A Method and System for Driving Behavior Safety Assessment Based on Particle Sphere Computation

Abstract

This invention discloses a driving behavior safety assessment method and system based on particle sphere computation, comprising: acquiring a compliant driving sample set; constructing a scenario benchmark space and establishing a set of scenario and behavior benchmark pairs composed of benchmark scenario particles and their corresponding benchmark safety behavior particles; collecting the environmental condition feature vector and physical state parameter vector of the target vehicle at the current moment, and generating a dynamic benchmark behavior particle matching the current condition based on the environmental condition feature vector and the scenario benchmark space; extracting the target driver's behavior feature vector sequence in real time within the same sliding time window as the current moment to construct an individual behavior particle; performing physical constraint compensation on the boundary radius of the individual behavior particle to obtain a corrected individual behavior particle; calculating the topological deviation measure between the corrected individual behavior particle and the dynamic benchmark behavior particle in the behavior feature space, identifying the dominant physical dimension, and generating a safety score and driving behavior correction suggestions.

Description

Technical Field

[0001] This invention relates to the fields of automotive safe driving assessment, driving behavior analysis, and intelligent transportation data processing, specifically to a driving behavior safety assessment method and system based on particle sphere computing. Background Technology

[0002] Existing driving performance evaluation methods typically assign fixed thresholds to single physical quantities such as acceleration, braking deceleration, and steering angular velocity, and then apply points accordingly. While simple to implement, these methods suffer from two significant drawbacks: First, they fail to consider driving behavior within specific environmental conditions, neglecting contextual factors such as road curvature, traffic density, road surface adhesion coefficient, visibility, and relative distance. Second, they fail to account for differences in physical states such as vehicle load, brake fade, and tire wear, leading to the same driving action receiving the same score under different vehicle conditions, which can easily result in distorted evaluations. Especially under conditions of fully loaded vehicles, vehicles with brake fade, or low-adhesion road surfaces, traditional methods struggle to simultaneously meet the requirements of fairness and interpretability.

[0003] Therefore, existing driving scoring methods suffer from problems such as severe scenario discretization, insufficient context awareness, neglect of vehicle physical limits, and lack of interpretability of scoring results. Summary of the Invention

[0004] The technical problem this invention aims to solve is that existing driving score methods suffer from severe scenario discretization, insufficient context awareness, neglect of vehicle physical limits, and a lack of interpretability in the score results. The purpose of this invention is to provide a driving behavior safety assessment method and system based on particle sphere computation. Under vehicle dynamics constraints and environmental condition constraints, this invention utilizes particle sphere computation theory to achieve quantitative assessment and safety scoring of driving styles. This invention can achieve context-aware assessment of driving styles under different road, traffic, weather, and vehicle physical conditions, significantly improving the fairness, continuity, and interpretability of the score results.

[0005] This invention is achieved through the following technical solution:

[0006] In a first aspect, the present invention provides a driving behavior safety assessment method based on particle-sphere calculation, the method comprising:

[0007] Obtain a compliant driving sample set; based on the compliant driving sample set, perform time synchronization, outlier removal, and sliding time window slicing on each sample in the compliant driving sample set, and extract environmental condition feature vectors, physical state parameter vectors, and behavioral feature vectors;

[0008] Manifold mapping and particle clustering are performed on the environmental condition feature vectors to construct a scene benchmark space; and a set of scene and behavior benchmark pairs is established, consisting of benchmark scene particles and their corresponding benchmark safety behavior particles.

[0009] Collect the environmental condition feature vector and physical state parameter vector of the target vehicle at the current moment, and generate a dynamic benchmark behavior particle that matches the current condition based on the environmental condition feature vector and the scene reference space.

[0010] Within the same sliding time window as the current moment, the behavioral feature vector sequence of the target driver is extracted in real time to construct an individual behavior particle sphere; and the boundary radius of the individual behavior particle sphere is physically constrained and compensated based on the vehicle load factor and braking performance factor to obtain the corrected individual behavior particle sphere.

[0011] Calculate the topological deviation measure between the corrected individual behavior particle and the dynamic baseline behavior particle in the behavior feature space; perform partial derivative analysis based on the topological deviation measure to obtain the feature contribution of each behavior feature dimension and identify the dominant physical dimension.

[0012] Based on topological deviation measurement, a safety score is generated; and based on the mapping relationship between the dominant physical dimension and preset features and actions, driving behavior correction suggestions are generated.

[0013] This invention uses "scene and behavior benchmark pairs" as the core data structure, with benchmark scene granularity. Used to answer "What environmental conditions is the vehicle currently in?"; benchmark safety behavior particles Used to answer the question, "Under these environmental conditions, what driving method is compliant and safe?"; Dynamic benchmark behavior particles. Used to answer the question, "Under the current instantaneous operating conditions, what is the safe driving benchmark obtained by continuous interpolation?"; Individual behavior particles Used to answer the question "How is the driver's actual behavior distributed within the current time window?"; Correcting individual behavior granularity. This invention addresses the question of "what physical tolerance should be applied to driver behavior in comparisons, considering vehicle load and braking performance." Thus, it achieves a complete closed loop from "current scene recognition" to "safety behavior benchmark generation" and then to "interpretation of individual behavioral deviations."

[0014] Furthermore, each sample in the compliant driving sample set Represented as:

[0015] ;

[0016] in, The environmental condition feature vector includes at least one of the following: vehicle speed, road curvature, road slope, road surface adhesion coefficient, distance to the vehicle in front, relative speed, traffic density, visibility, road speed limit, and number of lanes.

[0017] The physical state parameter vector includes at least one of the following: total vehicle mass, load factor, brake disc temperature, braking efficiency coefficient, brake wear, tire wear rate, and tire pressure.

[0018] The behavioral feature vector includes at least one of the following: longitudinal impact intensity, throttle opening rate of change, peak braking deceleration, braking intervention collision time, steering angular velocity, peak lateral acceleration, and minimum following distance.

[0019] A compliance marker used to characterize the sample. Does it meet the constraints of no accidents, no violations, and vehicle stability?

