Automatic parking repositioning method, device, equipment, storage medium and computer program product

Abstract

The invention relates to the technical field of parking positioning, in particular to an automatic parking repositioning method, device and equipment, a storage medium and a computer program product. The method comprises the following steps: when a parking enable signal is received, acquiring a vehicle look-around image set and positioning data; convolution of different Gaussian kernels is carried out on key images in the vehicle all-round view image set, difference calculation is carried out, and key points in the images are generated; based on a Newton iteration method, screening the key points to obtain feature points; based on the Euclidean distance, the feature points are matched, and matched feature point pairs are determined; determining a camera pose change value according to the matched feature point pair, the camera internal reference and the camera external reference; based on the Kalman filtering algorithm, the camera pose change value and the positioning data are fused, the current pose is determined, and the precision of automatic parking repositioning is improved.

Description

Technical Field

[0001] The present application relates to the field of parking positioning technology, and in particular to an automatic parking repositioning method, device, equipment, storage medium and computer program product. Background Art

[0002] With the continuous development of automobile technology, intelligent driving systems have gradually become one of the important functions of vehicles. Especially in urban traffic and parking scenarios, intelligent parking systems can effectively reduce the burden on drivers and improve parking efficiency and safety. The characteristics of parking scenarios include: extremely low vehicle speed, small environmental space, and high positioning accuracy requirements. Therefore, accurate automatic parking relocation technology is the key to realizing advanced parking functions. The existing automatic parking relocation technology mainly relies on specific symbols such as visual guide lines or parking space numbers, but these solutions are highly dependent on the environment and cannot flexibly adapt to various parking scenarios. At the same time, these methods have deficiencies in handling false detections, system communication bandwidth, and computing resource optimization, which makes implementation difficult and difficult to meet the needs of precise positioning. Therefore, how to improve the accuracy of automatic parking relocation has become a technical problem that needs to be solved urgently. Summary of the invention

[0003] The main purpose of the present application is to provide an automatic parking repositioning method, device, equipment, storage medium and computer program product, aiming to solve the technical problem of how to improve the accuracy of automatic parking repositioning.

[0004] To achieve the above object, the present application provides an automatic parking relocation method, the method comprising the following steps:

[0005] When a parking enable signal is received, a vehicle surround view image set and positioning data are obtained;

[0006] Performing convolution of different Gaussian kernels on the key images in the vehicle surround view image set and taking differences to generate key points in the image;

[0007] Based on the Newton iteration method, the key points are screened to obtain feature points;

[0008] Based on the Euclidean distance, the feature points are matched to determine matching feature point pairs;

[0009] Determine the camera pose change value according to the matching feature point pair, camera intrinsic parameters and camera extrinsic parameters;

[0010] Based on the Kalman filter algorithm, the camera posture change value is fused with the positioning data to determine the current posture.

[0011] In one embodiment, after the step of acquiring the vehicle surround view image set and positioning data when the parking enable signal is received, the following step is further included:

[0012] Determining the time difference and position change between adjacent vehicle surround view images according to the vehicle surround view image set and the positioning data;

[0013] Comparing the time difference with a preset time difference threshold, and comparing the position change with a preset position change threshold;

[0014] If the time difference exceeds the preset time difference threshold or the position change exceeds the preset position change threshold, the vehicle surround image set is intercepted to obtain the key image.

[0015] In one embodiment, the step of performing convolution of different Gaussian kernels on the key image in the vehicle surround view image set and taking differences to generate key points in the image includes:

[0016] Based on multiple Gaussian kernel parameters, a convolution operation is performed on the key image in the vehicle surround view image set to obtain smooth images at different scales;

[0017] Based on the difference between adjacent smoothed images, a Gaussian difference image is obtained;

[0018] Comparing each pixel in the Gaussian difference image with all the pixels in a preset adjacent range;

[0019] If the corresponding pixel point in the Gaussian difference image is a maximum value or a minimum value, the corresponding pixel point is used as the key point.

[0020] In one embodiment, the step of screening the key points based on the Newton iteration method to obtain feature points includes:

[0021] Based on the Gaussian difference image, determine the gradient of the key point, and based on the key point, construct a Hessian matrix;

[0022] Based on the Newton iteration method, the gradient of the key point and the Hessian matrix, the position of the key point is updated until the position change is less than a preset position change threshold, thereby obtaining a first point set;

[0023] Eliminate points whose contrast values ​​are greater than a preset contrast threshold in the first point set to obtain a second point set;

[0024] Based on the eigenvalues ​​of the Hessian matrix, determining whether the corresponding key point is a linear edge point;

[0025] If so, the linear edge points in the second point set are eliminated to obtain the feature points.

[0026] In one embodiment, the step of determining the camera pose change value according to the matching feature point pairs, camera intrinsic parameters and camera extrinsic parameters includes:

[0027] Convert the pixel coordinates of the matching feature point pairs into normalized coordinates;

[0028] Determine a basic matrix based on the matched feature point pairs after coordinate transformation, the camera intrinsic parameters, and the camera extrinsic parameters;

[0029] Decomposing the basic matrix to obtain a rotation matrix and a displacement vector;

[0030] The camera position change value is determined according to the rotation matrix and the displacement vector.

[0031] In one embodiment, before the step of determining the camera pose change value according to the rotation matrix and the displacement vector, the step further includes:

[0032] Based on the positioning data, deblurring the rotation matrix and the displacement vector;

[0033] Based on the rotation matrix and displacement vector after blur removal, reproject the matched feature point pairs after coordinate transformation to the key image, and obtain a reprojection error;

[0034] A preset optimization algorithm is used to minimize the re-projection error.

