Camera external parameter calibration method and device, electronic equipment and readable storage medium

Abstract

The invention provides a camera external parameter calibration method and device, electronic equipment and a computer readable storage medium, and relates to the technical field of intelligent driving, and the method comprises the steps: obtaining N parking space detection results corresponding to the same parking space, N is greater than or equal to 2, each parking space detection result comprises a first angular point coordinate of the same parking space, and N is greater than or equal to 1; the N parking space detection results are obtained by aerial view images of N original images shot by N look-around cameras; projecting first angular point coordinates corresponding to the N surround-view cameras to original images corresponding to the N surround-view cameras based on the initial external parameters of the N surround-view cameras to obtain second angular point coordinates of the first angular point coordinates in the original images; and adjusting respective initial external parameters of the N look-around cameras, and performing angular point coordinate alignment on the second angular point coordinates in the original image in the aerial view image to obtain respective target external parameters of the N look-around cameras. According to the invention, calibration of the surround-view camera is realized by using the parking space in the parking scene, and the total cost of calibration work can be saved.

Description

Technical Field

[0001] This application relates to the field of intelligent driving technology, and more specifically, to a camera extrinsic parameter calibration method, apparatus, electronic device, and computer-readable storage medium in the field of intelligent driving technology. Background Technology

[0002] With the rapid development of vehicle technology, most vehicles today are equipped with surround-view cameras, including front-view, rear-view, left-view, and right-view cameras. The calibration of surround-view cameras plays a crucial role in autonomous driving; for example, features such as 360-degree surround-view imaging and automatic parking in intelligent driver assistance systems rely on their calibration. Therefore, how to calibrate the surround-view cameras in vehicles has become an urgent problem to be solved. Summary of the Invention

[0003] This application provides a camera extrinsic parameter calibration method, apparatus, electronic device, and computer-readable storage medium. This application utilizes a calibration scheme for surround-view cameras in a real-world parking scenario, eliminating the need for additional calibration aids. Calibration of the surround-view camera can be achieved using parking lines and space lines within the parking scene, reducing reliance on the surrounding environment and eliminating the need for a dedicated calibration scene setup. Therefore, this application allows calibration to be performed in scenarios with parking spaces, making the calibration process easy to implement and significantly reducing the overall cost of calibration.

[0004] Firstly, a method for calibrating camera extrinsic parameters is provided, the method comprising:

[0005] Acquire first detection data; wherein, the first detection data includes N target parking space detection results corresponding to the same parking space, each target parking space detection result includes the first true corner coordinates of the same parking space, the N target parking space detection results are obtained from bird's-eye view images corresponding to N original images, the N original images are captured by N surround-view cameras, the shooting angles of the N surround-view cameras are different, and N is a positive integer greater than or equal to 2; based on the initial extrinsic parameters of each of the N surround-view cameras, project the first true corner coordinates in the target parking space detection results corresponding to each of the N surround-view cameras onto the original images corresponding to each of the N surround-view cameras to obtain the second true corner coordinates in the original images corresponding to each of the N target parking space detection results; adjust the initial extrinsic parameters of each of the N surround-view cameras to align the second true corner coordinates in the original images corresponding to each of the N surround-view cameras with the corner coordinates in the bird's-eye view images corresponding to the original images corresponding to each of the N surround-view cameras to obtain the first target extrinsic parameters of each of the N surround-view cameras.

[0006] This application embodiment, by employing the above-described technical solution, achieves the calibration of surround-view cameras located at multiple different shooting angles within a parking space in a parking scenario. On one hand, the calibration work does not require the setup of a calibration scene or reliance on additional calibration auxiliary equipment, making the calibration work easier to implement. This not only reduces the dependence of the calibration work on the surrounding environment but also saves on the overall cost of the calibration work and improves the calibration accuracy of the surround-view camera's extrinsic parameters. On the other hand, compared to natural scene calibration schemes, the above technical solution can significantly reduce the driving difficulty for the driver, such as reducing the difficulty of maneuvering the vehicle around flower petals, thus improving the practicality of the camera extrinsic parameter calibration method provided in this application. It also reduces the computational power requirements of electronic devices, the consumption of computational power, and the complexity of the algorithm.

[0007] In conjunction with the first aspect, in some possible implementations, acquiring the first detection data includes: acquiring M original images with the same timestamp; wherein the M original images are captured by M surround-view cameras, the M surround-view cameras have different shooting angles, the M surround-view cameras include the N surround-view cameras, and M is a positive integer greater than or equal to N; generating bird's-eye view images of the M original images based on the initial extrinsic parameters of each of the M surround-view cameras, resulting in M ​​bird's-eye view images; detecting parking spaces in each of the M bird's-eye view images to obtain a first parking space detection result corresponding to each of the M bird's-eye view images, each first parking space detection result including the coordinates of multiple detection corner points of the parking space, the confidence level and visibility identifier of each of the multiple detection corner points, wherein the visibility... A visibility identifier is used to indicate whether a detected corner point is visible. Based on the confidence level and the visibility identifier, false corner point coordinates are filtered from the multiple detected corner point coordinates included in the M first parking space detection results to obtain M second parking space detection results. Each second parking space detection result includes true corner point coordinates, which are the detected corner point coordinates other than the false corner point coordinates among the multiple detected corner point coordinates. The same parking space is matched across cameras according to the M second parking space detection results to obtain a parking space matching list. The parking space matching list includes multiple parking spaces and N second parking space detection results corresponding to each parking space. The N target parking space detection results corresponding to the same parking space are obtained from the parking space matching list to obtain the first detection data.

[0008] Combining the first aspect and the above implementation methods, in some possible implementation methods, the step of matching the same parking space across cameras based on the M second parking space detection results to obtain a parking space matching list includes: determining the similarity between the parking space corresponding to the third parking space detection result and the parking space corresponding to each second parking space detection result in the detection result set, based on the third parking space detection result and each second parking space detection result in the detection result set, to obtain multiple similarities; wherein, the third parking space detection result is any second parking space detection result, and the detection result set includes the second parking space detection results other than the third parking space detection result among the M second parking space detection results; determining the similarity among the multiple similarities that is greater than a preset similarity as the target similarity, and determining the parking spaces corresponding to the N second parking space detection results corresponding to the target similarity as the same parking space; associating the same parking space and the N second parking space detection results corresponding to the target similarity to generate the parking space matching list.

[0009] In conjunction with the first aspect and the above implementation methods, in some possible implementation methods, determining the similarity between the parking space corresponding to the third parking space detection result and the parking space corresponding to each of the second parking space detection results in the detection result set, based on the third parking space detection result and each of the second parking space detection results in the detection result set, includes: constructing a first parking space bounding box for the parking space corresponding to the third parking space detection result based on the true corner coordinates and corner order identifiers included in the third parking space detection result; constructing a second parking space bounding box for the parking space corresponding to each of the second parking space detection results in the detection result set based on the true corner coordinates and corner order identifiers included in each of the second parking space detection results in the detection result set; determining the intersection-union ratio (IUGR) between the first parking space bounding box and the second parking space bounding box; and using the IUGR as the similarity between the parking space corresponding to the third parking space detection result and the parking space corresponding to each of the second parking space detection results in the detection result set.

[0010] In combination with the first aspect and the above implementation methods, in some possible implementation methods, adjusting the initial extrinsic parameters of each of the N surround-view cameras to align the second true corner coordinates in the original images corresponding to each of the N surround-view cameras with the corner coordinates in the bird's-eye view images of the original images corresponding to each of the N surround-view cameras, and obtaining the first target extrinsic parameters of each of the N surround-view cameras, includes: adjusting the initial extrinsic parameters of each of the N surround-view cameras to obtain the working extrinsic parameters of each of the N surround-view cameras; and based on the working extrinsic parameters of each of the N surround-view cameras, projecting the second true corner coordinates in the original images corresponding to each of the N surround-view cameras onto the bird's-eye view images of the original images corresponding to each of the N surround-view cameras, to obtain the first target extrinsic parameters of each of the N surround-view cameras. The second true corner coordinates in the original image corresponding to each of the N surround-view cameras, and the third true corner coordinates in the bird's-eye view image of the original image corresponding to each of the N surround-view cameras; determine whether the third true corner coordinates in the bird's-eye view image of the original image corresponding to each of the N surround-view cameras meet the preset conditions for corner coordinate alignment; if yes, determine the working extrinsic parameters of each of the N surround-view cameras when the preset conditions are met as the first target extrinsic parameters of each of the N surround-view cameras; if no, execute the step of taking the working extrinsic parameters of each of the N surround-view cameras when the preset conditions are not met as the initial extrinsic parameters of each of the N surround-view cameras, and execute the step of adjusting the initial extrinsic parameters of each of the N surround-view cameras to obtain the working extrinsic parameters of each of the N surround-view cameras.

[0011] Combining the first aspect and the above implementation methods, in some possible implementation methods, determining whether the coordinates of the third true corner points in the bird's-eye view images of the original images corresponding to the N surround-view cameras satisfy the preset conditions for corner point coordinate alignment includes: determining the corner point connecting lines formed by the third true corner point coordinates in each of the N bird's-eye view images according to the corner point sequence identifiers in the detection results of the N target parking spaces; determining the point-to-line distance between the target corner point coordinates and the target connecting lines to obtain K point-to-line distances; where K is a positive integer greater than or equal to 1, the target corner point coordinates are any third true corner point coordinates in any bird's-eye view image, and the target... The connecting line is a line connecting the corner points corresponding to the target corner point coordinates in each of the bird's-eye view images included in the bird's-eye view image set, which includes bird's-eye view images other than the bird's-eye view image containing the target corner point coordinates; the distances of the K points and lines are summed to obtain the distance sum; if the distance sum is greater than or equal to a preset threshold, it is determined that the third true corner point coordinates in the bird's-eye view images of the original images corresponding to the N surround-view cameras do not meet the preset condition; if the distance sum is greater than or equal to the preset threshold, it is determined that the third true corner point coordinates in the bird's-eye view images of the original images corresponding to the N surround-view cameras meet the preset condition.

