Enhanced closed scene capability evaluation method and system applied to intelligent driving training

Abstract

The invention discloses an ability evaluation method and system applied to an enhanced closed scene in intelligent driving training. A super-realistic simulation environment is constructed, a physical engine and high-precision modeling are utilized, a dynamic weather system and a user-defined scene editor are integrated, and diversified realistic scenes are provided; multi-dimensional data acquisition and preprocessing are carried out, traditional operation and biological characteristic data are combined, and support is provided for risk assessment through cleaning and characteristic extraction; a reinforcement learning risk assessment model is constructed, and the driving behavior risk is assessed in real time; safety response and personalized teaching are realized, and teaching is intervened and adjusted in real time according to a risk assessment result. Through environment simulation, data processing, model evaluation, safety and teaching effect evaluation and the like, the driving skill of trainees can be comprehensively improved, driving practice safety is guaranteed, a personalized learning path is provided, an efficient, safe and personalized solution is brought to the intelligent driving training industry, and the method has wide application prospects and remarkable technical innovation value.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent driving training, and particularly relates to an ability evaluation method and system for strengthening closed scenarios in intelligent driving training. Background Art

[0002] In recent years, the rapid development of autonomous driving technology has profoundly changed the pattern of the automotive industry and also had a profound impact on the field of driving education. With the increasing maturity of intelligent driving technology, higher requirements are put forward for the driver's skill level, reaction speed, and safety awareness.

[0003] Traditional driving training methods mainly rely on on-site coach guidance and limited actual road practice, and there are many limitations to this method. First of all, it is difficult to comprehensively cover various complex driving scenarios in on-site practice, such as extreme weather conditions (such as heavy rain, heavy snow, thick fog, etc.), night driving, highway emergency avoidance and other special situations, resulting in students may lack sufficient coping abilities when encountering these situations. Secondly, the potential safety risks in on-site practice are relatively high, especially for beginners. An inadvertent operation error may lead to serious accidents. In addition, it is difficult for traditional training methods to provide personalized teaching plans according to the individual differences of students and difficult to meet the learning needs of different students.

[0004] In order to address the deficiencies of traditional driving training, intelligent driving training systems have emerged as the times require. Currently, there are already some intelligent driving training systems based on virtual reality on the market, and they usually consist of the following parts:

[0005] Virtual scene construction: Using computer graphics technology to create realistic driving scenarios such as urban roads, highways, and rural roads, attempting to simulate the real driving environment.

[0006] Interactive devices: Equipped with simulation driving devices such as steering wheels, throttles, and brakes, enabling students to experience the driving process immersive.

[0007] Driving evaluation system: By monitoring the driving operations of students, such as speed control, direction control, lane change operations, etc., to evaluate the driving skills of students.

[0008] Safety mechanism: Some systems have built-in simple safety mechanisms, such as issuing warnings or automatically decelerating when the vehicle approaches an obstacle, but these mechanisms are often based on preset rules and lack flexibility and intelligence.

[0009] Although these intelligent driving training systems have improved the efficiency and safety of driving training to a certain extent, there are still many problems. For example, it is impossible to dynamically adjust the training difficulty and scenarios according to the actual driving level and habits of students; it is difficult to simulate complex and changeable traffic environments and emergencies, such as bad weather, sudden failures, etc.;

[0010] The safety mechanism is relatively simple and cannot effectively predict and respond to potential driving risks.

[0011] In summary, the existing intelligent driving training systems still face many challenges in ensuring the safety of trainees. There is an urgent need for new solutions to improve the driving skills and response capabilities of trainees to meet the ever-developing driving training needs. This also provides directions for innovation and improvement for those skilled in the art, prompting the development of more effective intelligent driving training technologies and methods. Summary of the Invention

[0012] In view of the deficiencies of the current related prior arts, an ability evaluation method for enhancing closed scenarios in intelligent driving training according to the present invention aims to solve the defects of the existing intelligent driving training systems. By analyzing the driving behaviors and performances of trainees through reinforcement learning algorithms, it realizes dynamic adjustment of training content and difficulty to achieve personalized teaching; constructs diverse closed scenarios and simulates complex and changeable traffic environments and emergencies to improve the response capabilities of trainees; develops an intelligent risk assessment model to monitor and predict risks in real time and trigger a safety response mechanism to ensure the safety of trainees during driving practice, thereby comprehensively improving the quality and effect of intelligent driving training.

[0013] To achieve the above object, the first aspect of the present invention provides an ability evaluation method for enhancing closed scenarios in intelligent driving training, including the following steps:

[0014] S1: Construct a hyper-realistic simulation environment, including selecting a physics engine and modeling, integrating a dynamic weather system, and constructing a custom scene editor to create diverse and realistic closed driving scenarios;

[0015] S2: Conduct data collection and preprocessing, collect traditional driving operation data and biometric data, and clean and extract features from the collected data to provide data support for risk assessment;

[0016] S3: Construct a reinforcement learning risk assessment model, including selecting a deep reinforcement learning framework, defining the state space and action space, training the model, and deploying the model to achieve real-time assessment of trainees' driving behaviors;

[0017] S4: Implement safety response and personalized teaching, detect high-risk behaviors based on the risk assessment results and intervene immediately, and at the same time adjust the teaching difficulty according to the performance of trainees and provide personalized learning suggestions.

[0018] Further, in step S1, the selection of the physics engine and modeling is specifically as follows:

[0019] S11: Select the Unity or UnrealEngine physics engine to ensure that the physical laws and visual effects of the simulation environment are highly realistic;

[0020] S12: Create one or more elements including vehicles, roads, buildings, and weather elements using high-precision 3D modeling technology, ensuring that every detail conforms to the physical characteristics of the real world.

