Bad driving behavior feedback intervention method and device based on driving behavior big data

Abstract

The invention relates to the technical field of traffic safety, in particular to a bad driving behavior feedback intervention method and device based on driving behavior big data. Vehicle operation related data, driving environment data and driver behavior and physiological data are acquired. After original data is preprocessed, key features are extracted, a driving behavior feature database is constructed, and a driving behavior mode is generated. The bad driving behaviors are judged in a mode of combining rule judgment, a machine learning algorithm and mode matching, the rule judgment is based on historical big data statistical analysis, corresponding intervention measures are taken for different types of bad driving behaviors, and driving behavior change data after intervention are recorded and transmitted back to a cloud end to optimize a subsequent strategy. According to the method, multiple technologies are fused, high-precision judgment and personalized feedback intervention on bad driving behaviors are achieved, the method has the beneficial effects of real-time performance, high accuracy, personalized adaptation, multi-mode feedback, privacy protection, data safety and the like, and the driving safety can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic safety, and particularly to a method and device for feedback intervention of bad driving behaviors based on big data of driving behaviors. Background Art

[0002] In the current field of transportation, with the continuous increase in the vehicle ownership, the traffic safety situation has become increasingly severe. Traditional means of monitoring driving behaviors mainly rely on a single data source, such as simply tracking the driving trajectory of a vehicle based on GPS data, or conducting post-event analysis based on relevant data after an accident occurs. However, this approach has many drawbacks.

[0003] On the one hand, the monitoring dimensions are extremely limited, and only partial driving information can be obtained from a specific angle, unable to comprehensively and synthetically consider the complex behavior performance of a driver during actual driving. For example, it is difficult to accurately determine whether a driver has distracted operations or is fatigued while in the vehicle only relying on GPS, because it cannot obtain the operation conditions inside the vehicle and the physiological and behavioral state information of the driver.

[0004] On the other hand, the timeliness of feedback is severely lagged. Usually, data collection and analysis are carried out only after an accident occurs. At this time, the bad driving behavior has already caused serious consequences, and it is impossible to give the driver a warning and correction in a timely manner during driving, making it difficult to effectively prevent accidents from occurring.

[0005] In addition, the intervention measures lack pertinence and effectiveness. Since the types and degrees of bad driving behaviors cannot be accurately identified, the intervention means adopted are often general and cannot be adjusted individually according to the individual differences of different drivers and specific driving scenarios, making it difficult to truly play the role of intervention and reduce the incidence of bad driving behaviors.

[0006] In summary, the existing methods for monitoring and intervening in driving behaviors can no longer meet the growing traffic safety needs. There is an urgent need for an innovative system and method that can combine multi-source data and use advanced technologies to achieve comprehensive, real-time, and accurate monitoring and effective intervention of driving behaviors, thereby enhancing the road traffic safety guarantee ability. Summary of the Invention

[0007] In view of the deficiencies of the currently related existing technologies, the present invention provides a method and device for feedback intervention of bad driving behaviors based on big data of driving behaviors, which uses multi-source driving behavior data to achieve accurate and real-time identification and classification of bad driving behaviors, provides personalized and diversified feedback based on the driver's historical records and preferences to help them correct in a timely manner, and dynamically optimizes the intervention strategy based on behavior data and feedback effects to improve the intervention effectiveness, reduce the incidence of traffic accidents, ensure the safety of road users, and overcome the defects of insufficient monitoring dimensions, untimely feedback, and lack of pertinence in the existing technologies.

[0008] To achieve the above object, a first aspect of the present invention provides a method for feedback intervention of bad driving behaviors based on driving behavior big data, including the following steps:

[0009] Multi-source data collection step: Use the built-in data collection interface of the vehicle to collect vehicle operation-related data, collect driving environment data with the help of positioning and environment perception devices, and collect driver behavior and physiological data through image collection and wearable devices;

[0010] Data processing and analysis step: Preprocess the collected raw data to remove noise and outliers, extract key features including vehicle operation dynamics, driving environment conditions, driver operations and physiological states, and summarize a large amount of driving behavior data through a cloud platform to build a driving behavior feature database, and use data mining algorithms to generate driving behavior patterns;

[0011] Bad driving behavior determination step: Adopt a combination of rule determination based on historical data statistical analysis, machine learning algorithms with multi-source data as input, and pattern matching methods for building models based on the driver's historical behavior to determine whether the driving behavior is bad;

[0012] Personalized feedback step: According to the type and severity of bad driving behaviors, provide feedback to the driver through one or more feedback channels of vision, voice, and touch, and dynamically adjust the intensity and form of the feedback based on the driver's historical driving data and behavior tendencies;

[0013] Intervention strategy implementation and optimization step: Implement corresponding intervention measures for different types of bad driving behaviors, and record the driving behavior change data after the intervention and transmit it back to the cloud for continuously optimizing subsequent intervention strategies.

[0014] Further, in the multi-source data collection step, collecting vehicle operation-related data includes collecting one or more of vehicle speed, acceleration, brake depth, throttle depth, clutch depth, steering wheel angle, gear position, and engine speed through the vehicle bus;

[0015] Collecting driving environment data includes collecting one or more of real-time position, driving trajectory, and road speed limit through GPS; obtaining surrounding environment data through external sensors;

[0016] Collecting driver behavior and physiological data includes capturing one or more behavior characteristics of the driver's facial expression, head posture, and line of sight direction by a camera; recording the driver's physiological signals by a wearable device, at least including heart rate and fatigue level.

