An autonomous driving vehicle positioning system and method based on lane line recognition

Abstract

The present invention discloses a positioning system and method for an autonomous driving vehicle based on lane line recognition, including: a GPS unit, an IMU unit, a wheel odometer unit, a lane line detection unit, and an extended Kalman filter unit; the present invention enables the autonomous driving vehicle to determine the noise change amount of the inertial sensor according to the distance from itself to the lane line, and then correct its cumulative error, providing sufficient pre-positioning data for the decision-making, planning and control of the autonomous driving vehicle.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle positioning, and particularly relates to an autonomous driving vehicle positioning system and method based on lane line recognition. Background Art

[0002] An autonomous driving vehicle, also known as a driverless vehicle, a computer-driven vehicle, or a wheeled mobile robot, is a cutting-edge technology that relies on computer and artificial intelligence technologies to complete a complete, safe, and effective driving without human operation. As one of the most basic links in autonomous driving, positioning aims to clarify the absolute position of the vehicle relative to the global environment or the relative position relative to traffic participants, and its accuracy directly affects the safety of the autonomous driving system. Currently, autonomous driving technology is having an increasingly profound impact on our travel. However, in a complex real road environment, the positioning methods for autonomous driving are not yet mature enough.

[0003] Fusing the global satellite navigation system and inertial sensors is a relatively typical positioning method. On the one hand, the global satellite navigation system depends on external satellite signals and is easily interfered by weather, electromagnetic waves, etc., facing problems such as no signal in tunnels and multi-path effects; while inertial sensors do not depend on external signals and are not affected by the above factors. On the other hand, inertial sensors have cumulative errors, while the errors of GPS do not accumulate. By complementing each other, relatively accurate positioning can be achieved. The positioning devices used in this method are GPS, IMU, and wheel odometer, and the algorithms commonly used are Kalman filter algorithm, extended Kalman filter and its variants. This positioning method is widely used because the required devices are simple and the cost is low. However, the noises of IMU and wheel odometer are not fixed. When the vehicle is driving in a complex urban environment, due to the influence of the external environment, the noises of IMU and wheel odometer change frequently, and the complex urban environment is likely to cause GPS signal loss. Without the correction of GPS signals, it will lead to filter divergence and ultimately result in positioning failure.

[0004] Therefore, how to ensure that an autonomous driving vehicle can achieve accurate and stable self-positioning in a complex urban environment is an urgent problem to be solved in the current field of autonomous driving. Summary of the Invention

[0005] Aiming at the deficiencies of the above-mentioned existing technologies, the purpose of the present invention is to provide an autonomous driving vehicle positioning system and method based on lane line recognition, so as to solve the problem in the existing technology that due to the lack of in-vehicle GPS signals, the changing IMU noise caused by complex urban environments cannot be corrected, resulting in positioning failure. The present invention enables an autonomous driving vehicle to determine the change amount of the noise of the inertial sensor according to the distance from itself to the lane line, and then correct its cumulative error, providing sufficient pre-positioning data for the decision-making, planning and control of the autonomous driving vehicle.

[0006] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] An autonomous driving vehicle positioning system based on lane line recognition of the present invention includes: a GPS unit, an IMU unit, a wheel odometer unit, a lane line detection unit, and an extended Kalman filter unit;

[0008] The GPS unit is installed outside the vehicle and is used to obtain the real-time longitude and latitude positioning information and heading angle information of the vehicle, and send the obtained information to the extended Kalman filter unit in real time;

[0009] The IMU unit is installed inside the vehicle and is used to obtain the instantaneous acceleration information and instantaneous angular velocity information of the vehicle in real time, and send the obtained information to the extended Kalman filter unit in real time;

[0010] The wheel odometer unit is installed on the vehicle wheels and is used to obtain the instantaneous speed information of the vehicle in real time, and send the obtained information to the extended Kalman filter unit in real time;

