Automatic driving performance evaluation method based on scene complexity quantification

Abstract

The invention discloses an automatic driving performance evaluation method based on scene complexity quantification, and the method comprises the following steps: carrying out an automatic driving test in an automatic driving scene, obtaining a driving implementation scene image under the automatic driving test, and obtaining an automatic driving test result under the automatic driving test; obtaining a standard driving scene image, and obtaining target scene complexity according to the real-time driving scene image and the standard driving scene image; obtaining an automatic driving performance evaluation result based on the automatic driving test result and the target scene complexity; by quantifying the complex scene, the quantification result of the complex scene can be considered when the automatic driving test result is evaluated, so that a more accurate automatic driving performance evaluation result is obtained.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and in particular to an autonomous driving performance evaluation method based on scenario complexity metric. Background Technology

[0002] Scene complexity quantification refers to the decomposition and quantification of the environment in which autonomous vehicles operate, such as quantifying road conditions and weather conditions, and assigning a certain complexity score to each factor; autonomous driving performance evaluation is the assessment of the vehicle's performance in such scenarios.

[0003] Because autonomous driving systems interact dynamically with scene elements (such as roads and environmental conditions) in real time, the more complex the scene elements, the higher the complexity of the driving scene. This poses greater challenges to the autonomous driving system in terms of perception and execution, potentially leading to a higher collision rate and a greater impact on driving performance. Traditional static scene-based driving performance evaluations rely on weak interaction indicators in typical environments; that is, the scenes are relatively simple and cannot assess driving performance in complex scenarios, thus having certain limitations. Summary of the Invention

[0004] In view of the above-mentioned prior art, the present invention provides an autonomous driving performance evaluation method based on scenario complexity quantification, which mainly solves the technical problems existing in the background art.

[0005] To achieve the above objectives, the technical solution of this invention is implemented as follows: An autonomous driving performance evaluation method based on scenario complexity measurement includes the following steps: Autonomous driving tests are conducted in autonomous driving scenarios to obtain images of the actual driving scenarios and the results of the autonomous driving tests. Acquire a standard driving scene image, and obtain the target scene complexity based on the real-time driving scene image and the standard driving scene image; Based on the autonomous driving test results and the complexity of the target scenario, the autonomous driving performance evaluation results are obtained.

[0006] Optionally, obtaining the autonomous driving test results under autonomous driving testing includes: In the aforementioned autonomous driving scenario, three consecutive autonomous driving tests of the same route were performed. Record the test results of the same trip three times, and calculate the average value based on the test results of the three trips to obtain the autonomous driving test result.

[0007] Optionally, the step of acquiring a standard driving scene image and obtaining the target scene complexity based on the real-time driving scene image and the standard driving scene image includes: A knowledge graph is constructed by acquiring standard driving scene images and environmental data information synchronized with the standard driving scene images, resulting in a standard driving scene graph-standard semantic vector-standard scene complexity triplet library; Semantic recognition is performed on the real-time driving scene images to obtain real-time semantic vectors; The Euclidean distance between the real-time semantic vector and the standard semantic vector of each standard driving scene image is calculated sequentially to obtain multiple calculation results; The multiple calculation results are sorted from low to high, and the driving scene in the standard driving scene image with the lowest calculation result is taken as the target driving scene, and the corresponding target scene complexity is obtained.

[0008] Optionally, the acquisition of standard driving scene images and environmental data information synchronized with the standard driving scene images to construct a knowledge graph, resulting in a standard driving scene graph-standard semantic vector-standard scene complexity triplet library, including: Create scene nodes for each standard driving scenario image; Entities are identified and relationships between entities are mined based on environmental data information synchronized with the standard driving scenario image. The confidence level of the entity relationships is determined, and the edge weights are obtained based on the confidence level of the entity relationships to complete the construction of the knowledge graph. Semantic segmentation and attribute extraction are performed on each standard driving scene image to obtain a standard semantic vector; Using scene nodes as the granularity, the confidence of the entity relationships associated with the scene nodes is weighted and summed to obtain the standard scene complexity of the scene node, thereby obtaining a standard driving scene graph-standard semantic vector-standard scene complexity triplet library.

