Panoramic parking space data generation method, device, equipment, medium and program product

Abstract

This application relates to the fields of computer vision and autonomous driving perception technology, and particularly to a method, apparatus, device, medium, and program product for generating panoramic parking space data. The method includes: acquiring an original AVM image and preprocessing it to obtain a clean background image; generating parameterized parking spaces based on a preset set of candidate parking space contour primitives and preset multi-dimensional parameters, and mapping the parameterized parking spaces to the clean background image based on the coordinate system of the original AVM image to obtain an initial parking space image set; calculating the corner detection loss of each parking space image in the initial parking space image set based on a preset parking space corner detector, and generating a final parking space image set based on the corner detection loss of each parking space image from the initial parking space image set. This solves the problems of difficulty in simultaneously achieving background realism and foreground controllability, and low efficiency of data selection mechanisms in related technologies, thus improving the realism of the synthesized data.

Description

Technical Field

[0001] This application relates to the fields of computer vision and autonomous driving perception technology, and in particular to a method, apparatus, device, medium and program product for generating panoramic parking space data. Background Technology

[0002] Parking space detection is a key technology in Advanced Driver Assistance Systems (ADAS) and autonomous driving, and is fundamental to achieving automated parking and improving parking safety. Current mainstream detection algorithms are typically based on deep learning methods, requiring large amounts of precisely labeled image data for model training to ensure robustness and generalization ability. However, acquiring real parking space image data using panoramic surround-view systems is costly, and the data acquisition process is complex and time-consuming.

[0003] Among related technologies, data synthesis technology has become one of the research hotspots in the field of autonomous driving. Current research focuses on methods such as Generative Adversarial Networks (GANs) or style transfer, aiming to narrow the gap between synthetic data and real data.

[0004] However, current methods in related technologies typically require large amounts of basic training data, resulting in poor controllability of the generated results and difficulty in accurately generating parking space elements that conform to specific geometric constraints. While rendering engine-based methods can generate highly controllable 3D scenes through parameterization, their rendering costs are high, and images generated by game engines often lack realism, leading to poor transfer effects from simulation to reality. Current synthetic data methods for parking space detection struggle to achieve both background realism and foreground controllability, lacking efficient data selection mechanisms, which urgently need to be addressed. Summary of the Invention

[0005] This application provides a method, apparatus, equipment, medium, and program product for generating panoramic parking space data, in order to solve the problems of difficulty in achieving both background authenticity and foreground controllability and low efficiency of data selection mechanisms in related technologies.

[0006] The first aspect of this application provides a method for generating panoramic parking space data, including the following steps:

[0007] Obtain the original AVM image and preprocess it to obtain a clean background image; Based on a preset set of candidate parking space contour primitives and preset multi-dimensional parameters, parameterized parking spaces are generated, and based on the original AVM image coordinate system, the parameterized parking spaces are mapped to the clean background image to obtain an initial parking space image set. Based on a preset parking space corner detector, the corner detection loss of each parking space image in the initial parking space image set is calculated, and based on the corner detection loss of each parking space image, a final parking space image set is generated according to the initial parking space image set.

[0008] Optionally, the preset parking space corner detector is: ; in, The corner detection loss is... The output layer grid size of the marker detector. The total number of cells in the grid. The index number of the grid cell. To predict confidence levels, For true confidence level, To predict the location, This is the actual location.

[0009] Optionally, generating a final parking space image set based on the corner detection loss of each parking space image includes: Based on the corner detection loss of each parking space image, each parking space image in the initial parking space image set is sorted in ascending order to obtain a sorted parking space image set. Based on preset ratio selection conditions, a target parking space image is selected from the sorted parking space image set, and the final parking space image set is obtained based on the target parking space image.

[0010] Optionally, the preprocessing of the original AVM image to obtain a clean background image includes: Extract road information and parking areas from the original AVM image; The road information and parking areas in the original AVM image are erased to obtain the clean background image.

[0011] Optionally, before calculating the corner detection loss for each parking space image in the initial parking space image set based on a preset parking space corner detector, the method further includes: Gaussian blurring is applied to each parking space image in the initial parking space image set.

