Vehicle distance measurement optimization method and system based on monocular camera

Abstract

The invention discloses a monocular camera-based vehicle distance measurement optimization method and system, and the method comprises the steps: collecting the motion characteristics of a vehicle, and continuously obtaining an environment perception image containing a license plate and a target object in the axial direction of the vehicle through a vehicle-mounted monocular camera; extracting a license plate area image and target object features from the image, determining a vehicle approximate distance by using a triangle method, obtaining a vehicle estimation distance by using a monocular depth estimation model, and determining a distance correction coefficient based on a license plate; determining a distance correction coefficient of the target object by using a pre-trained linear regression model based on the distance correction coefficient of the license plate, the vehicle motion characteristics and the target object characteristics; and determining the estimated distance of the target object by using a monocular depth estimation model, and determining the corrected distance of the target object in combination with the distance correction coefficient of the target object. According to the invention, the complementarity of the two methods is comprehensively utilized to reduce errors, corresponding correction coefficients are determined for different target objects, and more accurate target object distance measurement is realized.

Description

Technical Field

[0001] The present application belongs to the technical field of vehicle ranging, and specifically relates to a vehicle ranging optimization method and system based on a monocular camera. Background Art

[0002] The commonly used solutions for measuring distance in autonomous driving include LiDAR, using cameras to take frame-by-frame images and combine them with algorithms to calculate object distances, millimeter-wave radar, and ultrasonic sensors. Each solution has its own specific advantages and applicable scenarios. Each distance measurement method has its own source of error. For example, in actual scenarios, the driving environment of vehicles is complex and changeable. In traffic congestion scenarios, vehicles block each other, which may affect the use of cameras to take frame-by-frame images, and the accuracy of the object distance determined thereby needs to be improved. Summary of the invention

[0003] The technical purpose of this application is to address the technical problem that the accuracy of the current single distance measurement method needs to be improved, and to provide a vehicle distance measurement optimization method and system based on a monocular camera, combining two different distance measurement methods to form complementary advantages.

[0004] In order to achieve the above technical objectives, this application adopts the following technical solutions.

[0005] In a first aspect, an embodiment of the present application provides a vehicle ranging optimization method based on a monocular camera, comprising: During the driving process of the vehicle, the motion characteristics of the vehicle are collected, and the environment perception image on the axis of the vehicle is continuously collected using the on-board monocular camera, wherein the environment perception image includes the license plate and target object of the vehicle on the axis; Extracting a license plate area image based on the environmental perception image, determining an approximate distance of the vehicle in the axial direction using a similar triangle method based on the license plate area image; determining an estimated distance of the vehicle in the axial direction using a pre-acquired monocular depth estimation model; obtaining a distance correction coefficient based on the license plate according to the estimated distance and the approximate distance; extracting features of the target object based on the environmental perception image; According to the distance correction coefficient based on the license plate, the motion characteristics of the vehicle and the characteristics of the target object, a linear regression model is obtained by pre-training to predict the distance correction coefficient of the target object; The estimated distance to the target object is determined by using the monocular depth estimation model, and the corrected target object distance is determined in combination with the distance correction coefficient of the target object.

[0006] In a second aspect, the embodiment of the present application further provides a vehicle ranging optimization system based on a monocular camera, comprising: The data acquisition module is used to collect the motion characteristics of the vehicle during the driving process, and continuously collect the environment perception image on the axis of the vehicle using the on-board monocular camera, wherein the environment perception image includes the license plate and target object of the vehicle on the axis; A feature extraction module is used to extract a license plate area image based on the environmental perception image, determine the approximate distance of the vehicle in the axis direction using a triangle method based on the license plate area image; determine the estimated distance of the vehicle in the axis direction using a pre-acquired monocular depth estimation model; obtain a distance correction coefficient based on the license plate according to the estimated distance and the approximate distance; and extract the features of the target object based on the environmental perception image; An adaptive module, configured to predict the distance correction coefficient of the target object by using a pre-trained linear regression model according to the distance correction coefficient based on the license plate, the motion characteristics of the vehicle and the characteristics of the target object; The distance correction module is used to determine the estimated distance to the target object by using the monocular depth estimation model, and determine the corrected target object distance in combination with the distance correction coefficient of the target object.

