Method for evaluating interactive cognitive ability of unmanned driving
By determining the interactive cognitive dimension set and calculating the comprehensive score in the evaluation method of unmanned driving interactive cognitive ability, the problem of multi-index evaluation in complex environments in the existing technology is solved, and the accuracy of unmanned driving interactive cognitive ability is achieved. Improvement of assessment and system generalization capabilities.
Patent Information
- Application Number
- CN202411017578.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-29
- Publication Date
- 2024-11-15
AI Technical Summary
The existing unmanned interactive cognitive ability evaluation method is difficult to accurately evaluate the correlation and nonlinear characteristics between multiple indicators in complex environments, resulting in weak generalization ability in complex traffic environments and high dependence on hardware performance. .
A method is proposed to calculate the comprehensive evaluation score by determining the dimension set of unmanned driving interaction cognitive ability, including standardized data matrix, data volatility and conflict in calculation indicators, information volume and weight, and calculation based on the positive and negative ideal solution gap Comprehensive scores can take into account the intrinsic correlation and nonlinear characteristics between indicators.
It realizes effective evaluation of driverless interactive cognitive capabilities, can accurately evaluate multi-indicator capabilities in complex environments, improves the generalization capabilities of the system, and reduces the dependence on hardware performance.
Smart Images

Figure CN118966883A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of unmanned driving, and in particular to a method for evaluating the interactive cognitive ability of unmanned driving. Background Art
[0002] Unmanned driving technology is changing people's travel, transportation, and even production and living styles. Unmanned driving is inseparable from the interaction between cars and people, cars and cars, and cars and roads. However, in actual scenarios, it often faces challenges such as multiple perception sources, heterogeneous data, severe weather and other complex environments and semantic diversity, which leads to people's unclear understanding of unmanned driving behavior and its decisions, and it is difficult for it to reach an agreement with our intentions. Therefore, it is particularly important to achieve unmanned driving interactive cognition and accurately evaluate its interactive cognitive ability in complex and uncertain environments. In terms of perceptual intelligence, it includes the following dimensions: the ability of unmanned driving vehicle cameras, radars, navigation and other sensors to obtain perceptual data; the ability to align multi-channel cross-modal data; the ability to select attention mechanisms for perceptual information; and the ability to recognize and understand visual information. In terms of behavioral intelligence, it includes the following dimensions: the ability to control embodied power such as steering, braking and acceleration of the vehicle itself; the ability to interact with the body language of various people; the ability to interact with the body language of cars; and the ability to interact with the environment. How to effectively evaluate the above multi-index capability model has become a key issue to be solved in unmanned driving interactive cognition. There are many evaluation methods currently available, and representative methods include TOPSIS and CRITIC.
[0003] TOPSIS (Technique for Order Preference by Similarity to an Ideal Solution) method, also known as the superior and inferior solution distance method, was proposed by CL Hwang and K. Yoon. It is a method for evaluating and ranking based on the degree of proximity between a limited number of evaluation objects and the idealized goal. This method can make full use of the original data information, and its results can accurately reflect the gap between the objects of the capability model. It has no strict restrictions on data distribution and sample content, and the data calculation is simple and easy. However, the TOPSIS method assumes that the weights of each indicator are the same, but in actual situations, the importance of each indicator is not consistent, resulting in the traditional TOPSIS method not being able to evaluate objects with multiple indicators well. The TOPSIS method based on entropy weight can calculate the weights of each indicator based on the degree of dispersion of data in each indicator and use information entropy to complete the multi-indicator comprehensive evaluation task. In the unmanned driving interactive cognitive capability model, there is correlation between the indicators. The CRITIC method can fully reflect the correlation between the indicators. The CRITIC method is an objective weight assignment method based on data volatility proposed by Diakoulaki et al. This method combines the correlation coefficient and the information weight. It reflects the amount of information based on the comparison strength of the evaluation indicators and the conflict between the indicators. The greater the comparison strength and the conflict, the more information is reflected, and the greater the corresponding indicator weight. The traditional CRITIC method based on the Pearson correlation coefficient can effectively measure the linear correlation between indicators. The evaluation of the unmanned driving interactive cognitive ability model is a complex nonlinear process, and the calculation and statistical methods of the indicators are not the same. Therefore, the traditional CRITIC method based on the Pearson correlation coefficient is difficult to accurately evaluate the unmanned driving interactive cognitive ability.