[0020] Furthermore, manifold mapping and particle-sphere clustering are performed on the environmental condition feature vectors to construct a scene baseline space, including:

[0021] Perform a manifold mapping on the environmental condition feature vectors to obtain a low-dimensional embedding space;

[0022] In the low-dimensional embedding space, recursive particle clustering is used to split particles that meet at least one of the following conditions: the particle radius is greater than a first preset threshold, the sample density within the particle is lower than a second preset threshold, and the covariance trace within the particle is greater than a third preset threshold; until leaf particles are obtained as the reference scene particles.

[0023] Furthermore, the baseline scene sphere Represented as:

[0024] ;

[0025] in, For the first The center of each baseline scene particle in the original environmental condition feature space. Let covariance matrix be the variance matrix. The radius of the sphere, This is the set of sample indices that fall into the reference scene's particles;

[0026] Benchmark safety behavior particle It is composed of reference scene particles The behavioral feature vectors of all compliant samples are constructed as follows: ;

[0027] in: ;

[0028] ;

[0029] in, The baseline safety behavior center vector; For the first The boundary radius of each behavioral feature dimension; This is a behavioral feature vector; For the first The boundary radius of each behavioral feature dimension. For this dimension in the baseline scene sphere within the standard deviation, To match the preset information level The corresponding confidence coefficient.

[0030] Furthermore, based on the environmental condition feature vector and the scene reference space, a dynamic reference behavior particle matching the current operating condition is generated, including:

[0031] Calculate the Mahalanobis distance between the environmental condition feature vector and each reference scene particle in the scene reference space, and select the nearest reference scene particle from them;

[0032] Based on the inverse distance weighted interpolation method, the reference safety behavior particles corresponding to the neighboring reference scene particles are weighted and mapped to generate dynamic reference behavior particles that match the current working conditions.

[0033] Furthermore, based on the vehicle load factor and braking performance factor, physical constraint compensation is applied to the boundary radius of the individual behavior particle to obtain the corrected individual behavior particle, including:

[0034] Calculate the vehicle load factor based on the current total vehicle mass and the reference total mass;

[0035] The braking performance factor is calculated based on the current brake disc temperature, the reference brake disc temperature, the amount of brake wear, and the minimum braking performance threshold.

[0036] Calculate the physical correction coefficient based on the vehicle load factor and braking performance factor;

[0037] Based on the physical correction coefficient, boundary radius compensation is performed on the behavioral feature dimensions that are highly related to physical constraints (related to safe following distance and braking deceleration) to obtain the compensated boundary radius;

[0038] Based on the compensated boundary radius and the average behavioral level of the driver in each behavioral characteristic dimension within the current time window, the corrected individual behavior particle is obtained.

[0039] Furthermore, topological deviation measure The topological deviation measure is obtained by jointly calculating the directional risk center deviation, radius expansion deviation, and non-overlapping gap deviation between the individual behavior particle and the dynamic baseline behavior particle. The expression is:

[0040] ;

[0041] ;

[0042] ;

[0043] ;

[0044] in, For the first Directional risk center deviation in each behavioral characteristic dimension; For the first Radius expansion bias of each behavioral feature dimension; For the first Non-overlapping gap deviation of each behavioral feature dimension; ; For the first The safety direction indication value for each behavioral feature dimension is taken as follows: the larger the value of this dimension, the more dangerous it is. When the smaller the value of this dimension, the more dangerous it is, take... ; For the first Risk weights for each behavioral characteristic dimension; , , These are the weighting coefficients for directional risk center deviation, radius expansion deviation, and non-overlapping gap deviation, respectively. To correct individual behavioral particles in the first... The center of dimensional; For the dynamic reference behavior of the particle in the first The center of dimensional; To prevent positive numbers with a denominator of zero; To correct individual behavioral particles in the first... The radius of the dimension; For the dynamic reference behavior of the particle in the first The radius of the dimension.

[0045] Furthermore, partial derivative analysis is performed based on the topological deviation measure to obtain the feature contribution of each behavioral feature dimension, and the dominant physical dimension is identified, including:

[0046] Partial derivative analysis is performed based on the topological deviation measure to obtain the feature contribution of each behavioral feature dimension; the expression for the feature contribution is: ,in, For the first Feature contribution of each behavioral feature dimension For topological deviation measure, To correct individual behavioral particles in the first... The center of dimensional, To correct individual behavioral particles in the first... The radius of the dimension, Contribution coefficient to radius deviation;

[0047] The feature contribution is normalized to obtain the normalized feature contribution; the expression for the normalized feature contribution is: ,in, The normalized feature contribution. To prevent positive numbers with a denominator of zero; For the first Feature contribution of each behavioral feature dimension;

[0048] When normalized feature contribution If the contribution is greater than the preset contribution threshold or is among the top few maximum values, then the first value is determined to be... The behavioral characteristic dimension is the dominant physical dimension.

[0049] Furthermore, based on the mapping relationship between the dominant physical dimension and preset features and actions, driving behavior correction suggestions are generated, including:

[0050] When the dominant physical dimension corresponds to the rate of change of throttle opening or longitudinal impact, a throttle smoothness correction suggestion is output.

[0051] When the dominant physical dimension corresponds to the minimum following distance, braking intervention collision time, or peak braking deceleration, then the braking intervention timing and following distance correction suggestions are output.

[0052] When the dominant physical dimension corresponds to the peak lateral acceleration or steering angular velocity, the output provides steering smoothness and cornering centripetal acceleration control correction suggestions.

[0053] Secondly, the present invention provides a driving behavior safety assessment system based on particle-based computing, the system comprising:

[0054] The acquisition and extraction unit is used to acquire a compliant driving sample set; and to extract environmental condition feature vectors, physical state parameter vectors, and behavioral feature vectors based on the compliant driving sample set.

[0055] The scenario baseline space construction unit is used to perform manifold mapping and particle clustering on the environmental condition feature vectors to construct the scenario baseline space; and to establish a set of scenario and behavior baseline pairs consisting of baseline scenario particles and their corresponding baseline safety behavior particles.

[0056] The dynamic baseline behavior particle generation unit is used to collect the environmental condition feature vector and physical state parameter vector of the target vehicle at the current moment, and generate a dynamic baseline behavior particle that matches the current condition based on the environmental condition feature vector and the scene reference space.

[0057] The individual behavior particle generation unit is used to extract the target driver's behavior feature vector sequence in real time within the same sliding time window as the current time, and construct the individual behavior particle; and to perform physical constraint compensation on the boundary radius of the individual behavior particle based on the vehicle load factor and braking performance factor to obtain the corrected individual behavior particle.