[0012] Combining the first aspect and the above implementation methods, in some possible implementation methods, after adjusting the initial extrinsic parameters of each of the N surround-view cameras to align the second true corner coordinates in the original images corresponding to each of the N surround-view cameras with the corner coordinates in the bird's-eye view images of the original images corresponding to each of the N surround-view cameras, and obtaining the first target extrinsic parameters of each of the N surround-view cameras, the camera extrinsic parameter calibration method further includes: acquiring N second detection data; wherein, the timestamps corresponding to each of the N second detection data are different; for each second detection data, the second detection data is used as the first detection data, and the step of projecting the first true corner coordinates in the target parking space detection results corresponding to each of the N surround-view cameras onto the original images corresponding to each of the N surround-view cameras based on the initial extrinsic parameters of each of the N surround-view cameras is performed to obtain the first target extrinsic parameters of each of the N surround-view cameras. The first true corner coordinates in each of the target parking space detection results are adjusted to the second true corner coordinates in the original images corresponding to each of the N surround view cameras. This is done by aligning the second true corner coordinates in the original images corresponding to each of the N surround view cameras with the corner coordinates in the bird's-eye view images of the original images corresponding to each of the N surround view cameras, to obtain N second target extrinsic parameters for each of the N surround view cameras. For each surround view camera, a histogram is constructed based on the N second target extrinsic parameters. The second target extrinsic parameter corresponding to the highest peak value in the histogram is determined as the third target extrinsic parameter for the surround view camera, to obtain the third target extrinsic parameters for each of the N surround view cameras. The first target extrinsic parameters for each of the N surround view cameras are updated using the third target extrinsic parameters for each of the N surround view cameras.

[0013] Secondly, a camera extrinsic parameter calibration device is provided, the camera extrinsic parameter calibration device comprising:

[0014] The data acquisition module is used to acquire first detection data; wherein, the first detection data includes the detection results of N target parking spaces corresponding to the same parking space, each target parking space detection result includes the first true corner coordinates of the same parking space, the N target parking space detection results are obtained from bird's-eye view images corresponding to N original images, the N original images are obtained by N surround view cameras, the shooting angles of the N surround view cameras are different, and N is a positive integer greater than or equal to 2;

[0015] The coordinate projection module is used to project the first true corner coordinates in the target parking space detection results of the N surround view cameras to the original images corresponding to the N surround view cameras based on the initial extrinsic parameters of the N surround view cameras, so as to obtain the second true corner coordinates in the original images corresponding to the N surround view cameras.

[0016] The coordinate alignment module is used to adjust the initial extrinsic parameters of each of the N surround-view cameras, so as to align the second true corner coordinates in the original images corresponding to each of the N surround-view cameras with the corner coordinates in the bird's-eye view images of the original images corresponding to each of the N surround-view cameras, thereby obtaining the first target extrinsic parameters of each of the N surround-view cameras.

[0017] In conjunction with the second aspect, in some possible implementations, the data acquisition module includes:

[0018] The first processing unit is used to acquire M original images with the same timestamp; wherein the M original images are captured by M surround-view cameras, the shooting angles of the M surround-view cameras are different, the M surround-view cameras include the N surround-view cameras, and M is a positive integer greater than or equal to N;

[0019] The second processing unit is used to generate bird's-eye view images of the M original images based on the initial extrinsic parameters of the M surround-view cameras, thereby obtaining M bird's-eye view images;

[0020] The third processing unit is used to detect parking spaces in each of the M bird's-eye view images to obtain the first parking space detection result corresponding to each of the M bird's-eye view images. Each first parking space detection result includes the coordinates of multiple detection corner points of the parking space, the confidence level and visibility identifier of each of the multiple detection corner points, and the visibility identifier is used to indicate whether the detection corner point is a visible corner point.

[0021] The fourth processing unit is used to filter the false corner coordinates among the multiple detected corner coordinates included in the M first parking space detection results based on the confidence level and the visibility identifier, so as to obtain M second parking space detection results; wherein, each second parking space detection result includes true corner coordinates, and the true corner coordinates are the detected corner coordinates other than the false corner coordinates among the multiple detected corner coordinates;

[0022] The fifth processing unit is used to perform cross-camera matching of the same parking space based on the M second parking space detection results to obtain a parking space matching list; wherein, the parking space matching list includes multiple parking spaces and N second parking space detection results corresponding to each of the multiple parking spaces;

[0023] The sixth processing unit is used to obtain the detection results of N target parking spaces corresponding to the same parking space from the parking space matching list, and obtain the first detection data.

[0024] In combination with the second aspect and the above implementation methods, in some possible implementations, the fifth processing unit includes:

[0025] The calculation subunit is used to determine the similarity between the parking space corresponding to the third parking space detection result and the parking space corresponding to each second parking space detection result in the detection result set, based on the third parking space detection result and each second parking space detection result in the detection result set, thereby obtaining multiple similarities; wherein, the third parking space detection result is any second parking space detection result, and the detection result set includes the second parking space detection results other than the third parking space detection result among the M second parking space detection results;

[0026] The determination subunit is used to determine the similarity among the multiple similarities that is greater than a preset similarity as the target similarity, and to determine the parking spaces corresponding to the N second parking space detection results corresponding to the target similarity as the same parking space;

[0027] A sub-unit is generated to associate the detection results of the same parking space and N second parking spaces corresponding to the target similarity to generate the parking space matching list.

[0028] In conjunction with the second aspect and the above implementation methods, in some possible implementation methods, the calculation subunit is specifically used to construct a first parking space bounding box corresponding to the third parking space detection result based on the actual corner coordinates and corner order identifiers included in the third parking space detection result; construct a second parking space bounding box corresponding to the parking space corresponding to each second parking space detection result in the detection result set based on the actual corner coordinates and corner order identifiers included in each second parking space detection result in the detection result set; determine the intersection-union ratio (IUGR) between the first parking space bounding box and the second parking space bounding box; and use the IUGR as the similarity between the parking space corresponding to the third parking space detection result and the parking space corresponding to each second parking space detection result in the detection result set.

[0029] In combination with the second aspect and the above implementation methods, in some possible implementations, the coordinate alignment module includes:

[0030] An adjustment unit is used to adjust the initial extrinsic parameters of each of the N surround-view cameras to obtain the operating extrinsic parameters of each of the N surround-view cameras;

[0031] The mapping unit is used to project the second true corner coordinates in the original images corresponding to the N surround-view cameras to the bird's-eye view image of the original images corresponding to the N surround-view cameras based on the working extrinsic parameters of the N surround-view cameras, so as to obtain the second true corner coordinates in the original images corresponding to the N surround-view cameras and the third true corner coordinates in the bird's-eye view image of the original images corresponding to the N surround-view cameras.

[0032] The judgment unit is used to determine whether the third true corner coordinates in the bird's-eye view of the original images corresponding to each of the N surround-view cameras meet the preset conditions for corner coordinate alignment; if yes, the working extrinsic parameters of each of the N surround-view cameras when the preset conditions are met are determined as the first target extrinsic parameters of each of the N surround-view cameras; if no, the working extrinsic parameters of each of the N surround-view cameras when the preset conditions are not met are used as the initial extrinsic parameters of each of the N surround-view cameras, and the step of adjusting the initial extrinsic parameters of each of the N surround-view cameras to obtain the working extrinsic parameters of each of the N surround-view cameras is executed.

[0033] In conjunction with the second aspect and the above implementation methods, in some possible implementation methods, the judgment unit, in determining whether the third true corner coordinates in the bird's-eye view images of the original images corresponding to the N surround-view cameras meet the preset conditions for corner coordinate alignment, is specifically used to determine the corner line formed by the third true corner coordinates in each of the N bird's-eye view images based on the corner order identifiers in the N target parking space detection results; determine the point-to-line distance between the target corner coordinates and the target line to obtain K point-to-line distances; where K is a positive integer greater than or equal to 1, and the target corner coordinates are any third true corner coordinates in any bird's-eye view image. The target line is the line connecting the corner points corresponding to the target corner point coordinates in each of the bird's-eye view images included in the bird's-eye view image set. The bird's-eye view image set includes bird's-eye view images other than the bird's-eye view image containing the target corner point coordinates. The distances of the K points and lines are summed to obtain a distance sum. If the distance sum is greater than or equal to a preset threshold, it is determined that the third true corner point coordinates in the bird's-eye view images of the original images corresponding to the N surround-view cameras do not meet the preset condition. If the distance sum is greater than or equal to the preset threshold, it is determined that the third true corner point coordinates in the bird's-eye view images of the original images corresponding to the N surround-view cameras meet the preset condition.

[0034] In conjunction with the second aspect and the above-described implementations, in some possible implementations, the camera extrinsic parameter calibration device further includes:

[0035] An optimization unit is used to acquire N second detection data points, wherein the timestamps corresponding to the N second detection data points are different. For each second detection data point, the second detection data point is used as the first detection data point, and the process of projecting the first true corner coordinates in the target parking space detection results corresponding to the N surround view cameras onto the original images corresponding to the N surround view cameras is performed based on the initial extrinsic parameters of the N surround view cameras. This process obtains the second true corner coordinates in the original images corresponding to the N surround view cameras, and adjusts the initial extrinsic parameters of the N surround view cameras to adjust the first true corner coordinates in the target parking space detection results. The second true corner coordinates of the original images corresponding to the N surround-view cameras are used to perform corner coordinate alignment in the bird's-eye view of the original images corresponding to the N surround-view cameras to obtain N second target extrinsic parameters for each of the N surround-view cameras. For each surround-view camera, a histogram corresponding to the surround-view camera is constructed based on the N second target extrinsic parameters. The second target extrinsic parameter corresponding to the highest peak value in the histogram is determined as the third target extrinsic parameter of the surround-view camera to obtain the third target extrinsic parameters for each of the N surround-view cameras. The first target extrinsic parameters of each of the N surround-view cameras are updated using the third target extrinsic parameters of each of the N surround-view cameras.

[0036] Thirdly, an electronic device is provided, including a memory and a processor. The memory is used to store executable program code, and the processor is used to call and run the executable program code from the memory, causing the electronic device to perform the camera extrinsic calibration method in the first aspect or any possible implementation thereof.

