmentation information corresponding to the internal node is determined, including:

[0028] If the noise distribution information corresponding to all preset segmentation information does not meet the preset conditions, an internal node is generated, and the preset segmentation information with the highest noise similarity indicated by the noise distribution information is taken as the optimal segmentation information corresponding to the internal node.

[0029] In this implementation, the noise distribution information can be used to accurately generate nodes of the decision tree model. If the noise distribution information corresponding to the preset segmentation information does not meet the preset conditions, it means that the noise distribution corresponding to the current node is not concentrated enough and cannot be used as a leaf node. Therefore, the preset segmentation information with the highest noise similarity indicated by the noise distribution information can be directly used as the optimal segmentation information corresponding to the internal node, that is, the internal node is generated.

[0030] In one optional implementation, a leaf node is generated based on the noise distribution information corresponding to each preset segmentation information, including:

[0031] If the noise distribution information corresponding to at least one preset segmentation information meets the preset conditions, a leaf node is generated.

[0032] In this implementation, noise distribution information can be used to accurately generate nodes of the decision tree model. If the noise distribution information corresponding to at least one preset segmentation information meets the preset conditions, it means that the noise of the sample measurement data in the multiple data subsets segmented by the preset segmentation information follows the same distribution, that is, the noise is similar. Therefore, leaf nodes can be directly generated, and the leaf nodes include noise parameter information corresponding to multiple data subsets.

[0033] In one optional implementation, it further includes:

[0034] In the Nth iteration, if the noise distribution information corresponding to all preset segmentation information in the current iteration does not meet the preset conditions, the sample dataset is divided into multiple sample data subsets based on the optimal segmentation information corresponding to the internal nodes generated in the current iteration, and each sample data subset is used as the sample dataset in the N+1th iteration; where N is a positive integer.

[0035] In this implementation, the optimal segmentation information can divide the sample dataset into multiple sample data subsets. Based on this segmentation method, a decision tree model that accurately finds the noise parameter information corresponding to the measurement data can be constructed.

[0036] In one optional implementation, the sample dataset includes multiple sample measurement data, and the sample measurement data includes feature values ​​of multiple dimensions; the preset segmentation information includes preset feature thresholds and preset dimensions;

[0037] The step of dividing the sample dataset into multiple data subsets based on the preset segmentation information includes:

[0038] The sample measurement data in the sample dataset whose feature values ​​of a preset dimension are less than or equal to a preset feature threshold are taken as target measurement data, and a data subset is generated based on the selected target measurement data;

[0039] Based on the sample measurement data in the sample dataset other than the target measurement data, another data subset is generated.

[0040] This implementation method, by using preset feature thresholds and preset dimensions, can accurately filter out target measurement data belonging to the same noise distribution from the sample dataset, determine the data subset, and set another data subset of the remaining sample measurement data after filtering.

[0041] In one optional implementation, the noise parameter information includes the noise mean and noise variance; the sample dataset includes multiple sample measurement data and measurement error data corresponding to each sample measurement data.

[0042] The step of determining the noise parameter information corresponding to each leaf node includes:

[0043] For each leaf node, the noise mean is determined based on the measurement error data in the sample dataset corresponding to the leaf node;

[0044] Based on the noise mean and various measurement error data, the noise variance is determined.

[0045] In this implementation, since the measurement error data is the error between the sample measurement data and its associated measurement data (such as data collected by the same preset device over a period of time), the noise mean of the sample measurement data belonging to the same noise distribution can be accurately determined by using the measurement error data in the sample dataset corresponding to the leaf node; the noise variance is determined by using the noise mean and the measurement error data, and this noise variance can characterize the noise variance corresponding to the sample measurement data belonging to the same noise distribution.

[0046] In an optional implementation, after determining the target noise information of the measurement data based on the noise parameter information corresponding to the target leaf node, the method further includes:

[0047] Determine the degree of deviation between the measured data and the noise indicated by the target noise information;

[0048] Obtain a preset probability threshold for the measured data to belong to abnormal data;

[0049] Based on the preset probability threshold, a deviation threshold is determined;

[0050] If the deviation is greater than or equal to the deviation threshold, the measured data is determined to be abnormal data.

[0051] In this implementation, since the degree of deviation follows a regular distribution, a preset probability threshold for the measured data to be considered abnormal can be used to find the indicator used to judge whether the measured data is abnormal from the regular distribution table, namely the deviation threshold. Based on the deviation threshold and the degree of deviation, it is possible to accurately determine whether the measured data is abnormal; that is, if the degree of deviation is greater than or equal to the deviation threshold, the measured data is determined to be abnormal; if the degree of deviation is less than or equal to the deviation threshold, the measured data is determined not to be abnormal.

[0052] In an optional implementation, after determining the target noise information of the measurement data, the method further includes:

[0053] For measurement data collected by multiple sensor devices in the vehicle, the target noise information of each measurement data is determined;

[0054] Based on the target noise information of each measurement data, the fusion weight of each measurement data is determined;

[0055] Based on the fusion weight of each measurement data, multiple measurement data are fused to determine measurement fusion data, so that the vehicle can determine operating environment information based on the measurement fusion data.

[0056] This implementation method utilizes more accurate target noise information to improve the accuracy of the fusion weights of measurement data during multi-sensor fusion, thereby improving the accuracy of the measurement fusion data and ensuring that the vehicle can accurately predict the operating environment information of the vehicle's operating environment based on accurate measurement fusion data.

[0057] Secondly, embodiments of this disclosure also provide a noise determination apparatus, comprising:

[0058] The information acquisition module is used to acquire measurement data collected by a preset device and a pre-constructed decision tree model corresponding to the preset device; the decision tree model includes multiple internal nodes, optimal splitting information corresponding to each internal node, multiple leaf nodes, and noise parameter information corresponding to each leaf node.

[0059] The node determination module is used to determine the target leaf node corresponding to the measurement data based on the optimal splitting information corresponding to each internal node.

[0060] The noise estimation module is used to determine the target noise information of the measurement data based on the noise parameter information corresponding to the target leaf node.

[0061] In one optional implementation, the device further includes a model building module, used to acquire a sample dataset collected by the preset device, and based on the sample dataset, determine multiple internal nodes, optimal splitting information corresponding to each internal node, and multiple leaf nodes.

[0062] Determine the noise parameter information corresponding to each leaf node;

[0063] The decision tree model is constructed based on the determined multiple internal nodes, multiple leaf nodes, the optimal splitting information corresponding to each internal node, and the noise parameter information corresponding to each leaf node.