[0017] Further, in the data processing and analysis step, the vehicle operation dynamics at least include the number of accelerations, the intensity of hard braking, and the lane departure amplitude; the driving environment conditions at least include the ratio of the current vehicle speed to the speed limit and the rate of change of the distance from the vehicle ahead; the driver's operations and physiological states at least include one or more behavioral characteristics such as whether the driver's line of sight deviates from the front, whether the facial expression shows fatigue, whether the driver looks down to shift gears, and whether the driver steers the wheel with one hand.

[0018] Further, in the step of determining bad driving behaviors, the classification of bad driving behaviors at least includes distracted driving, drowsy driving, and dangerous driving; among them, the judgment steps are as follows:

[0019] Rule judgment: Based on the statistical analysis of historical big data, such as speeding judgment: current speed > road speed limit + 20, if this condition is met, it is determined as speeding dangerous driving;

[0020] Hard acceleration judgment: the change amount of acceleration within a short time > 3m / s 2 , the time window is 2 seconds; lane departure judgment: the deviation angle of the vehicle from the lane line > 15° and the duration > 1.5 seconds;

[0021] Drowsy driving judgment: the detected eye - closing time of the facial state exceeds 1.5 seconds and there are frequent dozes;

[0022] Machine learning model: Use the random forest algorithm to classify bad driving behaviors. The input features include vehicle data, driving environment data, and driver behavior data. Use a large amount of driving behavior data to train the model, determine the optimal number of trees, depth, and splitting criterion through grid search, determine the category of driving behavior, and output the probability value of bad behavior;

[0023] Behavior pattern matching: Construct a behavior model based on the driver's historical data, and match the real - time behavior with the historical behavior pattern to determine abnormal behaviors; specifically: collect the driver's long - term historical driving data, construct a driving behavior model based on time - series features; collect the speed change pattern, the frequency of hard acceleration and hard braking, and the lane - keeping stability features, use a time - series model (LSTM) to extract the time - dependence of driving behaviors, match the real - time driving behavior data with the historical behavior data, and calculate the similarity: 1 - |current behavior pattern - historical behavior pattern| > 0.8, then it is determined as an abnormal behavior.

[0024] Further, in the data processing and analysis step, the data summary and storage operations are as follows:

[0025] Construction of the individual feature database: For each driver, build a personalized model by recording and analyzing their long - term personal driving data, covering regular behavior patterns and abnormal behavior patterns;

[0026] Global feature database construction: Summarize large-scale driver behavior data, extract the behavior patterns and distribution characteristics of all drivers, use this as the core to provide a global reference benchmark, and generate a baseline model for judgment, including the average hard braking frequency distribution of drivers in different road scenarios, and information on common driving behavior differences in specific traffic environments in urban or rural road areas;

[0027] The data mining and behavior pattern generation process is as follows:

[0028] Regular behavior pattern generation: Use statistical analysis and clustering algorithms to process historical driving data of drivers. Input features such as vehicle speed range, braking force, acceleration change, and lane keeping offset amplitude. First, clean and normalize the data to remove outliers, then extract features and reduce dimensions, and finally run the clustering algorithm to classify the behavior data, and output a personalized regular driving behavior template for each driver containing vehicle speed-time distribution, braking force range, and steering frequency;

[0029] Abnormal behavior pattern generation: Use association rule algorithms and anomaly detection algorithms to mine potential abnormal patterns; construct a driving behavior feature vector containing elements such as hard acceleration amplitude, lane offset angle, and speeding duration. Use the association rule algorithm to analyze the combination of behavior features, use the isolation forest algorithm to detect abnormal points in driving behavior data, mark the behavior patterns that deviate from the global feature distribution, and output an abnormal behavior rule table covering abnormal behavior feature combinations and their corresponding probabilities.

[0030] Furthermore, it also includes the specific implementation details of the big data processing method as follows:

[0031] Data processing steps and algorithms:

[0032] Data cleaning: Use imputation methods based on mean replacement or time series interpolation to clean missing values, outliers, and sensor noise;

[0033] Feature extraction: Extract core driving features from the original data, including vehicle features, environmental features, and driver features;

[0034] Feature normalization: Use Min-Max standardization to scale the data to the [0,1] interval;

[0035] Feature storage: Store the cleaned and normalized feature data in individual and global databases according to a hierarchical structure;

[0036] Output data format:

[0037] The data is stored in a multi-dimensional feature table, and each row corresponds to a driving behavior record;

[0038] The generated rules are stored as a rule condition table, which details the conditions and rules of the behavior patterns;

[0039] The application method of the processed data is as follows:

[0040] Support personalized determination: Combine the individual feature database to customize the determination criteria that suit the driver's driving habits; for example, if a driver's hard braking force threshold is higher than the group average, the system can dynamically adapt the abnormal determination standard for him;

[0041] Optimize the global model: The global feature database supplies the basic rules and behavior baselines for the determination module, regularly integrates new driving data into the global database, optimizes the model parameters, and improves the determination accuracy.

[0042] The second aspect of the present invention provides a device for feedback intervention of bad driving behaviors based on driving behavior big data. This device is used to implement the above method, including:

[0043] Data acquisition unit: Used to execute the multi-source data acquisition step, collect vehicle operation-related data using the built-in data acquisition interface of the vehicle, collect driving environment data with the help of positioning and environment perception devices, and collect driver behavior and physiological data through image acquisition and wearable devices;

[0044] Data processing and analysis unit: Preprocess the raw data collected by the acquisition unit to remove noise and outliers, extract key features including vehicle operation dynamics, driving environment conditions, driver operations and physiological states, and summarize a large amount of driving behavior data through the cloud platform to build a driving behavior feature database, and use data mining algorithms to generate driving behavior patterns;

[0045] Bad driving behavior determination unit: Adopt a combination of rule determination based on historical data statistical analysis, machine learning algorithms with multi-source data as input, and pattern matching methods for building models based on the driver's historical behavior to determine whether the driving behavior is bad;

