AIoT-based driving behavior identification and risk assessment system
Vehicle data is collected and analyzed through the AIoT system, and driving risk assessment is carried out in combination with entropy weight method and hierarchical analysis method, which solves the problem that existing technology is difficult to objectively evaluate driving risks, and improves the accuracy of traffic safety and insurance assessment.
Patent Information
- Application Number
- CN202411162562.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-23
- Publication Date
- 2024-12-03
AI Technical Summary
Existing on-board sensors are difficult to objectively evaluate driving risk behavior, and lack effective assessment of driving risks, which affects road traffic safety management and insurance companies' UBI auto insurance assessment.
Adopt AIoT-based driving behavior identification and risk assessment system to collect vehicle data through sensor modules, extract key indicators from the data processing module, and the risk assessment module combines entropy weight method and hierarchical analysis method for fuzzy evaluation, and uses machine learning model for risk assessment to provide Risk Assessment Score.
It realizes an objective assessment of driving risks, improves drivers' safety awareness and urban traffic safety, provides an assessment basis for insurance companies, and improves traffic management efficiency and commercial application prospects.
Smart Images

Figure CN119066340A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of driving behavior recognition, and specifically to a driving behavior recognition and risk assessment system based on AIoT. Background Art
[0002] By the end of 2023, China's car ownership will reach 336 million, and China's car production and sales have ranked first in the world for ten consecutive years. Use new technologies to improve road traffic safety management, among which the application of Internet of Things and AI technology is particularly important.
[0003] With the rapid development of Internet of Vehicles technology and the popularization of smart terminals, sensor technology that collects real-time data representing vehicle motion patterns and owner behaviors has developed rapidly, providing strong support for the analysis of risky driving behaviors.
[0004] However, the traditional on-board sensors on the market currently focus on recording vehicle behavior and providing real-time feedback to the driver and other audiences, but lack the evaluation of objective standards for driving risk behavior, which makes it inconvenient to visualize the abstract factor of driving risk. Based on this, this project combines sensor technology, uses portable sensors to obtain data, predicts vehicle behavior risks, and gives risk assessment scores as an objective evaluation of driving behavior, which is used to improve driver safety awareness, strengthen road management, and provide a basis for insurance companies' UBI auto insurance.
[0005] In summary, it is necessary to propose an AIoT-based driving behavior recognition and risk assessment system to solve the above problems. Summary of the invention
[0006] The purpose of the present invention is to provide a driving behavior recognition and risk assessment system based on AIoT to solve the problems raised in the above background technology.
[0007] To achieve the above object, the present invention provides the following technical solutions:
[0008] Driving behavior recognition and risk assessment system based on AI oT, which includes sensor module, data processing module, risk assessment module, IoT communication module, system control and display module and website back-end and front-end modules;
[0009] The sensor module is used to collect various data during vehicle driving, including attitude angle, acceleration, azimuth, temperature and humidity, and atmospheric pressure;
[0010] The data processing module is used to pre-process and analyze the data collected by the sensor module and extract key indicators;
[0011] The risk assessment module is used to combine driving behavior data and risk assessment models to assess driving risks;
[0012] The IoT communication module is used to upload processed data and risk assessment results to the server and receive instructions from the server;
[0013] The system control and display module is used for overall control and information display of the system;
[0014] The website back-end front-end module is used to provide user interaction interface and data visualization functions.
[0015] Preferably, the sensor module further comprises an MPU6050 unit, an HMC5883L unit and an AHT20_BMP280 unit;
[0016] The MPU6050 unit is used to collect attitude angle and acceleration information to determine the uphill and downhill and bumpy behavior of the vehicle;
[0017] The HMC5883L unit is used to collect azimuth information for determining the steering behavior of the vehicle;
[0018] The AHT20_BMP280 unit is used to collect temperature, humidity and atmospheric pressure information to provide environmental parameters for risk assessment.
[0019] Preferably, the data processing module further includes a data preprocessing unit, a feature extraction unit and a behavior recognition unit;
[0020] The data preprocessing unit is used to filter and remove noise from the original data to improve the data quality;
[0021] The feature extraction unit is used to extract the key indicators of acceleration change, rotation angle change, and pitch angle change according to a preset algorithm;
[0022] The behavior recognition unit is used to determine the current behavior state of the vehicle according to the extracted features, including uphill and downhill, and bumpy.
