Vehicle trafficability risk judgment method and system fusing driver ability and confidence

Abstract

The invention belongs to the field of road vehicle control systems, and relates to a vehicle trafficability risk judgment method and system fusing driver ability and confidence. The system comprises a driving behavior and environment data acquisition module, a scene matching module, a traffic capacity estimation module, a traffic capacity scoring module, a driving behavior feature analysis module, a traffic confidence scoring module, a trafficability risk calculation module and a database. The scene matching module is used for carrying out similarity calculation on the basis of a historical traffic sample set and the structural features of the current traffic scene and screening similar samples; the traffic capacity estimation module is used for carrying out weighted fusion on the screened traffic capacity and the calculated matching weight to obtain a traffic capacity estimation result; and the trafficability risk calculation module is used for correcting the original trafficability risk based on the structure complexity and the vehicle performance by using the traffic capacity score and the traffic confidence score. According to the system, personalized and dynamic adjustment of traffic risks is realized, and the accuracy of trafficability risk prediction is improved.

Description

Technical Field

[0001] The present invention belongs to the field of road vehicle control systems, involves the calculation or judgment of driving parameters, and specifically relates to a method and system for determining the passability risk of a vehicle by integrating driver ability and confidence. Background Art

[0002] During the driving process in complex terrains or high-risk areas such as steep slopes, gravel roads, and sections with concentrated obstacles, drivers often need to quickly judge the passability of the vehicle and adjust operations within a very short time to ensure smooth and safe passage. In this process, the driver's perception ability, confidence level, and actual operation skills directly affect their subjective judgment of passability risk and the choice of control strategies. If the driver's ability does not match the current driving scenario or the driver lacks confidence, it is very easy to lead to operation errors or judgment deviations, and then cause safety problems such as jamming and loss of control. Therefore, evaluating the driver's ability and state in a specific passability scenario and correcting the passability risk judgment accordingly are of great significance for improving vehicle operation safety.

[0003] At present, the methods for judging the passability risk of vehicles mainly include the following two categories: (1) Modeling of the passing ability based on vehicle physical performance and terrain environment: Such methods build models of the vehicle chassis structure, power system, and tire parameters, and combine terrain information such as slope, obstacle size, and soil type to theoretically analyze or simulate and evaluate whether the vehicle has the physical passing ability. For example, Chinese Patent CN 117087675 B judges the passing conditions through image recognition of the terrain and ground height detection, and Chinese Patent CN 114061969 B estimates whether there is a risk of rubbing or jamming of the vehicle based on parameters such as approach angle and departure angle. These methods can evaluate the adaptability of the vehicle to the environment, but generally ignore the influence of the driver's control ability and state on the actual passing result.

[0004] (2) Obstacle recognition and passability prediction methods based on images, lidar, and simulation platforms: Such methods use cameras or lidar to extract features such as the height, width, and spacing of obstacles, and combine vehicle parameters and passing strategies to predict the passing probability. For example, Chinese Patent CN 117087675 A judges the passability by combining real-time images and chassis height, Chinese Patent CN 116698157 A reconstructs the shape of obstacles through lidar point clouds to assist in decision-making, and Chinese Patent CN 115221672 A uses video scene reconstruction to create a virtual simulation environment and simulates the vehicle driving process to evaluate risks. These methods have advantages in static scene reconstruction and path prediction, but usually regard the driver as an ideal controller during the modeling process and lack modeling of the interactive influence of their actual operation behavior and the scene.

[0005] In addition, some studies have attempted to introduce the driver's mental state to correct risk judgment, and often use physiological signal acquisition methods such as heart rate, galvanic skin response, and blood pressure to infer their emotional stress or confidence level. Such methods have certain reference value under experimental conditions, but in practical applications, there are generally problems such as discomfort in wearing, signal delay, and environmental interference, making it difficult to meet the requirements of real-time and non-invasive in high-risk scenarios.

[0006] To sum up, the existing vehicle passability risk judgment methods still have the following deficiencies: First, the physical modeling method ignores the differences in the driver's manipulation ability, and it is difficult to depict the individual's risk contribution in operation only through driving style; Second, the obstacle recognition and simulation prediction method does not model the dynamic intervention of driving behavior on the passing process, and cannot realize the risk prediction at the level of vehicle-road interaction, resulting in insufficient risk prediction accuracy; Third, the psychological state recognition method based on physiological signals is limited by the perception method and deployment conditions, and its practical application value is limited. Therefore, it is urgent to construct a method that can non-invasively obtain driving behavior characteristics during actual driving, and dynamically evaluate the driver's confidence and ability in combination with the scene requirements and passing task requirements, so as to realize a more adaptable passability risk determination method. Summary of the Invention

[0007] In view of the shortcomings and deficiencies of the existing technology, the purpose of the present invention is to provide a vehicle passability risk determination method that integrates the driver's ability and confidence. This method non-invasively obtains driving behavior characteristics during actual driving, and dynamically evaluates the driver's confidence and ability in combination with the scene requirements and passing task requirements. Then, the passing ability and the driver's passing confidence score results are used as risk adjustment factors to realize the personalized and dynamic adjustment of passing risk and improve the accuracy of passability risk prediction.

[0008] To achieve the above purpose, the present invention adopts the following technical solutions: A vehicle passability risk determination method that integrates the driver's ability and confidence, the method includes the following steps: Step 1. Collect driving behavior and environmental data; Step 2. Evaluate the passing ability of the driver's current passing scene; Dynamically construct the structural characteristics of the current passing scene according to the collected environmental data, and match the structural characteristics of the current passing scene with the structural characteristics of the historical passing scenes in the historical passing sample set, and screen the passing scenes with a similarity greater than the set threshold; then normalize the screened similarity data to obtain the matching weight vector required for weighted fusion; finally, perform weighted fusion according to the passing ability vector in the screened historical passing sample set and the calculated matching weight vector to obtain the passing ability estimation result under the current passing scene; Step 3. Calculate the passing ability score of the driver's current passing scene: Construct a basic ability scoring function based on the static attributes of the driver, and combine the ability estimation results and structural characteristics in the current passing scenario. Use a structure-aware logistic regression model to predict the passing success probability. Finally, the passing ability scoring function fuses the results of the two parts and outputs the passing ability score of the driver in the current passing scenario; Step 4. Extract driving behavior feature indicators from the collected driving behavior data. The driving behavior feature indicators include four-dimensional grip force feature vectors of the left and right hands, micro-operation behavior feature vectors, and back posture feature vectors; Step 5. Calculate the passing confidence score of the driver in the current passing scenario based on the driving behavior feature indicators extracted in Step 4; Step 6. Incorporate the driver's state to modify the passing risk: Use the passing ability score in Step 3 and the passing confidence score in Step 5 to modify the original passing risk based on structural complexity and vehicle performance to obtain the final passing risk.

[0009] Preferably, in the present invention, the historical passing sample set in Step 2 is ; where is the passing ability vector of the i-th passing scenario in history; is the structural feature vector of the i-th passing scenario; is the passing result label, success = 1, failure = 0; , is the longitudinal operation smoothness; is the lateral control stability; is the lane-changing coherence; is the space utilization ability; is the ramp response ability; , is the effective passing width of the road; is the road curvature; is the degree of line of sight obstruction; is the road surface adhesion coefficient; is the road longitudinal slope angle, and N represents that the driver has experienced N representative passing scenarios; In Step 2, the Euclidean similarity function with a smoothing term is used to calculate the similarity.

[0010] Preferably, in the present invention, the expression of the basic ability scoring function in Step 3 is: ; where , indicating that the neutral driving style is the best; is the Sigmoid function; is a learnable parameter or an empirical setting; is the driving experience is the driving style label, and the output , representing the general passing ability score of the driver; The structure-aware logistic regression model includes a structure-aware weight generation model and a logistic regression model. The structure-aware weight generation model takes the structural features of the passing scenario as input and automatically generates the ability score weights in the current structural environment . The expression of the scoring function of the logistic regression model is: ; where is the Sigmoid function, which is used to compress the output to the interval to form a probability score. The superscript T represents transpose, and b represents the bias parameter in the model; is the estimated result of the ability vector of the current passing scenario, and the output , representing the predicted passing success probability of the driver under the current ability and structural scenario conditions; The expression of the passing ability scoring function is: ; where is the weight parameter; represents the passing ability score of the driver in the current structural scenario.