[0021] Further, in step S1, the integrated dynamic weather system is specifically:

[0022] S13: Develop a dynamic weather system to simulate one or more environments such as rain, snow, fog, dusk, and night through algorithms;

[0023] S14: Provide adjustable weather parameters, and coaches can adjust one or more parameters such as the amount of rainfall, the density of snowflakes, and the concentration of fog through the system interface.

[0024] Further, in step S1, the construction of the custom scenario editor is specifically:

[0025] S15: Develop an intuitive and easy-to-use scenario editor, provide a graphical interface, and support coaches to quickly construct or adjust specific scenarios;

[0026] S16: Establish a scenario element library including one or more of school areas, construction sections, highways, and rural roads, and coaches can select and combine these elements as needed.

[0027] Further, in step S2, the data collection is specifically:

[0028] S21: Collect one or more traditional driving operation data such as steering wheel angle and accelerator and brake pedals through sensors;

[0029] S22: Use a high-precision eye tracking system and facial expression recognition technology to collect one or more biometric data such as the attention distribution, emotional state, and tension level of the trainee.

[0030] Further, in step S2, the data cleaning and feature extraction are specifically:

[0031] S23: Use machine learning algorithms, including one or more of clustering analysis and outlier detection, to automatically filter out noise data and ensure data quality;

[0032] S24: Extract one or more key features such as driving habits, reaction speed, and attention concentration through feature engineering techniques.

[0033] Further, in step S3, the construction of the reinforcement learning risk assessment model is specifically:

[0034] S31: Select a suitable deep reinforcement learning framework, including TensorFlow or PyTorch, to build a risk assessment model;

[0035] S32: Define the state space of the model according to the characteristics of driving behavior, including one or more of vehicle speed, acceleration, and steering wheel angle, and the action space, including one or more of braking, accelerating, and steering;

[0036] S33: Divide the preprocessed data into a training set and a test set, and optimize the model parameters through iterative training to enable it to accurately evaluate the driving behavior risk of trainees;

[0037] S34: During the training process, comprehensively consider evaluation accuracy, real-time performance, and fairness, and design a multi-objective optimization function to ensure the model performance;

[0038] S35: Deploy the trained model to the intelligent driving training system to implement the real-time evaluation function. The system collects the driving data of trainees in real time, inputs it into the risk assessment model, and obtains the real-time risk assessment result.

[0039] Further, in step S4, the safety response and personalized teaching are specifically as follows:

[0040] S41: According to the results of the risk assessment model, detect one or more high-risk behaviors of trainees, including hard braking, hard accelerating, and excessive steering wheel rotation;

[0041] S42: When a high-risk behavior is detected, the system immediately activates the safety response mechanism and intervenes through one or more of simulated vehicle control, sound warning, visual prompt, and voice guidance;

[0042] S43: According to the real-time performance of trainees and the results of the risk assessment, the system automatically adjusts the difficulty and complexity of the virtual scenario to provide trainees with a personalized learning path;

[0043] S44: According to the driving behavior characteristics of trainees, the system provides personalized learning suggestions, including one or more of strengthening practice in a certain aspect and adjusting driving habits.

[0044] The second aspect of the present invention provides an ability evaluation system for strengthening closed scenarios in intelligent driving training, characterized in that the system is used for the above method, and the system includes:

[0045] A simulated environment construction module for constructing a hyper-realistic simulation environment, and this module includes:

[0046] A physical engine and modeling unit for selecting a physical engine and performing high-precision 3D modeling to create realistic driving scene elements. The physical engine is a Unity or Unreal Engine physical engine, and the modeling elements include one or more of vehicles, roads, buildings, and weather elements;

[0047] A dynamic weather system unit for developing a dynamic weather system and providing adjustable parameters to simulate one or more environments among rain, snow, fog, dusk, and night, and the coach can adjust one or more parameters such as the amount of rain, the density of snowflakes, and the concentration of fog through the system interface;

[0048] A custom scenario editor unit for developing an intuitive and easy-to-use scenario editor and establishing a scenario element library to support the coach in quickly constructing or adjusting specific scenarios. The scenario element library includes one or more of school areas, construction sections, highways, and rural roads;

[0049] A data collection and processing module for collecting and preprocessing trainee driving data. This module includes:

[0050] A data collection unit for collecting one or more traditional driving operation data such as steering wheel angle and accelerator and brake pedals through sensors, and collecting one or more biometric data such as the attention distribution, emotional state, and tension level of the trainee using a high-precision eye movement tracking system and facial expression recognition technology;

[0051] A data cleaning and feature extraction unit for automatically filtering noise data using machine learning algorithms, including one or more of clustering analysis and outlier detection, to ensure data quality, and extracting one or more key features such as driving habits, reaction speed, and attention concentration through feature engineering techniques;

[0052] A risk assessment model construction and operation module for constructing and operating a reinforcement learning risk assessment model. This module includes:

[0053] A model construction unit for selecting a suitable deep reinforcement learning framework, including TensorFlow or PyTorch, to construct a risk assessment model, and defining the state space and action space of the model according to the characteristics of driving behavior. The state space includes one or more of vehicle speed, acceleration, and steering wheel angle, and the action space includes one or more of braking, accelerating, and steering;

[0054] A model training unit for dividing the preprocessed data into a training set and a test set, optimizing the model parameters through iterative training to enable it to accurately evaluate the driving behavior risk of the trainee, and comprehensively considering evaluation accuracy, real-time performance, and fairness during the training process to design a multi-objective optimization function to ensure the model performance;

[0055] A model deployment and evaluation unit for deploying the trained model into the intelligent driving training system to achieve real-time evaluation function. The system collects the driving data of the trainee in real time, inputs it into the risk assessment model, and obtains real-time risk assessment results;

[0056] A safety response and teaching adjustment module, which is used to perform safety responses and personalized teaching adjustments according to the risk assessment results. This module includes:

[0057] A high-risk behavior detection unit, which is used to detect one or more high-risk behaviors of the trainee, including sudden braking, sudden acceleration, and excessive steering wheel rotation, according to the results of the risk assessment model;

[0058] A safety response unit, which is used to immediately activate the safety response mechanism and intervene by one or more of the ways of simulating vehicle control, sound warning, visual prompt, and voice guidance when a high-risk behavior is detected;

[0059] A teaching adjustment unit, which is used to automatically adjust the difficulty and complexity of the virtual scenario according to the real-time performance of the trainee and the risk assessment results, provide a personalized learning path for the trainee, and provide personalized learning suggestions according to the driving behavior characteristics of the trainee, including one or more of strengthening practice in a certain aspect and adjusting driving habits.

