Driving environment sensing method for autonomous vehicle

Abstract

The invention discloses a driving environment sensing method for an automatic driving vehicle, and particularly relates to the technical field of automatic driving vehicles. Signal-to-noise ratio abnormal information of laser signals received by a laser radar, processing complexity information of images shot by a camera and fusion information of a data fusion engine are acquired; according to the method, an emergency braking misjudgment risk assessment model is constructed, the misjudgment risk possibly existing in the current emergency braking process is quantified, a reliable reference basis is provided for automatic driving control of the vehicle, potential risks are predicted in advance, misjudgment caused by sensing system abnormity or inaccurate data fusion is prevented, the overall safety of automatic driving is improved, and the safety of the vehicle is improved. When the potential misjudgment risk exists in emergency braking of the automatic driving vehicle, potential misjudgment risk fluctuation information and tire attachment information existing in emergency braking of the automatic driving vehicle are obtained, a BP neural network is constructed to conduct intelligent regulation and control on the current emergency braking behavior, and accidents caused by braking failure or inconsistency are effectively reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of autonomous driving vehicles, and more specifically, to a method for perceiving a driving environment of an autonomous driving vehicle. Background Art

[0002] With the rapid development of autonomous driving technology, the accuracy and reliability of environmental perception systems have become core elements to ensure safe driving of vehicles. These systems are usually composed of multiple modules such as lidar, cameras and sensor fusion engines to achieve comprehensive perception of the surrounding environment. By emitting laser beams and measuring the returned reflected signals, lidar can accurately construct three-dimensional environmental maps and operate stably under a variety of lighting conditions, especially in low-light environments. In contrast, the image recognition module uses cameras and deep learning algorithms to identify and classify surrounding objects, including pedestrians, traffic signs and other vehicles; the sensor fusion engine fuses data from the lidar and image recognition modules to generate comprehensive environmental perception information, identify obstacles, lane markings, road conditions and other vehicles;

[0003] Although environmental perception systems have advanced technical features, they still face many challenges in practical applications. Limitations in sensor accuracy, interference from bad weather on signal transmission, and the impact of dynamic environmental changes on algorithm performance may all weaken the effectiveness of the system. In emergency braking scenarios, the lack of intelligent perception of potential operational hazards of lidar and cameras and fusion hazards of the data fusion process before each emergency braking may significantly increase the risk of misjudgment in emergency braking. If there is no systematic solution to these potential hazards in the future, misjudgment may occur, posing a serious threat to vehicle safety. Therefore, strengthening the intelligent monitoring and risk assessment mechanism of the environmental perception system has become an urgent need to improve the safety of autonomous driving. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a driving environment perception method for an autonomous driving vehicle to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A method for sensing a driving environment of an autonomous vehicle comprises the following steps:

[0007] Step S1, obtaining signal-to-noise ratio abnormality information of laser radar receiving laser signals, image processing complexity information of camera images, and fusion information of data fusion engine;

[0008] Step S2, constructing an emergency braking misjudgment risk assessment model, generating an emergency braking misjudgment risk assessment index, and assessing the potential misjudgment risk level of emergency braking of the autonomous driving vehicle;

[0009] Step S3, comparing the emergency braking misjudgment risk assessment index with a preset emergency braking misjudgment risk assessment index threshold to determine whether there is a potential misjudgment risk in emergency braking of the autonomous driving vehicle;

[0010] Step S4, obtaining the potential misjudgment risk fluctuation information of the emergency braking of the autonomous driving vehicle and the vehicle tire adhesion information, and constructing a BP neural network to effectively control the current emergency braking behavior.

[0011] In a preferred embodiment, the signal-to-noise ratio abnormality information of the laser signal received by the laser radar includes a signal-to-noise ratio abnormality coefficient, the image processing complexity information captured by the camera includes an image blur coefficient, and the fusion information of the data fusion engine includes a fusion degree coefficient.

[0012] In a preferred embodiment, the logic for obtaining the signal-to-noise ratio anomaly coefficient is as follows:

[0013] Get the laser signal power received by the laser radar and calculate the laser signal intensity Pr. The expression is as follows Where Preceived represents the laser signal power received by the lidar, and Preference represents the preset reference power;

[0014] Get the background noise power measured when the laser radar does not emit laser, and calculate the ambient noise intensity Nl. The expression is as follows Where Pnoise represents the background noise power measured when no laser is emitted;

[0015] Calculate the signal-to-noise ratio Snr, the expression is as follows

[0016] Calculate the average signal-to-noise ratio Snrbase, the expression is as follows Where Snr n represents the laser signal power received by the laser radar for the nth time, n = {1, 2, 3, ..., N}, N is a positive integer;

[0017] The signal-to-noise ratio calculated each time the laser radar receives a laser signal is compared with the average signal-to-noise ratio. When the signal-to-noise ratio calculated when the laser radar receives a laser signal is greater than the average signal-to-noise ratio, the signal-to-noise ratio greater than the average signal-to-noise ratio is marked as an abnormal signal-to-noise ratio.

[0018] Calculate the signal-to-noise ratio anomaly coefficient Xzby, the expression is as follows Where Snry irepresents the signal-to-noise ratio marked as abnormal signal-to-noise ratio for the i-th time, i = {1, 2, 3, ..., I}, and I is a positive integer.

