Factor identification method and device for limiting vehicle perception performance, and storage medium

Abstract

The invention provides a factor identification method and device for limiting vehicle perception performance and a storage medium. According to the method, a factor analysis network model is constructed for any sensing target of a vehicle sensing system, a conditional probability table is obtained through statistics in a training sample set, observation frequencies and expected frequencies of factor nodes are obtained based on the conditional probability table, chi-square statistics is calculated according to the multiple observation frequencies and the multiple expected frequencies, and the vehicle sensing target is obtained. The factor analysis network model and the conditional probability table are updated according to chi-square statistics until the factor analysis network model is stable, factor nodes are extracted from the stable factor analysis network model to serve as factors for limiting the vehicle perception system to perceive the target, and according to conditional probabilities in the conditional probability table, the vehicle perception system is determined according to the conditional probabilities in the conditional probability table. And the false positive rate or the false negative rate of the vehicle sensing system to the sensing target in different states of each factor node is calculated, so that the factors influencing the vehicle sensing system can be comprehensively and accurately identified.

Description

Technical Field

[0001] The present application relates to the field of vehicle intelligent driving technology, and in particular to a method, device and storage medium for identifying factors that limit vehicle perception performance. Background Art

[0002] With the rapid development of intelligent driving technology, the performance of the perception system has become a key factor affecting vehicle safety.

[0003] The performance of the perception system in different driving scenarios is affected by many factors, such as traffic density, weather conditions, and obstructions.

[0004] Existing solutions usually evaluate risk factor events based on existing test scenarios, and do not explore and update risk factors that affect vehicle perception functions. Therefore, it is difficult to comprehensively and accurately identify factors that affect vehicle perception systems. Summary of the invention

[0005] The embodiments of the present application provide a method, device and storage medium for identifying factors that limit vehicle perception performance, so as to address the defect that existing solutions are difficult to comprehensively and accurately identify factors that affect the vehicle perception system.

[0006] In a first aspect, an embodiment of the present application provides a method for identifying factors that limit vehicle perception performance, including:

[0007] A factor analysis network model is constructed for any perception target of a vehicle perception system, wherein the factor analysis network model includes a plurality of nodes and a plurality of directed edges, each of the directed edges connects two of the nodes, the two nodes include a parent node and a child node, the directed edge points from the parent node to the child node, and the plurality of nodes include: a factor node affecting the perception target, a performance limitation node corresponding to the perception target, and the performance limitation node is used to indicate a false negative rate or a false positive rate of the perception target;

[0008] Acquire a training sample set for the factor analysis network model, wherein the training sample set includes a plurality of training samples, and each training sample includes the state of each of the nodes in the factor analysis network model;

[0009] Counting a conditional probability table of the factor analysis network model from the training sample set, wherein the conditional probability table includes a probability that the child node is in a second state when the parent node of the child node is in a first state;

[0010] Determine the actual number of simultaneous occurrences of any state of the factor node and any state of the performance constraint node in the training sample set as the observed frequency, and determine the expected number of simultaneous occurrences of any state of the factor node and any state of the performance constraint node in the training sample set as the expected frequency, and calculate a chi-square statistic based on a plurality of the observed frequencies and a plurality of the expected frequencies;

[0011] The factor analysis network model and the conditional probability table are updated according to the chi-square statistic until the factor analysis network model is stable, so as to extract factor nodes from the stable factor analysis network model as factors that limit the vehicle perception system to the perception target, and according to each conditional probability in the conditional probability table, calculate the false positive rate or false negative rate of the vehicle perception system to the perception target under different states of each factor node.

[0012] In a possible implementation manner, determining the expected number of simultaneous occurrences of any state of the factor node and any state of the performance constraint node in the training sample set includes:

[0013] Counting the number of occurrences of any state of the factor node in the training sample set;

[0014] Counting the number of occurrences of any state of the performance-limiting node in the training sample set;

[0015] According to the number of occurrences of any state of the factor node in the training sample set, the number of occurrences of any state of the performance constraint node in the training sample set, and the number of training samples in the training sample set, the expected number of simultaneous occurrences of any state of the factor node and any state of the performance constraint node in the training sample set is calculated, the expected number of simultaneous occurrences is positively correlated to the number of occurrences of any state of the factor node in the training sample set, the number of occurrences of any state of the performance constraint node in the training sample set, and the expected number of simultaneous occurrences is negatively correlated to the number of training samples.

[0016] In a possible implementation, the calculating the chi-square statistic according to the plurality of observed frequencies and the plurality of expected frequencies includes:

[0017] The chi-square statistic is calculated by the following formula:

[0018]

[0019] Among them, X 2 is the chi-square statistic, O ijis the observation frequency corresponding to the i-th state of the factor node and the j-th state of the performance constraint node, E ij is the expected frequency corresponding to the i-th state of the factor node and the j-th state of the performance constraint node, r is the number of states of the factor node, and c is the number of states of the performance constraint node.

[0020] In a possible implementation, updating the factor analysis network model and the conditional probability table according to the chi-square statistic includes:

[0021] Searching for a test result parameter corresponding to the chi-square statistic from a mapping table, and determining a target hypothesis from two mutually exclusive hypotheses according to the test result parameter;

[0022] If the target hypothesis indicates that there is no association relationship between the factor node and the performance constraint node, the factor node and the directed edge connecting the factor node are deleted from the factor analysis network model, and the conditional probability related to the factor node is deleted from the conditional probability table.

