Automatic emergency braking system test scene construction method
By collecting accident samples and using fusion clustering algorithms to construct AEB testing scenarios, the problem of deviation between test scenarios and actual situations in the existing technology is solved, and comprehensive verification and efficient coverage of the functions of the AEB system are achieved.
Patent Information
- Application Number
- CN202510564073.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-30
- Publication Date
- 2025-08-08
AI Technical Summary
The existing AEB system test scenario construction method fails to fully consider the real traffic accident data, resulting in deviations from the test scenarios and the actual situation, and the system performance cannot be fully verified.
By collecting accident samples, selecting feature elements, using fusion clustering algorithms to cluster the accident samples, and constructing test scenarios for AEB system functional verification, including hierarchical clustering and K-means clustering, combining environmental and vehicle elements to accurately model real traffic accident scenarios.
It has realized the targeted testing scenario construction for the functional verification of AEB system, which can fully cover the testing needs, improve the scientificity and accuracy of the test, reduce the number of iterations and clustering fluctuations, and improve data clustering efficiency.
Smart Images
Figure CN120448855A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of intelligent vehicle testing, and in particular to a method for constructing a test scenario for an automatic emergency braking system. Background Art
[0002] With the accelerated advancement of global automotive intelligence, the automatic emergency braking (AEB) system has become a core indicator for measuring the active safety performance of vehicles. International mainstream evaluation organizations (such as Euro NCAP and C-NCAP) have included AEB in mandatory test items, and my country's "Intelligent Connected Vehicle Road Testing and Demonstration Application Management Specifications" also explicitly require strict verification of AEB systems. However, existing testing methods have significant limitations in the scenario construction dimension: traditional closed-field testing mainly relies on standard regulatory scenarios (such as C-NCAP's CCRs / CCRm conditions), and its scenario coverage is less than 40% of real accident types; test cases constructed based on experience are difficult to reflect the complex interactions between traffic participants. This gap between test scenarios and actual road safety needs directly restricts the effectiveness of functional safety verification of AEB systems, and it is urgent to build a new data-driven scenario construction method system.
[0003] Current AEB test scenario construction technology exhibits a dual-track development trend. On the one hand, real-vehicle testing based on physical sites still dominates, employing testing methods with preset obstacles and fixed trajectories. While this approach offers the advantage of reproducible test results, it also suffers from low scenario complexity and high testing costs (a single crash test can cost over 10,000 yuan). On the other hand, virtual simulation testing technology is rapidly developing. Digital twin scenarios constructed using tools such as Prescan and CarSim enable parameterized scenario adjustments, but most simulation models lack effective coupling with real-world accident data, resulting in insufficient correlation between virtual test results and actual road performance. Furthermore, existing AEB system test scenario construction methods often fail to fully consider real-world accident data. This is because existing AEB test scenario construction primarily relies on functional verification of the AEB system itself, lacking a robust AEB test scenario construction system and methodology based on real-world accident scenarios. Consequently, test scenarios lack a strong correlation with real-world accident scenarios, leading to discrepancies between test scenarios and actual conditions and an inability to fully verify the performance of the AEB system. Summary of the Invention
[0004] The present invention aims to provide a method for constructing a test scenario for an automatic emergency braking system, which can specifically construct a test scenario for AEB system function verification and can fully cover AEB testing requirements.
[0005] The basic solution provided by the present invention is: a method for constructing a test scenario for an automatic emergency braking system, comprising the following steps:
[0006] Step 1, collect accident samples;
[0007] Step 2: Select characteristic elements from the accident samples based on the test scenario requirements and the functional characteristics of the AEB system;
[0008] Step 3: Use a fusion clustering algorithm to perform cluster analysis on the accident samples and output the clustering results. The fusion clustering algorithm includes: first performing hierarchical clustering, and then performing K-means clustering based on the results of the hierarchical clustering;
[0009] Step 4: Construct a test scenario for AEB system function verification based on the clustering results.