[0011] As a preference of the present invention, the extraction method of the four-dimensional grip force feature vectors of the left and right hands in step 4 is: the left and right hand grip forces of the driver are collected in real time through the flexible resistive grip force sensors installed at the 3 o'clock and 9 o'clock positions of the steering wheel, and then the original signals are preprocessed by adopting a two-stage filtering strategy. According to the preprocessed data, the intensity deviation degree, fluctuation stability, normalized mutation index, and change sharpness are extracted to construct the four-dimensional grip force feature vectors of the left and right hands and , and the expression is: ; ; where represents the four-dimensional grip force feature vector of the left hand, represents the four-dimensional grip force feature vector of the right hand, and respectively represent the intensity deviation degrees of the left and right hands, and respectively represent the fluctuation stabilities of the left and right hands, and respectively represent the normalized mutation indices of the left and right hands, and represent the sharpness changes of the left hand and the right hand respectively, and the superscript T represents transpose.

[0012] As a preference of the present invention, the extraction method of the micro-operation behavior feature vector in step 4 is: for the original throttle pedal pressure P a (t), the brake pedal pressure P b (t) and the steering wheel angle signals are processed by sliding window mean filtering. Then, the frequency domain features are extracted from the filtered signals using short-time Fourier transform to obtain the power spectral density of the signals, and the proportion of the main frequency energy is calculated according to the power spectral density , the peak frequency , the spectral entropy value . If the determination condition is satisfied, and <1.2, it is determined that the current time window is a micro-operation behavior segment. Then, the signals in the micro-operation behavior segment are extracted, and a damped oscillation function is used for non-linear fitting to extract the physical parameters after fitting. Based on the obtained physical parameters, the feature vector of the micro-operation behavior is constructed , and the expression is: where, is the initial amplitude, is the damping ratio, is the oscillation frequency, is the amplitude ratio.

[0013] As a preference of the present invention, the extraction method of the back posture feature vector in step 4 is: collect the three-dimensional coordinate information of the key points of the driver's left shoulder, the key points of the right shoulder and the middle point of the back spine, and compare and calculate the Euclidean distance with the three-dimensional coordinate information of the reference points corresponding to the left shoulder of the seat, the reference points corresponding to the right shoulder of the seat, and the center point of the seat backrest respectively to obtain the backrest posture distance index; then calculate the driver correction factor according to the driver's height and body type index, and standardize the backrest posture distance index using the driver correction factor to obtain the back distance ; calculate the shoulder angle according to the key points of the driver's left shoulder and the key points of the right shoulder; calculate the back posture stability according to the position of the back key points in the th frame and the average coordinates within the time window, and construct the standardized back posture feature vector , and the expression is: ; where, the superscript T represents transpose.

[0014] As a preference of the present invention, the expression of the passing confidence score of the driver's current passing scenario in step 5 is: ; Among them: , , are the normalized sub - scores of grip strength feature, micro - operation behavior feature, and back posture feature respectively; , , are the corresponding importance weights, satisfying the constraint .

[0015] As a preference of the present invention, the final passing risk in step 6 has the following expression: ; Among them, and are the risk sensitivity coefficients of ability deviation and confidence deviation respectively; represents the degree of relative insufficiency of the driver's ability; represents the degree of lack of confidence of the driver in the passing task, is the original passing risk based on structural complexity and vehicle performance; DCCI represents the passing confidence score.

[0016] As a further preference of the present invention, the intensity deviation has the following expression: ; Among them, is the grip strength value at the i - th sampling point within the time window, and are the mean and standard deviation of the baseline period respectively, and n represents the number of samples; The fluctuation stability has the following expression: ; The normalized mutation index has the following expression: ; Among them, is the range of grip strength in the baseline period; is the maximum grip strength value in the current window; is the minimum grip strength value in the current window; The change sharpness has the following expression: ; Among them, and represent the grip strength values at two adjacent sampling points.

[0017] As a further preference of the present invention, the main - frequency energy ratio The expression is: ; Wherein, is the power spectral density at the signal frequency ; The expression of the peak frequency is: ; The expression of the spectral entropy value is: ; Wherein, represents the power spectral density at the frequency point, represents the probability distribution normalized by the power spectral density at the frequency point, and N is the window size.

[0018] As a further preference of the present invention, the expression of the damped oscillation function in step 4 is: ; Wherein, is the initial amplitude, is the damping ratio, is the natural frequency of the system, is the actual oscillation frequency after damping; is the initial phase angle, set to a constant or zero; is the time variable after the start of the micro-operation, represents the signal of the extracted micro-operation behavior segment; The oscillation frequency ; The amplitude ratio is the ratio of the two peaks before and after.

[0019] As a further preference of the present invention, the expression of the normalized sub-score of the grip force feature is: ; Wherein, is the input vector of the left or right hand grip force feature; is the symmetric weight matrix, modeling the cross-action between features; is the first-order linear weight vector; is the scalar bias term; is the Sigmoid activation function, represents transpose vector of represents transpose vector of The expression of the normalized sub-score of the micro-operation behavior feature is: ; Among them, is the original micro-operation behavior feature vector, and are the hidden feature vector and the gating vector respectively; and are the hidden layer parameters, is the weight matrix of the fully connected layer, is the corresponding bias term; the gating unit consists of the weight matrix of the linear fully connected layer and the bias term , and the generated gating vector is used for the per-channel control feature fusion strategy; , are the output linear combination weights; is the non-linear Tanh activation function; is the Sigmoid function; ∘ represents element-wise multiplication, represents the fused micro-operation behavior feature vector, represents the micro-operation behavior feature score; The expression for the normalized sub-score of the back posture feature is: ; Among them, is the spatial posture feature; is the dynamic posture fluctuation feature; and are the fully connected layer parameters of their respective branches; is the non-linear Tanh activation function of each branch; is the intermediate feature representation of the outputs of the two branches; is the final fusion weight and bias, acting on the overall representation after the branches are concatenated; is the Sigmoid function.

[0020] As a further preference of the present invention, a maximum upper limit is set for the corrected passability risk value in step 6 , and when the calculated passability risk value is greater than the maximum upper limit , the output passability risk value is the maximum upper limit .

[0021] The present invention also provides a vehicle passability risk determination system that fuses driver ability and confidence. The system is used to implement the vehicle passability risk determination method that fuses driver ability and confidence described above. The system includes a driving behavior and environment data collection module, a scene matching module, a passing ability estimation module, a passing ability scoring module, a driving behavior feature analysis module, a passing confidence scoring module, a passability risk calculation module, and a database; Among them, the driving behavior and environment data acquisition module is used to collect the driver's behavior data, the driver's static characteristics and historical passing data, and the structural feature information of the current passing scene in real time. The driver's behavior data includes grip strength, micro-operations, and postures. The scene matching module calculates the similarity based on the historical passing sample set and the structural features of the current passing scene, and filters the top-K similar samples. The passing ability estimation module is used to normalize the filtered similarity data to obtain a matching weight vector; perform weighted fusion based on the passing ability vector in the filtered historical passing sample set and the calculated matching weight vector to obtain the passing ability estimation result under the current passing scene. The passing ability scoring module is used to calculate the passing ability score of the driver in the current passing scene. The driving behavior feature analysis module is used to filter and perform time-frequency analysis on the original behavior signals of grip strength, micro-operations, and postures, and construct four-dimensional grip strength feature vectors, micro-operation behavior feature vectors, and back posture feature vectors for both hands. The passing confidence scoring module includes a grip strength feature scoring module, a micro-operation behavior feature scoring module, a back posture feature scoring module, and a total confidence scoring module. The passability risk calculation module is used to obtain the original passability risk based on structural complexity and vehicle performance, and use the passing ability score and the passing confidence score to correct the original passability risk based on structural complexity and vehicle performance, and output the passability risk considering the individual state. The database stores the driver's historical passing sample set and personal feature data. The historical passing sample set stores the structural feature vectors and passing ability vectors of historical passing scenes.