[0004] The Chinese invention patent application with application publication number CN103335853A discloses a computer-based unmanned vehicle cognitive ability test system, which includes: a solar power supply module, a system self-check module, a test instruction input module, a single-chip control module, a test module, a test storage module, and a test display module. The disadvantage of this method is that it does not explain in detail how to build a systematic standard test environment that comprehensively covers all possible traffic conditions, resulting in weak generalization ability in complex environments and under different conditions.
[0005] The Chinese invention patent application with application publication number CN108563169A discloses a system and method for testing cognitive ability of unmanned vehicles, the system includes an unmanned vehicle and an on-board recording device, the two of which are connected by wire through a communication interface, wherein the on-board recording device is used to input cognitive ability test tasks to the unmanned vehicle, and receive and store data output by the cognitive ability test of the unmanned vehicle; after the unmanned vehicle receives the test task through the test task input program, it re-establishes the serial port connection, starts the corresponding program to output the test data to the on-board recording device, and the test task input and test data output of the unmanned vehicle cognitive ability test adopt a unified interface, parameters and data transmission format, but the specific format of the data content is different; the communication interface adopts an RS-232 interface. The disadvantage of this method is that it cannot adapt to different weather conditions and complex actual traffic environments, and it has a high degree of dependence on hardware, and the performance limitations of the hardware may affect the performance of the entire system. Summary of the invention
[0006] In order to solve the above technical problems, the present invention proposes a method for evaluating the interactive cognitive ability of unmanned driving, which can simultaneously consider the intrinsic correlation and nonlinear characteristics between indicators, and can effectively evaluate the interactive cognitive ability of unmanned driving in complex, multi-indicator situations.
[0007] The present invention provides a method for evaluating the interactive cognitive ability of an unmanned driving vehicle, comprising determining a set of dimensions of the interactive cognitive ability of an unmanned driving vehicle, and further comprising the following steps:
[0008] Step 1: Determine the set of objects for evaluating the interactive cognitive ability of unmanned driving;
[0009] Step 2: Calculate the comprehensive evaluation score C of autonomous driving interaction cognition.
[0010] Preferably, the unmanned driving interactive cognitive ability dimension set U is
[0011] U={D,S,L,A,C,P,T,I}
[0012] Among them, D is the ability dimension of the unmanned vehicle's cameras, radars, navigation and other sensors to obtain perception data, S is the multi-channel cross-modal data registration ability dimension, L is the visual recognition and understanding ability dimension, A is the selective attention mechanism ability dimension for perceived information, C is the embodied power control ability dimension for steering, braking and acceleration of the vehicle itself, P is the body language interaction cognitive ability dimension between the vehicle and various types of people, T is the body language interaction cognitive ability dimension between vehicles, and I is the collaborative interaction cognitive ability dimension between the vehicle and the environment. Each dimension includes a varying number of indicators.
[0013] In any of the above solutions, preferably, the unmanned driving interactive cognitive ability evaluation object set V is
[0014] V={V 1 ,V 2 ,V 3 ,…,V m}
[0015] Among them, V 1 ~V m are the m evaluation objects in the unmanned driving interactive cognitive ability evaluation object set.
[0016] In any of the above schemes, preferably, step 2 comprises the following sub-steps:
[0017] Step 21: According to the m evaluation objects and n indicators, form the original data matrix R;
[0018] Step 22: Standardize the initial data matrix R;
[0019] Step 23: Calculate the volatility S of the data within each indicator j ;
[0020] Step 24: Calculate the conflict between indicators;
[0021] Step 25: Calculate the information content D of each indicator j ;
[0022] Step 26: Calculate the indicator weight w based on the information content of each indicator j ;
[0023] Step 27: Based on the standardized matrix R' and the weights of each indicator w j Get the weighted normalization matrix Z;
[0024] Step 28: Calculate the comprehensive score C of each evaluation object based on the gap between each indicator and the positive and negative ideal solutions i .