[0011] The lane line detection unit is installed at the front windshield of the vehicle, obtains the road image information in front of the vehicle in real time, and identifies the lane line information around the vehicle;

[0012] The extended Kalman filter unit is installed inside the vehicle. When the GPS unit signal is available, it obtains the longitude and latitude positioning information and heading angle information sent by the GPS unit, the instantaneous angular velocity information and instantaneous acceleration information sent by the IMU unit, and the instantaneous speed information sent by the wheel odometer unit in real time, and then fuses the above information using the extended Kalman filtering algorithm to finally obtain the real-time positioning information of the vehicle; when the GPS unit signal is unavailable, it fuses the instantaneous angular velocity information and instantaneous acceleration information sent by the IMU unit and the instantaneous speed information sent by the wheel odometer unit using the extended Kalman filtering algorithm, and uses the lane line information sent by the lane line detection unit to eliminate the cumulative error caused by the sensor noise in the IMU unit and the wheel odometer unit.

[0013] Further, the lane line detection unit identifies the lane line information around the vehicle through a lane line recognition algorithm. The neural network models specifically used in the lane line recognition algorithm include two network models, LaneNet and H-Net. Among them, LaneNet is a multi-task model that combines semantic segmentation and vector representation of pixels, and is used for instance segmentation of lane lines in images. H-Net is a network model composed of convolutional layers and fully connected layers, and is used to predict the transformation matrix, and the transformation matrix is used to perform regression on pixel points belonging to the same lane line.

[0014] The present invention also provides an autonomous driving vehicle positioning method based on lane line recognition. Based on the above system, the steps are as follows:

[0015] 1) Obtain the longitude and latitude positioning information, heading angle information, instantaneous acceleration information and instantaneous angular velocity information of the vehicle itself, the instantaneous speed information of the vehicle, and the lane line information around the vehicle in real time;

[0016] 2) Determine whether the information received by the GPS unit is available: Decode the received GPS raw data to obtain the status information of the GPS. If the decoded status information is non-negative, the information received by the GPS unit is available, and go to step 3); if the decoded status information is negative, the information received by the GPS unit is unavailable, and go to step 4);

[0017] 3) Send the longitude and latitude positioning information, heading angle information, instantaneous acceleration information and instantaneous angular velocity information, and instantaneous speed information of the vehicle obtained in step 1) to the extended Kalman filter unit, and fuse the above information through the extended Kalman filter in the unit to obtain the real-time positioning information of the vehicle;

[0018] 4) According to the lane line information around the vehicle obtained, calculate the distance from the vehicle to the nearest lane line, calculate the error of the IMU in the current IMU unit in the direction perpendicular to the lane line through the lane line distance constraint algorithm, and then deduce the current noise values of the IMU in the IMU unit and the wheel odometer in the wheel odometer unit. Use the noise values to update the system noise variance matrix of the extended Kalman filter in the extended Kalman filter unit, and use the updated extended Kalman filter to fuse the instantaneous acceleration information and instantaneous angular velocity information of the vehicle sent by the IMU unit and the instantaneous speed information of the vehicle sent by the wheel odometer unit to obtain the real-time positioning information of the vehicle.

[0019] Further, the GPS decoding method used in step 2) is: Use the function package nmea_navsat_driver under the LINUX system to decode the GPS raw data.

[0020] Furthermore, step 3) specifically includes:

[0021] The specific algorithm of the extended Kalman filter used is as follows:

[0022] X n = f(X n-1 , u n )

[0023] Z n = h(X n )

[0024] In the formula, X n-1 represents the state quantity of the vehicle at the (n - 1)th moment, x n-1 represents the latitude value at the (n - 1)th moment, y n-1 represents the longitude value at the (n - 1)th moment, represents the heading angle at the (n - 1)th moment, X n represents the state quantity of the vehicle at the nth moment, x n represents the latitude value at the nth moment, y n represents the longitude value at the nth moment, represents the heading angle at the nth moment, u n represents the input quantity of the system at the nth moment, Z n represents the observed quantity of the vehicle at the nth moment, f() and h() respectively represent the state transition equation and the observation equation;