[0007] Compared with the prior art, the vehicle distance measurement optimization method and system based on a monocular camera provided by the present application calculates the distance by two different methods, one is to obtain the estimated distance by using a monocular depth estimation model, and the other is to obtain the approximate distance by using a similar triangle method according to the monocular camera parameters and the license plate pixel height. This two-way calculation method can comprehensively utilize the advantages of different methods, and use the complementarity of the two methods to reduce errors, make up for the shortcomings of a single method, and improve the reliability of distance measurement. By using a linear regression model, combining the vehicle motion characteristics, the target characteristics and the license plate-based distance correction coefficient, the distance correction coefficient of the target is adaptively determined, and the estimated distance of the target obtained by the monocular depth estimation model is further corrected. It can perform personalized distance correction for different targets, and can achieve good performance when facing various types of targets. It can compensate for the influence of environmental factors on different targets to a certain extent, and can still provide accurate distance measurement results stably in complex and changing environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The drawings described herein are for explanation purposes only and are not intended to limit the scope of the present application in any way. In addition, the shapes and proportional dimensions of the components in the drawings are only for illustration purposes and are used to help understand the present application. They do not specifically limit the shapes and proportional dimensions of the components of the present application. Under the guidance of the present application, those skilled in the art can select various possible shapes and proportional dimensions to implement the present application according to specific circumstances. In the drawings: Figure 1 A schematic diagram of a process flow of a vehicle ranging optimization method based on a monocular camera provided in an embodiment of the present application; Figure 2 Schematic diagram of the principle of the similar triangle method used in the embodiment; Figure 3 is a schematic diagram of a depth map in an embodiment; Figure 4 is a schematic diagram of a depth map in another embodiment; Figure 5 A schematic diagram of the structure of a vehicle ranging optimization system based on a monocular camera provided in an embodiment of the present application. DETAILED DESCRIPTION

[0009] In order to enable those skilled in the art to better understand the technical solutions in the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this application.

[0010] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more features.

[0011] The commonly used solutions for measuring distance in autonomous driving currently include Lidar distance measurement method, method of calculating object distance by combining frame-by-frame image shooting with camera and algorithm, millimeter-wave radar distance measurement method and ultrasonic sensor distance measurement method. Among them, the camera-based distance measurement method is often used in basic vehicle safety systems, such as reversing radar, because it is easy to implement and has a low price. However, if a single method is used for distance measurement, it is easily affected by environmental factors, and the accuracy and reliability are relatively poor. Therefore, this application provides a vehicle distance measurement optimization method and system based on a monocular camera.

[0012] The present application embodiment provides a vehicle ranging optimization method based on a monocular camera, such as Figure 1 As shown, including: Step S1: During the driving process of the vehicle, the motion characteristics of the vehicle are collected, and the environment perception image along the axis of the vehicle is continuously collected using the on-board monocular camera, where the environment perception image includes the license plate of the vehicle and the target object along the axis; Step S2: extracting a license plate area image based on the environment perception image, determining an approximate distance of the vehicle in the axis direction using a similar triangle method based on the license plate area image; determining an estimated distance of the vehicle in the axis direction using a pre-acquired monocular depth estimation model; obtaining a distance correction coefficient based on the license plate according to the estimated distance and the approximate distance; extracting features of the target object based on the environment perception image; Step S3: According to the distance correction coefficient based on the license plate, the motion characteristics of the vehicle and the characteristics of the target object, a linear regression model is obtained by pre-training to predict the distance correction coefficient of the target object; Step S4: using a monocular depth estimation model to determine the estimated distance to the target object, and combining the distance correction coefficient of the target object to determine the corrected target object distance.

[0013] In a specific embodiment, a monocular camera is installed at a suitable position of the vehicle, such as above the inner side of the front windshield or near the rearview mirror inside the vehicle, to ensure that the environmental perception image in the axial direction (directly in front or directly behind) can be clearly captured. The environmental perception image may include license plates of other vehicles or other targets, such as pedestrians, traffic facilities (such as traffic signal poles, road signs, toll booth railings) or road obstacles.

[0014] In the embodiments, the vehicle's driving motion characteristics, such as speed or acceleration, are extracted. In some embodiments, the vehicle's speed or acceleration can be estimated based on the vehicle's displacement and time interval.