[0012] A second aspect of this application provides a panoramic parking space data generation device, comprising: The acquisition module is used to acquire the original AVM image and preprocess the original AVM image to obtain a clean background image; The mapping module is used to generate parameterized parking spaces based on a preset set of candidate parking space contour primitives and preset multi-dimensional parameters, and to map the parameterized parking spaces onto the clean background image based on the original AVM image coordinate system to obtain an initial parking space image set. The calculation module is used to calculate the corner detection loss of each parking space image in the initial parking space image set based on a preset parking space corner detector, and to generate a final parking space image set based on the corner detection loss of each parking space image.

[0013] Optionally, the preset parking space corner detector is: ; in, The corner detection loss is... The output layer grid size of the marker detector. The total number of cells in the grid. The index number of the grid cell. To predict confidence levels, For true confidence level, To predict the location, This is the actual location.

[0014] Optionally, the computing module is specifically used for: Based on the corner detection loss of each parking space image, each parking space image in the initial parking space image set is sorted in ascending order to obtain a sorted parking space image set. Based on preset ratio selection conditions, a target parking space image is selected from the sorted parking space image set, and the final parking space image set is obtained based on the target parking space image.

[0015] Optionally, the acquisition module is specifically used for: Extract road information and parking areas from the original AVM image; The road information and parking areas in the original AVM image are erased to obtain the clean background image.

[0016] Optionally, before calculating the corner detection loss for each parking space image in the initial parking space image set based on a preset parking space corner detector, the calculation module is further configured to: Gaussian blurring is applied to each parking space image in the initial parking space image set.

[0017] A third aspect of this application provides an electronic device, including: 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, the instructions being configured to perform the panoramic parking space data generation method as described in the above embodiments.

[0018] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement the panoramic parking space data generation method as described in the above embodiments.

[0019] A fifth aspect of this application provides a computer program product storing a computer program that, when executed by a processor, implements the panoramic parking space data generation method as described in the above embodiments.

[0020] Therefore, this embodiment of the application acquires the original AVM image and preprocesses it to obtain a clean background image; based on a preset set of candidate parking space contour primitives and preset multi-dimensional parameters, parameterized parking spaces are generated, and based on the coordinate system of the original AVM image, the parameterized parking spaces are mapped to the clean background image to obtain an initial parking space image set; based on a preset parking space corner detector, the corner detection loss of each parking space image in the initial parking space image set is calculated, and based on the corner detection loss of each parking space image, a final parking space image set is generated according to the initial parking space image set. This solves the problems of difficulty in simultaneously achieving background realism and foreground controllability and low efficiency of data selection mechanisms in related technologies, thus improving the realism of the synthesized data.

[0021] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0022] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of a panoramic parking space data generation method provided according to an embodiment of this application; Figure 2 This is a flowchart of a panoramic parking space data generation method according to an embodiment of this application; Figure 3 This is a schematic diagram of a panoramic parking space data generation device provided according to an embodiment of this application; Figure 4 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0023] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0024] Before introducing the panoramic parking space data generation method of the embodiments of this application, let's briefly introduce the panoramic parking space data generation method in related technologies.

[0025] Specifically, parking space detection is a key technology in Advanced Driver Assistance Systems (ADAS) and autonomous driving, and is fundamental to achieving automated parking and improving parking safety. Current mainstream detection algorithms are typically based on deep learning methods, requiring a large amount of precisely labeled image data for model training to ensure robustness and generalization ability. However, acquiring real parking space image data using panoramic surround-view systems is costly, involving the calibration of multiple fisheye cameras, image projection and stitching, and complex and time-consuming hardware deployment and annotation processes. Furthermore, due to the diversity of parking space geometry, ground texture, lighting conditions, and environmental noise in real-world scenarios, constructing datasets using real-world data is extremely difficult.

[0026] To address the problem of data scarcity, data synthesis technology has become a research hotspot in the field of autonomous driving. Current research largely focuses on Generative Adversarial Networks (GANs) or style transfer methods, aiming to narrow the gap between synthetic and real data. However, these methods typically require large amounts of basic training data, have poor controllability of the generated results, and struggle to accurately generate parking space elements that conform to specific geometric constraints. While rendering engine-based methods can generate highly controllable 3D scenes through parameterization, their rendering costs are high, and images generated by game engines often lack realism, resulting in poor transfer effects from simulation to reality.

