Driver state multi-dimensional identification system and device

Abstract

The invention discloses a driver state multi-dimensional identification system and device, and relates to the technical field of vehicle safety. The system comprises an initial individual baseline construction module, a real-time data acquisition module, a stable driving period identification module, an individual baseline dynamic updating module and a driver state judgment module. The initial individual baseline construction module collects stable data when a driver uses the initial individual baseline for the first time under multiple typical scenes, and generates scene-divided personalized baselines; the real-time data acquisition module acquires real-time multi-dimensional data and scene feature data and identifies a scene; a smooth driving period identification module matches the scene baseline, screens smooth segments and combines periods; the individual baseline dynamic updating module periodically corrects a baseline based on stationary data; and the driver state judgment module calculates a single-dimensional and multi-dimensional collaborative deviation value and judges the state in a grading manner. The method solves a problem that a group standard neglects individual differences, improves the recognition precision and scene adaptability, and guarantees the driving safety.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of vehicle safety, in particular to a driver state multi-dimensional recognition system and device. BACKGROUND

[0002] In the field of road traffic safety, accurately identifying the driver state is crucial for preventing traffic accidents. Currently, most driver state recognition systems and devices on the market mainly rely on group standards to determine whether the driver is in an abnormal state. For example, in terms of heart rate monitoring, the average threshold of the group, such as "normal heart rate 60-100 times / min", is commonly used as the basis for judgment; in other dimensions such as blink frequency and operation habits, general group data is also used as the standard. However, this recognition method based on group standards has obvious defects. The driver group has significant individual differences. Drivers of different ages, occupations and physical conditions have different physiological indicators and behavior habits in their normal state. For example, athletes, after long-term training, their resting heart rate is often lower than 60 times / min, and if they are judged according to the traditional group standard, they are likely to be misjudged as having an abnormal heart rate; elderly drivers have different reaction times and operation forces from young drivers due to their declining physical functions; professional drivers have formed unique operation habits and fatigue tolerance levels due to long-term driving. Traditional systems ignore these individual differences, making it difficult to accurately identify the true state of special individuals, resulting in frequent misjudgments and omissions. Misjudgment may cause unnecessary interference and alarms, affecting the driving experience; omission cannot timely detect the potential dangerous state of the driver, increasing the risk of traffic accidents. Therefore, the existing driver state recognition technology and equipment based on group standards cannot meet the growing demand for traffic safety, and there is an urgent need for a new technology and equipment that can fully consider individual differences and accurately identify the driver state to improve road traffic safety. SUMMARY

[0003] The purpose of the present application is to provide a driver state multi-dimensional recognition system and device to solve the problems raised in the background art.

[0004] In order to solve the above technical problems, the present application provides the following technical solutions: A driver state multi-dimensional recognition system, comprising: an initial individual baseline construction module, a real-time data acquisition module, a smooth driving period identification module, an individual baseline dynamic updating module and a driver state judgment module. The initial individual baseline construction module collects multi-dimensional data of the driver when the driver uses the system for the first time in various typical driving scenarios based on a preset baseline calibration process, and generates an initial individual baseline set of the driver in the corresponding scenarios according to the collected multi-dimensional data; wherein the multi-dimensional data are physiological characteristic data, behavior characteristic data and operation characteristic data of the driver in a stable driving state in each scenario, and the initial individual baseline set includes physiological baseline, behavior baseline and operation baseline in different scenarios; The real-time data acquisition module acquires real-time multi-dimensional data and real-time scene characteristic data of the driver in the subsequent driving process, wherein the real-time multi-dimensional data include real-time physiological characteristic data, real-time behavior characteristic data and real-time operation characteristic data of the driver, and the real-time scene characteristic data are used to represent the current driving scene type; The stable driving period identification module performs hierarchical analysis on the real-time multi-dimensional data and the real-time scene characteristic data, first matches the corresponding scene baseline based on the real-time scene characteristic data, then screens the stable driving segments in the real-time multi-dimensional data in combination with the scene baseline, and combines the continuous stable driving segments into a stable driving period; wherein the stable driving segment is a period in which the deviation of the real-time multi-dimensional data from the corresponding scene baseline is within a preset range and the driving state has no fluctuation; The individual baseline dynamic updating module periodically adjusts the corresponding baseline in the initial individual baseline set according to the real-time multi-dimensional data in the identified stable driving period in the physiological, behavior and operation dimensions, to form an updated individual baseline set; wherein the periodic adjustment includes independent correction of the characteristic parameters of each dimension baseline and collaborative correction of the cross-dimension associated parameters; The driver state judgment module compares the real-time multi-dimensional data with the baseline in the updated individual baseline set that matches the current scene, calculates the single-dimensional deviation value and the multi-dimensional collaborative deviation value of each dimension data, and determines whether the current state of the driver is abnormal according to the combination result of the single-dimensional deviation value and the multi-dimensional collaborative deviation value.

[0005] Further, the initial individual baseline construction module includes a scene data acquisition unit, a baseline calculation unit and a baseline set integration unit; The scene data acquisition unit determines a typical driving scene set, guides the driver to complete stable driving operation in each scene, and synchronously acquires a multi-dimensional data sequence in the process through a multi-dimensional sensor; wherein the typical driving scene set includes an urban road scene, an expressway scene, a night road scene and a rainy and snowy weather scene, and the multi-dimensional data sequence includes a physiological characteristic data sequence, a behavior characteristic data sequence and an operation characteristic data sequence; wherein the physiological characteristic data sequence includes continuously acquired heart rate data and blink duration data, the behavior characteristic data sequence includes continuously acquired eye gaze angle data and head rotation angle data, and the operation characteristic data sequence includes continuously acquired steering wheel rotation angle data and accelerator pedal depression depth data; The baseline calculation unit performs statistical analysis on the multi-dimensional data sequence in a typical driving scene, so as to obtain the mean and standard deviation of each characteristic parameter, and generate a physiological baseline, a behavior baseline and an operation baseline for each scene; wherein the expression of the physiological baseline generated according to the physiological characteristic data sequence is: Bp(s)={(μp1(s),σp1(s)),(μp2(s),σp2(s)),...,(μpn(s),σpn(s))}, wherein s represents a typical driving scene, μpi(s) is the mean of the i-th physiological characteristic under the scene s, σpi(s) is the standard deviation of the i-th physiological characteristic under the scene s, and i is 1 to n; similarly, the same processing is performed on the behavior characteristic data sequence and the operation characteristic data sequence to generate a behavior baseline Bb(s) and an operation baseline Bo(s) for each scene, and the expression form of the behavior baseline and the operation baseline is consistent with that of the physiological baseline; The baseline set integration unit integrates the physiological baseline, the behavior baseline and the operation baseline for each scene to form an initial individual baseline set B0 containing the multi-dimensional baseline corresponding to each scene, and B0={Bp(s),Bb(s),Bo(s)|s∈typical driving scene set}.

[0006] Further, the real-time data acquisition module includes a scene environment parameter acquisition unit, a driving scene classification unit and a multi-dimensional real-time data acquisition unit. The scene environment parameter acquisition unit acquires real-time scene characteristic data through a scene recognition sensor, wherein the real-time scene characteristic data includes a road type parameter, an illumination intensity parameter and a precipitation intensity parameter; the driving scene classification unit classifies and matches the real-time scene characteristic data acquired by the scene environment parameter acquisition unit, and identifies the current driving scene type according to the matching result; the multi-dimensional real-time data acquisition unit synchronously acquires real-time physiological characteristic data, real-time behavior characteristic data and real-time operation characteristic data of the driver under the current driving scene type identified by the driving scene classification unit.