[0010] Furthermore, in step 1, when collecting accident samples, the following principles are followed: rear-end collision traffic accidents, oblique collision traffic accidents, and head-on collision traffic accidents are selected; only two vehicles collide in the accident; and the vehicle types are passenger cars and commercial vehicles.
[0011] Furthermore, the characteristic elements include environmental elements and vehicle elements; the environmental elements include weather elements, lighting elements and accident location; the vehicle elements include accident type, collision direction, vehicle type, collision speed, movement state and collision location.
[0012] Furthermore, the hierarchical clustering includes: randomly selecting N samples from the accident samples;
[0013] Let the initial K value be 2, run the hierarchical clustering algorithm, calculate the distance between each two samples in N samples, merge the two samples with the closest distance into one sample, repeatedly calculate and merge until the N samples are reduced to K samples, and then stop the hierarchical clustering algorithm.
[0014] Furthermore, the K-means clustering includes:
[0015] S1, the K samples output by hierarchical clustering are used as the initial cluster centers of K-means clustering;
[0016] S2, run the K-means clustering algorithm, calculate the distance between each sample in the overall data sample and each initial cluster center, assign each sample to the category represented by each cluster center according to the principle of closest distance and establish a category label;
[0017] S3, compare the current category label of each sample with the previous one. When the two category labels of a sample are inconsistent, update the cluster center and repeat S2. Iterate the K-means clustering algorithm repeatedly until the two category labels of all samples are the same, and then stop the iteration to obtain the clustered data set.
[0018] S4, draw the clustering curve and judge the convergence of the algorithm. Set the initial K value to 2. If the clustering curve does not converge, increase the K value and repeat the hierarchical clustering and S1-S3 until the curve converges to the required effect, then terminate the algorithm.
[0019] S5, according to the clustering curve, the clustering data set corresponding to the optimal K value is selected as the final clustering data set, that is, K types of typical traffic scenes are clustered as the clustering results.
[0020] Furthermore, the clustering curve is a curve with the number of clusters K as the horizontal axis and the sum of the distances from all samples in the cluster to the cluster center as the vertical axis.
[0021] Furthermore, in step 4, based on the clustering results and comparing the feature codes, K types of scenario descriptions are interpreted to construct test scenarios for AEB system function verification.
[0022] The working principle and advantages of the present invention are:
[0023] The present invention provides a method for constructing test scenarios for an automatic emergency braking system, which can specifically construct test scenarios for AEB system function verification and fully cover AEB testing requirements. The key points are:
[0024] First, this solution achieves accurate modeling of real-world traffic accident scenarios through a systematic data collection and feature extraction mechanism. During the accident sample collection phase, three typical accident types were clearly defined: rear-end collisions, oblique collisions, and head-on collisions. Both passenger cars and commercial vehicles were covered, ensuring comprehensive coverage of the test scenarios. Compared to traditional methods that focus solely on a single collision configuration or vehicle type, this solution utilizes multi-dimensional data classification (such as collision direction, vehicle speed, and weather conditions) to more comprehensively reflect the complexity of real-world road conditions. For example, weather and lighting conditions are incorporated into environmental factors to simulate the impact of sensor degradation on AEB systems in rainy and foggy weather. Vehicle factors distinguish between motion states (acceleration / braking / constant speed) and collision locations, infusing dynamic interactive features into the test scenarios. Furthermore, limiting the number of accident participants to two vehicles effectively avoids overfitting in multi-vehicle chain collision scenarios, ensuring scenario representativeness while improving algorithm processing efficiency.