[0027] In related technologies, method (1) discloses a parking space detection method, device and electronic device based on multi-feature fusion. It uses a pre-trained model to extract parking space line instances, parking space area instances and parking space corner points in the environmental image, and constructs three types of candidate regions respectively. Finally, it determines the target parking space by comprehensively considering the vehicle position confidence of each region, which solves the problem of high detection difficulty in scenes with unclear or incomplete parking space lines. Method (2) discloses a parking space detection technology based on weather perception adaptive enhancement and multi-task joint learning. It identifies the weather through the weather perception mechanism and implements adaptive image enhancement. It uses a conditional attention module to guide the model to focus on key areas. At the same time, it constructs a multi-task deep learning network and introduces cross-task constraint loss to collaboratively optimize the detection tasks of parking spaces, targets and markings, thereby improving the robustness and accuracy of detection under different weather and lighting conditions. Method (3) discloses a parking space detection technique based on deep learning and low-light image enhancement. It enhances parking lot images in low-light environments using the Zero-DCE++ model and improves the YOLOv8 model structure by introducing a BiFPN bidirectional feature pyramid network. It uses feature extraction and fusion to achieve real-time classification and localization of accurate parking space coordinates and occupancy status. Method (4) proposes a lightweight parking space detection technique that integrates spatiotemporal consistency and self-supervised learning. It extracts the foreground mask by predefining the Region of Interest (ROI) and combining it with Gaussian mixture model background modeling. It uses spatiotemporal consistency analysis to perform depth detection only on changing regions and uses self-supervised learning and knowledge distillation to improve model performance. Thus, it ensures efficient and robust parking space status recognition while adapting to dynamic changes in lighting and environment.

[0028] However, current synthetic data methods for parking space detection suffer from the following shortcomings: First, achieving both background realism and foreground controllability is difficult: simple image synthesis methods struggle to guarantee the realism and lighting consistency of the synthesized image. High-realism rendering methods, on the other hand, lack precise control over specific targets. Second, there is a lack of efficient data selection mechanisms: while synthetic frameworks can generate large amounts of data, simply using all the synthetic data to train the model may lead to performance degradation or even negative transfer. Existing data selection methods often rely on coarse domain selection, but the samples in the dataset that truly have positive value for model training may be scattered across various domains; coarse domain selection strategies cannot achieve refined sample selection.

[0029] This application addresses the aforementioned problems by proposing a panoramic parking space data generation method. In this method, the original AVM image is acquired and preprocessed to obtain a clean background image. Based on a preset set of candidate parking space contour primitives and preset multi-dimensional parameters, parameterized parking spaces are generated. Using the coordinate system of the original AVM image, the parameterized parking spaces are mapped onto the clean background image to obtain an initial parking space image set. Based on a preset parking space corner detector, the corner detection loss for each parking space image in the initial parking space image set is calculated. Based on the corner detection loss of each parking space image, a final parking space image set is generated from the initial parking space image set. This solves the problems of difficulty in simultaneously achieving background realism and foreground controllability, and low efficiency of data selection mechanisms in related technologies, thereby improving the realism of the synthesized data.

[0030] Specifically, Figure 1 This is a flowchart illustrating a panoramic parking space data generation method provided in an embodiment of this application.

[0031] like Figure 1 As shown, the panoramic parking space data generation method includes the following steps: In step S101, the original AVM image is acquired, and the original AVM image is preprocessed to obtain a clean background image.

[0032] The original AVM image refers to a 360° panoramic image of the vehicle's surroundings, generated from a bird's-eye view perspective, after the images are synchronously captured by the Around View Monitor (AVM) through a wide-angle camera installed in the vehicle and preprocessed by distortion correction, projection transformation, and panoramic stitching.

[0033] Specifically, a clean, realistic background image is extracted from the original AVM image. This clean background image serves as the "canvas" for subsequent virtual parking space composition, ensuring that all compositional elements are built upon a real-world physical foundation.