[0023] Preferably, the risk assessment module further comprises a weight determination unit, a fuzzy evaluation unit and a machine learning model unit;
[0024] The weight determination unit is used to determine weights for various driving behaviors using an entropy weight method and a hierarchical analysis method;
[0025] The fuzzy evaluation unit is used to perform fuzzy comprehensive evaluation based on the weights and driving behavior data to obtain a preliminary risk score;
[0026] The machine learning model unit is used to fit and optimize the preliminary score using random forest and extreme gradient boosting algorithms to obtain the final risk assessment result.
[0027] Preferably, the Internet of Things communication module also includes a NB-I oT communication unit, a GPS positioning unit and a SIM card slot unit;
[0028] The NB-IoT communication unit is used to realize low-power and long-distance communication between device servers using NB-I oT technology;
[0029] The GPS positioning unit is used to provide real-time location information of the vehicle for map visualization and driving path analysis;
[0030] The SIM card slot unit is used to provide an Internet of Things card for NB-IoT communication to realize the networking function.
[0031] Preferably, the system control and display module further includes an MCU control unit, an indicator light display unit and a debugging programming interface unit;
[0032] The MCU control unit is used as the main control chip of the system and is responsible for the coordination and control between various modules;
[0033] The indicator light display unit is used to display the working status of the GPS, NB-I oT module and the server through the LED indicator light;
[0034] The debugging and programming interface unit is used to provide an ST_Link interface to facilitate debugging and programming of the system.
[0035] Preferably, the website back-end front-end module further includes a back-end server unit and a front-end display unit;
[0036] The back-end server unit is used to receive and store the uploaded driving behavior data and risk assessment results, and provide an API interface for the front-end to call;
[0037] The front-end display unit is used to display real-time data panels, map visualization, and statistical analysis chart information, providing users with intuitive driving behavior analysis and risk assessment results.
[0038] Preferably, the application process of the system is as follows:
[0039] S1. Driving behavior recognition: driving behavior recognition is performed by feature extraction after data collection and data preprocessing;
[0040] S2. Driving risk assessment: fuzzy evaluation is performed through data preprocessing, entropy weight determination, and hierarchical analysis method optimization, and finally the driving risk assessment prediction results are output through hybrid model training.
[0041] Preferably, the step S1 specifically includes the following steps:
[0042] S1.1. Data collection: Utilize various sensors installed on the vehicle, including accelerometers and gyroscopes, to collect data in real time during the vehicle's driving process, including acceleration, turning angle, and pitch angle;
[0043] S1.2. Data preprocessing: Clean, denoise, format and synchronize the collected data to ensure data accuracy and consistency;
[0044] S1.3. Feature extraction: Use deep learning algorithms to extract features from preprocessed data and extract key features related to driving behavior, including acceleration change rate, turning angle change rate, and pitch angle change rate;
[0045] S1.4. Driving behavior recognition: Build a driving behavior recognition model and use the extracted features for training. The model calculates key indicators and combines them with timestamp division to determine what kind of changing posture the current vehicle belongs to, including uphill and downhill and bumpy.
[0046] Preferably, the step S2 specifically includes the following steps:
[0047] S2.1. Data preprocessing: Collect and integrate multiple public driving behavior datasets, including DBnet, common-ai, and generated driver behavior data, normalize all data, set driver dangerous behaviors to 1 or 0, and compress other behavior data between 0 and 1;
[0048] S2.2. Determination of entropy weight: The processed data set is analyzed using the entropy weight formula to determine the preliminary weights of driving risk factors such as smoking, drinking, playing with mobile phones, and fatigue driving;
[0049] S2.3. Optimization of AHP:
[0050] S2.3.1. Use the analytic hierarchy process to optimize the weights determined by the entropy weight method;
[0051] S2.3.2. Construct a hierarchical structure of the driving risk model, construct a judgment matrix through expert scoring, calculate the total ranking weight, and obtain the final weight;
[0052] S2.4. Fuzzy evaluation:
[0053] S2.4.1. Define the evaluation indicators as driver behavior and vehicle driving behavior, and determine the evaluation set as the specific performance of driver behavior and vehicle driving behavior;
[0054] S2.4.2. Map the final weights to the corresponding membership values, perform fuzzy comprehensive evaluation, and obtain the fuzzy evaluation results;
[0055] S2.4.3. Through defuzzification processing, the fuzzy evaluation results are converted into specific evaluation values, that is, percentage scores;
[0056] S2.5. Hybrid model training:
[0057] S2.5.1. Use random forest and extreme gradient boosting algorithms to build a hybrid model to fit and regress the specific evaluation values obtained by fuzzy evaluation;
[0058] S2.5.2. Train the mixed model by weighted averaging to obtain the highest coefficient of determination;
[0059] S2.6. Output prediction results:
[0060] S2.6.1. Use the trained hybrid model to predict the input driving behavior data and output two scores: the minimum score and the average score;
[0061] S2.6.2. The lowest score refers to the score represented by the interval with the lowest score after the video is divided into multiple 40-second intervals; the average score refers to the average of all interval scores;
[0062] S2.6.3. These two scores are used to assess the level of risk while driving.