[0046] Personalized feedback unit: According to the type and severity of bad driving behaviors, provide feedback to the driver through one or more feedback channels such as vision, voice, and touch, and dynamically adjust the intensity and form of the feedback according to the driver's historical driving data and behavior tendencies;

[0047] Intervention strategy implementation unit: Implement corresponding intervention measures for different types of bad driving behaviors, record the changed data of driving behaviors after the intervention, and transmit it back to the cloud for continuously optimizing subsequent intervention strategies;

[0048] System architecture unit: It includes vehicle-end devices, cloud platforms, and edge computing units. Among them, the vehicle-end devices deploy real-time judgment and feedback functions to ensure the real-time performance and efficiency of the system; the cloud platform conducts large-scale data training and analysis to optimize model parameters; the edge computing unit realizes real-time data preprocessing and preliminary judgment through in-vehicle computing units.

[0049] Furthermore, the data acquisition unit includes:

[0050] Vehicle data acquisition sub-unit: It acquires one or more types of data such as vehicle speed, acceleration, brake depth, throttle depth, clutch depth, steering wheel angle, gear position, and engine speed through the vehicle bus;

[0051] Driving environment data acquisition sub-unit: It acquires one or more types of real-time position, driving trajectory, and road speed limit through GPS, and obtains surrounding environment data through external sensors;

[0052] Driver data acquisition sub-unit: The camera captures one or more behavioral characteristics such as the driver's facial expression, head posture, and line of sight direction, and the wearable device records the driver's physiological signals, at least including heart rate and fatigue level.

[0053] Furthermore, the data processing and analysis unit includes:

[0054] Data cleaning and preprocessing sub-unit: It is used to clean missing values, outliers, and sensor noise in the acquired data;

[0055] Feature extraction sub-unit: It extracts vehicle features such as hard brake intensity, acceleration change, and steering wheel rotation angle from the original data, environmental features such as the ratio of vehicle speed to speed limit and the change rate of the distance to the vehicle in front, and driver features such as the driver's head posture and line of sight direction;

[0056] Big data processing sub-unit: It aggregates data through the cloud platform to build a driving behavior feature database, and uses data mining algorithms to generate driving behavior patterns, including generating regular behavior patterns by using statistical analysis and clustering algorithms, generating abnormal behavior patterns by using association rule algorithms and anomaly detection algorithms, and storing the data in a multi-dimensional feature table and a rule condition table after Min-Max normalization processing.

[0057] Furthermore, in the system architecture unit:

[0058] The vehicle-end device includes an in-vehicle processor and a storage device. The in-vehicle processor is used to run the real-time judgment and feedback program, and the storage device is used to store temporary data and some key algorithms;

[0059] The cloud platform includes a large-scale data storage server and a high-performance computing cluster, which are used to store massive driving behavior data and perform complex data training and analysis;

[0060] The edge computing unit includes in-vehicle edge computing chips and related software modules, which implement real-time data preprocessing and preliminary determination functions, and data is transmitted between the vehicle-side device, the cloud platform, and the edge computing unit through a secure communication network.

[0061] Through the integration of rule determination, machine learning models, and behavior pattern matching, the present invention realizes high-precision determination and personalized feedback intervention for bad driving behaviors, and has the following beneficial effects:

[0062] 1. Real-time and high accuracy: Combining cloud-edge collaborative computing and machine learning algorithms to achieve fast and efficient behavior determination.

[0063] 2. Personalized adaptation: Dynamically adjust system parameters through behavior pattern matching and adaptive learning to adapt to the habits of different drivers.

[0064] 3. Multimodal feedback: Provide feedback through multiple methods such as voice, vision, and touch to enhance the intervention effect.

[0065] 4. Privacy protection and data security: Adopt federated learning and data encryption technologies to ensure the data security and privacy of drivers. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0067] Figure 1 It is a flowchart of the method for feedback intervention of bad driving behaviors of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0068] Here, the exemplary embodiments will be described in detail, and the examples are shown in the drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. On the contrary, they are only examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0069] Such as Figure 1As shown in the figure, the first aspect of this embodiment provides a method for feedback intervention of bad driving behaviors based on driving behavior big data, including the following steps:

[0070] Multi-source data collection step: Use the built-in data collection interface of the vehicle to collect data related to vehicle operation, collect driving environment data with the help of positioning and environment perception devices, and collect driver behavior and physiological data through image collection and wearable devices;

[0071] Data processing and analysis step: Preprocess the collected raw data to remove noise and outliers, extract key features including vehicle operation dynamics, driving environment conditions, driver operations and physiological states, and summarize a large amount of driving behavior data through a cloud platform to build a driving behavior feature database, and use data mining algorithms to generate driving behavior patterns;

[0072] Bad driving behavior determination step: Adopt a combination of rule determination based on historical data statistical analysis, machine learning algorithms with multi-source data as input, and pattern matching methods for building models based on the driver's historical behavior to determine whether the driving behavior is bad;

[0073] Personalized feedback step: According to the type and severity of bad driving behaviors, provide feedback to the driver through one or more feedback channels of vision, voice, and touch, and dynamically adjust the intensity and form of the feedback based on the driver's historical driving data and behavior tendencies;

[0074] Intervention strategy implementation and optimization step: Implement corresponding intervention measures for different types of bad driving behaviors, and record the changed data of driving behaviors after the intervention and transmit it back to the cloud for continuously optimizing subsequent intervention strategies.