[0064] In one optional implementation, the model building module is used to iteratively perform the following steps:

[0065] Based on the sample dataset, multiple preset segmentation information is determined;

[0066] For each preset segmentation information, the sample dataset is divided into multiple data subsets based on the preset segmentation information, and noise sub-information corresponding to each data subset is determined respectively;

[0067] Based on the noise sub-information corresponding to each data subset, the noise distribution information of the sample dataset corresponding to the preset segmentation information is determined; wherein, the noise distribution information is used to characterize the noise similarity of the sample dataset corresponding to the preset segmentation information;

[0068] Based on the noise distribution information corresponding to each preset segmentation information, an internal node or a leaf node is generated, and in the case of generating the internal node, the optimal segmentation information corresponding to the internal node is determined.

[0069] In one optional implementation, the sample dataset includes multiple sample measurement data and measurement error data corresponding to each sample measurement data.

[0070] The model building module is used to determine the noise sub-information corresponding to the data subset based on the measurement error data corresponding to the measurement data of each sample in the data subset.

[0071] In one optional implementation, the model building module is used to generate an internal node when the noise distribution information corresponding to all preset segmentation information does not meet the preset conditions, and to take the preset segmentation information with the highest similarity indicated by the noise distribution information as the optimal segmentation information corresponding to the internal node.

[0072] In one optional implementation, the model building module is used to generate a leaf node when the noise distribution information corresponding to at least one preset segmentation information meets preset conditions.

[0073] In an optional implementation, the model building module is further configured to, during the Nth iteration, if the noise distribution information corresponding to all preset segmentation information in the current iteration does not meet the preset conditions, divide the sample dataset into multiple sample data subsets based on the optimal segmentation information corresponding to the internal nodes generated in the current iteration, and use each sample data subset as the sample dataset used in the N+1th iteration; where N is a positive integer.

[0074] In one optional implementation, the sample dataset includes multiple sample measurement data, and the sample measurement data includes feature values ​​of multiple dimensions; the preset segmentation information includes preset feature thresholds and preset dimensions;

[0075] The model building module is used to take the sample measurement data in the sample dataset whose feature values ​​of a preset dimension are less than or equal to a preset feature threshold as target measurement data, and generate a data subset based on the selected target measurement data.

[0076] Based on the sample measurement data in the sample dataset other than the target measurement data, another data subset is generated.

[0077] In one optional implementation, the noise parameter information includes the noise mean and noise variance; the sample dataset includes multiple sample measurement data and measurement error data corresponding to each sample measurement data.

[0078] The model building module is used to determine the noise mean for each leaf node based on the measurement error data in the sample dataset corresponding to the leaf node.

[0079] Based on the noise mean and various measurement error data, the noise variance is determined.

[0080] In an optional embodiment, the device further includes an abnormal data determination module, which is used to determine the degree of deviation between the measured data and the noise indicated by the target noise information after determining the target noise information of the measured data based on the noise parameter information corresponding to the target leaf node;

[0081] Obtain a preset probability threshold for the measured data to belong to abnormal data;

[0082] Based on the preset probability threshold, a deviation threshold is determined;

[0083] If the deviation is greater than or equal to the deviation threshold, the measured data is determined to be abnormal data.

[0084] In one optional embodiment, the device further includes a data fusion module, which, after determining the target noise information of the measurement data, determines the target noise information of each measurement data collected by multiple sensor devices in the vehicle.

[0085] Based on the target noise information of each measurement data, the fusion weight of each measurement data is determined;

[0086] Based on the fusion weight of each measurement data, multiple measurement data are fused to determine measurement fusion data, so that the vehicle can determine operating environment information based on the measurement fusion data.

[0087] Thirdly, embodiments of this disclosure also provide a computer device, including: a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the computer device is running, the processor communicates with the memory via the bus, and when the machine-readable instructions are executed by the processor, the steps of the first aspect above, or any possible noise determination method in the first aspect, are performed.

[0088] Fourthly, embodiments of this disclosure also provide a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of the first aspect or any possible noise determination method described in the first aspect.

[0089] For a description of the effects of the noise determination device, computer equipment, and storage medium, please refer to the description of the noise determination method above; it will not be repeated here.

[0090] To make the above-mentioned objects, features and advantages of this disclosure more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0091] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings used in the embodiments will be briefly described below. These drawings are incorporated in and constitute a part of this specification. They illustrate embodiments conforming to this disclosure and, together with the specification, serve to explain the technical solutions of this disclosure. It should be understood that the following drawings only show some embodiments of this disclosure and should not be considered as limiting the scope. Those skilled in the art can obtain other related drawings based on these drawings without creative effort.

[0092] Figure 1 A flowchart of a noise determination method provided by an embodiment of this disclosure is shown;

[0093] Figure 2 This illustration shows a schematic diagram of the specific structure of the decision tree provided in an embodiment of the present disclosure;

[0094] Figure 3 This diagram illustrates the specific structure of the decision tree model provided in this embodiment of the disclosure.

[0095] Figure 4 A schematic diagram of a noise determination apparatus provided in an embodiment of this disclosure is shown;

[0096] Figure 5 A schematic diagram of the structure of a computer device provided in an embodiment of this disclosure is shown. Detailed Implementation

[0097] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. The components of the embodiments of this disclosure described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this disclosure provided in the accompanying drawings is not intended to limit the scope of the claimed disclosure, but merely represents selected embodiments of this disclosure. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without inventive effort are within the scope of protection of this disclosure.

[0098] Furthermore, the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein.

[0099] In this article, "multiple or several" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0100] Research has revealed that existing noise estimation methods mainly fall into two categories: online noise estimation and offline noise estimation. Online noise estimation utilizes an adaptive Kalman filter algorithm to estimate the noise of measurement data online. However, online noise estimation of the current measurement data requires historical measurement data; for example, the noise of the current measurement data might be the average of historical measurement error data. However, this calculated noise of the current measurement data is time-delayed and inaccurate, thus failing to accurately detect abnormal data in the measurement data and reducing the effectiveness of sensor fusion.

[0101] Offline noise estimation can store sample measurement data and the corresponding measurement error data offline in a k-dimensional tree (kd-tree). When the algorithm runs online, for each sensor's measurement data, it searches for the k nearest sample measurement data and the corresponding sample error data in the kd-tree, and uses the sample error data to estimate the noise. However, such algorithms often require a large amount of memory space to store a large amount of offline sample measurement data and the corresponding measurement error data, which limits the algorithm's running speed and reduces the efficiency of noise estimation.