[0060] Furthermore, the simulation environment construction module, the data collection and processing module, the risk assessment model construction and operation module, and the safety response and teaching adjustment module are communicatively connected to each other to achieve data transmission and collaborative work;

[0061] It further includes a storage module, which is used to store data during the operation of the system, including one or more of simulation environment data, trainee driving data, risk assessment model data, and teaching adjustment record data;

[0062] It further includes a display module, which is used to display one or more of the simulation driving scenario, trainee driving data, risk assessment results, and personalized learning suggestions for the trainee and the coach to view;

[0063] The system can also be communicatively connected to external devices or systems, including connecting to a driving school management system to achieve trainee information sharing and training progress synchronization; connecting to a vehicle manufacturer system to obtain vehicle performance data and updated driving safety standards information; connecting to a traffic management department system to upload trainee training data and ability assessment results for supervision and statistical analysis.

[0064] The present invention adopts the above technical solutions and has at least the following beneficial effects:

[0065] 1. Enhance learning experience and engagement: By leveraging advanced VR and AR technologies combined with a high-precision physics engine to construct a hyper-realistic simulation environment, various driving scenarios are vividly reproduced, providing trainees with a highly immersive experience in terms of vision, touch, and operation. Trainees seem to be in a real road environment, practicing driving in the face of various road conditions and weather conditions, greatly enhancing the fun and immersion of learning, thus effectively improving trainees' learning enthusiasm and active participation, and enabling them to focus more on the learning and improvement of driving skills.

[0066] 2. Cover all driving scenarios comprehensively and improve driving skills:

[0067] By integrating a dynamic weather system and constructing a rich library of scenario elements, it is possible to simulate diverse driving scenarios including extreme weather (such as heavy rain, heavy snow, thick fog, etc.), different time periods (such as dusk, night), and various complex road conditions (such as school areas, construction sections, highways, rural roads, etc.). Trainees can train in this comprehensive and diverse simulation environment, encounter various situations that may occur in actual driving, familiarize themselves with the driving key points and coping strategies in different scenarios, which helps to comprehensively improve their driving skills and enables them to respond more calmly and proficiently when facing real complex road conditions.

[0068] 3. Reduce training costs and resource consumption: The virtual scenario training mode does not rely on a large number of physical sites, real vehicles, and resources such as fuel like traditional on-site training. There is no need to specially find suitable road sites for practice in different scenarios, reducing site rental or construction costs; at the same time, the use of simulation driving equipment replaces the frequent wear and tear of real vehicles and fuel consumption, greatly reducing the investment in hardware facilities and operating costs during the training process. This gives driving training a significant advantage in cost control, bringing economic benefits to both training institutions and trainees, and also contributing to the wider popularization of intelligent driving training.

[0069] 4. Collect multi-dimensional data and comprehensively evaluate driving behavior: By combining the collection of traditional driving operation data (such as steering wheel angle, accelerator and brake pedal operations, etc.) with biometric data (such as attention distribution obtained through eye tracking, emotional state and tension level reflected by facial expression recognition, etc.), the behavior performance and psychological state of trainees during driving are comprehensively recorded and analyzed from multiple dimensions. Compared with the evaluation method that only relies on traditional operation data, it can more deeply and accurately understand trainees' driving habits, reaction speed, attention concentration, and psychological changes when facing different situations, providing richer and more comprehensive data support for comprehensively evaluating trainees' driving skills and potential risks, helping coaches to more accurately understand the strengths and weaknesses of trainees, and thus formulating more targeted teaching plans.

[0070] 5. Improve the accuracy and reliability of assessment: Based on the multi-dimensional collected data, intelligent data cleaning techniques (such as clustering analysis, outlier detection, etc.) are used to effectively remove noise and redundant information to ensure data quality. Then, through feature engineering techniques, key features such as driving habits, reaction speed, and attention concentration are accurately extracted to provide high-quality input data for the risk assessment model. At the same time, the risk assessment model constructed using the deep reinforcement learning framework has powerful learning and optimization capabilities and can continuously improve the accuracy of the risk assessment of trainees' driving behaviors through iterative training. This all-round optimization from data collection to processing and then to model assessment effectively improves the accuracy and reliability of the assessment results and provides a more scientific and objective ability evaluation for trainees.

[0071] 6. Implement personalized teaching and improve learning efficiency: According to the real-time driving performance and risk assessment results of trainees, the system can automatically and dynamically adjust the difficulty and complexity of the virtual scenarios, and customize personalized learning paths for each trainee. For trainees with relatively weak driving skills or deficiencies in certain aspects, the system will provide relatively simple and basic scenarios for intensive practice; while for trainees with better skills and faster progress, the difficulty will be appropriately increased to challenge higher-level driving scenarios to ensure that the teaching content always matches the actual capabilities of the trainees. At the same time, the system can also provide targeted learning suggestions based on the driving behavior characteristics of trainees, such as guiding trainees to strengthen the practice of specific driving operations or adjust bad driving habits. This personalized teaching model can maximize the learning needs of different trainees, enable trainees to learn and grow at their most suitable pace, thus significantly improving learning efficiency, shortening the training cycle, and mastering solid driving skills faster.