[0023] In a possible implementation, extracting factor nodes from the stable factor analysis network model as factors limiting the vehicle perception system to the perception target includes:

[0024] The verification sample set is verified by the factor analysis network model to obtain a verification result, wherein the verification result includes: the number of first samples correctly predicted as positive in the verification sample set, the number of second samples correctly predicted as negative in the verification sample set, the number of third samples incorrectly predicted as positive in the verification sample set, and the number of fourth samples incorrectly predicted as negative in the verification sample set; when the performance limiting node is used to indicate the false negative rate of the perception target, the positive class is a false negative, and the negative class is a non-false negative; when the performance limiting node is used to indicate the false positive rate of the perception target, the positive class is a false positive, and the negative class is a non-false positive;

[0025] Determine the accuracy and / or recall of the factor analysis network model according to the verification result, wherein the accuracy is positively correlated to the sum of the first sample number and the second sample number, the accuracy is negatively correlated to the sum of the first sample number, the second sample number, the third sample number and the fourth sample number, and the recall is positively correlated to the first sample number and negatively correlated to the sum of the first sample number and the fourth sample number;

[0026] If the accuracy is greater than or equal to a preset accuracy threshold, and / or the recall is greater than or equal to a preset recall threshold, factor nodes are extracted from the stable factor analysis network model as factors for limiting the vehicle perception system to the perception target.

[0027] In a possible implementation, updating the factor analysis network model and the conditional probability table according to the chi-square statistic until the factor analysis network model is stable includes:

[0028] updating the factor analysis network model and the conditional probability table according to the chi-square statistic;

[0029] When receiving the target factor selected from the training sample set, adding the target factor to the factor analysis network model, and returning to the step of statistically calculating the conditional probability table of the factor analysis network model from the training sample set;

[0030] When the target factor selected from the training sample set is not received and each of the factor nodes in the factor analysis network model has been verified, it is determined that the factor analysis network model is stable, and the verification of the factor node indicates that it has been verified whether the factor node has an impact on the perception target.

[0031] In a second aspect, an embodiment of the present application provides a device for identifying factors that limit vehicle perception performance, including:

[0032] An acquisition module is used to construct a factor analysis network model for any perception target of a vehicle perception system, wherein the factor analysis network model includes a plurality of nodes and a plurality of directed edges, each of the directed edges connects two of the nodes, the two nodes include a parent node and a child node, the directed edge points from the parent node to the child node, and the plurality of nodes include: a factor node affecting the perception target, a performance limitation node corresponding to the perception target, and the performance limitation node is used to indicate a false negative rate or a false positive rate of the perception target;

[0033] The acquisition module is further used to acquire a training sample set for the factor analysis network model, wherein the training sample set includes a plurality of training samples, and each training sample includes the state of each node in the factor analysis network model;

[0034] A processing module, configured to generate a conditional probability table of the factor analysis network model from the training sample set, wherein the conditional probability table includes a probability that the child node is in a second state when the parent node of the child node is in a first state;

[0035] The processing module is further used to determine the actual number of simultaneous occurrences of any state of the factor node and any state of the performance constraint node in the training sample set as the observed frequency, and to determine the expected number of simultaneous occurrences of any state of the factor node and any state of the performance constraint node in the training sample set as the expected frequency, and to calculate a chi-square statistic based on a plurality of the observed frequencies and a plurality of the expected frequencies;

[0036] An execution module is used to update the factor analysis network model and the conditional probability table according to the chi-square statistic until the factor analysis network model is stable, so as to extract factor nodes from the stable factor analysis network model as factors that limit the vehicle perception system to the perception target, and calculate the false positive rate or false negative rate of the vehicle perception system to the perception target under different states of each factor node according to each conditional probability in the conditional probability table.

[0037] In a third aspect, an embodiment of the present application provides an electronic device, including: a memory, a processor;

[0038] The memory stores computer-executable instructions;

[0039] The processor executes the computer-executable instructions stored in the memory, so that the processor executes the first aspect and / or various possible implementations of the first aspect as described above.

[0040] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the first aspect above and / or various possible implementations of the first aspect.

[0041] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the above first aspect and / or various possible implementation methods of the first aspect.

[0042] The method, device and storage medium for identifying factors that limit vehicle perception performance provided in the embodiments of the present application construct a factor analysis network model for any perception target of the vehicle perception system, and obtain a conditional probability table by statistics in a training sample set, and obtain the observed frequency and expected frequency of each factor node based on the conditional probability table, calculate the chi-square statistic according to multiple observed frequencies and multiple expected frequencies to determine the significance of each factor, and update the factor analysis network model and the conditional probability table according to the chi-square statistic until the factor analysis network model is stable, so as to extract factor nodes from the stable factor analysis network model as factors that limit the vehicle perception system to the perception target, and calculate the false positive rate or false negative rate of the vehicle perception system to the perception target under different states of each factor node according to each conditional probability in the conditional probability table, thereby achieving comprehensive and accurate identification of factors affecting the vehicle perception system. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0044] Figure 1 Schematic diagram of the process of identifying factors limiting vehicle perception performance provided by this application Figure 1 ;

[0045] Figure 2 Schematic diagram of the process of identifying factors limiting vehicle perception performance provided by this application Figure 2 ;

[0046] Figure 3 Schematic diagram of the initial structure of the factor analysis network provided for this application;

[0047] Figure 4 A schematic diagram of the updated network structure of the factor analysis network provided for this application;

[0048] Figure 5 A schematic diagram of the structure of a device for identifying factors limiting vehicle perception performance provided by the present application;

[0049] Figure 6 A schematic diagram of the structure of the electronic device provided in this application.

[0050] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0051] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0052] The performance of the vehicle perception system is a key factor that can affect vehicle safety. The vehicle perception system can perceive information such as the vehicle's position, driving speed, posture and orientation, as well as detailed information about the surrounding environment such as the location, shape, type and speed of obstacles and other obstacle-related data information.