[0015] During the driving process of the vehicle, environmental perception images including the license plate and target objects along the vehicle axis are continuously collected. The frame rate of the collection is determined according to actual needs and computing resources, such as 25 frames per second.

[0016] Optionally, noise may appear in the image due to various interference factors during the driving of the vehicle, such as vibration, light changes, etc. In the embodiment, a filtering technology may be used to remove the noise of the environment perception image, such as Gaussian filtering.

[0017] In the embodiment, step S2 may adopt existing technology to obtain the license plate area image, such as first performing edge detection, then performing morphological operations or the license plate contour, and finally performing feature matching and corner point detection to accurately locate the key points of the license plate.

[0018] As an example, step S2 extracts the features of the target object based on the environmental perception image, including: based on the environmental perception image, using a pre-trained deep learning model to obtain the category of the target object and the pixel coordinates of the target object; feature encoding the category of the target object and converting it into category parameters; and using the pixel coordinates of the target object and the category parameters as the features of the target object.

[0019] In an embodiment, the deep learning model may adopt YOLOv10.

[0020] In an embodiment, other features of the target object may also be extracted, such as texture, color, or the position of the target object in the environmental perception image.

[0021] In an embodiment, the acquired numerical feature data may be normalized, and non-numerical data such as target object categories may be converted into numerical category parameters by using methods such as one-hot encoding to facilitate linear regression model learning.

[0022] Since the size of the license plate is fixed (large cars and small cars may have different sizes), the vehicle is classified as a large car or a small car during detection. The key points of the license plate are detected during the license plate detection process: upper left point, upper right point, lower left point, and lower right point. The distance from both sides of the license plate to the camera is calculated based on the fixed size of the license plate, such as Figure 2 The pinhole imaging principle shown in FIG. 1 can be obtained based on the relationship between similar triangles in step S3: d = A*(H*f) / h; Among them, A is the internal parameter matrix of the monocular camera, f is the focal length of the monocular camera, H is the actual height of the license plate, h is the actual height of the pixel point of the license plate, d is the approximate distance from the license plate to the monocular camera, where h can be h1 or h2, h1 is the actual height of the pixel point on the image on the left side of the license plate, and h2 is the actual height of the pixel point on the image on the right side of the license plate.

[0023] After the above matrix multiplication and division operations, a scalar distance is finally obtained, that is, the approximate distance d of the vehicle on the axis.

[0024] When h is h1, the calculated approximate height d1 is the approximate distance from the left side of the license plate to the camera; when h is h2, the calculated approximate height d2 is the approximate distance from the right side of the license plate to the camera.

[0025] according to Figure 2 It can be seen that cos(α) = (w² + d2² -d1²) / (2 * w * d2); Where W is the pixel width on the license plate image, and α is the angle between the left and right sides of the license plate and the camera.

[0026] In an embodiment, the angle α between the left and right sides of the license plate and the camera can be used to estimate the vehicle posture, such as inferring the lateral posture of the vehicle in front. If the angle is symmetrical, it means that the vehicle in front is driving in the direction of the camera. If the angle is biased to one side, it means that the vehicle in front has a certain offset in the lateral direction. It can be applied to scenarios such as automatic parking and lane keeping assist.

[0027] In an embodiment, the pre-obtained monocular depth estimation model may be a depth map, such as Figure 3 and Figure 4 As shown. Each pixel value of the depth map represents the distance information from the viewpoint (i.e., the position of the monocular camera) to the surface of the object in the scene (which can be other license plates or other targets in the axis). The depth map can be generated using existing technologies. If the depth map is used to determine the distance of a target object on the environment perception image, the target detection algorithm (such as Faster R-CNN, YOLO series based on deep learning, etc.) can be used to identify the target object in the driving environment perception image and determine its position in the image (usually represented by a bounding box); the target position information obtained by the target detection is mapped to the depth map. If the position of the target object in the image is represented by a bounding box (x1, y1, x2, y2), then the depth value can be extracted in the corresponding position area of ​​the depth map. The average depth value can usually be used to represent the depth information of the target location, that is, the average value of all depth values ​​in the area is calculated. There is a certain correspondence between the depth value in the depth map and the actual distance, which can be converted according to the calibration parameters of the camera to obtain the target distance. This is not the invention point of the present application and is not described in detail.

