Driving style recognition system based on natural driving database

Abstract

The invention discloses a driving style recognition system based on a natural driving database, and relates to the technical field of automobile driver driving style recognition, and the key point of the technical scheme is that the driving style recognition system comprises a feature extraction module, an offline clustering analysis module and an online recognition and verification module. According to the invention, a driving style identification system based on a natural driving database is established, and feature extraction and clustering analysis are carried out on a high-dimensional large-volume database, so that the driving style of a driver is effectively identified; according to the system, the comfort of the vehicle is considered, the vehicle state response representing the riding comfort is introduced into a driving style recognition model, the recognition precision of the driving style is effectively improved, a feasible scheme is provided for analysis and processing of vehicle high-dimensional mass sample data in China, and meanwhile, the driving style recognition precision is improved. The system can provide technical support for safe driving and riding comfort of the vehicle.

Description

Technical Field

[0001] The invention relates to the technical field of automobile driver driving style recognition, and more specifically, to a driving style recognition system based on a natural driving database. Background Art

[0002] As the number of cars and drivers in China continues to grow, road traffic safety issues are becoming increasingly important. There is a close relationship between the driver's driving style and the incidence of traffic accidents, and aggressive driving styles often lead to a higher probability of accidents. Therefore, effectively identifying the driver's driving style is of great significance for improving traffic safety, promoting energy efficiency, and improving the driving experience. In addition, the development of autonomous vehicles also requires a deep understanding of the driver's driving style to achieve a more humane and personalized autonomous driving system.

[0003] At present, the evaluation methods of driver's driving style are mainly divided into two types: subjective evaluation and objective evaluation. Subjective evaluation methods, such as questionnaire surveys and expert scoring, are simple and easy to implement, but are easily affected by subjective factors and external environment. In contrast, objective evaluation methods are based on driving data obtained from driving simulators or real car road tests, which can more intuitively reflect the performance characteristics of driving style.

[0004] Among objective evaluation methods, style recognition models based on driving data have been widely used. These models usually use statistical and machine learning methods, such as hidden Markov model (HMM), support vector machine (SVM), K-means clustering, etc., to derive style recognition models from historical driving data. These models have the advantages of high accuracy, low complexity, and strong generalization ability, and can effectively identify the driver's driving style. In particular, driving style recognition systems based on natural driving databases have become a research hotspot. Such systems collect vehicle driving data such as speed, acceleration, steering angle, etc. during natural driving, and extract characteristic indicators that can reflect driving style. Then, these characteristic indicators are clustered and analyzed using machine learning algorithms to accurately identify the driver's driving style. This method based on big data and machine learning not only improves the accuracy of recognition, but also provides strong support for the humanized and personalized design of autonomous driving vehicles.

[0005] At present, there are still obvious defects in the research of driving style recognition: the subjectivity of questionnaire survey data, the limitations of data mining algorithms in processing high-dimensional and large-volume data, the dependence of supervised learning methods on large amounts of labeled data, and the neglect of the impact of driver comfort on driving style and driving safety. These problems jointly restrict the further development and application of driving style recognition technology. Summary of the invention

[0006] The purpose of the present invention is to provide a driving style recognition system based on a natural driving database to solve the problems of low efficiency in processing high-dimensional and large-volume data, low driving style recognition accuracy and lack of real-time online recognition capability in the prior art.

[0007] The above technical objectives of the present invention are achieved through the following technical solutions: a driving style recognition system based on a natural driving database, comprising a feature extraction module, an offline clustering analysis module, and an online recognition and verification module;

[0008] The feature extraction module obtains original vehicle data from the natural driving database, processes the collected original vehicle data, extracts the mean, variance, maximum value, and minimum value of the collected original vehicle data as corresponding feature parameters after processing, combines the key feature parameters to form a new data feature set, and performs dimensionality reduction processing through Laplace feature mapping;

[0009] The offline clustering analysis module performs clustering analysis on the feature data after dimensionality reduction by using the DBSCAN algorithm to identify different driving style categories;

[0010] The online recognition and verification module completes the online recognition of driving style by analyzing the relationship between the feature data after dimensionality reduction and the cluster center in the driving style recognition model.