[0072] 7. Provide real-time feedback and adjustment to promote the improvement of learning effects: The reinforcement learning risk assessment model can monitor the driving behaviors of trainees in real time, evaluate them at the moment of trainees' operations, and immediately feedback the results to the trainees. Trainees can timely understand whether their driving behaviors are correct, what risks exist, and how to improve, so as to correct wrong operations and adjust driving strategies in the first time. At the same time, the system will automatically adjust the assessment difficulty according to the real-time progress of trainees to maintain an appropriate level of challenge and encourage trainees to continuously improve themselves. This real-time feedback and adjustment mechanism effectively promotes the interaction between trainees and the teaching system, forms a virtuous learning cycle, helps trainees master correct driving skills faster, improve driving ability, and continuously strengthen safety awareness and the ability to handle emergencies during the learning process.

[0073] 8. Ensure the safety of trainees and reduce accident risks: When the intelligent safety response mechanism equipped in the system detects high-risk behaviors of trainees (such as sudden braking, sudden acceleration, excessive steering wheel rotation, etc.), it can quickly respond and conduct immediate intervention through various means such as simulated vehicle control, sound warnings, visual cues, and voice guidance. Such timely and effective intervention measures can remind trainees and correct their wrong operations at the first moment of potential danger, avoid the further development of dangerous behaviors leading to serious consequences, thus effectively ensuring the personal safety of trainees during driving practice and reducing the risk of accidents caused by operational errors. At the same time, through repeated simulation and response to various emergency situations, trainees can gradually accumulate emergency handling experience, improve their emergency response ability and psychological quality when facing sudden dangers in actual driving, and further enhance driving safety.

[0074] 9. Improve emergency handling ability and psychological quality: During the simulation training process, by deliberately setting and simulating various complex and changeable traffic environments and emergencies (such as vehicle failures, sudden road obstacles, dangerous driving behaviors of other vehicles, etc.), and guiding trainees to respond correctly, trainees can repeatedly experience and handle these emergencies in a safe environment, gradually familiarize themselves with the emergency handling process and skills, and improve their ability to respond to sudden crises. At the same time, in the face of simulated emergencies, the psychological quality of trainees can also be exercised and improved, learning to stay calm and make correct decisions under tension and pressure, which is crucial for safety in actual driving. After systematic emergency handling training, trainees can respond more calmly when facing real emergencies, reduce wrong operations caused by panic, and thus ensure their own and road traffic safety.

[0075] 10. Promote the development and innovation of the intelligent driving training industry: The ability evaluation method of the present invention comprehensively uses a variety of advanced technologies (such as VR, AR, artificial intelligence, big data analysis, etc.), innovatively solves many problems existing in the existing intelligent driving training systems, and provides a more efficient, safer, and more personalized training mode and solution for the intelligent driving training industry. Its successful application will lead the intelligent driving training industry to develop in a more intelligent, precise, and user-friendly direction, stimulate more relevant enterprises and research institutions to invest in technology research and development and product innovation, promote the technological progress and service quality improvement of the entire intelligent driving training industry, make positive contributions to cultivating more high-quality drivers with strong safety awareness, and thus have a positive and profound impact on the safety and development of the entire transportation industry. Brief Description of the Drawings

[0076] 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, without creative efforts, other drawings can be obtained based on these drawings.

[0077] Figure 1 It is a flowchart of the ability evaluation method of the present invention. Detailed implementation manners

[0078] Here, the exemplary embodiments will be described in detail, and the examples are shown in the drawings. When the following description involves the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementation manners described in the following exemplary embodiments do not represent all implementation manners 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.

[0079] Embodiment 1

[0080] The present invention aims to propose an ability evaluation method for enhancing closed scenarios, and is particularly applied to intelligent driving training to solve the following problems: realizing personalized teaching: analyzing the driving behaviors and performances of trainees through reinforcement learning algorithms, dynamically adjusting the training content and difficulty, and providing personalized learning paths for each trainee. Comprehensively simulating complex scenarios: constructing diverse closed driving scenarios, and using reinforcement learning algorithms to simulate complex and changeable traffic environments and emergencies to improve the coping abilities of trainees. Enhancing the intelligence of the safety mechanism: developing an intelligent risk assessment model based on reinforcement learning, real-time monitoring and predicting potential driving risks, triggering corresponding safety response mechanisms, and ensuring the safety of trainees during driving practice.

[0081] As Figure 1 shown, this embodiment provides an ability evaluation method for enhancing closed scenarios in intelligent driving training, including the following steps:

[0082] S1: Construct a hyper-realistic simulation environment, including selecting a physics engine and modeling, integrating a dynamic weather system, and constructing a custom scene editor to create diverse and realistic closed driving scenarios;

[0083] S2: Conduct data collection and preprocessing, collect traditional driving operation data and biometric data, and clean and extract features from the collected data to provide data support for risk assessment;

[0084] S3: Build a reinforcement learning risk assessment model, including selecting a deep reinforcement learning framework, defining the state space and action space, training the model, and deploying the model to achieve real-time assessment of the driving behavior of trainees;

[0085] S4: Implement safety response and personalized teaching, detect high-risk behaviors based on the risk assessment results and intervene immediately, and at the same time adjust the teaching difficulty and provide personalized learning suggestions according to the performance of trainees.

[0086] Further, in step S1, the selection of the physics engine and modeling is specifically as follows:

[0087] S11: Select the Unity or UnrealEngine physics engine to ensure that the physical laws and visual effects of the simulation environment are highly realistic;

[0088] S12: Use high-precision 3D modeling technology to create one or more elements including vehicles, roads, buildings, weather elements, etc., ensuring that every detail conforms to the physical characteristics of the real world.