[0007] Further, the driving scene classification unit classifies and matches the real-time environment parameters acquired by the scene environment parameter acquisition unit, and identifies the current driving scene type according to the matching result, and the specific process is as follows: A scene classification rule library is constructed in advance, wherein the scene classification rule library contains the threshold range of the environment parameters corresponding to each typical driving scene (urban road scene, highway scene, night road scene and rainy and snowy weather scene), and the threshold range of the road type parameter includes the road width interval, the lane number interval and the speed limit interval corresponding to the urban road, the threshold range of the illumination intensity parameter includes the illumination intensity interval corresponding to the night road, and the threshold range of the precipitation intensity parameter includes the precipitation intensity interval corresponding to the rainy and snowy weather; The real-time scene feature data output by the scene environment parameter collection unit is received, wherein the real-time scene feature data includes a real-time road type parameter R, a real-time light intensity parameter L, and a real-time precipitation intensity parameter W; the real-time road type parameter R is matched with the road type parameter threshold range of each scene in the scene classification rule library to obtain a road type matching result; the real-time light intensity parameter L is matched with the light intensity parameter threshold range of each scene in the scene classification rule library to obtain a light intensity matching result; and the real-time precipitation intensity parameter W is matched with the precipitation intensity parameter threshold range of each scene in the scene classification rule library to obtain a precipitation intensity matching result. According to the above three matching results, a scene matching degree calculation model is used to determine the current driving scene type. When the three matching results corresponding to a certain typical driving scene all meet the threshold requirements, or two core matching results (the road type matching result and the light intensity matching result) meet the threshold requirements and the third matching result does not conflict, it is determined that the current driving scene type is the typical driving scene. The calculation formula of the scene matching degree S(s) is: S(s) = α·S_R(s) + β·S_L(s) + γ·S_W(s), wherein s represents a typical driving scene, S_R(s) is the matching degree of the real-time road type parameter R and the road type parameter threshold range of the scene s; S_L(s) is the matching degree of the real-time light intensity parameter L and the light intensity parameter threshold range of the scene s; S_W(s) is the matching degree of the real-time precipitation intensity parameter W and the precipitation intensity parameter threshold range of the scene s; α, β, and γ are weight coefficients of the road type matching degree, the light intensity matching degree, and the precipitation intensity matching degree, respectively, and α + β + γ = 1, wherein α ≥ β > γ, the value range of α is 0.4-0.6, the value range of β is 0.3-0.4, and the value range of γ is 0.1-0.2; when S(s) ≥ 0.8, it is determined that the current driving scene type is the scene s; when there are multiple scenes s with S(s) ≥ 0.8, the scene with the maximum S(s) and the further refined road type parameter of S_R(s) corresponding to α is selected as the current driving scene type.

[0008] Further, the smooth driving period identification module includes a scene baseline matching unit, a multi-dimensional deviation degree calculation unit, a smooth driving segment screening unit, and a smooth driving period merging unit. The scene baseline matching unit receives the current driving scene type s_current output by the driving scene classification unit and calls the sub-scene baseline group B(s_current) corresponding to the current driving scene type from the initial individual baseline set B0 generated by the baseline set integration unit in the initial individual baseline construction module. The sub-scene baseline group B(s_current) includes the physiological baseline Bp(s_current), behavioral baseline Bb(s_current), and operational baseline Bo(s_current) for each scene. The multi-dimensional deviation calculation unit receives the real-time multi-dimensional data output by the multi-dimensional real-time data acquisition unit, and calculates the real-time deviation of each dimension respectively in combination with the sub-scene baseline group B (s_current) called by the scene baseline matching unit; The stable driving segment screening unit screens stable driving segments in the real-time multi-dimensional data based on the real-time deviation of each dimension output by the multi-dimensional deviation calculation unit and the driving state fluctuation judgment condition; The smooth driving period merging unit merges the multiple smooth driving segments output by the smooth driving segment screening unit to form a complete smooth driving period.

[0009] Furthermore, the specific process of calculating the real-time deviation of each dimension in the multi-dimensional deviation calculation unit is as follows: Calculate the physiological deviation Dp(t) between the real-time physiological characteristic number P(t) and the physiological baseline Bp(s_current). The corresponding calculation formula is: Dp(t)=(1 / n)∑ i∈[1,n] [|pi(t)-μ_pi(s_current)| / (k·σ_pi(s_current))], where pi(t) is the real-time value of the i-th physiological characteristic at time t, μ_pi(s_current) is the baseline mean of the i-th physiological characteristic in the s_current scenario, σ_pi(s_current) is the baseline standard deviation of the i-th physiological characteristic in the s_current scenario, and k is the preset deviation correction coefficient; using the same calculation logic and formula as the above physiological deviation, calculate the behavioral deviation Db(t) between the real-time behavioral characteristic data B(t) and the behavioral baseline Bb(s_current), and the operational deviation Do(t) between the real-time operational characteristic data O(t) and the operational baseline Bo(s_current); The specific process of filtering smooth driving segments from real-time multi-dimensional data in the smooth driving segment filtering unit is as follows: A preset smooth deviation threshold D and a driving state fluctuation threshold Z, for each timestamp t, if Dp(t)≤D, Db(t)≤D and Do(t)≤D are satisfied, and the variance Var(O(t), O(t-1), O(t-2)) of the real-time operation feature data O(t) within the continuous three timestamps is less than or equal to Z, it is determined that the t moment is a smooth driving point; the smooth driving points with continuous timestamps are combined to form a continuous data segment, and when the continuous data segment has a duration greater than or equal to a preset segment duration threshold, the continuous data segment is marked as a smooth driving segment; The specific analysis process of the smooth driving period merging unit is as follows: A preset segment interval threshold ε and a period minimum duration threshold θ, if the time interval between two adjacent smooth driving segments is less than or equal to the segment interval threshold ε, the two segments are merged into a new smooth driving segment; repeat the above merging operation until the interval of all adjacent segments is greater than the segment interval threshold ε; filter the segments with a duration greater than or equal to the period minimum duration threshold θ after merging, and mark them as the final smooth driving period.

[0010] Further, the individual baseline dynamic updating module includes a smooth data extraction unit, a single-dimension baseline independent correction unit, a cross-dimension association and collaborative correction unit, and an updated baseline set integration unit. The smooth data extraction unit receives the smooth driving period output by the smooth driving period identification module and the current driving scene type s_current output by the driving scene classification unit, and extracts the multi-dimensional smooth data sequence of the current driving scene type in the smooth driving period from the real-time multi-dimensional data, including the physiological feature smooth subsequence P_stable, the behavior feature smooth subsequence B_stable and the operation feature smooth subsequence O_stable. The single-dimension baseline independent correction unit independently corrects the baseline corresponding to the scene in the initial individual baseline set B_0 according to the multi-dimensional smooth data sequence output by the smooth data extraction unit based on the physiological, behavioral and operational dimensions; the specific analysis process is as follows: A preset baseline update period T and a correction coefficient λ are set, and 0 < λ < 1; when the system running time reaches the update period T, an independent correction process is triggered, the physiological characteristic stable subsequence P_stable is statistically analyzed, and the stable mean μp_new(s_current) and the stable standard deviation σp_new(s_current) of the i-th physiological characteristic are calculated; based on the initial physiological baseline Bp_old(s_current) = {(μp_old(s_current), σp_old(s_current))}, the corrected physiological baseline parameter is calculated, wherein the calculation formula of the corrected mean μp-update(s_current) and the standard deviation σp-update(s_current) is: μp-update(s_current) = (1-λ)·μp_old(s_current)+λ·μp_new(s_current), σp-update(s_current) = (1-λ)·σp_old(s_current)+λ·σp_new(s_current); thus, the corrected physiological baseline is Bp_update(s_current) = {(μp-update, σp-update(s_current))}; The same logic and formula as the above independent correction of the physiological baseline are used to process the behavior characteristic stable subsequence B_stable and the operation characteristic stable subsequence O_stable, respectively, so as to obtain the corrected behavior baseline Bb_update(s_current) and the operation baseline Bo_update(s_current); The cross-dimension association cooperative correction unit cooperatively adjusts the single-dimension independently corrected baseline based on the change of the association relationship of the multi-dimension stable data sequence; the specific analysis process is as follows: The Pearson correlation coefficient calculation method is used to calculate the physiological-behavior association correlation r_p-b, the physiological-operation association correlation r_p-o and the behavior-operation association correlation r_b-o; the historical cross-dimension association correlation r_old corresponding to the scene s_current in the initial individual baseline set B0 is extracted, which includes r_p-b-old, r_p-o-old and r_b-o-old; the corresponding association correlation change amount Δr = |r_new-r_old| is calculated, wherein r_new is the new association correlation calculated based on the stable data; a preset association change threshold Δr0 is set; if Δr ≥ Δr0, the cooperative correction is triggered; if Δr < Δr0, it is determined that the association relationship has no significant change, and the single-dimension independently corrected baseline parameter does not need to be cooperatively corrected; The update baseline set integration unit integrates the baselines of each dimension after independent correction of a single dimension or collaborative correction across dimensions to form an updated baseline group B_update(s_current) for the corresponding scenario, which is expressed as: B_update(s_current)={Bp_final(s_current),Bb_final(s_current),Bo_final(s_current)}, where Bp_final is the physiological baseline after independent or collaborative correction, Bb_final is the behavioral baseline after independent or collaborative correction, and Bo_final is the operational baseline after independent or collaborative correction. The updated baseline groups of all scenarios are integrated according to the structure of the initial individual baseline set B0 to form an updated individual baseline set Bnew covering all typical driving scenarios, and Bnew={B_update(s|s∈typical driving scenario set)}.