[0009] Optionally, the step of determining entities based on environmental data information synchronized with the standard driving scene image, mining relationships between entities, determining the confidence level of entity relationships, obtaining edge weights based on the confidence level of entity relationships, and completing the construction of the knowledge graph includes: The environmental data information includes weather environmental data information, road environmental data information, and traffic environmental data information; Multiple weather entities are determined based on the weather environment data, multiple road entities are determined based on the road environment data, and multiple traffic entities are determined based on the traffic environment data. The weather entities, road entities, and traffic entities are then connected to their respective scene nodes. The Bayesian algorithm was used to mine relationships between three types of entity pairs: weather-road, weather-traffic, and road-traffic, to obtain the first entity relationship, the second entity relationship, and the third entity relationship. The first entity relationship, the second entity relationship, and the third entity relationship are merged, and duplicate entity relationships are removed to obtain multiple candidate entity relationships. Bayesian posterior probabilities are assigned to the multiple candidate entity relationships as confidence levels. After normalizing the confidence level, the edge weights are obtained. Directed edges are then established between the weather-road-traffic entity nodes to complete the knowledge graph construction.

[0010] Optionally, the step of sequentially calculating the Euclidean distance between the real-time semantic vector and the standard semantic vector of the standard driving scene image to obtain the calculation result includes: Calculate the weather Euclidean distance between the real-time semantic vector and the standard semantic vector to obtain the first calculation result; Calculate the road Euclidean distance between the real-time semantic vector and the standard semantic vector to obtain a second calculation result; Calculate the traffic Euclidean distance between the real-time semantic vector and the standard semantic vector to obtain a third calculation result; The calculation result is obtained by weighted summation of the first calculation result, the second calculation result, and the third calculation result.

[0011] Optionally, the process of obtaining the autonomous driving performance evaluation result based on the autonomous driving test results and the complexity of the target scenario includes: Based on the autonomous driving test results, construct an evaluation index set and an evaluation result set; Based on the evaluation index set, the evaluation result set, and the target scene complexity, a fuzzy evaluation matrix is ​​constructed. The evaluation index set is then evaluated based on the fuzzy evaluation matrix to obtain the autonomous driving performance evaluation result.

[0012] Optionally, the evaluation index set includes a primary evaluation index set and a secondary evaluation index set; The primary evaluation index set includes: safety, comfort, efficiency, and driving standardization; The set of secondary evaluation indicators includes: number of collisions, number of emergency brakings, number of lateral swayings, parking impact, travel completion time, start time, number of traffic rule violations, and number of speedings; The evaluation result set includes: excellent, good, qualified, and unqualified.

[0013] Optionally, the step of constructing a fuzzy evaluation matrix based on the evaluation index set, the evaluation result set, and the target scene complexity, and evaluating the evaluation index set based on the fuzzy evaluation matrix to obtain the autonomous driving performance evaluation result, includes: The weights of the secondary evaluation index and the primary evaluation index are determined based on the complexity of the target scenario. Construct a fuzzy evaluation matrix, and obtain a fuzzy comprehensive matrix based on the fuzzy evaluation matrix, the weights of the first-level evaluation indicators, and the weights of the second-level evaluation indicators; The membership values ​​of autonomous driving performance under different evaluation results are obtained based on the fuzzy comprehensive matrix. The evaluation result of autonomous driving performance is determined based on the membership value.