[0075] As a preferred implementation method, in the multi-source data collection step of this embodiment, collecting data related to vehicle operation includes collecting one or more of vehicle speed, acceleration, brake depth, throttle depth, clutch depth, steering wheel angle, gear position, and engine speed through the vehicle bus;

[0076] Collecting driving environment data includes collecting one or more of real-time position, driving trajectory, and road speed limit through GPS; obtaining surrounding environment data through external sensors;

[0077] Collecting driver behavior and physiological data includes capturing one or more behavior characteristics such as the driver's facial expression, head posture, and line of sight direction by a camera; the wearable device records the driver's physiological signals, including at least heart rate and fatigue.

[0078] As a preferred embodiment, in the data processing and analysis step of this embodiment, the vehicle running dynamics at least include the number of accelerations, the intensity of emergency braking, and the lane departure amplitude; the driving environment conditions at least include the ratio of the current vehicle speed to the speed limit and the change rate of the distance from the vehicle ahead; the driver's operations and physiological states at least include one or more behavioral characteristics such as whether the driver's line of sight deviates from the front, whether the facial expression shows fatigue, whether the driver looks down to shift gears, and whether the driver steers the wheel with one hand.

[0079] As a preferred embodiment, in the step of determining bad driving behaviors of this embodiment, the classification of bad driving behaviors at least includes distracted driving, such as not looking ahead for a long time, operating a mobile phone, etc.

[0080] Drowsy driving, such as frequent eye closing, slow reaction, frequent dozing, etc.

[0081] And dangerous driving, such as speeding, sudden acceleration, emergency braking, changing lanes without turning on the turn signal, not wearing a seat belt, etc.

[0082] Among them, the judgment steps are as follows:

[0083] Rule judgment: Based on the statistical analysis of historical big data, for example, speeding judgment: current speed > road speed limit + 20, if this condition is met, it is determined as speeding dangerous driving;

[0084] Sudden acceleration judgment: the change amount of acceleration within a short time > 3m / s 2 , the time window is 2 seconds; lane departure judgment: the deviation angle between the vehicle and the lane line > 15° and the duration > 1.5 seconds;

[0085] Drowsy driving judgment: the detected eye closing time on the facial state exceeds 1.5 seconds, with frequent dozing;

[0086] Machine learning model: Use the random forest algorithm to classify bad driving behaviors. The input features include vehicle data, driving environment data, and driver behavior data. Use large-scale driving behavior data to train the model, and determine the optimal number of trees, depth, and splitting criterion through grid search, determine the category of driving behavior, and output the probability value of bad behaviors;

[0087] Behavior pattern matching: Construct a behavior model based on the driver's historical data, and match the real-time behavior with the historical behavior pattern to determine abnormal behaviors; specifically: collect the driver's long-term historical driving data, construct a driving behavior model based on time series features; collect the speed change pattern, the frequency of sudden acceleration and emergency braking, and the lane keeping stability features, use the time series model (LSTM) to extract the time dependence of driving behaviors, match the real-time driving behavior data with the historical behavior data, and calculate the similarity: 1 - |current behavior pattern - historical behavior pattern| > 0.8, then it is determined as an abnormal behavior.

[0088] Personalized module feedback;

[0089] Dynamic feedback mechanism;

[0090] According to the severity of the type of bad behavior, provide diverse feedback methods:

[0091] Visual feedback: Present specific warning information through the in-vehicle display screen (such as "You are speeding by 20%").

[0092] Voice broadcast: Play voice warnings, such as "Please stay focused".

[0093] Haptic feedback: Remind the driver of potential dangers through the vibration of the steering wheel or seat.

[0094] Personalized adjustment;

[0095] According to the driver's historical driving records and behavior preferences, adjust the intensity and form of feedback. For example, for drivers with a more serious habit of driving while fatigued, increase the frequency of haptic feedback.

[0096] Intervention strategy module;

[0097] Intervention plan design;

[0098] For different bad driving behaviors, adopt differentiated intervention strategies:

[0099] Distracted driving: The system plays a warning sound in real time and pauses the entertainment system (such as music playback).

[0100] Fatigued driving: Remind the driver to rest and accompany with visual warnings.

[0101] Dangerous driving: Reduce the vehicle speed by gentle braking.

[0102] Intervention effect evaluation;

[0103] Record the changes in driving behavior after each intervention and upload the data to the cloud to optimize subsequent intervention strategies.

[0104] As a preferred implementation method, in the data processing and analysis steps of this embodiment, the data summarization and storage operations are as follows:

[0105] Construction of individual characteristic database: For each driver, the individual database forms a personalized model through long-term recording and analysis of personal driving data, including regular behavior patterns (recording the typical behavior characteristics of the driver, such as the speed distribution range, braking frequency, throttle response time, etc.) and abnormal behavior patterns (setting personalized abnormal behavior thresholds based on personal historical data. For example, for a certain driver, the speeding behavior occurs when exceeding the speed limit by +30, rather than the general +20 standard).

[0106] Global feature database construction: Based on the aggregation of large-scale driver behavior data, extract the behavior patterns and distribution characteristics of all drivers. The core function of the global database is to provide a global reference standard and form a baseline model for judgment. For example, the average hard braking frequency distribution of drivers under different road scenarios, the differences in common driving behaviors in specific regional traffic environments (urban and rural roads), etc.

[0107] The data mining and behavior pattern generation process is as follows: Through a variety of algorithms and analysis steps, generate the regular behavior patterns and abnormal behavior patterns of drivers to support the subsequent bad driving behavior judgment module.

[0108] Regular behavior patterns:

[0109] Algorithm: Use statistical analysis and clustering algorithms (such as K-Means clustering) to cluster the historical driving data of drivers. The input features of the algorithm include speed range, braking force, acceleration change, lane keeping offset amplitude, etc.

[0110] Steps:

[0111] Clean and normalize the driving data to remove outliers (such as sensor noise).