[0053] The performance of the vehicle perception system in different driving scenarios is affected by different types of factors, namely boundary conditions. Existing traditional solutions usually analyze the expected functional safety corresponding to risk events in existing fixed vehicle test scenarios, and do not explore, analyze and evaluate new boundary conditions of vehicle perception functions. Therefore, it is difficult for existing traditional solutions to comprehensively and accurately identify the boundary conditions that limit vehicle perception performance.

[0054] Therefore, the present application provides a method for identifying factors that limit vehicle perception performance, by constructing a factor analysis network model for any perception target of the vehicle perception system, and obtaining a conditional probability table by statistics in a training sample set, and obtaining the observed frequency and expected frequency of each factor node based on the conditional probability table, and calculating the chi-square statistic according to multiple observed frequencies and multiple expected frequencies to judge the significance of each factor, and updating the factor analysis network model and the conditional probability table according to the chi-square statistic until the factor analysis network model is stable, so as to extract factor nodes from the stable factor analysis network model as factors that limit the vehicle perception system to the perception target, and according to each conditional probability in the conditional probability table, calculate the false positive rate or false negative rate of the vehicle perception system to the perception target under different states of each factor node, thereby realizing comprehensive and accurate identification of factors affecting the vehicle perception system.

[0055] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0056] Figure 1 Schematic diagram of the process of identifying factors limiting vehicle perception performance provided by this application Figure 1 ,like Figure 1 As shown, the method includes:

[0057] S101. Construct a factor analysis network model for any perception target of a vehicle perception system, wherein the factor analysis network model includes a plurality of nodes and a plurality of directed edges, each of the directed edges connects two of the nodes, the two nodes include a parent node and a child node, the directed edge points from the parent node to the child node, and the plurality of nodes include: a factor node affecting the perception target, a performance limitation node corresponding to the perception target, and the performance limitation node is used to indicate a false negative rate or a false positive rate of the perception target;

[0058] Specifically, the factor analysis network model includes factor nodes that affect the perception target and performance limitation nodes corresponding to the perception target. The factor nodes that affect the perception target include scene nodes and boundary condition nodes. Scene nodes refer to vehicles in different driving scenarios, such as vehicles running on national roads, provincial roads, expressways, and tunnels. Boundary condition nodes refer to factors that can affect vehicle perception errors, such as occlusion, weather, and traffic flow.

[0059] Furthermore, the performance limitation node reflects the performance of the vehicle perception system under the influence of the factor node. The performance limitation node can be a false negative node or a false positive node. For different performance limitation nodes, the corresponding Bayesian network can be reconstructed, and the false negative rate or false positive rate of the corresponding perception target can be obtained through the performance limitation node calculation. The nodes in the factor analysis network model are connected through a directed acyclic graph, that is, multiple nodes and multiple directed edges represent the conditional dependency relationship.

[0060] S102, obtaining a training sample set for the factor analysis network model, wherein the training sample set includes a plurality of training samples, and each training sample includes the state of each node in the factor analysis network model;

[0061] Specifically, after completing the construction of the factor analysis network model, a training sample set including multiple training samples is obtained, and the state of each node in the training sample is a specific value, such as the unobstructed state of the occluded node is zero, the partially occluded state is one, and the fully occluded state is two, so that according to the state of each node in the training sample, the probability of each node being in each state is statistically obtained.

[0062] S103, calculating a conditional probability table of the factor analysis network model from the training sample set, wherein the conditional probability table includes a probability that the child node is in a second state when the parent node of the child node is in a first state;

[0063] Specifically, after obtaining a training sample set including multiple training samples, a conditional probability table is obtained according to the node state of each training sample in the training sample set. The conditional probability table is used in factor analysis network models such as Bayesian networks to represent the probability distribution of child node states under a given parent node state, that is, the conditional probability of the corresponding node is obtained by counting the frequency of the combination of parent node and child node states.

[0064] S104, determining the actual number of simultaneous occurrences of any state of the factor node and any state of the performance constraint node in the training sample set as the observed frequency, and determining the expected number of simultaneous occurrences of any state of the factor node and any state of the performance constraint node in the training sample set as the expected frequency, and calculating a chi-square statistic based on a plurality of the observed frequencies and a plurality of the expected frequencies;

[0065] Specifically, after obtaining the conditional probability table, in the training sample set, the number of times any state of the factor node, such as the road line type is solid line, dashed line, or dashed-solid line, and the true false negative or true false positive of the performance limitation node knowledge simultaneously occur is counted as the observation frequency, and based on the number of times any state of the factor node occurs in the training sample set, the number of times any state of the performance limitation node occurs in the training sample set, and the number of training samples in the training sample set, the expected number of simultaneous occurrences of any state of the factor node and any state of the performance limitation node in the training sample set is calculated.

[0066] S105, updating the factor analysis network model and the conditional probability table according to the chi-square statistic until the factor analysis network model is stable, extracting factor nodes from the stable factor analysis network model as factors that restrict the vehicle perception system from sensing the target, and calculating the false positive rate or false negative rate of the vehicle perception system for the target under different states of each factor node according to each conditional probability in the conditional probability table;

[0067] Specifically, after calculating the chi-square statistic, the corresponding p-value, that is, the test result parameter, is found in the preset mapping table according to the chi-square statistic, and the test result parameter is used to detect whether the factor analysis network model and the conditional probability table are updated. If the factor analysis network model and the conditional probability table are updated, the false positive rate or false negative rate of the vehicle perception system for the perceived target under different states of each factor node is output based on the updated stable factor analysis network model and conditional probability table.