[0037] Fourthly, a computer program product is provided, comprising: computer program code, which, when run on a computer, causes the computer to execute the camera extrinsic calibration method in the first aspect or any possible implementation thereof.

[0038] Fifthly, a computer-readable storage medium is provided that stores computer program code, which, when executed on a computer, causes the computer to perform the camera extrinsic calibration method of the first aspect or any possible implementation thereof. Attached Figure Description

[0039] Figure 1 A schematic flowchart of a camera extrinsic parameter calibration method provided in an embodiment of this application is shown;

[0040] Figure 2 A schematic diagram of a vehicle equipped with a surround-view camera, according to an embodiment of this application, is shown;

[0041] Figure 3 An exemplary parking detection result diagram is shown;

[0042] Figure 4 This shows the effect of matching the same parking space across different cameras;

[0043] Figure 5 A schematic diagram of the IPM stitching effect of the panoramic parking space images is shown;

[0044] Figure 6 This illustration shows a schematic diagram of a camera extrinsic parameter calibration device provided in an embodiment of this application;

[0045] Figure 7 A schematic diagram of the structure of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0046] The technical solutions in this application will be clearly and thoroughly described below with reference to the accompanying drawings. In the description of the embodiments of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B. "And / or" in the text is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Furthermore, in the description of the embodiments of this application, "multiple" refers to two or more than two.

[0047] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature.

[0048] With the rapid development of vehicle technology, most vehicles today are equipped with surround-view cameras, which include front-view, rear-view, left-view, and right-view cameras. Surround-view camera calibration plays a crucial role in autonomous driving; for example, features such as 360-degree surround-view imaging and automatic parking in intelligent driver assistance systems rely on surround-view camera calibration.

[0049] Current surround-view camera calibration schemes, such as active vision camera calibration schemes, require the use of additional equipment, such as LiDAR or encoders, to work in conjunction with the camera and provide additional positioning information to aid in calibration. On the one hand, using additional equipment increases the overall cost of the calibration process. On the other hand, the accuracy of the calibration results is highly dependent on the precision of the additional equipment itself; if these devices are not properly calibrated or malfunction, the calibration results may be inaccurate.

[0050] Addressing the shortcomings of existing surround-view camera calibration schemes, this application provides a camera extrinsic parameter calibration method, apparatus, electronic device, and computer-readable storage medium. This application utilizes a real-world parking scenario for surround-view camera calibration, eliminating the need for additional calibration aids. Calibration can be achieved using parking lines and space lines within the parking scene, reducing reliance on the surrounding environment and eliminating the need for a dedicated calibration scene setup. Therefore, this application allows calibration to be performed in scenarios with parking spaces, simplifying implementation and significantly reducing overall calibration costs.

[0051] The following is an embodiment of a camera extrinsic parameter calibration method provided in this application.

[0052] Figure 1 A schematic flowchart illustrating a camera extrinsic parameter calibration method provided in an embodiment of this application is shown, such as... Figure 1 As shown, the camera extrinsic parameter calibration method provided in this application embodiment is applied to electronic devices with computing power, such as computers. The camera extrinsic parameter calibration method provided in this application embodiment can be applied to the calibration of surround-view cameras, which are mounted on vehicles. Figure 2 As shown, Figure 2 This illustration shows a vehicle equipped with a surround-view camera according to an embodiment of this application. 100 represents the vehicle. The surround-view camera system includes four cameras: a front-view camera 101 at the front of the vehicle, a rear-view camera 102 at the rear, a left-view camera 103 on the left side of the vehicle, and a right-view camera 104 on the right side of the vehicle. The front-view camera 101 overlaps with both the left-view and right-view cameras 103 and 104, and the rear-view camera 102 overlaps with both. That is, the front-view camera 101 and the left-view camera 103 can capture the same object, the front-view camera 101 and the right-view camera 104 can capture the same object, the rear-view camera 102 and the left-view camera 103 can capture the same object, and the rear-view camera 102 and the right-view camera 104 can capture the same object.

[0053] The above-mentioned camera extrinsic parameter calibration methods include the following schemes:

[0054] S110: Obtain the first detection data.

[0055] In an exemplary embodiment, the first detection data includes the detection results of N target parking spaces corresponding to the same parking space Ci, where N is a positive integer greater than or equal to 2. Each target parking space detection result includes the coordinates of the first true corner point of the same parking space Ci. The first true corner point coordinates refer to the coordinates of the true corner point of the same parking space Ci. A true corner point can be understood as a corner point of the same parking space Ci that actually exists and is visible in the image. There are multiple true corner points of the same parking space Ci, that is, there are also multiple first true corner point coordinates. Generally, the parking space is rectangular or parallelogram-shaped, that is, it has four corner points. Therefore, there are 4 true corner points of the same parking space Ci, and there are also 4 first true corner point coordinates. The N target parking space detection results are obtained from the bird's-eye view images corresponding to the N original images. The N original images are obtained by N surround-view cameras, and the shooting angles of the N surround-view cameras are different.

[0056] like Figure 2 As shown, assuming N=2, the N surround-view cameras can include a front-view camera 101 and a left-view camera 103, or a front-view camera 101 and a right-view camera 104, or a rear-view camera 102 and a left-view camera 103, or a rear-view camera 102 and a right-view camera 104. For example, if the N surround-view cameras include a front-view camera 101 and a left-view camera 103, then both the front-view camera 101 and the left-view camera 103 have captured the same parking space Ci. The difference is that their shooting perspectives are different. That is, the detection result of one target parking space corresponding to the same parking space Ci is obtained through the bird's-eye view image corresponding to the original image captured by the front-view camera 101, and the detection result of the other target parking space corresponding to the same parking space Ci is obtained through the bird's-eye view image corresponding to the original image captured by the left-view camera 103. In other words, the same parking space Ci is a parking space obtained through cross-camera matching, and the two target parking space detection results corresponding to the same parking space Ci are the two target parking space detection results of the parking space obtained through cross-camera matching.

[0057] When N surround-view cameras include a front-view camera 101 and a right-view camera 104, or a rear-view camera 102 and a left-view camera 103, or a rear-view camera 102 and a right-view camera 104, the understanding of the detection results for the same parking space Ci and the two target parking spaces corresponding to the same parking space Ci is the same as the understanding of the detection results for the same parking space Ci and the two target parking spaces corresponding to the same parking space Ci when N surround-view cameras include a front-view camera 101 and a left-view camera 103. This application will not repeat the details.

[0058] S120: Based on the initial extrinsic parameters of each of the N surround view cameras, project the first true corner coordinates in the target parking space detection results of each of the N surround view cameras onto the original images corresponding to each of the N surround view cameras, so as to obtain the second true corner coordinates in the original images corresponding to each of the N target parking space detection results.

[0059] Obtain the initial extrinsic parameters of each of the N surround-view cameras. For example, if the N surround-view cameras include a front-view camera 101 and a left-view camera 103, then obtain the initial extrinsic parameters of the front-view camera 101 and the initial extrinsic parameters of the left-view camera 103.

[0060] After obtaining the initial extrinsic parameters of each of the N surround-view cameras, based on the initial extrinsic parameters of the first surround-view camera, the first true corner coordinates in the target parking space detection result corresponding to the first surround-view camera are projected onto the original image corresponding to the first surround-view camera, thus obtaining the second true corner coordinates in the original image corresponding to the first surround-view camera based on the initial extrinsic parameters of the first surround-view camera; and based on the initial extrinsic parameters of the second surround-view camera, the first true corner coordinates in the target parking space detection result corresponding to the second surround-view camera are projected onto the original image corresponding to the first surround-view camera. The coordinates of the first true corner point in the target parking space detection result of the second surround-view camera are projected onto the original image corresponding to the second surround-view camera. This process is repeated to achieve the same result. Based on the initial extrinsic parameters of each of the N surround-view cameras, the coordinates of the first true corner point in the target parking space detection result of each of the N surround-view cameras are projected onto the original images corresponding to the N surround-view cameras, thus obtaining the second true corner point coordinates of the first true corner point in each of the N target parking space detection results within the original images corresponding to the N surround-view cameras. Here, the second true corner point coordinates refer to the pixel coordinates of the first true corner point in the original image.

[0061] For example, if there are N surround-view cameras, including a front-view camera 101 and a left-view camera 103, then based on the initial extrinsic parameters of the front-view camera 101, the first true corner coordinates in the target parking space detection result corresponding to the front-view camera 101 are projected onto the original image corresponding to the front-view camera 101 to obtain the second true corner coordinates in the original image corresponding to the front-view camera 101. Similarly, based on the initial extrinsic parameters of the left-view camera 103, the first true corner coordinates in the target parking space detection result corresponding to the left-view camera 103 are projected onto the original image corresponding to the left-view camera 103 to obtain the second true corner coordinates in the original image corresponding to the left-view camera 103.

[0062] S130: Adjust the initial extrinsic parameters of each of the N surround-view cameras to align the second true corner coordinates in the original images corresponding to each of the N surround-view cameras with the corner coordinates in the bird's-eye view images corresponding to the original images of each of the N surround-view cameras, so as to obtain the first target extrinsic parameters of each of the N surround-view cameras.

[0063] After S120 is completed, the initial extrinsic parameters of each of the N surround-view cameras are adjusted. Then, using the adjusted initial extrinsic parameters of the N surround-view cameras, the coordinates of the second true corner points in the original images of the N surround-view cameras are mapped to the bird's-eye view images of the original images of the N surround-view cameras. This aligns the second true corner point coordinates in the original images of the N surround-view cameras within the bird's-eye view images of the original images of the N surround-view cameras. This alignment can be understood as adjusting the initial extrinsic parameters of the N surround-view cameras and performing coordinate mapping, so that the coordinates of the second true corner points in the original images of the N surround-view cameras coincide in the bird's-eye view images of the original images of the N surround-view cameras.