[0102] Based on the above research, this disclosure provides a noise determination method. A decision tree model is pre-constructed, where leaf nodes represent noise parameter information of measurement data collected by a pre-defined device. Therefore, it can be built once and used repeatedly to estimate noise. The decision tree model only stores optimal segmentation information and noise parameter information. Since the amount of noise parameter information is relatively small compared to the measurement data (e.g., multiple measurement data points correspond to one noise parameter), the algorithm's efficiency can be improved by storing less data in the decision tree model. Furthermore, the optimal segmentation information allows for quick retrieval of the noise parameter information corresponding to the measurement data, thus estimating the noise and improving noise estimation efficiency. Additionally, the noise parameter information stored in the decision tree model is based on sample measurement data and provides a relatively comprehensive characterization of the noise in the measurement data, eliminating the need for historical measurement data and ensuring zero delay in noise estimation.

[0103] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0104] To facilitate understanding of this embodiment, a noise determination method disclosed in this disclosure will first be described in detail. The execution subject of the noise determination method provided in this disclosure is generally a computer device with a certain computing power. In some possible implementations, the noise determination method can be implemented by a processor calling computer-readable instructions stored in memory.

[0105] The noise determination method provided in this disclosure is illustrated below using a computer device as an example.

[0106] See Figure 1 The diagram shows a flowchart of a noise determination method provided in an embodiment of this disclosure. The method includes steps S101 to S103, wherein:

[0107] S101: Acquire measurement data collected by a preset device and a pre-built decision tree model corresponding to the preset device; the decision tree model includes multiple internal nodes, optimal splitting information corresponding to each internal node, multiple leaf nodes, and noise parameter information corresponding to each leaf node.

[0108] In this step, the preset equipment may include sensors capable of collecting information about the surrounding environment, or devices containing information collection programs, such as lidar, cameras, ultrasonic sensors, odometers, inertial measurement units (IMUs), and global positioning systems (GPS).

[0109] For autonomous driving applications, measurement data can include environmental images collected by sensors, navigation location, vehicle mileage, etc.

[0110] The decision tree model corresponding to the preset device can be constructed using the sample dataset collected by the preset device. Different preset devices correspond to different decision tree models. The decision tree model, for example, can be a decision tree algorithm model, which is a tree structure including multiple internal nodes and multiple leaf nodes. In this embodiment, the pre-constructed decision tree model includes multiple internal nodes and multiple leaf nodes at different levels, with leaf nodes being the final-level nodes. Internal nodes store optimal splitting information, and leaf nodes store noise parameter information. By utilizing the optimal splitting information in multiple internal nodes at different levels, the noise distribution of the measurement data (which cannot accurately represent the actual noise of the measurement data) can be continuously indicated until the noise parameter information corresponding to the leaf node is determined, that is, the noise distribution of the measurement data indicated by the noise parameter information is determined to be more accurate (which accurately represents the actual noise of the measurement data).

[0111] Here, multiple data collected by the same preset device can conform to the noise distribution determined by the same noise parameter information. Therefore, multiple measurement data can correspond to one noise parameter information.

[0112] For example, the optimal segmentation information may include a preset feature threshold and a preset dimension; wherein, the preset dimension is used to indicate the feature value corresponding to the preset dimension. Based on the preset feature threshold and preset dimension in the current internal node, the noise distribution to which the measurement data noise belongs can be determined, that is, the next internal node of the current internal node in the decision tree model can be determined; or, the next level leaf node of the current internal node.

[0113] For example, noise parameter information may include noise mean and noise variance, or noise distribution, etc. If the noise parameter information includes noise mean and noise variance, the noise distribution can be determined using the noise mean and noise variance, and the noise distribution can characterize the noise of the data collected by the preset device. If the noise parameter information includes noise distribution, the noise distribution can be directly used to characterize the noise. Here, noise distribution can be one way of representing noise.

[0114] S102: Based on the optimal splitting information corresponding to each internal node, determine the target leaf node corresponding to the measurement data.

[0115] See Figure 2 The diagram shows the specific structure of a decision tree, including multiple internal nodes 21 and multiple leaf nodes 22. Each internal node 21 includes a root node 211. For a specific internal node 212, the associated internal nodes include its parent node 212-1, its child nodes 212-2 and 212-3. Here, all internal nodes except the root node contain both parent and child nodes. The root node has only child nodes.

[0116] Based on the optimal segmentation information corresponding to each internal node, starting from the root node, the internal nodes corresponding to the measurement data can be determined sequentially until the target leaf node is determined.

[0117] For example, measurement data can be input into a decision tree model. Starting from the root node, based on the optimal splitting information corresponding to the root node, the child node A of the root node corresponding to the measurement data is determined. Here, the child node A of the root node is also an internal node A. Based on the optimal splitting information corresponding to the internal node A, the child node of the internal node A corresponding to the measurement data is determined. The decision tree model is traversed downwards sequentially until the target leaf node B corresponding to the measurement data is found.

[0118] S103: Determine the target noise information of the measurement data based on the noise parameter information corresponding to the target leaf node.

[0119] The target noise information may include noise parameter information, or noise determined based on the noise parameter information. For example, if the noise parameter information includes the noise mean and noise variance, the target noise information may be the noise distribution determined by the noise mean and noise variance, whereby the noise distribution characterizes the noise. If the noise parameter information includes a noise distribution, the target noise information may be the noise parameter information itself, i.e., the noise distribution.

[0120] For example, the noise distribution can be a Gaussian distribution that the noise of the measurement data detected by the sensors in an autonomous driving scenario follows (the noise parameters include the noise mean and noise variance determined by the measurement data).

[0121] The above steps S101-S103 pre-construct a decision tree model. The leaf nodes in this model represent noise parameter information of the measurement data collected by the prediction device. Therefore, it can be built once and used repeatedly to estimate noise. The decision tree model only stores optimal splitting information and noise parameter information. Since the amount of noise parameter information is relatively small compared to the measurement data (e.g., multiple measurement data points correspond to one noise parameter), the algorithm's running efficiency can be improved by storing less data in the decision tree model. Furthermore, the optimal splitting information allows for quick retrieval of the noise parameter information corresponding to the measurement data, thus estimating the noise of the measurement data and improving noise estimation efficiency. Additionally, the noise parameter information stored in the decision tree model is based on sample measurement data and provides a relatively comprehensive characterization of the noise in the measurement data, eliminating the need for historical measurement data and ensuring zero delay in noise estimation.