[0011] The present invention is further configured as follows: the original vehicle data includes the longitudinal speed, lateral speed, longitudinal acceleration, lateral acceleration and vertical acceleration of the vehicle, and the mean, variance, maximum value and minimum value of the above data are extracted as corresponding characteristic parameters.

[0012] The present invention is further configured as follows: the key characteristic parameters include the maximum value of the product of velocity and acceleration and the acceleration zero-crossing rate, the maximum value of the product of velocity and acceleration includes the maximum value of the product of longitudinal velocity and longitudinal acceleration, and the product of lateral velocity and lateral acceleration, and the acceleration zero-crossing rate includes the zero-crossing rate of longitudinal acceleration, lateral acceleration and vertical acceleration.

[0013] The present invention is further configured as follows: the offline clustering analysis module specifically selects representative data for initial clustering to determine the initial cluster center, and on this basis gradually adds new data feature sample points, iteratively optimizes the clustering results, and realizes efficient clustering analysis of large-volume databases.

[0014] The present invention is further configured as follows: the online identification and verification module also performs visualization processing on the data through the t-SNE algorithm, displays the topological structure of the feature data, and verifies the clustering effect of the DBSCAN clustering algorithm.

[0015] The present invention is further configured as follows: the online recognition and verification module also verifies the effectiveness of the recognition system through comparative analysis with a k-means clustering algorithm.

[0016] By adopting the above technical solution, the present invention not only extracts the mean, variance, maximum value and minimum value of the original vehicle data as basic characteristic parameters, but also specially considers the vertical movement of the vehicle as an evaluation index of ride comfort. This comprehensive consideration makes the recognition of driving style more comprehensive and accurate, and can more truly reflect the driver's behavioral habits and the passenger's riding experience; the present invention uses the DBSCAN clustering algorithm to perform cluster analysis on the data after dimensionality reduction. The algorithm can find clusters of any shape and has strong robustness to noise and outliers. Even in a complex driving data environment, the present invention can accurately identify different driving style categories, thereby improving the recognition accuracy of driving style categories;

[0017] The present invention uses Laplace eigenmap to reduce data dimension, effectively removes the strongly correlated parts in the feature data, and simplifies the data structure. This not only reduces the complexity of subsequent clustering analysis, but also improves the computational efficiency of the model, making the processing of large-scale data faster and more efficient, solving the problem of low efficiency in processing high-dimensional and large-volume data; and constructs the topological structure of the original data through the t-SNE algorithm, realizing data visualization, which enables researchers to intuitively observe and understand the distribution characteristics of the data, and then verify the pros and cons of the clustering effect. At the same time, the comparative analysis with the K-means clustering algorithm further enhances the scientificity and persuasiveness of the model verification;

[0018] The present invention is built based on a natural driving database, which means that the extracted driving style and recognition model are closer to the actual driving scene and have higher practical value. As the database is updated and expanded, the online recognition and verification module of the present invention can also be adjusted and optimized accordingly to adapt to more diverse driving style recognition needs, while also providing the system with real-time online recognition capabilities.

[0019] In summary, the present invention has the following beneficial effects:

[0020] 1. The present invention realizes a more comprehensive and accurate recognition of driving style by comprehensively considering the multivariate statistical characteristics of the original vehicle data and the vertical movement of the vehicle as the evaluation index of ride comfort. The present invention not only reflects the driving habits of the driver, but also reflects the riding experience of the passengers, thereby more realistically revealing the diversity of driving behavior; and uses the DBSCAN clustering algorithm to process the data after dimensionality reduction, improves the adaptability to complex data shapes and the robustness to noise data, and further improves the accuracy of driving style recognition;

[0021] 2. The present invention uses Laplace eigenmaps to reduce data dimensionality, effectively removing the redundant and strongly correlated parts in the feature data, improving the computational efficiency of subsequent cluster analysis, making the present invention more suitable for processing high-dimensional and large-volume driving data, and realizing rapid processing of large-scale data; and realizes data visualization through the t-SNE algorithm, providing researchers with an intuitive data analysis tool, enhancing the intuitiveness and persuasiveness of model verification;

[0022] 3. The present invention is constructed based on a natural driving database, which ensures that the recognition model is closely aligned with the actual driving scenario, thereby enhancing the practical value of the model. With the continuous updating and expansion of the database, the online recognition and verification module of the present invention can be flexibly adjusted and optimized to adapt to the ever-changing driving style recognition needs, while ensuring the real-time online recognition capability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is an overall flow chart of driving style recognition based on a natural driving database in an embodiment of the present invention. DETAILED DESCRIPTION

[0024] The following is combined with Figure 1 The present invention is described in further detail.