[0025]

[0026]

[0027]

[0028] In the formula, F n represents the state transition matrix of the system at the nth moment, G n represents the mapping matrix from the input of the system at the nth moment to the system state, H n represents the observation matrix of the system at the nth moment; According to the above three matrix formulas, the process of the standard EKF algorithm is described as follows: Given the initial value of the vehicle state quantity and the initial value of the system probability covariance matrix Then, continuously iterate according to the following formula to perform state estimation on the driverless vehicle:

[0029]

[0030]

[0031]

[0032]

[0033]

[0034]

[0035] wherein, represents the prior state estimator of the vehicle at time n, represents the posterior state estimator of the vehicle at time n, represents the posterior probability covariance matrix of the system state of the vehicle at time n-1, represents the transpose of the state transition matrix of the system at time n, represents the transpose of the mapping matrix from the system input to the system state at time n, represents the transpose of the observation matrix of the system at time n, K n represents the Kalman gain at time n, represents the prior probability covariance matrix of the system state of the vehicle at time n, represents the posterior probability covariance matrix of the system state of the vehicle at time n, R n represents the covariance matrix corresponding to the uncertainty of the observed value, R n is determined by the sensors used, Q n represents the system noise variance matrix at time n, e v,n represents the speed measurement noise of the wheel odometer at time n, e a,n represents the acceleration measurement noise of the IMU at time n, e ω,n represents the angular velocity measurement noise of the IMU at time n, e v,n , e a,n , e ω,n The initial value of is obtained from the calibration test;

[0036] After the extended Kalman filter fusion algorithm is completed, the fusion of the vehicle's longitude and latitude positioning information, heading angle information, vehicle instantaneous acceleration information, instantaneous angular velocity information, and vehicle instantaneous speed information can be completed, and finally the posterior state estimator of the vehicle at time n is output That is, the position and heading angle information of the vehicle.

[0037] Furthermore, the specific lane line distance constraint algorithm in step 4) is as follows:

[0038]

[0039]

[0040]

[0041]

[0042]

[0043]

[0044]

[0045]

[0046]

[0047]

[0048]

[0049] wherein, represents the change in the lateral position of the vehicle detected by the camera from time n - 1 to time n; represents the distance from the vehicle detected by the camera at time n - 1 to the lane line; represents the distance from the vehicle detected by the camera at time n to the lane line; represents the change in the lateral position of the vehicle detected by the IMU and the wheel odometer from time n - 1 to time n; V n-1,y represents the lateral speed of the vehicle at time n - 1; dt represents the time elapsed from time n - 1 to time n; represents the lateral acceleration of the vehicle at time n - 1, obtained through the IMU; V n-1 represents the speed of the vehicle at time n - 1; V n-1,x represents the longitudinal speed of the vehicle at time n - 1, obtained through the wheel speed odometer; represents the yaw angular velocity of the vehicle at time n - 1, obtained through the IMU; R represents the turning radius with the vehicle's center of mass as the origin; E n-1,n represents the lateral error value of the vehicle from time n - 1 to time n; h() represents the mapping relationship from the speed noise of the wheel odometer, the acceleration noise and the angular velocity noise of the IMU to the total lateral error value; represents the acceleration noise measured by the IMU from time n - 1 to time n, following a normal distribution with variance and mean of μ acc , σ acc , obtained through a calibration test; represents the angular velocity noise measured by the IMU from time n - 1 to time n, following a normal distribution with variance and mean of μ gyro , σ gyro , obtained through a calibration test; It represents the speed noise measured by the wheel odometer from time n-1 to time n, which conforms to a normal distribution, and the variance and mean are μ od , σ od , respectively, and are obtained from the calibration test; Q n-1,n It represents the system noise variance matrix obtained after the system is updated from time n-1 to time n. Use this updated system noise variance matrix to replace the system noise variance matrix of the extended Kalman filter used in step 3), and then use the updated extended Kalman filter for fusion positioning to obtain the real-time positioning information of the vehicle.