[0011] Furthermore, the driver state judgment module includes a scene baseline calling unit, a single-dimensional deviation value calculation unit, a multi-dimensional collaborative deviation value calculation unit, an abnormal state judgment unit and a judgment result output unit; The scene baseline calling unit receives the current driving scene type s_current output by the driving scene classification unit and the updated individual baseline set Bnew output by the individual baseline dynamic update module, and calls the target scene baseline group B_target(s_current) that matches s_current from the updated individual baseline set Bnew; the target scene baseline group includes the updated physiological baseline Bp_final(s_current), behavioral baseline Bb_final(s_current), and operational baseline Bo_final(s_current); The single-dimensional deviation value calculation unit receives the real-time multi-dimensional data output by the multi-dimensional real-time data acquisition unit and the target scene baseline group B_target(s_current) output by the scene baseline call unit, and calculates the single-dimensional deviation value of each feature by dimension. The specific analysis process is as follows: The deviation correction coefficient k1 is preset, and the real-time deviation calculation formula in the multi-dimensional deviation calculation unit is referred to to calculate the single-dimensional deviation value Dp_single(t) of the physiological dimension, the single-dimensional deviation value Db_single(t) of the behavioral dimension, and the single-dimensional deviation value Do_single(t) of the operational dimension respectively; The multi-dimensional collaborative deviation value calculation unit calculates the multi-dimensional collaborative deviation value based on the single-dimensional deviation value and the scenario-based weight of each dimension, and introduces the correction result of the inter-dimensional collaborative coefficient. The specific analysis content is as follows: A preset scene-based dimension weight library is configured to call corresponding weights w_p(s_current), w_b(s_current), and w_o(s_current) for the current scene s_current, and satisfy w_p(s_current)+w_b(s_current)+w_o(s_current)=1, and the weight distribution matches the scene characteristics; an initial multi-dimensional coordination deviation value Dc_init(t) is calculated according to the calculation formula: Dc_init(t)=w_p(s_current)·Dp_single(t)+w_b(s_current)·Db_single(t)+w_o(s_current)·Do_single(t); according to a preset coordination analysis time window T0, the synchronicity of real-time single-dimensional deviation values is calculated by using a Pearson correlation coefficient, thereby obtaining a physiological-operation coordination coefficient Cp-o(t), a physiological-behavior coordination coefficient Cp-b(t), and a behavior-operation coordination coefficient Cb-o(t), and taking the average of the three as a final coordination coefficient C(t); if C(t)≥0.6, the coefficient of Dc_init(t) is corrected, and the corresponding formula is: Dc_final(t)=Dc_init(t)·(1+f·C(t)), wherein f represents the coefficient; if C(t)<0.6, Dc_final(t)=Dc_init(t); The abnormal state determination unit determines the current state of the driver based on the single-dimensional deviation value and the corrected multi-dimensional coordination deviation value in combination with a preset threshold value; if Dp_single(t)<D1, Db_single(t)<D1, Do_single(t)<D1, and Dc_final(t)<D2, it is determined as “normal state”, wherein D1 is a single-dimensional abnormal threshold value, and D2 is a multi-dimensional abnormal threshold value; If any single-dimensional deviation value is greater than or equal to D1 and Dc_final(t)<D2, it is determined as “mild abnormal state”; If Dc_final(t)≥D2 and C(t)<D3, it is determined as “moderate abnormal state”, wherein D3 is a coordination enhancement threshold value; If Dc_final(t)≥D2 and C(t)≥D3, it is determined as “severe abnormal state”; The determination result output unit receives the state determination result output by the abnormal state determination unit, and outputs the determination result including “current state type (normal / mild abnormal / moderate abnormal / severe abnormal), abnormal dimension identifier (such as “physiological dimension-heart rate abnormality”), single-dimensional deviation value details, multi-dimensional coordination deviation value, and coordination coefficient” in a preset format.

[0012] A driver state multi-dimensional identification device includes a multi-dimensional sensor group and a system warning module; The multi-dimensional sensor group includes physiological characteristic sensors, behavioral characteristic sensors, operational characteristic sensors and scene recognition sensors, such as DMS cameras; the system warning module receives the judgment results output by the driver status judgment module and triggers corresponding warning interventions according to different abnormal status types. The warning interventions include light prompts, seat vibrations and voice warnings.

[0013] Compared with existing technologies, the present invention offers the following advantages: Through its initial individual baseline construction module, it collects driver-specific stable data under various typical driving scenarios to generate scenario-specific personalized baselines. Combined with the periodic adjustments (including cross-dimensional collaborative corrections) of the individual baseline dynamic update module, the baseline continuously adapts to changes in individual driver characteristics, fundamentally replacing group standards and significantly improving the accuracy of state recognition for specific groups, reducing the risk of misjudgments and missed detections. The present invention accurately identifies the current scene through a driving scenario classification unit. Modules such as stable driving period recognition and driver state determination all utilize individual baselines matching the scenario. Furthermore, multi-dimensional collaborative deviation calculation incorporates scenario-specific weights, aligning recognition logic with scenario characteristics, avoiding biases caused by insufficient scenario adaptation, and maintaining stable recognition performance in complex environments. The present invention utilizes the individual baseline dynamic update module to perform both single-dimensional independent and cross-dimensional collaborative corrections based on stable driving data, balancing the weights of historical baselines and new data, capturing dimensional correlation changes, and ensuring that the baseline consistently adapts to the driver's current state. This addresses the "outdated" issue of traditional static baselines and ensures the system's recognition timeliness and stability over the long term. This invention divides the driver's status into four levels: normal, mildly abnormal, moderately abnormal, and severely abnormal. The abnormality dimension and coordination coefficient are labeled. Based on the classification results, the system can trigger differentiated warnings (such as light prompts for mild abnormalities and voice and vibration warnings for severe abnormalities). This not only prevents minor abnormalities from interfering with driving, but also strengthens intervention for high-risk conditions, thereby ensuring traffic safety while improving the driving experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 This is a schematic diagram of a multi-dimensional driver status recognition system of the present invention. DETAILED DESCRIPTION

[0015] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0016] Referring to Figure 1 The application provides a technical solution: A driver state multi-dimensional recognition system comprises an initial individual baseline construction module, a real-time data acquisition module, a smooth driving period identification module, an individual baseline dynamic updating module and a driver state judgment module. The initial individual baseline construction module collects multi-dimensional data of a driver when the driver uses the system for the first time under multiple typical driving scenarios based on a preset baseline calibration process, and generates an initial individual baseline set of the driver under the corresponding scenarios according to the collected multi-dimensional data; wherein the multi-dimensional data are physiological feature data, behavior feature data and operation feature data of the driver under a smooth driving state in each scenario, and the initial individual baseline set comprises a physiological baseline, a behavior baseline and an operation baseline in each scenario. The real-time data acquisition module acquires real-time multi-dimensional data and real-time scene feature data of the driver in a subsequent driving process, wherein the real-time multi-dimensional data comprise real-time physiological feature data, real-time behavior feature data and real-time operation feature data of the driver, and the real-time scene feature data are used to represent the type of the current driving scenario. The smooth driving period identification module performs hierarchical analysis on the real-time multi-dimensional data and the real-time scene feature data, first matches the corresponding scene baseline based on the real-time scene feature data, then screens the smooth driving segments in the real-time multi-dimensional data in combination with the scene baseline, and combines the continuous smooth driving segments into a smooth driving period; wherein the smooth driving segment is a period in which the deviation of the real-time multi-dimensional data from the corresponding scene baseline is within a preset range and the driving state has no fluctuation. The individual baseline dynamic updating module periodically adjusts the corresponding baseline in the initial individual baseline set according to the real-time multi-dimensional data in the identified smooth driving period in the physiological, behavior and operation dimensions respectively, and forms an updated individual baseline set; wherein the periodic adjustment comprises independent correction of the feature parameters of each dimension baseline and collaborative correction of the cross-dimension associated parameters. The driver state judgment module compares the real-time multi-dimensional data with the baseline in the updated individual baseline set that matches the current scenario, calculates the single-dimensional deviation value and the multi-dimensional collaborative deviation value of each dimension data, and determines whether the current state of the driver is abnormal according to the combination result of the single-dimensional deviation value and the multi-dimensional collaborative deviation value.