[0112] Extract the driving data features and perform data dimensionality reduction (such as PCA).

[0113] Run the clustering algorithm to divide the driver's behavior data into different categories (such as normal driving, mild speeding, high-frequency hard braking, etc.).

[0114] Output data: The personalized regular driving behavior template for each driver, including speed-time distribution, braking force range, steering frequency, etc.

[0115] Abnormal behavior patterns

[0116] Algorithm: Use association rule algorithms and anomaly detection algorithms to identify potential abnormal patterns in driving behaviors.

[0117] Association rule algorithm: Used to analyze the combination relationship between behavior characteristics, such as whether hard braking and lane departure occur simultaneously.

[0118] Anomaly detection algorithm: Use the Isolation Forest algorithm to extract behavior characteristics that do not conform to the conventional pattern from the data.

[0119] Steps: Construct a driving behavior feature vector (such as hard acceleration amplitude, lane deviation angle, speeding duration).

[0120] Apply the association rule algorithm to analyze the combination of behavior characteristics (such as whether speeding behavior is accompanied by hard braking).

[0121] Perform outlier detection on driving behavior data and mark the behavior patterns that deviate from the global feature distribution.

[0122] Output data: Abnormal behavior rule table, including the combination of abnormal behavior features (such as "sudden acceleration + lane departure") and their corresponding probabilities.

[0123] As a preferred implementation, the specific implementation details of the big data processing method in this embodiment are as follows:

[0124] Data processing steps and algorithms:

[0125] Data cleaning: Clean missing values, outliers, and sensor noise. The algorithms include imputation methods based on mean replacement or time series interpolation.

[0126] Feature extraction: Extract the core driving features from the original data, such as:

[0127] Vehicle features: Sudden braking intensity, acceleration change, steering wheel rotation angle, etc.

[0128] Environmental features: Ratio of vehicle speed to speed limit, rate of change of the distance to the vehicle ahead, etc.

[0129] Driver features: Driver's head posture, line of sight direction, etc.

[0130] Feature normalization: Scale the data to the range [0, 1] through Min-Max normalization.

[0131] Feature storage: The cleaned and normalized feature data is stored hierarchically in the individual and global databases.

[0132] Output data format:

[0133] The data is stored as a multi-dimensional feature table (in the AnalysisDB database of Alibaba Cloud), and each row represents a driving behavior record.

[0134] The generated rules are stored as a rule condition table, describing the conditions and rules of the behavior patterns (such as "IF sudden acceleration > 3m / s 2 AND braking force > X THEN determined as abnormal behavior").

[0135] The application method of the processed data is as follows:

[0136] Support personalized determination: Combining the individual feature database, personalized determination criteria more in line with the driver's driving habits can be provided for the driver.

[0137] For example, if a driver's braking force threshold during sudden braking is usually higher than the group average, the system can dynamically adjust the abnormal determination criteria for him.

[0138] Optimize the global model: The global feature database provides the basic rules and behavior baselines (such as the group speeding determination threshold) for the determination module.

[0139] Regularly integrate new driving data into the global database, optimize the model parameters, and improve the determination accuracy.

[0140] The second aspect of the present invention provides a device for feedback intervention of bad driving behaviors based on driving behavior big data. This device is used to implement the above method and includes:

[0141] Data acquisition unit: Used to execute the multi-source data acquisition step, collect vehicle operation-related data using the built-in data acquisition interface of the vehicle, collect driving environment data with the help of positioning and environment perception devices, and collect driver behavior and physiological data through image acquisition and wearable devices;

[0142] Data processing and analysis unit: Preprocess the original data collected by the acquisition unit to remove noise and outliers, extract key features including vehicle operation dynamics, driving environment conditions, driver operations, and physiological states, and summarize a large amount of driving behavior data through the cloud platform to build a driving behavior feature database, and use data mining algorithms to generate driving behavior patterns;

[0143] Bad driving behavior determination unit: Adopt a combination of rule determination based on historical data statistical analysis, machine learning algorithms with multi-source data as input, and pattern matching methods based on models constructed according to the driver's historical behavior to determine whether the driving behavior is bad;

[0144] Personalized feedback unit: Provide feedback to the driver through one or more feedback channels of vision, voice, and touch according to the type and severity of the bad driving behavior, and dynamically adjust the intensity and form of the feedback based on the driver's historical driving data and behavior tendencies;

[0145] Intervention strategy implementation unit: Implement corresponding intervention measures for different types of bad driving behaviors, and record the changed driving behavior data after the intervention and send it back to the cloud for continuously optimizing subsequent intervention strategies;

[0146] System architecture unit: Includes vehicle-end devices, cloud platforms, and edge computing units. Among them, the vehicle-end devices deploy real-time determination and feedback functions to ensure the real-time performance and efficiency of the system; the cloud platform conducts large-scale data training and analysis to optimize the model parameters; the edge computing unit realizes real-time data preprocessing and preliminary determination through in-vehicle computing units.

[0147] Furthermore, the data acquisition unit includes:

[0148] Vehicle data acquisition sub-unit: Collect one or more types of data including vehicle speed, acceleration, brake depth, throttle depth, clutch depth, steering wheel angle, gear position, and engine speed through the vehicle bus;

[0149] Driving environment data acquisition sub-unit: Collect one or more of real-time position, driving trajectory, and road speed limit through GPS, and obtain surrounding environment data through external sensors;

[0150] Driver data acquisition sub-unit: The camera captures one or more behavioral characteristics such as the driver's facial expression, head posture, and line of sight direction, and the wearable device records the driver's physiological signals, including at least heart rate and fatigue level.