[0063] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention provides a driving risk assessment system based on universal sensor technology, which can collect driver behavior and vehicle behavior in a certain road section, and conduct a driving risk assessment system based on this, introduces risk assessment indicators such as driver behavior habits, "four emergency driving situations", and road conditions, and objectively evaluates the risk level of the driver's driving behavior within a certain period of time. The system can be applied to groups such as urban traffic management personnel and drivers themselves to improve the safety of urban traffic and the rationality of driving behavior. It can also be applied to commercial fields such as insurance companies and vehicle manufacturers as an evaluation indicator of vehicle insurance costs and vehicle performance. It can not only effectively improve the efficiency of urban traffic management and the rationality of driving behavior, but also has broad prospects in the fields of smart transportation and commercial applications. BRIEF DESCRIPTION OF THE DRAWINGS
[0064] Figure 1 The driving behavior recognition algorithm flow chart of the present invention is shown;
[0065] Figure 2 shows a risk assessment algorithm flow chart of the present invention;
[0066] Figure 3 A schematic diagram of hierarchical analysis of the present invention is shown;
[0067] Figure 4A schematic diagram of fuzzy evaluation of the present invention is shown;
[0068] Figure 5 The topology diagram of the driving behavior recognition and risk assessment system based on AI oT of the present invention is shown. DETAILED DESCRIPTION
[0069] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0070] See also Figure 1 , Figure 2 , Figure 3 , Figure 4 as well as Figure 5 The present invention proposes a driving behavior recognition and risk assessment system based on AIoT, which includes a sensor module, a data processing module, a risk assessment module, an Internet of Things communication module, a system control and display module, and a website back-end front-end module. It should be noted that:
[0071] In this system, the sensor module is used to collect various data during the vehicle's driving process, including attitude angle, acceleration, azimuth, temperature and humidity, and atmospheric pressure. The data processing module is used to preprocess and analyze the data collected by the sensor module and extract key indicators. The risk assessment module is used to combine driving behavior data and risk assessment models to assess driving risks. The Internet of Things communication module is used to upload processed data and risk assessment results to the server and receive instructions from the server. The system control and display module is used for overall control of the system and information display. The website back-end front-end module is used to provide a user interaction interface and data visualization functions.
[0072] It should also be noted that the sensor module also includes an MPU6050 unit, an HMC5883L unit and an AHT20_BMP280 unit. The MPU6050 unit is used to collect attitude angle and acceleration information for judging the uphill and downhill and bumpy behavior of the vehicle. The HMC5883L unit is used to collect azimuth information for judging the steering behavior of the vehicle. The AHT20_BMP280 unit is used to collect temperature, humidity and atmospheric pressure information to provide environmental parameters for risk assessment.
[0073] It should also be noted that the data processing module also includes a data preprocessing unit, a feature extraction unit and a behavior recognition unit. The data preprocessing unit is used to filter and denoise the original data to improve data quality. The feature extraction unit is used to extract key indicators of acceleration change, turning angle change, and pitch angle change according to a preset algorithm. The behavior recognition unit is used to judge the current behavior state of the vehicle based on the extracted features, including uphill and downhill and bumpy.
[0074] It should also be noted that the risk assessment module also includes a weight determination unit, a fuzzy evaluation unit and a machine learning model unit. The weight determination unit is used to determine the weights for various driving behaviors using the entropy weight method and the hierarchical analysis method. The fuzzy evaluation unit is used to perform fuzzy comprehensive evaluation based on the weights and driving behavior data to obtain a preliminary risk score. The machine learning model unit is used to fit and optimize the preliminary score using the random forest and extreme gradient boosting algorithms to obtain the final risk assessment result.