[0064] If the coordinates of the second true corner points in the original images corresponding to each of the N surround-view cameras are determined, and the corner point coordinates are aligned in the bird's-eye view of the original images corresponding to each of the N surround-view cameras, then when the corner point coordinates are aligned, the current extrinsic parameters of each of the N surround-view cameras are determined as the first target extrinsic parameters of each of the N surround-view cameras. That is, the first target extrinsic parameters of each of the N surround-view cameras are the final extrinsic parameters of each of the N surround-view cameras, and the extrinsic parameter calibration of the N surround-view cameras is completed.

[0065] If the coordinates of the second true corner points in the original images corresponding to each of the N surround-view cameras are determined, and the corner point coordinates are not aligned in the bird's-eye view of the original images corresponding to each of the N surround-view cameras, then the initial extrinsic parameters of each of the N surround-view cameras are adjusted again (that is, the extrinsic parameters are adjusted again based on the previous adjustment). Then, the above operation of mapping the second true corner point coordinates to the bird's-eye view is performed until the coordinates of the second true corner points in the original images corresponding to each of the N surround-view cameras are determined, and the corner point coordinates are aligned in the bird's-eye view of the original images corresponding to each of the N surround-view cameras, thereby completing the extrinsic parameter calibration of the N surround-view cameras.

[0066] This application embodiment, by employing the above-described technical solution, achieves the calibration of surround-view cameras located at multiple different shooting angles within a parking space in a parking scenario. On one hand, the calibration work does not require the setup of a calibration scene or reliance on additional calibration auxiliary equipment, making the calibration work easier to implement. This not only reduces the dependence of the calibration work on the surrounding environment but also saves the overall cost of the calibration work and improves the calibration accuracy of the surround-view camera's extrinsic parameters. On the other hand, compared to Simultaneous Localization and Mapping (SLAM) calibration schemes, the above technical solution can significantly reduce the driving difficulty for the driver, such as reducing the difficulty of maneuvering the vehicle around flower petals, thus improving the practicality of the camera extrinsic parameter calibration method provided in this application. It also reduces the computational power requirements of electronic devices, the consumption of computational power, and the complexity of the algorithm.

[0067] The following are Figure 1 The specific implementation methods of each step in the illustrated embodiment will be explained below:

[0068] In one possible implementation, the above-mentioned S110, obtaining the first detection data includes the following schemes:

[0069] Obtain M original images with the same timestamp;

[0070] Based on the initial extrinsic parameters of each of the M surround-view cameras, generate M original images of bird's-eye view, thus obtaining M bird's-eye view images;

[0071] Detect parking spaces in each of the M bird's-eye view images to obtain the first parking space detection result for each of the M bird's-eye view images;

[0072] Based on confidence and visibility indicators, the coordinates of false corner points among the multiple detected corner point coordinates included in the M first parking space detection results are filtered to obtain the M second parking space detection results;

[0073] Based on the detection results of M second parking spaces, the same parking space is matched across cameras to obtain a parking space matching list;

[0074] Obtain the detection results of N target parking spaces corresponding to the same parking space from the parking space matching list to obtain the first detection data.

[0075] M original images were captured by M surround-view cameras, each with a different shooting angle. The M surround-view cameras comprise N other surround-view cameras, where M is a positive integer greater than or equal to N. For example... Figure 2As shown, for example, M=4, the M surround view cameras are the front view camera 101, the rear view camera 102, the left view camera 103 and the right view camera 104, that is, the M surround view cameras are all the surround view cameras on the vehicle.

[0076] Each original image captured by a surround-view camera has a timestamp. Based on the same timestamp, the original images captured by M surround-view cameras are obtained, resulting in M ​​original images with the same timestamp. This ensures that the M original images are consistent in time, which helps to ensure the accuracy of the calibration of each surround-view camera.

[0077] Regarding the generation of bird's-eye view images: After obtaining M original images, inverse perspective mapping (IPM) image processing technology is used, combined with the initial extrinsic parameters of the M surround-view cameras, to generate bird's-eye view images of the M original images, resulting in M ​​bird's-eye view images. The following explanation uses the generation of a bird's-eye view image from any one of the original images as an example. The generation steps are as follows:

[0078] Step 1: Set the width and height of the IPM image and the physical distance represented by a single pixel;

[0079] Step 2: Convert the IPM image pixels into physical points in the vehicle coordinate system;

[0080] Step 3: Based on the initial extrinsic parameters of the camera and vehicle, transform the 3D points in the vehicle coordinate system to the camera coordinate system;

[0081] Step 4: Project the 3D points in the camera coordinate system into the image coordinate system using the camera intrinsic parameters to obtain the corresponding 2D pixel coordinates.

[0082] Step 5: Obtain the pixel RGB value through 2D pixel bilinear interpolation;

[0083] Step 6: Generate a mapping table for IPM image conversion;

[0084] Step 7: Complete the transformation from the original image to the IPM image.

[0085] After obtaining the mapping table, the original image can be traversed. Based on the information in the mapping table, the pixel values ​​of the original image can be copied or interpolated to the corresponding positions in the IPM image, thereby generating a complete IPM image, that is, obtaining a bird's-eye view of any original image.

[0086] Regarding parking space detection: After generating M bird's-eye view images, these images are input into a pre-trained deep learning model. The deep learning model detects parking spaces in each of the M bird's-eye view images, thus outputting the first parking space detection result for each of the M bird's-eye view images. Each first parking space detection result includes the coordinates of multiple detected corner points of the parking space, the confidence score of each corner point, and a visibility flag. The confidence score indicates the probability that the detected corner point is a corner point of the parking space, and the visibility flag indicates whether the detected corner point is visible. For example... Figure 3 As shown, Figure 3 An exemplary parking detection result diagram is shown. Figure 3 The four detection results from left to right are detection result 301, detection result 302, detection result 303 and detection result 304. The surround view camera corresponding to detection result 301 is the front-view camera, the surround view camera corresponding to detection result 302 is the left-view camera, the surround view camera corresponding to detection result 303 is the rear-view camera, and the surround view camera corresponding to detection result 304 is the right-view camera.

[0087] Regarding the selection of deep learning models, suitable deep learning architectures for object detection can be chosen, such as Faster R-CNN, YOLO (You Only Look Once), SSD (Single Shot MultiBox Detector), or Mask R-CNN. For training the deep learning model, a labeled IPM image dataset is used. This dataset contains information such as the bounding boxes, corner coordinates, confidence scores, and visibility of parking spaces. During training, the model learns how to identify the features of parking spaces and predict their location and attributes. Using deep learning models for parking space detection can improve detection efficiency.

[0088] After obtaining the first parking space detection results for each of the M bird's-eye view images, the false corner coordinates among the multiple detected corner coordinates included in the M first parking space detection results are filtered out using confidence and visibility indicators to obtain M second parking space detection results. A visibility indicator of "0" indicates that the corresponding detected corner is invisible, while a visibility indicator of "1" indicates that the corresponding detected corner is visible. A confidence level greater than a preset confidence threshold indicates that the corresponding detected corner is a parking space corner, and a confidence level less than or equal to the preset confidence threshold indicates that the corresponding detected corner is not a parking space corner. Each second parking space detection result includes true corner coordinates, which are the coordinates of the detected corners other than the false corner coordinates. False corner coordinates are determined by the fact that the detected corner coordinates with a confidence level less than or equal to the preset confidence threshold and a visibility indicator of "0" are false corner coordinates. Therefore, the other detected corner coordinates other than the false corner coordinates are the true corner coordinates. By filtering out false corner coordinates from the multiple detection corner coordinates included in the M first parking space detection results, it is beneficial to improve the efficiency of matching the same parking space across cameras.

[0089] Regarding the matching of the same parking space between cameras: The operation of matching the same parking space between cameras is to determine the same parking space captured by multiple surround-view cameras located at different shooting angles, thereby obtaining multiple parking space detection results corresponding to the same parking space. That is, the multiple parking space detection results corresponding to the same parking space are obtained by the bird's-eye view image corresponding to the original image captured by multiple surround-view cameras located at different shooting angles.

[0090] After obtaining M second parking space detection results, cross-camera matching of the same parking space is performed based on the M second parking space detection results to obtain a parking space matching list. The parking space matching list includes multiple parking spaces and N second parking space detection results corresponding to each parking space. For example, parking space A corresponds to 2 second parking space detection results. These 2 second parking space detection results are obtained from the bird's-eye view images corresponding to the original images captured by the front-view camera 101 and the left-view camera 103.

[0091] After generating the parking space matching list, the detection results of N target parking spaces corresponding to the same parking space are obtained from the parking space matching list to obtain the first detection data. For example, if the detection results of two second parking spaces corresponding to parking space A are obtained from the parking space matching list, then the first detection data includes the detection results of the two second parking spaces corresponding to parking space A.

[0092] Given a fixed resolution and physical extent of the entire bird's-eye view image, the parking space bounding boxes detected in the bird's-eye view image are used as the area occupied by the entire parking space. Theoretically, parking spaces in the same physical space should have completely overlapping bounding boxes detected in bird's-eye view images from different surround-view cameras. However, due to errors in the extrinsic parameters of each surround-view camera, there may be slight misalignment areas between the physical parking spaces in the 3D space in the bird's-eye view images corresponding to different surround-view cameras. This application uses the similarity between parking spaces across cameras as the error calculation method for the Hungarian matching algorithm, modeling the problem as an optimal allocation problem to obtain the matching results of parking spaces across cameras. This result is then used for subsequent optimization of the extrinsic parameters of the parking spaces. That is, this application uses the similarity between parking spaces across cameras as the error calculation method for the Hungarian matching algorithm to generate a parking space matching list. The above-mentioned matching of the same parking space across cameras based on the detection results of M second parking spaces yields a parking space matching list including the following schemes:

[0093] First, any one of the M second parking space detection results is used as the third parking space detection result, meaning the third parking space detection result is any one of the M second parking space detection results. All other second parking space detection results from the M second parking space detection results (excluding the third parking space detection result) are placed into a detection result set, meaning the detection result set includes all second parking space detection results from the M second parking space detection results (excluding the third parking space detection result). Second, based on the third parking space detection result and each second parking space detection result in the detection result set, the similarity between the parking space corresponding to the third parking space detection result and the parking space corresponding to each second parking space detection result in the detection result set is determined, resulting in multiple similarity scores. Then, the similarity scores greater than a preset similarity score are determined as the target similarity score, and the parking spaces corresponding to the N second parking space detection results corresponding to the target similarity score are determined to be the same parking space. Finally, the same parking space and the N second parking space detection results corresponding to the target similarity score are associated to generate a parking space matching list, meaning the parking space matching list includes multiple parking spaces and the N second parking space detection results corresponding to each of the multiple parking spaces.