[0022] Advantages and beneficial effects of the present invention: (1) To solve the problem that the existing passability evaluation results do not consider the influence of the driver's subjective state, the present invention proposes to use the passing ability and the driver's passing confidence scoring results as risk adjustment factors. By introducing the driver's confidence and ability state and the current structure matching result, the original passability risk index based on vehicle physical performance is corrected, realizing personalized and dynamic adjustment of the passing risk, and enhancing the system's safety perception ability in unstructured complex scenes.

[0023] (2) To solve the problem that it is difficult to accurately identify the current driving confidence state of drivers, the present invention proposes a state perception method that integrates grip strength, micro-operation frequency, and attitude stability. This method can jointly identify the driving operation state and mental state of drivers in a low-invasive and unobtrusive manner without relying on interfering physiological sensors, and then construct a driver's passing confidence index to achieve both practicality and real-time performance of state recognition.

[0024] (3) Aiming at the problem that existing passability risk judgment methods do not fully consider driver individual differences and ability states, the present invention proposes a capacity scoring method based on structural perception, which matches and models the driver's passing ability score with terrain structure features. This method enhances the individual adaptability of the system to passing judgments in complex scenarios and improves the personalization and accuracy of passability assessment.

[0025] (4) To improve the generalization effect of driver ability assessment in complex scenarios, the present invention proposes a capacity transfer mechanism based on structural similarity, which performs Top-K screening and weighted fusion through the matching degree between historical passing samples and the current structural scenario, effectively solving the problem that driver ability cannot be directly transferred to new scenarios, enhancing the adaptability of the model to unseen structural combinations, and improving the stability and accuracy of ability estimation.

[0026] (5) To achieve a fast response link from risk perception to control execution, the present invention constructs a multi-level auxiliary control strategy system triggered by risk levels, and through measures such as voice prompts, throttle limit, speed limit control, and path replanning, it realizes corresponding interventions for different risk levels from L0 to L4, ensuring that the system provides effective compensation when the driver's cognition is insufficient, and significantly improving the safety guarantee ability of the vehicle in complex passing scenarios.

[0027] (6) By introducing a damped oscillation function to model micro-operation behaviors, the present invention can accurately reflect the periodic adjustment, operation hesitation, and amplitude attenuation characteristics shown by drivers under specific passing pressures, improving the model's ability to depict driving behaviors and the sensitivity of passing risk judgment.

[0028] (7) Through the spatial comparison between the attitude points and the seat reference points, the present invention realizes attitude standardization processing without additional calibration conditions, improving the comparability of attitude features among different drivers. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Through the following description in conjunction with the drawings, and with a more comprehensive understanding of the present invention, other objects and results of the present invention will become more obvious and easier to understand. In the drawings: Figure 1 is the flow chart of the vehicle passability risk judgment method that integrates driver ability and confidence of the present invention; Figure 2 Structural block diagram of the vehicle passability risk determination system that integrates driver ability and confidence according to the present invention. Specific implementation manners

[0030] To enable those skilled in the art to better understand the technical solutions and advantages of the present invention, the present application will be described in detail below with reference to the accompanying drawings, but is not used to limit the protection scope of the present invention.

[0031] Example 1:

[0032] The present invention proposes a vehicle passability risk determination method that integrates driver ability and confidence. Figure 1 Flow chart of the vehicle passability risk determination method that integrates driver ability and confidence for this example, as Figure 1 shown, the vehicle passability risk determination method that integrates driver ability and confidence includes the following steps: Step 1. Collect driving behavior and environmental data: Step 1.1. Collect short-term driving state data:

[0033] In this example, flexible resistive grip sensors are respectively embedded at the 3 o'clock and 9 o'clock positions of the steering wheel. The flexible resistive grip sensors are used to collect the left and right hand grips of the driver in real time, and the collection frequency f s = 10 Hz, specifically including the left hand grip F L (t), and the right hand grip F R (t).

[0034] In this example, through the original sensors of the vehicle, longitudinal and lateral micro-operation data are collected, and the collection frequency f s = 10 Hz, specifically including the throttle pedal pressure P a (t), the brake pedal pressure P b (t), and the steering wheel angle .

[0035] Through the RGB-D camera and the human pose recognition algorithm, the three-dimensional coordinate information of three key skeleton points on the upper body of the driver is collected, specifically including the key point of the driver's left shoulder, denoted as ; the key point of the driver's right shoulder, denoted as ; the middle point of the driver's back spine, denoted as .

[0036] To achieve an accurate judgment of the driving posture state, three fixed spatial reference points corresponding to the above points are also set on the seat backrest in this example, specifically including the reference point corresponding to the left shoulder of the seat, denoted as ; the reference point corresponding to the right shoulder of the seat, denoted as ; The center point of the seat back, located on the vertical axis and at the same height as the middle section of the driver's back, is denoted as .

[0037] Step 1.2. Collect historical driving data of the vehicle: To model the driver's capabilities and evaluate adaptability in specific passability scenarios, collect their passing records in multiple historical scenarios and construct their passing ability characteristics. On this basis, combined with the structural information of the current scenario to be passed, calculate the passability ability vector of the driver in this scenario for subsequent passing risk determination.

[0038] Step 1.2.1. Record of historical scenario structure characteristics: Suppose the driver has experienced representative passing scenarios and construct their historical scenario set : ; Among them, is the i-th passing scenario, and each scenario records five structural characteristics. is the effective passing width of the road; is the road curvature; is the degree of line-of-sight obstruction; is the road surface adhesion coefficient; is the road longitudinal slope angle; each group describes the structural requirements for driving behavior in this scenario, and the data sources include maps, visual perception, and ground condition recognition modules.

[0039] Step 1.2.2. Record of historical scenario behaviors and ability extraction: For the driving process of the driver in the th historical scenario, record their complete trajectory and control behavior data to form a time series , and use the CNN-LSTM network to extract five types of driving behavior indicators as the basis for special capabilities, and construct their historical scenario ability set : ; Among them, is the passing ability vector of the i-th passing scenario, is the longitudinal operation smoothness; is the lateral control stability; is the lane-changing coherence, that is, the smoothness of the angular velocity curve in the lane-changing section; is the space utilization ability, that is, the mean and variance of the closest distance to the obstacle; is the ramp response ability, that is, the average acceleration adjustment rate in the ramp section.

[0040] Meanwhile, record the passing result label of the driver in this scenario , which is used to indicate whether the passing is successful (1 means passing smoothly, 0 means failure, jamming or detour in the middle).

[0041] Step 1.3. Collect the driver's personal characteristic data: Specifically, preset the collection of the driver's basic information through the in-vehicle identity recognition system or the driver login interface, including the driver's height ; the driver's BMI body type factor ; the driving years ; the driving style label (conservative = 0, neutral = 0.5, aggressive = 1). The collected data is stored in the vehicle user profile for subsequent attitude normalization and risk judgment model calls.

[0042] Step 2. Evaluate the passing ability of the driver in the current passing scenario: Step 2.1. Structure matching between the current passing scenario and historical scenarios: Extract the structural features of the current passing scenario from the sensor perception information or high-definition map , including the road width , the road curvature , the degree of line-of-sight occlusion , the road surface adhesion coefficient and the road longitudinal slope angle , which respectively correspond to five dimensions in the structure vector. For different scenario types or road environments, these structural feature values will show differences. In this embodiment, the structure vector of the current passing scenario is dynamically constructed and used as the core matching basis.