[0035] In addition, to achieve the above-mentioned purpose, the present application also proposes an automatic parking repositioning device, the automatic parking repositioning device comprising:

[0036] A data acquisition module, used to acquire a vehicle surround view image set and positioning data when receiving a parking enable signal;

[0037] A convolution difference module is used to perform convolution of different Gaussian kernels on the key images in the vehicle surround view image set and calculate the difference to generate key points in the image;

[0038] A key point screening module, used to screen the key points based on Newton iteration method to obtain feature points;

[0039] A feature point matching module, used to match the feature points based on the Euclidean distance to determine matching feature point pairs;

[0040] A posture change module, used to determine the camera posture change value according to the matching feature point pair, camera intrinsic parameters and camera extrinsic parameters;

[0041] The target module is used to fuse the camera posture change value with the positioning data based on the Kalman filter algorithm to determine the current posture.

[0042] In addition, to achieve the above-mentioned purpose, the present application also proposes an automatic parking relocation device, which includes: a memory, a processor, and an automatic parking relocation program stored on the memory and executable on the processor, wherein the automatic parking relocation program is configured to implement the steps of the automatic parking relocation method as described above.

[0043] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, on which an automatic parking relocation program is stored. When the automatic parking relocation program is executed by a processor, the steps of the automatic parking relocation method described above are implemented.

[0044] In addition, to achieve the above-mentioned purpose, the present application also proposes a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps of the automatic parking repositioning method as described above are implemented.

[0045] When the parking enable signal is received, the present application obtains the vehicle surround view image set and positioning data; performs convolution of different Gaussian kernels on the key images in the vehicle surround view image set and calculates the difference to generate the key points in the image; based on the Newton iteration method, the key points are screened to obtain feature points; based on the Euclidean distance, the feature points are matched to determine the matching feature point pairs; according to the matching feature point pairs, the camera intrinsic parameters and the camera extrinsic parameters, the camera posture change value is determined; based on the Kalman filter algorithm, the camera posture change value is fused with the positioning data to determine the current posture. The present application extracts key points through Gaussian kernel convolution and difference, and uses the Newton iteration method to screen feature points to improve the accuracy of feature matching; based on the matching feature point pairs, the camera internal and external parameters and the basic matrix, the posture change of the vehicle is solved, and the posture change and positioning data are fused through the Kalman filter algorithm to continuously correct the vehicle position and improve the accuracy of automatic parking repositioning. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a flow chart of the first embodiment of the automatic parking relocation method of the present application;

[0047] Figure 2 This is a schematic diagram of a sub-process in the second embodiment of the automatic parking relocation method of the present application;

[0048] Figure 3 This is a schematic diagram of a sub-process in the third embodiment of the automatic parking relocation method of the present application;

[0049] Figure 4 This is a schematic diagram of the module structure of the automatic parking relocation device according to an embodiment of the present application;

[0050] Figure 5Schematic diagram of the device structure of the hardware operating environment involved in the automatic parking relocation method in the embodiment of the present application.

[0051] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0052] It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0053] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0054] It should be noted that with the continuous development of automobile technology, intelligent driving systems have gradually become one of the important functions of vehicles. Especially in urban traffic and parking scenarios, intelligent parking systems can effectively reduce the burden on drivers and improve parking efficiency and safety. The characteristics of parking scenarios include: extremely low vehicle speed, small environmental space, and high positioning accuracy requirements. Therefore, accurate automatic parking relocation technology is the key to realizing advanced parking functions. The existing automatic parking relocation technology mainly relies on specific symbols such as visual guide lines or parking space numbers, but these solutions are highly dependent on the environment and cannot flexibly adapt to various parking scenarios. At the same time, these methods have deficiencies in handling false detections, system communication bandwidth, and computing resource optimization, which makes implementation difficult and difficult to meet the needs of precise positioning. Therefore, how to improve the accuracy of automatic parking relocation has become a technical problem that needs to be solved urgently.

[0055] The main solution of the present application is: when a parking enable signal is received, a vehicle surround view image set and positioning data are obtained; the key images in the vehicle surround view image set are convolved with different Gaussian kernels and differentiated to generate key points in the image; based on the Newton iteration method, the key points are screened to obtain feature points; based on the Euclidean distance, the feature points are matched to determine the matching feature point pairs; according to the matching feature point pairs, the camera intrinsic parameters and the camera extrinsic parameters, the camera posture change value is determined; based on the Kalman filter algorithm, the camera posture change value is fused with the positioning data to determine the current posture.

[0056] This application extracts key points through Gaussian kernel convolution and difference, and uses Newton iteration method to screen feature points to improve the accuracy of feature matching; based on the matched feature point pairs, the camera's internal and external parameters and the basic matrix solve the vehicle's posture changes, and use the Kalman filter algorithm to fuse the posture changes and positioning data to continuously correct the vehicle position and improve the accuracy of automatic parking repositioning.

[0057] It should be noted that the execution subject of the method of this embodiment can be a computing service device with data processing, network communication and program running functions, or it can be the above-mentioned automatic parking relocation device with the same or similar functions. This embodiment and the following embodiments will be described by taking the automatic parking relocation device as an example.

[0058] Based on this, the first embodiment of the automatic parking relocation method of the present application is proposed, please refer to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the automatic parking relocation method of the present application.

[0059] In this embodiment, the automatic parking relocation method includes the following steps:

[0060] S1: When a parking enable signal is received, a vehicle surround view image set and positioning data are obtained;

[0061] It should be noted that the parking enable signal refers to a control signal triggered by the system when the vehicle enters the parking mode, notifying the system to start the automatic parking process. It is a sign to turn on the parking function in the vehicle system, usually issued by the driver or after the system automatically detects environmental conditions. The surround view image set is a collection of multiple frames of images collected by multiple cameras around the vehicle (such as the front, rear, left and right cameras), which is used to perceive the vehicle's surrounding environment in real time. This image set is the main source of information for obtaining surrounding obstacles and reference objects during the parking process. The positioning data obtains the vehicle's current position and direction information through the on-board positioning module (such as GPS, IMU, odometer, etc.), which is used to determine the vehicle's spatial position and posture (position) to provide support for path planning and position adjustment.