[0050] Advantages of the present invention:

[0051] 1. The present invention ensures the stability of the positioning signal output of the autonomous driving vehicle when the GPS signal is missing, and improves the safety of the autonomous driving vehicle when driving on urban roads;

[0052] 2. The present invention can automatically adjust the system noise variance matrix of the filter by detecting the lane line distance, ensure the stability of the filter, and thus stably output more accurate positioning information. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is the schematic diagram of the positioning method of the present invention.

[0054] Figure 2 It is the schematic diagram of 5 GPS missing areas specifically selected in the verification experiment of the present invention.

[0055] Figure 3 It is the comparison diagram of the positioning effect of a specific path in the verification experiment of the present invention.

[0056] Figure 4 It is the comparison diagram of the maximum error values of five GPS missing areas in the verification experiment of the present invention.

[0057] Figure 5 It is the comparison diagram of the root mean square error values of five GPS missing areas in the verification experiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0058] For the convenience of those skilled in the art to understand, the present invention will be further described below in conjunction with the embodiments and the drawings. The content mentioned in the embodiments does not limit the present invention.

[0059] A positioning system for an autonomous driving vehicle based on lane line recognition according to the present invention includes: a GPS unit, an IMU unit, a wheel odometer unit, a lane line detection unit, and an extended Kalman filter unit;

[0060] The GPS unit is installed outside the vehicle and is used to obtain the real-time longitude and latitude positioning information and heading angle information of the vehicle, and send the obtained information to the extended Kalman filter unit in real time;

[0061] The IMU unit is installed inside the vehicle and is used to obtain the instantaneous acceleration information and instantaneous angular velocity information of the vehicle in real time, and send the obtained information to the extended Kalman filter unit in real time;

[0062] The wheel odometer unit is installed on the vehicle wheels and is used to obtain the instantaneous speed information of the vehicle in real time, and send the obtained information to the extended Kalman filter unit in real time;

[0063] The lane line detection unit is installed at the front windshield of the vehicle, obtains the road image information in front of the vehicle in real time, and identifies the lane line information around the vehicle;

[0064] The extended Kalman filter unit is installed inside the vehicle. When the GPS unit signal is available, it obtains the longitude and latitude positioning information and heading angle information sent by the GPS unit, the instantaneous angular velocity information and instantaneous acceleration information sent by the IMU unit, and the instantaneous speed information sent by the wheel odometer unit in real time, and then uses the extended Kalman filtering algorithm to fuse the above information to finally obtain the real-time positioning information of the vehicle; when the GPS unit signal is unavailable, it uses the extended Kalman filtering algorithm to fuse the instantaneous angular velocity information and instantaneous acceleration information sent by the IMU unit and the instantaneous speed information sent by the wheel odometer unit, and uses the lane line information sent by the lane line detection unit to eliminate the cumulative error caused by the sensor noise in the IMU unit and the wheel odometer unit.

[0065] Among them, the lane line detection unit identifies the lane line information around the vehicle through a lane line recognition algorithm. The neural network models specifically used in the lane line recognition algorithm include two network models, LaneNet and H-Net; among them, LaneNet is a multi-task model that combines semantic segmentation and vector representation of pixels and is used for instance segmentation of lane lines in pictures; H-Net is a network model composed of convolutional layers and fully connected layers and is used to predict the transformation matrix, and the transformation matrix is used to perform regression on pixel points belonging to the same lane line.