[0043] The historical passing sample set of the driver is ; among them, represents the i-th passing scenario, recording five structural features, represents the passing ability vector of the i-th passing scenario, is the passing result label.

[0044] To measure the degree of structure matching of the current passing scenario, use the Euclidean similarity function with a smoothing term for calculation: ; Among them, is the similarity between the i-th passing scenario and the current passing scenario; is a small constant to avoid the denominator being zero, usually taking the value of .

[0045] Step 2.2. Top-K similar scenario screening and weighting: To ensure that the historical samples referred to in the estimation of the driver's passing ability are representative and credible in terms of structural attributes, a structural similarity threshold control mechanism is introduced in the similar scenario sample screening stage of this embodiment. This mechanism sets a matching degree threshold , and only retains the historical passing scenario samples that meet to participate in the ability estimation, excluding samples with too large structural differences, and improving the effectiveness and generalization stability of the judgment model.

[0046] According to all similarity scores , sort the historical samples in descending order, and select the first most similar passing scenario numbers to form a set , where ; represents the similarity score of the most similar passing scenario , represents the similarity score of the most similar passing scenario , represents the similarity score of the most similar passing scenario . The scenarios in this Top-K subset are considered to be the closest to the current passing scenario in terms of structural attributes and can be used as a reference basis for ability transfer. It can be set that ≤ as an empirical value to avoid excessive averaging.

[0047] To ensure the numerical stability and interpretability in the ability fusion process, it is necessary to normalize the similarities in the above Top-K samples to obtain the matching weight coefficients required for weighted fusion. This weight vector satisfies , so that the more similar historical scenarios will obtain larger weights, represents the weight coefficient of the most similar passing scenario , represents the weight coefficient of the most similar passing scenario .

[0048] Step 2.3. Calculation of the passing ability of the current passing scenario: Using the historical passing ability vectors of the drivers in the selected Top-K samples, perform weighted fusion to obtain the ability estimation result under the current passing scenario. The expression is: ; Among them, represents the passing ability vector of the most similar passing scenario . Finally, the ability vector under the current passing scenario is obtained, where , respectively represent the longitudinal operation smoothness, lateral control stability, lane-changing coherence, space utilization ability, and ramp response ability in the current traffic scenario.

[0049] Step 3. Calculate the traffic capacity score of the driver in the current traffic scenario: To achieve a quantitative evaluation of the traffic capacity of a driver in a traffic scenario with a specific structure, the present invention proposes a traffic capacity scoring function that integrates individual attribute characteristics and the ability-structure matching relationship. This traffic capacity scoring function constructs a basic ability scoring function based on the driver's static attributes, combines the ability estimation results and structure characteristics in the current traffic scenario, uses a structure-aware logistic regression model to predict the traffic success probability, and finally integrates the results of the two parts to output the traffic capacity score of the driver in this traffic scenario.

[0050] Specifically, in this embodiment, the driver's static individual attributes, including driving years and driving style label , are used to construct a basic ability scoring function for characterizing the general ability tendency of the driver. The expression of the basic ability scoring function is: ; where , indicating that the neutral driving style is the best; is the Sigmoid function; is a learnable parameter or empirical setting; the output represents the general traffic capacity score of the driver.

[0051] In this embodiment, the structure-aware logistic regression model includes a structure-aware weight generation model and a logistic regression model. The structure-aware weight generation model takes the structure feature S of the traffic scenario as input, automatically generates the ability scoring weights in the current structure environment, and then constructs a logistic regression model according to the ability scoring weights.

[0052] Specifically, according to the structure feature vector , , , , of the current traffic scenario and the ability vector estimation result of the current traffic scenario, first bring the structure feature vector into the structure-aware weight generation model , generate the ability scoring weights from the structure features, and then construct a logistic regression model. The expression of the scoring function of the logistic regression model is: ; where is the Sigmoid function, which is used to compress the output to the interval to form a probabilistic score. The superscript T represents transpose, and b represents the bias parameter in the model, which is used to adjust the passing score benchmark value under the current input conditions; the model score output , indicating the predicted passing success probability of the driver under the current ability and structural scenario conditions.

[0053] Finally, the passing ability scoring function fuses the basic ability and the scenario scoring results, and the expression is: ; where is the weight parameter, which can be set fixedly or adjusted dynamically based on the structural complexity; represents the passing ability score of the driver under the current structural scenario.

[0054] In this embodiment, by introducing a structure-aware weight generation model, taking the structural feature S of the passing scenario as the input, an ability scoring weight vector under the current structural environment is automatically generated, thereby constructing a mapping mechanism between ability-structure-result. This structure-aware weight generation model has the generalization ability for unseen structural conditions, enabling it to still output reasonable scoring results when facing new structural combinations, improving the scenario adaptability and inference robustness; specifically, ; ;

[0055] where to respectively represent the weights of road width, road curvature, sight distance occlusion degree, road surface adhesion coefficient and road longitudinal slope angle, represents the ability scoring weight generated according to the structural features, to respectively represent road width, road curvature, sight distance occlusion degree, road surface adhesion coefficient and road longitudinal slope angle; In passing scenarios with different structures, there are significant differences in the requirements for driving ability. For example, in complex curves, lateral control ability is particularly crucial; while in ramps or narrow lanes, speed control and micro-operation accuracy are more important; if the scoring function only relies on fixed empirical weights for scoring, it cannot accurately express the relative importance of ability indicators under different structural features, resulting in a lack of adaptability and predictability in the evaluation.

[0056] To solve this problem, this embodiment trains the model and introduces the passing result label As a supervision signal, automatically learn the joint influence relationship between the traffic capacity vector and the structural features. The training process enables the model to identify which combinations of capabilities are more likely to lead to successful passage and which are more risky under different structural feature conditions.

[0057] In the training stage, use historical traffic sample data to perform supervised learning on the traffic capacity scoring function. Among them, is the traffic capacity vector of the i-th traffic scenario in history; is the structural feature vector of the i-th traffic scenario; is the traffic result label (success = 1, failure = 0); represents the passing probability calculated for the i-th traffic scenario in history, where b is the bias parameter, and the training objective is to minimize the cross-entropy loss function: ; Step 4. According to the collected driving behavior data, extract driving behavior feature indicators: Step 4.1. Extract grip force features: Suppose the system samples the grip force signal within the time window at the sampling frequency to obtain samples, denoted as .

[0058] The system requires the driver to collect grip force data for a certain period of time in a normal driving state. At this time, the driver should be in a normal and relaxed driving posture, without emotional fluctuations or stress responses; by calculating the mean , standard deviation and range of the grip force in the normal state, represents the maximum grip force collected, represents the minimum grip force collected, which can more accurately identify abnormal situations where the grip force exceeds the normal fluctuation range, thereby effectively distinguishing abnormal grip force performances in specific situations and further improving the evaluation accuracy of traffic confidence.

[0059] In this embodiment, to improve the smoothness of the grip force signal and the accuracy of feature extraction, a two-stage filtering strategy is adopted to preprocess the original signal. First, use a first-order IIR low-pass filter to suppress high-frequency noise above 10 Hz and remove interference due to outliers. Subsequently, introduce a Kalman filter to recursively smooth the signal, and by constructing a state transition equation and an observation model, effectively fuse historical information and current observations to dynamically estimate the change trend of the grip force.

[0060] On this basis, the system calculates the real-time collected left hand grip force F L (t), right hand grip force FR (t) data, extract the following four types of characteristic indicators: 1. Strength deviation : Measure the degree of deviation of the overall strength of the grip force during the period relative to the driver's personal normal grip force level.

[0061] ; Among them, is the grip force value at the i-th sampling point within the time window, and are the mean and standard deviation during the baseline period (under normal conditions), respectively.

[0062] 2. Fluctuation stability : Characterize the degree of stability of the current grip force signal's fluctuation amplitude relative to the baseline period. When is close to 1, it indicates that the fluctuation is consistent with the normal state; when the grip force remains highly stable, which may be due to highly concentrated or rigid gripping; when the fluctuation is significantly enhanced, which may represent behavioral hesitation, lack of confidence, or excessive corrective actions.