[0019] In a preferred embodiment, the logic for obtaining the image blur coefficient is as follows:

[0020] Get the original image captured by the camera, and convert the original image into a grayscale image Igray(x,y), where x and y represent the horizontal and vertical coordinates of the pixel in the grayscale image, and Igray(x,y) represents the pixel value at position (x,y) in the grayscale image;

[0021] Calculate the difference between each pixel in the grayscale image and the surrounding pixel values ​​to obtain the edge intensity image. The expression is as follows: L(x,y) = |Igray(x+1,y)+Igray(x-1,y)+Igray(x,y+1)+Igray(x,y-1)-4*Igray(x,y)|, where L(x,y) represents the pixel value at position (x,y) in the edge intensity image, Igray(x+1,y) represents the pixel value at position (x+1,y) in the grayscale image, Igray(x-1,y) represents the pixel value at position (x-1,y) in the grayscale image, Igray(x,y+1) represents the pixel value at position (x,y+1) in the grayscale image, and Igray(x,y-1) represents the pixel value at position (x,y-1) in the grayscale image;

[0022] Calculate the pixel mean Xsj of the edge strength image, the expression is as follows Where K represents the total number of pixels in the edge intensity image, and X and Y represent the width and height of the edge intensity image, respectively;

[0023] Calculate the image blur coefficient Xsdm, the expression is as follows

[0024] In a preferred embodiment, the logic for obtaining the fusion degree coefficient is as follows:

[0025] Get the fusion success rate Pfusion of data fusion, the expression is as follows Where Nsuccess represents the number of objects that the data fusion engine has identified based on the data collected by the lidar and the camera, and Ntotal represents the total number of data fusions.

[0026] The Euclidean distance calculation method is used to calculate the distance error Jfusion between the same objects identified together;

[0027] Calculate the standard deviation of the distance error Bfusion, the expression is as follows Jfusionb represents the distance error between the bth commonly identified objects, b = {1, 2, ..., Nsuccess}, Represents the average value of the distance error, and the expression is as follows

[0028] Calculate the fusion quality factor Qfusion, the expression is as follows

[0029] Calculate the fusion coefficient Rhdx, the expression is as follows: Rhdx = Pfusion*Qfusion.

[0030] In a preferred embodiment, an emergency braking misjudgment risk assessment model is constructed according to the signal-to-noise ratio abnormality coefficient, the image blur coefficient, and the fusion coefficient, and an emergency braking misjudgment risk assessment index RiskIndex is generated according to the following formula: RiskIndex = ω1*Xzby+ω2*Xsdm-ω2*Rhdx, where ω1, ω2, and ω2 represent preset proportional coefficients of the signal-to-noise ratio abnormality coefficient, the image blur coefficient, and the fusion coefficient, respectively, and ω1, ω2, and ω2 are all greater than 0.

[0031] In a preferred embodiment, the emergency braking misjudgment risk assessment index is compared with a preset emergency braking misjudgment risk assessment index threshold to determine whether there is a potential misjudgment risk in the emergency braking of the autonomous driving vehicle, as follows:

[0032] If the emergency braking misjudgment risk assessment index is greater than the emergency braking misjudgment risk assessment index threshold, a misjudgment risk signal is generated, and the existing emergency braking strategy needs to be improved; if the emergency braking misjudgment risk assessment index is less than or equal to the emergency braking misjudgment risk assessment index threshold, a braking stability signal is generated.

[0033] In a preferred embodiment, the potential misjudgment risk fluctuation information of emergency braking of the autonomous driving vehicle includes the misjudgment risk difference fluctuation coefficient, and the vehicle tire adhesion information includes the tire adhesion difference coefficient.

[0034] In a preferred embodiment, the logic for obtaining the misjudgment risk difference fluctuation coefficient is as follows:

[0035] When a misjudgment risk signal is generated, the difference wpc between the emergency braking misjudgment risk assessment index and the emergency braking misjudgment risk assessment index threshold is calculated, and the expression is as follows: wpc = RiskIndex - RiskIndex z , where RiskIndex z Indicates the preset emergency braking misjudgment risk assessment index threshold;

[0036] Calculate the standard deviation Bwpc of the difference between the emergency braking misjudgment risk assessment index and the emergency braking misjudgment risk assessment index threshold, and the expression is as follows: Where wpc m represents the difference between the emergency braking misjudgment risk assessment index and the emergency braking misjudgment risk assessment index threshold when the misjudgment risk signal is generated for the mth time, m = {1, 2, ..., M}, M is a positive integer, It represents the average value of the difference between the emergency braking misjudgment risk assessment index and the emergency braking misjudgment risk assessment index threshold, and the expression is as follows

[0037] Calculate the fluctuation coefficient of misjudgment risk difference Wpxd, the expression is as follows: Wpxd = ln(Bwpc + 1);

[0038] The logic for obtaining the tire adhesion difference coefficient is as follows:

[0039] Randomly number the tires of the autonomous driving vehicle, and mark the tire numbers of the autonomous driving vehicle as a={1,2,3,...,A}, where a represents the tire number of the autonomous driving vehicle, and A is the total number of tires of the autonomous driving vehicle;

[0040] Calculate the adhesion coefficient F of the ath tire a , the expression is as follows Where Ftraction a Indicates the traction of the ath tire, Fnormal a represents the normal force of the ath tire;

[0041] Calculate the average Favg of the tire adhesion coefficient of the autonomous driving vehicle. The expression is as follows

[0042] Calculate the tire adhesion difference coefficient Ltfc, the expression is as follows Ltfc = max(|Fzhuo a -Favg|), where max() represents the maximum value acquisition function.