[0094] like Figure 4 As shown, Figure 4 The image shows the result of matching the same parking space across cameras, for example. Figure 4401 indicates the cross-camera parking space matching result A for the same parking space group A (including 2 parking spaces) located on the left front side of the vehicle, as captured by the left-view camera and the front-view camera. The light-colored parking space line 4011 in the cross-camera parking space matching result A corresponds to the actual parking space group A captured by the left-view camera, and the dark-colored parking space line 4012 corresponds to the actual parking space group A captured by the front-view camera. Due to the different settings of the front-view camera and the left-view camera on the vehicle, the light-colored parking space line 4011 and the dark-colored parking space line 4012 in the cross-camera parking space matching result A do not completely overlap, but have a deviation.

[0095] Figure 4 402 indicates the cross-camera parking space matching result B for the same parking space group B (including 2 parking spaces) located on the right front side of the vehicle, as captured by the right-view camera and the front-view camera. The light-colored parking space line 4021 in the cross-camera parking space matching result B corresponds to the actual parking space group B captured by the right-view camera, and the dark-colored parking space line 4022 corresponds to the actual parking space group B captured by the front-view camera. Due to the different settings of the front-view camera and the right-view camera on the vehicle, the light-colored parking space line 4021 and the dark-colored parking space line 4022 in the cross-camera parking space matching result B do not completely overlap, but have a deviation.

[0096] Figure 4 403 represents the cross-camera parking space matching result C of the same parking space group C (including 2 parking spaces) located on the left rear side of the vehicle, as captured by the left-view camera and the rear-view camera. The light-colored parking space line 4031 in the cross-camera parking space matching result C corresponds to the actual parking space group C captured by the left-view camera, and the dark-colored parking space line 4022 corresponds to the actual parking space group C captured by the rear-view camera. Due to the different settings of the rear-view camera and the left-view camera on the vehicle, the light-colored parking space line 4031 and the dark-colored parking space line 4032 in the cross-camera parking space matching result C do not completely overlap, but have a deviation.

[0097] Figure 4 404 indicates the cross-camera parking space matching result D for the same parking space group D (including 2 parking spaces) located on the right rear side of the vehicle, as captured by the right-view camera and the rear-view camera. The light-colored parking space line 4041 in the cross-camera parking space matching result D corresponds to the actual parking space group D captured by the right-view camera, and the dark-colored parking space line 4042 corresponds to the actual parking space group D captured by the rear-view camera. Due to the different settings of the rear-view camera and the right-view camera on the vehicle, the light-colored parking space line 4041 and the dark-colored parking space line 4042 in the cross-camera parking space matching result D do not completely overlap, but have a deviation.

[0098] In the process of matching the same parking space across cameras, to improve the accuracy and reliability of matching and avoid erroneous matching, the Intersection over Union (IoU) ratio is used as the error calculation method for the Hungarian matching algorithm, that is, the IoU ratio represents the aforementioned similarity. Therefore, the method for determining the similarity between the parking space corresponding to the third parking space detection result and the parking space corresponding to each second parking space detection result in the detection result set, based on the third parking space detection result and each second parking space detection result in the detection result set, includes the following schemes:

[0099] The first parking space detection result obtained through a deep learning model includes not only the coordinates of detected corner points, confidence scores, and visibility indicators, but also a corner point order indicator for each detected corner point. This order indicator represents the sequence of the detected corner points, facilitating the subsequent construction of the parking space bounding box. For example, connecting the detected corner points corresponding to each parking space in a clockwise or counterclockwise direction yields the parking space bounding box. Because the first parking space detection result includes corner point order indicators, the third parking space detection result also includes corner point order indicators.

[0100] For similarity calculation: First, based on the true corner coordinates and corner order identifiers included in the third parking space detection result, construct the first parking space bounding box corresponding to the third parking space detection result. Then, based on the true corner coordinates and corner order identifiers included in the second parking space detection result in the detection result set, construct the second parking space bounding box corresponding to the second parking space detection result in the detection result set. Next, determine the intersection-union ratio (IUU) between the first and second parking space bounding boxes. Finally, use the IUU as the similarity between the parking space corresponding to the third parking space detection result and the parking space corresponding to each second parking space detection result in the detection result set.

[0101] In one possible implementation, in S130 above, adjusting the initial extrinsic parameters of each of the N surround-view cameras to align the second true corner coordinates in the original images corresponding to each of the N surround-view cameras with the corner coordinates in the bird's-eye view images of the original images corresponding to each of the N surround-view cameras, and obtaining the first target extrinsic parameters of each of the N surround-view cameras, includes the following schemes:

[0102] Adjust the initial extrinsic parameters of each of the N surround-view cameras to obtain the operating extrinsic parameters of each of the N surround-view cameras;

[0103] Based on the working extrinsic parameters of each of the N surround-view cameras, the coordinates of the second true corner points in the original images corresponding to each of the N surround-view cameras are projected onto the bird's-eye view images corresponding to the original images corresponding to each of the N surround-view cameras, so as to obtain the coordinates of the second true corner points in the original images corresponding to each of the N surround-view cameras and the coordinates of the third true corner points in the bird's-eye view images corresponding to the original images corresponding to each of the N surround-view cameras.

[0104] Determine whether the coordinates of the third true corner point in the bird's-eye view of the original images corresponding to N surround-view cameras satisfy the preset conditions for corner point coordinate alignment;

[0105] If so, the working extrinsic parameters of each of the N surround-view cameras when the preset conditions are met are determined as the first target extrinsic parameters of each of the N surround-view cameras;

[0106] If not, execute the steps of taking the working extrinsic parameters of each of the N surround-view cameras as their initial extrinsic parameters when the preset conditions are not met, and then adjust the initial extrinsic parameters of each of the N surround-view cameras to obtain their working extrinsic parameters.

[0107] The following explanation will be based on the example of N=2, with two surround-view cameras: a front-view camera 101 and a left-view camera 103.

[0108] Adjust the initial extrinsic parameters of the forward-looking camera 101 and the left-looking camera 103 respectively to obtain the working extrinsic parameters of the forward-looking camera 101 and the left-looking camera 103.

[0109] Based on the operating extrinsic parameters of the forward-looking camera 101, the second true corner coordinates in the original image corresponding to the forward-looking camera 101 are projected onto the bird's-eye view image of the original image corresponding to the forward-looking camera 101, thereby obtaining the second true corner coordinates in the original image corresponding to the forward-looking camera 101 and the third true corner coordinates in the bird's-eye view image of the original image corresponding to the forward-looking camera 101. Then, it is determined whether the third true corner coordinates in the bird's-eye view image of the original image corresponding to the forward-looking camera 101 satisfy the preset conditions for corner coordinate alignment. If they satisfy the preset conditions, then the result is... The working extrinsic parameters of the forward-looking camera 101 are determined as the first target extrinsic parameters of the forward-looking camera 101. If the conditions are not met, the working extrinsic parameters of the forward-looking camera 101 when the preset conditions are not met are taken as the initial extrinsic parameters of the forward-looking camera 101. The process of adjusting the initial extrinsic parameters of the forward-looking camera 101 to obtain the working extrinsic parameters of the forward-looking camera 101 is carried out until the preset conditions are met, thereby completing the extrinsic parameter calibration of the forward-looking camera 101.

[0110] Similarly, based on the operating extrinsic parameters of the left-view camera 103, the second true corner coordinates in the original image corresponding to the left-view camera 103 are projected onto the bird's-eye view image of the original image corresponding to the left-view camera 103, thereby obtaining the second true corner coordinates in the original image corresponding to the left-view camera 103 and the third true corner coordinates in the bird's-eye view image of the original image corresponding to the left-view camera 103. Then, it is determined whether the third true corner coordinates in the bird's-eye view image of the original image corresponding to the left-view camera 103 satisfy the preset conditions for corner coordinate alignment. If they do, then the preset conditions are satisfied. The working extrinsic parameters of the left-view camera 103 are determined as the first target extrinsic parameters of the left-view camera 103. If the conditions are not met, the working extrinsic parameters of the left-view camera 103 that do not meet the preset conditions are used as the initial extrinsic parameters of the left-view camera 103. The process of adjusting the initial extrinsic parameters of the left-view camera 103 to obtain the working extrinsic parameters of the left-view camera 103 is carried out until the preset conditions are met, thereby completing the extrinsic parameter calibration of the left-view camera 103.

[0111] In one possible implementation, the determination of whether the third true corner coordinates in the bird's-eye view images of the original images corresponding to the N surround-view cameras satisfy the preset conditions for corner coordinate alignment includes the following schemes:

[0112] Based on the corner point sequence identifiers in the detection results of N target parking spaces, determine the corner point connection line formed by the coordinates of the third true corner point in each of the N bird's-eye view images;

[0113] Determine the point-to-line distance between the target corner coordinates and the line connecting the target, to obtain K point-to-line distances, where K is a positive integer greater than or equal to 1;

[0114] Sum the distances between K points to obtain the total distance.

[0115] If the sum of distances is greater than or equal to a preset threshold, the coordinates of the third true corner point in the bird's-eye view of the original images corresponding to the N panoramic cameras do not meet the preset conditions.

[0116] If the sum of distances is greater than or equal to a preset threshold, determine the coordinates of the third true corner point in the bird's-eye view of the original images corresponding to the N panoramic cameras, which meet the preset conditions.