[0034] Optionally, in some embodiments, the original AVM image is preprocessed to obtain a clean background image, including: extracting road information and parking areas from the original AVM image; and erasing the road information and parking areas in the original AVM image to obtain a clean background image.

[0035] Understandably, the original AVM image is semantically segmented using a semantic segmentation model to obtain a segmentation mask, extract road information, and determine parking areas. Subsequently, a LaMa network (Large Mask Inpainting Network) integrating Fast Fourier Convolutions is used to erase and inpaint lane lines and existing parking space markings in the original AVM image, generating a clean background image without any parking markings. .

[0036] In step S102, parameterized parking spaces are generated based on a preset set of candidate parking space contour primitives and preset multi-dimensional parameters. Then, based on the original AVM image coordinate system, the parameterized parking spaces are mapped onto a clean background image to obtain an initial parking space image set.

[0037] The preset candidate parking space contour primitive set is a set of parking space contour structures that actually exist in the original image and are described in vector geometric form (such as corner coordinates and line segment equations).

[0038] Specifically, the Douglas-Peucker algorithm is used to approximate the segmentation mask, transforming the pixel-level mask into vector geometry. An adaptive threshold is then set. While smoothing noise, the rectangular geometric features and corner points are preserved to generate a set of candidate parking space contour primitives. : ; Furthermore, diverse parking space foreground elements are generated based on preset multi-dimensional parameters and integrated into the repaired background in a physically consistent manner. By introducing a domain randomization strategy, diverse parking spaces are generated, and the preset multi-dimensional parameters cover the following dimensions: 1) Geometric shape: Randomly select different types such as vertical, parallel, and diagonal.

[0039] 2) Color and Texture: RGB values ​​are adjusted based on real samples. The perturbation is 30, and ZenBG is used to generate diverse line segment textures.

[0040] 3) Appearance deformation: Curvature deformation in the range of 100 to 200 is randomly introduced through elastic transformation to simulate wear or non-standard scribing.

[0041] 4) Environmental interference: Gaussian noise, Laplace noise, salt and pepper noise, etc. are introduced to simulate real imaging noise.

[0042] 5) Lighting conditions: The global brightness is randomly adjusted between 0.5 and 1.5 to simulate different lighting conditions.

[0043] Furthermore, subsequently, using 3 Homography matrix of 3 The generated parameterized parking spaces are mapped to the AVM image coordinate system (original AVM image coordinate system) to obtain the initial parking space image set, where the mapping formula is: ; in, The coordinates for generating the pattern, These are the coordinates of the corresponding point in the AVM image; It is a 3×3 homography transformation matrix. This is the scaling factor.

[0044] In step S103, based on the preset parking space corner detector, the corner detection loss of each parking space image in the initial parking space image set is calculated, and based on the corner detection loss of each parking space image, the final parking space image set is generated according to the initial parking space image set.

[0045] Optionally, in some embodiments, the preset parking space corner detector is: ; in, For corner detection loss, The output layer grid size of the marker detector. The total number of cells in the grid. The index number of the grid cell. To predict confidence levels, For true confidence level, To predict the location, This is the actual location.

[0046] Among them, corner detection loss refers to the quantified value of the deviation between the predicted value and the true labeled value of the corner point of the parking space in each synthetic image, using a preset parking space corner detector as a proxy evaluation model.

[0047] Specifically, to select the subset with the most "supplementary value" to a small sample of real data from a massive synthetic dataset, this application proposes a "Close-to-Far Data Selection" (CFDS) strategy. The core of this strategy is to measure the potential value of each synthetic sample for training the current base model. By using a lightweight parking space corner detector (a pre-defined parking space corner detector) as an evaluation surrogate model, the value of each synthetic image is calculated. Corner detection loss The loss function takes into account both corner confidence and position error, and determines the final parking space image set from the initial parking space image set.

[0048] Optionally, in some embodiments, generating a final parking space image set based on the corner detection loss of each parking space image includes: sorting each parking space image in the initial parking space image set in ascending order based on the corner detection loss of each parking space image to obtain a sorted parking space image set; selecting a target parking space image from the sorted parking space image set based on a preset ratio selection condition, and obtaining the final parking space image set based on the target parking space image.