[0066] Refer to Figure 1 As shown, the present invention also provides a positioning method for an autonomous driving vehicle based on lane line recognition. Based on the above system, the steps are as follows:

[0067] 1) Obtain the longitude and latitude positioning information and heading angle information of the vehicle itself, the instantaneous acceleration information and instantaneous angular velocity information of the vehicle itself, the instantaneous speed information of the vehicle, and the lane line information around the vehicle in real time;

[0068] 2) Determine whether the information received by the GPS unit is available: Decode the received GPS raw data to obtain the status information of the GPS. If the decoded status information is non - negative, the information received by the GPS unit is available, and proceed to step 3); if the decoded status information is negative, the information received by the GPS unit is unavailable, and proceed to step 4);

[0069] Among them, the GPS decoding method used is: Use the function package nmea_navsat_driver under the LINUX system to decode the GPS raw data.

[0070] 3) Send the vehicle's longitude and latitude positioning information, heading angle information, vehicle instantaneous acceleration information, instantaneous angular velocity information, and vehicle instantaneous speed information obtained in step 1) to the extended Kalman filter unit. Through the extended Kalman filter in this unit, fuse the above - mentioned information to obtain the vehicle's real - time positioning information;

[0071] Specifically, the specific algorithm of the extended Kalman filter used is as follows:

[0072] X n = f(X n-1 , u n )

[0073] Z n = h(X n )

[0074] In the formula, X n-1 represents the state quantity of the vehicle at the (n - 1)th moment, x n-1 represents the latitude value at the (n - 1)th moment, y n-1 represents the longitude value at the (n - 1)th moment, represents the heading angle at the (n - 1)th moment, X n represents the state quantity of the vehicle at the nth moment, x n represents the latitude value at the nth moment, y n represents the longitude value at the nth moment, represents the heading angle at the nth moment, u n represents the input quantity of the system at the nth moment, Z n represents the observed quantity of the vehicle at the nth moment, f() and h() represent the state transition equation and the observation equation respectively;

[0075]

[0076]

[0077]

[0078] In the formula, F n represents the state transition matrix of the system at time n, G n represents the mapping matrix from the system input to the system state at time n, H n represents the observation matrix of the system at time n; According to the above three matrix formulas, the process of the standard EKF algorithm is described as follows: Given the initial value of the vehicle state quantity and the initial value of the system probability covariance matrix Then, continuously iterate according to the following formula to perform state estimation on the driverless vehicle:

[0079]

[0080]

[0081]

[0082]

[0083]

[0084]

[0085] In the formula, represents the prior state estimation quantity of the vehicle at time n, represents the posterior state estimation quantity of the vehicle at time n, represents the posterior probability covariance matrix of the system state of the vehicle at time n - 1, represents the transpose of the state transition matrix of the system at time n, represents the transpose of the mapping matrix from the system input to the system state at time n, represents the transpose of the observation matrix of the system at time n, K n represents the Kalman gain at time n, represents the prior probability covariance matrix of the system state of the vehicle at time n, represents the posterior probability covariance matrix of the system state of the vehicle at time n, R n represents the covariance matrix corresponding to the uncertainty of the observed value, R n is determined by the sensors used, Q n represents the system noise variance matrix at time n, e v,n represents the speed measurement noise of the wheel odometer at time n, e a,n represents the acceleration measurement noise of the IMU at time n, e ω,n represents the angular velocity measurement noise of the IMU at time n, e v,n , e a,n , e ω,nThe initial value is obtained from the calibration test;

[0086] After the extended Kalman filter fusion algorithm is completed, the fusion of the vehicle's longitude and latitude positioning information, heading angle information, vehicle instantaneous acceleration information and instantaneous angular velocity information, and vehicle instantaneous speed information can be completed, and finally the posterior state estimator of the vehicle at time n is output That is, the position and heading angle information of the vehicle.

[0087] 4) According to the lane line information around the vehicle obtained, calculate the distance from the vehicle to the nearest lane line, calculate the error of the IMU in the current IMU unit in the direction perpendicular to the lane line through the lane line distance constraint algorithm, and then deduce the current noise values of the IMU in the IMU unit and the wheel odometer in the wheel odometer unit. Use this noise value to update the system noise variance matrix of the extended Kalman filter in the extended Kalman filter unit, and use the updated extended Kalman filter to fuse the vehicle instantaneous acceleration information and instantaneous angular velocity information sent by the IMU unit and the vehicle instantaneous speed information sent by the wheel odometer unit to obtain the real-time positioning information of the vehicle.