[0075] It should also be noted that the IoT communication module also includes an NB-IoT communication unit, a GPS positioning unit and a SIM card slot unit. The NB-IoT communication unit is used to utilize NB-I oT technology to achieve low-power, long-distance communication between device servers. The GPS positioning unit is used to provide real-time location information of the vehicle for map visualization and driving path analysis. The SIM card slot unit is used to provide an IoT card for NB-I oT communication to achieve networking functions.
[0076] It should also be noted that the system control and display module also includes an MCU control unit, an indicator light display unit and a debugging programming interface unit. The MCU control unit is used as the main control chip of the system and is responsible for the coordination and control between various modules. The indicator light display unit is used to display the working status of the GPS, NB-I oT module and the server through LED indicators. The debugging programming interface unit is used to provide an ST_L i nk interface to facilitate debugging and programming of the system.
[0077] It should also be noted that the website's back-end front-end module also includes a back-end server unit and a front-end display unit. The back-end server unit is used to receive and store uploaded driving behavior data and risk assessment results, and provide an API interface for the front-end to call. The front-end display unit is used to display real-time data panels, map visualization, and statistical analysis chart information to provide users with intuitive driving behavior analysis and risk assessment results.
[0078] In summary, the AIoT-based driving behavior recognition and risk assessment system, which is implemented based on the sensor module, data processing module, risk assessment module, IoT communication module, system control and display module, and website back-end and front-end modules, has the following process in actual application:
[0079] S1. Driving behavior recognition:
[0080] S1.1. Data collection: Utilize various sensors installed on the vehicle, including accelerometers and gyroscopes, to collect data in real time during the vehicle's driving process, including acceleration, turning angle, and pitch angle;
[0081] S1.2. Data preprocessing: Clean, denoise, format and synchronize the collected data to ensure data accuracy and consistency;
[0082] S1.3. Feature extraction: Use deep learning algorithms to extract features from preprocessed data and extract key features related to driving behavior, including acceleration change rate, turning angle change rate, and pitch angle change rate;
[0083] S1.4. Driving behavior recognition: Build a driving behavior recognition model and use the extracted features for training. The model calculates key indicators and combines timestamp division to determine what kind of changing posture the current vehicle belongs to, including uphill and downhill and bumpy.
[0084] S2. Driving risk assessment:
[0085] S2.1. Data preprocessing: Collect and integrate multiple public driving behavior datasets, including DBnet, common-ai, and generated driver behavior data, normalize all data, set driver dangerous behaviors to 1 or 0, and compress other behavior data between 0 and 1;
[0086] S2.2. Determination of entropy weight: The processed data set is analyzed using the entropy weight formula to determine the preliminary weights of driving risk factors such as smoking, drinking, playing with mobile phones, and fatigue driving;
[0087] S2.3. Optimization of AHP:
[0088] S2.3.1. Use the analytic hierarchy process to optimize the weights determined by the entropy weight method;
[0089] S2.3.2. Construct a hierarchical structure of the driving risk model, construct a judgment matrix through expert scoring, calculate the total ranking weight, and obtain the final weight;
[0090] S2.4. Fuzzy evaluation:
[0091] S2.4.1. Define the evaluation indicators as driver behavior and vehicle driving behavior, and determine the evaluation set as the specific performance of driver behavior and vehicle driving behavior;
[0092] S2.4.2. Map the final weights to the corresponding membership values, perform fuzzy comprehensive evaluation, and obtain the fuzzy evaluation results;
[0093] S2.4.3. Through defuzzification processing, the fuzzy evaluation results are converted into specific evaluation values, that is, percentage scores;
[0094] S2.5. Hybrid model training:
[0095] S2.5.1. Use random forest and extreme gradient boosting algorithms to build a hybrid model to fit and regress the specific evaluation values obtained by fuzzy evaluation;
[0096] S2.5.2. Train the mixed model by weighted averaging to obtain the highest coefficient of determination;
[0097] S2.6. Output prediction results:
[0098] S2.6.1. Use the trained hybrid model to predict the input driving behavior data and output two scores: the minimum score and the average score;
[0099] S2.6.2. The lowest score refers to the score represented by the interval with the lowest score after the video is divided into multiple 40-second intervals; the average score refers to the average of all interval scores;
[0100] S2.6.3. These two scores are used to assess the level of risk while driving.