[0068] Therefore, the method for identifying factors that limit vehicle perception performance provided in the embodiment of the present application constructs a factor analysis network model for any perception target of the vehicle perception system, and statistically obtains a conditional probability table in a training sample set, and obtains the observed frequency and expected frequency of each factor node based on the conditional probability table, calculates the chi-square statistic according to multiple observed frequencies and multiple expected frequencies to determine the significance of each factor, and updates the factor analysis network model and the conditional probability table according to the chi-square statistic until the factor analysis network model is stable, so as to extract factor nodes from the stable factor analysis network model as factors that limit the vehicle perception system to the perception target, and calculates the false positive rate or false negative rate of the vehicle perception system to the perception target under different states of each factor node according to each conditional probability in the conditional probability table, thereby achieving comprehensive and accurate identification of factors affecting the vehicle perception system.

[0069] Figure 2 Schematic diagram of the process of identifying factors limiting vehicle perception performance provided by this application Figure 2 , Figure 3 The initial structure diagram of the factor analysis network provided for this application, Figure 4 The schematic diagram of the factor analysis network structure provided for this application is combined with Figure 2 , Figure 3 and Figure 4 As shown, in this embodiment Figure 1 Based on the embodiment, a method for identifying factors that limit vehicle perception performance is described in detail, and the method includes:

[0070] S201, constructing a factor analysis network model for any perception target of the vehicle perception system;

[0071] Specifically, according to expert knowledge and national standards, for any perception target of the vehicle perception system, such as identifying road lines, identifying adjacent vehicles, or identifying obstacles, when it is necessary to analyze the influencing factors of a certain perception target, a corresponding factor analysis network model such as a Bayesian network structure is constructed for the perception target. The factor analysis network model includes all factors that may affect the perception target and a false negative or false positive table of the perception target. Of course, the factor analysis network model can also be an analysis model such as a hidden Markov model, a decision tree, or a random forest.

[0072] Furthermore, taking the performance of the vehicle perception system for lane line detection as an example, based on expert knowledge that it is affected by road conditions, reflections and occlusions, experts establish hierarchical dependencies between the performance limitations of the scene and the factor nodes that affect the perception target. Based on these dependencies, propositions are established, such as high occlusion may lead to higher false negatives; weather affects the false negative rate by affecting roads and lighting; road and lighting conditions affect reflections and thus affect the false negative rate. Based on the established propositions, factor analysis network models such as Bayesian networks are constructed to achieve quantification of the interaction between the environment and vehicle perception systems as indicated by national standards such as ISO21448.

[0073] S202, obtaining a training sample set for the factor analysis network model, wherein the training sample set includes a plurality of training samples, and each training sample includes the state of each node in the factor analysis network model;

[0074] Specifically, the content of this step is the same as that of S102 and will not be repeated here.

[0075] S203, calculating a conditional probability table of the factor analysis network model from the training sample set, wherein the conditional probability table includes a probability that the child node is in a second state when the parent node of the child node is in a first state;

[0076] Specifically, for each node X in a factor analysis network model such as a Bayesian network, assume that it has n parent nodes P a (X) = {Y1, Y2, ..., Y n}, each parent node Y i Yes i possible states, and the child node X has m possible states. Then the conditional probability table CPT(X) is a A matrix where each element CPT(X) jk It indicates the probability that the child node X is in state k under the parent node state combination j, which is obtained by the following calculation formula:

[0077] CPT(X) jk =P(X=k|P a (X) = j)

[0078] Among them, j represents an index of the parent node state combination, and k represents a possible state of the child node X. In actual calculation, the probability of each node is obtained by counting the frequency of the corresponding state combination in the training data set, and the probability of each node is obtained by the following calculation formula:

[0079]

[0080] Therefore, in the perception and detection of lane lines, the conditional probability table of the network model based on statistical factors in the training sample set is as follows:

[0081]

[0082] Among them, CPT (false negative) {反射(是)，遮挡(完全)，道线类型(实线)}假阴性(是) Indicates the probability of false negative being yes, reflection being yes, occlusion being complete, and road line type being solid line in the training dataset.

[0083] S204, determining the actual number of simultaneous occurrences of any state of the factor node and any state of the performance constraint node in the training sample set as the observation frequency;

[0084] Specifically, after obtaining the conditional probability table, the null hypothesis and alternative hypothesis are set. The null hypothesis H0 indicates that there are significant differences in the false negative indicators of the road line types solid line, dashed line and dashed-solid line, and the alternative hypothesis H1 indicates that there are no significant differences in the false negative indicators of the road line types solid line, dashed line and dashed-solid line, where the significance level is set to 0.05;

[0085] Furthermore, the number of occurrences of any state of the statistical factor node, such as the road type being a solid line, a dashed line, or a dashed-solid line in the training sample set, and the number of false negatives or false positives indicated by the corresponding performance limitation nodes of the road type being a solid line, a dashed line, or a dashed-solid line, are calculated. An observation frequency table is established for the frequency values ​​of different categories of road types, as shown in the following table:

[0086] False negative (yes) False negative (no) sum Solid Line A1 A2 A dotted line B1 B2 B Dashed and solid lines C1 C2 C sum D1 D2 D

[0087] Among them, A1 is the number of samples with a solid line and false negatives, that is, the number of occurrences of factor nodes such as solid line type in the training sample set;

[0088] Furthermore, in the training sample set, the number of times any state of the statistical factor node, such as the line type being a solid line, a dashed line, or a dashed-solid line, and the true false negative or true false positive indicated by the performance limitation node appear at the same time is taken as the observation frequency O ij As A1, where O ij is the observation frequency in the i-th row and j-th column of the observation frequency table.

[0089] S205, counting the number of occurrences of any state of the factor node in the training sample set, and counting the number of occurrences of any state of the performance restriction node in the training sample set;

[0090] Specifically, based on the above observation frequency table, the number of occurrences of corresponding factor nodes such as road type status as solid line, dashed line or dashed-solid line in the training sample set and the false negative or false positive indicated by the performance limitation node of the corresponding status are obtained.