[0062] Specifically, when the parking enable signal is triggered, the system enters the automatic parking mode and first calls the onboard surround view system to collect surround view images around the vehicle. These images are provided by multiple cameras, including real-time video frames from the front, back, left and right of the vehicle. The system stitches and processes these image frames to form a 360-degree environmental view around the vehicle. This process ensures that the system fully perceives the obstacles and sign information in the parking environment, preparing for the subsequent extraction and matching of feature points.

[0063] Furthermore, the positioning data of the vehicle is obtained synchronously, including position and attitude information (such as the current direction angle and driving distance of the vehicle). These data are provided by on-board sensors (such as IMU or GPS) and vehicle odometers, ensuring that the system has an accurate grasp of the vehicle's own motion state while processing image data. Through this combination of image and positioning data, the system can determine the position of the vehicle in three-dimensional space and provide a basis for subsequent posture change calculations.

[0064] First, the simultaneous acquisition of surround view image sets and positioning data ensures that the system can perceive the dynamic changes of the vehicle's surrounding environment in real time in the parking lot, and quickly adjust the perception range according to the vehicle's motion state. This real-time performance provides strong support for the precise navigation of the automatic parking system, reduces the reliance on a single sensor, and improves the robustness of the system. Secondly, the fusion of image and positioning data in this step lays a solid foundation for feature point matching and posture change calculation in subsequent steps. By accurately obtaining the initial position and direction of the vehicle, the system can perform path planning and error correction faster, improving the safety and efficiency of the parking process. At the same time, since it does not rely on the external guidance facilities of the parking lot, this method has high universality and is suitable for different types of parking scenarios.

[0065] S2: performing convolution with different Gaussian kernels on the key images in the vehicle surround view image set and calculating the difference to generate key points in the image;

[0066] It should be noted that key images refer to image frames with important information selected from the surround image set. They are usually selected based on time intervals, vehicle motion status or posture changes to ensure that computing resources are concentrated on key images. Gaussian kernel is a mathematical filter used for image smoothing to reduce noise and details. The convolution result is a smoothed version of the image, which provides a basis for subsequent image difference and key point detection. Difference of Gaussian (DOG) is to differentiate adjacent smooth images to enhance the edges and important features in the image and highlight potential key points.

[0067] Specifically, after acquiring the vehicle surround image set, multiple Gaussian kernels of different scales are applied to each key image for convolution. The purpose of convolution is to reduce the noise interference of the image while retaining the main structural information. In this process, different Gaussian kernels (i.e., standard deviations of different scales) are used to smooth the image to form a series of smoothed images.

[0068] Furthermore, the smooth images at different scales are subjected to difference calculation (DOG). Specifically, the difference between smooth images at every two adjacent scales is calculated. This process can highlight the significant change areas such as edges and corners in the image. In the difference image, each pixel is compared with its adjacent pixels. If the pixel is a local maximum or minimum in a specific area, it is marked as a potential key point.

[0069] Through multi-scale Gaussian kernel convolution and smoothed differential operations, not only the noise in the image is effectively reduced, but also the important features in the image are highlighted, making the key points more obvious. This feature extraction method ensures the robustness of the system in complex parking scenarios. Whether it is light changes or noise interference, the system can accurately capture key environmental information. In addition, the design of multi-scale convolution ensures that the system can detect feature points of different sizes, so that the system can accurately identify and match when facing various obstacles (such as vehicles of different sizes, roadblocks, etc.). This lays the foundation for subsequent feature point screening and matching, and improves the accuracy and efficiency of parking relocation.

[0070] S3: Based on the Newton iteration method, the key points are screened to obtain feature points;

[0071] It should be noted that key points are potential important points detected in an image by the Difference of Gaussian (DOG) algorithm. These points are usually edges, corners or significant change areas of the image. Feature points are important points with high stability and high discrimination selected from key points. They can be matched between different images and used to calculate the posture change of the vehicle.

[0072] Specifically, in the set of key points initially detected, the system uses the Newton iteration method to optimize the position of each key point. By calculating the gradient of the key point in the image and constructing the corresponding Hessian matrix, the system can analyze the change trend of each key point in the local area. Based on this information, the Newton iteration method continuously adjusts the position of the key point so that it converges to the local maximum or minimum, thereby determining a more accurate key point position. The system will perform iterations multiple times until the position change is less than the set threshold.

[0073] Furthermore, after completing the position optimization, the system screens these key points. First, the system removes low-contrast points based on the contrast threshold to avoid interference from noise points. Then, the system determines which key points are linear edge points (such as the edges of walls or curbs) by analyzing the eigenvalues ​​of the Hessian matrix and removes these edge points. Finally, the remaining key points are feature points with high stability and high discrimination, which will be used for subsequent image matching and pose calculations.

[0074] By screening key points using the Newton iteration method, the system can quickly optimize the position of each key point and improve the accuracy and stability of the key points. The process of eliminating low-contrast points and linear edge points further improves the reliability of feature points and avoids interference from invalid points. The optimized feature points not only have a good matching effect between different images, but also remain stable under changes in lighting or viewing angles. This provides high-quality data support for subsequent feature point matching and pose calculations, and improves the repositioning accuracy and robustness of the automatic parking system.

[0075] S4: matching the feature points based on the Euclidean distance to determine matching feature point pairs;

[0076] It should be noted that Euclidean distance refers to the straight-line distance between two points and is a common method for measuring the similarity between two points. In the field of image processing, Euclidean distance is used to calculate the similarity between feature points. The smaller the distance, the more similar the two feature points are. Matching feature point pairs refer to corresponding feature point pairs found in different images, which are used to describe the same part of the same object or scene.

[0077] Specifically, in the automatic parking system, for the feature point sets of different images, the system calculates the Euclidean distance between them one by one to determine similar feature point pairs. For each feature point in the first image, the system will find the closest feature point in the second image (i.e., the nearest neighbor point). In order to ensure the accuracy of the match, a two-way matching check can be performed: not only the feature point in the first image must find the nearest corresponding point, but the corresponding point should also find the same feature point in the second image for the best match.