[0025] In any of the above solutions, preferably, the formula of the original data matrix R is:
[0026]
[0027] Among them, r ij Represents the original data of the jth indicator of the ith evaluation object, m is the number of evaluation objects, and n is the number of evaluation indicators.
[0028] In any of the above solutions, preferably, the formula for normalizing the initial data matrix R is:
[0029]
[0030] Among them, r'ij is the element in the matrix after normalization, min(r ij ) is the smallest element in the original data matrix, max(r ij ) is the largest element in the original data matrix.
[0031] In any of the above solutions, preferably, the volatility S of the data within each indicator j The calculation formula is
[0032]
[0033] Among them, mean(r' j ) is the mean value of each indicator column data after standardization.
[0034] In any of the above solutions, preferably, step 24 includes the following sub-steps:
[0035] Step 241: Calculate the correlation coefficient ρ between indicators tj ;
[0036] Step 242: Calculate the conflict B between indicators j .
[0037] In any of the above schemes, preferably, the correlation coefficient ρ between the indicators is tj The calculation formula is
[0038]
[0039] Among them, r it is the level of the data corresponding to the tth indicator of the i-th evaluation object in the standardized matrix, r” t is the level of the data in the tth row of the standardized matrix, r" j is the rank of the data in the jth column of the standardized matrix, r” ij is the level of the data corresponding to the jth indicator of the i-th evaluation object in the standardized matrix, mean(r” t ) is the average level of the t-th row data after standardization, mean(r” j ) is the average rank of the data in the jth column.
[0040] In any of the above schemes, preferably, the conflict between the indicators B j The calculation formula is
[0041]
[0042] Among them, t is the number of rows in the correlation coefficient matrix, and h is the total number of rows in the correlation coefficient matrix.
[0043] In any of the above solutions, preferably, the information amount D of each indicator isj The calculation formula is
[0044] D j =S j ·B j .
[0045] In any of the above solutions, preferably, the indicator weight w j The calculation formula is
[0046]
[0047] In any of the above solutions, preferably, the calculation formula of the weighted normalization matrix Z is:
[0048] Z=w j R′.
[0049] In any of the above schemes, preferably, the comprehensive score C of each evaluation object is i The calculation formula is
[0050]
[0051] Among them, max(z 1j ,z 2j ,…z mj ) is the positive ideal solution for each index, min(z 1j ,z 2j ,…z mj ) is the negative ideal solution of each indicator.
[0052] The present invention proposes a method for evaluating the interactive cognitive ability of unmanned driving, which realizes the evaluation of the interactive cognitive ability of different algorithms, thereby further analyzing the interactive cognitive ability of unmanned driving. BRIEF DESCRIPTION OF THE DRAWINGS
[0053] Figure 1 The present invention is a flowchart of a preferred embodiment of a method for evaluating the interactive cognitive ability of an unmanned driving system according to the present invention.
[0054] Figure 2 The present invention is a flowchart of an embodiment of the comprehensive evaluation of unmanned driving interaction cognition according to the method for evaluating unmanned driving interaction cognition ability of the present invention. DETAILED DESCRIPTION
[0055] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.
[0056] Embodiment 1
[0057] like Figure 1As shown, step 1000 is executed to determine the unmanned driving interaction cognitive capability dimension set, wherein the unmanned driving interaction cognitive capability dimension set U is
[0058] U={D,S,L,A,C,P,T,I}
[0059] Among them, D is the ability dimension of the unmanned vehicle's cameras, radars, navigation and other sensors to obtain perception data, S is the multi-channel cross-modal data registration ability dimension, L is the visual recognition and understanding ability dimension, A is the selective attention mechanism ability dimension for perceived information, C is the embodied power control ability dimension for steering, braking and acceleration of the vehicle itself, P is the body language interaction cognitive ability dimension between the vehicle and various types of people, T is the body language interaction cognitive ability dimension between vehicles, and I is the collaborative interaction cognitive ability dimension between the vehicle and the environment. Each dimension includes a varying number of indicators.
[0060] Execute step 1100 to determine a set of unmanned driving interaction cognitive ability evaluation objects, wherein the set V of unmanned driving interaction cognitive ability evaluation objects is
[0061] V={V 1 ,V 2 ,V 3 ,…,V m}
[0062] Among them, V 1 ~V m are the m evaluation objects in the unmanned driving interactive cognitive ability evaluation object set.