[0091] S206, according to the number of occurrences of any state of the factor node in the training sample set, the number of occurrences of any state of the performance constraint node in the training sample set, and the number of training samples in the training sample set, calculate the expected number of simultaneous occurrences of any state of the factor node and any state of the performance constraint node in the training sample set as the expected frequency;

[0092] Specifically, the expected number of simultaneous occurrences is positively correlated to the number of occurrences of any state of the factor node in the training sample set, such as the total number of solid lines, and the number of occurrences of any state of the performance limit node in the training sample set, such as the total number of false negatives. The expected number of simultaneous occurrences is negatively correlated to the number of training samples, that is, the total number of samples. For the expected frequency of the line type being a solid line and a false negative, it is obtained by the following formula:

[0093]

[0094] Among them, A is the total number of solid lines, D1 is the total number of false negatives, and D is the total number of samples. For the expected frequency of false negatives when the line type is a dotted line, it is obtained by the following formula:

[0095]

[0096] Among them, B is the total number of dotted lines, D1 is the total number of false negatives, and D is the total number of samples.

[0097] S207, calculating a chi-square statistic according to the plurality of observed frequencies and the plurality of expected frequencies;

[0098] Specifically, after obtaining the observed frequency and expected frequency of each factor node, the chi-square statistic is calculated by the following formula:

[0099]

[0100] Among them, X 2 is the chi-square statistic, O ij is the observation frequency corresponding to the i-th state of a factor node and the j-th state of a performance constraint node, E ij is the expected frequency corresponding to the i-th state of a factor node and the j-th state of a performance constraint node, r is the number of states of a factor node, and c is the number of states of a performance constraint node.

[0101] S208, searching for the test result parameters corresponding to the chi-square statistic from the mapping table, and determining the target hypothesis from two mutually exclusive hypotheses according to the test result parameters;

[0102] Specifically, after the chi-square statistic is calculated, the corresponding p-value, i.e., the test result parameter, is found in the preset mapping table according to the chi-square statistic. The larger the chi-square statistic, the greater the difference between the observed frequency and the expected frequency, indicating that there may be an association between the categorical variables, i.e., the null hypothesis may not hold.

[0103] Further, if the p-value is less than the significance level such as 0.05, the null hypothesis is rejected, and the road line type has no significant difference on the false negative index, and it is determined that there is no correlation relationship between the target hypothesis indicator factor node and the performance limitation node;

[0104] If the p-value is greater than the significance level, the null hypothesis cannot be rejected, and there is insufficient evidence to show that the road line type has no significant difference in the false negative indicator. Then the target hypothesis indicates that there is an association between the factor node and the performance limitation node.

[0105] S209, if the target hypothesis indicates that there is no association relationship between the factor node and the performance constraint node, deleting the factor node and the directed edge connecting the factor node from the factor analysis network model, and deleting the conditional probability related to the factor node from the conditional probability table;

[0106] Specifically, if the target hypothesis indicates that there is no association between the factor node and the performance limitation node, the factor analysis network model and the conditional probability table need to be updated, the factor node is deleted from the factor analysis network model, and the conditional probability corresponding to the factor node is deleted from the conditional probability table. If the target hypothesis indicates that there is an association between the factor node and the performance limitation node, the factor node is retained in the factor analysis network model, and the conditional probability corresponding to the factor node is retained in the conditional probability table.

[0107] S210, when receiving the target factor selected from the training sample set, adding the target factor to the factor analysis network model, and returning to the step of statistically calculating the conditional probability table of the factor analysis network model from the training sample set;

[0108] Specifically, when updating the factor analysis network model and the conditional probability table, if it is necessary to add factor nodes to the factor analysis network model, such as adding road status and traffic flow status, that is, when receiving the target factor selected from the training sample set, the target factor is added to the factor analysis network model, and based on each node in the factor analysis network model after adding the target factor, the probability of each node is updated to obtain an updated conditional probability table.

[0109] S211, when the target factor selected from the training sample set is not received and each of the factor nodes in the factor analysis network model has been verified, determining that the factor analysis network model is stable, and the verification of the factor node indicates that it has been verified whether the factor node has an impact on the perception target;

[0110] Specifically, when the factor analysis network model and the conditional probability table are updated, if it is no longer necessary to add factor nodes to the factor analysis network model, that is, when the target factor selected from the training sample set is not received, and all factor nodes in the factor analysis network model and the performance constraint nodes are verified to have a correlation relationship, the factor analysis network model is determined to be stable.

[0111] S212, verifying the verification sample set by using the factor analysis network model to obtain a verification result, and determining the accuracy and / or recall rate of the factor analysis network model according to the verification result;

[0112] Specifically, the verification results include the number of first samples correctly predicted as positive in the verification sample set, the number of second samples correctly predicted as negative in the verification sample set, the number of third samples incorrectly predicted as positive in the verification sample set, and the number of fourth samples incorrectly predicted as negative in the verification sample set. When the performance limit node is used to indicate the false negative rate of the perception target, the positive class is false negative and the negative class is non-false negative. When the performance limit node is used to indicate the false positive rate of the perception target, the positive class is false positive and the negative class is non-false positive.

[0113] Furthermore, the accuracy is positively correlated to the sum of the first sample size and the second sample size, the accuracy is negatively correlated to the sum of the first sample size, the second sample size, the third sample size and the fourth sample size, the recall rate is positively correlated to the first sample size and negatively correlated to the sum of the first sample size and the fourth sample size. The accuracy is obtained by the following formula:

[0114]

[0115] Among them, Accuracy is the accuracy rate, TP is the number of samples correctly predicted by the model as the first class, TN is the number of samples correctly predicted by the model as the second class as the negative class, FP is the number of samples incorrectly predicted by the model as the third class as the positive class, and FN is the number of samples incorrectly predicted by the model as the fourth class as the negative class. The recall rate is obtained by the following formula:

[0116]

[0117] Among them, Recall is the recall rate, TP is the number of samples correctly predicted by the model as the first class, and FN is the number of samples incorrectly predicted by the model as the fourth class as the negative class.