[0043] In a preferred embodiment, the obtained misjudgment risk difference fluctuation coefficient and tire adhesion difference coefficient are normalized, and the normalized misjudgment risk difference fluctuation coefficient and tire adhesion difference coefficient are combined into a feature vector TZ=[Wpxd, Favg], a BP neural network is constructed, the existing emergency braking strategy is improved, and an emergency braking control signal is output.

[0044] The BP neural network includes an input layer, a hidden layer, and an output layer. The output layer has two neurons, corresponding to the misjudgment risk difference fluctuation coefficient and the tire adhesion difference coefficient. The obtained misjudgment risk difference fluctuation coefficient and the tire adhesion difference coefficient are used as inputs. The number of neurons in the hidden layer is set to SL = 2*tz+1, where SL represents the number of neurons in the hidden layer, and tz represents the number of input features. The output layer has one neuron, which is used to output the emergency braking control signal.

[0045] The obtained misjudgment risk difference fluctuation coefficient and tire adhesion difference coefficient are passed as input to the hidden layer through the input layer. Each neuron in the hidden layer receives the input vector and calculates the weighted sum of each neuron. c =qz 1j *Wpxd+qz 2j *Favg+p c , where jq c represents the weighted sum of the cth neuron, qz 1j represents the weight of the input misjudgment risk difference fluctuation coefficient to the cth neuron in the hidden layer, qz 2j represents the weight of the input tire adhesion difference coefficient to the cth neuron in the hidden layer, p c Represents the bias term to the cth neuron in the hidden layer. The hidden layer outputs the calculated weighted sum through the Sigmoid activation function sc c Represents the output of the cth neuron in the hidden layer. The output layer receives the output of the hidden layer and generates the final emergency braking control signal And by calculating the gradient of the loss function, the BP neural network parameters are continuously adjusted and optimized, so that the value of the loss function is gradually reduced, and the construction of the BP neural network is completed;

[0046] By constructing the BP neural network, the existing emergency braking strategy is improved and the emergency braking control signal is output.

[0047] Technical effects and advantages of the present invention:

[0048] 1. The present invention constructs an emergency braking misjudgment risk assessment model by acquiring abnormal signal-to-noise ratio information of laser signals received by the laser radar, image processing complexity information captured by the camera, and fusion information of the data fusion engine, quantifies the misjudgment risks that may exist in the current emergency braking process, provides a reliable reference for the vehicle's automatic driving control, predicts potential risks in advance, prevents misjudgments caused by abnormal perception systems or inaccurate data fusion, and improves the overall safety of automatic driving. When there is a potential misjudgment risk in emergency braking of an automatic driving vehicle, the present invention obtains potential misjudgment risk fluctuation information and tire adhesion information in emergency braking of the automatic driving vehicle, constructs a BP neural network to intelligently regulate the current emergency braking behavior, adapts to various road conditions and driving scenarios, and controls the vehicle in real time to avoid unsafe braking behaviors caused by changes in tire adhesion or fluctuations in misjudgment risks, and effectively reduces accidents caused by braking failure or inconsistency. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to facilitate understanding by those skilled in the art, the present invention is further described below in conjunction with the accompanying drawings;

[0050] Figure 1 Schematic diagram of the structure of the method of the embodiment of the present invention. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0052] Example: Figure 1 The present invention provides a method for sensing the driving environment of an autonomous driving vehicle, comprising the following steps:

[0053] Step S1, obtaining signal-to-noise ratio abnormality information of laser radar receiving laser signals, image processing complexity information of camera images, and fusion information of data fusion engine;

[0054] Step S2, constructing an emergency braking misjudgment risk assessment model, generating an emergency braking misjudgment risk assessment index, and assessing the potential misjudgment risk level of emergency braking of the autonomous driving vehicle;

[0055] Step S3, comparing the emergency braking misjudgment risk assessment index with a preset emergency braking misjudgment risk assessment index threshold to determine whether there is a potential misjudgment risk in emergency braking of the autonomous driving vehicle;

[0056] Step S4, obtaining the potential misjudgment risk fluctuation information of the emergency braking of the autonomous driving vehicle and the vehicle tire adhesion information, and constructing a BP neural network to effectively control the current emergency braking behavior;

[0057] Step S1, obtaining signal-to-noise ratio abnormality information of laser radar receiving laser signals, image processing complexity information of camera images, and fusion information of data fusion engine;

[0058] The signal-to-noise ratio abnormality information of the laser radar receiving the laser signal includes the signal-to-noise ratio abnormality coefficient, the image processing complexity information captured by the camera includes the image fuzziness coefficient, and the fusion information of the data fusion engine includes the fusion degree coefficient;