[0028] In some embodiments, obtaining the license plate-based distance correction factor based on the estimated distance and the approximate distance includes: determining an average value of the estimated distances of vehicles on a preset number of axes; ; Determine the approximate average distance of vehicles on the corresponding preset number of axes ; The average of the approximate distance The average distance from the estimated The ratio is used as the distance correction coefficient k based on the license plate, expressed as: .

[0029] In the embodiment, the linear regression model in step S3 is expressed as follows: ;

[0030] in, is the coefficient of the independent variable obtained through training, are the independent variables, which can be determined according to the motion characteristics of the vehicle, the characteristics of the target object and the distance correction coefficient based on the license plate; y is the predicted distance correction coefficient of the target object, and e is the error term.

[0031] As an example, a linear regression model looks like this: ; Among them, y is the distance correction coefficient of the predicted target object, is the Euclidean distance determined by the pixel coordinates of the target object, is the category parameter of the target object, is the speed of the vehicle, is the distance correction factor based on the license plate, , , and are all coefficients obtained through training, and e is the error term.

[0032] As an example, the relative depth information of each target can be obtained by simply relying on the depth map MH. The size of the depth map is the same as the size of the camera input image. The position of the corresponding license plate in the depth map is found and the distance of the depth map is corrected. The specific steps of step S4 are as follows: A block matrix is ​​intercepted from the depth map MH, and the position of the block matrix is ​​the license plate position. According to the depth map MH, the estimated distance MH_dobject from the target position corresponding to the depth map to the monocular camera is obtained; based on the predicted distance correction coefficient kobject of the target object, the corrected target object distance is determined as: dobject = kobject* MH_dobject.

[0033] In an embodiment, the vehicle distance measurement optimization method based on a monocular camera also includes training a linear regression model, specifically including: constructing an initial linear regression model; collecting data for training, including the corresponding vehicle motion characteristics, the characteristics of the target in the vehicle axis environment perception image, the category parameters of the target, and the corresponding license plate-based distance correction coefficient; using a pre-acquired monocular depth estimation model to determine the estimated distance of the target; obtaining the real distance between the vehicle and the target; using the ratio of the real distance of the target to the estimated distance as the real distance correction coefficient of the target. In an embodiment, the real distance between the vehicle and the target can be actually measured, or measured using a high-precision sensor such as a laser radar.

[0034] After preprocessing the data used for training, the data is divided into training set, validation set and test set according to a preset ratio; the training set is used to train model parameters, the validation set is used to adjust model hyperparameters, and the test set is used to evaluate model performance; the training set is input into the linear regression model, and the error between the predicted distance correction coefficient of the target object and the actual distance correction coefficient is calculated according to the selected loss function (such as the mean square error loss function), and the model parameters are continuously adjusted through the optimization algorithm to gradually reduce the loss function value; in each training round, the validation set is used to evaluate the performance of the linear regression model, and the model hyperparameters are adjusted according to the performance on the validation set to prevent overfitting; the test set is used to evaluate the performance of the linear regression model. If the performance meets the requirements (or reaches the maximum number of iterations), the final trained linear regression model is obtained.

[0035] In some embodiments, the method further includes performing a first data fusion based on a Kalman filter algorithm for the approximate distance to obtain an optimized approximate distance; and obtaining a corrected current distance measurement result based on the estimated distance and the optimized approximate distance.

[0036] As an example, data fusion based on the Kalman filter algorithm includes: Calculate the Kalman gain Kk, expressed as: ; The Kalman gain is used to update the prior state estimate to obtain the posterior state estimate: + ; Update the estimated error ; in, is the estimation error of the kth iteration, is the estimation error of the k-1th iteration, is the measurement error of the kth iteration.

[0037] With the new measurement distance (i.e., the approximate distance of the vehicle on the axis obtained by the similar triangle method) keeps coming, and the prediction step and update step are repeated. Each time, the distance measured by the new similar triangle method is incorporated into the state estimation, and the estimate of the real distance from the camera to the front vehicle is continuously optimized. Using the Kalman gain iteration formula mentioned above, the iteration is stopped when the convergence condition is reached, and the measurement error remains unchanged. During the iteration process, the estimation error and the final corrected approximate distance estimate are corrected. For example, during the driving process of the vehicle, the measured distance of the similar triangle method is obtained once per second, and the prediction and update operation of the Kalman filter is performed once per second, so that the distance estimate is continuously optimized.