[0049] Understandably, the corner detection losses of all synthetic images are sorted in ascending order. Samples with lower losses are considered "near" data, meaning they are closer to the real data distribution in the current model's view. Including them in the training set helps stabilize the model. Samples with higher losses are considered unrealistic and are discarded because they deviate too much from the real data. CFDS uses dynamic thresholding or quantity control to select a certain proportion (based on preset proportion selection conditions) of synthetic images (target parking space images) according to the sorting order, forming a high-quality training subset. This refers to the set of parking space images ultimately used for model training.

[0050] Therefore, this invention enhances the realism and controllability of synthetic data. It obtains realistic scene backgrounds through image inpainting techniques and parametrically overlays controllable foreground elements to generate realistic data that meets diverse needs. This invention achieves refined data selection and optimizes training efficiency by introducing a sample-level data selection strategy based on active learning, replacing the traditional coarse-grained domain selection. This strategy accurately selects the most effective subset for improving model performance from massive amounts of synthetic data.

[0051] Optionally, in some embodiments, before calculating the corner detection loss of each parking space image in the initial parking space image set based on a preset parking space corner detector, the method further includes: performing Gaussian blur processing on each parking space image in the initial parking space image set.

[0052] Understandably, to eliminate synthetic artifacts, the mask edges of each parking space image in the initial parking space image set are subjected to 5... 5. Gaussian blur is applied to eliminate high-frequency aliasing; then Poisson image editing is used, which solves the Poisson partial differential equation under Dirichlet boundary conditions to seamlessly blend the lighting and color gradient of the parking lines with the real background.

[0053] Therefore, this invention uses a LaMa insulation network to erase the original markings in the AVM image to obtain a semantically consistent clean road surface background, and uses homography transformation to project virtual parking spaces with controllable geometric features onto the insulated background. Poisson image editing is then used to eliminate synthesis artifacts. The background consistency synthesis mechanism based on "semantic insulation-parametric redrawing" in this invention solves the problem of background semantic conflict in the traditional overlay method, and significantly enhances the fidelity of the synthesized samples by preserving real-world background details.

[0054] This invention constructs a parameterized system for explicit geometric and visual control of parking spaces. Virtual samples are generated by changing parameters including the number and lateral offset of parking spaces, RGB color perturbation, ZenBG-based texture mode, line curvature introduced by elastic transformation, and global brightness scaling of 0.5-1.5 times. Compared with generative models, the domain randomization generation method of the multi-dimensional parameterized definition of this invention ensures that parking space lines strictly conform to physical space logic and traffic rules.

[0055] This invention proposes a sample-level selection strategy based on active learning. It uses a corner detection loss to evaluate and rank all synthetic samples, and selects subsets with lower loss or covering a specific distribution according to a dynamic strategy. This achieves intelligent screening of synthetic datasets to maximize their training gain for the target model.

[0056] To facilitate those skilled in the art to further understand the panoramic parking space data generation method of the embodiments of this application, the following is combined with... Figure 2 The embodiments shown will be described in detail.

[0057] Specifically, such as Figure 2 As shown, Figure 2 This document presents a flowchart of a panoramic parking space data generation method according to an embodiment of this application. The invention constructs a data synthesis framework based on a "repair-combination" mechanism, and is equipped with a "near-to-far" data selection strategy incorporating active learning principles. The method mainly includes the following four core processes: background restoration and candidate region generation, parameterized foreground generation and fusion, active learning data selection, and joint training and optimization of virtual and real data.

[0058] Furthermore, the original input AVM image is first semantically segmented and scene understood to identify existing parking spaces and lines on the road surface. These are then erased using LaMa network-based image inpainting techniques, generating a clean road background image and a reconstructed foreground image. Subsequently, newly generated parking spaces, parameterized by shape, color, and curvature, are placed in this background using homography transformation and an image synthesis module. These are then processed with Poisson fusion and edge smoothing to construct a high-fidelity original synthetic dataset. Finally, a "near-to-far" data filtering (CFDS) strategy is employed to select high-quality samples by calculating and ranking the marker detection loss, thereby constructing the final training dataset. This achieves efficient training and performance optimization of the parking space detection model.