[0025] Second, this solution uses a fusion clustering algorithm to perform scientific and accurate cluster analysis on accident samples and designs AEB test scenarios based on the clustering results, effectively improving the scientific nature and accuracy of AEB system testing. By using the results of hierarchical clustering as the initial cluster centers for K-means clustering, compared to the traditional K-means clustering algorithm, which randomly selects initial cluster centers, the number of iterations can be reduced. The entire fusion clustering algorithm reduces the number of iterations by an average of 8, and the average fluctuation in clustering results is reduced by 3%. This achieves scientific and accurate clustering of accident data samples, improves data clustering efficiency, and effectively covers existing AEB test scenarios. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] Figure 1 A schematic diagram of a method flow in an embodiment of a method for constructing a test scenario for an automatic emergency braking system according to the present invention;
[0027] Figure 2 This is a structural diagram of an embodiment of a method for constructing a test scenario for an automatic emergency braking system according to the present invention. DETAILED DESCRIPTION
[0028] The following is a further detailed description through specific implementation methods:
[0029] The embodiment is basically as shown in the attached Figure 1 A method for constructing a test scenario for an automatic emergency braking system is shown, comprising the following steps:
[0030] Step 1: Collect accident samples.
[0031] When collecting accident samples, refer to the following principles: select rear-end collision traffic accidents, oblique collision traffic accidents, and head-on collision traffic accidents; only two vehicles collide in the accident; the vehicle types are passenger cars and commercial vehicles.
[0032] In this embodiment, 6,600 traffic accident samples suitable for AEB system testing are collected from the public National Highway Traffic Safety Administration (NHTSA) Crash Report Sampling System (CRSS) database.
[0033] Step 2: Select characteristic elements from the accident samples based on the test scenario requirements and the functional characteristics of the AEB system.
[0034] Among them, the test scenario requirements include: typical scenario testing, which is used to verify the AEB system's target recognition ability, risk decision-making ability and braking execution ability; complex extreme scenario testing, which is used to verify the AEB system's scenario adaptability, system stability and robustness to abnormal working conditions in special environments.
[0035] The functional features of the AEB system include: collision risk identification function, human-computer interaction function, and automatic emergency braking function. Among them, the collision risk identification function refers to the ability of the AEB system to identify the possible collision between the vehicle and other traffic participants. The human-computer interaction function refers to the generation of multi-level warning signals by the AEB system when a collision risk is detected, including visual warnings, acoustic alarms and tactile feedback, to warn the driver and prompt them to take emergency braking or other risk avoidance measures. The automatic emergency braking function means that if the driver fails to take effective braking measures in time, that is, when the driver's response delay exceeds a preset threshold, the AEB system will intervene in a timely manner and automatically perform emergency braking to reduce the risk of collision or mitigate the severity of the collision.
[0036] In this embodiment, feature correlation analysis can be used to select feature elements based on the above-mentioned test scenario requirements and AEB system functional characteristics.
[0037] like Figure 2 As shown, the characteristic elements include environmental elements and vehicle elements; the environmental elements include weather elements, lighting elements and accident location; the vehicle elements include accident type, collision direction, vehicle type, collision speed, motion state and collision location.
[0038] The 6,600 accident samples were statistically analyzed and coded according to characteristic factors, such as weather conditions: clear (characteristic coding value 1), cloudy (characteristic coding value 2), rainy (characteristic coding value 3), snowy (characteristic coding value 4), and other (characteristic coding value 5).
[0039] Step 3: Use the fusion clustering algorithm to perform cluster analysis on the accident samples; and output the clustering results; the fusion clustering algorithm includes: first performing hierarchical clustering, and then performing K-means clustering based on the results of the hierarchical clustering.
[0040] The hierarchical clustering includes: randomly extracting N samples from the accident samples; in this embodiment, N=500.
[0041] Let the initial K value be 2, run the hierarchical clustering algorithm, calculate the distance between each two samples in N samples, merge the two samples with the closest distance into one sample, repeatedly calculate and merge until the N samples are reduced to K samples, and then stop the hierarchical clustering algorithm.
[0042] The distance of the sample is calculated using the following formula:
[0043]
[0044] Where: X i is a single data sample, C j is the j-th cluster center, N is the number of accident samples, and K is the number of clusters.