[0038] Based on the above embodiments, some embodiments also include performing a second data fusion based on the Kalman filter algorithm for the corrected target distance to obtain a final optimized current distance measurement result.

[0039] The advantages of using the size of the license plate as a fixed value as the distance reference value and using the imaging principle to calculate the distance and angle from the license plate to the camera are: accurate distance calculation, low computing power requirements, and high robustness. The disadvantages are: only the distance from the surrounding license plates to the camera can be detected, and the distance perception is very limited. The relative value accuracy of the depth map distance is high, and the absolute value accuracy is poor. The advantage is that the distance perception range is full and is not affected by the presence or absence of license plates. This application corrects the two distances to obtain a reliable global range perception distance. There are many types of targets in actual driving scenarios, such as vehicles, pedestrians, traffic signs, etc. By adaptively determining the correction coefficient, this method does not need to deploy high-precision sensors, reduces the complexity of the overall vehicle system, and can adapt to the distance measurement requirements of different targets in various complex scenarios, ensuring that more accurate distance information can be obtained in all situations.

[0040] The vehicle ranging optimization method based on a monocular camera provided in the above embodiment has the same inventive concept, and the embodiment of the present application also provides a vehicle ranging optimization system based on a monocular camera, such as Figure 5 As shown, it includes a data acquisition module, a feature extraction module, an adaptive module and a distance correction module; wherein the data acquisition module is used to collect the motion characteristics of the vehicle during the driving process of the vehicle, and use the on-board monocular camera to continuously collect the environmental perception image on the axis of the vehicle, and the environmental perception image includes the license plate and target object of the vehicle on the axis.

[0041] The feature extraction module is used to extract the license plate area image based on the environmental perception image, determine the approximate distance of the vehicle on the axis using the triangle method based on the license plate area image; determine the estimated distance of the vehicle on the axis using the pre-acquired monocular depth estimation model; obtain the distance correction coefficient based on the license plate according to the estimated distance and the approximate distance; and extract the features of the target object based on the environmental perception image.

[0042] The adaptive module is used to predict the distance correction coefficient of the target object based on the distance correction coefficient based on the license plate, the movement characteristics of the vehicle and the characteristics of the target object by using a pre-trained linear regression model.

[0043] The distance correction module is used to determine the estimated distance to the target object using the monocular depth estimation model, and to determine the corrected target object distance in combination with the distance correction coefficient of the target object.

[0044] It should be understood that the contents of each method step described in the above embodiment 1 can be implemented by the vehicle ranging optimization system based on a monocular camera in this embodiment. For the sake of brevity, they will not be repeated in this embodiment.

[0045] The above is a detailed introduction to the vehicle ranging optimization method and system based on a monocular camera provided by the present application. This article uses specific examples to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the concept of the present application and should not be understood as limiting the scope of protection of the present application.

Claims

1. A vehicle ranging optimization method based on a monocular camera, characterized in that: include: During the driving process of the vehicle, the motion characteristics of the vehicle are collected, and the environment perception image on the axis of the vehicle is continuously collected using the on-board monocular camera, wherein the environment perception image includes the license plate and target object of the vehicle on the axis; Extracting a license plate area image based on the environmental perception image, determining an approximate distance of the vehicle in the axial direction using a similar triangle method based on the license plate area image; determining an estimated distance of the vehicle in the axial direction using a pre-acquired monocular depth estimation model; and obtaining a distance correction coefficient based on the license plate according to the estimated distance and the approximate distance; Extracting features of the target object based on the environment perception image; According to the distance correction coefficient based on the license plate, the motion characteristics of the vehicle and the characteristics of the target object, a linear regression model is obtained by pre-training to predict the distance correction coefficient of the target object; The estimated distance to the target object is determined by using the monocular depth estimation model, and the corrected target object distance is determined in combination with the distance correction coefficient of the target object.