[0025] Embodiment: A driving style recognition system based on a natural driving database, such as Figure 1 As shown, it includes a feature extraction module, an offline clustering analysis module, and an online recognition and verification module.

[0026] 1) Feature parameter extraction module based on natural driving database

[0027] Taking the impact of vehicle comfort on driving behavior into account, driving style recognition is achieved by establishing a natural driving database including the vehicle's speed, longitudinal acceleration, lateral acceleration and vertical acceleration.

[0028] First, we collect raw signal data from the real-time operation of the vehicle through sensors, including but not limited to the longitudinal and lateral speeds, longitudinal acceleration, lateral acceleration and vertical acceleration of the vehicle. We perform preliminary data cleaning and preprocessing on these data to eliminate the impact of noise and outliers on subsequent analysis.

[0029] In the feature extraction stage, we calculate the statistical characteristics of each signal data: the mean is used to characterize the overall level of the signal, the variance is used to measure volatility, and the maximum and minimum values ​​are used to describe the extreme states of the signal. These primary features characterize the basic characteristics of driving behavior from different dimensions.

[0030] Next, we calculate the key feature parameters and obtain the maximum value of the vehicle power change by multiplying the speed and acceleration. This feature reveals the peak of the power intensity during driving. In addition, the acceleration zero rate (i.e., the frequency of acceleration changing from positive to negative or from negative to positive) provides a measure of driving smoothness, which can reflect the frequency and consistency of the driver's acceleration and deceleration. These advanced features and primary statistical features together constitute a more comprehensive data feature set.

[0031] In order to deal with the redundant information that may exist between multi-dimensional features, the present invention introduces Laplace eigenmap for dimensionality reduction processing. This step effectively identifies and removes high-correlation features by calculating the correlation matrix between features, and retains the features that have the greatest influence on driving style recognition. This not only simplifies the complexity of the feature set and reduces the amount of calculation, but also enhances the generalization ability and recognition accuracy of the model.

[0032] 2) Driving style offline recognition module based on DBSCAN algorithm

[0033] In order to realize the recognition of driving style, a driving style offline recognition module based on the DBSCAN algorithm was established. This module is suitable for processing high-dimensional and large-volume data. The specific process is as follows:

[0034] Firstly, the DBSCAN (Density--Based Spatial Clustering of Applications with Noise) algorithm is used to perform preliminary clustering analysis on the characteristic parameters of the natural driving database after dimensionality reduction processing. DBSCAN is particularly suitable for data with noise and clusters of irregular shapes.

[0035] In high-dimensional and large-volume environments, DBSCAN may have high computational complexity and cause problems with computer memory support and I / O consumption. Therefore, to solve this problem, we adopted a method of iteratively optimizing cluster centers. In this method, we first preliminarily identify cluster centers through DBSCAN, and then in subsequent iterations, we gradually optimize and adjust these cluster centers by exploring the relationship between new samples and the neighborhood of the current cluster center. This optimization not only helps improve the clustering effect without significantly increasing the computational burden, but also effectively reduces the pressure of sample points on memory.

[0036] Specifically, we define a neighborhood relationship. When a new sample point enters the system, we check its distance relationship with the existing cluster center. If the new sample belongs to the neighborhood of the existing cluster center, the parameters of the cluster are updated; otherwise, the new sample will be marked as noise or a potential new cluster center. Through this process, the initialized cluster center can more accurately describe the data distribution of different driving styles.

[0037] Through such iterative optimization, when realizing driving style clustering analysis, not only the efficiency and feasibility of the algorithm in high-dimensional and large-volume databases are improved, but also the accuracy and robustness of clustering are improved, laying a solid foundation for subsequent driving style identification and analysis.