[0014] The beneficial effects of this invention are as follows: This invention provides an autonomous driving performance evaluation method based on scene complexity quantification. It quantifies standard driving scene images to obtain the scene complexity corresponding to the standard driving scene image, then acquires real-time driving scene images under autonomous driving testing, and matches the most similar standard driving scene image to obtain the scene complexity corresponding to the real-time driving scene image. Finally, based on the autonomous driving test results obtained under autonomous driving testing and the scene complexity corresponding to the real-time driving scene image, an autonomous driving performance evaluation result is obtained. By quantifying complex scenes, the quantification results of complex scenes can be considered when evaluating autonomous driving test results, thus obtaining a more accurate autonomous driving performance evaluation result. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating an autonomous driving performance evaluation method based on scenario complexity metric provided in an embodiment of the present invention. Detailed Implementation

[0016] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. In the following description, the expression "some embodiments" refers to a subset of all possible embodiments; however, it should be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments and can be combined with each other without conflict.

[0017] In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention can be practiced without one or more of these details. In other instances, certain technical features well-known in the art have not been described in order to avoid obscuring the invention.

[0018] It should be understood that the present invention can be embodied in various forms and should not be construed as being limited to the embodiments set forth herein. Rather, providing these embodiments will make the disclosure thorough and complete, and will fully convey the scope of the invention to those skilled in the art. Furthermore, the terminology used herein is intended only to describe particular embodiments and is not intended to limit the invention. When used herein, the singular forms “a,” “an,” and “the” are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the terms “compose” and / or “comprising,” when used in this specification, identify the presence of the stated features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups. When used herein, the term “and / or” includes any and all combinations of the associated listed items.

[0019] It should also be noted that when an element is referred to as being "fixed to" another element, it can be directly attached to the other element or there may be an intervening element. When an element is referred to as being "connected to" another element, it can be directly connected to the other element or there may be an intervening element. The terms "vertical," "horizontal," "inner," "outer," "left," "right," and similar expressions used herein are for illustrative purposes only and do not represent the only possible implementation.

[0020] To fully understand this invention, a detailed structure will be presented in the following description to illustrate the technical solution proposed by this invention. Optional embodiments of the invention are described in detail below; however, in addition to these detailed descriptions, the invention may have other embodiments.

[0021] Example Please refer to the attached document. Figure 1 This application provides a method for evaluating the performance of autonomous driving based on scenario complexity measurement, including the following steps: Autonomous driving tests are conducted in autonomous driving scenarios to obtain images of the actual driving scenarios and the results of the autonomous driving tests. Acquire a standard driving scene image, and obtain the target scene complexity based on the real-time driving scene image and the standard driving scene image; Based on the autonomous driving test results and the complexity of the target scenario, the autonomous driving performance evaluation results are obtained.

[0022] Specifically, driving scenarios are diverse and contain a variety of elements, such as road conditions, traffic participants, and weather. Generally, the higher the complexity of the driving scenario, the greater the challenge for the autonomous driving system in terms of perception, decision-making, and execution, which may lead to lower safety and comfort scores. Scene complexity is a numerical value obtained by quantifying each element in the scene. The quantification results provide a basis for subsequent evaluation of driving performance, thereby enabling the autonomous driving system to demonstrate its driving performance in complex and dynamic scenarios. Specifically, this involves: acquiring autonomous driving test videos under the same scene, performing frame-by-frame processing on the autonomous driving test videos, and selecting the image of the middle frame of the autonomous driving video as the real-time driving scene image; and obtaining the autonomous driving test results under this scene based on road test equipment, vehicle OBD system, etc. Standard driving scenario images are standard images that can represent scene elements. Standard driving scenario images are collected for various scenarios. Then, the scene corresponding to the standard driving scenario image is quantified to obtain the scene complexity, thus obtaining multiple standard driving images and their respective scene complexities. Then, the real-time driving scenario image is compared with multiple standard driving scenario images, and the standard driving scenario image most similar to the real-time driving scenario image is taken as the target driving scenario. The corresponding target scene complexity is obtained based on the target driving scenario, that is, the target scene complexity is taken as the real-time driving scenario complexity. Finally, the driving performance is evaluated based on the autonomous driving test results and the target scene complexity to obtain the autonomous driving performance evaluation result.