[0059] Therefore, the embodiments of this application significantly reduce the cost of data acquisition and annotation. Specifically, by generating high-quality training data through an automated synthesis mechanism, it effectively solves the problems of complex AVM image acquisition and extremely time-consuming manual annotation. This method can generate samples with precise labels in batches, providing a scalable and low-cost solution for the development of industrial-grade perception systems. The generated data in the embodiments of this application possesses both high realism and strong controllability. Through the "repair-combination" framework, the authenticity of the synthesized image background is ensured, while parameterized generation enables precise control over various attributes such as the shape, color, and texture of the foreground parking space, resulting in a diverse training set. The embodiments of this application introduce an active learning mechanism to achieve efficient data filtering. The proposed CFDS strategy overcomes the limitations of traditional domain selection methods, accurately filtering the most effective synthetic data for improving the target detection model at the sample-level granularity, avoiding data redundancy and negative transfer, and significantly improving training efficiency.

[0060] According to the panoramic parking space data generation method proposed in this application, the original AVM image is acquired and preprocessed to obtain a clean background image; based on a preset set of candidate parking space contour primitives and preset multi-dimensional parameters, parameterized parking spaces are generated, and based on the coordinate system of the original AVM image, the parameterized parking spaces are mapped to the clean background image to obtain an initial parking space image set; based on a preset parking space corner detector, the corner detection loss of each parking space image in the initial parking space image set is calculated, and based on the corner detection loss of each parking space image, a final parking space image set is generated from the initial parking space image set. This solves the problems of difficulty in simultaneously achieving background realism and foreground controllability and low efficiency of data selection mechanisms in related technologies, thus improving the realism of the synthesized data.

[0061] Next, the panoramic parking space data generation apparatus according to the embodiments of this application is described with reference to the accompanying drawings.

[0062] Figure 3 This is a block diagram of a panoramic parking space data generation device according to an embodiment of this application.

[0063] like Figure 3 As shown, the panoramic parking space data generation device 10 includes: an acquisition module 100, a mapping module 200, and a calculation module 300.

[0064] The acquisition module 100 is used to acquire the original AVM image and preprocess the original AVM image to obtain a clean background image. The mapping module 200 is used to generate parameterized parking spaces based on a preset set of candidate parking space contour primitives and preset multi-dimensional parameters, and to map the parameterized parking spaces onto a clean background image based on the original AVM image coordinate system to obtain an initial parking space image set. The calculation module 300 is used to calculate the corner detection loss of each parking space image in the initial parking space image set based on the preset parking space corner detector, and generate the final parking space image set based on the corner detection loss of each parking space image.

[0065] Optionally, the preset parking space corner detector is: ; in, For corner detection loss, The output layer grid size of the marker detector. The total number of cells in the grid. The index number of the grid cell. To predict confidence levels, For true confidence level, To predict the location, This is the actual location.

[0066] Optionally, the calculation module 300 is specifically used to: sort each parking space image in the initial parking space image set in ascending order based on the corner detection loss of each parking space image to obtain a sorted parking space image set; select a target parking space image from the sorted parking space image set based on a preset ratio selection condition, and obtain the final parking space image set based on the target parking space image.

[0067] Optionally, the acquisition module 100 is specifically used to: extract road information and parking areas from the original AVM image; and erase the road information and parking areas in the original AVM image to obtain a clean background image.

[0068] Optionally, before calculating the corner detection loss of each parking space image in the initial parking space image set based on a preset parking space corner detector, the calculation module 300 is further configured to: perform Gaussian blur processing on each parking space image in the initial parking space image set.

[0069] It should be noted that the foregoing explanation of the panoramic parking space data generation method embodiment also applies to the panoramic parking space data generation device of this embodiment, and will not be repeated here.

[0070] According to the panoramic parking space data generation device proposed in this application, the present application acquires the original AVM image and preprocesses it to obtain a clean background image; based on a preset set of candidate parking space contour primitives and preset multi-dimensional parameters, parameterized parking spaces are generated, and based on the coordinate system of the original AVM image, the parameterized parking spaces are mapped to the clean background image to obtain an initial parking space image set; based on a preset parking space corner detector, the corner detection loss of each parking space image in the initial parking space image set is calculated, and based on the corner detection loss of each parking space image, a final parking space image set is generated from the initial parking space image set. This solves the problems of difficulty in simultaneously achieving background realism and foreground controllability and low efficiency of data selection mechanisms in related technologies, thus improving the realism of the synthesized data.