[0017] The initial individual baseline construction module comprises a scene data acquisition unit, a baseline calculation unit and a baseline set integration unit. The scenario data acquisition unit determines a set of typical driving scenarios, guides the driver to complete smooth driving operations in each scenario, and synchronously collects multi-dimensional data sequences in the process through multi-dimensional sensors; the typical driving scenario set includes urban road scenarios, highway scenarios, nighttime road scenarios, and rainy and snowy weather scenarios; the multi-dimensional data sequences include physiological characteristic data sequences, behavioral characteristic data sequences, and operational characteristic data sequences; the physiological characteristic data sequences include continuously collected heart rate data and blink duration data, the behavioral characteristic data sequences include continuously collected eye gaze angle data and head rotation angle data, and the operational characteristic data sequences include continuously collected steering wheel rotation angle data and accelerator pedal depression depth data; The baseline calculation unit performs statistical analysis on the multi-dimensional data sequence under typical driving scenarios to obtain the mean and standard deviation of each feature parameter, and generates the physiological baseline, behavioral baseline and operation baseline for each scenario. The expression of the physiological baseline generated based on the physiological feature data sequence is: Bp(s)={(μp1(s),σp1(s)),(μp2(s),σp2(s)),...,(μpn(s),σpn(s))}, where s represents a typical driving scenario, μpi(s) is the mean of the i-th physiological feature in scenario s, σpi(s) is the standard deviation of the i-th physiological feature in scenario s, and i ranges from 1 to n. Similarly, the behavioral feature data sequence and the operation feature data sequence are processed in the same way to generate the behavioral baseline Bb(s) and the operation baseline Bo(s) for each scenario, and the expression form of the behavioral baseline and the operation baseline is consistent with that of the physiological baseline. The baseline set integration unit integrates the physiological baselines, behavioral baselines, and operational baselines of different scenarios to form an initial individual baseline set B0 containing the multi-dimensional baselines corresponding to each scenario, and B0={Bp(s),Bb(s),Bo(s)|s∈typical driving scenario set}.

[0018] The real-time data acquisition module includes a scene environment parameter acquisition unit, a driving scene classification unit, and a multi-dimensional real-time data acquisition unit; The scene environment parameter acquisition unit collects real-time scene feature data through the scene recognition sensor, and the real-time scene feature data includes road type parameters, light intensity parameters and precipitation intensity parameters; the driving scene classification unit classifies and matches the real-time scene feature data obtained by the scene environment parameter acquisition unit, and identifies the current driving scene type according to the matching results; after the driving scene classification unit identifies the current driving scene type, the multi-dimensional real-time data acquisition unit synchronously collects the driver's real-time physiological feature data, real-time behavioral feature data and real-time operation feature data in the scene.

[0019] The driving scene classification unit classifies and matches the real-time environment parameters obtained by the scene environment parameter acquisition unit, and identifies the current driving scene type according to the matching result. The specific process is as follows: A scene classification rule library is constructed in advance, wherein the scene classification rule library contains the environment parameter threshold range corresponding to each typical driving scene (urban road scene, highway scene, night road scene, and rain and snow weather scene), and the road type parameter threshold range includes the road width interval, the lane number interval, and the speed limit interval corresponding to the urban road, the light intensity parameter threshold range includes the light intensity interval corresponding to the night road, and the precipitation intensity parameter threshold range includes the precipitation intensity interval corresponding to the rain and snow weather; The real-time scene feature data output by the scene environment parameter acquisition unit is received, wherein the real-time scene feature data includes the real-time road type parameter R, the real-time light intensity parameter L, and the real-time precipitation intensity parameter W; the real-time road type parameter R is matched with the road type parameter threshold range of each scene in the scene classification rule library to obtain a road type matching result; the real-time light intensity parameter L is matched with the light intensity parameter threshold range of each scene in the scene classification rule library to obtain a light intensity matching result; and the real-time precipitation intensity parameter W is matched with the precipitation intensity parameter threshold range of each scene in the scene classification rule library to obtain a precipitation intensity matching result; According to the above three matching results, the scene matching degree calculation model is used to determine the current driving scene type. When the three matching results corresponding to a certain typical driving scene all meet the threshold requirement, or when two core matching results (road type matching result and light intensity matching result) meet the threshold requirement and the third matching result does not conflict, it is determined that the current driving scene type is the typical driving scene. The calculation formula of the scene matching degree S(s) is: S(s)=α·S_R(s)+β·S_L(s)+γ·S_W(s), wherein s represents a typical driving scene, S_R(s) is the matching degree of the real-time road type parameter R and the road type parameter threshold range of the scene s; S_L(s) is the matching degree of the real-time light intensity parameter L and the light intensity parameter threshold range of the scene s; S_W(s) is the matching degree of the real-time precipitation intensity parameter W and the precipitation intensity parameter threshold range of the scene s; α, β, and γ are the weight coefficients of the road type matching degree, the light intensity matching degree, and the precipitation intensity matching degree, respectively, and α+β+γ=1, wherein α≥β>γ, the value range of α is 0.4-0.6, the value range of β is 0.3-0.4, and the value range of γ is 0.1-0.2; when S(s)≥0.8, it is determined that the current driving scene type is the scene s; when there are multiple scenes s with S(s)≥0.8, the scene with the maximum S(s) and the further refined road type parameter corresponding to α of S_R(s) is selected as the current driving scene type.

[0020] In this embodiment, the specific calculation process of the matching degree is as follows, taken S_R(s) as an example: S_R(s) = 1 when R is within the threshold range of the road type parameter of scene s, S_R(s) = 0.6-0.9 when at least two sub-parameters in R are within the threshold range of the road type parameter of scene s and the deviation rate of the remaining sub-parameters from the threshold is ≤20%, S_R(s) = 0.1-0.5 when only one sub-parameter in R is within the threshold range of the road type parameter of scene s or the deviation rate of any sub-parameter from the threshold is >20%, and S_R(s) = 0 when all sub-parameters in R are not within the threshold range of the road type parameter of scene s and the deviation rate of any sub-parameter from the threshold is >30%; wherein the deviation rate = |real-time sub-parameter value-threshold boundary value| / threshold boundary value, if the real-time sub-parameter value is higher than the upper threshold, the upper threshold is taken as the threshold boundary value; if the real-time sub-parameter value is lower than the lower threshold, the lower threshold is taken as the threshold boundary value.

[0021] In this embodiment, it is assumed that there are night city road s1 and night highway s2 in the scene classification rule library, and the corresponding road type parameter threshold range, light intensity parameter threshold range and precipitation intensity parameter threshold range are respectively: s1: road width 15-25m, number of lanes 2-4, speed limit 40-60km / h; light intensity ≤200lux (night); precipitation intensity ≤5mm / h (no rain and snow); s2: road width 28-40m, number of lanes 4-6, speed limit 80-120km / h; light intensity ≤200lux (night); precipitation intensity ≤5mm / h (no rain and snow); It is assumed that the corresponding weight coefficients are: α=0.5 (road type), β=0.3 (light intensity) and γ=0.2 (precipitation intensity), and α+β+γ=1.

