Image data acquisition method for driving scene

By establishing risk assessment models for multiple driving scenarios and building image noise reduction models, the problem of incomplete data capture in a single perspective is solved, driving safety and perception capabilities are improved, and efficient early warning of dangerous behaviors is achieved.

CN119091281AActive Publication Date: 2024-12-06BEIJING ZHONGKE RUITU TECH CO LTD

Patent Information

Application Number
CN202411138601.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-19
Publication Date
2024-12-06
Estimated Expiration
2044-08-19

AI Technical Summary

Technical Problem

Data from a single perspective cannot fully, accurately and efficiently capture the driver's critical driving behavior characteristics, limiting the performance of the current predictive model and the perception ability of the driving scenario, resulting in poor driving safety improvement.

Method used

By establishing risk evaluation models for multiple different driving scenarios, dynamically adjusting the image data proportion and processing parameters of the on-board monitoring submodule, and building image noise reduction models and correlation models based on disturbance parameters to enhance the complete perception of driving scenarios and promptly warning of potential risk.

Benefits of technology

It improves driving safety, and through dynamic adjustment and complementary correction of image data, an efficient warning of dangerous behavior is achieved, and the perception of driving scenarios is enhanced.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119091281A_ABST
    Figure CN119091281A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of automobile driving, in particular to an image data acquisition method for a driving scene. Comprising the following steps: establishing a vehicle-mounted monitoring sub-module sequence and a driving scene sequence, and generating a risk evaluation model in each driving scene; acquiring vehicle real-time state parameters, and selecting a target risk evaluation model according to the vehicle real-time state parameters; and processing image data acquired by each vehicle-mounted monitoring sub-module according to the target risk evaluation model, and judging whether a risk early warning instruction is generated or not. A plurality of different driving scenes are established according to different driving parameters and environment parameters, corresponding risk evaluation models are constructed for the different driving scenes, the corresponding risk evaluation models are called according to real-time vehicle operation parameters, and image data proportions and image data processing parameters of the vehicle-mounted monitoring sub-modules are dynamically adjusted. Therefore, dangerous behaviors can be early warned more efficiently, and the driving safety is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of automobile driving technology, and in particular to an image data acquisition method for driving scenes. Background Art

[0002] With the application of artificial intelligence in smart transportation and autonomous driving, the construction of predictive models of driver driving behavior has become an important support for achieving traffic safety and efficient driving. These predictive models can help autonomous driving systems better understand and predict driver behavior during driving, thereby improving the performance and safety of autonomous driving systems.

[0003] However, data from a single perspective often cannot fully, accurately, and efficiently capture the characteristics of the driver's critical driving behavior, which limits the performance of the current prediction model. Its ability to fully perceive the driving scene is weak, and its ability to improve driving safety is poor. Summary of the invention

[0004] The purpose of this application is: to solve the above-mentioned technical problems, this application provides an image data acquisition method for driving scenes, aiming to improve the complete perception capability of driving scenes, thereby improving driving safety.

[0005] In some embodiments of the present application, a plurality of different driving scenarios are established according to different driving parameters and environmental parameters, and a corresponding risk assessment model is constructed for each different driving scenario. The corresponding risk assessment model is called according to the real-time vehicle operating parameters, and the image data proportion and image data processing parameters of each on-board monitoring sub-module are dynamically adjusted, so as to more efficiently warn of dangerous behaviors and improve driving safety.

[0006] In some embodiments of the present application, a corresponding image denoising model is constructed based on disturbance parameters in different scenarios, and a corresponding image association model is constructed based on vehicle posture parameters in different scenarios, so that each on-board monitoring sub-module collects image data for complementary correction, thereby increasing the complete perception capability of the driving scene, timely warning of potential driving risks, and improving driving safety.

[0007] In some embodiments of the present application, a method for collecting image data for a driving scene is provided, comprising:

[0008] Establish a series of vehicle monitoring submodules and driving scenarios, and generate risk assessment models for each driving scenario;

[0009] Obtain the real-time status parameters of the vehicle, and select the target risk assessment model according to the real-time status parameters of the vehicle;

[0010] Process the image data collected by each vehicle monitoring submodule according to the target risk assessment model to determine whether to generate a risk warning instruction;

[0011] The vehicle-mounted monitoring submodule sequence A, A = (a1, a2 ... a n ), where a i is the i-th vehicle monitoring submodule; n is the number of vehicle monitoring submodules.