[0059] The signal-to-noise ratio anomaly coefficient is used to measure the degree of anomaly between the ratio of the laser signal and noise received by the laser radar on the autonomous driving vehicle deviates from the normal value. The larger the signal-to-noise ratio anomaly coefficient, the worse the quality of the laser signal received by the laser radar, and the data may be interfered by more noise. In this case, the laser radar may not be able to accurately detect the distance and shape of surrounding objects, especially in complex environments, and missed detection or false detection may occur. The obstacle information detected by the laser radar is inaccurate, which may cause the autonomous driving system to mistakenly believe that there is an obstacle ahead, thereby triggering unnecessary emergency braking, significantly increasing the risk of misjudgment of the environment, and leading to erroneous emergency braking decisions. On the contrary, when the signal-to-noise ratio anomaly coefficient is small, the system's perception of the environment is more accurate, the risk of misjudgment is significantly reduced, and the decision on emergency braking is more robust. By calculating the signal-to-noise ratio anomaly coefficient, the operating hazards of the laser radar can be effectively identified, and data enhancement or compensation measures can be taken in a timely manner to improve the overall reliability of the perception system.

[0060] The logic for obtaining the signal-to-noise ratio anomaly coefficient is as follows:

[0061] Get the laser signal power received by the laser radar and calculate the laser signal intensity Pr. The expression is as follows Where Preceived represents the laser signal power received by the lidar, and Preference represents the preset reference power;

[0062] It should be noted that the laser signal power can be obtained by connecting the laser radar with a power meter, oscilloscope and other measuring equipment in real time to obtain the laser signal power data received by the laser radar. The connected measuring equipment can also be used to measure the background noise power. The reference power can be preset by a technician in this field according to the technical specifications, industry standards, environment and other factors of the laser radar. A reasonable reference power setting will help improve the accuracy of the signal-to-noise ratio, thereby enhancing the reliability and safety of the system.

[0063] Get the background noise power measured when the laser radar does not emit laser, and calculate the ambient noise intensity Nl. The expression is as follows Where Pnoise represents the background noise power measured when no laser is emitted;

[0064] Calculate the signal-to-noise ratio Snr, the expression is as follows

[0065] Calculate the average signal-to-noise ratio Snrbase, the expression is as follows Where Snr n represents the laser signal power received by the laser radar for the nth time, n = {1, 2, 3, ..., N}, N is a positive integer;

[0066] The signal-to-noise ratio calculated each time the laser radar receives a laser signal is compared with the average signal-to-noise ratio. When the signal-to-noise ratio calculated when the laser radar receives a laser signal is greater than the average signal-to-noise ratio, the signal-to-noise ratio greater than the average signal-to-noise ratio is marked as an abnormal signal-to-noise ratio.

[0067] Calculate the signal-to-noise ratio anomaly coefficient Xzby, the expression is as follows Where Snry i represents the signal-to-noise ratio marked as abnormal signal-to-noise ratio for the i-th time, i = {1, 2, 3, ..., I}, I is a positive integer;

[0068] The image blur coefficient is used to quantify the clarity of the image captured by the camera and to evaluate the impact of the image blur on the processing complexity. The larger the image blur coefficient, the blurrier the image and the higher the processing complexity. More image preprocessing steps may be required. An increase in the image blur coefficient may lead to inaccurate object recognition, decision delays, and information loss, thereby increasing the risk of emergency braking misjudgment. Therefore, in the autonomous driving system, it is very important to monitor and control the image blur to ensure safe driving and system reliability. The blur of the image captured by the camera is analyzed and the image blur coefficient is calculated so that necessary adjustments can be made during the recognition and decision-making process.

[0069] The logic for obtaining the image blur coefficient is as follows:

[0070] Get the original image captured by the camera, and convert the original image into a grayscale image Igray(x,y), where x and y represent the horizontal and vertical coordinates of the pixel in the grayscale image, and Igray(x,y) represents the pixel value at position (x,y) in the grayscale image;

[0071] In an optional example, the original image may be converted into a grayscale image by the following formula: Igray(x,y)=0.2989*R(x,y)+0.5870*G(x,y)+0.1140*B(x,y), where R, G, and B represent pixel values ​​of red, green, and blue channels, respectively;

[0072] Calculate the difference between each pixel in the grayscale image and the surrounding pixel values ​​to obtain the edge intensity image. The expression is as follows: L(x,y) = |Igray(x+1,y)+Igray(x-1,y)+Igray(x,y+1)+Igray(x,y-1)-4*Igray(x,y)|, where L(x,y) represents the pixel value at position (x,y) in the edge intensity image, Igray(x+1,y) represents the pixel value at position (x+1,y) in the grayscale image, Igray(x-1,y) represents the pixel value at position (x-1,y) in the grayscale image, Igray(x,y+1) represents the pixel value at position (x,y+1) in the grayscale image, and Igray(x,y-1) represents the pixel value at position (x,y-1) in the grayscale image;

[0073] Calculate the pixel mean Xsj of the edge strength image, the expression is as follows Where K represents the total number of pixels in the edge intensity image, and X and Y represent the width and height of the edge intensity image, respectively;

[0074] Calculate the image blur coefficient Xsdm, the expression is as follows

[0075] The fusion coefficient is used to evaluate the accuracy and consistency of the laser radar and camera data during the fusion process, and to measure the success rate of data fusion. The higher the fusion coefficient, the better the data fusion effect, the more consistent the data between sensors, the smaller the error, and the more accurate the final environmental perception effect. The smaller the fusion coefficient, the worse the effect of laser radar and camera data fusion, the greater the data inconsistency between sensors, and the higher the data error, which will lead to an increased risk of misjudgment during emergency braking. On the contrary, the larger the fusion coefficient, the better the data fusion effect, the more consistent the sensor information, and the lower the risk of misjudgment during emergency braking.