[0118] S213: If the accuracy is greater than or equal to a preset accuracy threshold, and / or the recall is greater than or equal to a preset recall threshold, extracting factor nodes from the stable factor analysis network model as factors for limiting the vehicle perception system to the perception target;

[0119] Specifically, after obtaining the accuracy and recall, if the accuracy is not less than the preset accuracy threshold and the recall is less than the preset recall threshold, it is confirmed that the factor analysis network model meets the actual needs, otherwise the factor analysis network model and the conditional probability table are continued to be updated.

[0120] S214, calculating the false positive rate or false negative rate of the vehicle perception system for the perception target under different states of each of the factor nodes according to each conditional probability in the conditional probability table;

[0121] Specifically, after obtaining the factors that limit the vehicle perception system's perception of the target through the factor analysis network model, the perception error rate is calculated according to the conditional probabilities of each factor in the conditional probability table. The perception error rate includes the false negative rate and the false positive rate. The perception error rate indicates the error probability of the perception target under the factors affecting the vehicle perception system, so as to provide quantitative support for the expected functional safety analysis.

[0122] The factor analysis network model constructed in the embodiment of the present application can systematically consider multiple factors and their interactions, so as to more accurately evaluate the performance of the perception system and enhance the overall reliability of the intelligent driving technology. In addition, based on the accurate identification of the boundary conditions that limit the performance of the perception system, timely measures can be taken before or when these conditions occur, such as reducing the vehicle speed, issuing warnings, or taking over vehicle control, thereby effectively avoiding potential safety risks.

[0123] Safety of intended functions (SOTIF) analysis of intelligent driving vehicles is an important part of ensuring vehicle safety. The method for identifying factors that limit vehicle perception performance provided in the embodiments of the present application can provide a quantitative boundary condition model for SOTIF analysis, help engineers more comprehensively evaluate the performance of intelligent driving systems in various scenarios, and provide a scientific basis for vehicle design and improvement.

[0124] The method for identifying factors that limit vehicle perception performance provided in the embodiment of the present application constructs a factor analysis network model for any perception target of the vehicle perception system, obtains a conditional probability table by statistics in a training sample set, obtains the observed frequency and expected frequency of each factor node based on the conditional probability table, calculates a chi-square statistic based on multiple observed frequencies and multiple expected frequencies to determine the significance of each factor, updates the factor analysis network model and the conditional probability table based on the chi-square statistic until the factor analysis network model is stable, extracts factor nodes from the stable factor analysis network model as factors that limit the vehicle perception system for the perception target, and calculates the false positive rate or false negative rate of the vehicle perception system for the perception target under different states of each factor node based on each conditional probability in the conditional probability table, thereby achieving comprehensive and accurate identification of factors affecting the vehicle perception system.

[0125] Figure 5 A schematic diagram of the structure of the device for identifying factors limiting vehicle perception performance provided by the present application, such as Figure 5 As shown, the device 50 for identifying factors limiting vehicle perception performance provided in this embodiment includes:

[0126] The acquisition module 501 is used to construct a factor analysis network model for any perception target of the vehicle perception system, wherein the factor analysis network model includes a plurality of nodes and a plurality of directed edges, each of the directed edges connects two of the nodes, the two nodes include a parent node and a child node, the directed edge points from the parent node to the child node, and the plurality of nodes include: a factor node affecting the perception target, a performance limitation node corresponding to the perception target, and the performance limitation node is used to indicate a false negative rate or a false positive rate of the perception target;

[0127] The acquisition module is further used to acquire a training sample set for the factor analysis network model, wherein the training sample set includes a plurality of training samples, and each training sample includes the state of each node in the factor analysis network model;

[0128] Processing module 502, for collecting statistics of a conditional probability table of the factor analysis network model from the training sample set, wherein the conditional probability table includes a probability that the child node is in a second state when the parent node of the child node is in a first state;

[0129] The processing module 502 is further configured to determine the actual number of simultaneous occurrences of any state of the factor node and any state of the performance constraint node in the training sample set as the observed frequency, and to determine the expected number of simultaneous occurrences of any state of the factor node and any state of the performance constraint node in the training sample set as the expected frequency, and to calculate a chi-square statistic based on a plurality of the observed frequencies and a plurality of the expected frequencies;

[0130] The execution module 503 is used to update the factor analysis network model and the conditional probability table according to the chi-square statistic until the factor analysis network model is stable, so as to extract factor nodes from the stable factor analysis network model as factors that limit the vehicle perception system to the perception target, and calculate the false positive rate or false negative rate of the vehicle perception system to the perception target under different states of each factor node according to each conditional probability in the conditional probability table.

[0131] Furthermore, the processing module 502 is specifically used to count the number of occurrences of any state of the factor node in the training sample set;

[0132] Counting the number of occurrences of any state of the performance-limiting node in the training sample set;

[0133] According to the number of occurrences of any state of the factor node in the training sample set, the number of occurrences of any state of the performance constraint node in the training sample set, and the number of training samples in the training sample set, the expected number of simultaneous occurrences of any state of the factor node and any state of the performance constraint node in the training sample set is calculated, the expected number of simultaneous occurrences is positively correlated to the number of occurrences of any state of the factor node in the training sample set, the number of occurrences of any state of the performance constraint node in the training sample set, and the expected number of simultaneous occurrences is negatively correlated to the number of training samples.