[0101] Example
[0102] The application examples of the above AIoT-based driving behavior recognition and risk assessment system are:
[0103] 1. Driving behavior recognition:
[0104] (1) Driving behavior recognition algorithm: Analyze key driving behavior data based on sensor data;
[0105] Programming language: Python
[0106] See also Figure 1 ,Driving behavior recognition is mainly based on various driving behaviors obtained by sensors, and mathematical formulas are used for calculation to extract key indicators for judging driving behaviors: acceleration change, turning angle change, pitch angle change, etc., and they are combined to determine what kind of change posture the current vehicle belongs to;
[0107] (2) Driving behavior recognition algorithm process:
[0108] Since the vehicle's behavior is a continuous action, we divide it into 5-second intervals based on the timestamp when the sensor acquires the data. The judgment of each behavior is based on the changes within 5 seconds.
[0109] (2.1) Uphill and downhill:
[0110] ① First, take 5s as a time interval and calculate the multiple continuous pitch angle change rates within this time interval;
[0111] ② When the pitch angle change rate is greater than (less than) the threshold value of 0.8 (-0.8) rad / s, the upslope (downslope) factor is increased by 1;
[0112] ③ Since the average number of pitch angle change rate segments recorded in the 5-second period is 40, if the uphill (downhill) factor number exceeds 20, it can indicate that the vehicle is in an uphill (downhill) state during this period.
[0113] (2.2) Bumping:
[0114] ①First, record multiple consecutive pitch angle change rates at 5s intervals;
[0115] ② Then use a sliding window with a length of 5 consecutive pitch angle changes to traverse the time interval;
[0116] ③Record the number of times the pitch angle change rate in a sliding window is greater than or less than the threshold;
[0117] ④ If the number of times that the pitch angle change rate is greater than or less than the threshold in a window is not 0, it is recorded as a turbulence factor.
[0118] ⑤If the number of bumpy factors within a time interval is greater than the threshold, it means that this section of the journey is a bumpy section.
[0119] 2. Driving risk assessment:
[0120] Driving risk assessment algorithm: entropy weight-hierarchy analysis-fuzzy evaluation model combined with random forest and extreme gradient boosting algorithm;
[0121] See also Figure 2 ,The driving risk assessment algorithm first uses the public driving behavior dataset common-ai ,and uses the entropy weight and hierarchical analysis method to add weights to various behaviors that affect driving risks. Based on the frequency and weights of different behaviors, a fuzzy evaluation model can be used to make a preliminary 100-point evaluation. After obtaining the preliminary score;
[0122] Since the rating records are discrete, we can preprocess them to obtain a data set, and then use the random forest algorithm to fit it into a regression model as the final rating model;
[0123] Entropy weight method and analytic hierarchy process
[0124] 1. Entropy Weight Method:
[0125] The entropy method is an objective weighting method with high accuracy and credibility, which can obtain the indicator weight value;
[0126] By calculating the information entropy of the indicator, the weight of the indicator is determined according to the relative development and change degree of the relevant indicators;
[0127] The indicator system with a large relative degree of change has a larger weight;
[0128] The smaller the entropy value, the greater the change in the indicator value, the more information it provides, and the greater the weight;
[0129] On the contrary, the larger the entropy value of an important indicator is, the smaller the influence of the variation of its indicator value is, the less information it provides, and the smaller its weight coefficient is.
[0130] The entropy method has higher accuracy and credibility than the subjective weighting method;
[0131] 2. Analytical Hierarchy Process:
[0132] The Analytical Hierarchy Process (AHP) is a method that transforms complex decision-making problems into information systems.