[0063] Execute step 1200 to calculate the comprehensive evaluation score C of the autonomous driving interaction cognition, such as Figure 2 As shown, it includes the following sub-steps:
[0064] Execute step 1210, and form the original data matrix R according to m evaluation objects and n indicators, and the formula is
[0065]
[0066] Among them, r ij Represents the original data of the jth indicator of the ith evaluation object, m is the number of objects to be evaluated, and n is the number of evaluation indicators.
[0067] Execute step 1220 to standardize the initial data matrix R. The formula is:
[0068]
[0069] Among them, r' ij is the element in the matrix after normalization, min(r ij) is the smallest element in the original data matrix, max(r ij ) is the largest element in the original data matrix.
[0070] Execute step 1230 to calculate the volatility S of the data in each indicator j , the calculation formula is
[0071]
[0072] Among them, mean(r' j ) is the mean of each indicator column data after standardization, and n represents the number of indicators.
[0073] Execute step 1240 to calculate the conflict between indicators, including the following sub-steps:
[0074] Execute step 1241 to calculate the correlation coefficient ρ between the indicators tj , the calculation formula is
[0075]
[0076] Among them, r it is the level of the data corresponding to the tth indicator of the i-th evaluation object in the standardized matrix, r” t is the level of the data in the tth row of the standardized matrix, r" j is the rank of the data in the jth column of the standardized matrix, r” ij is the level of the data corresponding to the jth indicator of the i-th evaluation object in the standardized matrix, mean(r” t ) is the average level of the t-th row data after standardization, mean(r” j ) is the average rank of the data in the jth column.
[0077] Execute step 1242 to calculate the conflict B between the indicators j , the calculation formula is
[0078]
[0079] Among them, t is the number of rows in the correlation coefficient matrix, and h is the total number of rows in the correlation coefficient matrix.
[0080] Execute step 1250 to calculate the information amount D of each indicator j , the calculation formula is
[0081] D j =S j ·B j .
[0082] Execute step 1260 to calculate the indicator weight w according to the information content of each indicator j , the calculation formula is
[0083]
[0084] Execute step 1270, based on the standardized matrix R' and the weights w of each indicator j Get the weighted normalization matrix Z, the calculation formula is
[0085] Z=w j R′.
[0086] Execute step 1280 to calculate the comprehensive score C of each evaluation object based on the gap between each indicator and the positive and negative ideal solutions. i , the calculation formula is
[0087]
[0088] Among them, max(z 1j ,z 2j ,…z mj ) is the positive ideal solution for each index, min(z 1j ,z 2j ,…z mj ) is the negative ideal solution of each indicator.
[0089] Embodiment 2
[0090] The research purpose of this invention is to achieve effective evaluation of multi-index, nonlinear unmanned driving interactive cognitive ability. In view of the limitations of the above-mentioned popular index evaluation methods in evaluating the intrinsic correlation between multiple indicators and the difficulty in applying to complex nonlinear models, the method of this invention can simultaneously consider the intrinsic correlation and nonlinear characteristics between indicators, and can effectively evaluate the complex, multi-index unmanned driving interactive cognitive ability.
[0091] In the process of autonomous driving interactive cognition, autonomous driving interactive cognition based on the Wiener model forms a situation graph cluster, including a perception graph cluster and a behavior graph cluster. The interactive cognition capability model forms an octet based on the situation graph cluster, and the interactive cognition multi-dimensional capability evaluation method is as follows:
[0092] 1. Determine the set of dimensions of autonomous driving interactive cognitive capabilities.
[0093] U={D,S,L,A,C,P,T,I} (Formula 1)
[0094] In the formula, U is the set of interactive cognitive ability dimensions of unmanned vehicles. D is the ability dimension of sensors such as cameras, radars and navigation of unmanned vehicles to obtain perception data; S is the ability dimension of multi-channel cross-modal data registration; L is the visual recognition and understanding ability dimension; A is the ability dimension of the selective attention mechanism for perception information; C is the ability dimension of embodied dynamic control such as steering, braking and acceleration of the vehicle itself; P is the cognitive ability dimension of the body language interaction between the vehicle and various people; T is the cognitive ability dimension of the body language interaction between vehicles; I is the cognitive ability dimension of the collaborative interaction between the vehicle and the environment; each dimension includes different numbers of indicators.