[0117] This application constructs point-line constraints between parking spaces by obtaining multiple parking space detection results and corner point sequence identifiers corresponding to the same parking space through cross-camera matching of the same parking space. The purpose of using point-line constraints instead of point-point constraints in this application is that the corner points of parking spaces are often obstructed or invisible due to the field of view. Using point-line constraints can significantly increase the algorithm's tolerance to corner point occlusion, thereby improving the overall robustness of the program. Specifically, the above-mentioned process of projecting the second true corner point coordinates in the original images corresponding to the N surround-view cameras into the bird's-eye view images corresponding to the original images of the N surround-view cameras, based on the working extrinsic parameters of the N surround-view cameras respectively, to obtain the second true corner point coordinates in the original images of the N surround-view cameras and the third true corner point coordinates in the bird's-eye view images corresponding to the original images of the N surround-view cameras, belongs to the original observation process of constructing point-line constraints.

[0118] This application determines the coordinates of the second true corner points in the original images corresponding to N surround-view cameras. Whether corner point coordinate alignment is achieved in the bird's-eye view of the original images corresponding to the N surround-view cameras is determined using point-line constraints. The specific determination process is as follows:

[0119] Take any third true corner coordinate from any of the N bird's-eye view images as the target corner coordinate. That is, the target corner coordinate is any third true corner coordinate from any of the N bird's-eye view images. Place all bird's-eye view images except the one containing the target corner coordinate into a bird's-eye view image set. That is, the bird's-eye view image set includes all bird's-eye view images except the one containing the target corner coordinate. Take the line connecting the corner points corresponding to the target corner coordinate in each of the bird's-eye view images in the bird's-eye view image set as the target line. That is, the target line is the line connecting the corner points corresponding to the target corner coordinate in each of the bird's-eye view images in the bird's-eye view image set. Note that the bird's-eye view image containing the target corner coordinate and the bird's-eye view image containing the line connecting the target corner coordinate are not the same bird's-eye view image. Within the same bird's-eye view image, there may be one or two lines connecting the corner points corresponding to the target corner coordinate.

[0120] Taking N=2 as an example, based on the corner point sequence identifiers in the two target parking space detection results, the corner point lines formed by the coordinates of the third true corner points in each of the two bird's-eye view images are determined. For example, the four true corner points described by the corner point sequence identifiers in the first target parking space detection result are in a counter-clockwise order. The first bird's-eye view image corresponding to the first target parking space detection result includes four third true corner point coordinates, namely A11, A12, A13, and A14. The order of the true corner points is A11-A12-A13-A14, where A11 is the starting point. Therefore, the corner point lines formed include A11A12, A12A13, A13A14, and A14A11. Similarly, the order of the four real corner points described by the corner point sequence identifier in the second target parking space detection result is also counterclockwise. The second bird's-eye view image corresponding to the second target parking space detection result includes four third real corner point coordinates, namely A21, A22, A23, and A24. The order of the real corner points is A21-A22-A23-A24, where A21 is the starting point, and the corner point connection lines include A21A22, A22A23, A23A24, and A24A21. If the target corner point coordinate is A11, then the target connection lines include A21A22 and A24A21. Where N=2, the surround-view cameras corresponding to the two bird's-eye view images are adjacent cameras.

[0121] After determining the line connecting the third true corner points in each of the N bird's-eye view images, the point-to-line distance between the target corner point coordinates and the target line is calculated to obtain K point-to-line distances. These K point-to-line distances are then summed to obtain the total distance. The point-to-line distance can be understood as the point-to-line error of the parking space between two surround-view cameras. The summation operation can be a weighted summation or a direct summation.

[0122] After obtaining the distance sum, compare it with a preset threshold. If the distance sum is greater than or equal to the preset threshold, it indicates that the point-to-line error of the parking space is large. In this case, the coordinates of the third true corner point in the bird's-eye view of the original images corresponding to each of the N surround-view cameras do not meet the preset conditions. Therefore, it is necessary to continue adjusting the extrinsic parameters of each of the N surround-view cameras and map the second true corner point coordinates to the bird's-eye view. If the distance sum is less than the preset threshold, it indicates that the point-to-line error of the parking space is small and has met expectations. In this case, it is considered that the coordinates of the third true corner point in the bird's-eye view of the original images corresponding to each of the N surround-view cameras meet the preset conditions. Therefore, it is not necessary to continue adjusting the extrinsic parameters of each of the N surround-view cameras, and the current extrinsic parameters of the N surround-view cameras are considered to be the desired final result.

[0123] Because the parking space detection results output by deep learning models typically contain some detection noise, the corner points of the detected parking spaces will fluctuate around the ground truth, and in extreme cases, false detections may even occur. The extrinsic parameter calibration method performed in S110-S130 above is an extrinsic parameter calibration method using single-frame surround view data, which is inevitably affected by the detection effect of the current frame, resulting in low consistency of the overall calibration results and even erroneous calibration results. Therefore, in order to improve calibration accuracy, this application uses multi-frame surround view data for extrinsic parameter calibration to improve the accuracy of extrinsic parameter calibration for each surround view camera.

[0124] In one possible implementation, in S130 above, the initial extrinsic parameters of each of the N surround-view cameras are adjusted to align the second true corner coordinates in the original images corresponding to each of the N surround-view cameras with the corner coordinates in the bird's-eye view images corresponding to the original images of the N surround-view cameras, and after obtaining the first target extrinsic parameters of each of the N surround-view cameras, the above camera extrinsic parameter calibration method further includes the following scheme:

[0125] Obtain N second detection data points; where each of the N second detection data points has a different timestamp.

[0126] For each second detection data, the second detection data is used as the first detection data, and the first true corner coordinates in the target parking space detection results of the N surround view cameras are projected onto the original images of the N surround view cameras based on the initial extrinsic parameters of the N surround view cameras, so as to obtain the second true corner coordinates of the first true corner coordinates in the target parking space detection results of the N surround view cameras in the original images of the N surround view cameras. The initial extrinsic parameters of the N surround view cameras are adjusted to align the second true corner coordinates in the original images of the N surround view cameras in the bird's-eye view images of the original images of the N surround view cameras, so as to obtain the N second target extrinsic parameters of the N surround view cameras.

[0127] For each surround-view camera, construct the histogram corresponding to the surround-view camera based on the N extrinsic parameters of the second target of the surround-view camera;

[0128] The second target extrinsic parameter corresponding to the highest peak value in the histogram is determined as the third target extrinsic parameter of the surround-view camera, so as to obtain the third target extrinsic parameters of each of the N surround-view cameras;

[0129] The first target extrinsic parameters of each of the N surround-view cameras are updated using the third target extrinsic parameters of each of the N surround-view cameras.

[0130] Regarding the extrinsic parameter calibration using multi-frame surround view data: N second detection data points are acquired. Each second detection data point includes the detection results of N target parking spaces corresponding to the same parking space. The timestamps of the second detection data points are different from those of the first detection data points. After acquiring the N second detection data points, each second detection data point is processed in the same way as the first detection data points. That is, the second detection data points are used as the first detection data points, and steps S110 to S130 are executed to obtain the N second target extrinsic parameters for each of the N surround view cameras. The N second target extrinsic parameters for each of the N surround view cameras are then calculated. Furthermore, for each surround-view camera, a histogram corresponding to each surround-view camera is constructed using N second target extrinsic parameters of each surround-view camera. Specifically, the N second target extrinsic parameters of each surround-view camera are input into the histogram parameter estimation model to obtain the histogram corresponding to each surround-view camera. Then, it is analyzed whether the histogram conforms to the expected noise distribution. If it does, the second target extrinsic parameter corresponding to the highest peak value in the histogram is determined as the third target extrinsic parameter of each surround-view camera, thereby obtaining the third target extrinsic parameters of each of the N surround-view cameras. Finally, the third target extrinsic parameters of each of the N surround-view cameras are used to update the first target extrinsic parameters of each of the N surround-view cameras. That is, the third target extrinsic parameters of each of the N surround-view cameras are used as the first target extrinsic parameters of each of the N surround-view cameras. For example, the third target extrinsic parameter of the first surround-view camera is used as the first target extrinsic parameter of the first surround-view camera, the third target extrinsic parameter of the second surround-view camera is used as the first target extrinsic parameter of the second surround-view camera, and so on, until the third target extrinsic parameter of the Nth surround-view camera is used as the first target extrinsic parameter of the Nth surround-view camera.

[0131] Based on the above-mentioned technical solution for extrinsic parameter calibration using multi-frame surround view data, the consistency of the extrinsic parameter solution results can be improved, and it exhibits high robustness to noise interference, achieving the optimal extrinsic parameter solution for the surround view camera, which is beneficial to improving the accuracy of extrinsic parameter calibration.

[0132] The extrinsic parameter calibration solution using the aforementioned multi-frame surround view data achieves exemplary accuracy indicators including: Roll of 0.25 degrees, Pitch of 0.15 degrees, and Yaw of 0.2 degrees, which is sufficient to meet the accuracy requirements of functions such as 360-degree surround view imaging and automatic assisted parking. Based on the calibrated extrinsic parameters, the IPM stitching effect of parking space images captured by M surround view cameras is shown below. Figure 5 As shown, Figure 5 This diagram illustrates the IPM stitching effect of a surround-view parking space image. Figure 5 The center arrow indicates the direction the vehicle is facing. Figure 5The diagram shows the partial parking spaces detected by the vehicle on the left and right sides of the vehicle in reality. The global parking spaces detected by the vehicle are not shown. Dashed box 501 represents a portion of the parking spaces in a row of parking spaces on the left side of the vehicle in reality, and dashed box 502 represents a portion of the parking spaces in a row of parking spaces on the right side of the vehicle in reality.

[0133] The technical solutions provided in the above embodiments are all based on the external parameter calibration implemented by parking spaces. This application can also replace parking spaces with lane lines in driving scenarios, which can also achieve the same external parameter calibration and have the same technical effect as the external parameter calibration implemented by parking spaces.

[0134] The following are embodiments of the apparatus described in this application, which can be used to execute the embodiments of the method described in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in this application.