[0063] ; 3. Normalized mutation index : Measure the instantaneous extreme change degree of the grip force within the current time window, and reflect whether there are obvious short-term mutations or jump operations. The index is particularly sensitive to capturing local sudden psychological changes.

[0064] ; Among them, is the range of grip force during the baseline period (under normal conditions); is the maximum value of the grip force in the current window; is the minimum value of the grip force in the current window.

[0065] 4. Change sharpness : Measure the relative change amplitude between adjacent sampling points in the time evolution process of the grip force signal, and use it to evaluate the smoothness of its change process. When the value of the change sharpness r is larger, it indicates that the change is more discontinuous, which may represent rapid actions, continuous fine-tuning, or mental tension. Attention should be paid to the driver's passing confidence state or potential risks.

[0066] ; Among them, and represent the grip force values of two adjacent sampling points; Based on the above calculation process, based on the left and right hand grip force signal sequences within the set time window, denoted as respectively, construct four-dimensional grip force feature vectors for the left and right hands: ; ; Among them, the definition method of each item is the same as the original feature, only replacing the signal with the grip force sequence of the corresponding hand or , where the subscript L represents the left hand and R represents the right hand.

[0067] Step 4.2. Extract the micro-operation frequency features: For the original accelerator pedal pressure P a (t), brake pedal pressure P b (t) and steering wheel angle signals, perform moving window mean filtering with a window length of 0.2 seconds (i.e., 2 sampling points) to suppress high-frequency noise. The specific formula is as follows: ; Among them, is the filtered signal; is the window size, indicating that each time window contains 2 sampling points; is the sampling interval; is the original signal, , or one of them; perform a first-order difference on the filtered signal to calculate the gradient for capturing the instantaneous changes of micro-operations and identifying sudden adjustments of acceleration, braking, or steering.

[0068] To further analyze the frequency components of micro-operation behaviors, the present invention performs time-domain to frequency-domain conversion on the smoothed signal, and uses the short-time Fourier transform (STFT) on the filtered signal to extract frequency-domain features. The window length is 5 seconds (i.e., 50 sampling points), and the overlap rate is 50% to analyze the frequency changes of the signal in different time periods.

[0069] ; Among them, is the result of the short-time Fourier transform, indicating the frequency component of the signal at time and frequency ; is the window function, used to define the time window for each transformation; is the signal sequence, and the signal value at time (the observed value of the signal at time ).

[0070] Through the Fourier transform, the power spectral density of the signal is obtained , representing the energy distribution of the signal at different frequencies.

[0071] ; Among them, is the power spectral density at frequency , representing the energy of this frequency component; is the observed value of the signal at time ; is the frequency, with the unit of Hz.

[0072] After converting to frequency-domain features, the present invention further extracts key frequency-domain judgment indicators for judging whether the signal exhibits the characteristics of micro-operations. The present invention sets the micro-operation frequency range to 0.5 Hz to 3.0 Hz, based on the understanding of the controllable action frequencies of human hands and feet in human factors engineering. This frequency band covers the stable adjustment frequency range that can be achieved by humans during continuous active operations, which can not only reflect frequent small-amplitude correction behaviors but also effectively distinguish non-controllable noise from normal operations. On this basis, the system calculates the following micro-operation judgment indicators.

[0073] 1. Proportion of main frequency energy : , defining the proportion of the energy of the signal in the frequency band from 0.5 Hz to 3 Hz to the total energy as the key indicator for judging the micro-operation frequency component. This frequency band usually reflects micro-operation behaviors such as frequent small-amplitude acceleration or correction of the steering wheel, etc.; among them, is the power spectral density of the signal at frequency ; 0.5 Hz ≤ ≤ 3 Hz is the micro-operation frequency band, representing frequent small-amplitude adjustments.

[0074] 2. Peak frequency : Used to identify the frequency corresponding to the maximum amplitude in the power spectrum. If falls within the frequency band from 0.5 Hz to 3 Hz, it is determined as the micro-operation dominant frequency, indicating that there are frequent small-amplitude operations in the signal, such as continuous small-amplitude acceleration, braking, or steering adjustments. The maximum spectral energy corresponding frequency point can be determined by traversing the power spectral density sequence , , that is, looking for the frequency corresponding to the maximum value of ; if the peak frequency , it is considered that the current operation signal is characterized by micro-operation dominance, that is, frequent small-amplitude repeated adjustments; if is not within this range, it is considered that the signal does not have typical micro-operation characteristics.

[0075] 3. Spectral entropy value : To exclude hybrid behaviors or noise domination, the power spectral density at each frequency point is normalized to a probability distribution . If is close to a uniform distribution, that is, all frequency energies are similar, the entropy value is close to the maximum; if is highly concentrated, that is, only a few frequencies dominate, the entropy value is small; the entropy value of this distribution is calculated using the Shannon entropy formula , which is the spectral entropy value and measures the "dispersion" or "chaos" of the frequency energy distribution.

[0076] ; In the present invention, if the determination condition is satisfied, and < 1.2, it is determined that the current time window is a micro-operation behavior segment.

[0077] The present invention uses a damped oscillation function to extract the micro-operation characteristics of the driver. The core advantage of the damped oscillation function is to simulate the behavior of "gradually decaying amplitude and periodic adjustment", which is highly consistent with the actual control actions of the driver during micro-operation. During the time period determined as micro-operation, the signal of this segment is extracted and nonlinearly fitted using the following damped oscillation function: ; where is the initial amplitude, used to reflect the initial adjustment amplitude of the micro-operation; is the damping ratio, measuring the attenuation speed of the operation and reflecting the convergence and determination degree of the operation; is the natural frequency of the system, reflecting the potential operation response rhythm; is the actual oscillation frequency after damping; is the initial phase angle, set to a constant or zero; is the time variable after the start of the micro-operation.

[0078] The fitting method is implemented using a least squares nonlinear optimization tool to extract the physical parameters after fitting.

[0079] Based on the damping model, key behavior characteristic indicators are calculated, including the initial amplitude ; the damping ratio ; the oscillation frequency ; the amplitude ratio which represents the ratio of two consecutive peaks to judge whether there is repeated fine-tuning.

[0080] Based on the above frequency determination and dynamic modeling process, the feature vector of the micro-operation behavior is constructed as follows: ; Step 4.3. Extract the driver's back posture features: 1. Driving back - backrest distance: Calculate the three - dimensional Euclidean distances between the above three pairs of points respectively for the three - dimensional coordinate information extracted in Step 1.1: ; Among them, represents the Euclidean distance operation, which is used to measure the absolute displacement between point pairs in three - dimensional space; represents the Euclidean distance between the key point of the driver's left shoulder and the corresponding position on the backrest, represents the Euclidean distance between the key point of the right shoulder and the corresponding position on the backrest, represents the Euclidean distance between the mid - point of the driver's back spine and the corresponding position on the backrest. By fusing the above three items, construct the total backrest posture distance index: ; Among them, is the overall backrest posture distance index; , , are the weighting coefficients of, and set , and can be adjusted according to the feature stability.

[0081] To avoid the influence of individual body size differences on the backrest posture features, introduce the driver correction factor , and standardize the driver's back distance according to height , body type index . The calculation formula is as follows: ; Among them, the parameter values are = 170cm, = 22, = 0.15, = 0.10, and the calculated is used to standardize the back distance .

[0082] ; 2. Shoulder angle : Reflects the angle between the line connecting the left and right shoulders and the horizontal direction, and judges whether there are "shoulder shrugging" or asymmetric twisting movements. The change of shoulder posture is a non - verbal feedback of the driver's assessment of environmental risks. Under normal driving conditions, the driver's shoulders should maintain a high degree of left - right symmetry. When passing through narrow roads, going up and down slopes or encountering sudden obstacles, the pressure increases, and some drivers will have stress reactions such as shoulder tension, slight shoulder shrugging or unilateral sinking.