[0089] Further, in step S1, the integration of the dynamic weather system is specifically as follows:

[0090] S13: Develop a dynamic weather system to simulate one or more environments such as rain, snow, fog, dusk, and night through algorithms;

[0091] S14: Provide adjustable weather parameters, and coaches can adjust one or more parameters such as the amount of rainfall, the density of snowflakes, and the concentration of fog through the system interface.

[0092] As an implementation method, in step S1 of this embodiment, the construction of the custom scene editor is specifically as follows:

[0093] S15: Develop an intuitive and easy-to-use scene editor, provide a graphical interface, and support coaches to quickly build or adjust specific scenes;

[0094] S16: Establish a scene element library including one or more of school areas, construction sections, highways, and rural roads, and coaches can select and combine these elements according to needs.

[0095] As an implementation method, in step S2 of this embodiment, the data collection is specifically as follows:

[0096] S21: Collect one or more traditional driving operation data such as steering wheel angle and accelerator brake pedal through sensors;

[0097] S22: Use a high-precision eye movement tracking system and facial expression recognition technology to collect one or more biometric data such as the attention distribution, emotional state, and tension level of trainees.

[0098] As an implementation, in step S2 of this embodiment, the data cleaning and feature extraction are specifically as follows:

[0099] S23: Use machine learning algorithms, including one or more of clustering analysis and outlier detection, to automatically filter out noise data and ensure data quality;

[0100] S24: Extract one or more key features such as driving habits, reaction speed, and attention concentration through feature engineering techniques.

[0101] As an implementation, in step S3 of this embodiment, the construction of the reinforcement learning risk assessment model is specifically as follows:

[0102] S31: Select a suitable deep reinforcement learning framework, including TensorFlow or PyTorch, to construct a risk assessment model;

[0103] S32: Define the state space of the model according to the characteristics of driving behavior, including one or more of vehicle speed, acceleration, and steering wheel angle, and the action space, including one or more of braking, accelerating, and steering;

[0104] S33: Divide the preprocessed data into a training set and a test set, and optimize the model parameters through iterative training so that it can accurately evaluate the driving behavior risk of trainees;

[0105] S34: During the training process, design a multi-objective optimization function by comprehensively considering evaluation accuracy, real-time performance, and fairness to ensure the model performance;

[0106] S35: Deploy the trained model to the intelligent driving training system to implement the real-time evaluation function. The system collects the driving data of trainees in real time, inputs it into the risk assessment model, and obtains the real-time risk assessment result.

[0107] As an implementation, in step S4 of this embodiment, the safety response and personalized teaching are specifically as follows:

[0108] S41: Detect one or more high-risk behaviors of trainees, including one or more of sudden braking, sudden acceleration, and excessive steering wheel rotation, according to the results of the risk assessment model;

[0109] S42: When a high-risk behavior is detected, the system immediately activates the safety response mechanism and intervenes by means of one or more of simulated vehicle control, sound warning, visual prompt, and voice guidance;

[0110] S43: According to the real-time performance of trainees and the results of the risk assessment, the system automatically adjusts the difficulty and complexity of the virtual scenario to provide a personalized learning path for trainees;

[0111] S44: Based on the driving behavior characteristics of the trainee, the system provides personalized learning suggestions, including strengthening practice in certain aspects, adjusting one or more of the driving habits.

[0112] Embodiment 2

[0113] The second aspect of the present invention provides an ability evaluation system for strengthening closed scenarios in intelligent driving training. This system is used for the above-mentioned method and includes:

[0114] A simulation environment construction module, used to construct a hyper-realistic simulation environment. This module includes:

[0115] A physical engine and modeling unit, used to select a physical engine and perform high-precision 3D modeling to create realistic driving scene elements. The physical engine is a Unity or Unreal Engine physical engine, and the modeling elements include one or more of vehicles, roads, buildings, and weather elements;

[0116] A dynamic weather system unit, used to develop a dynamic weather system and provide adjustable parameters to simulate one or more environments of rain, snow, fog, dusk, and night, and the coach can adjust one or more parameters such as the amount of rain, the density of snowflakes, and the concentration of fog through the system interface;

[0117] A custom scene editor unit, used to develop an intuitive and easy-to-use scene editor and establish a scene element library, supporting the coach to quickly construct or adjust specific scenarios. The scene element library includes one or more of school areas, construction sections, highways, and rural roads;

[0118] A data collection and processing module, used to collect and preprocess the driving data of the trainee. This module includes:

[0119] A data collection unit, used to collect one or more traditional driving operation data such as steering wheel angle and accelerator and brake pedals through sensors, and collect one or more biometric data such as the attention distribution, emotional state, and tension degree of the trainee by using a high-precision eye movement tracking system and facial expression recognition technology;

[0120] A data cleaning and feature extraction unit, used to automatically filter out noise data by using machine learning algorithms, including one or more of clustering analysis and outlier detection, to ensure the data quality, and extract one or more key features such as driving habits, reaction speed, and attention concentration through feature engineering techniques;

[0121] A risk assessment model construction and operation module, used to construct and operate a reinforcement learning risk assessment model. This module includes:

[0122] A model construction unit for selecting a suitable deep reinforcement learning framework, including TensorFlow or PyTorch, to construct a risk assessment model, and defining the state space and action space of the model according to the characteristics of driving behavior. The state space includes one or more of vehicle speed, acceleration, and steering wheel angle, and the action space includes one or more of braking, accelerating, and steering;

[0123] A model training unit for dividing the preprocessed data into a training set and a test set, and optimizing the model parameters through iterative training to enable it to accurately evaluate the driving behavior risks of trainees. During the training process, the evaluation accuracy, real-time performance, and fairness are comprehensively considered, and a multi-objective optimization function is designed to ensure the model performance;