[0023] As an optional implementation, obtaining the autonomous driving test results under autonomous driving testing includes: In the aforementioned autonomous driving scenario, three consecutive autonomous driving tests of the same route were performed. Record the test results of the same trip three times, and calculate the average value based on the test results of the three trips to obtain the autonomous driving test result.

[0024] Specifically, to avoid bias in autonomous driving test results, three autonomous driving tests with the same route are performed under the same autonomous driving scenario to obtain three autonomous driving test results. The average of the three autonomous driving test results is calculated and the result of the average calculation is taken as the autonomous driving test result.

[0025] As an optional implementation, the step of acquiring a standard driving scene image and obtaining the target scene complexity based on the real-time driving scene image and the standard driving scene image includes: A knowledge graph is constructed by acquiring standard driving scene images and environmental data information synchronized with the standard driving scene images, resulting in a standard driving scene graph-standard semantic vector-standard scene complexity triplet library; Semantic recognition is performed on the real-time driving scene images to obtain real-time semantic vectors; The Euclidean distance between the real-time semantic vector and the standard semantic vector of each standard driving scene image is calculated sequentially to obtain multiple calculation results; The multiple calculation results are sorted from low to high, and the driving scene in the standard driving scene image with the lowest calculation result is taken as the target driving scene, and the corresponding target scene complexity is obtained.

[0026] Specifically, by first constructing a triplet library of standard driving scene graphs, standard semantic vectors, and standard scene complexity, it is equivalent to constructing a benchmark scene library based on multiple standard driving scene graphs. Each standard driving scene graph in the benchmark scene library corresponds to a standard semantic vector and a standard scene complexity. The benchmark scene library can provide a foundation for the subsequent quantification of scene complexity of real-time driving scene images. Acquire real-time driving scene images, perform semantic recognition on the real-time driving scene images, and obtain real-time semantic vectors with the same features as standard semantic vectors. For example, if the standard semantic vectors include weather vectors, road vectors, and traffic vectors, then the real-time semantic vectors also include weather vectors, road vectors, and traffic vectors. The Euclidean distance between the real-time semantic vector and the standard semantic vector corresponding to each standard driving scenario graph in the benchmark scenario library is calculated sequentially, and multiple calculation results are obtained. The smaller the calculated Euclidean distance, the more similar the real-time semantic vector and the standard semantic vector are. The standard driving scenario graph associated with the standard semantic vector corresponding to the minimum Euclidean distance is taken as the target driving scenario graph, and the corresponding standard scenario complexity is taken as the target scenario complexity.

[0027] As an optional implementation, the step of acquiring standard driving scene images and constructing a knowledge graph using environmental data information synchronized with the standard driving scene images, resulting in a standard driving scene graph-standard semantic vector-standard scene complexity triplet library, includes: Create scene nodes for each standard driving scenario image; Entities are identified and relationships between entities are mined based on environmental data information synchronized with the standard driving scenario image. The confidence level of the entity relationships is determined, and the edge weights are obtained based on the confidence level of the entity relationships to complete the construction of the knowledge graph. Semantic segmentation and attribute extraction are performed on each standard driving scene image to obtain a standard semantic vector; Using scene nodes as the granularity, the confidence of the entity relationships associated with the scene nodes is weighted and summed to obtain the standard scene complexity of the scene node, thereby obtaining a standard driving scene graph-standard semantic vector-standard scene complexity triplet library.

[0028] Specifically, each standard driving scenario image is abstracted into an independent node, forming a set of scenario nodes. These nodes serve as the carriers for subsequent associated entities, semantic vectors, and complexity, ensuring that the scenario can be uniquely identified and traced. Next, semantic relationships are mined from environmental data to quantify the strength of associations between entities in the scenario, providing a basis for complexity calculation. Then, attributes are extracted from each standard driving scenario image to form a multi-dimensional attribute vector. This attribute vector is then transformed into a fixed-dimensional semantic vector, ensuring that the semantic vectors of different scenarios can be used to calculate similarity. Finally, the comprehensive complexity of the scenario nodes is quantified, and the scenario node, its corresponding standard semantic vector, and the calculated standard scenario complexity are associated and stored to form a triplet library.