[0078] Furthermore, after the matching is completed, the system will further screen the matching feature point pairs. First, point pairs with too large distances are eliminated according to the preset distance threshold to reduce the possibility of mismatching. In addition, the system can screen by nearest neighbor ratio, that is, compare the distance ratio between the nearest neighbor point and the next nearest neighbor point, and eliminate those matching points with too high distance ratios. In addition, the system can also combine geometric consistency checks (such as the RANSAC algorithm) to eliminate erroneous matching pairs that do not match the overall scene, ensuring that the matching feature point pairs finally retained are accurate and reliable.

[0079] By using Euclidean distance for feature point matching, the system can quickly find corresponding feature point pairs between different images, laying the foundation for subsequent pose calculations. Bidirectional matching and distance screening steps ensure the accuracy of matching and reduce false matches caused by noise and environmental changes. Through nearest neighbor ratio and geometric consistency checks, the system further eliminates invalid matching pairs and improves the robustness of matching results. This high-quality feature point matching provides reliable data support for the camera's pose change calculation, ensuring the accuracy and stability of the automatic parking system in different lighting conditions and complex environments.

[0080] S5: Determine the camera pose change value according to the matching feature point pair, camera intrinsic parameters and camera extrinsic parameters;

[0081] It should be noted that the camera intrinsic parameters are a set of parameters that describe the internal parameters of the camera, including focal length, principal point position, pixel ratio, etc., which reflect the optical characteristics of the camera and are used to convert pixel coordinates into normalized coordinates. The camera extrinsic parameters describe the position and orientation information of the camera in space, including rotation matrix and displacement vector, which reflect the posture of the camera in the world coordinate system. The posture change value describes the change in the spatial position and orientation of the camera between different time points, including rotation angle and displacement, which is used to represent the motion state of the vehicle.

[0082] Specifically, after obtaining the matching feature point pairs, the system first uses the camera's intrinsic parameters to convert the pixel coordinates of these feature points into points in the normalized coordinate system. The normalization process eliminates the influence of the camera's optical distortion and resolution, allowing the feature points to be accurately mapped to the three-dimensional space. This step ensures that the comparison and calculation between feature points are not interfered by the camera's hardware characteristics.

[0083] Furthermore, based on the converted matching feature point pairs, combined with the camera's internal and external parameters, a basic matrix or essential matrix between the two moments is constructed to describe the relative motion relationship between the camera at the two moments. Then, the system decomposes the matrix to obtain the camera's rotation matrix and displacement vector, which represent the camera's direction change and position change, respectively. These parameters constitute the camera's pose change value, that is, the system's motion state between different image frames.

[0084] By matching feature point pairs and combining the camera's internal and external parameters, the system can accurately calculate the camera's posture change values ​​at different times. This process provides key motion information for the automatic parking system, allowing the vehicle to accurately identify its position and direction in space. At the same time, normalization and matrix decomposition ensure the robustness of the calculation, maintaining stable results even under different lighting conditions and environments. This high-precision posture calculation provides reliable data support for subsequent path planning and error correction, improving the system's parking positioning accuracy and safety.

[0085] S6: Based on the Kalman filter algorithm, the camera posture change value is integrated with the positioning data to determine the current posture.

[0086] It should be noted that the Kalman filter algorithm is a recursive algorithm that is used to optimally estimate the system state (such as position, speed, direction, etc.) by fusing multiple sensor data in the presence of noise interference. It can continuously correct the current state based on historical data and new observation data to improve the estimation accuracy of the system. The current posture refers to the position and direction (i.e., posture) of the vehicle at a specific point in time. It is the real-time state of the vehicle in space and is used for planning and adjusting the parking path.

[0087] Specifically, the camera pose change value and vehicle positioning data are used as input data to establish a Kalman filter model. In the initial stage, the system predicts the current pose based on the historical positioning data and the initial state. This prediction result is used as the initial value of the Kalman filter. Subsequently, the system obtains the pose change value from the camera and the new data from the positioning sensor, and updates the predicted state based on these data.

[0088] Furthermore, the Kalman filter performs two operations at each moment: prediction phase: using the system's motion model to predict the vehicle's posture and infer the vehicle's position and direction at the current moment; update phase: combining the posture change value calculated by the camera with the positioning data of the sensor to correct the prediction result. The Kalman filter reduces noise and error by weighted fusion of the outputs of different data sources to obtain a more accurate current posture. During multiple iterations, the Kalman filter continuously optimizes the posture information, allowing the system to maintain the accuracy and stability of vehicle positioning under the interference of sensor errors and environmental noise.

[0089] Through the Kalman filter algorithm, the system can dynamically fuse the camera's posture change value with other sensor data to provide a more stable and accurate posture estimation when the sensor may have noise or errors. This data fusion method not only improves the system's anti-interference ability, but also corrects the path in real time when the vehicle is moving, providing reliable support for path planning and obstacle avoidance during automatic parking. Ultimately, this fusion of multi-source data ensures the safety, robustness and accuracy of the parking process.

[0090] When the parking enable signal is received, the present embodiment obtains the vehicle surround view image set and positioning data; performs convolution of different Gaussian kernels on the key images in the vehicle surround view image set and obtains the difference to generate the key points in the image; based on the Newton iteration method, the key points are screened to obtain the feature points; based on the Euclidean distance, the feature points are matched to determine the matching feature point pairs; according to the matching feature point pairs, the camera intrinsic parameters and the camera extrinsic parameters, the camera posture change value is determined; based on the Kalman filter algorithm, the camera posture change value is fused with the positioning data to determine the current posture. The present embodiment extracts key points through Gaussian kernel convolution and difference, and uses the Newton iteration method to screen feature points to improve the accuracy of feature matching; based on the matching feature point pairs, the camera internal and external parameters and the basic matrix, the posture change of the vehicle is solved, and the posture change and positioning data are fused through the Kalman filter algorithm to continuously correct the vehicle position, thereby improving the accuracy of automatic parking repositioning.