[0124] A model deployment and evaluation unit for deploying the trained model into the intelligent driving training system to achieve real-time evaluation. The system collects the driving data of trainees in real time, inputs it into the risk assessment model, and obtains real-time risk assessment results;

[0125] A safety response and teaching adjustment module for performing safety responses and personalized teaching adjustments according to the risk assessment results. This module includes:

[0126] A high-risk behavior detection unit for detecting one or more high-risk behaviors of trainees, including emergency braking, emergency acceleration, and excessive steering wheel rotation, according to the results of the risk assessment model;

[0127] A safety response unit for immediately activating the safety response mechanism when high-risk behaviors are detected and intervening through one or more of simulated vehicle control, sound warnings, visual cues, and voice guidance;

[0128] A teaching adjustment unit for automatically adjusting the difficulty and complexity of the virtual scenario according to the real-time performance of trainees and the risk assessment results, providing personalized learning paths for trainees, and providing personalized learning suggestions according to the driving behavior characteristics of trainees, including strengthening practice in a certain aspect, adjusting driving habits, etc.;

[0129] Furthermore, the simulation environment construction module, data collection and processing module, risk assessment model construction and operation module, and safety response and teaching adjustment module are interconnected for data transmission and collaborative work;

[0130] It also includes a storage module for storing data during the operation of the system, including one or more of simulation environment data, trainee driving data, risk assessment model data, and teaching adjustment record data;

[0131] It further includes a display module for displaying one or more of the simulated driving scenarios, trainee driving data, risk assessment results, and personalized learning suggestions for the trainee and the coach to view;

[0132] The system can also be communicatively connected to external devices or systems, including connecting to a driving school management system to achieve sharing of trainee information and synchronization of training progress; connecting to a vehicle manufacturer system to obtain vehicle performance data and updated driving safety standard information; and connecting to a traffic management department system to upload trainee training data and ability assessment results for one or more of supervision and statistical analysis.

[0133] Embodiment III

[0134] This embodiment provides a specific implementation manner of an ability evaluation method and system for an enhanced closed scenario in intelligent driving training:

[0135] 3 Detailed elaboration of the technical solution of the present invention

[0136] 3.1 Realization of a hyper-realistic simulation environment

[0137] Step 1: Select a physics engine and perform modeling

[0138] Select a physics engine: Adopt industry-leading physics engines such as Unity or Unreal Engine to ensure that the physical laws and visual effects of the simulation environment are highly realistic.

[0139] High-precision 3D modeling: Use 3D modeling technology to create various elements including vehicles, roads, buildings, weather elements, etc., ensuring that every detail conforms to the physical characteristics of the real world.

[0140] Step 2: Integrate a dynamic weather system

[0141] Weather simulation: Develop a dynamic weather system to simulate various extreme weather conditions such as rain, snow, fog, etc., as well as low-light environments such as dusk and night through algorithms.

[0142] Weather parameter adjustment: According to teaching needs, the coach can adjust weather parameters such as the amount of rain, the density of snowflakes, the concentration of fog, etc. through the system interface.

[0143] Step 3: Build a custom scenario editor

[0144] Scenario editor development: Develop an intuitive and easy-to-use scenario editor, provide a graphical interface, and support the coach to quickly build or adjust specific scenarios.

[0145] Scenario element library: Establish a rich scenario element library, including school areas, construction sections, highways, rural roads, etc., and the coach can select and combine these elements as needed.

[0146] 3.2 Implementation of Data Collection and Preprocessing

[0147] Step 1: Data Collection

[0148] Traditional driving operation data: Collect physical operation data such as steering wheel angle, accelerator and brake pedals through sensors.

[0149] Biometric data: Use a high-precision eye tracking system and facial expression recognition technology to collect data such as the attention distribution, emotional state and tension level of trainees.

[0150] Step 2: Data Cleaning and Feature Extraction

[0151] Data cleaning: Use machine learning algorithms such as clustering analysis and outlier detection to automatically filter out noise data and ensure data quality.

[0152] Feature extraction: Extract key features such as driving habits, reaction speed, and attention concentration through feature engineering techniques to provide data support for the risk assessment model.

[0153] 3.3 Implementation of the Reinforcement Learning Risk Assessment Model

[0154] Step 1: Model Construction

[0155] Select the DRL framework: Select a suitable deep reinforcement learning framework such as TensorFlow or PyTorch to build a risk assessment model.

[0156] Define the state space and action space: According to the characteristics of driving behavior, define the state space (such as vehicle speed, acceleration, steering wheel angle, etc.) and action space (such as braking, accelerating, steering, etc.) of the model.

[0157] Step 2: Model Training

[0158] Data preparation: Divide the preprocessed data into a training set and a test set for model training and validation.

[0159] Training process: Through iterative training, continuously optimize the parameters of the model so that it can accurately evaluate the driving behavior risk of trainees.

[0160] Multi-objective optimization: During the training process, comprehensively consider evaluation accuracy, real-time performance and fairness, and design a multi-objective optimization function to ensure model performance.

[0161] Step 3: Model Deployment and Real-time Evaluation

[0162] Model deployment: Deploy the trained model to the intelligent driving training system to achieve real-time evaluation function.

[0163] Real-time assessment: The system collects the driving data of trainees in real time and inputs it into the risk assessment model to obtain real-time risk assessment results.

[0164] 3.4 Implementation of Safety Response and Personalized Teaching

[0165] Step 1: Safety Response Mechanism

[0166] High-risk behavior detection: According to the results of the risk assessment model, detect the high-risk behaviors of trainees, such as sudden braking, sudden acceleration, excessive steering wheel turning, etc.

[0167] Immediate intervention: When high-risk behaviors are detected, the system immediately activates the safety response mechanism and intervenes through means such as simulated vehicle control, sound warnings, visual cues, and voice guidance.