[0133] Decompose the decision-making problem into different hierarchical structures according to the overall goal, evaluation criteria, sub-goals at each level and specific plans;
[0134] Then, by solving the eigenvector of the judgment matrix, the priority weight value of each element in each level relative to an element in the previous level is calculated;
[0135] Finally, the weighted summation method is used to recursively merge the weights of each alternative plan to the overall goal;
[0136] Finally, the optimal solution is obtained; the "priority" mentioned here is a relative measure, which means the relative measure of the superiority of each alternative solution among the secondary goals of the evaluation criteria or specific characteristics, and the relative measure of the importance of each secondary goal to the previous goal;
[0137] Specific algorithm implementation process:
[0138] (1) Data preprocessing:
[0139] Since the entropy weight method requires a large number of data sets as a basis, multiple public driving behavior data sets such as DBnet and common-ai were collected as driving behavior materials;
[0140] The SHNDS dataset collected in the paper "The Shangha i naturalistic driving study" randomly generates driver behavior data based on the frequency and average values of drivers' smoking, drinking, using mobile phones, distraction and other behaviors proposed in the paper;
[0141] Subsequently, all data were normalized, where the driver's dangerous behavior was set to 1 if it occurred, otherwise it was set to 0; for other behaviors, data standardization and normalization operations were used to compress the data between 0 and 1. The data standardization formula is shown in formula (1), and the normalization formula is shown in formula (2);
[0142]
[0143] (2) Determination of entropy weight:
[0144] Combined with the processed data set, the entropy weight method formula was used to analyze the weights of driving risk factors such as smoking, drinking, playing with mobile phones, fatigue driving, sudden acceleration, sudden deceleration, and sharp turns, and the preliminary weights were obtained;
[0145] (3) Analytical Hierarchy Process Optimization:
[0146] Since the entropy weight method determines the weights entirely from the objective data set, the accuracy is relatively general. On this basis, the hierarchical analysis method based on subjective perception is used to optimize it.
[0147] First, the driving risk model is divided into three levels: target level, criterion level, and solution level. On this basis, the judgment matrix of driver behavior and vehicle driving behavior is constructed by consulting traffic management personnel for expert scoring. The total ranking weight is calculated using the judgment matrix after the judgment matrix satisfies the consistency test shown below. Finally, the weights obtained by the entropy weight method and the hierarchical analysis are combined to obtain the final weights shown in the figure.
[0148] Fuzzy evaluation - calculating score:
[0149] Fuzzy evaluation is generally divided into the following five aspects: defining evaluation indicators, determining evaluation sets, fuzzy processing, fuzzy synthesis,
[0150] Comprehensive evaluation and defuzzification processing;
[0151] See also Figure 3 , the first three steps have been completed, and the sum of the driver's behavior and the vehicle's driving behavior is the evaluation index;
[0152] These behaviors are divided into driver behaviors and vehicle driving behaviors to determine the evaluation set, and the final weight obtained at the end is the mapping of the specific value of each indicator to the corresponding membership value;
[0153] The next step is to complete the last two steps: fuzzy comprehensive evaluation is to calculate the fuzzy evaluation result by using the fuzzy comprehensive method according to the weight and membership value of the evaluation index, and to obtain the fuzzy evaluation result by directly multiplying the evaluation index with the final weight;
[0154] Defuzzification is to convert the fuzzy evaluation results into specific evaluation values, that is, to map the fuzzy values to specific evaluation values;
[0155] Use the percentage system, which is easy to understand and train. Subtract the evaluation value just obtained from 100 to get the final specific evaluation score.
[0156] Random Forest and Extreme Gradient Boosting Algorithms
[0157] See also Figure 4 ,RandomForest is an integrated learning algorithm that combines the characteristics of decision trees and randomness;
[0158] Perform classification or regression tasks by building multiple decision trees and combining their prediction results;
[0159] Algorithm principle: Random forest consists of multiple decision trees, each of which is an independent classifier;
[0160] When constructing a random forest, we first use the bootstrap sampling method to generate multiple different training subsets from the original training set.