[0083] ; 3. Back Posture Stability : Analyze the stability of the back points within the time window, reflecting whether the driver's posture changes violently during consecutive time periods; define the sliding window size as ; then: ; Among them, is the position of the back key points in the th frame; is the average coordinate within the time window.

[0084] Based on the above calculation process, construct a standardized back posture feature vector , and the expression is: ; Step 5. Based on the driving behavior feature indicators extracted in Step 4, calculate the passing confidence score of the driver's current passing scenario: Step 5.1. Driver's Grip Force Feature Score: There is a significant interaction effect among the grip force feature dimensions. "High grip force intensity + high volatility" is often more significant in risk indication than a single factor. To capture such non-linear interaction relationships, a quadratic form combined with a neural network is used to model the grip force feature score. Respectively input and , and find the mean to obtain : ; Among them, is the grip force feature input vector of the left or right hand; Symmetric weight matrix, modeling the cross-action between features; is the first-order linear weight vector, retaining the independent linear influence of each feature on the score; is the scalar bias term, used to offset the overall score benchmark; is the Sigmoid activation function, represents transpose vector of, represents transpose vector of.

[0085] Step 5.2. Driver's Micro-operation Behavior Feature Score: Aiming at the problem of coexistence of features such as "amplitude, frequency, stability and repeated adjustment" in micro-operation behaviors, a gated residual network structure is used as the scoring function modeling method for micro-operation behavior features. In this structure, on the one hand, the original micro-operation features are retained, and on the other hand, potential non-linear combinations are extracted, and through the hidden feature vector and the gating vector Adjust the information flow for fusion to improve the balance performance of the model between stability modeling and generalization ability. The structure of the scoring function is as follows: ; Among them, is the original micro-operation behavior feature vector, which respectively reflects the initial response of the operation, damping convergence, frequency components, and amplitude fluctuation behavior; and are the hidden layer parameters. The former is the weight matrix of the fully connected layer, and the latter is the corresponding bias term. After extracting the high-order combined features through the activation function, the hidden feature vector is obtained. The two jointly determine the projection method of the input features in the hidden space; The gating unit consists of the weight matrix of the linear fully connected layer and the bias term . The generated gating vector is used for the per-channel control feature fusion strategy; , are the output linear combination weights; is the non-linear Tanh activation function, which can extract the potential high-order combined structure in the micro-operation behavior; is the Sigmoid function; ∘ represents element-wise multiplication, represents the fused micro-operation behavior feature vector, represents the micro-operation behavior feature score.

[0086] Step 5.3. Driver's back posture feature score: To accurately model the impact of the driver's back posture on the passing state, the present invention divides the posture features into two categories: spatial posture features and dynamic stability features, and designs a lightweight double-branch fusion neural structure accordingly. This structure can model the static position features and dynamic fluctuation features respectively while ensuring interpretability, and fuse them to generate the final score. The structure of the scoring function is as follows: ; Among them, are the spatial posture features, which respectively represent the distance between the back and the backrest, and the left and right skew angles of the shoulders; is the dynamic posture fluctuation feature, which represents the stability of the key points on the back within a set time window; and are the fully connected layer parameters of their respective branches, which perform linear mapping and non-linear extraction on different types of features respectively; are the non-linear Tanh activation functions of each branch; is the intermediate feature representation of the outputs of the two branches; are the final fusion weights and biases, which act on the overall representation after the branches are concatenated; is the Sigmoid function.

[0087] Step 5.4. Calculate the overall driver confidence: To integrate the comprehensive effects of grip force characteristics, micro-operation frequency characteristics, and back posture characteristics on the driver's passing confidence, the following non-linear enhanced scoring function is constructed to calculate the driver's passing confidence score (DCCI): ; Where: , , are the normalized sub-scores of grip force characteristics, micro-operation behavior characteristics, and back posture characteristics, respectively; , , are the importance weights of the corresponding modules, satisfying the constraint .

[0088] This function has a non-linear smooth growth characteristic in exponential form. When the driver performs well in multiple dimensions, the passing confidence will show an obvious increasing trend; while when any key dimension is seriously insufficient, the overall score will be significantly suppressed, effectively reflecting the sensitivity of the driver's passing ability to the combined action of multiple ability factors.

[0089] Step 6. Integrate the driver's state to correct the passing risk: To effectively integrate the driver's state into the passing risk assessment, let the original passing risk based on structural complexity and vehicle performance be , introduce the driver's passing ability score and the passing confidence score , and construct the corrected risk model as follows: ; Where, and are the risk sensitivity coefficients of ability deviation and confidence deviation, respectively; represents the degree of relative deficiency of the driver's ability; represents the degree of lack of confidence of the driver in the passing task.

[0090] To prevent the risk value from abnormally inflating due to too low driver state score or too large parameter setting, a maximum upper limit is set for the corrected risk value; to ensure the stability of the evaluation result and the controllability of the engineering system, the maximum boundary constraint of the risk value is set as follows: ; This corrected model has both interpretability and adjustability, and can dynamically reflect the impact of driver state changes on passing risk while ensuring the accuracy of the basic risk assessment.

[0091] In this embodiment, the original passability risk value based on structural complexity and vehicle performance , can directly call the existing traffic capacity evaluation model for calculation; such models are usually based on the adaptation relationship between vehicle chassis parameters and scene structure characteristics, and judge whether the vehicle has the passing ability under standard driving conditions by setting passing thresholds, constructing passable regions, or reconstructing physical scenes and replaying passing processes in a simulation platform. Therefore, without reconstructing the underlying physical model, the present invention directly uses the risk calculation framework formed by the above existing research to evaluate the current vehicle and structural scene, and obtains the basic passability risk value , which is used as the basis for subsequent personalized risk correction by integrating driver factors.

[0092] In this embodiment, after obtaining the corrected passability risk value , hierarchical management and auxiliary automatic control strategy triggering are implemented according to this risk value to improve the safety guarantee ability in the actual passing process.

[0093] Specifically, the risk levels are divided as follows: L0 (completely safe): ; L1 (low risk): ; L2 (medium risk): ; L3 (high risk): ; L4 (no passage): .

[0094] According to different risk levels, the control strategies are triggered as follows: (1) If the risk level is L1 (low risk), the system reminds the driver through voice prompt: "The front structure is complex, please drive carefully", and displays a blue low-risk indicator on the instrument interface, does not intervene in vehicle control, maintains the current throttle response and steering settings, and only continuously monitors and prompts for risks, without actively intervening in driving operations.

[0095] (2) If the risk level is L2 (medium risk), the system gives a voice reminder: "The passing risk ahead is relatively high, please operate carefully", and at the same time, a yellow medium-risk icon is displayed on the instrument interface, and a suggestion window pops up; at the same time, the upper limit of the throttle opening is compressed to 80% of the current value, and the fine steering assist mode is enabled to increase the steering response sensitivity by about 15% to assist the driver in smoothly controlling in complex areas.

[0096] (3) If the risk level is L3 (high risk), the system will give a voice prompt: "The traffic environment ahead is complex. It is recommended to slow down and observe." The instrument interface will display an orange high-risk warning and flash to prompt. At the same time, the upper limit of the throttle opening will be actively reduced to 70% of the current value, and the speed limit function will be enabled to limit the maximum speed to 20 km / h. The high-sensitivity braking warning mechanism will be activated to monitor and respond in advance to possible obstacles or sudden slope changes, prompting the driver to enter the manual low-speed and cautious operation state.

[0097] (4) If the risk level is L4 (no passage), the system will continuously give a voice prompt: "The traffic risk is too high. Do not continue to move forward." The instrument interface will display a red no-passage sign and keep flashing. At the same time, the throttle opening will be forcibly limited to 50% of the current value, the output of the sub-hydraulic braking system will be increased to 1.5 times the current value, and the speed limit control will be enabled (the maximum speed is 10 km / h, only for getting out of trouble). The emergency stop mechanism will be triggered when necessary, and the recommended route will be re-planned to prevent the driver from continuing to pass through the high-risk area.