[0168] Step 2: Personalized Teaching

[0169] Difficulty adjustment: According to the real-time performance of trainees and the results of risk assessment, the system automatically adjusts the difficulty and complexity of the virtual scenario to provide trainees with a personalized learning path.

[0170] Learning suggestions: According to the driving behavior characteristics of trainees, the system provides personalized learning suggestions, such as strengthening practice in a certain aspect, adjusting driving habits, etc.

[0171] 4. Evaluation Process and Implementation Logic

[0172] 4.1 Evaluation Process

[0173] Step 1: Simulation Environment Evaluation

[0174] Fidelity test: By comparing the real driving environment with the simulation environment, evaluate the fidelity of the simulation environment, including aspects such as visual effects and physical laws.

[0175] Scene diversity test: Test whether the simulation environment covers various weather conditions, time periods, and complex road conditions to ensure that trainees can be exposed to various situations that may be encountered in actual driving.

[0176] Step 2: Data Collection and Processing Evaluation

[0177] Data integrity test: Check whether the collected data is complete, including traditional driving operation data and biometric data.

[0178] Data accuracy test: By comparing the actual driving behavior with the collected data, evaluate the accuracy of the data.

[0179] Feature extraction effect evaluation: Verify the extracted features through machine learning algorithms and evaluate the effect of feature extraction.

[0180] Step 3: Risk Assessment Model Evaluation

[0181] Accuracy test: Using the test set data, evaluate the accuracy of the risk assessment model, including the accuracy rate and recall rate of risk assessment.

[0182] Real-time test: Test the performance of the model in real-time evaluation to ensure that the model can give evaluation results within the specified time.

[0183] Fairness test: By comparing the evaluation results of different students, evaluate the fairness of the model to ensure that each student can get a fair evaluation.

[0184] Step 4: Safety response and personalized teaching evaluation

[0185] Safety response effect test: By simulating high-risk behaviors, test whether the safety response mechanism of the system is timely and effective. Personalized teaching effect test: According to the learning progress and performance of students, test whether the system can provide personalized learning paths and suggestions.

[0186] 4.2 Implementation logic and principle

[0187] Ultra-realistic simulation environment: Using a physics engine and 3D modeling technology, combined with a dynamic weather system and a custom scene editor, construct a highly realistic driving scene so that students can encounter various situations that may occur in actual driving in a safe environment.

[0188] Data collection and preprocessing: Through sensors and biometric recognition technologies, collect students' driving data and biometric data, and use machine learning algorithms for cleaning and feature extraction to provide high-quality data support for the risk assessment model.

[0189] Reinforcement learning risk assessment model: Based on a deep reinforcement learning framework, construct a risk assessment model, optimize the model parameters through iterative training, realize real-time assessment of students' driving behavior risks, and adjust the teaching difficulty and provide personalized suggestions according to the assessment results.

[0190] Safety response and personalized teaching: According to the results of the risk assessment model, the system activates the safety response mechanism for immediate intervention, and at the same time provides personalized learning paths and suggestions according to the learning progress and performance of students to help students grow at their most suitable pace.

[0191] Through the above detailed implementation steps and evaluation processes, the present invention can comprehensively cover driving scenarios, improve the learning experience, reduce training costs, comprehensively evaluate driving behaviors, improve evaluation accuracy, provide real-time feedback and adjustment, ensure the safety of students, enhance emergency handling capabilities, and achieve personalized teaching and other goals, providing a set of efficient, safe, and personalized solutions for the intelligent driving training industry.

[0192] 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 method for evaluating the ability of strengthening closed scenes in intelligent driving training, characterized by: The following steps are involved: S1: Building a hyper-realistic simulation environment, including selecting a physics engine and modeling, integrating a dynamic weather system, and building a custom scene editor to create diverse and realistic closed driving scenarios; S2: Perform data collection and preprocessing, collect traditional driving operation data and biometric data, and clean and extract features from the collected data to provide data support for risk assessment; S3: Build a reinforcement learning risk assessment model, including selecting a deep reinforcement learning framework, defining the state space and action space, training the model, and deploying the model to achieve real-time assessment of student driving behavior; S4: Achieve safe response and personalized teaching, detect high-risk behaviors and intervene immediately based on risk assessment results, adjust teaching difficulty and provide personalized learning suggestions based on student performance.

2. The ability evaluation method according to claim 1, characterized in that: In step S1, the physical engine selection and modeling are specifically as follows: S11: Select a physics engine to ensure that the physical laws and visual effects of the simulated environment are highly realistic; S12: Elements are created using high-precision 3D modeling technology, ensuring that every detail conforms to real-world physics.

3. The ability evaluation method according to claim 1, characterized in that: In step S1, the integrated dynamic weather system is specifically: S13: Develop a dynamic weather system that simulates one or more environments through algorithms; S14: Provide adjustable weather parameters. The coach can adjust one or more parameters including rainfall amount, snowflake density, and fog concentration through the system interface.

4. The ability evaluation method according to claim 1, characterized in that: In step S1, the construction of a custom scene editor is specifically as follows: S15: Develop an intuitive and easy-to-use scenario editor that provides a graphical interface to support coaches to quickly build or adjust specific scenarios; S16: Establish a scene element library containing one or more of school areas, construction sections, highways, and rural roads. The instructor can select and combine these elements as needed.

5. The ability evaluation method according to claim 1, characterized in that: In step S2, the data collection specifically includes: S21: collecting one or more traditional driving operation data of a steering wheel angle, an accelerator pedal, and a brake pedal through a sensor; S22: Use high-precision eye tracking system and facial expression recognition technology to collect one or more biometric data of students' attention distribution, emotional state and tension level.