[0045] The K-means clustering includes:
[0046] S1, the K samples output by hierarchical clustering are used as the initial cluster centers of K-means clustering;
[0047] S2, run the K-means clustering algorithm, calculate the distance between each sample in the overall data sample and each initial cluster center, assign each sample to the category represented by each cluster center according to the principle of closest distance and establish a category label;
[0048] S3, compare the current category label of each sample with the previous one. When the two category labels of a sample are inconsistent, update the cluster center and repeat S2. Iterate the K-means clustering algorithm repeatedly until the two category labels of all samples are the same, and then stop the iteration to obtain the clustered data set.
[0049] S4, draw the clustering curve and judge the convergence of the algorithm. Set the initial K value to 2. If the clustering curve does not converge, increase the K value and repeat the hierarchical clustering and S1-S3 until the curve converges to the required effect, then terminate the algorithm.
[0050] S5, according to the clustering curve, the clustering data set corresponding to the optimal K value is selected as the final clustering data set, that is, K types of typical traffic scenes are clustered as the clustering results.
[0051] The clustering curve is a curve with the number of clusters K as the horizontal axis and the sum of the distances from all samples in the cluster to the cluster center (SSC) as the vertical axis.
[0052] The calculation formula of SSC is:
[0053]
[0054] Where: X ij is the i-th sample of the j-th class, C j is the j-th cluster center, K is the number of clusters, n j is the number of samples contained in the jth class.
[0055] By integrating these clustering algorithms, this solution significantly improves scenario clustering efficiency and stability, while effectively covering existing AEB test scenarios. Hierarchical clustering, by decomposing the complex correlations of accident samples (such as collision timing and the coupling of traffic participant behaviors) through a tree-like structure, effectively identifies potential subclasses in non-convex datasets. K-means offers computational efficiency advantages in spherical clustering. Combining these two approaches allows for a balanced approach to diverse data distributions and effectively reduces the risk of misclassification.
[0056] Secondly, the dendrogram output by hierarchical clustering can automatically determine the optimal number of clusters, K, for K-means, resolving the subjectivity inherent in traditional K-means, which requires manual presetting of the K value. The initial centroid coordinates generated by hierarchical clustering can also avoid the local optimality trap caused by random initialization in K-means. Furthermore, compared to traditional K-means clustering algorithms, this fusion clustering algorithm effectively reduces the number of algorithm iterations by an average of eight, and reduces the average fluctuation of clustering results by 3%, demonstrating improved clustering stability.
[0057] Furthermore, traditional clustering methods in the ADAS testing field often directly apply a single algorithm (such as the K-means algorithm alone, and existing technologies often do not integrate the K-means algorithm with the hierarchical clustering algorithm because traditional theory believes that hierarchical clustering focuses on inter-cluster heterogeneity, while K-means focuses on intra-cluster homogeneity. The two optimization goals conflict and are not suitable for integration). This fails to address the dual challenges of "long-tail distribution" and "sensitivity to outliers" in accident scenarios. This solution overcomes the technical bias of existing technologies against integrating the K-means and hierarchical clustering algorithms through algorithm cascade innovation. The breakthrough discovery is that the combination of the two can better achieve clustering by prioritizing the separation of high-risk scenarios through hierarchical clustering and then optimizing high-frequency scenarios with K-means. In the hierarchical clustering stage, rare extreme scenarios (such as the extremely small proportion of pedestrian crossing scenarios) are captured; in the K-means stage, the Euclidean distance is iteratively optimized to enhance the discrimination of high-incidence typical scenarios (such as the high proportion of forward collision scenarios), thereby achieving effective coverage of accident scenarios.
[0058] Step 4: Construct a test scenario for AEB system function verification based on the clustering results.
[0059] Specifically, based on the clustering results and comparing the feature codes, K types of scenario descriptions are interpreted to construct test scenarios for AEB system function verification.