[0012] In some embodiments of the present application, when establishing a driving scene sequence, it includes:

[0013] Extract multiple driving evaluation indicators based on historical driving parameters;

[0014] According to all driving evaluation indicators, multiple initial scenarios are established, and an initial scenario sequence B is established, B = (b1, b2…b m1 ), where b i is the i-th initial scene, m1 is the number of initial scenes;

[0015] The initial scenes are grouped and processed according to the preset similarity evaluation model;

[0016] According to the collection processing results, a driving scene sequence C is established, C = (c1, c2...c m ), where c i is the i-th driving scenario, m is the number of driving scenarios, and m≤m1.

[0017] In some embodiments of the present application, when the initial scene is aggregated, it includes:

[0018] Select the initial scene b1;

[0019] Generate the initial scene b1 and the similarity evaluation value of each initial scene in the initial scene sequence B according to the preset similarity evaluation model;

[0020] Establish a similar evaluation value sequence G, G = (g1, g2...g m1 ), g i is the similarity evaluation value between the initial scene b1 and the i-th initial scene;

[0021] Preset similarity evaluation value threshold Δg, if g i >Δg, the i-th initial scene is grouped with the initial scene b1;

[0022] Generate a single driving scenario based on the collected results;

[0023] According to the remaining uncollected initial scenes, an initial scene sequence B1 is established, B1 = (b 11 ,b 12 …b 1m2 ), where b 1iis the i-th unclassified initial scene, m2 is the number of unclassified initial scenes, and m2 <m1;

[0024] Select initial scene b 11;

[0025] The above collection steps are iterated cyclically, and a driving scene sequence C is established according to the iteration results.

[0026] In some embodiments of the present application, when establishing a driving scene sequence, it also includes:

[0027] Acquire historical driving parameters of the vehicle according to a preset update cycle;

[0028] Determining whether to generate an update instruction based on historical driving parameters of the vehicle;

[0029] If an update instruction is generated, the similarity evaluation value threshold Δg is corrected according to the update instruction, and the driving scene sequence C is updated according to the correction result.

[0030] In some embodiments of the present application, when establishing a similarity evaluation value sequence G, it includes:

[0031] Select the target initial scenes in turn;

[0032]

[0033] Among them, g is the similarity evaluation value between the initial scene b1 and the target initial scene; e1 is the preset first weight coefficient; e2 is the preset second weight coefficient; e3 is the preset third weight coefficient; Q1 is the preset first fixed coefficient; Q2 is the preset second fixed coefficient; Q3 is the preset third fixed coefficient; θ is the number of interference evaluation indicators; di is the interference evaluation value of the i-th interference evaluation indicator of the initial scene b1; d'i is the interference evaluation value of the i-th interference evaluation indicator of the target initial scene; αi is the influencing factor of the i-th interference evaluation indicator; U is the blind spot evaluation value set according to the visual blind spot overlap between the initial scene b1 and the target initial scene; ki is the difference in dependence of the i-th vehicle monitoring sub-module on the image collected by the i-th vehicle monitoring sub-module in the initial scene b1 and the target initial scene; βi is the influencing factor of the i-th vehicle monitoring sub-module.

[0034] In some embodiments of the present application, when generating a risk assessment model in each driving scenario, it includes:

[0035] Select a target driving scene according to the driving scene sequence C, and generate a driving monitoring range of the target driving scene;

[0036] The weight ratio of each vehicle monitoring submodule is set according to the blind spot range and the driving monitoring range;

[0037] Obtain vehicle posture parameters of the target driving scene;

[0038] Establish an image association model between various vehicle monitoring submodules according to vehicle posture parameters;

[0039] Generate an image denoising model based on all interference evaluation indicators of the target driving scene;

[0040] Generate a risk indicator series for the target driving scenario;

[0041] Establish a risk assessment model for the target driving scene based on the image association model and risk index series of the target driving scene;

[0042] Establish risk assessment model series J, J = (j1, j2…j m ), where j i is the risk assessment model for the i-th driving scenario.