[0088] Among them, the specific content of the lane line distance constraint algorithm is as follows:

[0089]

[0090]

[0091]

[0092]

[0093]

[0094]

[0095]

[0096]

[0097]

[0098]

[0099]

[0100] In the formula, represents the change in the vehicle's lateral position detected by the camera from time n - 1 to time n; represents the distance from the vehicle to the lane line detected by the camera at time n - 1; represents the distance from the vehicle to the lane line detected by the camera at time n; represents the change in the lateral position of the vehicle detected by the IMU and the wheel odometer from time n - 1 to time n; V n-1,y represents the speed of the vehicle in the lateral direction at time n - 1; dt represents the time elapsed from time n - 1 to time n; represents the acceleration of the vehicle in the lateral direction at time n - 1, obtained through the IMU; V n-1 represents the speed of the vehicle at time n - 1; V n-1,x represents the speed of the vehicle in the longitudinal direction at time n - 1, obtained through the wheel speed odometer; represents the yaw rate of the vehicle at time n - 1, obtained through the IMU; R represents the radius of turning with the vehicle's center of mass as the origin; E n-1,n represents the error value of the vehicle in the lateral direction from time n - 1 to time n; h() represents the mapping relationship from the speed noise of the wheel odometer, the acceleration noise and the angular velocity noise of the IMU to the total error value in the lateral direction; represents the acceleration noise measured by the IMU from time n - 1 to time n, which conforms to a normal distribution, and the variance and mean are μ acc , σ acc , obtained from the calibration test; represents the angular velocity noise measured by the IMU from time n - 1 to time n, which conforms to a normal distribution, and the variance and mean are μ gyro , σ gyro , obtained from the calibration test; represents the speed noise measured by the wheel odometer from time n - 1 to time n, which conforms to a normal distribution, and the variance and mean are μ od , σ od , obtained from the calibration test; Q n-1,n represents the system noise variance matrix obtained after the system is updated from time n - 1 to time n. Use this updated system noise variance matrix to replace the system noise variance matrix of the extended Kalman filter used in step 3), and then use the updated extended Kalman filter for fusion positioning to obtain the real-time positioning information of the vehicle.

[0101] To further verify the positioning accuracy of the designed positioning method, a road was selected for experimental verification. The specific path is as Figure 2 shown. The circled part is the part where the GPS signal is missing. One of the paths when the GPS signal is missing is selected as Figure 4 shown. It can be seen that the optimized positioning effect is closer to the reference path than before optimization. Analyze the positioning data when the GPS is missing five times: the maximum error value is as Figure 4As shown, it can be seen that the maximum positioning error values after optimization are all smaller than those before optimization. By analyzing the data, it is obtained that the average maximum error value has decreased by 42.16%; the root mean square error values are as Figure 5 shown. It can be seen that the root mean square error values of the positioning data after optimization are all smaller than those of the positioning data before optimization. By analyzing the data, it is obtained that the average root mean square error value has decreased by 32.47%.

[0102] There are many specific application ways of the present invention. The above description is only the preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art of this technology, without departing from the principle of the present invention, several improvements can still be made, and these improvements should also be regarded as the protection scope of the present invention.