[0076] The logic for obtaining the fusion coefficient is as follows:

[0077] Get the fusion success rate Pfusion of data fusion, the expression is as follows Where Nsuccess represents the number of objects that the data fusion engine has identified based on the data collected by the lidar and the camera, and Ntotal represents the total number of data fusions.

[0078] The Euclidean distance calculation method is used to calculate the distance error Jfusion between the same objects identified together;

[0079] In an optional example, the Euclidean distance calculation method calculates the distance error Jfusion between the same objects that are jointly identified. The calculation expression is as follows: Where (x lidar ,y lidar ,z lidar ) represents the three-dimensional coordinates of the object in the data collected by the lidar, (x camera ,y camera ,z camera ) represents the three-dimensional coordinates of the object in the data collected by the camera;

[0080] It should be noted that since the data collected by the camera is usually image data, it is necessary to convert the two-dimensional coordinates of the object in the image into three-dimensional space coordinates by using image processing software or image processing library. Commonly used image processing software or image processing libraries include OpenCV, MATLAB, 3DReconstruction Software, etc. Before calculating the distance error, the camera data and lidar data need to be projected into the same space coordinate system to make the two types of data have spatial consistency;

[0081] Calculate the standard deviation of the distance error Bfusion, the expression is as follows Jfusion b represents the distance error between the bth commonly identified objects, b = {1, 2, ..., Nsuccess}, Represents the average value of the distance error, and the expression is as follows

[0082] Calculate the fusion quality factor Qfusion, the expression is as follows

[0083] Calculate the fusion coefficient Rhdx, the expression is as follows Rhdx = Pfusion * Qfusion;

[0084] Step S2, constructing an emergency braking misjudgment risk assessment model, generating an emergency braking misjudgment risk assessment index, and assessing the potential misjudgment risk level of emergency braking of the autonomous driving vehicle;

[0085] According to the signal-to-noise ratio abnormality coefficient, image blur coefficient, and fusion coefficient, an emergency braking misjudgment risk assessment model is constructed to generate an emergency braking misjudgment risk assessment index RiskIndex, based on the following formula: RiskIndex = ω1*Xzby+ω2*Xsdm-ω2*Rhdx, where ω1, ω2, and ω2 represent the preset proportional coefficients of the signal-to-noise ratio abnormality coefficient, the image blur coefficient, and the fusion coefficient, respectively, and ω1, ω2, and ω2 are all greater than 0;

[0086] It should be noted that before constructing the emergency braking misjudgment risk assessment model, the signal-to-noise ratio abnormality coefficient, image blur coefficient, and fusion coefficient need to be normalized. Common normalization methods include Min-Max normalization, Z-Score standardization, etc. ω1, ω2, and ω2 are set according to actual conditions. For example, the expert empowerment method is adopted, that is, experts in related fields are invited to determine the preset proportional coefficients of various indicators through professional opinion surveys and comprehensive evaluations;

[0087] It can be seen from the above calculation expression that the larger the signal-to-noise ratio abnormality coefficient, the larger the image blur coefficient, and the smaller the fusion coefficient, the larger the emergency braking misjudgment risk assessment index, indicating that the potential misjudgment risk of emergency braking of the autonomous driving vehicle is greater. On the contrary, the smaller the signal-to-noise ratio abnormality coefficient, the smaller the image blur coefficient, and the larger the fusion coefficient, the smaller the emergency braking misjudgment risk assessment index, indicating that the potential misjudgment risk of emergency braking of the autonomous driving vehicle is less severe.

[0088] Step S3, comparing the emergency braking misjudgment risk assessment index with a preset emergency braking misjudgment risk assessment index threshold, to determine whether there is a potential misjudgment risk in the emergency braking of the autonomous driving vehicle, as follows:

[0089] If the emergency braking misjudgment risk assessment index is greater than the emergency braking misjudgment risk assessment index threshold, it indicates that there is a potential misjudgment risk in the emergency braking of the autonomous driving vehicle, and a misjudgment risk signal is generated, and the existing emergency braking strategy needs to be improved; if the emergency braking misjudgment risk assessment index is less than or equal to the emergency braking misjudgment risk assessment index threshold, it indicates that the potential misjudgment risk of the emergency braking of the autonomous driving vehicle is not perceived, and a braking stability signal is generated, and the emergency braking behavior of the autonomous driving vehicle can be implemented according to the existing emergency braking strategy;

[0090] Step S4, obtaining the potential misjudgment risk fluctuation information of the emergency braking of the autonomous driving vehicle and the vehicle tire adhesion information, and constructing a BP neural network to effectively control the current emergency braking behavior;

[0091] The potential misjudgment risk fluctuation information of the automatic driving vehicle emergency braking includes the misjudgment risk difference fluctuation coefficient, and the vehicle tire adhesion information includes the tire adhesion difference coefficient;

[0092] The misjudgment risk difference fluctuation coefficient is used to measure the risk depth of potential misjudgment risk of emergency braking of the autonomous driving vehicle each time the autonomous driving vehicle performs emergency braking;