[0043] In some embodiments of the present application, when selecting a target risk assessment model according to the real-time status parameters of the vehicle, it includes:

[0044] Extract scene evaluation indicators based on real-time vehicle status parameters;

[0045] The compatibility between the vehicle and each driving scenario is generated based on all scenario evaluation indicators, and a compatibility series K is established, K = (k1, k2…k m ), where k i is the compatibility between the vehicle and the i-th driving scenario;

[0046] Select the maximum value k in the fit degree sequence K max The risk assessment model corresponding to the driving scenario is the target risk assessment model.

[0047] In some embodiments of the present application, when determining whether to generate a warning instruction, the process includes:

[0048] Obtain the real-time image data collected by each vehicle monitoring submodule and establish an image sequence T, T = (t1, t2…t n ), where t i The real-time image collected by the i-th vehicle monitoring submodule;

[0049] Preprocess all real-time images according to the image denoising model corresponding to the target risk model;

[0050] Extract risk factors from each image according to the target risk assessment model, and generate initial risk assessment values ​​for each image;

[0051] Process all real-time images according to the image association model and correct the initial risk assessment value of each image;

[0052] Establish a risk evaluation value sequence F according to the correction result, F = (f1, f2... f n ), where f i is the risk evaluation value corresponding to the real-time image of the i-th vehicle-mounted monitoring sub-module;

[0053] Judge whether to generate a warning instruction according to the risk evaluation value sequence F.

[0054] In some embodiments of the present application, when judging whether to generate a warning instruction according to the risk evaluation value sequence F, it includes:

[0055] Set a first risk evaluation value threshold F1 and a second risk evaluation value threshold F2 according to the driving scenario corresponding to the target risk evaluation model; and F1 < F2;

[0056] If f i < F1, no warning instruction is generated;

[0057] If F1 < f i < F2, generate a first-level warning instruction for the i-th vehicle-mounted monitoring sub-module;

[0058] If f i > F2, generate a second-level warning instruction for the i-th vehicle-mounted monitoring sub-module;

[0059] Obtain the total number p of the first-level warning instruction and the second-level warning instruction;

[0060] If the total number p is greater than the preset first quantity threshold, generate a third-level warning instruction.

[0061] Compared with the prior art, the beneficial effects of an image data acquisition method for a driving scenario in an embodiment of the present application are as follows:

[0062] Establish multiple different driving scenarios according to different driving parameters and environmental parameters, construct corresponding risk evaluation models for each different driving scenario, and call the corresponding risk evaluation models according to the real-time vehicle operation parameters to dynamically adjust the image data ratio and image data processing parameters of each vehicle-mounted monitoring sub-module, so as to more efficiently warn of dangerous behaviors and improve driving safety.

[0063] Construct corresponding image noise reduction models based on the disturbance parameters in different scenarios, and construct corresponding image association models based on the vehicle attitude parameters in different scenarios, so that the image data collected by each vehicle-mounted monitoring sub-module is complementarily corrected, thereby increasing the complete perception ability of the driving scenario, warning of potential driving risks in a timely manner, and improving driving safety. Brief Description of the Drawings

[0064] Figure 1It is a flow chart of an image data acquisition method for driving scenes in a preferred embodiment of the present application. DETAILED DESCRIPTION

[0065] The specific implementation methods of the present application are further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present application but are not intended to limit the scope of the present application.

[0066] In the description of the present application, it should be understood that the terms "center", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present application.

[0067] The terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, unless otherwise specified, "plurality" means two or more.

[0068] In the description of this application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0069] like Figure 1 As shown, a method for collecting image data for a driving scene in a preferred embodiment of the present application includes:

[0070] S101: Establishing a vehicle monitoring submodule array and a driving scenario array, and generating a risk assessment model in each driving scenario;

[0071] S102: Acquire the real-time status parameters of the vehicle, and select a target risk assessment model according to the real-time status parameters of the vehicle;

[0072] S103: Processing the image data collected by each vehicle-mounted monitoring submodule according to the target risk assessment model to determine whether to generate a risk warning instruction;

[0073] The vehicle-mounted monitoring submodule sequence A, A = (a1, a2 ... a n ), where a i is the i-th vehicle monitoring submodule; n is the number of vehicle monitoring submodules.