[0095] 2. Determine the set of objects for evaluation of autonomous driving interactive cognitive capabilities.
[0096] V={V 1 ,V 2 ,V 3 ,…,V m} (Formula 2)
[0097] Where V 1~ V m It is the m evaluation objects of the octet set of autonomous driving interactive cognitive capabilities.
[0098] 3. Calculate the comprehensive evaluation score C of autonomous driving interaction cognition.
[0099] (1) Based on m evaluation objects and n indicators, the original data matrix R is formed.
[0100]
[0101] The i-th row and j-th column in R represent the original data of the j-th indicator of the i-th evaluation object. The value range of i is [1, m], and the value range of j is [1, n].
[0102] (2) For the initial data matrix R = (r ij ) m×n Perform standardization.
[0103]
[0104] (3) Calculate the volatility S of the data within each indicator j .
[0105]
[0106] Among them, mean(r' j ) is the mean value of each indicator (column) data after standardization, S j represents the standard deviation. n represents the number of indicators.
[0107] (4) Calculate the conflicts between indicators.
[0108] ① Calculate the correlation coefficient ρ between indicators tj .
[0109]
[0110] where r ij is the level (rank) of the data corresponding to the jth indicator of the i-th evaluation object in the standardized matrix, mean(r” t ) and mean(r” j ) represent the average rank of the data in the tth row and jth column after standardization, ρ tj The value range is [-1,1].
[0111] ② Calculate the conflict between indicators B j
[0112]
[0113] (5) Calculate the information amount D of each indicator j .
[0114] D j =S j ·B j (Formula 8)
[0115] (6) Calculate the indicator weight w according to the information content of each indicator j .
[0116]
[0117] (7) The standardized matrix R′ and the weights of each indicator w j Get the weighted normalization matrix Z.
[0118] Z=w j R′ (Formula 10)
[0119] (8) Calculate the comprehensive score C of each evaluation object based on the gap between each indicator and the positive and negative ideal solutions i .
[0120]
[0121] Among them, C i The value range is [0,1], max(z 1j ,z 2j ,…z mj ) is the positive ideal solution for each index, min(z 1j ,z 2j ,…z mj ) is the negative ideal solution of each indicator.
[0122] The comprehensive score obtained by calculation realizes the evaluation of the interactive cognitive ability of different algorithms, which can further analyze the interactive cognitive ability of unmanned driving.
[0123] Embodiment 3
[0124] During the driving process of unmanned vehicles, pedestrian postures, traffic police gestures, etc. are typical ways of interaction between people and vehicles. It is necessary to identify and understand people's behavioral intentions, and then make decisions such as parking and turning, so as to realize the body language interaction between vehicles and people. Therefore, the body language interaction between vehicles and people has become one of the key issues in the field of unmanned driving. Taking the body language interaction cognition between vehicles and people as an example, the present invention comprehensively evaluates and analyzes the performance of the current five mainstream behavior recognition algorithms: multi-view spatial-temporal graph convolution network (MV-STGCN), directed graph neural network (DGNN), distillation-trained spatial-temporal graph convolution model (DGCNNet), pose refinement graph convolution network (PR-GCN) and spatial temporal graph convolution network (ST-GCN). The results of the above five methods tested on the NTU RGB+D dataset are shown in Table 1.
[0125] Table 1 Comparison of different methods on the NTU RGB+D dataset
[0126]
[0127] There are 8 dimensions in the evaluation of body language interaction between vehicles and people: {D}, {S}, {A}, {L}, {C}, {P}, {V} and {I}, each of which contains different indicators. Here, 12 indicators in different dimensions are selected to conduct a comprehensive evaluation of different algorithms. The specific evaluation steps are as follows:
[0128] 1. Determine the cognitive ability dimension set of body language interaction between vehicles and people.