[0093] The logic for obtaining the volatility coefficient of misjudgment risk difference is as follows:

[0094] When a misjudgment risk signal is generated, the difference wpc between the emergency braking misjudgment risk assessment index and the emergency braking misjudgment risk assessment index threshold is calculated, and the expression is as follows: wpc = RiskIndex - RiskIndex z , where RiskIndex z Indicates the preset emergency braking misjudgment risk assessment index threshold;

[0095] Calculate the standard deviation Bwpc of the difference between the emergency braking misjudgment risk assessment index and the emergency braking misjudgment risk assessment index threshold, and the expression is as follows: Where wpc m represents the difference between the emergency braking misjudgment risk assessment index and the emergency braking misjudgment risk assessment index threshold when the misjudgment risk signal is generated for the mth time, m = {1, 2, ..., M}, M is a positive integer, It represents the average value of the difference between the emergency braking misjudgment risk assessment index and the emergency braking misjudgment risk assessment index threshold, and the expression is as follows

[0096] Calculate the fluctuation coefficient of misjudgment risk difference Wpxd, the expression is as follows: Wpxd = ln(Bwpc + 1);

[0097] The tire adhesion difference coefficient is used to quantify the difference in adhesion between different tires and reflect the degree of change in the current tire adhesion;

[0098] The logic for obtaining the tire adhesion difference coefficient is as follows:

[0099] Randomly number the tires of the autonomous driving vehicle, and mark the tire numbers of the autonomous driving vehicle as a={1,2,3,...,A}, where a represents the tire number of the autonomous driving vehicle, and A is the total number of tires of the autonomous driving vehicle;

[0100] Calculate the adhesion coefficient F of the ath tire a , the expression is as follows Where Ftraction a Indicates the traction of the ath tire, Fnormal a represents the normal force of the ath tire;

[0101] It should be noted that the traction of the ath tire isa A force sensor can be installed on the axle or suspension system of the vehicle to measure the traction of the tire under different conditions. The normal force of the a-th tire can be obtained in real time by installing a tire pressure sensor or a strain gauge sensor on the axle or suspension of the autonomous driving vehicle to obtain the normal force of the tire of the autonomous driving vehicle in different motion states.

[0102] Calculate the average Favg of the tire adhesion coefficient of the autonomous driving vehicle. The expression is as follows

[0103] Calculate the tire adhesion difference coefficient Ltfc, the expression is as follows Ltfc = max(|Fzhuo a -Favg|), where max() represents the maximum value acquisition function;

[0104] The obtained misjudgment risk difference fluctuation coefficient and tire adhesion difference coefficient are normalized, and the normalized misjudgment risk difference fluctuation coefficient and tire adhesion difference coefficient are combined into a feature vector TZ=[Wpxd,Favg], a BP neural network is constructed, the existing emergency braking strategy is improved, and the emergency braking control signal is output.

[0105] The BP neural network includes an input layer, a hidden layer, and an output layer. The output layer has two neurons, corresponding to the misjudgment risk difference fluctuation coefficient and the tire adhesion difference coefficient. The obtained misjudgment risk difference fluctuation coefficient and the tire adhesion difference coefficient are used as inputs. The number of neurons in the hidden layer is set to SL = 2*tz+1, where SL represents the number of neurons in the hidden layer, and tz represents the number of input features. The output layer has one neuron, which is used to output the emergency braking control signal.

[0106] The obtained misjudgment risk difference fluctuation coefficient and tire adhesion difference coefficient are passed as input to the hidden layer through the input layer. Each neuron in the hidden layer receives the input vector and calculates the weighted sum of each neuron. c =qz 1j *Wpxd+qz 2j *Favg+p c , where jq c represents the weighted sum of the cth neuron, qz 1j represents the weight of the input misjudgment risk difference fluctuation coefficient to the cth neuron in the hidden layer, qz 2j represents the weight of the input tire adhesion difference coefficient to the cth neuron in the hidden layer, p c Represents the bias term to the cth neuron in the hidden layer. The hidden layer outputs the calculated weighted sum through the Sigmoid activation function scc Represents the output of the cth neuron in the hidden layer. The output layer receives the output of the hidden layer and generates the final emergency braking control signal And by calculating the gradient of the loss function, the BP neural network parameters are continuously adjusted and optimized, so that the value of the loss function is gradually reduced, and the construction of the BP neural network is completed;

[0107] By constructing the BP neural network, the existing emergency braking strategy is improved and the emergency braking control signal is output;

[0108] The present invention constructs an emergency braking misjudgment risk assessment model by acquiring abnormal signal-to-noise ratio information of laser signals received by the laser radar, image processing complexity information captured by the camera, and fusion information of the data fusion engine, quantifies the misjudgment risks that may exist in the current emergency braking process, provides a reliable reference basis for the vehicle's automatic driving control, predicts potential risks in advance, prevents misjudgments caused by abnormal perception systems or inaccurate data fusion, and improves the overall safety of automatic driving. When there is a potential misjudgment risk in the emergency braking of the automatic driving vehicle, the potential misjudgment risk fluctuation information and tire adhesion information of the emergency braking of the automatic driving vehicle are obtained, and a BP neural network is constructed to perform intelligent regulation and control of the current emergency braking behavior, adapt to various road conditions and driving scenarios, and control the vehicle in real time to avoid unsafe braking behaviors caused by changes in tire adhesion or fluctuations in misjudgment risks, and effectively reduce accidents caused by braking failure or inconsistency.