[0029] As an optional implementation, the steps of determining entities based on environmental data information synchronized with the standard driving scene image, mining relationships between entities, determining the confidence level of entity relationships, obtaining edge weights based on the confidence level of entity relationships, and completing the construction of the knowledge graph include: The environmental data information includes weather environmental data information, road environmental data information, and traffic environmental data information; Multiple weather entities are determined based on the weather environment data, multiple road entities are determined based on the road environment data, and multiple traffic entities are determined based on the traffic environment data. The weather entities, road entities, and traffic entities are then connected to their respective scene nodes. The Bayesian algorithm was used to mine relationships between three types of entity pairs: weather-road, weather-traffic, and road-traffic, to obtain the first entity relationship, the second entity relationship, and the third entity relationship. The first entity relationship, the second entity relationship, and the third entity relationship are merged, and duplicate entity relationships are removed to obtain multiple candidate entity relationships. Bayesian posterior probabilities are assigned to the multiple candidate entity relationships as confidence levels. After normalizing the confidence level, the edge weights are obtained. Directed edges are then established between the weather-road-traffic entity nodes to complete the knowledge graph construction.

[0030] Specifically, the environment in an autonomous driving scenario includes weather, road, and traffic environments. For example, weather conditions include weather type, precipitation intensity, and visibility; road conditions include road type, road infrastructure, and road surface conditions; and traffic conditions include traffic participants, traffic flow, and traffic congestion. Bayesian algorithms can be used to model relationships between entities, inferring associations through probabilistic dependencies and mining relationships by representing conditional dependencies between entities. Bayesian algorithms are used to analyze three types of entities: weather-road, weather-traffic, and road-traffic, to mine directed relationships between entity pairs. We obtain the first, second, and third entity relationships, which reflect the influence between entities in different environments. This helps us understand the complexity of driving scenarios and provides a basis for subsequent quantification of scenario complexity. For example, the weather-road entity pair: snow cover will cause lane lines to be covered; the weather-traffic entity pair: heavy rain will cause vehicles to move slowly, leading to traffic congestion; the road-traffic entity pair: water accumulation on the road will cause traffic congestion. The confidence level can directly reflect the probability of the relationship being true. After normalizing the confidence level, we obtain the edge weights. Based on the entity nodes, entity relationships, and edge weights, we construct a knowledge graph.

[0031] As an optional implementation, the step of sequentially calculating the Euclidean distance between the real-time semantic vector and the standard semantic vector of the standard driving scene image to obtain the calculation result includes: Calculate the weather Euclidean distance between the real-time semantic vector and the standard semantic vector to obtain the first calculation result; Calculate the road Euclidean distance between the real-time semantic vector and the standard semantic vector to obtain a second calculation result; Calculate the traffic Euclidean distance between the real-time semantic vector and the standard semantic vector to obtain a third calculation result; The calculation result is obtained by weighted summation of the first calculation result, the second calculation result, and the third calculation result.

[0032] Specifically, autonomous driving scenarios generally include weather elements, road elements, and traffic elements. Real-time semantic vectors describe key information of real-time driving scenario images. By calculating the Euclidean distance of weather, the Euclidean distance of road, and the Euclidean distance of traffic, the first, second, and third calculation results can be obtained to measure the similarity between the real-time semantic vector obtained from the real-time driving scenario image and the standard semantic vector obtained from the standard driving scenario image. This allows us to determine the degree of similarity between the real-time driving scenario image and the standard driving scenario image in terms of weather, road, and traffic. The first, second, and third calculation results are then weighted and summed to obtain the final calculation result.