[0098] During the passage, if the system continuously detects a change in the risk level, the control strategy will only be updated when the risk level rises to a higher level, and the control instructions of the same level will only be triggered once to avoid repeated intervention and frequent control.

[0099] Embodiment 2:

[0100] As Figure 2 shown, the present invention also provides a vehicle passability risk assessment system that integrates driver ability and confidence. The system includes a driving behavior and environmental data acquisition module, a scene matching module, a passability estimation module, a passability scoring module, a driving behavior feature analysis module, a driving confidence scoring module, a passability risk calculation module, and a database; Among them, the driving behavior and environmental data acquisition module is used to collect in real time the behavior data of the driver such as grip force, micro-operations, and postures, the static characteristics of the driver and historical passage data, and the structural feature information of the current passage scene; The scene matching module calculates the similarity based on the historical passage sample set and the structural features of the current passage scene, and screens the Top-K similar samples; The passability estimation module is used to normalize the screened similarity data to obtain a matching weight vector; based on the passability vector in the screened historical passage sample set and the calculated matching weight vector, weighted fusion is performed to obtain the passability estimation result under the current passage scene; The passability scoring module is used to calculate the passability score of the driver in the current passage scene; The driving behavior feature analysis module is used to filter and perform time-frequency analysis on the original behavior signals of grip strength, micro-operations, and postures, and construct four-dimensional grip strength feature vectors, micro-operation behavior feature vectors, and back posture feature vectors for both hands. The passing confidence score module includes a grip strength feature score module, a micro-operation behavior feature score module, a back posture feature score module, and a total confidence score module. The passing risk calculation module is used to obtain the original passing risk based on structural complexity and vehicle performance, and use the passing ability score and passing confidence score to correct the original passing risk based on structural complexity and vehicle performance, and output the passing risk considering the individual state. The database stores data such as the driver's historical passing sample set and personal characteristics. The historical passing sample set stores the structural feature vectors and passing ability vectors of historical passing scenarios.

[0101] Furthermore, in this embodiment, the system further includes a risk assessment and strategy decision module. The risk assessment and strategy decision module is used to implement the L0-L4 hierarchical control strategy for the corrected passing risk value according to the risk level, and convey the risk level and control strategy to the driver and the vehicle automatic control execution unit through the vehicle-mounted system for voice reminder, instrument prompt, and automatic control execution, so as to ensure the synchronization and effectiveness of the system response.

[0102] The present invention also provides an electronic device, including: one or more processors and a memory; wherein, the memory is used to store one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned vehicle passing risk assessment method that integrates driver ability and confidence.

[0103] The present invention also provides a computer-readable medium, on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned vehicle passing risk assessment method that integrates driver ability and confidence is implemented.

[0104] Those skilled in the art can understand that all or part of the functions of the above-mentioned methods / modules can be implemented in a hardware manner or in a computer program manner. When all or part of the functions in the above-mentioned embodiments are implemented in a computer program manner, the program can be stored in a computer-readable storage medium. The storage medium can include: read-only memory, random access memory, magnetic disk, optical disk, hard disk, etc. The above functions are implemented by a computer executing the program. For example, when the program is stored in the memory of the device and the program in the memory is executed by the processor, all or part of the above functions can be implemented.

[0105] In addition, when all or part of the functions in the above embodiments are implemented in the form of a computer program, the program can also be stored in a storage medium such as a server, another computer, a magnetic disk, an optical disk, a flash drive or a mobile hard disk, downloaded or copied and saved to the memory of a local device, or the system of the local device can be updated. When the program in the memory is executed by a processor, all or part of the functions in the above embodiments can be realized.

[0106] The above uses specific examples to illustrate the present invention, which is only used to help understand the present invention and is not intended to limit the present invention. For those skilled in the technical field to which the present invention pertains, several simple deductions, deformations or substitutions can also be made based on the idea of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for determining the vehicle passability risk by integrating driver ability and confidence, characterized in that, Including the following steps: Step 1. Collect driving behavior and environmental data; Step 2. Evaluate the traffic capacity of the current driving scenario of the driver; Dynamically construct the structural features of the current traffic scenario according to the collected environmental data, and match the structural features of the current traffic scenario with those of the historical traffic scenarios in the historical traffic sample set, and filter out the traffic scenarios with a similarity greater than the set threshold; then normalize the filtered similarity data to obtain the matching weight vector required for weighted fusion; finally, perform weighted fusion according to the traffic capacity vector in the filtered historical traffic sample set and the calculated matching weight vector to obtain the traffic capacity estimation result under the current traffic scenario; Step 3. Calculate the traffic capacity score of the driver's current traffic scenario: Construct a basic capacity scoring function based on the driver's static attributes, and combine the capacity estimation result and structural features under the current traffic scenario, and use a structure-aware logistic regression model to predict the traffic success probability. Finally, the traffic capacity scoring function fuses the two parts of the results and outputs the traffic capacity score of the driver under the current traffic scenario; Step 4. According to the collected driving behavior data, extract driving behavior characteristic indicators, and the driving behavior characteristic indicators include four-dimensional grip force characteristic vectors of the left and right hands, micro-operation behavior characteristic vectors, and back posture characteristic vectors; Step 5. Calculate the traffic confidence score of the driver's current traffic scenario based on the driving behavior characteristic indicators extracted in Step 4; Step 6. Fuse the driver state to correct the traffic risk: Use the traffic capacity score in Step 3 and the traffic confidence score in Step 5 to correct the original traffic risk based on the structural complexity and vehicle performance to obtain the final traffic risk.

2. The method for determining the vehicle passability risk by integrating driver ability and confidence according to claim 1, characterized in that The historical passing sample set in step 2 is ; where is the passing capacity vector of the i-th passing scenario in history; is the structural feature vector of the i-th passing scenario; is the passing result label, success = 1, failure = 0; , is the longitudinal operation smoothness; is the lateral control stability; is the lane change coherence; is the space utilization ability; is the ramp response ability; , is the effective passing width of the road; is the road curvature; is the degree of sight obstruction; is the road surface adhesion coefficient; is the road longitudinal slope angle, and N represents that the driver has experienced N representative passing scenarios; In Step 2, the Euclidean similarity function with a smoothing term is used to calculate the similarity.

3. The vehicle passability risk determination method that integrates driver ability and confidence according to claim 1, wherein The expression of the basic capacity scoring function in Step 3 is: ； Among them, , indicating the best for the neutral driving style; is the Sigmoid function; is the learnable parameter or experience setting; is the driving years, is the driving style label, and the output , indicating the general passing ability score of the driver; The structure-aware logistic regression model includes a structure-aware weight generation model and a logistic regression model. The structure-aware weight generation model takes the structural features of the traffic scenario as input and automatically generates the ability score weights in the current structural environment . The expression of the scoring function of the logistic regression model is as follows: ； Among them, is the Sigmoid function, which is used to compress the output to interval to form a probability score. The superscript T represents transpose, and b represents the bias parameter in the model; is the estimated result of the ability vector of the current passing scenario, and the output represents the predicted passing success probability of the driver under the current ability and structural scenario conditions; The expression of the traffic capacity scoring function is: ； Among them, is a weight parameter; represents the passing ability score of the driver in the current structural scenario.