[0022] The real-time environmental parameters collected by a driver when driving on a "city expressway" at night are as follows: Real-time road type parameter R: road width 26m, number of lanes 4, speed limit 70km / h (between the threshold values of city road and highway); Real-time light intensity parameter L: 150lux (satisfies the threshold value of "night" ≤200lux); Real-time precipitation intensity parameter W: 0mm / h (no rain and snow, satisfies "precipitation intensity ≤5mm / h"); According to the matching degree formula S(s) = α·S_R(s) + β·S_L(s) + γ·S_W(s), the matching degrees of s1 (night city road) and s2 (night highway) are calculated respectively: The matching degree S(s1) of night city road s1: S_R(s1) (road type matching degree): the real-time road width 26m slightly exceeds the s1 threshold (15-25m), but the number of lanes 4 and the speed limit 70km / h partially meet the s1 characteristics, and the determination S_R(s1)=0.8 (not completely matched but close to the threshold); S_L(s1) (light intensity matching degree): 150lux≤200lux, complete match, S_L(s1)=1; S_W(s1) (precipitation intensity matching degree): 0mm / h≤5mm / h, complete match, S_W(s1)=1; S(s1)=0.5×0.8+0.3×1+0.2×1=0.4+0.3+0.2=0.9≥0.8; Similarly, for the matching degree S(s2) of the night highway s2, the same calculation is performed, and the following is obtained: S(s2)=0.5×0.8+0.3×1+0.2×1=0.4+0.3+0.2=0.9≥0.8; At this time, the matching degrees of the night urban road s1 and the night highway s2 are both 0.9, which meets the determination condition of S(s)≥0.8, that is, the case of "multiple scene matching degrees meeting the threshold" occurs. The essence of this example is that the road parameters of the urban expressway are between urban roads and highways, and the light and precipitation parameters of "no rain and snow at night" also meet the threshold requirements of the two types of scenes, finally resulting in the comprehensive matching degrees of the two scenes reaching the threshold. According to the "conflict processing rule" set before (when there are multiple scenes s with S(s)≥0.8, select the scene with the largest S(s) and further refine the road type parameter corresponding to α, so as to select the scene with larger S_R(s) as the current driving scene type"), if the road type parameter is further refined (such as adding "shoulder width", "exit density" and other sub-parameters), S_R(s) can be improved, and it is assumed that the final unique scene is determined (such as the urban expressway being classified as a "night urban road" sub-class).

[0023] The smooth driving period identification module includes a scene baseline matching unit, a multi-dimensional deviation calculation unit, a smooth driving segment screening unit, and a smooth driving period merging unit. The scene baseline matching unit receives the current driving scene type s_current output by the driving scene classification unit, and calls the scene-specific baseline group B(s_current) corresponding to the current driving scene type from the initial individual baseline set B0 generated by the baseline set integration unit in the initial individual baseline construction module; wherein the scene-specific baseline group B(s_current) includes the physiological baseline Bp(s_current), the behavior baseline Bb(s_current), and the operation baseline Bo(s_current) of the scene-specific baseline. The multi-dimensional deviation calculation unit receives the real-time multi-dimensional data output by the multi-dimensional real-time data acquisition unit, and calculates the real-time deviation of each dimension respectively in combination with the sub-scene baseline group B (s_current) called by the scene baseline matching unit; The stable driving segment screening unit screens stable driving segments in the real-time multi-dimensional data based on the real-time deviation of each dimension output by the multi-dimensional deviation calculation unit and the driving state fluctuation judgment condition; The smooth driving period merging unit merges the multiple smooth driving segments output by the smooth driving segment screening unit to form a complete smooth driving period.

[0024] The specific process of calculating the real-time deviation of each dimension in the multi-dimensional deviation calculation unit is as follows: Calculate the physiological deviation Dp(t) between the real-time physiological characteristic number P(t) and the physiological baseline Bp(s_current). The corresponding calculation formula is: Dp(t)=(1 / n)∑ i∈[1,n] [|pi(t)-μ_pi(s_current)| / (k·σ_pi(s_current))], where pi(t) is the real-time value of the i-th physiological characteristic at time t, μ_pi(s_current) is the baseline mean of the i-th physiological characteristic in the s_current scenario, σ_pi(s_current) is the baseline standard deviation of the i-th physiological characteristic in the s_current scenario, and k is the preset deviation correction coefficient; using the same calculation logic and formula as the above physiological deviation, calculate the behavioral deviation Db(t) between the real-time behavioral characteristic data B(t) and the behavioral baseline Bb(s_current), and the operational deviation Do(t) between the real-time operational characteristic data O(t) and the operational baseline Bo(s_current); The specific process of filtering smooth driving segments from real-time multi-dimensional data in the smooth driving segment filtering unit is as follows: A preset stability deviation threshold D and driving state fluctuation threshold Z are set. For each timestamp t, if Dp(t)≤D, Db(t)≤D, and Do(t)≤D are satisfied, and at the same time, the variance Var(O(t), O(t-1), O(t-2)) of the real-time operation feature data O(t) within three consecutive timestamps is ≤Z, then time t is determined to be a stable driving point. Stable driving points with consecutive timestamps are grouped into continuous data segments. When the duration of a continuous data segment is greater than or equal to the preset segment duration threshold, the continuous data segment is marked as a stable driving segment. The specific analysis process of the merging unit during the stable driving period is as follows: A preset segment interval threshold value ε and a time period minimum duration threshold value θ are set, if the time interval between two adjacent smooth driving segments is less than or equal to the segment interval threshold value ε, the two segments are combined into a new smooth driving segment; the above combining operation is repeated until the interval of all adjacent segments is greater than the segment interval threshold value ε; the segments with a duration greater than or equal to the time period minimum duration threshold value θ are screened and marked as the final smooth driving time period.

[0025] The individual baseline dynamic updating module comprises a smooth data extraction unit, a single-dimension baseline independent correction unit, a cross-dimension association and collaborative correction unit, and an updated baseline set integration unit. The smooth data extraction unit receives the smooth driving time period output by the smooth driving time period identification module and the current driving scene type s_current output by the driving scene classification unit, extracts the multi-dimension smooth data sequence of the current driving scene type in the smooth driving time period from the real-time multi-dimension data, and the multi-dimension smooth data sequence comprises a physiological feature smooth sub-sequence P_stable, a behavior feature smooth sub-sequence B_stable, and an operation feature smooth sub-sequence O_stable. The single-dimension baseline independent correction unit independently corrects the baseline corresponding to the scene in the initial individual baseline set B_0 according to the physiological, behavior, and operation dimensions based on the multi-dimension smooth data sequence output by the smooth data extraction unit; the specific analysis process is as follows: A preset baseline updating period T and a correction coefficient λ are set, and 0<λ<1; when the system running duration reaches the updating period T, the independent correction process is triggered, the physiological feature smooth sub-sequence P_stable is statistically analyzed, the stable mean μp_new(s_current) and the stable standard deviation σp_new(s_current) of the i-th physiological feature are calculated; based on the initial physiological baseline Bp_old(s_current)={(μp_old(s_current),σp_old(s_current))}, the corrected physiological baseline parameter is calculated, wherein the calculation formulas of the corrected mean μp-update(scurrent) and the standard deviation σp-update(s_current) are: μp-update(scurrent)=(1-λ)·μp_old(s_current)+λ·μp_new(s_current), σp-update(s_current)=(1-λ)·σp_old(s_current)+λ·σp_new(s_current); thus, the corrected physiological baseline is: Bp_update(s_current)={(μp-update,σp-update(s_current))}. The same logic and formula as the above physiological baseline independent correction are used to process the behavior characteristic stable sub-sequence B_stable and the operation characteristic stable sub-sequence O_stable, so as to obtain the corrected behavior baseline Bb_update(s_current) and the corrected operation baseline Bo_update(s_current). The cross-dimension correlation cooperative correction unit cooperatively adjusts the single-dimension independent corrected baseline based on the correlation relationship change of the multi-dimension stable data sequence. The specific analysis process is as follows: Pearson correlation coefficient is used to calculate the physiological-behavior correlation coefficient r_p-b, the physiological-operation correlation coefficient r_p-o and the behavior-operation correlation coefficient r_b-o. The historical cross-dimension correlation coefficients r_old in the initial individual baseline set B0 corresponding to the scene s_current are extracted, including r_p-b-old, r_p-o-old and r_b-o-old. The correlation coefficient change amount Δr = |r_new-r_old| is calculated, where r_new is the new correlation coefficient calculated based on the stable data. A preset correlation change threshold Δr0 is set. If Δr ≥ Δr0, the cooperative correction is triggered. If Δr < Δr0, it is determined that there is no significant change in the correlation relationship, and the single-dimension independent corrected baseline parameter is retained. In this embodiment, if Δr ≥ Δr0, the cooperative correction is triggered. Taking the physiological-behavior correlation as an example, if r_p-b-new > r_p-b-old (indicating that the correlation between the physiological and behavior characteristics is enhanced), the correction amplitude of the physiological baseline and the behavior baseline is adjusted, and the formula is as follows: The cooperative correction physiological mean μp_co(s_current) = μp-update(scurrent)·(1+e·Δr). The cooperative correction behavior mean μb_co(s_current) = μb-update(scurrent)·(1+e·Δr). Where e is a correlation influence coefficient (0 < e < 0.5, used to control the cooperative correction amplitude). If Δr < Δr0, it is determined that there is no significant change in the correlation relationship, and the single-dimension independent corrected baseline parameter is retained. The above process is repeated to complete the cooperative correction of the physiological-operation and behavior-operation dimensions, and the cooperative corrected baseline of each dimension (Bp_co(s_current), Bb_co(s_current), Bo_co(s_current)) is obtained.