[0109] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0110] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0111] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for sensing the driving environment of an autonomous driving vehicle, characterized in that: The steps include: Step S1, obtaining signal-to-noise ratio abnormality information of laser radar receiving laser signals, image processing complexity information of camera images, and fusion information of data fusion engine; Step S2, constructing an emergency braking misjudgment risk assessment model, generating an emergency braking misjudgment risk assessment index, and assessing the potential misjudgment risk level of emergency braking of the autonomous driving vehicle; Step S3, comparing the emergency braking misjudgment risk assessment index with a preset emergency braking misjudgment risk assessment index threshold to determine whether there is a potential misjudgment risk in emergency braking of the autonomous driving vehicle; Step S4, obtaining the potential misjudgment risk fluctuation information of the emergency braking of the autonomous driving vehicle and the vehicle tire adhesion information, and constructing a BP neural network to effectively control the current emergency braking behavior.

2. The method for sensing the driving environment of an autonomous driving vehicle according to claim 1, characterized in that: The signal-to-noise ratio abnormality information of the laser signal received by the laser radar includes the signal-to-noise ratio abnormality coefficient, the image processing complexity information captured by the camera includes the image blur coefficient, and the fusion information of the data fusion engine includes the fusion degree coefficient.

3. The method for sensing the driving environment of an autonomous driving vehicle according to claim 2, characterized in that: The logic for obtaining the signal-to-noise ratio anomaly coefficient is as follows: Get the laser signal power received by the laser radar and calculate the laser signal intensity Pr. The expression is as follows Where Preceived represents the laser signal power received by the lidar, and Preference represents the preset reference power; Get the background noise power measured when the laser radar does not emit laser, and calculate the ambient noise intensity Nl. The expression is as follows Where Pnoise represents the background noise power measured when no laser is emitted; Calculate the signal-to-noise ratio Snr, the expression is as follows Calculate the average signal-to-noise ratio Snrbase, the expression is as follows Where Snr n represents the laser signal power received by the laser radar for the nth time, n = {1, 2, 3, ..., N}, N is a positive integer; The signal-to-noise ratio calculated each time the laser radar receives a laser signal is compared with the average signal-to-noise ratio. When the signal-to-noise ratio calculated when the laser radar receives a laser signal is greater than the average signal-to-noise ratio, the signal-to-noise ratio greater than the average signal-to-noise ratio is marked as an abnormal signal-to-noise ratio. Calculate the signal-to-noise ratio anomaly coefficient Xzby, the expression is as follows Where Snry i represents the signal-to-noise ratio marked as abnormal signal-to-noise ratio for the i-th time, i = {1, 2, 3, ..., I}, and I is a positive integer.

4. The method for sensing the driving environment of an autonomous driving vehicle according to claim 2, characterized in that: The logic for obtaining the image blur coefficient is as follows: Get the original image captured by the camera, and convert the original image into a grayscale image Igray(x,y), where x and y represent the horizontal and vertical coordinates of the pixel in the grayscale image, and Igray(x,y) represents the pixel value at position (x,y) in the grayscale image; Calculate the difference between each pixel in the grayscale image and the surrounding pixel values ​​to obtain the edge intensity image. The expression is as follows: L(x,y) = |Igray(x+1,y)+Igray(x-1,y)+Igray(x,y+1)+Igray(x,y-1)-4*Igray(x,y)|, where L(x,y) represents the pixel value at position (x,y) in the edge intensity image, Igray(x+1,y) represents the pixel value at position (x+1,y) in the grayscale image, Igray(x-1,y) represents the pixel value at position (x-1,y) in the grayscale image, Igray(x,y+1) represents the pixel value at position (x,y+1) in the grayscale image, and Igray(x,y-1) represents the pixel value at position (x,y-1) in the grayscale image; Calculate the pixel mean Xsj of the edge strength image, the expression is as follows Where K represents the total number of pixels in the edge intensity image, and X and Y represent the width and height of the edge intensity image, respectively; Calculate the image blur coefficient Xsdm, the expression is as follows 5. The method for sensing the driving environment of an autonomous driving vehicle according to claim 2, characterized in that: The logic for obtaining the fusion coefficient is as follows: Get the fusion success rate Pfusion of data fusion, the expression is as follows Where Nsuccess represents the number of objects that the data fusion engine has identified based on the data collected by the lidar and the camera, and Ntotal represents the total number of data fusions. The Euclidean distance calculation method is used to calculate the distance error Jfusion between the same objects identified together; Calculate the standard deviation of the distance error Bfusion, the expression is as follows Jfusion b represents the distance error between the b-th commonly recognized objects, b = {1, 2, ..., Nsuccess}, Represents the average value of the distance error, and the expression is as follows Calculate the fusion quality factor Qfusion, the expression is as follows Calculate the fusion coefficient Rhdx, the expression is as follows: Rhdx = Pfusion*Qfusion.