[0122] The present disclosure embodiments can construct a decision tree model according to the following steps:

[0123] S1011: Obtain the sample dataset collected by the preset device, and based on the sample dataset, determine multiple internal nodes, the optimal segmentation information corresponding to each internal node, and multiple leaf nodes.

[0124] In this step, the sample dataset may include sample measurement data collected by a preset device, as well as measurement error data corresponding to the sample measurement data. To determine the measurement error data corresponding to the sample measurement data, specifically, multiple historical sample measurement data points from a continuous time period prior to the sample measurement data x collected by the preset device are obtained, denoted as x′. i Where i is a positive integer, i is less than M, and M represents the number of historical sample measurement data in the historical period; using Formula 1, the measurement error data y corresponding to the sample measurement data x is determined.

[0125]

[0126] Sample dataset D = {(x1,y1),(x2,y2),……,(x i ,y i ),……,(x N ,y N )}. Where, x i Let y represent the measurement data of the i-th sample. i This represents the measurement error data corresponding to the i-th sample measurement data; N represents that there are N sample measurement data in the sample dataset, and N is a positive integer.

[0127] Based on the sample measurement data and the corresponding measurement error data in the sample dataset, the root node and the optimal segmentation information corresponding to the root node can be determined. Then, according to the optimal segmentation information of the root node, the sample dataset is divided into multiple data subsets. Using the sample measurement data and the corresponding measurement error data in the multiple data subsets, the child nodes of the root node and the optimal segmentation information corresponding to each child node of the root node are determined. The child nodes of each internal node are determined in turn until the leaf nodes are obtained.

[0128] S1012: Determine the noise parameter information corresponding to each leaf node.

[0129] Optionally, if a leaf node is determined, its parent node is determined, and based on the parent node, a subset of sample data corresponding to the leaf node is determined; and based on the subset of sample data corresponding to the leaf node, noise parameter information is determined.

[0130] For example, the sample data subset corresponding to the leaf node may include multiple sample measurement data indicated by the optimal splitting information of the leaf node's parent node. Noise parameter information is determined based on the sample measurement data in the sample data subset and the measurement error data corresponding to the sample measurement data.

[0131] Noise parameter information may include noise mean and noise variance. Optionally, the noise mean may be determined based on the measurement error data in the sample dataset corresponding to the leaf node; and the noise variance may be determined based on the noise mean and the measurement error data.

[0132] For example, given the subset of sample data corresponding to the parent node of a leaf node, denoted as D1 = {(x1,y1),(x2,y2),……,(x K ,y K)}, where K represents the number of sample measurement data in the sample data subset, K is a positive integer, and K is less than or equal to N. Noise parameter information includes the noise mean μ and the noise variance ε. The noise mean μ is determined using Formula 2; the noise variance ε is determined using Formula 3.

[0133]

[0134]

[0135] Where T represents transpose, (y i -μ) T This indicates that for matrix (y) i -μ) is transposed.

[0136] S1013: Construct a decision tree model based on the determined number of internal nodes, multiple leaf nodes, the optimal splitting information corresponding to each internal node, and the noise parameter information corresponding to each leaf node.

[0137] In the process of determining internal nodes, the parent node and child node of each internal node are marked. The parent node is the internal node above the child node. According to the hierarchical relationship, a tree structure model, namely the decision tree model, can be constructed.

[0138] For example, after generating the decision tree model, let the number of nodes be Q. In practical applications, the time complexity of finding noise parameter information in the decision tree model for a given measurement data is T = log2Q.

[0139] The above steps S1011 to S1013 utilize multiple internal nodes, multiple leaf nodes, the optimal splitting information corresponding to each internal node, and the noise parameter information corresponding to each leaf node to construct a decision tree model with low time complexity, which can improve the efficiency of noise estimation in subsequent applications.

[0140] In some embodiments, the following steps are performed iteratively to determine multiple internal nodes, optimal segmentation information corresponding to each internal node, and multiple leaf nodes. Each iteration generates either an internal node or a leaf node. Specifically, multiple preset segmentation information can be determined based on the sample dataset. For each preset segmentation information, the sample dataset is divided into multiple data subsets, and noise sub-information corresponding to each data subset is determined. Based on the noise sub-information corresponding to each data subset, noise distribution information of the sample dataset corresponding to the preset segmentation information is determined. Based on the noise distribution information corresponding to each preset segmentation information, an internal node or a leaf node is generated, and if an internal node is generated, the optimal segmentation information corresponding to the internal node is determined.

[0141] Here, noise distribution information is used to characterize the noise similarity of the sample dataset to the preset segmentation information. Specifically, it characterizes whether the noise distributions of different data subsets after the same sample dataset is segmented are similar. The preset segmentation information can be used to segment the sample dataset; or it can be used to segment the sample data subsets after the sample dataset is segmented.

[0142] By utilizing pre-defined segmentation information, a sample dataset can be divided into multiple data subsets. For example, assuming the noise distribution of the measurement data from different samples within the same subset is similar, the noise information (e.g., noise parameters such as mean or variance, or noise distribution) of the multiple subsets obtained through the pre-defined segmentation information is assessed for similarity. If the assessment indicates dissimilar noise distributions, an internal node is generated, and the sample dataset is further segmented, repeating the assessment process. If the assessment indicates similar noise distributions, it means the noise of the measurement data from the samples within the multiple subsets segmented by the pre-defined segmentation information follows the same distribution, and a leaf node can be directly generated. This method of segmenting the sample dataset using pre-defined segmentation information can generate internal and leaf nodes relatively accurately.

[0143] In some embodiments, the preset segmentation information may include a preset feature threshold and a preset dimension, or a segmentation point. The segmentation point can represent the segmentation location of the sample measurement data. For example, a segmentation point of 3 can divide the third sample measurement data (or the measurement error data of the third sample measurement data) in the sample dataset into a data subset consisting of the first three sample measurement data and the measurement error data of the first three sample measurement data (including the third sample measurement data and the measurement error data of the third sample measurement data), and a data subset consisting of the remaining sample measurement data and the measurement error data of the sample measurement data.

[0144] A preset segmentation information may include multiple segmentation points. For example, when the segmentation points are 25 and 50, the 25th sample measurement data in the sample dataset is segmented into a data subset consisting of the first 25 sample measurement data and the measurement error data of the first 25 sample measurement data (including the 25th sample measurement data and the measurement error data of the 25th sample measurement data), the 26th to 50th sample measurement data and the measurement error data of the 26th to 50th sample measurement data and the measurement error data of the 26th to 50th sample measurement data and the measurement error data of the 50th sample measurement data and the remaining sample measurement data and the measurement error data of the ...