[0026] The update baseline set integration unit integrates the baselines of each dimension after independent correction of a single dimension or collaborative correction across dimensions to form an updated baseline group B_update(s_current) for the corresponding scenario, which is expressed as: B_update(s_current)={Bp_final(s_current),Bb_final(s_current),Bo_final(s_current)}, where Bp_final is the physiological baseline after independent or collaborative correction, Bb_final is the behavioral baseline after independent or collaborative correction, and Bo_final is the operational baseline after independent or collaborative correction. The updated baseline groups of all scenarios are integrated according to the structure of the initial individual baseline set B0 to form an updated individual baseline set Bnew covering all typical driving scenarios, and Bnew={B_update(s|s∈typical driving scenario set)}.

[0027] The driver status judgment module includes a scene baseline calling unit, a single-dimensional deviation value calculation unit, a multi-dimensional collaborative deviation value calculation unit, an abnormal state judgment unit and a judgment result output unit; The scene baseline calling unit receives the current driving scene type s_current output by the driving scene classification unit and the updated individual baseline set Bnew output by the individual baseline dynamic update module, and calls the target scene baseline group B_target(s_current) that matches s_current from the updated individual baseline set Bnew; the target scene baseline group includes the updated physiological baseline Bp_final(s_current), behavioral baseline Bb_final(s_current), and operational baseline Bo_final(s_current); The single-dimensional deviation value calculation unit receives the real-time multi-dimensional data output by the multi-dimensional real-time data acquisition unit and the target scene baseline group B_target(s_current) output by the scene baseline call unit, and calculates the single-dimensional deviation value of each feature by dimension. The specific analysis process is as follows: The deviation correction coefficient k1 is preset, and the real-time deviation calculation formula in the multi-dimensional deviation calculation unit is referred to to calculate the single-dimensional deviation value Dp_single(t) of the physiological dimension, the single-dimensional deviation value Db_single(t) of the behavioral dimension, and the single-dimensional deviation value Do_single(t) of the operational dimension respectively; The multi-dimensional collaborative deviation value calculation unit calculates the multi-dimensional collaborative deviation value based on the single-dimensional deviation value and the scenario-based weight of each dimension, and introduces the correction result of the inter-dimensional collaborative coefficient. The specific analysis content is as follows: A preset scene-based dimension weight library is configured to call corresponding weights w_p(s_current), w_b(s_current), and w_o(s_current) for the current scene s_current, and satisfy w_p(s_current)+w_b(s_current)+w_o(s_current)=1, and the weight distribution matches the scene characteristics; an initial multi-dimensional coordination deviation value Dc_init(t) is calculated according to the calculation formula: Dc_init(t)=w_p(s_current)·Dp_single(t)+w_b(s_current)·Db_single(t)+w_o(s_current)·Do_single(t); according to a preset coordination analysis time window T0, the synchronicity of real-time single-dimensional deviation values is calculated by using a Pearson correlation coefficient, thereby obtaining a physiological-operation coordination coefficient Cp-o(t), a physiological-behavior coordination coefficient Cp-b(t), and a behavior-operation coordination coefficient Cb-o(t), and taking the average of the three as a final coordination coefficient C(t); if C(t)≥0.6, the coefficient of Dc_init(t) is corrected, and the corresponding formula is: Dc_final(t)=Dc_init(t)·(1+f·C(t)), wherein f represents the coefficient; if C(t)<0.6, Dc_final(t)=Dc_init(t); The abnormal state determination unit determines the current state of the driver based on the single-dimensional deviation value and the corrected multi-dimensional coordination deviation value in combination with a preset threshold value; if Dp_single(t)<D1, Db_single(t)<D1, Do_single(t)<D1, and Dc_final(t)<D2, it is determined as “normal state”, wherein D1 is a single-dimensional abnormal threshold value, and D2 is a multi-dimensional abnormal threshold value; If any single-dimensional deviation value is greater than or equal to D1 and Dc_final(t)<D2, it is determined as “mild abnormal state”; If Dc_final(t)≥D2 and C(t)<D3, it is determined as “moderate abnormal state”, wherein D3 is a coordination enhancement threshold value; If Dc_final(t)≥D2 and C(t)≥D3, it is determined as “severe abnormal state”; The determination result output unit receives the state determination result output by the abnormal state determination unit, and outputs the determination result including “current state type (normal / mild abnormal / moderate abnormal / severe abnormal), abnormal dimension identifier (such as “physiological dimension-heart rate abnormality”), single-dimensional deviation value details, multi-dimensional coordination deviation value, and coordination coefficient” in a preset format.

[0028] A driver state multi-dimensional identification device includes a multi-dimensional sensor group and a system warning module; The multi-dimension sensor group includes physiological characteristic sensors, behavior characteristic sensors, operation characteristic sensors and scene recognition sensors such as DMS cameras; the system early warning module receives the determination result output by the driver state judgment module, triggers corresponding early warning intervention according to different abnormal state types, and the early warning intervention includes light prompt, seat vibration and voice warning.

[0029] It should be noted that the relational terms herein such as first and second and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any such actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0030] Finally, it should be noted that: the above only describes the preferred embodiments of the present application, and is not used to limit the present application, although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features. Any modification, equivalent replacement, improvement and the like made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A multi-dimensional driver status recognition system, characterized by: The system includes: an initial individual baseline construction module, a real-time data acquisition module, a stable driving period identification module, an individual baseline dynamic update module, and a driver status judgment module; The initial individual baseline construction module collects multi-dimensional data of the driver when he first uses the system in various typical driving scenarios based on a preset baseline calibration process, and generates an initial individual baseline set for the driver in the corresponding scenario based on the collected multi-dimensional data; The real-time data acquisition module collects real-time multi-dimensional data and real-time scene feature data during the driver's subsequent driving process; The stable driving period identification module performs a hierarchical analysis on the real-time multi-dimensional data and the real-time scene feature data, first matching the corresponding scene baseline, then filtering the stable driving segments in the real-time multi-dimensional data based on the scene baseline, and merging consecutive stable driving segments into stable driving periods; The individual baseline dynamic update module periodically adjusts the corresponding baselines in the initial individual baseline set based on real-time multi-dimensional data during the stable driving period to form an updated individual baseline set; the periodic adjustment includes independent correction of the characteristic parameters of the baselines in each dimension and coordinated correction of cross-dimensional correlation parameters; The driver status judgment module compares the real-time multi-dimensional data with the updated individual baseline that matches the baseline of the current scene, calculates the single-dimensional deviation value and multi-dimensional collaborative deviation value of each dimensional data, and determines whether the driver's current status is abnormal based on the combined result of the two.