[0161] For each training subset, build a decision tree independently;
[0162] When building a decision tree, for each node, a subset is randomly selected from all features, and then the best feature is selected for partitioning;
[0163] Each decision tree grows to the maximum depth possible without pruning;
[0164] Finally, the prediction results of all decision trees are combined by taking the average value to obtain the final prediction result, that is, the risk prediction score;
[0165] Extreme Gradient Boosting (XGBoost) is an improved version of Gradient Boosting Trees;
[0166] Some new techniques and strategies are introduced based on the gradient boosting tree to further improve the performance and efficiency of the model;
[0167] Algorithm principle: The core of XGBoost is to fit the model by optimizing the objective function;
[0168] The objective function consists of two parts: the loss function of the training data and the regularization term;
[0169] XGBoost uses the gradient boosting algorithm to minimize the objective function;
[0170] During the training process, XGBoost first initializes a simple model (such as a tree with only one leaf node), and then gradually improves the model in an iterative manner;
[0171] In each iteration, XGBoost calculates the gradient and second-order derivative of the current model, and then builds a new decision tree based on the gradient and second-order derivative;
[0172] New decision trees are added to the model and the model's predictions are updated based on the objective function;
[0173] Finally, by weighting and summing the prediction results of multiple decision trees, XGBoost obtains the final prediction result, which is the risk prediction score;
[0174] The random forest-extreme gradient boosting hybrid model was trained using a weighted average approach to obtain the highest coefficient of determination during model testing;
[0175] Final prediction results:
[0176] Output two scores, namely the minimum score and the average score;
[0177] The lowest score refers to the score represented by the interval with the lowest score after the video is divided into many 40-s intervals; the average score refers to the average of the scores of all intervals.
[0178] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. AIoT-based driving behavior recognition and risk assessment system, characterized by: The system includes a sensor module, a data processing module, a risk assessment module, an Internet of Things communication module, a system control and display module, and a website back-end and front-end module; The sensor module is used to collect various data during vehicle driving, including attitude angle, acceleration, azimuth, temperature and humidity, and atmospheric pressure; The data processing module is used to pre-process and analyze the data collected by the sensor module and extract key indicators; The risk assessment module is used to combine driving behavior data and risk assessment models to assess driving risks; The IoT communication module is used to upload processed data and risk assessment results to the server and receive instructions from the server; The system control and display module is used for overall control and information display of the system; The website back-end front-end module is used to provide user interaction interface and data visualization functions.
2. The AIoT-based driving behavior recognition and risk assessment system according to claim 1, characterized in that: The sensor module also includes the MPU6050 unit, the HMC5883L unit, and the AHT20_BMP280 unit; The MPU6050 unit is used to collect attitude angle and acceleration information to determine the uphill and downhill and bumpy behavior of the vehicle; The HMC5883L unit is used to collect azimuth information for determining the steering behavior of the vehicle; The AHT20_BMP280 unit is used to collect temperature, humidity and atmospheric pressure information to provide environmental parameters for risk assessment.
3. The AIoT-based driving behavior recognition and risk assessment system according to claim 2 is characterized in that: The data processing module also includes a data preprocessing unit, a feature extraction unit and a behavior recognition unit; The data preprocessing unit is used to filter and remove noise from the original data to improve the data quality; The feature extraction unit is used to extract key indicators of acceleration change, rotation angle change, and pitch angle change according to a preset algorithm; The behavior recognition unit is used to determine the current behavior state of the vehicle according to the extracted features, including uphill and downhill and bumpy.
4. The AIoT-based driving behavior recognition and risk assessment system according to claim 3 is characterized in that: The risk assessment module also includes a weight determination unit, a fuzzy evaluation unit and a machine learning model unit; The weight determination unit is used to determine weights for various driving behaviors using an entropy weight method and a hierarchical analysis method; The fuzzy evaluation unit is used to perform fuzzy comprehensive evaluation based on the weights and driving behavior data to obtain a preliminary risk score; The machine learning model unit is used to fit and optimize the preliminary score using random forest and extreme gradient boosting algorithms to obtain the final risk assessment result.
5. The AIoT-based driving behavior recognition and risk assessment system according to claim 4 is characterized in that: The IoT communication module also includes an NB-IoT communication unit, a GPS positioning unit and a SIM card slot unit; The NB-IoT communication unit is used to realize low-power consumption and long-distance communication between device servers by using NB-IoT technology; The GPS positioning unit is used to provide real-time location information of the vehicle for map visualization and driving path analysis; The SIM card slot unit is used to provide an Internet of Things card for NB-IoT communication to realize the networking function.
6. The AIoT-based driving behavior recognition and risk assessment system according to claim 5, characterized in that: The system control and display module also includes an MCU control unit, an indicator light display unit and a debugging programming interface unit; The MCU control unit is used as the main control chip of the system and is responsible for the coordination and control between various modules; The indicator light display unit is used to display the working status of the GPS, NB-IoT module and the server through the LED indicator light; The debugging and programming interface unit is used to provide an ST_Link interface to facilitate debugging and programming of the system.