[0135] Figure 6 This application provides a schematic diagram of the structure of a camera extrinsic parameter calibration device according to an embodiment of the present application. Figure 6 As shown, the camera extrinsic parameter calibration device 600 includes:

[0136] The data acquisition module 610 is used to acquire first detection data; wherein, the first detection data includes the detection results of N target parking spaces corresponding to the same parking space, each target parking space detection result includes the first true corner coordinates of the same parking space, the N target parking space detection results are obtained from bird's-eye view images corresponding to N original images, the N original images are obtained from N surround view cameras, the shooting angles of the N surround view cameras are different, and N is a positive integer greater than or equal to 2;

[0137] The coordinate projection module 620 is used to project the first true corner coordinates in the target parking space detection results of the N surround view cameras to the original images corresponding to the N surround view cameras based on the initial extrinsic parameters of the N surround view cameras, so as to obtain the second true corner coordinates in the original images corresponding to the N surround view cameras.

[0138] The coordinate alignment module 630 is used to adjust the initial extrinsic parameters of each of the N surround-view cameras, so as to align the second true corner coordinates in the original images corresponding to each of the N surround-view cameras with the corner coordinates in the bird's-eye view images of the original images corresponding to each of the N surround-view cameras, thereby obtaining the first target extrinsic parameters of each of the N surround-view cameras.

[0139] In one possible implementation, the data acquisition module 610 includes:

[0140] The first processing unit is used to acquire M original images with the same timestamp; wherein the M original images are captured by M surround-view cameras, the shooting angles of the M surround-view cameras are different, the M surround-view cameras include the N surround-view cameras, and M is a positive integer greater than or equal to N;

[0141] The second processing unit is used to generate bird's-eye view images of the M original images based on the initial extrinsic parameters of the M surround-view cameras, thereby obtaining M bird's-eye view images;

[0142] The third processing unit is used to detect parking spaces in each of the M bird's-eye view images to obtain the first parking space detection result corresponding to each of the M bird's-eye view images. Each first parking space detection result includes the coordinates of multiple detection corner points of the parking space, the confidence level and visibility identifier of each of the multiple detection corner points, and the visibility identifier is used to indicate whether the detection corner point is a visible corner point.

[0143] The fourth processing unit is used to filter the false corner coordinates among the multiple detected corner coordinates included in the M first parking space detection results based on the confidence level and the visibility identifier, so as to obtain M second parking space detection results; wherein, each second parking space detection result includes true corner coordinates, and the true corner coordinates are the detected corner coordinates other than the false corner coordinates among the multiple detected corner coordinates;

[0144] The fifth processing unit is used to perform cross-camera matching of the same parking space based on the M second parking space detection results to obtain a parking space matching list; wherein, the parking space matching list includes multiple parking spaces and N second parking space detection results corresponding to each of the multiple parking spaces;

[0145] The sixth processing unit is used to obtain the detection results of N target parking spaces corresponding to the same parking space from the parking space matching list, and obtain the first detection data.

[0146] In one possible implementation, the fifth processing unit includes:

[0147] The calculation subunit is used to determine the similarity between the parking space corresponding to the third parking space detection result and the parking space corresponding to each second parking space detection result in the detection result set, based on the third parking space detection result and each second parking space detection result in the detection result set, thereby obtaining multiple similarities; wherein, the third parking space detection result is any second parking space detection result, and the detection result set includes the second parking space detection results other than the third parking space detection result among the M second parking space detection results;

[0148] The determination subunit is used to determine the similarity among the multiple similarities that is greater than a preset similarity as the target similarity, and to determine the parking spaces corresponding to the N second parking space detection results corresponding to the target similarity as the same parking space;

[0149] A sub-unit is generated to associate the detection results of the same parking space and N second parking spaces corresponding to the target similarity to generate the parking space matching list.

[0150] In one possible implementation, the computational subunit is specifically configured to: construct a first parking space bounding box corresponding to the third parking space detection result based on the actual corner coordinates and corner order identifiers included in the third parking space detection result; construct a second parking space bounding box corresponding to each second parking space detection result in the detection result set based on the actual corner coordinates and corner order identifiers included in each second parking space detection result in the detection result set; determine the intersection-union ratio (IUGR) between the first parking space bounding box and the second parking space bounding box; and use the IUGR as the similarity between the parking space corresponding to the third parking space detection result and the parking space corresponding to each second parking space detection result in the detection result set.

[0151] In one possible implementation, the coordinate alignment module 630 includes:

[0152] An adjustment unit is used to adjust the initial extrinsic parameters of each of the N surround-view cameras to obtain the operating extrinsic parameters of each of the N surround-view cameras;

[0153] The mapping unit is used to project the second true corner coordinates in the original images corresponding to the N surround-view cameras to the bird's-eye view image of the original images corresponding to the N surround-view cameras based on the working extrinsic parameters of the N surround-view cameras, so as to obtain the second true corner coordinates in the original images corresponding to the N surround-view cameras and the third true corner coordinates in the bird's-eye view image of the original images corresponding to the N surround-view cameras.

[0154] The judgment unit is used to determine whether the third true corner coordinates in the bird's-eye view of the original images corresponding to each of the N surround-view cameras meet the preset conditions for corner coordinate alignment; if yes, the working extrinsic parameters of each of the N surround-view cameras when the preset conditions are met are determined as the first target extrinsic parameters of each of the N surround-view cameras; if no, the working extrinsic parameters of each of the N surround-view cameras when the preset conditions are not met are used as the initial extrinsic parameters of each of the N surround-view cameras, and the step of adjusting the initial extrinsic parameters of each of the N surround-view cameras to obtain the working extrinsic parameters of each of the N surround-view cameras is executed.

[0155] In one possible implementation, the judging unit, in determining whether the third true corner coordinates in the bird's-eye view images corresponding to the original images of the N surround-view cameras satisfy the preset conditions for corner coordinate alignment, is specifically used to determine the corner line formed by the third true corner coordinates in each of the N bird's-eye view images based on the corner order identifiers in the detection results of the N target parking spaces; and to determine the point-to-line distance between the target corner coordinates and the target line to obtain K point-to-line distances; where K is a positive integer greater than or equal to 1, the target corner coordinates are any third true corner coordinates in any bird's-eye view image, and the target line... A line is drawn connecting the corner points corresponding to the target corner point coordinates in each of the bird's-eye view images included in the bird's-eye view image set, which includes bird's-eye view images other than the one containing the target corner point coordinates. The distances between the K points and lines are summed to obtain a distance sum. If the distance sum is greater than or equal to a preset threshold, it is determined that the third true corner point coordinates in the bird's-eye view images of the original images corresponding to the N surround-view cameras do not meet the preset condition. If the distance sum is greater than or equal to the preset threshold, it is determined that the third true corner point coordinates in the bird's-eye view images of the original images corresponding to the N surround-view cameras meet the preset condition.

[0156] In one possible implementation, the camera extrinsic calibration device 600 further includes:

[0157] An optimization unit is used to acquire N second detection data points, wherein the timestamps corresponding to the N second detection data points are different. For each second detection data point, the second detection data point is used as the first detection data point, and the process of projecting the first true corner coordinates in the target parking space detection results corresponding to the N surround view cameras onto the original images corresponding to the N surround view cameras is performed based on the initial extrinsic parameters of the N surround view cameras. This process obtains the second true corner coordinates in the original images corresponding to the N surround view cameras, and adjusts the initial extrinsic parameters of the N surround view cameras to adjust the first true corner coordinates in the target parking space detection results. The second true corner coordinates of the original images corresponding to the N surround-view cameras are used to perform corner coordinate alignment in the bird's-eye view of the original images corresponding to the N surround-view cameras to obtain N second target extrinsic parameters for each of the N surround-view cameras. For each surround-view camera, a histogram corresponding to the surround-view camera is constructed based on the N second target extrinsic parameters. The second target extrinsic parameter corresponding to the highest peak value in the histogram is determined as the third target extrinsic parameter of the surround-view camera to obtain the third target extrinsic parameters for each of the N surround-view cameras. The first target extrinsic parameters of each of the N surround-view cameras are updated using the third target extrinsic parameters of each of the N surround-view cameras.

[0158] It should be noted that the camera extrinsic parameter calibration device provided in the above embodiments is only illustrated by the division of the above functional modules when performing the camera extrinsic parameter calibration method. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the camera extrinsic parameter calibration device and the camera extrinsic parameter calibration method embodiments provided in the above embodiments belong to the same concept. Therefore, for details not disclosed in the device embodiments of this application, please refer to the embodiments of the camera extrinsic parameter calibration method described above in this application, which will not be repeated here.

[0159] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0160] Figure 7 This application provides a schematic diagram of the structure of an electronic device according to an embodiment of the present application. Figure 7 As shown, the electronic device 700 includes a memory 701 and a processor 702. The memory 701 stores executable program code 7011, and the processor 702 is used to call and execute the executable program code 7011 to perform a camera extrinsic calibration method.

[0161] This embodiment can divide the electronic device into functional modules according to the above method example. For example, each module can correspond to a separate functional module, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware. It should be noted that the module division in this embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0162] When each functional module is divided according to its corresponding function, the electronic device may include: a data acquisition module, a coordinate projection module, a coordinate alignment module, etc. It should be noted that all relevant content of each step involved in the above method embodiments can be referenced from the functional description of the corresponding functional module, and will not be repeated here.

[0163] The electronic device provided in this embodiment is used to execute the above-described camera extrinsic parameter calibration method, and thus can achieve the same effect as the above-described implementation method.

[0164] When using integrated units, the electronic device may include a processing module and a storage module. The processing module is used to control and manage the operation of the electronic device. The storage module is used to support the execution of relevant program code and data by the electronic device.

[0165] The processing module may be a processor or a controller, which can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. The processor may also be a combination of functions that implement computing capabilities, such as a combination of one or more microprocessors, a combination of digital signal processing (DSP) and a microprocessor, etc., and the storage module may be a memory.

[0166] This embodiment also provides a computer-readable storage medium storing computer program code. When the computer program code is run on a computer, the computer executes the above-described related method steps to implement a camera extrinsic parameter calibration method in the above embodiment.

[0167] This embodiment also provides a computer program product that, when run on a computer, causes the computer to perform the aforementioned related steps to implement a camera extrinsic parameter calibration method as described in the above embodiment.

[0168] In addition, the electronic device provided in the embodiments of this application may specifically be a chip, component or module. The electronic device may include a connected processor and a memory. The memory is used to store instructions. When the electronic device is running, the processor may call and execute the instructions to make the chip execute a camera extrinsic calibration method in the above embodiments.