[0058] The topological deviation measurement calculation unit is used to calculate the topological deviation measure between the corrected individual behavior particle and the dynamic baseline behavior particle in the behavior feature space.

[0059] The safety score generation and correction unit is used to perform partial derivative analysis based on the topological deviation measure to obtain the feature contribution of each behavioral feature dimension and identify the dominant physical dimension; based on the dominant physical dimension, it generates a safety score and driving behavior correction suggestions.

[0060] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0061] This invention relates to a driving behavior safety assessment method and system based on particle sphere computation. By constructing a correlation computation mechanism between scenario-based benchmark particles, benchmark safe behavior particles, dynamic benchmark behavior particles, and individual behavior particles, this invention achieves context-aware quantitative assessment of driving style under different road, traffic, weather, and vehicle physical conditions. Compared to traditional methods that rely solely on fixed threshold deductions, this invention significantly improves the fairness, continuity, and interpretability of the assessment results, and can provide reliable data support for driver training, risk warning, insurance pricing, and fleet management.

[0062] (1) A continuous scene reference space is constructed by manifold mapping and particle clustering to avoid discrete boundary errors caused by hard-coded scene labels;

[0063] (2) Dynamic benchmark behavior particles are generated by multi-particle weighted interpolation, so that the safety benchmark can change continuously with the current working conditions;

[0064] (3) The boundary radius of individual behavior particles is compensated by vehicle load and braking performance to improve the fairness of the evaluation results under different vehicle conditions;

[0065] (4) Improve the sensitivity of risk detection by using topological deviation measurement to jointly evaluate center offset, fluctuation increase and safety envelope non-overlap.

[0066] (5) By decoupling through partial derivatives, the dominant physical dimension is identified, and correction suggestions with physical meaning are directly output, thereby improving interpretability and operability. Attached Figure Description

[0067] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings:

[0068] Figure 1 This is a flowchart of the driving behavior safety assessment method based on particle-sphere calculation according to the present invention;

[0069] Figure 2 This is a block diagram of the driving behavior safety assessment system based on particle-sphere calculation according to the present invention. Detailed Implementation

[0070] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0071] This invention uses "scene and behavior benchmark pairs" as the core data structure, with benchmark scene granularity. Used to answer "What environmental conditions is the vehicle currently in?"; benchmark safety behavior particles Used to answer the question, "Under these environmental conditions, what driving method is compliant and safe?"; Dynamic benchmark behavior particles. Used to answer the question, "Under the current instantaneous operating conditions, what is the safe driving benchmark obtained by continuous interpolation?"; Individual behavior particles Used to answer the question "How is the driver's actual behavior distributed within the current time window?"; Correcting individual behavior granularity. This invention addresses the question of "what physical tolerance should be applied to driver behavior in comparisons, considering vehicle load and braking performance." Thus, it achieves a complete closed loop from "current scene recognition" to "safety behavior benchmark generation" and then to "interpretation of individual behavioral deviations."

[0072] Here, "particle" refers to a set of locally similar samples characterized by its center and boundary radius. In this invention, a vector radius defined by feature dimensions is preferably used to describe the range of behavioral fluctuations.

[0073] Example 1

[0074] like Figure 1 As shown, the present invention provides a driving behavior safety assessment method based on particle-sphere computation, which includes:

[0075] S1, Obtain a compliant driving sample set According to the compliant driving sample set The environmental condition feature vectors were extracted by performing time synchronization, outlier removal, and sliding time window slicing on each sample in the compliant driving sample set. Physical state parameter vector and behavioral feature vector ;

[0076] In this embodiment, the compliant driving sample set The compliant driving samples were obtained from multiple sources of raw data through unified processing. The raw data included at least:

[0077] (1) Vehicle bus data: vehicle speed, longitudinal acceleration, lateral acceleration, braking pressure, throttle opening, steering wheel angle, steering wheel angular velocity, yaw rate, wheel speed, etc.

[0078] (2) Positioning and map data: GPS / IMU pose, road curvature, road slope, road speed limit, number of lanes, road type, etc.;

[0079] (3) Environmental perception data: distance to the vehicle in front, relative speed, traffic density, number of visible targets, lane position, weather and visibility estimation, etc.;

[0080] (4) Vehicle status data: total mass, load factor, axle load distribution, brake disc temperature, brake wear, tire wear rate, tire pressure, etc.

[0081] (5) Compliance label data: accident records, violation records, active safety system trigger records, and vehicle stability constraint records.

[0082] Preferably, each original signal is resampled to the same sampling frequency according to a unified time scale, and a length of [missing information] is used. The time window is sliced. Thus, the compliant driving sample set... Each sample Represented as:

[0083] ;

[0084] Among them, 1) Environmental condition feature vector:

[0085] ;

[0086] in, For the speed of this vehicle, For road curvature, For road slope, This is an estimated value for the road surface adhesion coefficient. Distance to the vehicle in front. For relative velocity, For traffic density, For visibility, Speed ​​limits for roads, This refers to the number of lanes.

[0087] 2) For the physical state parameter vector:

[0088] ;

[0089] in, For the total mass of the vehicle. For load factor, For brake disc temperature, The braking efficiency coefficient, This refers to brake wear. For tire wear rate, This refers to tire pressure.

[0090] 3) For behavioral feature vectors:

[0091] ;

[0092] in:

[0093] Vertical impact, representing the time window. The root mean square value of the rate of change of longitudinal acceleration of the vehicle reflects the smoothness / impact of the driver's longitudinal (forward and backward) operation. To be within the time window The longitudinal acceleration of the vehicle at each sampling time (in m / s²), with positive values ​​indicating acceleration and negative values ​​indicating deceleration; The sampling time interval, The number of sampling points within the time window (window length);

[0094] The rate of change of throttle opening. For the first time window Accelerator pedal opening at each sampling time;

[0095] For peak braking deceleration, For the $th time window At each sampling moment, the longitudinal acceleration of the vehicle when it is braking;

[0096] For braking intervention collision time, Braking start time At that time, the distance between this vehicle and the vehicle in front (unit: meters). Braking start time At that time, the relative speed of this vehicle to the vehicle in front (unit: m / s);

[0097] The root mean square of the steering angular velocity. For the first time window Steering wheel angle at each sampling time;

[0098] This represents the peak lateral acceleration. For the first time window The lateral acceleration of the vehicle at each sampling time;

[0099] Minimum following distance, For the first At each sampling moment, the longitudinal speed of this vehicle, It should be a very small positive number (to prevent division by zero errors);

[0100] 4) For compliance marking

[0101] Used to characterize the sample Does it meet the constraints of no accidents, no violations, and vehicle stability?