2. The multi-dimensional driver status recognition system according to claim 1, characterized in that: The initial individual baseline construction module includes a scene data acquisition unit, a baseline calculation unit and a baseline set integration unit; The scenario data acquisition unit determines a set of typical driving scenarios, guides the driver to complete smooth driving operations in each scenario, and synchronously collects multi-dimensional data sequences in the process through multi-dimensional sensors; The typical driving scene set includes urban road scenes, highway scenes, nighttime road scenes, and rainy and snowy weather scenes. The multi-dimensional data sequence includes physiological characteristic data sequence, behavioral characteristic data sequence, and operation characteristic data sequence. Among them, the physiological characteristic data sequence includes continuously collected heart rate data and blink duration data, the behavioral characteristic data sequence includes continuously collected eye gaze angle data and head rotation angle data, and the operation characteristic data sequence includes continuously collected steering wheel rotation angle data and accelerator pedal depression depth data. The baseline calculation unit performs statistical analysis on the multi-dimensional data sequence under the typical driving scenario, thereby obtaining the mean and standard deviation of each characteristic parameter, and generating the physiological baseline, behavioral baseline and operation baseline of the different scenarios; wherein, the expression of the physiological baseline generated according to the physiological characteristic data sequence is: Bp(s)={(μp1(s),σp1(s)),(μp2(s),σp2(s)),...,(μpn(s),σpn(s))}, wherein s represents the typical driving scenario, μpi(s) is the mean of the i-th physiological characteristic in scenario s, σpi(s) is the standard deviation of the i-th physiological characteristic in scenario s, and i ranges from 1 to n; similarly, the behavioral characteristic data sequence and the operation characteristic data sequence are processed in the same way to generate the behavioral baseline Bb(s) and the operation baseline Bo(s) of the different scenarios, and the expression form of the behavioral baseline and the operation baseline is consistent with that of the physiological baseline; The baseline set integration unit integrates the physiological baselines, behavioral baselines, and operational baselines of different scenarios to form an initial individual baseline set B0 containing the multi-dimensional baselines corresponding to each scenario, and B0={Bp(s),Bb(s),Bo(s)|s∈typical driving scenario set}.

3. The multi-dimensional driver status recognition system according to claim 1, characterized in that: The real-time data acquisition module includes a scene environment parameter acquisition unit, a driving scene classification unit and a multi-dimensional real-time data acquisition unit; The scene environment parameter acquisition unit collects real-time scene feature data through a scene recognition sensor, and the real-time scene feature data includes road type parameters, light intensity parameters, and precipitation intensity parameters; the driving scene classification unit classifies and matches the real-time scene feature data obtained by the scene environment parameter acquisition unit, and identifies the current driving scene type based on the matching results; after the driving scene classification unit identifies the current driving scene type, the multi-dimensional real-time data acquisition unit synchronously collects the driver's real-time physiological feature data, real-time behavioral feature data, and real-time operation feature data in the scene.

4. The multi-dimensional driver status recognition system according to claim 3, characterized in that: The driving scene classification unit classifies and matches the real-time environmental parameters acquired by the scene environmental parameter acquisition unit, and identifies the current driving scene type based on the matching results. The specific process is as follows: A scenario classification rule library is pre-built, which contains the environmental parameter threshold ranges corresponding to each typical driving scenario. The road type parameter threshold range includes the road width range, lane number range, and speed limit range corresponding to urban roads; the light intensity parameter threshold range includes the light intensity range corresponding to nighttime roads; and the precipitation intensity parameter threshold range includes the precipitation intensity range corresponding to rainy and snowy weather. receiving real-time scene feature data output by a scene environment parameter acquisition unit, wherein the real-time scene feature data includes a real-time road type parameter R, a real-time light intensity parameter L, and a real-time precipitation intensity parameter W; matching the real-time road type parameter R with a road type parameter threshold range for each scene in a scene classification rule library to obtain a road type matching result; matching the real-time light intensity parameter L with a light intensity parameter threshold range for each scene in the scene classification rule library to obtain a light intensity matching result; and matching the real-time precipitation intensity parameter W with a precipitation intensity parameter threshold range for each scene in the scene classification rule library to obtain a precipitation intensity matching result; According to the above three matching results, the scene matching degree calculation model is used to determine the current driving scene type. When the three matching results corresponding to a typical driving scene all meet the threshold requirements, or meet two core matching results and the third matching result has no conflict, the current driving scene type is determined to be the typical driving scene. The calculation formula of the scene matching degree S(s) is: S(s)=α·S_R(s)+β·S_L(s)+γ·S_W(s), where s represents the typical driving scene, S_R(s) is the matching degree between the real-time road type parameter R and the road type parameter threshold range of scene s; S_L(s) is the real-time road type parameter R. The matching degree of the illumination intensity parameter L with the illumination intensity parameter threshold range of scene s; S_W(s) is the matching degree of the real-time precipitation intensity parameter W with the precipitation intensity parameter threshold range of scene s; α, β, and γ are the weight coefficients of road type matching, illumination intensity matching, and precipitation intensity matching, respectively, and α+β+γ=1; when S(s)≥0.8, the current driving scene type is determined to be scene s; when there are multiple scenes s with S(s)≥0.8, the scene with the largest S(s) is selected and the road type parameter S_R(s) corresponding to α is further refined, so that the scene with the largest S_R(s) is selected as the current driving scene type.

5. The multi-dimensional driver status recognition system according to claim 3, characterized in that: The stable driving period identification module includes a scene baseline matching unit, a multi-dimensional deviation calculation unit, a stable driving segment screening unit and a stable driving period merging unit; The scene baseline matching unit receives the current driving scene type s_current output by the driving scene classification unit, and calls a sub-scene baseline group B(s_current) corresponding to the current driving scene type from the initial individual baseline set B0 generated by the baseline set integration unit in the initial individual baseline construction module; wherein the sub-scene baseline group B(s_current) includes a physiological baseline Bp(s_current), a behavioral baseline Bb(s_current), and an operational baseline Bo(s_current) of each scene; The multi-dimensional deviation calculation unit receives the real-time multi-dimensional data output by the multi-dimensional real-time data acquisition unit, and calculates the real-time deviation of each dimension respectively in combination with the sub-scene baseline group B (s_current) called by the scene baseline matching unit; The stable driving segment screening unit screens stable driving segments in the real-time multi-dimensional data based on the real-time deviation of each dimension output by the multi-dimensional deviation calculation unit and in combination with the driving state fluctuation judgment condition; The smooth driving period merging unit merges the multiple smooth driving segments output by the smooth driving segment screening unit to form a complete smooth driving period.

6. The multi-dimensional driver status recognition system according to claim 5, characterized in that: The specific process of calculating the real-time deviation of each dimension in the multi-dimensional deviation calculation unit is as follows: Calculate the physiological deviation Dp(t) between the real-time physiological characteristic number P(t) and the physiological baseline Bp(s_current). The corresponding calculation formula is: Dp(t)=(1 / n)∑ i∈[1,n] [|pi(t)-μ_pi(s_current)| / (k·σ_pi(s_current))], where pi(t) is the real-time value of the i-th physiological characteristic at time t, μ_pi(s_current) is the baseline mean of the i-th physiological characteristic in the s_current scenario, σ_pi(s_current) is the baseline standard deviation of the i-th physiological characteristic in the s_current scenario, and k is the preset deviation correction coefficient; using the same calculation logic and formula as the above physiological deviation, calculate the behavioral deviation Db(t) between the real-time behavioral characteristic data B(t) and the behavioral baseline Bb(s_current), and the operational deviation Do(t) between the real-time operational characteristic data O(t) and the operational baseline Bo(s_current); The specific process of screening the smooth driving segments in the real-time multi-dimensional data in the smooth driving segment screening unit is as follows: A stable deviation threshold D and a driving state fluctuation threshold Z are preset. For each timestamp t, if Dp(t)≤D, Db(t)≤D, and Do(t)≤D, and the variance of the real-time operation feature data O(t) within three consecutive timestamps Var(O(t), O(t-1), O(t-2))≤Z, then time t is determined to be a stable driving point. The smooth driving points with consecutive timestamps are grouped into continuous data segments. When the duration of the continuous data segment is greater than or equal to a preset segment duration threshold, the continuous data segment is marked as a smooth driving segment. The specific analysis process of the stable driving period merging unit is as follows: A segment interval threshold ε and a time segment minimum duration threshold θ are preset. If the time interval between two adjacent smooth driving segments is less than or equal to the segment interval threshold ε, the two segments are merged into a new smooth driving segment. The above merging operation is repeated until the interval between all adjacent segments is greater than the segment interval threshold ε. The segments with a duration greater than or equal to the time segment minimum duration threshold θ after merging are selected and marked as the final smooth driving segment.