[0074] Specifically, when establishing a driving scenario series, it includes:

[0075] Extract multiple driving evaluation indicators based on historical driving parameters;

[0076] According to all driving evaluation indicators, multiple initial scenarios are established, and an initial scenario sequence B is established, B = (b1, b2…b m1 ), where b i is the i-th initial scene, m1 is the number of initial scenes;

[0077] The initial scenes are grouped and processed according to the preset similarity evaluation model;

[0078] According to the collected processing results, a driving scene sequence is established, C = (c1, c2...c m ), where c i is the i-th driving scenario, m is the number of driving scenarios, and m≤m1.

[0079] Specifically, driving evaluation indicators include but are not limited to environmental categories such as rainy days, foggy days, strong light, etc., road categories, national highways, expressways, presence or absence of markings, traffic lights, etc., vehicle speed, vehicle posture parameters, such as left turn, right turn, forward, and reverse.

[0080] Specifically, multiple initial scenarios are constructed according to the random combination of different driving evaluation indicators, and a single initial scenario includes an environment category indicator, a road category indicator, a vehicle speed range, and a vehicle posture parameter.

[0081] Specifically, the vehicle monitoring submodule is a vehicle camera. By setting up a vehicle camera with multiple viewing angles, the limitations of the existing dangerous behavior prediction method based on a single camera are overcome.

[0082] In a preferred embodiment of the present application, when the initial scene is aggregated, it includes:

[0083] Select the initial scene b1;

[0084] Generate the initial scene b1 and the similarity evaluation value of each initial scene in the initial scene sequence B according to the preset similarity evaluation model;

[0085] Establish a similar evaluation value sequence G, G = (g1, g2...g m1 ), g i is the similarity evaluation value between the initial scene b1 and the i-th initial scene;

[0086] Preset similarity evaluation value threshold Δg, if g i >Δg, the i-th initial scene is grouped with the initial scene b1;

[0087] Generate a single driving scenario based on the collected results;

[0088] According to the remaining uncollected initial scenes, an initial scene sequence B1 is established, B1 = (b 11 ,b 12 …b 1m2 ), where b 1i is the i-th unclassified initial scene, m2 is the number of unclassified initial scenes, and m2 <m1;

[0089] Select initial scene b 11;

[0090] The above collection steps are iterated cyclically, and a driving scene sequence C is established according to the iteration results.

[0091] Specifically, when establishing a driving scenario series, it also includes:

[0092] Acquire historical driving parameters of the vehicle according to a preset update cycle;

[0093] Determining whether to generate an update instruction based on historical driving parameters of the vehicle;

[0094] If an update instruction is generated, the similarity evaluation value threshold Δg is corrected according to the update instruction, and the driving scene sequence C is updated according to the correction result.

[0095] Specifically, the initial scenes are grouped according to their similarity evaluation values, thereby reducing the modeling workload of the risk assessment model corresponding to the driving scenes and improving the efficiency of calling the model. At the same time, the driving scenes are periodically updated and the risk assessment model is corrected, thereby improving the evaluation accuracy of vehicle risk behaviors and improving vehicle driving safety.

[0096] Specifically, when establishing the similarity evaluation value sequence G, it includes:

[0097] Select the target initial scenes in turn;

[0098]

[0099] Among them, g is the similarity evaluation value between the initial scene b1 and the target initial scene; e1 is the preset first weight coefficient; e2 is the preset second weight coefficient; e3 is the preset third weight coefficient; Q1 is the preset first fixed coefficient; Q2 is the preset second fixed coefficient; Q3 is the preset third fixed coefficient; θ is the number of interference evaluation indicators; di is the interference evaluation value of the i-th interference evaluation indicator of the initial scene b1; d'i is the interference evaluation value of the i-th interference evaluation indicator of the target initial scene; αi is the influence factor of the i-th interference evaluation indicator; U is the blind spot evaluation value set according to the visual blind spot overlap between the initial scene b1 and the target initial scene; ki is the difference in the dependence of the i-th vehicle monitoring submodule on the image collected by the i-th vehicle monitoring submodule in the initial scene b1 and the target initial scene; βi is the influence factor of the i-th vehicle monitoring submodule. Specifically, by presetting the first fixed coefficient, the second fixed coefficient and the third fixed coefficient, the model is normalized so that each indicator is within the same value range.