[0102] A sample can be marked as a compliant sample when it meets all of the following conditions:

[0103] a. No accident records were recorded within the corresponding time window;

[0104] b. No traffic violation record;

[0105] c. No record of strong intervention by active safety systems;

[0106] d. Key dynamic indicators did not exceed the permissible limits for vehicle stability;

[0107] e. The local cluster where the sample is located meets the preset density threshold.

[0108] Therefore, the so-called "high confidence sample cluster" refers to a sample set with high local sample density and low internal variance under the constraints of no accidents, no illegal activities, and dynamic stability.

[0109] S2, for the environmental condition feature vector Perform manifold mapping and particle clustering to construct a scene baseline space; and establish a system based on the baseline scene particles. Its corresponding benchmark safety behavior particle The set of scenarios and behavioral benchmarks;

[0110] In this embodiment, manifold mapping and particle-sphere clustering are performed on the environmental condition feature vectors to construct a scene baseline space, including:

[0111] Perform a manifold mapping on the environmental condition feature vectors to obtain a low-dimensional embedding space;

[0112] In the low-dimensional embedding space, recursive particle clustering is used to split particles that meet at least one of the following conditions: the particle radius is greater than a first preset threshold, the sample density within the particle is lower than a second preset threshold, and the covariance trace within the particle is greater than a third preset threshold; until leaf particles are obtained as the reference scene particles.

[0113] Specifically, step S2 includes:

[0114] S21, manifold mapping

[0115] Because driving scenarios exhibit continuous changes in the original environmental condition feature space, directly using discrete labels to divide scenarios can easily lead to abrupt changes in scenario boundaries. This invention first addresses this by analyzing the environmental condition feature vectors... Perform manifold mapping and embed it into a low-dimensional manifold space, preferably using a local linear embedding method.

[0116] For the For each sample, first select the nearest neighbor set in the original environmental condition feature space. Solve for the reconstructed weights:

[0117] ;

[0118] in, Indicates the first The nearest neighbor set of each sample in the original environmental condition feature space For local reconstruction of weights; then solve for the low-dimensional embedding:

[0119] ;

[0120] in, For the first The embedding coordinates of each sample in the low-dimensional manifold space (low-dimensional representation vector).

[0121] Obtain a low-dimensional scene representation .

[0122] The principle is to preserve the local adjacency relationship between samples, so that similar scenes remain similar in low-dimensional space, thereby reducing the discontinuity problem caused by artificial discrete labels.

[0123] S22, Granulocyte Clustering

[0124] In a low-dimensional manifold space, for all Perform recursive sphere clustering. Initially, the root sphere is constructed using all samples. For any candidate sphere... ,calculate:

[0125] Pellet center:

[0126] ;

[0127] in, The center of the grain; This represents the number of samples within the granule;

[0128] spherical radius :

[0129] ;

[0130] Sample density :

[0131] ;

[0132] in, The number of samples within a granule. For the dimension of the low-dimensional manifold space, To prevent positive numbers with a denominator of zero;

[0133] Splitting is performed on a pellet when at least one of the following conditions is met:

[0134] ;

[0135] ;

[0136] .

[0137] in, The threshold for grain radius splitting. The sample density splitting threshold, The covariance trace splitting threshold;

[0138] The splitting is preferably achieved using binary clustering until all leaf granules meet the termination condition.

[0139] S23, Representation of the scene reference space

[0140] The final scene reference space Represented as:

[0141] ;

[0142] in: For the baseline scene, spheres The benchmark safety behavior particle;

[0143] ;

[0144] in, For the first The center of each baseline scene particle in the original environmental condition feature space. Let covariance matrix be the variance matrix. The radius of the sphere, This is the set of sample indices that fall into the reference scene's particles;

[0145] here, It's not a simple scene tag library, but a set of benchmark pairs composed of "scene indexes" and "safety behavior benchmarks." In other words:

[0146] Reference scene sphere The meaning of is a set of sample regions with similar environmental conditions and continuous proximity in manifold space;

[0147] Benchmark safety behavior particle The meaning of is the safety behavior envelope obtained from a large number of compliant driving samples under corresponding environmental conditions;

[0148] The relationship between the two is a one-to-one correspondence: each Uniquely associated with one .

[0149] S24, Benchmark Safety Behavior Particle

[0150] Benchmark safety behavior particle It is composed of reference scene particles The behavioral feature vectors of all compliant samples are constructed as follows: ;

[0151] For the benchmark scene sphere Extract the set of behavioral feature vectors from all compliant samples in the dataset. And calculate:

[0152] ;

[0153] ;

[0154] in, As the baseline safety behavior center vector, The baseline safety behavior boundary radius vector, For the first The boundary radius of each behavioral feature dimension. For the first The standard deviation of each behavioral feature dimension To preset the credit level The corresponding confidence coefficient.

[0155] therefore, The physical meaning is: in the scene of the sphere The safety center and permissible fluctuation range of driving behavior in various dimensions under the represented environmental conditions.

[0156] S3, Collect the environmental condition feature vector of the target vehicle at the current moment. With physical state parameter vector Based on the environmental condition feature vector Generate dynamic baseline behavior particles that match the current operating conditions within the scene baseline space. ;

[0157] In this embodiment, the target vehicle obtains the environmental condition feature vector at the current moment. Next, it is necessary to determine its matching relationship with the scene reference space. Since actual working conditions typically lie between multiple reference scene spheres, this invention does not use single nearest neighbor matching, but rather multi-sphere weighted interpolation. This includes:

[0158] First, calculate the feature vector of the current environmental conditions. Compared with the particle spheres of each benchmark scenario Mahalanobis distance :

[0159] ;

[0160] in, For the first The center of a baseline scene particle, Let its covariance matrix be denoted as Mahalanobis distance. The reason for using Mahalanobis distance is that the dimensions of environmental operating conditions are different and correlated, and Mahalanobis distance can simultaneously reflect scale differences and related structures.