2. The vehicle ranging optimization method based on a monocular camera according to claim 1, characterized in that: Extracting a license plate area image based on the environment perception image, and determining an approximate distance of the vehicle on an axial direction using a similar triangle method based on the license plate area image, including: Detecting key points of the license plate in the environmental perception image and extracting the license plate area image; Based on the license plate area image, determining the actual height of the pixel points of the license plate; The approximate distance of the vehicle in the axial direction is determined using a similar triangle method according to the focal length of the vehicle-mounted monocular camera, the camera internal parameter matrix, and the actual height of the pixel points of the license plate.

3. The vehicle ranging optimization method based on a monocular camera according to claim 1 is characterized in that: Obtaining a distance correction coefficient based on the license plate according to the estimated distance and the approximate distance, including: Determine the average estimated distance of the vehicle along a preset number of axes ; Determine the average approximate distance of the vehicle on the preset number of axes ; The approximate distance average The average distance from the estimated The ratio is used as the distance correction coefficient k based on the license plate, expressed as: .

4. The vehicle ranging optimization method based on a monocular camera according to claim 1, characterized in that: Extracting the features of the target object based on the environment perception image includes: Based on the environmental perception image, using a pre-trained deep learning model to obtain the category of the target object and the pixel coordinates of the target object; Characteristic encoding of the category of the target object is converted into category parameters; The pixel coordinates of the target object and the category parameter are used as features of the target object.

5. The vehicle ranging optimization method based on a monocular camera according to claim 1, characterized in that: The linear regression model is expressed as follows: ; in, is the coefficient of the independent variable obtained through training, are independent variables, y is the predicted distance correction coefficient of the target object, n is the number of independent variables, and e is the error term; wherein each variable is determined according to the motion characteristics of the vehicle, the characteristics of the target object and the distance correction coefficient based on the license plate.

6. The vehicle ranging optimization method based on a monocular camera according to claim 1, characterized in that: The method further includes pre-training the linear regression model, specifically comprising: Construct an initial linear regression model; Collecting data for training, including corresponding vehicle motion characteristics, characteristics of a target object in an environment perception image on the vehicle axis, and a corresponding license plate-based distance correction coefficient; determining an estimated distance of the target object using a pre-acquired monocular depth estimation model; obtaining a true distance between the vehicle and the target object; and using a ratio of the true distance to the estimated distance of the target object as a true distance correction coefficient of the target object; After preprocessing the data for training, the data are divided into a training set, a validation set and a test set according to a preset ratio; Input the training set into the linear regression model, calculate the error between the predicted distance correction coefficient of the target object and the actual distance correction coefficient according to the selected loss function, and continuously adjust the model parameters through the optimization algorithm so that the loss function value gradually decreases; In each training round, the performance of the linear regression model is evaluated using the validation set, and the model hyperparameters are adjusted according to the performance on the validation set to prevent overfitting; The test set is used to evaluate the performance of the linear regression model. If the performance meets the requirements, the final trained linear regression model is obtained.

7. The vehicle ranging optimization method based on a monocular camera according to claim 1, characterized in that: The method further includes: performing a first data fusion on the approximate distance based on a Kalman filter algorithm to obtain the optimized approximate distance.

8. The vehicle ranging optimization method based on a monocular camera according to claim 1, characterized in that: The method further includes: performing a second data fusion on the corrected target object distance based on a Kalman filter algorithm to obtain a final optimized target object distance.

9. The vehicle distance measurement optimization system based on a monocular camera is characterized by: include: The data acquisition module is used to collect the motion characteristics of the vehicle during the driving process, and continuously collect the environment perception image on the axis of the vehicle using the on-board monocular camera, wherein the environment perception image includes the license plate and target object of the vehicle on the axis; A feature extraction module is used to extract a license plate area image based on the environmental perception image, determine an approximate distance of the vehicle on the axis using a similar triangle method based on the license plate area image; determine an estimated distance of the vehicle on the axis using a pre-acquired monocular depth estimation model; and obtain a distance correction coefficient based on the license plate according to the estimated distance and the approximate distance; Extracting features of the target object based on the environment perception image; An adaptive module, configured to predict the distance correction coefficient of the target object by using a pre-trained linear regression model according to the distance correction coefficient based on the license plate, the motion characteristics of the vehicle and the characteristics of the target object; The distance correction module is used to determine the estimated distance to the target object by using the monocular depth estimation model, and determine the corrected target object distance in combination with the distance correction coefficient of the target object.

Images