[0100] Specifically, the reference value of the interference evaluation index refers to the factors generated based on the environment in the current initial scene, including haze, wind and sand, strong light, rain and snow, etc. that affect driving vision, and the corresponding interference evaluation value is set according to their intensity.

[0101] Specifically, in different initial scenarios, the degree of dependence on each on-board monitoring sub-module on the vehicle is different. The larger the difference in dependence, the more it indicates that the degree of use of the current on-board monitoring sub-module is different in the corresponding two initial scenarios. For example, the dependence on the on-board monitoring sub-module of the rear camera point is different when driving forward and reversing.

[0102] In a preferred embodiment of the present application, when generating a risk assessment model in each driving scenario, it includes:

[0103] Select a target driving scene according to the driving scene sequence C, and generate a driving monitoring range of the target driving scene;

[0104] The weight ratio of each vehicle monitoring submodule is set according to the blind spot range and the driving monitoring range;

[0105] Obtain vehicle posture parameters of the target driving scene;

[0106] Establish an image association model between various vehicle monitoring submodules according to vehicle posture parameters;

[0107] Generate an image denoising model based on all interference evaluation indicators of the target driving scene;

[0108] Generate a risk indicator series for the target driving scenario;

[0109] Establish a risk assessment model for the target driving scene based on the image association model and risk index series of the target driving scene;

[0110] Establish risk assessment model series J, J = (j1, j2…j m ), where j i is the risk assessment model for the i-th driving scenario.

[0111] It can be understood that in the above embodiments, a corresponding image denoising model is constructed based on the disturbance parameters in different scenarios, and a corresponding image association model is constructed according to the vehicle posture parameters in different scenarios, so that each on-board monitoring sub-module collects image data for complementary correction, thereby increasing the complete perception capability of the driving scene, timely warning of potential driving risks, and improving driving safety.

[0112] In a preferred embodiment of the present application, when selecting a target risk assessment model according to the real-time status parameters of the vehicle, it includes:

[0113] Extract scene evaluation indicators based on real-time vehicle status parameters;

[0114] The compatibility between the vehicle and each driving scenario is generated based on all scenario evaluation indicators, and a compatibility series K is established, K = (k1, k2…k m ), where k i is the compatibility between the vehicle and the i-th driving scenario;

[0115] Select the maximum value k in the fit degree sequence K max The risk assessment model corresponding to the driving scenario is the target risk assessment model.

[0116] Specifically, the degree of compatibility with each driving scenario is generated based on the vehicle's real-time driving parameters, so as to quickly call the corresponding target risk assessment model, and set the working parameters of each on-board monitoring sub-module according to the target risk assessment model.

[0117] Specifically, when determining whether to generate an early warning instruction, it includes:

[0118] Obtain the real-time image data collected by each vehicle monitoring submodule and establish an image sequence T, T = (t1, t2…t n ), where t i The real-time image collected by the i-th vehicle monitoring submodule;

[0119] Preprocess all real-time images according to the image denoising model corresponding to the target risk model;

[0120] Extract risk factors from each image according to the target risk assessment model, and generate initial risk assessment values ​​for each image;

[0121] Process all real-time images according to the image association model and correct the initial risk evaluation values of each image;

[0122] Establish a risk evaluation value sequence F based on the correction results, F = (f1, f2... f n ), where f i is the risk evaluation value corresponding to the real-time image of the i-th vehicle-mounted monitoring sub-module;

[0123] Judge whether to generate a warning instruction according to the risk evaluation value sequence F.

[0124] Specifically, when judging whether to generate a warning instruction according to the risk evaluation value sequence F, it includes:

[0125] Set a first risk evaluation value threshold F1 and a second risk evaluation value threshold F2 according to the driving scenario corresponding to the target risk evaluation model; and F1 < F2;

[0126] If f i < F1, do not generate a warning instruction;

[0127] If F1 < f i < F2, generate a first-level warning instruction for the i-th vehicle-mounted monitoring sub-module;

[0128] If f i > F2, generate a second-level warning instruction for the i-th vehicle-mounted monitoring sub-module;

[0129] Obtain the total number p of the first-level warning instructions and the second-level warning instructions;

[0130] If the total number p is greater than the preset first quantity threshold, generate a third-level warning instruction.