[0161] Select the one with the smallest distance Each benchmark scene particle forms a neighborhood set. And calculate the inverse distance weights. :

[0162] ;

[0163] Further weighted mapping is performed on the corresponding baseline safety behavior particles:

[0164] ;

[0165] ;

[0166] This yields the dynamic baseline behavior of the sphere. :

[0167] ;

[0168] in, The set of neighboring reference scene particles at the current moment. For the first The weights of each baseline scene particle. The Mahalanobis distance, To prevent positive numbers with a denominator of zero.

[0169] The technical principle behind this step is that the same road condition does not necessarily belong to a single discrete category, but may exist in the transition zone between multiple adjacent scenarios. By generating dynamic benchmark behavior particles through inverse distance weighted interpolation, the safe driving benchmark can be continuously changed with environmental conditions, thereby avoiding abrupt score changes at scenario switching boundaries.

[0170] S4. Within the same sliding time window as the current moment, extract the target driver's behavioral feature vector sequence in real time to construct an individual behavioral particle sphere. ; and based on vehicle load factor and braking performance factor, individual behavioral particles Physical constraint compensation is applied to the boundary radius to obtain the corrected individual behavior particle sphere. ;

[0171] In this embodiment, step S4 specifically includes:

[0172] S41, Constructing Individual Behavior Particles

[0173] In the current sliding time window Inside, collect the behavioral feature vector sequence of the target driver. Constructing individual behavioral spheres:

[0174] ;

[0175] in:

[0176] ;

[0177] ;

[0178] In the formula, The number of samples within the window. This represents the average behavioral level of the driver across various behavioral characteristic dimensions within the current time window. This indicates the degree of fluctuation in the driver's corresponding behavioral characteristic dimension; For individual behavioral particles in the first The center of each behavioral feature dimension For individual behavioral particles in the first Boundary radius on each behavioral feature dimension; For individual behavioral particles in the first Boundary radius on each behavioral feature dimension; For the first The behavioral feature vector at time n is at the... The value that a dimension can take;

[0179] Therefore, individual behavioral particles not only describe "how drivers drive on average", but also "whether drivers drive consistently".

[0180] S42, perform physical constraint compensation to obtain the corrected individual behavior particle.

[0181] Since increased vehicle load or decreased braking system efficiency alters the physical reachability boundaries of driving behavior, simply comparing the current behavior directly with a baseline behavior would lead to unfair evaluation. Therefore, this invention performs physical constraint compensation on individual behavior particles, specifically:

[0182] S421, Calculate the vehicle load factor based on the current total vehicle mass and the reference total mass. ;

[0183] ;

[0184] in, For the current total mass, For reference quality.

[0185] S422, calculate the braking performance factor based on the current brake disc temperature, reference brake disc temperature, brake wear amount, and minimum braking performance threshold. ;

[0186] ;

[0187] in, This is the current brake disc temperature. For reference brake disc temperature, This refers to brake wear. and For calibration coefficients, The minimum braking efficiency threshold;

[0188] This formula indicates that braking efficiency decreases as braking temperature increases and brake wear increases.

[0189] S423, calculate the physical correction factor based on the vehicle load factor and braking performance factor. ;

[0190] ;

[0191] The larger the value, the more conservative the following and braking behavior of the current vehicle needs.

[0192] S424, based on the physical correction coefficient, the set of behavioral feature dimensions highly correlated with physical constraints. Perform boundary radius compensation to obtain the compensated boundary radius. ;

[0193] ;

[0194] ;

[0195] in, The boundary radius; This represents the set of behavioral feature dimensions related to physical constraints. The preferred features include the minimum following distance, the braking intervention collision time, and the characteristic dimensions corresponding to the peak braking deceleration. For the first Compensation weights for each behavioral feature dimension.

[0196] S425, based on the compensated boundary radius and the driver's average behavioral level across various behavioral characteristic dimensions within the current time window, obtains the corrected individual behavioral sphere. .

[0197] ;

[0198] The technical principle behind this step is that the vehicle's physical state does not change the driver's behavior category, but it does change the "reasonable range of behavior fluctuations." Therefore, this invention, by modifying the boundary radius rather than simply shifting the center value, makes the assessment more consistent with the physical fact that "the tolerance boundaries for the same type of behavior differ under different physical states."

[0199] S5, Calculate the corrected individual behavior particle sphere With dynamic benchmark behavior of particles Topological deviation measure in behavioral feature space According to topological deviation measure Partial derivative analysis was performed to obtain the feature contribution of each behavioral feature dimension. And identify the dominant physical dimension;

[0200] In this embodiment, step S5 specifically includes:

[0201] S51, Topological Deviation Measure

[0202] This invention does not only compare the center difference, nor only compare single-point thresholds, but compares them simultaneously:

[0203] a. Whether individual behavior deviates in a dangerous direction;

[0204] b. Whether the individual's behavioral fluctuations exceed the baseline fluctuations;

[0205] c. Whether the individual behavior envelope is completely removed from the safety envelope.

[0206] Therefore, for the first Each behavioral feature dimension is defined as follows:

[0207] (1) Directional risk center deviation

[0208] ;

[0209] in: For individual behavioral particles in the first The center of dimensional; For the dynamic reference behavior of the particle in the first The center of dimensional; For safety direction indication; For the dynamic reference behavior of the particle in the first The radius of the dimension; ; For the first The safety direction indication value for each behavioral feature dimension is taken as follows: the larger the value of this dimension, the more dangerous it is. When the smaller the value of this dimension, the more dangerous it is, take... ;

[0210] When the When considering characteristics where "the larger the value, the more dangerous," such as longitudinal impact, throttle opening rate of change, peak braking deceleration, steering angular velocity, and peak lateral acceleration, take... When the first The characteristics that "the smaller the value, the more dangerous" include, for example, minimum following distance and braking intervention collision time. .

[0211] therefore, Deviations are only counted for those that deviate from the dangerous direction, without penalizing more conservative actions.