[0033] As an optional implementation, obtaining the autonomous driving performance evaluation result based on the autonomous driving test results and the target scene complexity includes: Based on the autonomous driving test results, construct an evaluation index set and an evaluation result set; Based on the evaluation index set, the evaluation result set, and the target scene complexity, a fuzzy evaluation matrix is ​​constructed. The evaluation index set is then evaluated based on the fuzzy evaluation matrix to obtain the autonomous driving performance evaluation result.

[0034] The evaluation index set includes a primary evaluation index set and a secondary evaluation index set; The primary evaluation index set includes: safety, comfort, efficiency, and driving standardization; The set of secondary evaluation indicators includes: number of collisions, number of emergency brakings, number of lateral swayings, parking impact, travel completion time, start time, number of traffic rule violations, and number of speedings; The evaluation result set includes: excellent, good, satisfactory, and unsatisfactory; Specifically, the fuzzy evaluation matrix can transform qualitative evaluation into computable quantitative evaluation, that is, it can make an overall evaluation of objects with multiple influencing factors through fuzzy mathematics, and provide more accurate evaluation results. By constructing an evaluation index set and an evaluation result set, the evaluation index set includes a primary evaluation index set and a secondary evaluation index set. The primary evaluation index set includes safety, comfort, efficiency, and driving discipline. The safety index includes two secondary evaluation indicators: number of collisions and number of emergency brakings. The comfort index includes two secondary evaluation indicators: number of lateral swaying events and stopping impact. The efficiency index includes two secondary evaluation indicators: trip completion time and start-up time. The driving discipline index includes two secondary evaluation indicators: number of traffic rule violations and number of speeding violations. The evaluation result set is the set of final evaluation results obtained after calculation based on the primary and secondary evaluation indicators, including excellent, good, satisfactory, and unsatisfactory. U = (U1, U2, U3, U4) = (Safety, Comfort, Efficiency, Driving Standards) U1=(u 11 ,u 12 = (Number of collisions, number of emergency braking operations) U2=(u 21 ,u 22 = (Number of lateral swings, impact of stopping) U3=(u 31 ,u 32 = (Trip completion time, start time) U4=(u 41 ,u42 = (Number of traffic violations and number of speeding violations) V = (v1, v2, v3, v4) = (Excellent, Good, Satisfactory, Unsatisfactory) Wherein, U is the evaluation index set, U1 is the safety index, U2 is the comfort index, U3 is the efficiency index, and U4 is the driving standardization index; V is the evaluation set.

[0035] As an optional implementation, the step of constructing a fuzzy evaluation matrix based on the evaluation index set, the evaluation result set, and the target scene complexity, and evaluating the evaluation index set based on the fuzzy evaluation matrix to obtain the autonomous driving performance evaluation result includes: The weights of the secondary evaluation index and the primary evaluation index are determined based on the complexity of the target scenario. Construct a fuzzy evaluation matrix, and obtain a fuzzy comprehensive matrix based on the fuzzy evaluation matrix, the weights of the first-level evaluation indicators, and the weights of the second-level evaluation indicators; The membership values ​​of autonomous driving performance under different evaluation results are obtained based on the fuzzy comprehensive matrix. The evaluation result of autonomous driving performance is determined based on the membership value.

[0036] Specifically, weights are set according to the complexity of the target scenario. In each primary evaluation indicator, the evaluation focus differs depending on the complexity of the scenario. For example, in more complex scenarios, more attention is paid to safety indicators. Therefore, when the scenario complexity is higher, the weight of safety indicators can be set higher. When the scenario complexity is lower, the weight of comfort and efficiency indicators can be set higher. Regardless of whether the scenario complexity is high or low, the focus on driving compliance indicators has a certain weight. By determining the weights of primary and secondary evaluation indicators based on the complexity of the target scenario, the weights can be dynamically adjusted based on the complexity of the target scenario to obtain more accurate evaluation results. A = (A1, A2, A3, A4) A1 = (a 11 ,a 12 ) A2 = (a 21 ,a 22 ) A3 = (a 31 ,a 32 ) A4 = (a 41 ,a 42 ) Where A is the weight vector of the primary evaluation index, A 1、 A 2、 A 3、A4 is the weight vector corresponding to each secondary evaluation indicator; A fuzzy evaluation matrix R is constructed using membership functions. The fuzzy evaluation matrix R is composed of each index factor U. i Membership ratio (R / U) of each evaluation level i The fuzzy evaluation matrix is ​​composed of several elements, each representing the membership degree of the evaluation index factor to the evaluation level. The rows of the fuzzy evaluation matrix correspond to the evaluation indexes, and the columns correspond to the evaluation results. The expression for the fuzzy evaluation matrix is:

[0037]

[0038]

[0039] First, a single-factor evaluation is performed on the subset of secondary evaluation indicators to obtain a fuzzy evaluation matrix, thereby obtaining the evaluation result vectors B1, B2, B3, and B4 of the secondary evaluation indicator subset. Then, a fuzzy comprehensive matrix R of the primary evaluation indicators is constructed based on the evaluation result vectors of the secondary evaluation indicator subset. Next, the evaluation result B is obtained through the fuzzy comprehensive matrix R of the primary evaluation indicators and the weight vector A of the primary evaluation indicators. Each element in the evaluation result represents the membership degree of the evaluation system to a certain evaluation result. That is to say, the obtained evaluation result B can obtain the score corresponding to each evaluation result (excellent, good, qualified, unqualified). Based on the score corresponding to each evaluation result, the result with the highest score is taken as the evaluation result of driving performance.

[0040] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for evaluating the performance of autonomous driving based on scenario complexity metric, characterized in that, Includes the following steps: Autonomous driving tests are conducted in autonomous driving scenarios to obtain images of the actual driving scenarios and the results of the autonomous driving tests. Acquire a standard driving scene image, and obtain the target scene complexity based on the real-time driving scene image and the standard driving scene image; Based on the autonomous driving test results and the complexity of the target scenario, the autonomous driving performance evaluation results are obtained.

2. The autonomous driving performance evaluation method based on scenario complexity quantification according to claim 1, characterized in that, The acquisition of autonomous driving test results under autonomous driving testing includes: In the aforementioned autonomous driving scenario, three consecutive autonomous driving tests of the same route were performed. Record the test results of the same trip three times, and calculate the average value based on the test results of the three trips to obtain the autonomous driving test result.

3. The autonomous driving performance evaluation method based on scenario complexity quantification according to claim 1, characterized in that, The step of acquiring a standard driving scene image and obtaining the target scene complexity based on the real-time driving scene image and the standard driving scene image includes: A knowledge graph is constructed by acquiring standard driving scene images and environmental data information synchronized with the standard driving scene images, resulting in a standard driving scene graph-standard semantic vector-standard scene complexity triplet library; Semantic recognition is performed on the real-time driving scene images to obtain real-time semantic vectors; The Euclidean distance between the real-time semantic vector and the standard semantic vector of each standard driving scene image is calculated sequentially to obtain multiple calculation results; The multiple calculation results are sorted from low to high, and the driving scene in the standard driving scene image with the lowest calculation result is taken as the target driving scene, and the corresponding target scene complexity is obtained.

4. The autonomous driving performance evaluation method based on scenario complexity quantification according to claim 3, characterized in that, The acquisition of standard driving scenario images and the environmental data information synchronized with the standard driving scenario images are used to construct a knowledge graph, resulting in a standard driving scenario graph-standard semantic vector-standard scenario complexity triplet library, including: Create scene nodes for each standard driving scenario image; Entities are identified and relationships between entities are mined based on environmental data information synchronized with the standard driving scenario image. The confidence level of the entity relationships is determined, and the edge weights are obtained based on the confidence level of the entity relationships to complete the construction of the knowledge graph. Semantic segmentation and attribute extraction are performed on each standard driving scene image to obtain a standard semantic vector; Using scene nodes as the granularity, the confidence of the entity relationships associated with the scene nodes is weighted and summed to obtain the standard scene complexity of the scene node, thereby obtaining a standard driving scene graph-standard semantic vector-standard scene complexity triplet library.