[0131] Specifically, different driving scenarios have different dangerous behaviors that need to be warned, and the safety response time of the vehicle is also different. By dynamically setting the corresponding risk evaluation value thresholds, the risk warning efficiency is improved and driving safety is enhanced.

[0132] Specifically, the first-level warning instruction indicates that the current vehicle driver has behaviors such as fatigue driving and non-standard driving, and the possibility of driving risks occurring is relatively high. The second-level warning instruction means that the uncertainty of the current environmental factors is relatively large, and there are potential driving risks in combination with driving behaviors. The third-level warning instruction indicates that the current behavior is very likely to cause driving risks and requires emergency braking to avoid further expanding driving risks.

[0133] According to the first concept of the present application, a plurality of different driving scenarios are established according to different driving parameters and environmental parameters, and a corresponding risk assessment model is constructed for each different driving scenario. The corresponding risk assessment model is called according to the real-time vehicle operating parameters, and the image data proportion and image data processing parameters of each on-board monitoring sub-module are dynamically adjusted, so as to more efficiently warn of dangerous behaviors and improve driving safety.

[0134] According to the second concept of the present application, a corresponding image denoising model is constructed based on the disturbance parameters under different scenarios, and a corresponding image association model is constructed based on the vehicle posture parameters under different scenarios, so that each on-board monitoring sub-module collects image data for complementary correction, thereby increasing the complete perception capability of the driving scene, timely warning of potential driving risks, and improving driving safety.

[0135] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and substitutions can be made without departing from the technical principles of the present application. These improvements and substitutions should also be regarded as the scope of protection of the present application.

Claims

1. A method for collecting image data for driving scenes, characterized in that: include: Establish a series of vehicle monitoring submodules and driving scenarios, and generate risk assessment models for each driving scenario; Obtain the real-time status parameters of the vehicle, and select the target risk assessment model according to the real-time status parameters of the vehicle; Process the image data collected by each vehicle monitoring submodule according to the target risk assessment model to determine whether to generate a risk warning instruction; The vehicle-mounted monitoring submodule sequence A, A = (a1, a2 ... a n ), where a i is the i-th vehicle monitoring submodule; n is the number of vehicle monitoring submodules.

2. The image data acquisition method for driving scenes according to claim 1, characterized in that: When establishing a driving scenario series, include: Extract multiple driving evaluation indicators based on historical driving parameters; According to all driving evaluation indicators, multiple initial scenarios are established, and an initial scenario sequence B is established, B = (b1, b2…b m1 ), where b i is the i-th initial scene, m1 is the number of initial scenes; The initial scenes are grouped and processed according to the preset similarity evaluation model; According to the collection processing results, a driving scene sequence C is established, C = (c1, c2...c m ), where c i is the i-th driving scenario, m is the number of driving scenarios, and m≤m1.

3. The image data acquisition method for driving scenes according to claim 2, characterized in that: When the initial scene is aggregated, it includes: Select the initial scene b1; Generate the initial scene b1 and the similarity evaluation value of each initial scene in the initial scene sequence B according to the preset similarity evaluation model; Establish a similar evaluation value sequence G, G = (g1, g2...g m1 ), g i is the similarity evaluation value between the initial scene b1 and the i-th initial scene; Preset similarity evaluation value threshold Δg, if g i >Δg, the i-th initial scene is grouped with the initial scene b1; Generate a single driving scenario based on the collected results; According to the remaining uncollected initial scenes, an initial scene sequence B1 is established, B1 = (b 11 ,b 12 …b 1m2 ), where b 1i is the i-th unclassified initial scene, m2 is the number of unclassified initial scenes, and m2 <m1; Select initial scene b 11; The above collection steps are iterated cyclically, and a driving scene sequence C is established according to the iteration results.

4. The image data acquisition method for driving scenes according to claim 3, characterized in that: When establishing a driving scenario series, it also includes: Acquire historical driving parameters of the vehicle according to a preset update cycle; Determining whether to generate an update instruction based on historical driving parameters of the vehicle; If an update instruction is generated, the similarity evaluation value threshold Δg is corrected according to the update instruction, and the driving scene sequence C is updated according to the correction result.