Claims

1. An autonomous driving vehicle positioning method based on lane line recognition, based on an autonomous driving vehicle positioning system based on lane line recognition, the system comprising: GPS unit, IMU unit, wheel odometer unit, lane line detection unit, extended Kalman filter unit; The GPS unit is installed outside the vehicle and is used to obtain the real-time longitude and latitude positioning information and heading angle information of the vehicle, and send the obtained information to the extended Kalman filter unit in real time; The IMU unit is installed inside the vehicle and is used to obtain the instantaneous acceleration information and instantaneous angular velocity information of the vehicle in real time, and send the obtained information to the extended Kalman filter unit in real time; The wheel odometer unit is installed on the vehicle wheels and is used to obtain the instantaneous speed information of the vehicle in real time, and send the obtained information to the extended Kalman filter unit in real time; The lane line detection unit is installed at the front windshield of the vehicle, obtains the road image information in front of the vehicle in real time, and identifies the lane line information around the vehicle; The extended Kalman filter unit is installed inside the vehicle. When the GPS unit signal is available, it obtains the longitude and latitude positioning information and heading angle information sent by the GPS unit, the instantaneous angular velocity information and instantaneous acceleration information sent by the IMU unit, and the instantaneous speed information sent by the wheel odometer unit in real time, and then uses the extended Kalman filtering algorithm to fuse the above information to finally obtain the real-time positioning information of the vehicle. When the GPS unit signal is unavailable, it uses the extended Kalman filtering algorithm to fuse the instantaneous angular velocity information and instantaneous acceleration information sent by the IMU unit and the instantaneous speed information sent by the wheel odometer unit, and uses the lane line information sent by the lane line detection unit to eliminate the cumulative error caused by the sensor noise in the IMU unit and the wheel odometer unit; It is characterized in that the method steps are as follows: 1) Obtain the longitude and latitude positioning information and heading angle information of the vehicle itself, the instantaneous acceleration information and instantaneous angular velocity information of the vehicle itself, the instantaneous speed information of the vehicle, and the lane line information around the vehicle in real time; 2) Determine whether the information received by the GPS unit is available: Decode the received GPS raw data to obtain the status information of the GPS. If the decoded status information is non-negative, the information received by the GPS unit is available, and go to step 3); if the decoded status information is negative, the information received by the GPS unit is unavailable, and go to step 4); 3) Send the longitude and latitude positioning information and heading angle information of the vehicle, the instantaneous acceleration information and instantaneous angular velocity information of the vehicle, and the instantaneous speed information of the vehicle obtained in step 1) to the extended Kalman filter unit, and fuse the above information through the extended Kalman filter in the unit to obtain the real-time positioning information of the vehicle; 4) Calculate the distance from the vehicle to the nearest lane line based on the obtained lane line information around the vehicle, calculate the error of the IMU in the current IMU unit in the direction perpendicular to the lane line through the lane line distance constraint algorithm, and then infer the current noise values of the IMU in the IMU unit and the wheel odometer in the wheel odometer unit. Use these noise values to update the system noise variance matrix of the extended Kalman filter in the extended Kalman filter unit, and use the updated extended Kalman filter to fuse the vehicle instantaneous acceleration information and instantaneous angular velocity information sent by the IMU unit and the vehicle instantaneous speed information sent by the wheel odometer unit to obtain the real-time positioning information of the vehicle; The specific content of step 3) is as follows: The specific algorithm of the used extended Kalman filter is as follows: X n = f(X n-1 , u n ) Z n = h(X n ) Where, X n-1 represents the state quantity of the vehicle at the (n - 1)th moment, x n-1 represents the latitude value at the (n - 1)th moment, y n-1 represents the longitude value at the (n - 1)th moment, represents the heading angle at the (n - 1)th moment, X n represents the state quantity of the vehicle at the nth moment, x n represents the latitude value at the nth moment, y n represents the longitude value at the nth moment, represents the heading angle at the nth moment, u n represents the input quantity of the system at the nth moment, Z n represents the observed quantity of the vehicle at the nth moment, f() and h() respectively represent the state transition equation and the observation equation; where F n represents the state transition matrix of the system at time n, G n represents the mapping matrix from the system input to the system state at time n, H n represents