[0134] Further, the processing module 502 is specifically configured to calculate the chi-square statistic using the following formula:

[0135]

[0136] Among them, X 2 is the chi-square statistic, O ij is the observation frequency corresponding to the i-th state of the factor node and the j-th state of the performance constraint node, E ij is the expected frequency corresponding to the i-th state of the factor node and the j-th state of the performance constraint node, r is the number of states of the factor node, and c is the number of states of the performance constraint node.

[0137] Further, the processing module 502 is specifically used to find the test result parameter corresponding to the chi-square statistic from the mapping table, and determine the target hypothesis from two mutually exclusive hypotheses according to the test result parameter;

[0138] If the target hypothesis indicates that there is no association relationship between the factor node and the performance constraint node, the factor node and the directed edge connecting the factor node are deleted from the factor analysis network model, and the conditional probability related to the factor node is deleted from the conditional probability table.

[0139] Further, the processing module 502 is specifically used to verify the verification sample set through the factor analysis network model to obtain a verification result, wherein the verification result includes: the number of first samples in the verification sample set that are correctly predicted as positive, the number of second samples in the verification sample set that are correctly predicted as negative, the number of third samples in the verification sample set that are incorrectly predicted as positive, and the number of fourth samples in the verification sample set that are incorrectly predicted as negative, when the performance limitation node is used to indicate the false negative rate of the perception target, the positive class is a false negative, and the negative class is a non-false negative, and when the performance limitation node is used to indicate the false positive rate of the perception target, the positive class is a false positive, and the negative class is a non-false positive;

[0140] Determine the accuracy and / or recall of the factor analysis network model according to the verification result, wherein the accuracy is positively correlated to the sum of the first sample number and the second sample number, the accuracy is negatively correlated to the sum of the first sample number, the second sample number, the third sample number and the fourth sample number, and the recall is positively correlated to the first sample number and negatively correlated to the sum of the first sample number and the fourth sample number;

[0141] If the accuracy is greater than or equal to a preset accuracy threshold, and / or the recall is greater than or equal to a preset recall threshold, factor nodes are extracted from the stable factor analysis network model as factors for limiting the vehicle perception system to the perception target.

[0142] Further, the execution module 503 is specifically used to update the factor analysis network model and the conditional probability table according to the chi-square statistic;

[0143] When receiving the target factor selected from the training sample set, adding the target factor to the factor analysis network model, and returning to the step of statistically calculating the conditional probability table of the factor analysis network model from the training sample set;

[0144] When the target factor selected from the training sample set is not received and each of the factor nodes in the factor analysis network model has been verified, it is determined that the factor analysis network model is stable, and the verification of the factor node indicates that it has been verified whether the factor node has an impact on the perception target.

[0145] The device for identifying factors limiting vehicle perception performance provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail in this embodiment.

[0146] Figure 6 This is a schematic diagram of the structure of the electronic device provided in this application. Figure 6 As shown, the electronic device 60 provided in this embodiment includes: at least one processor 601 and a memory 602. Optionally, the device 60 also includes a communication component 603. The processor 601, the memory 602 and the communication component 603 are connected via a bus 604.

[0147] In a specific implementation process, at least one processor 601 executes the computer execution instructions stored in the memory 602, so that at least one processor 601 executes the above method.

[0148] The specific implementation process of the processor 601 can be found in the above method embodiment, and its implementation principle and technical effect are similar, so this embodiment will not be repeated here.

[0149] In the above embodiments, it should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the invention may be directly implemented as being executed by a hardware processor, or may be executed by a combination of hardware and software modules in the processor.

[0150] The memory may include a high-speed memory (Random Access Memory, RAM), and may also include a non-volatile memory (Non-volatile Memory, NVM), such as at least one disk memory.

[0151] The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present application is not limited to only one bus or one type of bus.

[0152] The present application also provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0153] The present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the above method is implemented.

[0154] The above-mentioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general or special-purpose computer.

[0155] An exemplary readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (Application Specific Integrated Circuits, referred to as: ASIC). Of course, the processor and the readable storage medium can also exist in the device as discrete components.

[0156] The division of units is only a logical function division, and there may be other divisions in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.

[0157] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0158] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0159] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0160] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0161] Finally, it should be noted that those skilled in the art will readily conceive of other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention, which follow the general principles of the present invention and include common knowledge or customary technical means in the art not disclosed by the present invention, are not limited to the precise structure described above and shown in the drawings, and may be modified and changed in various ways without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A method for identifying factors that limit vehicle perception performance, characterized in that: include: A factor analysis network model is constructed for any perception target of a vehicle perception system, wherein the factor analysis network model includes a plurality of nodes and a plurality of directed edges, each of the directed edges connects two of the nodes, the two nodes include a parent node and a child node, the directed edge points from the parent node to the child node, and the plurality of nodes include: a factor node affecting the perception target, a performance limitation node corresponding to the perception target, and the performance limitation node is used to indicate a false negative rate or a false positive rate of the perception target; Acquire a training sample set for the factor analysis network model, wherein the training sample set includes a plurality of training samples, and each training sample includes the state of each of the nodes in the factor analysis network model; Counting a conditional probability table of the factor analysis network model from the training sample set, wherein the conditional probability table includes a probability that the child node is in a second state when the parent node of the child node is in a first state; Determine the actual number of simultaneous occurrences of any state of the factor node and any state of the performance constraint node in the training sample set as the observed frequency, and determine the expected number of simultaneous occurrences of any state of the factor node and any state of the performance constraint node in the training sample set as the expected frequency, and calculate a chi-square statistic based on a plurality of the observed frequencies and a plurality of the expected frequencies; The factor analysis network model and the conditional probability table are updated according to the chi-square statistic until the factor analysis network model is stable, so as to extract factor nodes from the stable factor analysis network model as factors that limit the vehicle perception system to the perception target, and according to each conditional probability in the conditional probability table, calculate the false positive rate or false negative rate of the vehicle perception system to the perception target under different states of each factor node.