[0091] Based on the above first embodiment, a second embodiment of the automatic parking relocation method of the present application is proposed. Figure 2 , Figure 2 This is a schematic diagram of a sub-process in the second embodiment of the automatic parking relocation method of the present application.

[0092] like Figure 2 As shown, in this embodiment, after step S1, the following is further included:

[0093] S1a: determining the time difference and position change between adjacent vehicle surround view images according to the vehicle surround view image set and the positioning data;

[0094] S1b: comparing the time difference with a preset time difference threshold, and comparing the posture change with a preset position change threshold;

[0095] S1c: If the time difference exceeds the preset time difference threshold or the position change exceeds the preset position change threshold, the vehicle surround image set is intercepted to obtain the key image.

[0096] It should be noted that the time difference refers to the time interval between two adjacent frames of surround view images, which is used to determine whether a new image needs to be collected again. The preset threshold refers to the limit standard of the time difference and posture change, which is used to control the frequency and accuracy of image capture.

[0097] Specifically, the system monitors the vehicle's surround view image set and obtains the vehicle's positioning data each time the surround view image is updated. For two adjacent surround view image frames, the system calculates the time difference between them and determines the time interval between the image frames. In addition, the system also calculates the vehicle's position change during this period based on the positioning data corresponding to the two frames of images, including the vehicle's position movement and direction change.

[0098] Furthermore, after the calculation is completed, the system compares the time difference with the preset time difference threshold, and compares the position change with the preset position change threshold. If the time difference exceeds the threshold or the position change exceeds the threshold, the system will trigger a new surround view image acquisition process, intercept the vehicle's surround view image set, and save the latest image frame as the key image. This ensures that the system updates the image data in a timely manner when the environment changes significantly or the vehicle moves significantly.

[0099] By judging the time difference and the amount of posture change, the system can dynamically control the frequency of image acquisition, avoid unnecessary repeated calculations, and improve the efficiency of the system. New images are captured only when the vehicle moves or the environment changes significantly, reducing the computational burden of the system. At the same time, this flexible image acquisition strategy ensures that the system always uses the latest and most relevant image information, thereby improving the accuracy of feature point matching and posture calculation, ensuring that the vehicle's automatic parking process is more stable, safe, and efficient.

[0100] Based on the above first embodiment, step S2 includes:

[0101] S21: performing a convolution operation on the key image in the vehicle surround view image set based on multiple Gaussian kernel parameters to obtain smooth images at different scales;

[0102] S22: obtaining a Gaussian difference image based on the difference between adjacent smoothed images;

[0103] S23: Compare each pixel point in the Gaussian difference image with all pixel points in a preset adjacent range;

[0104] S24: If the corresponding pixel point in the Gaussian difference image is a maximum value or a minimum value, the corresponding pixel point is used as the key point.

[0105] It should be noted that the Gaussian kernel parameter refers to the scale (standard deviation σ) of the Gaussian filter, which is used to adjust the blur degree of the filter. Different scales can highlight features of different sizes on the image. Smoothed image refers to the image obtained after Gaussian filtering, which is used to eliminate noise, retain the main structural information, and provide a basis for subsequent key point detection. The Gaussian difference image (DOG) is an image obtained by subtracting smooth images of two adjacent scales. It is mainly used to detect significant change areas in the image, such as edges and corners. The extreme value of a pixel point means that if the value of a pixel point is the local maximum or minimum value in its adjacent range, then the pixel point is an extreme point. Extreme points often correspond to obvious features in the image, such as corners.

[0106] Specifically, for the key images in the vehicle surround view image set, the system uses multiple different Gaussian kernel parameters (i.e., standard deviations σ at different scales) to perform convolution operations on the images. Gaussian convolution can smooth images to varying degrees, eliminating noise while retaining the main structural features of the image. After the convolution is completed, the system generates a series of smoothed images at different scales, which capture the features of the image at different scales. Adjacent smoothed images are differentially operated, that is, smoothed images of two adjacent scales are subtracted pixel by pixel to obtain a Gaussian difference image (DOG). Gaussian difference images can highlight the changing areas in the image, such as edges and corners, providing a basis for key point detection.

[0107] Furthermore, in the generated Gaussian difference image, the system will detect whether each pixel is a local maximum or minimum. Specifically, the system compares each pixel with its 26 adjacent pixels (within the current scale and the adjacent upper and lower scales). If the value of the pixel is larger or smaller than all the surrounding pixels, the pixel is considered to be an extreme point. These extreme points often correspond to important features in the image. Finally, the detected extreme points are used as key points, which will play an important role in the subsequent feature point screening and matching process.

[0108] Through multi-scale Gaussian convolution and differential operations of smoothed images, the system can detect significant features in the image at different scales, thereby generating stable and robust key points. This method not only effectively reduces noise interference, but also detects features of various sizes and shapes, improving the accuracy of key point detection. In addition, by comparing a pixel with multiple neighboring pixels around it, it can ensure that the selected key point has obvious local features, laying a solid foundation for subsequent feature point matching and pose calculation. This detection method improves adaptability under different lighting conditions and complex environments, providing high-precision perception capabilities for the vehicle's automatic parking system.

[0109] When the parking enable signal is received, the present embodiment obtains the vehicle surround view image set and positioning data; performs convolution of different Gaussian kernels on the key images in the vehicle surround view image set and obtains the difference to generate the key points in the image; based on the Newton iteration method, the key points are screened to obtain the feature points; based on the Euclidean distance, the feature points are matched to determine the matching feature point pairs; according to the matching feature point pairs, the camera intrinsic parameters and the camera extrinsic parameters, the camera posture change value is determined; based on the Kalman filter algorithm, the camera posture change value is fused with the positioning data to determine the current posture. The present embodiment extracts key points through Gaussian kernel convolution and difference, and uses the Newton iteration method to screen feature points to improve the accuracy of feature matching; based on the matching feature point pairs, the camera internal and external parameters and the basic matrix, the posture change of the vehicle is solved, and the posture change and positioning data are fused through the Kalman filter algorithm to continuously correct the vehicle position, thereby improving the accuracy of automatic parking repositioning.