6. The method for sensing the driving environment of an autonomous driving vehicle according to claim 2, characterized in that: An emergency braking misjudgment risk assessment model is constructed according to the signal-to-noise ratio abnormality coefficient, image blur coefficient and fusion coefficient, and an emergency braking misjudgment risk assessment index RiskIndex is generated according to the following formula: RiskIndex = ω1*Xzby+ω2*Xsdm-ω2*Rhdx, where ω1, ω2 and ω2 represent the preset proportional coefficients of the signal-to-noise ratio abnormality coefficient, image blur coefficient and fusion coefficient, respectively, and ω1, ω2 and ω2 are all greater than 0.

7. The method for sensing the driving environment of an autonomous driving vehicle according to claim 6, characterized in that: The emergency braking misjudgment risk assessment index is compared with the preset emergency braking misjudgment risk assessment index threshold to determine whether there is a potential misjudgment risk in the emergency braking of the autonomous driving vehicle, as follows: If the emergency braking misjudgment risk assessment index is greater than the emergency braking misjudgment risk assessment index threshold, a misjudgment risk signal is generated, and the existing emergency braking strategy needs to be improved; if the emergency braking misjudgment risk assessment index is less than or equal to the emergency braking misjudgment risk assessment index threshold, a braking stability signal is generated.

8. The method for sensing the driving environment of an autonomous driving vehicle according to claim 1, characterized in that: The potential misjudgment risk fluctuation information of emergency braking of the autonomous driving vehicle includes the misjudgment risk difference fluctuation coefficient, and the vehicle tire adhesion information includes the tire adhesion difference coefficient.

9. The method for sensing the driving environment of an autonomous driving vehicle according to claim 8, characterized in that: The logic for obtaining the volatility coefficient of misjudgment risk difference is as follows: When a misjudgment risk signal is generated, the difference wpc between the emergency braking misjudgment risk assessment index and the emergency braking misjudgment risk assessment index threshold is calculated, and the expression is as follows: wpc = RiskIndex - RiskIndex z , where RiskIndex z Indicates the preset emergency braking misjudgment risk assessment index threshold; Calculate the standard deviation Bwpc of the difference between the emergency braking misjudgment risk assessment index and the emergency braking misjudgment risk assessment index threshold, and the expression is as follows: Where wpc m represents the difference between the emergency braking misjudgment risk assessment index and the emergency braking misjudgment risk assessment index threshold when the misjudgment risk signal is generated for the mth time, m = {1, 2, ..., M}, M is a positive integer, It represents the average value of the difference between the emergency braking misjudgment risk assessment index and the emergency braking misjudgment risk assessment index threshold, and the expression is as follows Calculate the fluctuation coefficient of misjudgment risk difference Wpxd, the expression is as follows: Wpxd = ln(Bwpc + 1); The logic for obtaining the tire adhesion difference coefficient is as follows: Randomly number the tires of the autonomous driving vehicle, and mark the tire numbers of the autonomous driving vehicle as a={1,2,3,...,A}, where a represents the tire number of the autonomous driving vehicle, and A is the total number of tires of the autonomous driving vehicle; Calculate the adhesion coefficient F of the ath tire a , the expression is as follows Where Ftraction a Indicates the traction of the ath tire, Fnormal a represents the normal force of the ath tire; Calculate the average Favg of the tire adhesion coefficient of the autonomous driving vehicle. The expression is as follows Calculate the tire adhesion difference coefficient Ltfc, the expression is as follows Ltfc = max(|Fzhuo a -Favg|), where max() represents the maximum value acquisition function.

10. The method for sensing the driving environment of an autonomous driving vehicle according to claim 9, characterized in that: The obtained misjudgment risk difference fluctuation coefficient and tire adhesion difference coefficient are normalized, and the normalized misjudgment risk difference fluctuation coefficient and tire adhesion difference coefficient are combined into a feature vector TZ=[Wpxd,Favg], a BP neural network is constructed, the existing emergency braking strategy is improved, and the emergency braking control signal is output. The BP neural network includes an input layer, a hidden layer, and an output layer. The output layer has two neurons, corresponding to the misjudgment risk difference fluctuation coefficient and the tire adhesion difference coefficient. The obtained misjudgment risk difference fluctuation coefficient and the tire adhesion difference coefficient are used as inputs. The number of neurons in the hidden layer is set to SL = 2*tz+1, where SL represents the number of neurons in the hidden layer, and tz represents the number of input features. The output layer has one neuron, which is used to output the emergency braking control signal. The obtained misjudgment risk difference fluctuation coefficient and tire adhesion difference coefficient are passed as input to the hidden layer through the input layer. Each neuron in the hidden layer receives the input vector and calculates the weighted sum of each neuron. c =qz 1j *Wpxd+qz 2j *Favg+p c , where jq c represents the weighted sum of the cth neuron, qz 1j represents the weight of the input misjudgment risk difference fluctuation coefficient to the cth neuron in the hidden layer, qz 2j represents the weight of the input tire adhesion difference coefficient to the cth neuron in the hidden layer, p c Represents the bias term to the cth neuron in the hidden layer. The hidden layer outputs the calculated weighted sum through the Sigmoid activation function sc c Represents the output of the cth neuron in the hidden layer. The output layer receives the output of the hidden layer and generates the final emergency braking control signal And by calculating the gradient of the loss function, the BP neural network parameters are continuously adjusted and optimized, so that the value of the loss function is gradually reduced, and the construction of the BP neural network is completed; By constructing the BP neural network, the existing emergency braking strategy is improved and the emergency braking control signal is output.

Images