[0212] (2) Radius expansion deviation

[0213] ;

[0214] in: To correct individual behavioral particles in the first... The radius of the dimension;

[0215] This quantity reflects the current driver's position in the [missing information]. Does the volatility in a behavioral characteristic dimension exceed the baseline volatility? For example, if the mean is the same but the throttle changes are more abrupt, the steering is more shaky, and the braking is more unstable, this item will increase.

[0216] (3) Non-overlapping gap deviation

[0217] ;

[0218] in: To correct individual behavioral particles in the first... The center of dimensional; For the dynamic reference behavior of the particle in the first The center of dimensional; To prevent positive numbers with a denominator of zero;

[0219] This quantity reflects the individual behavior granules in the first... Whether the interval on the dimension completely deviates from the dynamic baseline behavior particle sphere interval. When this term is greater than zero, it indicates that there is a clear gap between the individual behavior envelope and the safety baseline envelope, which is a stronger risk signal.

[0220] (4) Feature weights

[0221] Preferably, feature weights are defined to highlight the sensitive dimensions in the contracted security envelope. :

[0222] ;

[0223] That is, the smaller the dynamic baseline radius and the stricter the tolerance of a dimension, the greater its weight.

[0224] (5) Total topological deviation measure

[0225] ;

[0226] in, This represents the total risk deviation of the driver relative to the safety baseline within the current time window. For the first Risk weights (importance) of each behavioral characteristic dimension; , , These are the weighting coefficients for directional risk center deviation, radius expansion deviation, and non-overlapping gap deviation, respectively. To prevent positive numbers with a denominator of zero.

[0227] S52, the dominant physical dimension

[0228] To obtain an explainable result regarding "which specific driving action led to the score decline," this invention... Taking the partial derivative with respect to the individual behavior particle center and the compensated radius, we define the... Feature contribution of each behavioral feature dimension for:

[0229] ;

[0230] in, For the first Feature contribution of each behavioral feature dimension For topological deviation measure, To correct individual behavioral particles in the first... The center of dimensional, To correct individual behavioral particles in the first... The radius of the dimension, Contribution coefficient to radius deviation;

[0231] Preferably, its explicit form can be written as:

[0232] ;

[0233] ;

[0234] in, This is an indicator function.

[0235] The feature contribution is then normalized to obtain the normalized feature contribution. ;

[0236] ;

[0237] in, The normalized feature contribution. To prevent positive numbers with a denominator of zero; For the first Feature contribution of each behavioral feature dimension;

[0238] When normalized feature contribution Greater than the preset contribution threshold Or located in front When the maximum value is reached, then the first value is determined. The behavioral characteristic dimension is the dominant physical dimension.

[0239] S6, generating a safety score based on topology deviation measure. Based on the dominant physical dimension and the mapping relationship between preset features and actions, driving behavior correction suggestions are generated.

[0240] In this embodiment, based on topological deviation measure Calculate the security score for the current time window. :

[0241] ;

[0242] in, This is the score decay coefficient. When... When the individual's behavior is completely consistent with the dynamic baseline behavior, the score reaches its highest level; when As the score increases, it decreases smoothly in an exponential manner.

[0243] In this embodiment, based on the mapping relationship between the dominant physical dimension and preset features and actions, driving behavior correction suggestions are generated, including:

[0244] When the dominant physical dimension corresponds to the rate of change of throttle opening or longitudinal impact, a throttle smoothness correction suggestion is output.

[0245] When the dominant physical dimension corresponds to the minimum following distance, braking intervention collision time, or peak braking deceleration, then the braking intervention timing and following distance correction suggestions are output.

[0246] When the dominant physical dimension corresponds to the peak lateral acceleration or steering angular velocity, the output provides steering smoothness and cornering centripetal acceleration control correction suggestions.

[0247] Specifically, this invention does not simply output "how many points were deducted," but rather maps the dominant physical dimension back to driving action suggestions. A preferred definition is the feature-action mapping relationship. for:

[0248] , Smoothness of throttle opening;

[0249] , , Corresponding braking intervention timing and safe following distance;

[0250] , Corresponding to steering smoothness and cornering acceleration control.

[0251] To provide more granular levels of recommendations, the [number] level is further defined. Comprehensive deviation level of each feature dimension :

[0252]

[0253] but:

[0254] when When this happens, output a level 1 prompt;

[0255] when At that time, a level two warning will be issued;

[0256] when At that time, a Level 3 strong warning will be issued.

[0257] Accordingly, the preferred generation rule is:

[0258] (1) Suggestions on throttle opening smoothness

[0259] When the dominant physical dimension is or hour:

[0260] Level 1 tip: Reduce the instantaneous rate of change of the accelerator pedal and avoid continuous sudden pressing and releasing;

[0261] Level 2 warning: Break down the acceleration action into two progressive acceleration stages to reduce longitudinal impact;

[0262] Level 3 Strong Warning: Restrict sudden acceleration during the starting, following, and overtaking phases.

[0263] (2) Recommendations on the timing of braking intervention

[0264] When the dominant physical dimension is , or hour:

[0265] like Smaller or Too small: Instructions to "brake earlier and increase following distance";

[0266] like Too high: Warning: "Avoid large deceleration braking when approaching obstacles";

[0267] If both exist Smaller and Larger than expected: This indicates that "the current risk mainly stems from the braking intervention being too late, and the deceleration process should be established earlier."

[0268] (3) Steering centripetal acceleration control recommendations

[0269] When the dominant physical dimension is or hour:

[0270] Level 1 tip: Reduce your entry speed appropriately and avoid making sharp turns while in the corner;

[0271] Level 2 warning: Move the main deceleration action forward to before entering the curve, and maintain small directional corrections during the curve;

[0272] Level 3 Strong Warning: The current cornering control has significantly deviated from the baseline. You should simultaneously reduce the entry speed and steering change rate.

[0273] Therefore, this invention directly converts abstract deviation measurements into corrective actions that can be performed by the driver, thereby achieving a closed-loop output of "scoring-cause-recommendation".

[0274] In practice, continuous operational data of a certain vehicle model fleet is selected as the original data source, with a collection period of [missing information]. After aligning the vehicle bus data, environmental perception data, map data, and vehicle status data, a length of [missing information] is generated. A sliding time window is used to construct samples. Samples that meet the criteria of no accidents, no violations, and no strong intervention from the stability control system are screened to form a compliant driving sample set. Subsequently, manifold mapping was performed on the environmental condition feature vectors, and recursive particle clustering was used in the low-dimensional manifold space to obtain multiple benchmark scene particles. Based on the behavioral characteristics of samples inside each benchmark scenario particle, a corresponding benchmark safety behavior particle is constructed. .