[0129] U={u 1 ,u 2 ,u 3 ,u 4 ,u5 ,u 6 ,u 7 ,u 8 ,u 9 ,u 10 ,u 11 ,u 12}
[0130] where u 1 to u 12 They are point cloud acquisition density, number of acquired image frames, average precision of multi-channel cross-modal data matching, absolute error of multi-channel cross-modal data matching, average precision of multi-channel cross-source data matching, absolute error of multi-channel cross-source data matching, number of spatiotemporal network model parameters, image quality assessment signal-to-noise ratio, image quality assessment mean square error, image recognition accuracy, image recognition recall rate, pedestrian action recognition accuracy, etc. Among these 12 evaluation indicators, u 1 and u 2 Belongs to dimension D; u 3 ,u 4 ,u 5 and u 6 Belongs to dimension S; u 7 ,u 8 ,u 9 ,u 10 ,u 11 and u 12 Belongs to dimension L.
[0131] D is the ability dimension of the unmanned vehicle's cameras, radars, navigation and other sensors to obtain perception data, S is the multi-channel cross-modal data registration ability dimension, L is the visual recognition and understanding ability dimension, A is the selective attention mechanism ability dimension for perception information, C is the embodied power control ability dimension for steering, braking and acceleration of the vehicle itself, P is the body language interaction cognitive ability dimension between the vehicle and various types of people, T is the body language interaction cognitive ability dimension between vehicles, and I is the collaborative interaction cognitive ability dimension between the vehicle and the environment.
[0132] 2. Determine the set of evaluation objects for different algorithms.
[0133] V={v 1 ,v 2 ,v 3 ,v 4 ,v 5}
[0134] Among them, v 1 to v 5They correspond to MV-STGCN, DGNN, DGCNNet, PR-GCN, and ST-GCN respectively. MV-STGCN is a multi-view spatial-temporal graph convolution network;
[0135] DGNN is a directed graph neural network.
[0136] DGCNNet is a distillation-trained spatial-temporal graph convolution model.
[0137] PR-GCN is a pose refinement graph convolutional network;
[0138] ST-GCN is a spatial temporal graph convolutional network.
[0139] 3. Different algorithms obtain the original data matrix R according to the indicator set
[0140]
[0141] 4. Standardize the original data matrix, calculate the volatility S and conflict B of each indicator, and then get the information content DS of each indicator = (0.442 0.389 0.368 0.384 0.393 0.386 0.382 0.369 0.383 0.360 0.359 0.358)
[0142] B=(6.721 6.440 8.064 8.025 6.352 7.977 8.028 6.339 8.0356.499 6.4656.401)
[0143] D=(2.971 2.511 2.972 3.084 2.497 3.081 3.065 2.343 3.080 2.340 2.325 2.295)
[0145] 5. Calculate the weight matrix W of the body language interaction cognitive ability dimension between vehicles and humans based on the amount of information of each indicator.
[0146] W=(0.091 0.077 0.092 0.094 0.076 0.095 0.094 0.072 0.095 0.071 0.069 0.070)
[0148] 6. Calculate the comprehensive evaluation score C of the performance of different algorithms for the 12 indicators of the 5 algorithms. C = (0.2750.235 0.199 0.151 0.140)
[0149] After experimental verification, the comprehensive evaluation score of MV-STGCN is 0.275, which is significantly higher than the other four behavior recognition algorithms. Compared with the results tested on the NTU RGB+D dataset, the effectiveness of the evaluation method is verified.
[0150] The NTU RGB+D dataset is a public dataset for action recognition based on RGB information and depth information collected and produced by Nanyang Technological University (NTU).
[0151] In order to better understand the present invention, the above is described in detail in conjunction with the specific embodiments of the present invention, but it is not intended to limit the present invention. Any simple modifications made to the above embodiments based on the technical essence of the present invention still fall within the scope of the technical solution of the present invention. Each embodiment in this specification focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referenced to each other. For the system embodiment, since it basically corresponds to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.
Claims
1. A method for evaluating the interactive cognitive ability of an unmanned driving vehicle, comprising determining a set of dimensions of the interactive cognitive ability of an unmanned driving vehicle, characterized in that: The following steps are also included: Step 1: Determine the set of objects for evaluating the interactive cognitive ability of unmanned driving; Step 2: Calculate the comprehensive evaluation score C of autonomous driving interaction cognition.