6. The ability evaluation method according to claim 1, characterized in that: In step S2, the data cleaning and feature extraction are specifically as follows: S23: Use machine learning algorithms, including one or more of cluster analysis and outlier detection, to automatically filter out noisy data and ensure data quality; S24: Extract one or more key features of driving habits, reaction speed, and concentration through feature engineering technology.

7. The ability evaluation method according to claim 1, characterized in that: In step S3, the construction of the reinforcement learning risk assessment model is specifically as follows: S31: Choose a suitable deep reinforcement learning framework, including TensorFlow or PyTorch, to build a risk assessment model; S32: according to the characteristics of the driving behavior, define the state space of the model, including one or more of vehicle speed, acceleration, and steering wheel angle, and the action space, including one or more of braking, acceleration, and steering; S33: Divide the preprocessed data into a training set and a test set, and optimize the model parameters through iterative training so that it can accurately assess the driving behavior risks of the trainees; S34: During the training process, comprehensive consideration is given to evaluation accuracy, real-time, and fairness, and a multi-objective optimization function is designed to ensure model performance; S35: Deploy the trained model to the intelligent driving training system to realize the real-time evaluation function. The system collects the trainees’ driving data in real time and inputs it into the risk assessment model to obtain real-time risk assessment results.

8. The ability evaluation method according to claim 1, characterized in that: In step S4, the safety response and personalized teaching are specifically as follows: S41: Detecting one or more high-risk behaviors of the trainee including sudden braking, sudden acceleration, and excessive steering wheel turning according to the result of the risk assessment model; S42: When high-risk behavior is detected, the system immediately initiates a safety response mechanism and intervenes through one or more of simulated vehicle control, sound warnings, visual prompts, and voice guidance; S43: Based on the students’ real-time performance and risk assessment results, the system automatically adjusts the difficulty and complexity of the virtual scene to provide students with a personalized learning path; S44: Based on the student's driving behavior characteristics, the system provides personalized learning suggestions, including one or more of strengthening practice in a certain aspect and adjusting driving habits.

9. A capability evaluation system for strengthening closed scenes in intelligent driving training, characterized in that: The system is used for the method described in any one of claims 1 to 8 above, and the system comprises: The simulation environment building module is used to build a hyper-realistic simulation environment. The module includes: Physics Engine and Modeling Unit, which is used to select a physics engine and perform high-precision 3D modeling to create realistic driving scene elements; A dynamic weather system unit, used to develop a dynamic weather system and provide adjustable parameters to simulate one or more environments, and the instructor can adjust one or more parameters of the amount of rain, snow density, and fog concentration through the system interface; Customized scenario editor unit, used to develop an intuitive and easy-to-use scenario editor and establish a scenario element library, which allows coaches to quickly build or adjust specific scenarios. The scenario element library contains one or more of school areas, construction sections, highways, and rural roads; The data acquisition and processing module is used to collect and pre-process the student driving data. The module includes: A data acquisition unit, used to collect one or more traditional driving operation data such as steering wheel angle, accelerator and brake pedal through sensors, and to collect one or more biometric data such as trainee's attention allocation, emotional state and tension level by using high-precision eye tracking system and facial expression recognition technology; A data cleaning and feature extraction unit, which is used to automatically filter out noisy data using machine learning algorithms, including one or more of cluster analysis and outlier detection, to ensure data quality, and to extract one or more key features of driving habits, reaction speed, and concentration through feature engineering techniques; The risk assessment model construction and operation module is used to build and run the reinforcement learning risk assessment model. This module includes: A model building unit is used to select a suitable deep reinforcement learning framework, including TensorFlow or PyTorch, to build a risk assessment model, and define the state space and action space of the model according to the characteristics of driving behavior. The state space includes one or more of vehicle speed, acceleration, and steering wheel angle, and the action space includes one or more of braking, acceleration, and steering. The model training unit is used to divide the preprocessed data into training sets and test sets, and optimize the model parameters through iterative training so that it can accurately assess the driving behavior risks of trainees. In the training process, the accuracy, real-time and fairness of the assessment are comprehensively considered, and a multi-objective optimization function is designed to ensure the model performance; The model deployment and evaluation unit is used to deploy the trained model into the intelligent driving training system to realize the real-time evaluation function. The system collects the driving data of the trainees in real time and inputs it into the risk assessment model to obtain real-time risk assessment results. The safety response and teaching adjustment module is used to make safety responses and personalized teaching adjustments based on risk assessment results. This module includes: A high-risk behavior detection unit, used to detect one or more high-risk behaviors of the trainee including sudden braking, sudden acceleration and excessive steering wheel turning according to the result of the risk assessment model; A safety response unit is used to immediately initiate a safety response mechanism when high-risk behavior is detected, intervening through one or more of simulated vehicle control, sound warnings, visual prompts, and voice guidance; The teaching adjustment unit is used to automatically adjust the difficulty and complexity of the virtual scene according to the student's real-time performance and risk assessment results, provide students with a personalized learning path, and provide personalized learning suggestions based on the student's driving behavior characteristics, including one or more of strengthening practice in a certain aspect and adjusting driving habits.

10. The capability evaluation system according to claim 9, characterized in that: The simulation environment construction module, the data acquisition and processing module, the risk assessment model construction and operation module and the safety response and teaching adjustment module are interconnected to achieve data transmission and collaborative work; It also includes a storage module for storing data during the operation of the system, including one or more of simulation environment data, student driving data, risk assessment model data, and teaching adjustment record data; It also includes a display module for displaying one or more of the following information: simulated driving scenarios, student driving data, risk assessment results, and personalized learning suggestions, so that students and coaches can view them; The system can also communicate with external devices or systems, including connecting with driving school management systems to achieve student information sharing and training progress synchronization; connecting with vehicle manufacturer systems to obtain vehicle performance data and driving safety standards update information; connecting with traffic management department systems to upload student training data and ability assessment results for one or more of supervision and statistical analysis.

Images