[0038] 3) Driving style online identification and verification module based on DBSCAN algorithm

[0039] In the process of realizing online recognition and verification of driving style, we built an online recognition and verification module using the recognition model based on the DBSCAN algorithm. The following are the design and execution steps of the module:

[0040] In order to cope with the high-dimensional nature of the data, we apply Laplace eigentransformation in real time to reduce the dimensionality of the newly acquired driving data during the data collection phase. This process can reduce the dimensionality of the data and reduce the computational complexity while retaining the main feature information of the data for online real-time processing. After dimensionality reduction, we calculate the Euclidean distance between each data sample point and each cluster center in the pre-established driving style recognition model. Through these distance measurements, the relationship between the new data and each driving style (i.e., each cluster) is determined, thereby realizing online driving style recognition and classification.

[0041] To verify the clustering effect of the DBSCAN algorithm on data, we introduced the t-SNE (t-distributed Stochastic Neighbor Embedding) algorithm to perform dimensionality reduction and visualization processing on the collected data. t-SNE can map high-dimensional data into two-dimensional or three-dimensional space while retaining the topological structure of the data to intuitively display the data distribution and clustering boundaries between different driving styles. This helps to qualitatively evaluate the effect of DBSCAN clustering.

[0042] The clustering results based on DBSCAN are compared with the traditional k-means algorithm. By comparing the clustering effects of the two on the same data set, such as the determination of cluster centers, noise point processing, and adaptability to clusters of different shapes, the effectiveness and advantages of our proposed model in driving style recognition are further verified. k-means may not be as good as DBSCAN in processing irregularly shaped data, so this comparison can reflect the superiority of DBSCAN under complex data structures.

[0043] This specific embodiment is merely an explanation of the present invention and is not a limitation of the present invention. After reading this specification, those skilled in the art may make non-creative modifications to the present embodiment as needed. However, as long as they are within the scope of the claims of the present invention, they are protected by the patent law.

Claims

1. A driving style recognition system based on a natural driving database, characterized by: Included Features Extraction module, offline clustering analysis module, online identification and verification module; The feature extraction module obtains original vehicle data from the natural driving database, processes the collected original vehicle data, extracts the mean, variance, maximum value, and minimum value of the collected original vehicle data as corresponding feature parameters after processing, combines the key feature parameters to form a new data feature set, and performs dimensionality reduction processing through Laplace feature mapping; The offline clustering analysis module performs clustering analysis on the feature data after dimensionality reduction by using the DBSCAN algorithm to identify different driving style categories; The online recognition and verification module completes the online recognition of driving style by analyzing the relationship between the feature data after dimensionality reduction and the cluster center in the driving style recognition model.

2. The driving style recognition system based on the natural driving database according to claim 1, characterized in that: The original vehicle data includes the longitudinal speed, lateral speed, longitudinal acceleration, lateral acceleration and vertical acceleration of the vehicle, and the mean, variance, maximum value and minimum value of the above data are extracted as corresponding characteristic parameters.

3. The driving style recognition system based on the natural driving database according to claim 1, characterized in that: The key characteristic parameters include the maximum value of the product of velocity and acceleration and the acceleration zero-crossing rate. The maximum value of the product of velocity and acceleration includes the maximum value of the product of longitudinal velocity and longitudinal acceleration and the product of lateral velocity and lateral acceleration. The acceleration zero-crossing rate includes the zero-crossing rate of longitudinal acceleration, lateral acceleration and vertical acceleration.

4. The driving style recognition system based on the natural driving database according to claim 1, characterized in that: The offline clustering analysis module specifically selects representative data for initial clustering to determine the initial cluster center, and gradually adds new data feature sample points on this basis to iteratively optimize the clustering results to achieve efficient clustering analysis of large-volume databases.

5. The driving style recognition system based on the natural driving database according to claim 1, characterized in that: The online identification and verification module also visualizes the data through the t-SNE algorithm, displays the topological structure of the feature data, and verifies the clustering effect of the DBSCAN clustering algorithm.

6. The driving style recognition system based on the natural driving database according to claim 1, characterized in that: The online recognition and verification module also verifies the effectiveness of the recognition system through comparative analysis with the k-means clustering algorithm.

Images