[0145] In other embodiments, for any one of the preset feature thresholds s within its range, and for any one of the preset dimensions j within its range, a preset segmentation information can be randomly combined to determine. Here, for the sample dataset, n×N preset segmentation information can be determined.

[0146] The sample dataset is segmented using a preset feature threshold s and a preset dimension j. Specifically, sample measurement data in the sample dataset whose feature values ​​of the preset dimension are less than or equal to the preset feature threshold can be used as target measurement data, and a data subset is generated based on the selected target measurement data; another data subset is generated based on the sample measurement data in the sample dataset other than the target measurement data.

[0147] Continuing the previous example, given a sample dataset D = {(x1,y1),(x2,y2),(x3,y3)}, determine a preset feature threshold s and a preset dimension j; the feature values ​​x of the preset dimension are... j ,include and if If the value is less than or equal to the preset feature threshold s, then the data subset R1 is determined to include x1; if... If the value is less than or equal to the preset feature threshold s, then x2 is determined to be included in the data subset R1; if... If the value is greater than the preset feature threshold s, then the data subset R2 is determined to include x3.

[0148] The preset feature threshold s and preset dimension j can be determined based on sample measurement data in the sample dataset. For example, the sample measurement data... in, Represents sample measurement data x i The nth feature value in the dataset. n represents the n-dimensional dimension, and n is a positive integer. The preset feature threshold s ranges between the minimum and maximum feature values ​​in all sample measurement data in the sample dataset. For example, if the sample dataset includes D = {(x1,y1),(x2,y2),(x3,y3)}, then... from Find the largest eigenvalue and minimum eigenvalue The range of the preset feature threshold s is: include and The preset value range of dimension j is between 1 and n, excluding 1 and n.

[0149] When multiple data subsets are obtained through segmentation, for each data subset, noise sub-information can be determined based on the measurement error data corresponding to each measurement data point within the subset. This noise sub-information can include the noise parameter μ and the noise variance ε; or, the noise distribution.

[0150] The noise sub-information corresponding to the data subset can be determined according to Formula 2 and Formula 3. Specifically, the segmented data subset is determined, the measurement error data of the sample measurement data in the data subset is obtained, and the noise parameter μ and noise variance ε corresponding to the data subset are calculated according to Formula 2 and Formula 3, thereby determining the noise sub-information.

[0151] The noise distribution information corresponding to the preset segmentation information of the sample dataset is determined. Specifically, the noise distribution corresponding to each data subset can be determined based on the noise sub-information corresponding to each data subset. Then, the noise distribution information is determined according to the noise distribution corresponding to each data subset, and L(j,s) is used to represent the noise distribution information. Here, the more similar the two noise distributions are, the smaller the value of L(j,s) and the higher the noise similarity.

[0152] Continuing the previous example, given data subsets R1 and R2, we determine the noise sub-information μ1(j,s) and ε1(j,s) corresponding to data subset R1, and the noise sub-information μ2(j,s) and ε2(j,s) corresponding to data subset R2, according to formulas 2 and 3. Using μ1(j,s), ε1(j,s), μ2(j,s), and ε2(j,s), we determine the noise distribution information L(j,s) between the noise distributions corresponding to each data subset according to formula 4.

[0153]

[0154] Where argmin represents iterating through the traversal to find the minimum value, that is, calculating the minimum. It can represent the noise distribution corresponding to the data subset R1; L(j, s) can represent the noise distribution corresponding to the data subset R2. The smaller L(j, s) is, the more similar the noise distribution corresponding to the data subset R1 is to the noise distribution corresponding to the data subset R2, that is, the higher the noise similarity.

[0155] For a sample dataset including D = {(x1, y1), (x2, y2), (x3, y3)}, n×3 preset segmentation information can be determined. Following the above method, the noise distribution information L(j, s) corresponding to each preset segmentation information can be determined.

[0156] Based on the noise distribution information L(j, s) corresponding to each preset segmentation information, an internal node is generated, and the optimal segmentation information corresponding to the internal node is determined, or a leaf node is generated. Specifically, if the noise distribution information corresponding to all preset segmentation information does not meet the preset conditions, an internal node is generated, and the preset segmentation information with the highest noise similarity indicated by the noise distribution information is taken as the optimal segmentation information corresponding to the internal node. That is, the preset segmentation information corresponding to the minimum value of L(j, s) indicated by the noise distribution information is taken as the optimal segmentation information corresponding to the internal node.

[0157] The preset conditions may include L(j,s) indicated by the noise distribution information being less than or equal to a preset threshold. Here, the preset threshold can be set according to the actual application scenario and empirical values, and this embodiment does not impose specific limitations.

[0158] Continuing with the previous example, if the noise distribution information indicating L(j,s) of each of the n×3 preset segmentation information is greater than the preset threshold, then the preset segmentation information corresponding to the smallest L(j,s) among the n×3 noise distribution information can be used as the optimal segmentation information corresponding to the internal node.

[0159] Alternatively, if the noise distribution information corresponding to at least one preset segmentation information meets the preset conditions, a leaf node is generated.

[0160] Nodes in a decision tree model can be accurately generated using noise distribution information. If the noise distribution information corresponding to at least one preset segmentation information meets preset conditions, it indicates that the noise of the sample measurement data in the multiple data subsets segmented by the preset segmentation information follows the same distribution, i.e., the noise is similar. Therefore, leaf nodes can be directly generated, and these leaf nodes include noise parameter information corresponding to multiple data subsets.

[0161] Continuing the previous example, if at least one of the n×3 preset segmentation information has a noise distribution information L(j,s) less than or equal to a preset similarity, then the preset segmentation information corresponding to the noise distribution information L(j,s) less than or equal to the preset similarity can be determined. Afterward, this preset segmentation information is no longer used to segment the sample dataset, and a leaf node is generated. Then, the measurement error data of each sample measurement data in the unsegmented sample dataset can be used to generate the noise parameter information corresponding to the leaf node. For example, the noise mean and noise variance can be calculated according to Formulas 2 and 3 to determine the noise parameter information.

[0162] In some embodiments, during the Nth iteration, if the noise distribution information corresponding to all preset segmentation information does not meet the preset conditions, the sample dataset is divided into multiple sample data subsets based on the optimal segmentation information corresponding to the internal nodes generated in the current iteration, and each sample data subset is used as the sample dataset in the N+1th iteration; where N is a positive integer. That is, return to the above step of "determining multiple preset segmentation information based on the sample dataset" to continue determining internal nodes or leaf nodes.