[0275] During the real-time scoring phase, the environmental condition feature vector of the target vehicle at the current moment is used. The Mahalanobis distances between the dynamic baseline behavior particle and multiple reference scene particle spheres are calculated, and the dynamic baseline behavior particle sphere is obtained by interpolation using an inverse distance weighting method. Simultaneously, individual behavior granules are constructed based on driving behavior data within the current sliding time window. Then, by combining the vehicle load mass and braking system status to perform boundary radius compensation, the corrected individual behavior particle sphere is obtained. Then calculate. and Topological deviation measure The dominant risk dimension is obtained by decoupling using partial derivatives, and a safety score is output. And corresponding driving correction suggestions.

[0276] In an example scenario, if the minimum following distance dimension and the peak braking deceleration dimension are detected to have the highest contribution, and the corresponding comprehensive deviation level is greater than the level 2 warning threshold, the system will output a joint suggestion of "braking in advance, increasing the following distance, and avoiding large deceleration braking when approaching obstacles".

[0277] Example 2

[0278] like Figure 2 As shown, the difference between this embodiment and Embodiment 1 is that this embodiment provides a driving behavior safety assessment system based on particle sphere computation, which corresponds one-to-one with the driving behavior safety assessment method based on particle sphere computation in Embodiment 1; the system includes:

[0279] The acquisition and extraction unit is used to acquire a compliant driving sample set; and to extract environmental condition feature vectors, physical state parameter vectors, and behavioral feature vectors based on the compliant driving sample set.

[0280] The scenario baseline space construction unit is used to perform manifold mapping and particle clustering on the environmental condition feature vectors to construct the scenario baseline space; and to establish a set of scenario and behavior baseline pairs consisting of baseline scenario particles and their corresponding baseline safety behavior particles.

[0281] The dynamic baseline behavior particle generation unit is used to collect the environmental condition feature vector and physical state parameter vector of the target vehicle at the current moment, and generate a dynamic baseline behavior particle that matches the current condition based on the environmental condition feature vector and the scene reference space.

[0282] The individual behavior particle generation unit is used to extract the target driver's behavior feature vector sequence in real time within the same sliding time window as the current time, and construct the individual behavior particle; and to perform physical constraint compensation on the boundary radius of the individual behavior particle based on the vehicle load factor and braking performance factor to obtain the corrected individual behavior particle.

[0283] The topological deviation measurement calculation unit is used to calculate the topological deviation measure between the corrected individual behavior particle and the dynamic baseline behavior particle in the behavior feature space.

[0284] The safety score generation and correction unit is used to perform partial derivative analysis based on the topological deviation measure to obtain the feature contribution of each behavioral feature dimension and identify the dominant physical dimension; based on the dominant physical dimension, it generates a safety score and driving behavior correction suggestions.

[0285] The execution process of each unit can be carried out according to the steps of the driving behavior safety assessment method based on particle calculation in Example 1, and will not be described in detail in this example.

[0286] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0287] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0288] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0289] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0290] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A driving behavior safety assessment method based on particle-sphere computation, characterized in that, include: Obtain a compliant driving sample set; based on the compliant driving sample set, extract environmental condition feature vectors, physical state parameter vectors, and behavioral feature vectors; Manifold mapping and particle clustering are performed on the environmental condition feature vectors to construct a scene reference space; and a set of scene and behavior reference pairs is established, consisting of reference scene particles and their corresponding reference safety behavior particles. Collect the environmental condition feature vector and physical state parameter vector of the target vehicle at the current moment, and generate a dynamic reference behavior particle that matches the current condition based on the environmental condition feature vector and the scene reference space. Within the same sliding time window as the current moment, the behavioral feature vector sequence of the target driver is extracted in real time to construct an individual behavior particle sphere; and the boundary radius of the individual behavior particle sphere is physically constrained and compensated based on the vehicle load factor and braking performance factor to obtain a corrected individual behavior particle sphere. Calculate the topological deviation measure between the corrected individual behavior particle and the dynamic baseline behavior particle in the behavior feature space; perform partial derivative analysis based on the topological deviation measure to obtain the feature contribution of each behavior feature dimension and identify the dominant physical dimension. Based on the topological deviation measure, a safety score is generated; and based on the mapping relationship between the dominant physical dimension and preset features and actions, driving behavior correction suggestions are generated.

2. The driving behavior safety assessment method based on particle-sphere calculation according to claim 1, characterized in that, Each sample in the compliant driving sample set Represented as: ； in, The environmental condition feature vector includes at least one of the following: vehicle speed, road curvature, road slope, road surface adhesion coefficient, distance to the vehicle in front, relative speed, traffic density, visibility, road speed limit, and number of lanes. The physical state parameter vector includes at least one of the following: total vehicle mass, load factor, brake disc temperature, braking efficiency coefficient, brake wear, tire wear rate, and tire pressure. The behavioral feature vector includes at least one of the following: longitudinal impact intensity, throttle opening rate of change, peak braking deceleration, braking intervention collision time, steering angular velocity, peak lateral acceleration, and minimum following distance. A compliance marker used to characterize the sample. Does it meet the constraints of no accidents, no violations, and vehicle stability? 3. The driving behavior safety assessment method based on particle-sphere calculation according to claim 1, characterized in that, The environmental condition feature vectors are subjected to manifold mapping and particle-sphere clustering to construct a scene reference space, including: Perform manifold mapping on the environmental condition feature vectors to obtain a low-dimensional embedding space; In the low-dimensional embedding space, recursive particle clustering is used to split particles that meet at least one of the following conditions: the particle radius is greater than a first preset threshold, the sample density within the particle is lower than a second preset threshold, and the covariance trace within the particle is greater than a third preset threshold; until leaf particles are obtained as the reference scene particles.