[0151] Furthermore, the data processing and analysis unit includes:

[0152] Data cleaning and preprocessing sub-unit: Used to clean missing values, outliers, and sensor noise in the collected data;

[0153] Feature extraction sub-unit: Extract vehicle features such as hard brake intensity, acceleration change, and steering wheel rotation angle from the original data, environmental features such as the ratio of vehicle speed to speed limit and the change rate of the distance to the vehicle ahead, and driver features such as the driver's head posture and line of sight direction;

[0154] Big data processing sub-unit: Aggregate data through the cloud platform to build a driving behavior feature database, and use data mining algorithms to generate driving behavior patterns, including generating regular behavior patterns using statistical analysis and clustering algorithms, generating abnormal behavior patterns using association rule algorithms and anomaly detection algorithms, and storing the data in a multi-dimensional feature table and rule condition table after Min-Max normalization processing.

[0155] Furthermore, in the system architecture unit:

[0156] The vehicle-end device includes an in-vehicle processor and a storage device. The in-vehicle processor is used to run the real-time determination and feedback program, and the storage device is used to store temporary data and some key algorithms;

[0157] The cloud platform includes a large-scale data storage server and a high-performance computing cluster, which are used to store massive driving behavior data and perform complex data training and analysis;

[0158] The edge computing unit includes an in-vehicle edge computing chip and related software modules, which realize the functions of real-time data preprocessing and preliminary determination, and data is transmitted between the vehicle-end device, the cloud platform, and the edge computing unit through a secure communication network.

[0159] In the field of driving behavior monitoring and intervention, the present invention brings significant improvements in many aspects.

[0160] In terms of the accuracy and timeliness of judgment, by integrating rule-based judgment, machine learning models, and behavior pattern matching, the ability to identify bad driving behaviors has been greatly enhanced. Rule-based judgment is based on precise conditions derived from the statistical analysis of a large amount of historical big data, such as clear quantitative criteria for speeding, sudden acceleration, lane departure, and fatigue driving. It can quickly screen out obvious signs of bad driving behaviors. The machine learning model uses the random forest algorithm. With its powerful classification ability, it is deeply trained by inputting a variety of vehicle data, driving environment data, and driver behavior data. It can handle complex driving scenarios and behavior patterns. The probability value of bad behaviors output provides a reliable reference basis for judgment, effectively making up for the limitations of single rule-based judgment. Behavior pattern matching relies on the historical data of drivers accumulated over a long time. By using the time series model (LSTM) to extract the time dependence of driving behaviors and through careful matching and similarity calculation with real-time behaviors, it can keenly capture subtle abnormal changes in drivers' behaviors. Even some uncommon or special bad driving behaviors are hard to hide. This multi-dimensional comprehensive judgment mechanism ensures that the system can quickly and accurately identify bad driving behaviors in the face of various driving situations, laying a solid foundation for subsequent timely intervention.

[0161] In terms of personalized services, the present invention demonstrates excellent adaptability. The individual feature database constructed for each driver details their unique behavior patterns during the driving process, including conventional behavior characteristics such as vehicle speed distribution range, braking frequency, and throttle response time, as well as personalized abnormal behavior thresholds set based on personal historical data. For example, the normal vehicle speed range of some drivers under specific road conditions may be different from the general standard. The system can make targeted judgments and provide feedback based on this, avoiding misjudgments or inappropriate feedback caused by a unified standard. At the same time, according to the driver's historical driving records and behavior preferences, the personalized feedback unit can precisely adjust the intensity and form when providing feedback. For drivers who are prone to fatigue, the tactile feedback frequency is increased or the tone and content of the voice prompt are adjusted to make it easier for them to accept and pay attention to the feedback information, thus more effectively helping drivers correct bad behaviors, improving driving safety, and enhancing the interaction experience and trust between the driver and the system.

[0162] The application of the multimodal feedback mechanism has greatly enhanced the intervention effect. Visual feedback intuitively displays specific and eye-catching warning messages such as "You are speeding by [X]%", "Lane departure, please adjust" etc. on the in-vehicle display screen, enabling the driver to clearly understand the types and severity of their bad driving behaviors in the first place and make a quick response. Voice announcements provide clear and explicit voice warnings, such as "Please stay focused, the road conditions ahead are complex", "You are driving fatigued, please rest as soon as possible" etc., timely conveying important reminder messages without disturbing the driver's line of sight, enhancing the timeliness and effectiveness of the feedback. Haptic feedback uses the vibration of the steering wheel or seat to give the driver a strong physical tactile stimulus in case of some emergency or critical bad driving behaviors, such as when there is a risk of collision or severe speeding etc., quickly arousing the driver's high alertness and prompting them to immediately take corrective measures. The three feedback methods complement each other, comprehensively covering the driver's perception range, effectively increasing the driver's attention to bad driving behaviors and willingness to correct them, and greatly reducing the risk of traffic accidents caused by bad driving behaviors.

[0163] At the data processing and system architecture level, the data collection and feature extraction module comprehensively collects data from multiple data sources, covering all key parameters of vehicle operation, detailed information of the driving environment, as well as the driver's behavior and physiological state, providing rich and comprehensive data support for the system. Big data processing technology efficiently aggregates, stores and deeply mines these massive data. The constructed driving behavior feature database not only contains personalized models of individual drivers, but also extracts the behavior patterns and distribution characteristics of all drivers, providing strong data support and decision-making basis for the accurate operation of the entire system. The cloud-vehicle collaborative system architecture design gives full play to the advantages of all parties. The real-time judgment and feedback function of the vehicle-side device ensures the immediate response of the system during the driving process, meeting the real-time requirements; the cloud platform, relying on its powerful computing power and large-scale data storage capacity, conducts in-depth data training and analysis, continuously optimizing the model parameters and improving the overall performance and accuracy of the system; the edge computing unit realizes real-time data preprocessing and preliminary judgment through the in-vehicle computing unit, reducing the burden on the cloud and vehicle sides, improving the data processing efficiency, and ensuring the smooth operation of the system.