2. The method according to claim 1, characterized in that The determining of the expected number of simultaneous occurrences of any state of the factor node and any state of the performance restriction node in the training sample set includes: Counting the number of occurrences of any state of the factor node in the training sample set; Counting the number of occurrences of any state of the performance-limiting node in the training sample set; According to the number of occurrences of any state of the factor node in the training sample set, the number of occurrences of any state of the performance constraint node in the training sample set, and the number of training samples in the training sample set, the expected number of simultaneous occurrences of any state of the factor node and any state of the performance constraint node in the training sample set is calculated, the expected number of simultaneous occurrences is positively correlated to the number of occurrences of any state of the factor node in the training sample set, the number of occurrences of any state of the performance constraint node in the training sample set, and the expected number of simultaneous occurrences is negatively correlated to the number of training samples.

3. The method according to claim 2, characterized in that The calculating the chi-square statistic according to the plurality of observed frequencies and the plurality of expected frequencies comprises: The chi-square statistic is calculated by the following formula: Among them, X 2 is the chi-square statistic, O ij is the observation frequency corresponding to the i-th state of the factor node and the j-th state of the performance constraint node, E ij is the expected frequency corresponding to the i-th state of the factor node and the j-th state of the performance constraint node, r is the number of states of the factor node, and c is the number of states of the performance constraint node.

4. The method according to any one of claims 1 to 3, characterized in that The updating of the factor analysis network model and the conditional probability table according to the chi-square statistic comprises: Searching for a test result parameter corresponding to the chi-square statistic from a mapping table, and determining a target hypothesis from two mutually exclusive hypotheses according to the test result parameter; If the target hypothesis indicates that there is no association relationship between the factor node and the performance constraint node, the factor node and the directed edge connecting the factor node are deleted from the factor analysis network model, and the conditional probability related to the factor node is deleted from the conditional probability table.

5. The method according to any one of claims 1 to 3, characterized in that The extracting factor nodes from the stable factor analysis network model as factors limiting the vehicle perception system to the perception target includes: The verification sample set is verified by the factor analysis network model to obtain a verification result, wherein the verification result includes: the number of first samples correctly predicted as positive in the verification sample set, the number of second samples correctly predicted as negative in the verification sample set, the number of third samples incorrectly predicted as positive in the verification sample set, and the number of fourth samples incorrectly predicted as negative in the verification sample set; when the performance limiting node is used to indicate the false negative rate of the perception target, the positive class is a false negative, and the negative class is a non-false negative; when the performance limiting node is used to indicate the false positive rate of the perception target, the positive class is a false positive, and the negative class is a non-false positive; Determine the accuracy and / or recall of the factor analysis network model according to the verification result, wherein the accuracy is positively correlated to the sum of the first sample number and the second sample number, the accuracy is negatively correlated to the sum of the first sample number, the second sample number, the third sample number and the fourth sample number, and the recall is positively correlated to the first sample number and negatively correlated to the sum of the first sample number and the fourth sample number; If the accuracy is greater than or equal to a preset accuracy threshold, and / or the recall is greater than or equal to a preset recall threshold, factor nodes are extracted from the stable factor analysis network model as factors for limiting the vehicle perception system to the perception target.

6. The method according to any one of claims 1 to 3, characterized in that The updating of the factor analysis network model and the conditional probability table according to the chi-square statistic until the factor analysis network model is stable includes: updating the factor analysis network model and the conditional probability table according to the chi-square statistic; When receiving the target factor selected from the training sample set, adding the target factor to the factor analysis network model, and returning to the step of statistically calculating the conditional probability table of the factor analysis network model from the training sample set; When the target factor selected from the training sample set is not received and each of the factor nodes in the factor analysis network model has been verified, it is determined that the factor analysis network model is stable, and the verification of the factor node indicates that it has been verified whether the factor node has an impact on the perception target.

7. A device for identifying factors that limit vehicle perception performance, characterized in that: include: An acquisition module is used to construct a factor analysis network model for any perception target of a vehicle perception system, wherein the factor analysis network model includes a plurality of nodes and a plurality of directed edges, each of the directed edges connects two of the nodes, the two nodes include a parent node and a child node, the directed edge points from the parent node to the child node, and the plurality of nodes include: a factor node affecting the perception target, a performance limitation node corresponding to the perception target, and the performance limitation node is used to indicate a false negative rate or a false positive rate of the perception target; The acquisition module is further used to acquire a training sample set for the factor analysis network model, wherein the training sample set includes a plurality of training samples, and each training sample includes the state of each node in the factor analysis network model; A processing module, configured to generate a conditional probability table of the factor analysis network model from the training sample set, wherein the conditional probability table includes a probability that the child node is in a second state when the parent node of the child node is in a first state; The processing module is further used to determine the actual number of simultaneous occurrences of any state of the factor node and any state of the performance constraint node in the training sample set as the observed frequency, and to determine the expected number of simultaneous occurrences of any state of the factor node and any state of the performance constraint node in the training sample set as the expected frequency, and to calculate a chi-square statistic based on a plurality of the observed frequencies and a plurality of the expected frequencies; An execution module is used to update the factor analysis network model and the conditional probability table according to the chi-square statistic until the factor analysis network model is stable, so as to extract factor nodes from the stable factor analysis network model as factors that limit the vehicle perception system to the perception target, and calculate the false positive rate or false negative rate of the vehicle perception system to the perception target under different states of each factor node according to each conditional probability in the conditional probability table.

8. An electronic device, characterized in that: include: Memory, processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory, so that the processor performs the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 6 when executed by a processor.

10. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 6 when being executed by a processor.

Images