4. The driving behavior safety assessment method based on particle-sphere calculation according to claim 3, characterized in that, The reference scene sphere Represented as: ； in, For the first The center of each baseline scene particle in the original environmental condition feature space. Let covariance matrix be the variance matrix. The radius of the sphere, This is the set of sample indices that fall into the reference scene's particles; The benchmark safety behavior particle It is composed of reference scene particles The behavioral feature vectors of all compliant samples are constructed as follows: ; in: ; ； in, The baseline safety behavior center vector; For the first The boundary radius of each behavioral feature dimension; This is a behavioral feature vector; For the first The boundary radius of each behavioral feature dimension. For this dimension in the baseline scene sphere within the standard deviation, To match the preset information level The corresponding confidence coefficient.

5. The driving behavior safety assessment method based on particle-sphere calculation according to claim 1, characterized in that, Based on the environmental condition feature vector and the scene reference space, a dynamic reference behavior particle matching the current operating condition is generated, including: Calculate the Mahalanobis distance between the environmental condition feature vector and each reference scene particle in the scene reference space, and select the nearest reference scene particle from them; Based on the inverse distance weighted interpolation method, the reference safety behavior particles corresponding to the neighboring reference scene particles are weighted and mapped to generate dynamic reference behavior particles that match the current working conditions.

6. The driving behavior safety assessment method based on particle sphere calculation according to claim 1, characterized in that, The boundary radius of the individual behavior particle is physically constrained and compensated based on the vehicle load factor and braking performance factor to obtain a corrected individual behavior particle, including: Calculate the vehicle load factor based on the current total vehicle mass and the reference total mass; The braking performance factor is calculated based on the current brake disc temperature, the reference brake disc temperature, the amount of brake wear, and the minimum braking performance threshold. Calculate the physical correction coefficient based on the vehicle load factor and the braking performance factor; Based on the physical correction coefficient, boundary radius compensation is performed on the behavioral feature dimension that is highly related to the physical constraints to obtain the compensated boundary radius. Based on the compensated boundary radius and the driver's average behavior level in each behavioral feature dimension within the current time window, the corrected individual behavior particle is obtained.

7. The driving behavior safety assessment method based on particle sphere calculation according to claim 1, characterized in that, The topological deviation measure The topological deviation measure is obtained by jointly calculating the directional risk center deviation, radius expansion deviation, and non-overlap gap deviation of the corrected individual behavior particle and the dynamic baseline behavior particle. The expression is: ； ； ； ； in, For the first Directional risk center deviation in each behavioral characteristic dimension; For the first Radius expansion bias of each behavioral feature dimension; For the first Non-overlapping gap deviation of each behavioral feature dimension; ; For the first The safety direction indication value for each behavioral feature dimension is taken as follows: the larger the value of this dimension, the more dangerous it is. When the smaller the value of this dimension, the more dangerous it is, take... ; For the first Risk weights for each behavioral characteristic dimension; , , These are the weighting coefficients for directional risk center deviation, radius expansion deviation, and non-overlapping gap deviation, respectively. To correct individual behavioral particles in the first... The center of dimensional; For the dynamic reference behavior of the particle in the first The center of dimensional; To prevent positive numbers with a denominator of zero; To correct individual behavioral particles in the first... The radius of the dimension; For the dynamic reference behavior of the particle in the first The radius of the dimension.

8. The driving behavior safety assessment method based on particle sphere calculation according to claim 1, characterized in that, Partial derivative analysis is performed based on the aforementioned topological deviation measure to obtain the feature contribution of each behavioral feature dimension, and the dominant physical dimension is identified, including: Partial derivative analysis is performed based on the aforementioned topological deviation measure to obtain the feature contribution of each behavioral feature dimension; the expression for the feature contribution is: ,in, For the first Feature contribution of each behavioral feature dimension For topological deviation measure, To correct individual behavioral particles in the first... The center of dimensional, To correct individual behavioral particles in the first... The boundary radius of the dimension, Contribution coefficient to radius deviation; The feature contribution is normalized to obtain the normalized feature contribution; the expression for the normalized feature contribution is: ,in, The normalized feature contribution. To prevent positive numbers with a denominator of zero; For the first Feature contribution of each behavioral feature dimension; When the normalized feature contribution If the contribution is greater than the preset contribution threshold or is among the top few maximum values, then the first value is determined to be... The behavioral characteristic dimension is the dominant physical dimension.

9. The driving behavior safety assessment method based on particle-sphere computation according to claim 1, characterized in that, Based on the mapping relationship between the dominant physical dimension and preset features and actions, driving behavior correction suggestions are generated, including: When the dominant physical dimension corresponds to the throttle opening change rate or longitudinal impact, a throttle smoothness correction suggestion is output. When the dominant physical dimension corresponds to the minimum following distance, braking intervention collision time, or peak braking deceleration, then braking intervention timing and following distance correction suggestions are output. When the dominant physical dimension corresponds to the peak lateral acceleration or steering angular velocity, steering smoothness and cornering centripetal acceleration control correction suggestions are output.

10. A driving behavior safety assessment system based on particle sphere computation, characterized in that, The system includes: The acquisition and extraction unit is used to acquire a compliant driving sample set; and extract environmental condition feature vectors, physical state parameter vectors, and behavioral feature vectors based on the compliant driving sample set. The scenario baseline space construction unit is used to perform manifold mapping and particle clustering on the environmental condition feature vectors to construct the scenario baseline space; and to establish a set of scenario and behavior baseline pairs composed of baseline scenario particles and their corresponding baseline safety behavior particles. The dynamic reference behavior particle generation unit is used to collect the environmental condition feature vector and physical state parameter vector of the target vehicle at the current moment, and generate a dynamic reference behavior particle that matches the current condition based on the environmental condition feature vector and the scene reference space. The individual behavior particle generation unit is used to extract the target driver's behavior feature vector sequence in real time within the same sliding time window as the current time, construct an individual behavior particle, and perform physical constraint compensation on the boundary radius of the individual behavior particle based on the vehicle load factor and braking performance factor to obtain the corrected individual behavior particle. The topological deviation measurement calculation unit is used to calculate the topological deviation measurement between the corrected individual behavior particle and the dynamic benchmark behavior particle in the behavior feature space. The safety score generation and correction unit is used to perform partial derivative analysis based on the topological deviation measure to obtain the feature contribution of each behavioral feature dimension and identify the dominant physical dimension; based on the dominant physical dimension, it generates a safety score and driving behavior correction suggestions.

Images