the observation matrix of the system at time n; According to the above three matrix formulas, the process of the standard EKF algorithm is described as follows: Given the initial value of the vehicle state quantity and the initial value of the system probability covariance matrix Iterate continuously according to the following formula to perform state estimation on the driverless vehicle: In the formula, represents the prior state estimator of the vehicle at time n, represents the posterior state estimator of the vehicle at time n, represents the posterior probability covariance matrix of the system state of the vehicle at time n - 1, represents the transpose of the state transition matrix of the system at time n, represents the transpose of the mapping matrix from the system input to the system state at time n, represents the transpose of the observation matrix of the system at time n, K n represents the Kalman gain at time n, represents the prior probability covariance matrix of the system state of the vehicle at time n, represents the posterior probability covariance matrix of the system state of the vehicle at time n, R n represents the covariance matrix corresponding to the uncertainty of the observed value, R n is determined by the sensors used, Q n represents the system noise variance matrix at time n, e v,n represents the speed measurement noise of the wheel odometer at time n, e a,n represents the acceleration measurement noise of the IMU at time n, e ω,n represents the angular velocity measurement noise of the IMU at time n, e v,n , e a,n , e ω,n The initial value of is obtained from the calibration test; After the extended Kalman filter fusion algorithm is completed, the fusion of the vehicle's longitude and latitude positioning information, heading angle information, instantaneous acceleration information and instantaneous angular velocity information, and instantaneous velocity information can be completed, and finally the posterior state estimator of the vehicle at time n is output That is, the position and heading angle information of the vehicle; The specific content of the lane line distance constraint algorithm in step 4) is as follows: Wherein, represents the change in the lateral position of the vehicle detected by the camera from time n-1 to time n; represents the distance from the vehicle detected by the camera at time n-1 to the lane line; represents the distance from the vehicle detected by the camera at time n to the lane line; represents the change in the lateral position of the vehicle detected by the IMU and wheel odometer from time n-1 to time n; V n-1,y represents the lateral speed of the vehicle at time n-1; dt represents the time elapsed from time n-1 to time n; represents the lateral acceleration of the vehicle at time n-1, obtained by the IMU; V n-1 represents the speed of the vehicle at time n-1; V n-1,x represents the longitudinal speed of the vehicle at time n-1, obtained by the wheel speed odometer; represents the yaw rate of the vehicle at time n-1, obtained by the IMU; R represents the turning radius with the vehicle's center of mass as the origin; E n-1,n represents the lateral error value of the vehicle from time n-1 to time n; h() represents the mapping relationship from the speed noise of the wheel odometer, the acceleration noise and angular velocity noise of the IMU to the total lateral error value; represents the acceleration noise measured by the IMU from time n-1 to time n, which conforms to a normal distribution, and the variance and mean are μ acc , σ acc , obtained from the calibration test; represents the angular velocity noise measured by the IMU from time n-1 to time n, which conforms to a normal distribution, and the variance and mean are μ gyro , σ gyro , obtained from the calibration test; represents the speed noise measured by the wheel odometer from time n-1 to time n, which conforms to a normal distribution, and the variance and mean are μ od , σ od , obtained from the calibration test; Q n-1,n represents the system noise variance matrix obtained after the system is updated from time n-1 to time n. Replace the system noise variance matrix of the extended Kalman filter used in step 3) with this updated system noise variance matrix, and then use the updated extended Kalman filter for fusion positioning to obtain the real-time positioning information of the vehicle.

2. The method for positioning an autonomous driving vehicle based on lane line recognition according to claim 1, wherein, The lane line detection unit identifies the lane line information around the vehicle through the lane line recognition algorithm. The neural network models specifically used in the lane line recognition algorithm include two network models, LaneNet and H-Net. Among them, LaneNet is a multi-task model that combines semantic segmentation and vector representation of pixels, and is used for instance segmentation of lane lines in pictures; H-Net is a network model composed of convolutional layers and fully connected layers, and is used to predict the transformation matrix, and the transformation matrix is used to perform regression on pixel points belonging to the same lane line.

3. The method for positioning an autonomous vehicle based on lane line recognition according to claim 1, wherein The GPS decoding method used in step 2) is: use the function package nmea_navsat_driver under the LINUX system to decode the GPS raw data.

Images