[0060] The present embodiment provides a method for constructing an automatic emergency braking system test scenario, which can specifically construct a test scenario for AEB system function verification and can fully cover AEB test requirements.
[0061] The above is only an embodiment of the present invention. Common knowledge such as the specific structure and characteristics of the scheme is not described in detail here. Ordinary technicians in the relevant field are aware of all common technical knowledge in the technical field of the invention before the application date or priority date, can obtain all existing technologies in the field, and have the ability to apply conventional experimental means before that date. Ordinary technicians in the relevant field can improve and implement this scheme in combination with their own abilities under the guidance of this application. Some typical well-known structures or well-known methods should not become obstacles for ordinary technicians in the relevant field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several variations and improvements can be made, which should also be regarded as the scope of protection of the present invention. These will not affect the effect of the implementation of the present invention and the practicality of the patent.
Claims
1. A method for constructing a test scenario for an automatic emergency braking system, characterized in that: The following steps are involved: Step 1, collect accident samples; Step 2: Select characteristic elements from the accident samples based on the test scenario requirements and the functional characteristics of the AEB system; Step 3: Use the fusion clustering algorithm to perform cluster analysis on the accident samples; And output the clustering results; The fusion clustering algorithm includes: first performing hierarchical clustering, and then performing K-means clustering based on the results of the hierarchical clustering; Step 4: Construct a test scenario for AEB system function verification based on the clustering results.
2. The method for constructing an automatic emergency braking system test scenario according to claim 1, characterized in that: In step 1, when collecting accident samples, refer to the following principles: select rear-end collision traffic accidents, oblique collision traffic accidents, and head-on collision traffic accidents; only two vehicles collide in the accident; the vehicle types are passenger cars and commercial vehicles.
3. The method for constructing an automatic emergency braking system test scenario according to claim 1, characterized in that: The characteristic elements include environmental elements and vehicle elements; the environmental elements include weather elements, lighting elements and accident location; the vehicle elements include accident type, collision direction, vehicle type, collision speed, movement state and collision location.
4. The method for constructing an automatic emergency braking system test scenario according to claim 1, characterized in that: The hierarchical clustering includes: randomly selecting N samples from the accident samples; Let the initial K value be 2, run the hierarchical clustering algorithm, calculate the distance between each two samples in N samples, merge the two samples with the closest distance into one sample, repeatedly calculate and merge until the N samples are reduced to K samples, and then stop the hierarchical clustering algorithm.
5. The method for constructing an automatic emergency braking system test scenario according to claim 4, characterized in that: The K-means clustering includes: S1, the K samples output by hierarchical clustering are used as the initial cluster centers of K-means clustering; S2, run the K-means clustering algorithm, calculate the distance between each sample in the overall data sample and each initial cluster center, assign each sample to the category represented by each cluster center according to the principle of closest distance and establish a category label; S3, compare the current category label of each sample with the previous one. When the two category labels of a sample are inconsistent, update the cluster center and repeat S2. Iterate the K-means clustering algorithm repeatedly until the two category labels of all samples are the same, and then stop the iteration to obtain the clustered data set. S4, draw the clustering curve and judge the convergence of the algorithm. Set the initial K value to 2. If the clustering curve does not converge, increase the K value and repeat the hierarchical clustering and S1-S3 until the curve converges to the required effect, then terminate the algorithm. S5, according to the clustering curve, the clustering data set corresponding to the optimal K value is selected as the final clustering data set, that is, K types of typical traffic scenes are clustered as the clustering results.
6. The method for constructing an automatic emergency braking system test scenario according to claim 5, characterized in that: The clustering curve is a curve with the number of clusters K as the horizontal axis and the sum of the distances from all samples in the cluster to the cluster center as the vertical axis.
7. The method for constructing an automatic emergency braking system test scenario according to claim 1, characterized in that: In step 4, based on the clustering results and comparing the feature codes, K types of scenario descriptions are interpreted to construct test scenarios for AEB system function verification.