[0163] For example, if the noise distribution information L(j,s) corresponding to the determined preset segmentation information is greater than the preset threshold, the sample dataset can be divided into multiple sample data subsets based on the segmentation points indicated by the optimal segmentation information; alternatively, the sample dataset can also be divided into multiple sample data subsets based on the preset feature threshold and preset dimension in the optimal segmentation information. Here, the process of segmenting the sample dataset based on the optimal segmentation information can be referred to the above process of segmenting and dividing the sample dataset based on the preset segmentation information, and the repeated parts will not be repeated.

[0164] If the noise distribution information corresponding to all preset segmentation information does not meet the preset conditions, each time the sample dataset is segmented, the resulting multiple sample data subsets correspond to multiple next-level child nodes to be generated in the currently generated internal node.

[0165] Calculating the noise parameter information in the leaf node requires measurement error data from multiple sample measurement data points within the sample data subset. In some embodiments, if the number of sample measurement data points included in the data subset after segmentation is less than a preset number, a leaf node is directly generated, and the sample data subset is not further segmented. Here, the preset number can be set based on empirical values, and this embodiment does not impose a specific limitation.

[0166] For the above embodiments, please refer to Figure 3 The diagram shown illustrates the specific structure of a decision tree model. Noise parameter information corresponding to leaf nodes can be found using this decision tree model.

[0167] For example, known measurement data Measurement data x v It includes multiple eigenvalues, namely like Figure 3 As shown, for root node 31, if the measured data x v The w-th eigenvalue in the eigenvalue is less than s. w ,Right now Then proceed to the child node corresponding to the "yes" path of root node 31, which is internal node 32; if the measured data x v The i-th eigenvalue in the eigenvalue is not less than s.i ,Right now Then proceed to the child node corresponding to the "No" path of internal node 32; continue to judge according to the above rules until the noise parameter information corresponding to the leaf node is determined, namely the noise parameter μ and the noise variance ε.

[0168] In some embodiments, after determining the target noise information, it can also be used to determine whether the measurement data is abnormal data. Specifically, the degree of deviation between the measurement data and the noise indicated by the target noise information can be determined; a preset probability threshold for the measurement data to be abnormal data can be obtained; a deviation threshold can be determined based on the preset probability threshold; and if the deviation is greater than or equal to the deviation threshold, the measurement data can be determined to be abnormal data.

[0169] The deviation between the measured data and the noise indicated by the target noise information can include the Mahalanobis distance between the measured data and the mean noise in the corresponding target noise information. Since the Mahalanobis distance follows a chi-square distribution, a Mahalanobis distance threshold, i.e., the deviation threshold, can be obtained from the chi-square distribution table based on a preset probability threshold. Then, if the deviation is greater than or equal to the deviation threshold, the measured data can be determined to be abnormal; if the deviation is less than the deviation threshold, the measured data can be determined not to be abnormal.

[0170] The Mahalanobis distance r can be determined using Formula 5:

[0171] r=(μ-x i ) T ε -1 (μ-x i )……………Formula 5

[0172] Where, ε -1 This represents the inverse of the noise variance ε. μ represents the measurement data x. i The mean noise value in the corresponding target noise information; ε represents the measured data x i The noise variance in the corresponding target noise information.

[0173] In some embodiments, after determining the target noise information, the multiple measurement data can be fused based on the target noise information corresponding to each of the multiple measurement data. Specifically, for the measurement data collected by multiple sensor devices in the vehicle, the target noise information for each measurement data is determined; based on the target noise information of each measurement data, the fusion weight of each measurement data is determined; based on the fusion weight of each measurement data, the multiple measurement data are fused to determine the fused measurement data, so that the vehicle can determine the operating environment information based on the fused measurement data.

[0174] The fusion weights can be determined using the Kalman filter algorithm. The noise indicated by the target noise information corresponding to each of the multiple measurement data points is processed by Kalman filtering to obtain the fusion weight for each measurement data point. Using these fusion weights, the multiple measurement data points are then fused to determine the final fused measurement data. Here, utilizing more accurate target noise information can improve the accuracy of the fusion weights in the multi-sensor fusion process, thereby improving the accuracy of the fused measurement data.

[0175] Taking autonomous driving scenarios as an example, measurement fusion data can be information about the surrounding operating environment perceived by the vehicle through multiple sensor devices. Using this measurement fusion data, accurate environmental perception data can be obtained, enabling high-precision autonomous driving.

[0176] Those skilled in the art will understand that, in the above-described method of the specific implementation, the order in which each step is written does not imply a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic.

[0177] Based on the same inventive concept, this disclosure also provides a noise determination device corresponding to the noise determination method. Since the principle of the device in this disclosure is similar to that of the noise determination method described above, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0178] Reference Figure 4 The diagram shown is a schematic representation of a noise determination device provided in an embodiment of this disclosure. The device includes: an information acquisition module 401, a node determination module 402, and a noise estimation module 403; wherein,

[0179] The information acquisition module 401 is used to acquire measurement data collected by a preset device and a pre-constructed decision tree model corresponding to the preset device; the decision tree model includes multiple internal nodes, optimal splitting information corresponding to each internal node, multiple leaf nodes, and noise parameter information corresponding to each leaf node.

[0180] The node determination module 402 is used to determine the target leaf node corresponding to the measurement data based on the optimal segmentation information corresponding to each internal node;

[0181] The noise estimation module 403 is used to determine the target noise information of the measurement data based on the noise parameter information corresponding to the target leaf node.

[0182] In one optional implementation, the device further includes a model building module 404, used to acquire a sample dataset collected by the preset device, and based on the sample dataset, determine multiple internal nodes, optimal splitting information corresponding to each internal node, and multiple leaf nodes.

[0183] Determine the noise parameter information corresponding to each leaf node;

[0184] The decision tree model is constructed based on the determined multiple internal nodes, multiple leaf nodes, the optimal splitting information corresponding to each internal node, and the noise parameter information corresponding to each leaf node.

[0185] In one optional implementation, the model building module 404 is used to iteratively execute the following steps:

[0186] Based on the sample dataset, multiple preset segmentation information is determined;

[0187] For each preset segmentation information, the sample dataset is divided into multiple data subsets based on the preset segmentation information, and noise sub-information corresponding to each data subset is determined respectively;

[0188] Based on the noise sub-information corresponding to each data subset, the noise distribution information of the sample dataset corresponding to the preset segmentation information is determined; wherein, the noise distribution information is used to characterize the noise similarity of the sample dataset corresponding to the preset segmentation information;

[0189] Based on the noise distribution information corresponding to each preset segmentation information, an internal node or a leaf node is generated, and in the case of generating the internal node, the optimal segmentation information corresponding to the internal node is determined.