[0110] Based on the above second embodiment, a third embodiment of the automatic parking relocation method of the present application is proposed. Figure 3 , Figure 3 This is a schematic diagram of a sub-process in the third embodiment of the automatic parking relocation method of the present application.

[0111] In this embodiment, step S3 includes:

[0112] S31: determining the gradient of the key point based on the Gaussian difference image, and constructing a Hessian matrix based on the key point;

[0113] S32: Based on the Newton iteration method, the gradient of the key point and the Hessian matrix, the position of the key point is updated until the position change is less than a preset position change threshold, thereby obtaining a first point set;

[0114] S33: eliminating points in the first point set whose contrast values ​​are greater than a preset contrast threshold, to obtain a second point set;

[0115] S34: Based on the eigenvalue of the Hessian matrix, determine whether the corresponding key point is a linear edge point;

[0116] S35: If yes, remove the linear edge points in the second point set to obtain the feature points.

[0117] It should be noted that the gradient describes the direction and amplitude of the change in pixel value. In key point detection, the gradient is used to determine the significance of the point in the image and reflect the change trend of the key point in the local area. The Hessian matrix is ​​a matrix that describes the second-order derivatives in the image and is used to analyze the curvature changes around the key points. It helps to determine the extreme value properties of the key points and whether they are linear edge points. The contrast threshold is a pre-set intensity threshold used to remove key points with too low contrast to avoid interference from noise points. Linear edge points refer to points that change linearly in a single direction in the image. They usually do not have significant characteristics and are prone to mismatching, so they need to be removed.

[0118] Specifically, the gradient of each key point is first calculated based on the Gaussian difference image to describe its change direction and intensity in the local area. Subsequently, the system constructs a Hessian matrix for each key point. The matrix captures the curvature information near the key point through the second-order derivative, which is used to judge the nature of the key point. Next, the Newton iteration method is used to optimize the position of the key point. Specifically, the system continuously updates the position of the key point based on the gradient and Hessian matrix of the key point, so that it gradually approaches the local maximum or minimum. The system will continue to iterate until the change in the position of the key point is less than the preset change threshold, ensuring that the position of the key point reaches the optimal state, and these key points are used as the first point set.

[0119] Furthermore, after the first point set is generated, points with too low contrast are eliminated according to the preset contrast threshold to avoid interference from noise points. The points after contrast screening are retained as the second point set. Subsequently, based on the eigenvalue analysis of the Hessian matrix, it is determined whether these points are linear edge points. If the curvature of a point in a certain direction is low (i.e., the change along a single direction is obvious), the point is marked as a linear edge point. These linear edge points will be eliminated from the second point set, and finally a stable and high-quality feature point set will be obtained.

[0120] Through gradient calculation, Hessian matrix construction and Newton iteration optimization, the position of key points can be accurately determined to ensure high stability and accuracy. Using contrast screening and linear edge detection, noise points and misleading edge points are eliminated, further improving the reliability of feature points. This screening and optimization process ensures that the final feature points have high discrimination and robustness, providing high-quality data support for subsequent feature point matching and pose calculation.

[0121] Based on the above second embodiment, in this embodiment, step S5 includes:

[0122] S51: Convert the pixel coordinates of the matching feature point pair into normalized coordinates;

[0123] S52: Determine a basic matrix based on the matched feature point pairs after coordinate transformation, the camera intrinsic parameters, and the camera extrinsic parameters;

[0124] S53: Decomposing the basic matrix to obtain a rotation matrix and a displacement vector;

[0125] S54: Determine the camera posture change value according to the rotation matrix and the displacement vector.

[0126] It should be noted that the normalized coordinates are converted from pixel coordinates to points in a standard coordinate system that does not depend on camera parameters by removing the influence of the camera's intrinsic parameters. This conversion makes the comparison of feature points unaffected by factors such as image resolution and focal length. The base matrix is ​​a matrix that describes the geometric relationship between two images and reflects the relative position and direction of the camera between two perspectives. The rotation matrix represents the change in direction of the camera between different perspectives and is used to describe the rotation relationship in space. The displacement vector describes the change in position of the camera between different perspectives and reflects the translation of the camera in three-dimensional space.

[0127] Specifically, the pixel coordinates of the matching feature point pairs are obtained, and the camera's intrinsic parameters (such as focal length and principal point position) are used to convert these pixel coordinates into points in a normalized coordinate system. The normalization process removes the effects of camera optical distortion and resolution differences, allowing these feature points to be accurately compared between different images. After completing the coordinate transformation, the matching feature point pairs are used to solve the basic matrix based on the camera's intrinsic and extrinsic parameters. The basic matrix captures the relative motion relationship between the camera at two different time points and is an important geometric tool for describing camera motion.

[0128] Furthermore, the obtained basic matrix is ​​decomposed to obtain the rotation matrix and displacement vector of the camera. The rotation matrix describes the change in direction of the camera between different perspectives, while the displacement vector reflects the translation of the camera in space. Finally, the rotation matrix and the displacement vector are combined to calculate the change in the camera's position between two time points, that is, the specific direction and position change of the camera. These change in position will provide key motion information for subsequent path planning and positioning.

[0129] By converting the pixel coordinates of matching feature point pairs into normalized coordinates, the system ensures the accuracy of feature point matching between different images, regardless of camera parameters. The calculation and decomposition of the basic matrix provides accurate rotation and displacement information, enabling the system to accurately obtain the motion state of the camera at different time points. This information provides high-precision pose estimation for the vehicle's automatic parking system, ensuring that the vehicle's path planning and obstacle avoidance in complex environments are more accurate and efficient.

[0130] Based on the above second embodiment, in this embodiment, before step S54, the following is further included:

[0131] S54a: Based on the positioning data, deblur the rotation matrix and the displacement vector;

[0132] S54b: Based on the rotation matrix and displacement vector after blur removal, reproject the matching feature point pairs after coordinate transformation to the key image, and obtain a reprojection error;

[0133] S54c: Using a preset optimization algorithm to minimize the re-projection error.