[0071] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: The memory 401, the processor 402, and the computer program stored on the memory 401 and capable of running on the processor 402.

[0072] When processor 402 executes the program, it implements the panoramic parking space data generation method provided in the above embodiments.

[0073] Furthermore, electronic devices also include: Communication interface 403 is used for communication between memory 401 and processor 402.

[0074] The memory 401 is used to store computer programs that can run on the processor 402.

[0075] Memory 401 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0076] If the memory 401, processor 402, and communication interface 403 are implemented independently, then the communication interface 403, memory 401, and processor 402 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 4 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0077] Optionally, in a specific implementation, if the memory 401, processor 402, and communication interface 403 are integrated on a single chip, then the memory 401, processor 402, and communication interface 403 can communicate with each other through an internal interface.

[0078] Processor 402 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0079] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described panoramic parking space data generation method.

[0080] This application also provides a computer program product, which stores a computer program that, when executed by a processor, implements the above-described panoramic parking space data generation method.

[0081] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0082] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0083] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0084] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0085] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

Claims

1. A method for generating panoramic parking space data, characterized in that, Includes the following steps: Obtain the original AVM image and preprocess it to obtain a clean background image; Based on a preset set of candidate parking space contour primitives and preset multi-dimensional parameters, parameterized parking spaces are generated, and based on the original AVM image coordinate system, the parameterized parking spaces are mapped to the clean background image to obtain an initial parking space image set. Based on a preset parking space corner detector, the corner detection loss of each parking space image in the initial parking space image set is calculated, and based on the corner detection loss of each parking space image, a final parking space image set is generated according to the initial parking space image set.

2. The method according to claim 1, characterized in that, The preset parking space corner detector is: ； in, The corner detection loss is... The output layer grid size of the marker detector. The total number of cells in the grid. The index number of the grid cell. To predict confidence levels, For true confidence level, To predict the location, This is the actual location.

3. The method according to claim 1 or 2, characterized in that, The step of generating a final parking space image set based on the corner detection loss of each parking space image, according to the initial parking space image set, includes: Based on the corner detection loss of each parking space image, each parking space image in the initial parking space image set is sorted in ascending order to obtain a sorted parking space image set. Based on preset ratio selection conditions, a target parking space image is selected from the sorted parking space image set, and the final parking space image set is obtained based on the target parking space image.

4. The method according to claim 1, characterized in that, The preprocessing of the original AVM image to obtain a clean background image includes: Extract road information and parking areas from the original AVM image; The road information and parking areas in the original AVM image are erased to obtain the clean background image.

5. The method according to claim 1, characterized in that, Before calculating the corner detection loss for each parking space image in the initial parking space image set based on a preset parking space corner detector, the method further includes: Gaussian blurring is applied to each parking space image in the initial parking space image set.

6. A panoramic parking space data generation device, characterized in that, include: The acquisition module is used to acquire the original AVM image and preprocess the original AVM image to obtain a clean background image; The mapping module is used to generate parameterized parking spaces based on a preset set of candidate parking space contour primitives and preset multi-dimensional parameters, and to map the parameterized parking spaces onto the clean background image based on the original AVM image coordinate system to obtain an initial parking space image set. The calculation module is used to calculate the corner detection loss of each parking space image in the initial parking space image set based on a preset parking space corner detector, and to generate a final parking space image set based on the corner detection loss of each parking space image.

7. The apparatus according to claim 6, characterized in that, The preset parking space corner detector is: ； in, The corner detection loss is... The output layer grid size of the marker detector. The total number of cells in the grid. The index number of the grid cell. To predict confidence levels, For true confidence level, To predict the location, This is the actual location.

8. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the panoramic parking space data generation method as described in any one of claims 1-5.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the panoramic parking space data generation method as described in any one of claims 1-5.

10. A computer program product, said computer program product storing a computer program, characterized in that, When the program is executed by the processor, it implements the panoramic parking space data generation method as described in any one of claims 1-5.

Images