[0190] In one optional implementation, the sample dataset includes multiple sample measurement data and measurement error data corresponding to each sample measurement data.

[0191] The model building module 404 is used to determine the noise sub-information corresponding to the data subset based on the measurement error data corresponding to the measurement data of each sample in the data subset.

[0192] In one optional implementation, the model building module 404 is used to generate an internal node when the noise distribution information corresponding to all preset segmentation information does not meet the preset conditions, and to take the preset segmentation information with the highest similarity indicated by the noise distribution information as the optimal segmentation information corresponding to the internal node.

[0193] In one optional implementation, the model building module 404 is used to generate a leaf node when the noise distribution information corresponding to at least one preset segmentation information meets preset conditions.

[0194] In an optional implementation, the model building module 404 is further configured to, during the Nth iteration, if the noise distribution information corresponding to all preset segmentation information in the current iteration does not meet the preset conditions, divide the sample dataset into multiple sample data subsets based on the optimal segmentation information corresponding to the internal nodes generated in the current iteration, and use each sample data subset as the sample dataset used in the N+1th iteration; where N is a positive integer.

[0195] In one optional implementation, the sample dataset includes multiple sample measurement data, and the sample measurement data includes feature values ​​of multiple dimensions; the preset segmentation information includes preset feature thresholds and preset dimensions;

[0196] The model building module 404 is used to take the sample measurement data in the sample dataset whose feature values ​​of a preset dimension are less than or equal to a preset feature threshold as target measurement data, and generate a data subset based on the selected target measurement data.

[0197] Based on the sample measurement data in the sample dataset other than the target measurement data, another data subset is generated.

[0198] In one optional implementation, the noise parameter information includes the noise mean and noise variance; the sample dataset includes multiple sample measurement data and measurement error data corresponding to each sample measurement data.

[0199] The model building module 404 is used to determine the noise mean for each leaf node based on the measurement error data in the sample dataset corresponding to the leaf node.

[0200] Based on the noise mean and various measurement error data, the noise variance is determined.

[0201] In an optional embodiment, the device further includes an abnormal data determination module 405, which is used to determine the degree of deviation between the measured data and the noise indicated by the target noise information after determining the target noise information of the measured data based on the noise parameter information corresponding to the target leaf node;

[0202] Obtain a preset probability threshold for the measured data to belong to abnormal data;

[0203] Based on the preset probability threshold, a deviation threshold is determined;

[0204] If the deviation is greater than or equal to the deviation threshold, the measured data is determined to be abnormal data.

[0205] In an optional embodiment, the device further includes a data fusion module 406, which, after determining the target noise information of the measurement data, determines the target noise information of each measurement data collected by multiple sensor devices in the vehicle.

[0206] Based on the target noise information of each measurement data, the fusion weight of each measurement data is determined;

[0207] Based on the fusion weight of each measurement data, multiple measurement data are fused to determine measurement fusion data, so that the vehicle can determine operating environment information based on the measurement fusion data.

[0208] The processing flow of each module in the device and the interaction flow between each module can be referred to the relevant descriptions in the above method embodiments, and will not be detailed here.

[0209] Based on the same technical concept, embodiments of this application also provide a computer device. (Refer to...) Figure 5 The diagram shown is a structural schematic of a computer device provided in an embodiment of this application, including:

[0210] The system includes a processor 51, a memory 52, and a bus 53. The memory 52 stores machine-readable instructions executable by the processor 51. The processor 51 executes these machine-readable instructions, and when executed, it performs the following steps: S101: Acquire measurement data collected by a preset device and a pre-constructed decision tree model; the decision tree model includes multiple internal nodes, optimal segmentation information corresponding to each internal node, multiple leaf nodes, and noise parameter information corresponding to each leaf node; S102: Based on the optimal segmentation information corresponding to each internal node, determine the target leaf node corresponding to the measurement data; S103: Based on the noise parameter information corresponding to the target leaf node, determine the target noise information of the measurement data.

[0211] The aforementioned memory 52 includes a main memory 521 and an external memory 522. The main memory 521, also known as internal memory, is used to temporarily store the computational data in the processor 51, as well as the data exchanged with external memory such as a hard disk. The processor 51 exchanges data with the external memory 522 through the main memory 521. When the computer device is running, the processor 51 and the memory 52 communicate through the bus 53, so that the processor 51 executes the execution instructions mentioned in the above method embodiments.

[0212] This disclosure also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the noise determination method described in the above-described method embodiments. The storage medium may be a volatile or non-volatile computer-readable storage medium.

[0213] This disclosure also provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the noise determination method described above. The computer program product can be any product capable of implementing the noise determination method described above. Part or all of the solutions in the computer program product that contribute to the prior art can be embodied in the form of a software product (e.g., a software development kit, SDK). This software product can be stored in a storage medium, and the included computer instructions cause a relevant device or processor to execute part or all of the steps of the noise determination method described above.

[0214] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the device described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here. In the several embodiments provided in this disclosure, it should be understood that the disclosed device and method can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division; in actual implementation, there may be other division methods. Furthermore, multiple modules or components can be combined, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some communication interfaces; the indirect coupling or communication connection of devices or modules can be electrical, mechanical, or other forms.

[0215] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0216] In addition, the functional modules in the various embodiments of this disclosure can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0217] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this disclosure, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this disclosure. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0218] Finally, it should be noted that the above-described embodiments are merely specific implementations of this disclosure, used to illustrate the technical solutions of this disclosure, and not to limit it. The protection scope of this disclosure is not limited thereto. Although this disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this disclosure. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this disclosure, and should all be covered within the protection scope of this disclosure. Therefore, the protection scope of this disclosure should be determined by the protection scope of the claims.