[0169] In this embodiment, the electronic device, computer-readable storage medium, computer program product or chip are all used to execute the corresponding camera extrinsic calibration method provided above. Therefore, the beneficial effects that can be achieved can be referred to the beneficial effects in the corresponding camera extrinsic calibration method provided above, and will not be repeated here.

[0170] Through the above description of the embodiments, those skilled in the art will understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0171] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0172] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for calibrating camera extrinsic parameters, characterized in that, The camera extrinsic parameter calibration method comprises: obtaining first detection data; wherein the first detection data comprises N target parking space detection results corresponding to the same parking space, each target parking space detection result comprises a first real corner point coordinate of the same parking space, the N target parking space detection results are obtained from bird's eye view images corresponding to N original images, the N original images are captured by N surround view cameras, the shooting angles of the N surround view cameras are different, and N is a positive integer greater than or equal to 2; projecting the first real corner point coordinates in the target parking space detection results corresponding to the N surround view cameras respectively based on the initial extrinsic parameters of the N surround view cameras respectively to obtain second real corner point coordinates of the first real corner point coordinates in the N target parking space detection results respectively in the original images corresponding to the N surround view cameras respectively; adjusting the initial extrinsic parameters of the N surround view cameras respectively to align the second real corner point coordinates in the original images corresponding to the N surround view cameras respectively in the bird's eye view images of the original images corresponding to the N surround view cameras respectively to obtain the first target extrinsic parameters of the N surround view cameras respectively.

2. The camera extrinsic calibration method of claim 1, wherein, The first detection data comprises: obtaining M original images with the same timestamp; wherein the M original images are captured by M surround view cameras, the shooting angles of the M surround view cameras are different, the M surround view cameras comprise the N surround view cameras, and M is a positive integer greater than or equal to N; generating bird's eye view images of the M original images respectively according to the initial extrinsic parameters of the M surround view cameras to obtain M bird's eye view images; detecting parking spaces in the M bird's eye view images respectively to obtain first parking space detection results corresponding to the M bird's eye view images respectively, each first parking space detection result comprising detection corner point coordinates of a plurality of detection corners of a parking space, confidence degrees of the plurality of detection corners respectively, and visibility identifiers, the visibility identifiers being used to indicate whether the detection corners are visible corners; screening false corner point coordinates in a plurality of detection corner point coordinates included in the M first parking space detection results based on the confidence degrees and the visibility identifiers to obtain M second parking space detection results; wherein each second parking space detection result comprises real corner point coordinates, the real corner point coordinates being detection corner point coordinates other than the false corner point coordinates in the plurality of detection corner point coordinates; performing same parking space matching across cameras based on the M second parking space detection results to obtain a parking space matching list; wherein the parking space matching list comprises a plurality of parking spaces and N second parking space detection results corresponding to the plurality of parking spaces respectively; obtaining N target parking space detection results corresponding to the same parking space from the parking space matching list to obtain the first detection data.

3. The camera extrinsic calibration method of claim 2, wherein, The same parking space matching across cameras based on the M second parking space detection results to obtain a parking space matching list comprises: According to the third parking space detection result and each second parking space detection result in the detection result set, a similarity between the parking space corresponding to the third parking space detection result and the parking space corresponding to each second parking space detection result in the detection result set is determined, to obtain a plurality of similarities; wherein the third parking space detection result is any one of the second parking space detection results, and the detection result set includes the second parking space detection results other than the third parking space detection result in the M second parking space detection results; The similarity greater than the preset similarity in the plurality of similarities is determined as a target similarity, and the parking space corresponding to each of the N second parking space detection results corresponding to the target similarity is determined as the same parking space; The same parking space and the N second parking space detection results corresponding to the target similarity are associated to generate the parking space matching list.

4. The camera extrinsic calibration method of claim 3, wherein, The similarity between the parking space corresponding to the third parking space detection result and the parking space corresponding to each second parking space detection result in the detection result set is determined according to the third parking space detection result and each second parking space detection result in the detection result set, including: According to the real corner point coordinates and the corner point sequence identifiers included in the third parking space detection result, a first parking space frame of the parking space corresponding to the third parking space detection result is constructed; According to the real corner point coordinates and the corner point sequence identifiers included in each second parking space detection result in the detection result set, a second parking space frame of the parking space corresponding to each second parking space detection result in the detection result set is constructed; The intersection over union between the first parking space frame and the second parking space frame is determined; The intersection over union is taken as the similarity between the parking space corresponding to the third parking space detection result and the parking space corresponding to each second parking space detection result in the detection result set.

5. The camera extrinsic parameter calibration method of claim 1, wherein, The adjustment of the initial extrinsic parameters of the N surround view cameras to align the second real corner point coordinates in the original images corresponding to the N surround view cameras with the third real corner point coordinates in the bird's eye view images corresponding to the N surround view cameras to obtain the first target extrinsic parameters of the N surround view cameras includes: Adjusting the initial extrinsic parameters of the N surround view cameras to obtain the working extrinsic parameters of the N surround view cameras; Based on the working extrinsic parameters of the N surround view cameras, projecting the second real corner point coordinates in the original images corresponding to the N surround view cameras to the bird's eye view images corresponding to the N surround view cameras to obtain the third real corner point coordinates in the bird's eye view images corresponding to the N surround view cameras from the second real corner point coordinates in the original images corresponding to the N surround view cameras; Judging whether the third real corner point coordinates in the bird's eye view images corresponding to the N surround view cameras satisfy the preset condition for corner point coordinate alignment; If yes, the working extrinsic parameters of the N surround view cameras when the preset condition is satisfied are determined as the first target extrinsic parameters of the N surround view cameras; If not, the execution will not meet the preset condition when the N ring view cameras each work outside the reference, as the initial outside reference of the N ring view cameras each, and execute the step of adjusting the initial outside reference of the N ring view cameras each to obtain the working outside reference of the N ring view cameras each.

6. The camera extrinsic parameter calibration method of claim 5, wherein, The judgment whether the third real corner point coordinates in the bird's eye view image of the original image corresponding to the N ring view cameras each meet the preset condition about the alignment of the corner point coordinates includes: According to the corner point sequence identifier in the N target parking space detection result each, determine the corner point connection line composed of the third real corner point coordinates in the N bird's eye view image each; Determine the point-line distance between the target corner point coordinates and the target connection line to obtain K point-line distances; wherein K is a positive integer greater than or equal to 1, the target corner point coordinates are any one of the third real corner point coordinates in any one bird's eye view image, and the target connection line is the corner point connection line corresponding to the target corner point coordinates in each bird's eye view image included in the bird's eye view image set, and the bird's eye view image set includes the bird's eye view image other than the bird's eye view image where the target corner point coordinates are located in the N bird's eye view images; Sum the K point-line distances to obtain a distance sum; If the distance sum is greater than or equal to a preset threshold, it is determined that the third real corner point coordinates in the bird's eye view image of the original image corresponding to the N ring view cameras each do not meet the preset condition; If the distance sum is greater than or equal to a preset threshold, it is determined that the third real corner point coordinates in the bird's eye view image of the original image corresponding to the N ring view cameras each meet the preset condition.

7. The camera extrinsic parameter calibration method of any one of claims 1 to 6, wherein, After adjusting the initial outside reference of the N ring view cameras each to align the second real corner point coordinates in the original image corresponding to the N ring view cameras each in the bird's eye view image corresponding to the N ring view cameras each, the camera outside reference calibration method further includes: Obtain N second detection data; wherein the time stamps corresponding to the N second detection data are different; For each second detection data, the second detection data is taken as the first detection data, and the step of projecting the first real corner point coordinates in the target parking space detection result corresponding to the N ring view cameras each to the original image corresponding to the N ring view cameras each based on the initial outside reference of the N ring view cameras each to obtain the second real corner point coordinates of the first real corner point coordinates in the N target parking space detection result each in the original image corresponding to the N ring view cameras each is executed to adjust the initial outside reference of the N ring view cameras each to align the second real corner point coordinates in the original image corresponding to the N ring view cameras each in the bird's eye view image corresponding to the N ring view cameras each to obtain the N second target outside reference of the N ring view cameras each; For each ring view camera, construct the histogram corresponding to the ring view camera based on the N second target outside reference of the ring view camera; The second target extrinsic parameter corresponding to the highest peak value in the histogram is determined as a third target extrinsic parameter of the surround-view camera, to obtain the third target extrinsic parameter of each of the N surround-view cameras. The first target extrinsic parameter of each of the N surround-view cameras is updated by using the third target extrinsic parameter of each of the N surround-view cameras.

8. A camera extrinsic parameter calibration apparatus, characterized in that, The camera extrinsic parameter calibration apparatus comprises: The data acquisition module is configured to acquire first detection data, wherein the first detection data comprises N target parking space detection results corresponding to a same parking space, each target parking space detection result comprises a first real corner point coordinate of the same parking space, the N target parking space detection results are obtained from bird's-eye images corresponding to N original images, the N original images are captured by N surround-view cameras, the N surround-view cameras have different shooting angles, and N is a positive integer greater than or equal to 2. The coordinate projection module is configured to project the first real corner point coordinate in the target parking space detection result corresponding to each of the N surround-view cameras to an original image corresponding to each of the N surround-view cameras based on an initial extrinsic parameter of each of the N surround-view cameras, to obtain a second real corner point coordinate of the first real corner point coordinate in the N target parking space detection results in the original image corresponding to each of the N surround-view cameras. The coordinate alignment module is configured to adjust the initial extrinsic parameter of each of the N surround-view cameras, to perform corner point coordinate alignment on the second real corner point coordinate in the original image corresponding to each of the N surround-view cameras in a bird's-eye image of the original image corresponding to each of the N surround-view cameras, to obtain the first target extrinsic parameter of each of the N surround-view cameras.

9. An electronic device, comprising: The electronic device comprises: The memory is configured to store executable program code. The processor is configured to call and run the executable program code from the memory, so that the electronic device performs the camera extrinsic parameter calibration method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, when the computer program is executed, the camera extrinsic parameter calibration method according to any one of claims 1 to 7 is realized.

Images