[0164] In terms of privacy protection, encryption technology is used to encrypt the uploaded data, effectively preventing the data from being stolen or tampered with during the transmission process, ensuring the confidentiality of the driver's data. The application of federated learning technology is even more of an innovation. It realizes the joint optimization of the judgment model without sharing the original data, not only making full use of the value of multi-source data, but also maximizing the protection of the driver's privacy and security, enabling the driver to have no worries when using the present invention, and enhancing the credibility and acceptability of the system.

[0165] The present invention has significant advantages in the monitoring, feedback, and intervention of bad driving behaviors, can effectively improve driving safety, and provides an efficient, intelligent, and reliable solution for road traffic safety management.

[0166] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A bad driving behavior feedback intervention method based on driving behavior big data, characterized by: The following steps are involved: Multi-source data collection steps: Use the vehicle's built-in data collection interface to collect vehicle operation-related data, use positioning and environmental perception equipment to collect driving environment data, and use image acquisition and wearable devices to collect driver behavior and physiological data; Data processing and analysis steps: Preprocess the collected raw data to remove noise and outliers, extract key features including vehicle operation dynamics, driving environment conditions, driver operation and physiological state, and aggregate a large amount of driving behavior data through the cloud platform to build a driving behavior feature database, and use data mining algorithms to generate driving behavior patterns; Bad driving behavior determination steps: Use a combination of rule determination based on historical data statistical analysis, machine learning algorithms with multi-source data as input, and pattern matching methods that build models based on the driver's historical behavior to determine whether the driving behavior is bad; Personalized feedback step: Provide feedback to the driver through one or more feedback channels such as vision, voice, and touch based on the type and severity of the bad driving behavior, and dynamically adjust the intensity and form of the feedback based on the driver's historical driving data and behavioral tendencies; Intervention strategy implementation and optimization steps: Implement corresponding intervention measures for different types of bad driving behaviors, and record the driving behavior change data after the intervention and send it back to the cloud for continuous optimization of subsequent intervention strategies.

2. The method according to claim 1, characterized in that: In the multi-source data collection step, collecting vehicle operation related data includes collecting one or more data of vehicle speed, acceleration, brake depth, throttle depth, clutch depth, steering wheel angle, gear position and engine speed through a vehicle bus; Collecting driving environment data includes collecting one or more of real-time location, driving trajectory and road speed limit through GPS; obtaining surrounding environment data through external sensors; Collecting driver behavior and physiological data includes capturing one or more behavioral characteristics of the driver's facial expression, head posture and sight direction by a camera; Wearable devices record the driver's physiological signals, including at least heart rate and fatigue.

3. The method according to claim 1, characterized in that: In the data processing and analysis steps, the vehicle operation dynamics include at least the number of accelerations, the intensity of emergency braking and the lane deviation range; the driving environment conditions include at least the ratio of the current vehicle speed to the speed limit and the rate of change of the distance to the vehicle in front; the driver's operation and physiological state include at least one or more behavioral characteristics such as whether the driver's line of sight deviates from the front, whether the facial expression shows fatigue, whether the driver lowers his head to shift gears and steers the wheel with one hand.

4. The method according to claim 1, characterized in that: In the bad driving behavior determination step, the bad driving behavior classification includes at least distracted driving, fatigue driving and dangerous driving; wherein the determination step is: Rule determination: Based on historical big data statistical analysis, such as speeding judgment: current speed> road speed limit+20, if this condition is met, it is judged as speeding and dangerous driving; Rapid acceleration judgment: acceleration change in a short period of time> 3m / s 2 ,The time window is 2 seconds; Lane departure judgment: The deviation angle between the vehicle and the lane line is greater than 15° and the duration is greater than 1.5 seconds; Fatigue driving judgment: facial status check shows eyes closed for more than 1.5 seconds, frequent dozing off; Machine learning model: Use the random forest algorithm to classify bad driving behaviors. Input features include vehicle data, driving environment data, and driver behavior data. Use large-scale driving behavior data to train the model. Use grid search to determine the number, depth, and splitting criteria of the best trees, determine the driving behavior category, and output the probability value of bad behavior. Behavior pattern matching: Build a behavior model based on the driver's historical data, match the real-time behavior with the historical behavior pattern, and determine abnormal behavior; specifically: collect the driver's long-term historical driving data, build a driving behavior model based on time series features; collect speed change patterns, sudden acceleration and braking frequency, lane keeping stability features, use a time series model (LSTM) to extract the time dependency of driving behavior, match the real-time driving behavior data with the historical behavior data, and calculate the similarity: 1-|current behavior pattern-historical behavior pattern|>0.8, then it is determined to be abnormal behavior.