4. The method for determining the vehicle passability risk by integrating driver ability and confidence according to claim 1, characterized in that, The extraction method of the four-dimensional grip force feature vectors of the left and right hands in step 4 is as follows: The grip forces of the driver's left and right hands are collected in real time by flexible resistive grip sensors installed at the 3 o'clock and 9 o'clock positions of the steering wheel. Then, a two-stage filtering strategy is used to preprocess the original signal. Based on the preprocessed data, the intensity deviation, fluctuation stability, normalized mutation index, and change sharpness are extracted to construct the four-dimensional grip force feature vectors of the left and right hands. and , and the expression is: ； ； Among them, represents the four-dimensional grip force feature vector of the left hand, represents the four-dimensional grip force feature vector of the right hand, and represent the strength deviation degrees of the left hand and the right hand respectively, and represent the fluctuation stabilities of the left hand and the right hand respectively, and represent the normalized mutation indices of the left hand and the right hand respectively, and represent the change sharpness of the left hand and the right hand respectively, and the superscript T represents transpose; The extraction method of the micro-operation behavior feature vector in step 4 is as follows: for the original throttle pedal pressure P a (t), the brake pedal pressure P b (t) and the steering wheel angle signals are processed by sliding window mean filtering. Then, the frequency domain features are extracted from the filtered signals using the short-time Fourier transform to obtain the power spectral density of the signals, and the ratio of the main frequency energy , the peak frequency , and the spectral entropy value are calculated. If the determination conditions are satisfied, and < 1.2, it is determined that the current time window is a micro-operation behavior segment. Then, the signals in the micro-operation behavior segment are extracted, and a damped oscillation function is used for non-linear fitting to extract the physical parameters after fitting. Based on the obtained physical parameters, the feature vector of the micro-operation behavior is constructed. The expression is as follows: ； Among them, is the initial amplitude, is the damping ratio, is the oscillation frequency, is the amplitude ratio; The method for extracting the back posture feature vector in step 4 is as follows: collect the three-dimensional coordinate information of the key points of the driver's left shoulder, right shoulder, and the midpoint of the back spine, and compare them with the three-dimensional coordinate information of the reference points corresponding to the left shoulder of the seat, the reference points corresponding to the right shoulder of the seat, and the center point of the seat backrest respectively, and calculate the Euclidean distance to obtain the backrest posture distance index; then calculate the driver correction factor according to the driver's height and body type index, and standardize the backrest posture distance index by using the driver correction factor to obtain the back distance ; calculate the shoulder angle according to the key points of the driver's left shoulder and right shoulder ; according to the frame back key point position and the average coordinates within the time window, calculate the back posture stability , and construct a standardized back posture feature vector , and the expression is: ； Among them, the superscript T represents transpose.

5. The vehicle passability risk determination method that integrates driver ability and confidence according to claim 1, characterized in that The passing confidence score of the driver's current passing scenario in step 5 The expression is as follows: ； Wherein: , , are the normalized sub-scores of the grip force feature, the micro-operation behavior feature, and the back posture feature respectively; , , are the corresponding importance weights, satisfying the constraint .

6. The vehicle passability risk determination method that integrates driver ability and confidence according to claim 1, characterized in that Final passability risk in Step 6 The expression is: ； Among them, and are the risk sensitivity coefficients of ability deviation and confidence deviation respectively; represents the degree of relative lack of driver ability; represents the degree of lack of confidence of the driver in the passing task, is the original passability risk based on structural complexity and vehicle performance; DCCI represents the passing confidence score.

7. The method for determining the vehicle passability risk by integrating driver ability and confidence according to claim 4, characterized in that Strength deviation degree The expression is as follows: ； wherein, is the grip strength value at the i-th sampling point within the time window, and are the mean and standard deviation of the baseline period respectively, and n represents the sample size; Fluctuation stability The expression is as follows: ； Normalized Mutation Index The expression is as follows: ； in, The grip strength in the baseline period was extremely poor; is the maximum grip strength of the current window; is the minimum grip strength of the current window; Change sharpness The expression is: ； Among them, and represent the grip strength values of two adjacent sampling points; Main frequency energy ratio The expression is as follows: ； Among them, is the power spectral density at the signal frequency; Peak frequency The expression is as follows: ； Spectrum entropy value The expression is as follows: ； ； Among them, represents the power spectral density at the frequency point, represents the probability distribution of the power spectral density at the frequency point normalized, where N is the window size; The expression of the damped oscillation function in Step 4 is: ； Among them, is the initial amplitude, is the damping ratio, is the natural frequency of the system, is the actual oscillation frequency after damping; is the initial phase angle, set to a constant or zero; is the time variable after the start of the micro-operation, represents the signal of the extracted micro-operation behavior segment; Oscillation frequency ; Amplitude ratio is the ratio of two peaks before and after.

8. The method for determining the vehicle passability risk by integrating driver ability and confidence according to claim 5, characterized in that, The expression of the normalized sub-score of the grip force characteristic is: ； Among them, is the input vector of the left or right hand grip force feature; is a symmetric weight matrix that models the cross-action between features; is the first-order linear weight vector; is the scalar bias term; is the Sigmoid activation function, represents the transposed vector of, represents the transposed vector of; The expression of the normalized sub-score of the micro-operation behavior characteristic is: ； Among them, is the original micro-operation behavior feature vector, and are the hidden feature vector and the gating vector respectively; and are the hidden layer parameters, is the weight matrix of the fully connected layer, is the corresponding bias term; the gating unit is composed of the weight matrix and the bias term of the linear fully connected layer, and the generated gating vector is used for the per-channel control feature fusion strategy; , are the output linear combination weights; is the non-linear Tanh activation function; is the Sigmoid function; ∘ represents element-wise multiplication, represents the fused micro-operation behavior feature vector, represents the micro-operation behavior feature score; The expression of the normalized sub-score of the back posture characteristic is: ； Among them, is the spatial attitude feature; is the dynamic attitude fluctuation feature; and are the fully connected layer parameters of their respective branches; is the non-linear Tanh activation function of each branch; is the intermediate feature representation of the outputs of the two branches; is the final fusion weight and bias, acting on the overall representation after branch concatenation; is the Sigmoid function.

9. The method for determining the vehicle passability risk by integrating driver ability and confidence according to claim 6, wherein, Set a maximum upper limit for the corrected passability risk value in Step 6 , when the calculated passability risk value is greater than the maximum upper limit , the output passability risk value is the maximum upper limit .

10. A vehicle passability risk determination system that integrates driver ability and confidence, characterized in that, The system is used to implement the vehicle traffic risk determination method for fusing the driver's ability and confidence described in any one of claims 1 to 9. The system includes a driving behavior and environmental data collection module, a scenario matching module, a traffic capacity estimation module, a traffic capacity scoring module, a driving behavior characteristic analysis module, a traffic confidence scoring module, a traffic risk calculation module, and a database; Among them, the driving behavior and environmental data collection module is used to collect the driver's behavior data, driver's static characteristics and historical traffic data, and the structural feature information of the current traffic scenario in real time. The driver's behavior data includes grip force, micro-operation, and posture; The scenario matching module calculates the similarity based on the historical traffic sample set and the structural features of the current traffic scenario, and filters out the Top-K similar samples; The traffic capacity estimation module is used to normalize the selected similarity data to obtain a matching weight vector; perform weighted fusion based on the traffic capacity vector in the selected historical traffic sample set and the calculated matching weight vector to obtain the traffic capacity estimation result in the current traffic scenario; The traffic capacity scoring module is used to calculate the traffic capacity score of the driver in the current traffic scenario; The driving behavior feature analysis module is used to filter and perform time-frequency analysis on the original behavior signals of grip strength, micro-operations, and postures, and construct four-dimensional grip strength feature vectors, micro-operation behavior feature vectors, and back posture feature vectors for both hands; The traffic confidence scoring module includes a grip strength feature scoring module, a micro-operation behavior feature scoring module, a back posture feature scoring module, and a total confidence scoring module; The passability risk calculation module is used to obtain the original passability risk based on structural complexity and vehicle performance, and use the traffic capacity score and traffic confidence score to correct the original passability risk based on structural complexity and vehicle performance, and output the passability risk considering the individual state; The database stores the driver's historical traffic sample set and personal feature data, and the historical traffic sample set stores the structural feature vector and traffic capacity vector of the historical traffic scenario.

Images