[0134] It should be noted that blur removal refers to improving the accuracy of motion estimation by fusing positioning data, correcting the uncertainty (ambiguity) in the rotation matrix and displacement vector. Reprojection error refers to projecting the calculated feature point position onto the image and comparing it with the actual observed position. The difference between the two is the reprojection error. Preset optimization algorithms include least squares method, RANSAC, etc., which are used to minimize errors in multiple iterations to ensure the accuracy of the model.

[0135] Specifically, after obtaining the preliminary rotation matrix and displacement vector, the vehicle's positioning data is used to remove blur. Since the rotation and displacement results calculated by image matching may be subject to scale uncertainty or noise, positioning data (such as GPS and IMU data) can provide additional constraints for these parameters, eliminate uncertainty, and improve the accuracy of pose estimation. Through this step, the rotation matrix and displacement vector after blur removal are generated. These corrected parameters are more reliable and can reflect the vehicle's motion state in real space.

[0136] Furthermore, based on the corrected rotation matrix and displacement vector, the matching feature point pairs after the previous coordinate transformation are reprojected onto the key image. After reprojection, the difference between the actual position and the projected position of the feature point is calculated, i.e., the reprojection error. In order to minimize the error, a preset optimization algorithm (such as the least squares method) is used for multiple iterations to continuously adjust the rotation matrix and displacement vector until the reprojection error is minimized. This optimization process ensures the accuracy of the matching results and provides reliable support for the vehicle's pose estimation.

[0137] By fusing positioning data with image matching data, the system can eliminate ambiguity in the rotation matrix and displacement vector, ensuring that motion estimation is more stable and accurate. At the same time, based on the optimization algorithm of reprojection error, the system can minimize the deviation in the model and improve the accuracy of pose calculation. This high-precision pose estimation provides reliable data support for vehicle path planning and motion control, ensuring that the vehicle can accurately locate and move in complex parking environments. The optimization of this process not only improves the safety and accuracy of the automatic parking system, but also enhances the robustness of the system under different environmental conditions.

[0138] When the parking enable signal is received, the present embodiment obtains the vehicle surround view image set and positioning data; performs convolution of different Gaussian kernels on the key images in the vehicle surround view image set and obtains the difference to generate the key points in the image; based on the Newton iteration method, the key points are screened to obtain the feature points; based on the Euclidean distance, the feature points are matched to determine the matching feature point pairs; according to the matching feature point pairs, the camera intrinsic parameters and the camera extrinsic parameters, the camera posture change value is determined; based on the Kalman filter algorithm, the camera posture change value is fused with the positioning data to determine the current posture. The present embodiment extracts key points through Gaussian kernel convolution and difference, and uses the Newton iteration method to screen feature points to improve the accuracy of feature matching; based on the matching feature point pairs, the camera internal and external parameters and the basic matrix, the posture change of the vehicle is solved, and the posture change and positioning data are fused through the Kalman filter algorithm to continuously correct the vehicle position, thereby improving the accuracy of automatic parking repositioning.

[0139] The present application also provides an automatic parking repositioning device, please refer to Figure 4 , Figure 4 This is a schematic diagram of the module structure of the automatic parking relocation device according to an embodiment of the present application. The automatic parking relocation device includes:

[0140] The data acquisition module 401 is used to acquire the vehicle surround view image set and positioning data when receiving the parking enable signal;

[0141] A convolution difference module 402 is used to perform convolution of different Gaussian kernels on the key images in the vehicle surround view image set and obtain differences to generate key points in the image;

[0142] A key point screening module 403 is used to screen the key points based on Newton iteration method to obtain feature points;

[0143] A feature point matching module 404 is used to match the feature points based on the Euclidean distance to determine matching feature point pairs;

[0144] A posture change module 405 is used to determine a camera posture change value according to the matching feature point pairs, camera intrinsic parameters and camera extrinsic parameters;

[0145] The target module 406 is used to fuse the camera posture change value with the positioning data based on the Kalman filter algorithm to determine the current posture.

[0146] The automatic parking relocation device provided in the embodiment of the present application adopts the automatic parking relocation method in the above embodiment, which can solve the technical problem of how to improve the accuracy of automatic parking relocation. Compared with the prior art, the beneficial effects of the automatic parking relocation device provided in the embodiment of the present application are the same as the beneficial effects of the automatic parking relocation method provided in the above embodiment, and other technical features in the automatic parking relocation device are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0147] The present application provides an automatic parking relocation device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the automatic parking relocation method in the above-mentioned embodiment.

[0148] Reference below Figure 5 , Figure 5 It is a schematic diagram of the device structure of the hardware operating environment involved in the automatic parking relocation method in the embodiment of the present application, which shows a schematic diagram of the structure of the automatic parking relocation device suitable for implementing the embodiment of the present application. Figure 5 The automatic parking relocation device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0149] like Figure 5As shown, the automatic parking relocation device may include a processing device 1001 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. Various programs and data required for the operation of the automatic parking relocation device are also stored in the RAM 1004. The processing device 1001, the ROM 1002, and the RAM 1004 are connected to each other via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems may be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 1003 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 1009. The communication device 1009 may allow the automatic parking relocation device to communicate with other devices wirelessly or by wire to exchange data. Although the figure shows an automatic parking relocation device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or have instead.

[0150] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. When the computer program is executed by the processing device 1001, the above functions defined in the method of the embodiments disclosed in the present application are executed.

[0151] The automatic parking relocation device provided by the present application adopts the automatic parking relocation method in the above embodiment, which can solve the technical problem of how to improve the accuracy of automatic parking relocation. Compared with the prior art, the beneficial effects of the automatic parking relocation device provided by the present application are the same as the beneficial effects of the automatic parking relocation method provided by the above embodiment, and other technical features in the automatic parking relocation device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0152] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0153] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0154] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the automatic parking repositioning method in the above-mentioned embodiment.