5. The method according to any one of claims 1 to 4, characterized in that: The data processing and analysis steps include data aggregation and storage operations as follows: Individual feature database construction: For each driver, a personalized model is created by recording and analyzing their long-term personal driving data, covering both normal and abnormal behavior patterns; Global feature database construction: Aggregate large-scale driver behavior data, refine the behavior patterns and distribution characteristics of all drivers, and use this as the core to provide a global reference benchmark to generate a baseline model for judgment, including the average frequency distribution of emergency braking of drivers under different road scenarios, and common driving behavior differences in specific traffic environments in urban or rural road areas; The data mining and behavior pattern generation process is: Generate regular behavior patterns: Use statistical analysis and clustering algorithms to process drivers’ historical driving data, input speed range, braking force, acceleration change, lane keeping deviation amplitude features, first clean and normalize the data to remove outliers, then extract features and reduce dimension), and finally run clustering algorithms to classify the behavior data, and output each driver’s personalized regular driving behavior template containing speed-time distribution, braking force range, and steering frequency content; Abnormal behavior pattern generation: Association rule algorithms and anomaly detection algorithms are used to mine potential abnormal patterns; a driving behavior feature vector containing elements such as sudden acceleration amplitude, lane deviation angle, and speeding duration is constructed, and the combination of behavioral features is analyzed with the help of association rule algorithms. The isolation forest algorithm is used to detect abnormal points in driving behavior data, mark behavior patterns that deviate from the global feature distribution, and output an abnormal behavior rule table covering abnormal behavior feature combinations and their corresponding probabilities.

6. The method according to claim 5, characterized in that: The specific implementation details of the big data processing method are as follows: Data processing steps and algorithms: Data cleaning: Use interpolation methods based on mean replacement or time series interpolation to clean up missing values, outliers, and sensor noise; Feature extraction: extracting core driving features from raw data, including vehicle features, environmental features, and driver features; Feature normalization: Min-Max normalization is used to scale the data to the [0,1] interval; Feature storage: The cleaned and normalized feature data is stored in individual and global databases according to the hierarchical structure; Output data format: The data is stored in a multidimensional feature table, where each row corresponds to a driving behavior record; The generated rules are stored as a rule condition table, which describes the conditions and rules of the behavior pattern in detail; The processed data is applied as follows: Support personalized judgment: Combined with the individual feature database, the system can tailor judgment criteria that suit the driver's driving habits. For example, if a driver's emergency braking force threshold is higher than the group average, the system can dynamically adapt the abnormal judgment criteria for him. Optimize the global model: The global feature database provides basic rules and behavior baselines for the judgment module, regularly integrates new driving data into the global database, optimizes model parameters, and improves judgment accuracy.

7. A bad driving behavior feedback intervention device based on driving behavior big data, characterized in that: The device is used to implement the method described in any one of claims 1 to 6, including: Data acquisition unit: used to perform multi-source data acquisition steps, using the vehicle's built-in data acquisition interface to collect vehicle operation-related data, using positioning and environmental perception equipment to collect driving environment data, and collecting driver behavior and physiological data through image acquisition and wearable devices; Data processing and analysis unit: pre-processes the raw data collected by the acquisition unit to remove noise and outliers, extracts key features including vehicle operation dynamics, driving environment conditions, driver operation and physiological status, and aggregates a large amount of driving behavior data through the cloud platform to build a driving behavior feature database, and uses data mining algorithms to generate driving behavior patterns; Bad driving behavior determination unit: It uses a combination of rule determination based on historical data statistical analysis, machine learning algorithms with multi-source data as input, and pattern matching methods that build models based on the driver's historical behavior to determine whether the driving behavior is bad; Personalized feedback unit: Provides feedback to the driver through one or more feedback channels such as vision, voice, and touch according to the type and severity of bad driving behavior, and dynamically adjusts the intensity and form of feedback based on the driver's historical driving data and behavioral tendencies; Intervention strategy implementation unit: implements corresponding intervention measures for different types of bad driving behaviors, and records the driving behavior change data after intervention and transmits it back to the cloud for continuous optimization of subsequent intervention strategies; System architecture unit: includes vehicle-side equipment, cloud platform and edge computing unit. The vehicle-side equipment deploys real-time judgment and feedback functions to ensure the real-time and high efficiency of the system; the cloud platform conducts large-scale data training and analysis to optimize model parameters; the edge computing unit realizes real-time data preprocessing and preliminary judgment through the vehicle-mounted computing unit.

8. The device according to claim 7, characterized in that The data acquisition unit comprises: Vehicle data collection subunit: collects one or more data including vehicle speed, acceleration, brake depth, throttle depth, clutch depth, steering wheel angle, gear position and engine speed through the vehicle bus; Driving environment data collection subunit: collects one or more of the real-time position, driving trajectory and road speed limit through GPS, and obtains surrounding environment data through external sensors; Driver data collection subunit: The camera captures one or more behavioral characteristics of the driver's facial expression, head posture and line of sight direction, and the wearable device records the driver's physiological signals, including at least heart rate and fatigue.

9. The device according to claim 8, characterized in that The data processing and analysis unit comprises: Data cleaning and preprocessing subunit: used to clean up missing values, outliers and sensor noise in collected data; Feature extraction subunit: extracts vehicle features from raw data such as emergency braking intensity, acceleration change, steering wheel rotation angle, environmental features such as the ratio of vehicle speed to speed limit, rate of change of distance to the vehicle in front, and driver features such as driver's head posture and line of sight direction; Big data processing subunit: Aggregate data through the cloud platform to build a driving behavior feature database, and use data mining algorithms to generate driving behavior patterns, including using statistical analysis and clustering algorithms to generate regular behavior patterns, using association rule algorithms and anomaly detection algorithms to generate abnormal behavior patterns, and performing Min-Max normalization on the data before storing it in a multidimensional feature table and rule condition table.

10. The device according to claim 9, characterized in that In the system architecture unit: The vehicle-side equipment includes an on-board processor and a storage device. The on-board processor is used to run real-time judgment and feedback programs, and the storage device is used to store temporary data and some key algorithms. The cloud platform includes large-scale data storage servers and high-performance computing clusters for storing massive amounts of driving behavior data and conducting complex data training and analysis; The edge computing unit includes an on-board edge computing chip and related software modules to realize real-time data preprocessing and preliminary judgment functions, and data is transmitted between the vehicle-side equipment, cloud platform and edge computing unit through a secure communication network.

Images