7. The multi-dimensional driver status recognition system according to claim 5, characterized in that: The individual baseline dynamic update module includes a stable data extraction unit, a single-dimensional baseline independent correction unit, a cross-dimensional correlation and collaborative correction unit, and an updated baseline set integration unit; The stable data extraction unit receives the stable driving period output by the stable driving period identification module and the current driving scene type s_current output by the driving scene classification unit, and extracts a multi-dimensional stable data sequence of the current driving scene type within the stable driving period from the real-time multi-dimensional data, wherein the multi-dimensional stable data sequence includes a stable subsequence P_stable of physiological characteristics, a stable subsequence B_stable of behavioral characteristics, and a stable subsequence O_stable of operational characteristics; The single-dimensional baseline independent correction unit independently corrects the baselines of the corresponding scenarios in the initial individual baseline set B_0 according to the physiological, behavioral, and operational dimensions based on the multi-dimensional stationary data sequence output by the stationary data extraction unit. The specific analysis process is as follows: The baseline update period T and correction coefficient λ are preset, and 0<λ<1; when the system operation time reaches the update period T, the independent correction process is triggered, and the stable subsequence P_stable of the physiological characteristics is statistically analyzed to calculate the stable mean μp_new(s_current) and stable standard deviation σp_new(s_current) of the i-th physiological characteristic; based on the initial physiological baseline Bp_old(s_current)={(μp_old(s_current),σp_old(s_current))}, the corrected physiological baseline parameters are calculated, where the corrected mean μp-update(scurrent The calculation formulas corresponding to the standard deviation σp-update(s_current) are: μp-update(s_current)=(1-λ)·μp_old(s_current)+λ·μp_new(s_current), σp-update(s_current)=(1-λ)·σp_old(s_current)+λ·σp_new(s_current); the corrected physiological baseline is obtained as: Bp_update(s_current)={(μp-update,σp-update(s_current))}; The same logic and formula as the independent correction of the physiological baseline are used to process the behavioral characteristic stable subsequence B_stable and the operational characteristic stable subsequence O_stable respectively, thereby obtaining the corrected behavioral baseline Bb_update(s_current) and operational baseline Bo_update(s_current); The cross-dimensional correlation collaborative correction unit collaboratively adjusts the baseline after independent correction of a single dimension based on the correlation relationship changes of the multi-dimensional stationary data sequence. The specific analysis process is as follows: The Pearson correlation coefficient calculation method was used to calculate the physiological-behavioral correlation coefficient r_p-b, the physiological-operational correlation coefficient r_p-o, and the behavioral-operational correlation coefficient r_b-o. The historical cross-dimensional correlation coefficient r_old of the corresponding scene s_current in the initial individual baseline set B0 was extracted, including r_p-b-old, r_p-o-old, and r_b-o-old. The corresponding correlation coefficient change Δr = |r_new - r_old| was calculated, where r_new is the new correlation coefficient calculated based on the stationary data. A preset correlation change threshold Δr0 was set. If Δr ≥ Δr0, collaborative correction was triggered. If Δr < Δr0, no collaborative correction was required, and the baseline parameters after independent correction of the single dimension were retained. The updated baseline set integration unit integrates the baselines of each dimension after independent correction of a single dimension or collaborative correction across dimensions to form an updated baseline group B_update(s_current) for the corresponding scenario, which is expressed as: B_update(s_current)={Bp_final(s_current),Bb_final(s_current),Bo_final(s_current)}, where Bp_final is the physiological baseline after independent or collaborative correction, Bb_final is the behavioral baseline after independent or collaborative correction, and Bo_final is the operational baseline after independent or collaborative correction. The updated baseline groups of all scenarios are integrated according to the structure of the initial individual baseline set B0 to form an updated individual baseline set Bnew covering all typical driving scenarios, and Bnew={B_update(s|s∈typical driving scenario set)}.

8. The multi-dimensional driver status recognition system according to claim 7, characterized in that: The driver state judgment module includes a scene baseline calling unit, a single-dimensional deviation value calculation unit, a multi-dimensional collaborative deviation value calculation unit, an abnormal state judgment unit and a judgment result output unit; The scene baseline calling unit receives the current driving scene type s_current output by the driving scene classification unit and the updated individual baseline set Bnew output by the individual baseline dynamic update module, and calls the target scene baseline group B_target(s_current) that matches s_current from the updated individual baseline set Bnew; the target scene baseline group includes the updated physiological baseline Bp_final(s_current), behavioral baseline Bb_final(s_current), and operational baseline Bo_final(s_current); The single-dimensional deviation value calculation unit receives the real-time multi-dimensional data output by the multi-dimensional real-time data acquisition unit and the target scene baseline group B_target(s_current) output by the scene baseline call unit, and calculates the single-dimensional deviation value of each feature by dimension. The specific analysis process is as follows: The deviation correction coefficient k1 is preset, and the real-time deviation calculation formula in the multi-dimensional deviation calculation unit is referred to to calculate the single-dimensional deviation value Dp_single(t) of the physiological dimension, the single-dimensional deviation value Db_single(t) of the behavioral dimension, and the single-dimensional deviation value Do_single(t) of the operational dimension respectively; The multi-dimensional collaborative deviation value calculation unit calculates the multi-dimensional collaborative deviation value based on the single-dimensional deviation value and the scenario-based weight of each dimension, and introduces the correction result of the inter-dimensional collaborative coefficient. The specific analysis content is as follows: Preset scenario-based dimension weight library, call the corresponding weights w_p(s_current), w_b(s_current), w_o(s_current) for the current scenario s_current, satisfy w_p(s_current)+w_b(s_current)+w_o(s_current)=1, and the weight distribution matches the scenario characteristics; calculate the initial multi-dimensional collaborative deviation value Dc_init(t), the calculation formula is: Dc_init(t)=w_p(s_current)·Dp_single(t)+w_b(s_current)·Db_single(t)+w_o(s_curr ent)·Do_single(t); According to the preset collaborative analysis time window T0, the Pearson correlation coefficient is used to calculate the synchronization of the real-time single-dimensional deviation value, thereby obtaining the physiological-operational synergy coefficient Cp-o(t), the physiological-behavioral synergy coefficient Cp-b(t), and the behavioral-operational synergy coefficient Cb-o(t), and the average of the three is taken as the final synergy coefficient C(t); If C(t) ≥ 0.6, the coefficient of Dc_init(t) is corrected, and the corresponding formula is: Dc_final(t) = Dc_init(t)·(1+f·C(t)), where f represents the coefficient; If C(t) < 0.6, then Dc_final(t) = Dc_init(t); The abnormal state determination unit determines the driver's current state based on the single-dimensional deviation value and the corrected multi-dimensional coordinated deviation value, combined with the preset threshold. If Dp_single(t) < D1, Db_single(t) < D1, Do_single(t) < D1, and Dc_final(t) < D2, it is determined to be a "normal state", where D1 is the single-dimensional abnormal threshold and D2 is the multi-dimensional abnormal threshold; If any single-dimension deviation value is greater than or equal to D1 and Dc_final(t)<D2, it is determined to be a "mild abnormal state"; If Dc_final(t)≥D2 and C(t)<D3, it is judged as "moderate abnormal state", where D3 is the synergistic enhancement threshold; If Dc_final(t)≥D2 and C(t)≥D3, it is determined to be a "severe abnormal state"; The determination result output unit receives the state determination result output by the abnormal state determination unit, and outputs the determination result including "current state type, abnormal dimension identifier, single-dimensional deviation value details, multi-dimensional collaborative deviation value and collaborative coefficient" in a preset format.

9. A multi-dimensional driver status identification device, characterized in that: A multi-dimensional driver status recognition system according to any one of claims 1 to 8, further comprising a multi-dimensional sensor group and a system warning module; The multi-dimensional sensor group includes physiological characteristic sensors, behavioral characteristic sensors, operational characteristic sensors and scene recognition sensors; the system warning module receives the judgment results output by the driver status judgment module, and triggers corresponding warning interventions according to different abnormal status types. The warning interventions include light prompts, seat vibrations and voice warnings.

Images