[0155] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the automatic parking repositioning device, the automatic parking repositioning device: when receiving a parking enable signal, obtains a vehicle surround view image set and positioning data; performs convolution of key images in the vehicle surround view image set with different Gaussian kernels and obtains differences to generate key points in the image; based on the Newton iteration method, the key points are screened to obtain feature points; based on the Euclidean distance, the feature points are matched to determine matching feature point pairs; according to the matching feature point pairs, camera intrinsic parameters and camera extrinsic parameters, the camera posture change value is determined; based on the Kalman filter algorithm, the camera posture change value is fused with the positioning data to determine the current posture. The computer program code for performing the operations of the present application can be written in one or more programming languages ​​or a combination thereof, and the programming language includes an object-oriented programming language such as Java, Smalltalk, C++, and a conventional procedural programming language such as "C" language or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0156] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0157] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0158] The readable storage medium provided in the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned automatic parking relocation method, and can solve the technical problem of how to improve the accuracy of automatic parking relocation. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in the present application are the same as the beneficial effects of the automatic parking relocation method provided in the above-mentioned embodiment, and will not be repeated here.

[0159] An embodiment of the present application provides a computer program product, including a computer program, which implements the steps of the automatic parking relocation method as described above when executed by a processor.

[0160] The computer program product provided in this application can solve the technical problem of how to improve the accuracy of automatic parking relocation. Compared with the prior art, the beneficial effects of the computer program product provided in the embodiment of this application are the same as the beneficial effects of the automatic parking relocation method provided in the above embodiment, which will not be repeated here.

[0161] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent processing scope of the present application.

Claims

1. An automatic parking relocation method, characterized in that: The method comprises: When a parking enable signal is received, a vehicle surround view image set and positioning data are obtained; Performing convolution of different Gaussian kernels on the key images in the vehicle surround view image set and taking differences to generate key points in the image; Based on the Newton iteration method, the key points are screened to obtain feature points; Based on the Euclidean distance, the feature points are matched to determine matching feature point pairs; Determine the camera pose change value according to the matching feature point pair, camera intrinsic parameters and camera extrinsic parameters; Based on the Kalman filter algorithm, the camera posture change value is fused with the positioning data to determine the current posture.

2. The method according to claim 1, characterized in that After the step of acquiring the vehicle surround view image set and positioning data when the parking enable signal is received, the method further includes: Determining the time difference and position change between adjacent vehicle surround view images according to the vehicle surround view image set and the positioning data; Comparing the time difference with a preset time difference threshold, and comparing the position change with a preset position change threshold; If the time difference exceeds the preset time difference threshold or the position change exceeds the preset position change threshold, the vehicle surround image set is intercepted to obtain the key image.

3. The method according to claim 1, characterized in that The step of performing convolution of different Gaussian kernels on the key images in the vehicle surround view image set and obtaining differences to generate key points in the image comprises: Based on multiple Gaussian kernel parameters, a convolution operation is performed on the key image in the vehicle surround view image set to obtain smooth images at different scales; Based on the difference between adjacent smoothed images, a Gaussian difference image is obtained; Comparing each pixel in the Gaussian difference image with all the pixels in a preset adjacent range; If the corresponding pixel point in the Gaussian difference image is a maximum value or a minimum value, the corresponding pixel point is used as the key point.

4. The method according to claim 3, characterized in that The step of screening the key points based on the Newton iteration method to obtain feature points includes: Based on the Gaussian difference image, determine the gradient of the key point, and based on the key point, construct a Hessian matrix; Based on the Newton iteration method, the gradient of the key point and the Hessian matrix, the position of the key point is updated until the position change is less than a preset position change threshold, thereby obtaining a first point set; Eliminate points whose contrast values ​​are greater than a preset contrast threshold in the first point set to obtain a second point set; Based on the eigenvalues ​​of the Hessian matrix, determining whether the corresponding key point is a linear edge point; If so, the linear edge points in the second point set are eliminated to obtain the feature points.

5. The method according to claim 1, characterized in that The step of determining the camera pose change value according to the matching feature point pair, camera intrinsic parameters and camera extrinsic parameters comprises: Convert the pixel coordinates of the matching feature point pairs into normalized coordinates; Determine a basic matrix based on the matched feature point pairs after coordinate transformation, the camera intrinsic parameters, and the camera extrinsic parameters; Decomposing the basic matrix to obtain a rotation matrix and a displacement vector; The camera position change value is determined according to the rotation matrix and the displacement vector.

6. The method according to claim 5, characterized in that Before the step of determining the camera posture change value according to the rotation matrix and the displacement vector, the method further includes: Based on the positioning data, deblurring the rotation matrix and the displacement vector; Based on the rotation matrix and displacement vector after blur removal, reproject the matched feature point pairs after coordinate transformation to the key image, and obtain a reprojection error; A preset optimization algorithm is used to minimize the re-projection error.

7. An automatic parking repositioning device, characterized in that: The device comprises: A data acquisition module, used to acquire a vehicle surround view image set and positioning data when receiving a parking enable signal; A convolution difference module is used to perform convolution of different Gaussian kernels on the key images in the vehicle surround view image set and calculate the difference to generate key points in the image; A key point screening module, used to screen the key points based on Newton iteration method to obtain feature points; A feature point matching module, used to match the feature points based on the Euclidean distance to determine matching feature point pairs; A posture change module, used to determine the camera posture change value according to the matching feature point pair, camera intrinsic parameters and camera extrinsic parameters; The target module is used to fuse the camera posture change value with the positioning data based on the Kalman filter algorithm to determine the current posture.

8. A computer device, characterized in that: The device comprises: a memory, a processor, and an automatic parking relocation program stored in the memory and executable on the processor, wherein the automatic parking relocation program is configured to implement the steps of the automatic parking relocation method according to any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium stores an automatic parking repositioning program, and when the automatic parking repositioning program is executed by the processor, the steps of the automatic parking repositioning method according to any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the automatic parking repositioning method according to any one of claims 1 to 6 are implemented.

Images