5. The image data acquisition method for driving scenes according to claim 3, characterized in that: When establishing the similarity evaluation value sequence G, it includes: Select the target initial scenes in turn; Among them, g is the similarity evaluation value between the initial scene b1 and the target initial scene; e1 is the preset first weight coefficient; e2 is the preset second weight coefficient; e3 is the preset third weight coefficient; Q1 is the preset first fixed coefficient; Q2 is the preset second fixed coefficient; Q3 is the preset third fixed coefficient; θ is the number of interference evaluation indicators; di is the interference evaluation value of the i-th interference evaluation indicator of the initial scene b1; d'i is the interference evaluation value of the i-th interference evaluation indicator of the target initial scene; αi is the influencing factor of the i-th interference evaluation indicator; U is the blind spot evaluation value set according to the visual blind spot overlap between the initial scene b1 and the target initial scene; ki is the difference in dependence of the i-th vehicle monitoring sub-module on the image collected by the i-th vehicle monitoring sub-module in the initial scene b1 and the target initial scene; βi is the influencing factor of the i-th vehicle monitoring sub-module.

6. The image data acquisition method for driving scenes according to claim 5, characterized in that: When generating risk assessment models for various driving scenarios, it includes: Select a target driving scene according to the driving scene sequence C, and generate a driving monitoring range of the target driving scene; The weight ratio of each vehicle monitoring submodule is set according to the blind spot range and the driving monitoring range; Obtain vehicle posture parameters of the target driving scene; Establish an image association model between various vehicle monitoring submodules according to vehicle posture parameters; Generate an image denoising model based on all interference evaluation indicators of the target driving scene; Generate a risk indicator series for the target driving scenario; Establish a risk assessment model for the target driving scene based on the image association model and risk index series of the target driving scene; Establish risk assessment model series J, J = (j1, j2…j m ), where j i is the risk assessment model for the i-th driving scenario.

7. The image data acquisition method for driving scenes according to claim 6, characterized in that: When selecting a target risk assessment model based on the real-time status parameters of the vehicle, it includes: Extract scene evaluation indicators based on real-time vehicle status parameters; The compatibility between the vehicle and each driving scenario is generated based on all scenario evaluation indicators, and a compatibility series K is established, K = (k1, k2…k m ), where k i is the compatibility between the vehicle and the i-th driving scenario; Select the maximum value k in the fit degree sequence K max The risk assessment model corresponding to the driving scenario is the target risk assessment model.

8. The image data acquisition method for driving scenes according to claim 7, characterized in that: When judging whether to generate an early warning instruction, it includes: Obtain the real-time image data collected by each vehicle monitoring submodule and establish an image sequence T, T = (t1, t2…t n ), where t i The real-time image collected by the i-th vehicle monitoring submodule; Preprocess all real-time images according to the image denoising model corresponding to the target risk model; Extract risk factors from each image according to the target risk assessment model, and generate initial risk assessment values ​​for each image; Process all real-time images according to the image association model and correct the initial risk assessment value of each image; According to the correction results, a risk assessment value series F is established, F = (f1, f2…f n ), where f i is the risk assessment value corresponding to the real-time image of the i-th vehicle monitoring submodule; Determine whether to generate an early warning instruction based on the risk assessment value sequence F.

9. The image data acquisition method for driving scenes according to claim 8, characterized in that: When judging whether to generate an early warning instruction according to the risk assessment value sequence F, it includes: According to the driving scenario corresponding to the target risk assessment model, a first risk assessment value threshold F1 and a second risk assessment value threshold F2 are set; and F1 <F2; If f i <F1, no warning command is generated; If F1 < f i < F2, generate a first-level warning instruction for the i-th vehicle-mounted monitoring sub-module; If f i >F2, generate the secondary warning instruction of the i-th vehicle monitoring submodule; Get the total number p of first-level warning instructions and second-level warning instructions; If the total number p is greater than the preset first number threshold, a third-level warning instruction is generated.

Citation Information

Patent Citations

  • Anti-collision early warning system and method based on 360-degree look-around input

    CN108638999A

  • Automatic driving test scene evaluation method based on perceptual defects

    CN110020797A

  • Application scene modeling method and modeling system for automatic driving

    CN115374550A

  • Adaptive generation and evaluation of autonomous vehicle key scenarios

    CN116490858A

  • Test evaluation method and device for automatic driving scene and vehicle

    CN117633426A

Cited By

  • Infrared temperature measurement observation system and control method thereof

    CN119688075A