7. The AIoT-based driving behavior recognition and risk assessment system according to claim 6, characterized in that: The website back-end front-end module also includes a back-end server unit and a front-end display unit; The back-end server unit is used to receive and store the uploaded driving behavior data and risk assessment results, and provide an API interface for the front-end to call; The front-end display unit is used to display real-time data panels, map visualization, and statistical analysis chart information, providing users with intuitive driving behavior analysis and risk assessment results.
8. The AIoT-based driving behavior recognition and risk assessment system according to claim 7, characterized in that: The application process of the system is as follows: S1. Driving behavior recognition: driving behavior recognition is performed by feature extraction after data collection and data preprocessing; S2. Driving risk assessment: fuzzy evaluation is performed through data preprocessing, entropy weight determination, and hierarchical analysis method optimization, and finally the driving risk assessment prediction results are output through hybrid model training.
9. The AIoT-based driving behavior recognition and risk assessment system according to claim 8, characterized in that: The step S1 specifically includes the following steps: S1.
1. Data collection: Utilize various sensors installed on the vehicle, including accelerometers and gyroscopes, to collect data in real time during the vehicle's driving process, including acceleration, turning angle, and pitch angle; S1.
2. Data preprocessing: Clean, denoise, format and synchronize the collected data to ensure data accuracy and consistency; S1.
3. Feature extraction: Use deep learning algorithms to extract features from preprocessed data and extract key features related to driving behavior, including acceleration change rate, turning angle change rate, and pitch angle change rate; S1.
4. Driving behavior recognition: Build a driving behavior recognition model and use the extracted features for training. The model calculates key indicators and combines them with timestamp division to determine what kind of changing posture the current vehicle belongs to, including uphill and downhill and bumpy.
10. The AIoT-based driving behavior recognition and risk assessment system according to claim 8, characterized in that , the step S2 specifically includes the following steps: S2.
1. Data preprocessing: Collect and integrate multiple public driving behavior datasets, including DBnet, common-ai, and generated driver behavior data, normalize all data, set driver dangerous behaviors to 1 or 0, and compress other behavior data between 0 and 1; S2.
2. Determination of entropy weight: The processed data set is analyzed using the entropy weight formula to determine the preliminary weights of driving risk factors such as smoking, drinking, playing with mobile phones, and fatigue driving; S2.
3. Optimization of AHP: S2.3.
1. Use the analytic hierarchy process to optimize the weights determined by the entropy weight method; S2.3.
2. Construct a hierarchical structure of the driving risk model, construct a judgment matrix through expert scoring, calculate the total ranking weight, and obtain the final weight; S2.
4. Fuzzy evaluation: S2.4.
1. Define the evaluation indicators as driver behavior and vehicle driving behavior, and determine the evaluation set as the specific performance of driver behavior and vehicle driving behavior; S2.4.
2. Map the final weights to the corresponding membership values, perform fuzzy comprehensive evaluation, and obtain the fuzzy evaluation results; S2.4.
3. Through defuzzification processing, the fuzzy evaluation results are converted into specific evaluation values, that is, percentage scores; S2.
5. Hybrid model training: S2.5.
1. Use random forest and extreme gradient boosting algorithms to build a hybrid model to fit and regress the specific evaluation values obtained by fuzzy evaluation; S2.5.
2. Train the mixed model by weighted averaging to obtain the highest coefficient of determination; S2.
6. Output prediction results: S2.6.
1. Use the trained hybrid model to predict the input driving behavior data and output two scores: the minimum score and the average score; S2.6.
2. The lowest score refers to the score represented by the interval with the lowest score after the video is divided into multiple 40-second intervals; the average score refers to the average of the scores of all intervals; S2.6.
3. These two scores are used to assess the level of risk while driving.
Citation Information
Patent Citations
Driver state indicating device and method based on vehicle moving state
CN105599773A
Driving behavior comprehensive evaluation method based on multi-index fusion
CN115204636A
Method for evaluating driving behavior of mine driver
CN117033903A
Random forest classification method based on fuzzy comprehensive evaluation
CN117494008A
Cited By
Driving behavior risk assessment method and system
CN119961773A
A driving behavior risk assessment method and system
CN119961773B
Driving behavior big data-based driver speed control ability analysis method and device
CN120408389A