5. The autonomous driving performance evaluation method based on scenario complexity metric as described in claim 4, characterized in that, The process of determining entities based on environmental data information synchronized with the standard driving scenario image, mining relationships between entities, determining the confidence level of entity relationships, obtaining edge weights based on the confidence level of entity relationships, and completing the construction of the knowledge graph includes: The environmental data information includes weather environmental data information, road environmental data information, and traffic environmental data information; Multiple weather entities are determined based on the weather environment data, multiple road entities are determined based on the road environment data, and multiple traffic entities are determined based on the traffic environment data. The weather entities, road entities, and traffic entities are then connected to their respective scene nodes. The Bayesian algorithm was used to mine relationships between three types of entity pairs: weather-road, weather-traffic, and road-traffic, to obtain the first entity relationship, the second entity relationship, and the third entity relationship. The first entity relationship, the second entity relationship, and the third entity relationship are merged, and duplicate entity relationships are removed to obtain multiple candidate entity relationships. Bayesian posterior probabilities are assigned to the multiple candidate entity relationships as confidence levels. After normalizing the confidence level, the edge weights are obtained. Directed edges are then established between the weather-road-traffic entity nodes to complete the knowledge graph construction.

6. The autonomous driving performance evaluation method based on scenario complexity metric as described in claim 5, characterized in that, The step of sequentially calculating the Euclidean distance between the real-time semantic vector and the standard semantic vector of the standard driving scene image to obtain the calculation result includes: Calculate the weather Euclidean distance between the real-time semantic vector and the standard semantic vector to obtain the first calculation result; Calculate the road Euclidean distance between the real-time semantic vector and the standard semantic vector to obtain a second calculation result; Calculate the traffic Euclidean distance between the real-time semantic vector and the standard semantic vector to obtain a third calculation result; The calculation result is obtained by weighted summation of the first calculation result, the second calculation result, and the third calculation result.

7. The autonomous driving performance evaluation method based on scenario complexity quantification according to claim 1, characterized in that, The autonomous driving performance evaluation result is obtained based on the autonomous driving test results and the complexity of the target scenario, including... Based on the autonomous driving test results, construct an evaluation index set and an evaluation result set; Based on the evaluation index set, the evaluation result set, and the target scene complexity, a fuzzy evaluation matrix is ​​constructed. The evaluation index set is then evaluated based on the fuzzy evaluation matrix to obtain the autonomous driving performance evaluation result.

8. The autonomous driving performance evaluation method based on scenario complexity quantification according to claim 7, characterized in that, The evaluation index set includes a primary evaluation index set and a secondary evaluation index set; The primary evaluation index set includes: safety, comfort, efficiency, and driving standardization; The set of secondary evaluation indicators includes: number of collisions, number of emergency brakings, number of lateral swayings, parking impact, travel completion time, start time, number of traffic rule violations, and number of speedings; The evaluation result set includes: excellent, good, qualified, and unqualified.

9. The autonomous driving performance evaluation method based on scenario complexity quantification according to claim 8, characterized in that, The step of constructing a fuzzy evaluation matrix based on the evaluation index set, the evaluation result set, and the target scene complexity, and evaluating the evaluation index set based on the fuzzy evaluation matrix to obtain the autonomous driving performance evaluation result includes: The weights of the secondary evaluation index and the primary evaluation index are determined based on the complexity of the target scenario. Construct a fuzzy evaluation matrix, and obtain a fuzzy comprehensive matrix based on the fuzzy evaluation matrix, the weights of the first-level evaluation indicators, and the weights of the second-level evaluation indicators; The membership values ​​of autonomous driving performance under different evaluation results are obtained based on the fuzzy comprehensive matrix. The evaluation result of autonomous driving performance is determined based on the membership value.

Images