2. The method for evaluating the interactive cognitive ability of unmanned driving according to claim 1, characterized in that: The unmanned driving interactive cognitive ability dimension set U is U={D,S,L,A,C,P,T,I} Among them, D is the ability dimension of the unmanned vehicle's cameras, radars, navigation and other sensors to obtain perception data, S is the multi-channel cross-modal data registration ability dimension, L is the visual recognition and understanding ability dimension, A is the selective attention mechanism ability dimension for perceived information, C is the embodied power control ability dimension for steering, braking and acceleration of the vehicle itself, P is the body language interaction cognitive ability dimension between the vehicle and various types of people, T is the body language interaction cognitive ability dimension between vehicles, and I is the collaborative interaction cognitive ability dimension between the vehicle and the environment. Each dimension includes a varying number of indicators.
3. The method for evaluating the interactive cognitive ability of unmanned driving according to claim 2, characterized in that: The unmanned driving interactive cognitive ability evaluation object set V is V={V1,V2,V3,…,V m } Among them, V1~V m are the m evaluation objects in the unmanned driving interactive cognitive ability evaluation object set.
4. The method for evaluating the interactive cognitive ability of unmanned driving according to claim 3, characterized in that: The step 2 includes the following sub-steps: Step 21: According to the m evaluation objects and n indicators, form the original data matrix R; Step 22: Standardize the initial data matrix R; Step 23: Calculate the volatility S of the data within each indicator j ; Step 24: Calculate the conflict between indicators; Step 25: Calculate the information content D of each indicator j ; Step 26: Calculate the indicator weight w based on the information content of each indicator j ; Step 27: Based on the standardized matrix R' and the weights of each indicator w j Get the weighted normalization matrix Z; Step 28: Calculate the comprehensive score C of each evaluation object based on the gap between each indicator and the positive and negative ideal solutions i .
5. The method for evaluating the interactive cognitive ability of unmanned driving according to claim 4, characterized in that: The formula of the original data matrix R is Among them, r ij Represents the original data of the jth indicator of the ith evaluation object, m is the number of evaluation objects, and n is the number of evaluation indicators.
6. The method for evaluating the interactive cognitive ability of unmanned driving according to claim 5, characterized in that: The formula for standardizing the initial data matrix R is: Among them, r' ij is the element in the matrix after normalization, min(r ij ) is the smallest element in the original data matrix, max(r ij ) is the largest element in the original data matrix.
7. The method for evaluating the interactive cognitive ability of unmanned driving according to claim 6, characterized in that: The volatility of the data within each indicator S j The calculation formula is Among them, mean(r' j ) is the mean value of each indicator column data after standardization.
8. The method for evaluating the interactive cognitive ability of unmanned driving according to claim 7, characterized in that: The step 24 includes the following sub-steps: Step 241: Calculate the correlation coefficient ρ between indicators tj ; Step 242: Calculate the conflict B between indicators j .
9. The method for evaluating the interactive cognitive ability of unmanned driving according to claim 8, characterized in that: The correlation coefficient ρ between the indicators tj The calculation formula is Among them, r it is the level of the data corresponding to the tth indicator of the i-th evaluation object in the standardized matrix, r” t is the level of the data in the tth row of the standardized matrix, r" j is the rank of the data in the jth column of the standardized matrix, r” ij is the level of the data corresponding to the jth indicator of the i-th evaluation object in the standardized matrix, mean(r” t ) is the average level of the t-th row data after standardization, mean(r” j ) is the average rank of the data in the jth column.
10. The method for evaluating the interactive cognitive ability of unmanned driving according to claim 9, characterized in that: Conflict between the indicators B j The calculation formula is Among them, t is the number of rows in the correlation coefficient matrix, and h is the total number of rows in the correlation coefficient matrix.
Citation Information
Cited By
Method for representing interactive cognitive ability of unmanned vehicle
CN121341181A
Evaluation method for interactive cognitive ability of unmanned vehicle
CN121580644A
Interaction cognitive ability evaluation method for unmanned driving
CN121580646A