Claims

1. A method for determining noise, characterized in that, include: Acquire measurement data collected by a preset device, and a pre-constructed decision tree model corresponding to the preset device; The decision tree model includes multiple internal nodes, optimal splitting information corresponding to each internal node, multiple leaf nodes, and noise parameter information corresponding to each leaf node. Based on the optimal segmentation information corresponding to each internal node, the target leaf node corresponding to the measurement data is determined; Based on the noise parameter information corresponding to the target leaf node, the target noise information of the measurement data is determined; For measurement data collected by multiple sensor devices in the vehicle, the target noise information of each measurement data is determined; Based on the target noise information of each measurement data, the fusion weight of each measurement data is determined; Based on the fusion weight of each measurement data, multiple measurement data are fused to determine measurement fusion data, so that the vehicle can determine operating environment information based on the measurement fusion data. The steps for constructing the decision tree model include: acquiring a sample dataset collected by the preset device, and based on the sample dataset, determining multiple internal nodes, optimal splitting information corresponding to each internal node, and multiple leaf nodes; determining noise parameter information corresponding to each leaf node; and constructing the decision tree model based on the determined multiple internal nodes, multiple leaf nodes, optimal splitting information corresponding to each internal node, and noise parameter information corresponding to each leaf node. The step of determining multiple internal nodes, the optimal segmentation information corresponding to each internal node, and multiple leaf nodes based on the sample dataset includes: iteratively executing the following steps to determine multiple internal nodes, the optimal segmentation information corresponding to each internal node, and multiple leaf nodes, wherein each iteration generates one internal node or one leaf node: determining multiple preset segmentation information based on the sample dataset; for each preset segmentation information, dividing the sample dataset into multiple data subsets based on the preset segmentation information, and determining noise sub-information corresponding to each data subset; determining noise distribution information of the sample dataset corresponding to the preset segmentation information based on the noise sub-information corresponding to each data subset; wherein the noise distribution information is used to characterize the noise similarity of the sample dataset corresponding to the preset segmentation information; generating one internal node or one leaf node based on the noise distribution information corresponding to each preset segmentation information, and determining the optimal segmentation information corresponding to the internal node when the internal node is generated.

2. The method according to claim 1, characterized in that, The sample dataset includes multiple sample measurement data and measurement error data corresponding to each sample measurement data; Determining the noise sub-information corresponding to the data subset includes: Based on the measurement error data corresponding to each sample measurement data in the data subset, the noise sub-information corresponding to the data subset is determined.

3. The method according to claim 1, characterized in that, Based on the noise distribution information corresponding to each preset segmentation information, an internal node is generated, and the optimal segmentation information corresponding to the internal node is determined, including: If the noise distribution information corresponding to all preset segmentation information does not meet the preset conditions, an internal node is generated, and the preset segmentation information with the highest noise similarity indicated by the noise distribution information is taken as the optimal segmentation information corresponding to the internal node.

4. The method according to claim 1, characterized in that, Based on the noise distribution information corresponding to each preset segmentation information, a leaf node is generated, including: If the noise distribution information corresponding to at least one preset segmentation information meets the preset conditions, a leaf node is generated.

5. The method according to claim 3, characterized in that, Also includes: In the Nth iteration, if the noise distribution information corresponding to all preset segmentation information in the current iteration does not meet the preset conditions, the sample dataset is divided into multiple sample data subsets based on the optimal segmentation information corresponding to the internal nodes generated in the current iteration, and each sample data subset is used as the sample dataset in the N+1th iteration; where N is a positive integer.

6. The method according to any one of claims 1 to 5, characterized in that, The sample dataset includes multiple sample measurement data, and the sample measurement data includes feature values ​​in multiple dimensions. The preset segmentation information includes preset feature thresholds and preset dimensions; The step of dividing the sample dataset into multiple data subsets based on the preset segmentation information includes: The sample measurement data in the sample dataset whose feature values ​​of a preset dimension are less than or equal to a preset feature threshold are taken as target measurement data, and a data subset is generated based on the selected target measurement data; Based on the sample measurement data in the sample dataset other than the target measurement data, another data subset is generated.

7. The method according to claim 4, characterized in that, The noise parameter information includes the noise mean and noise variance; the sample dataset includes multiple sample measurement data and measurement error data corresponding to each sample measurement data. The step of determining the noise parameter information corresponding to each leaf node includes: For each leaf node, the noise mean is determined based on the measurement error data in the sample dataset corresponding to the leaf node; Based on the noise mean and various measurement error data, the noise variance is determined.

8. The method according to any one of claims 1 to 5, characterized in that, After determining the target noise information of the measurement data based on the noise parameter information corresponding to the target leaf node, the method further includes: Determine the degree of deviation between the measured data and the noise indicated by the target noise information; Obtain a preset probability threshold for the measured data to belong to abnormal data; Based on the preset probability threshold, a deviation threshold is determined; If the deviation is greater than or equal to the deviation threshold, the measured data is determined to be abnormal data.

9. A noise determination device, characterized in that, include: The information acquisition module is used to acquire measurement data collected by preset devices and pre-built decision tree models; The decision tree model includes multiple internal nodes, optimal splitting information corresponding to each internal node, multiple leaf nodes, and noise parameter information corresponding to each leaf node. The node determination module is used to determine the target leaf node corresponding to the measurement data based on the optimal splitting information corresponding to each internal node. The noise estimation module is used to determine the target noise information of the measurement data based on the noise parameter information corresponding to the target leaf node. The data fusion module is used to determine the target noise information of each measurement data point collected by multiple sensor devices in the vehicle after determining the target noise information of the measurement data. Based on the target noise information of each measurement data, the fusion weight of each measurement data is determined; Based on the fusion weight of each measurement data, multiple measurement data are fused to determine measurement fusion data, so that the vehicle can determine operating environment information based on the measurement fusion data. The model building module is used to acquire the sample dataset collected by the preset device, and based on the sample dataset, determine multiple internal nodes, the optimal segmentation information corresponding to each internal node, and multiple leaf nodes. Determine the noise parameter information corresponding to each leaf node; Based on the determined multiple internal nodes, multiple leaf nodes, optimal splitting information corresponding to each internal node, and noise parameter information corresponding to each leaf node, the decision tree model is constructed. The model building module is used to iteratively execute the following steps: based on the sample dataset, determine multiple preset segmentation information; for each preset segmentation information, divide the sample dataset into multiple data subsets based on the preset segmentation information, and determine the noise sub-information corresponding to each data subset; based on the noise sub-information corresponding to each data subset, determine the noise distribution information of the sample dataset corresponding to the preset segmentation information; wherein, the noise distribution information is used to characterize the noise similarity of the sample dataset corresponding to the preset segmentation information; Based on the noise distribution information corresponding to each preset segmentation information, an internal node or a leaf node is generated, and in the case of generating the internal node, the optimal segmentation information corresponding to the internal node is determined.

10. A computer device, characterized in that, include: The computer device includes a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor, and the processor communicates with the memory via the bus when the computer device is running, and the machine-readable instructions, when executed by the processor, perform the steps of the noise determination method as described in any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the noise determination method as described in any one of claims 1 to 8.

Images