Parking space number detection method, detection device, vehicle and storage medium

Abstract

This application discloses a parking space number detection method, detection device, vehicle, and storage medium. The parking space number detection method includes acquiring multiple images of parking space numbers to be detected; inputting the multiple images of parking space numbers to be detected into a preset deep learning network to obtain corresponding feature sets, the feature sets including at least the character set corresponding to the parking space number frame, and an embedded feature set of the center point position of the parking space number frame and the center point position of the character; based on the embedded feature set, clustering the character set to obtain multiple character clusters, each character cluster corresponding to a string of a parking space number frame; and outputting the target string of the target parking space number frame. This method realizes automatic detection of parking space numbers in images of parking space numbers to be detected through a deep learning network, without the need for manual operation by the user, thus improving the detection efficiency and accuracy of parking space numbers.

Description

Technical Field

[0001] This application relates to the field of autonomous driving technology, and more specifically, to a parking space number detection method, detection device, vehicle, and storage medium. Background Technology

[0002] Currently, parking spaces in parking lots are typically assigned corresponding parking space numbers for parking management purposes. These parking space numbers usually consist of a string of characters, such as A001. In the field of autonomous driving, self-driving vehicles need to identify parking space numbers for location or navigation in parking lots.

[0003] Existing technologies typically involve manually extracting features from the acquired parking space number images, and then inputting the extracted features into a classifier for character recognition to obtain the parking space number. However, this traditional image processing method suffers from technical problems such as slow detection speed and low accuracy. Summary of the Invention

[0004] In view of the above problems, this application proposes a parking space number detection method, detection device, vehicle and storage medium to overcome or at least partially solve the problems of the prior art.

[0005] In a first aspect, embodiments of this application provide a parking space number detection method, comprising: acquiring multiple images of parking space numbers to be detected; inputting the multiple images of parking space numbers to be detected into a preset deep learning network to obtain a corresponding feature set, wherein the feature set includes at least the character set corresponding to the parking space number frame, and an embedded feature set of the center point position of the parking space number frame and the center point position of the character; clustering the character set based on the embedded feature set to obtain multiple character clusters, wherein each character cluster corresponds to a string of a parking space number frame; and outputting the target string of the target parking space number frame.

[0006] In some implementations, the feature set includes the character set corresponding to the parking space number frame, and the embedded feature set of the center point position of the parking space number frame and the center point position of the character. Multiple parking space number images to be detected are input into a preset deep learning network to obtain the corresponding feature set. This includes: inputting multiple parking space number images to be detected into a preset deep learning network, causing the deep learning network to detect the character corresponding to the parking space number frame, the center point position of the parking space number frame, and the center point position of the character, respectively, to obtain the character set corresponding to the parking space number frame, the center point position set of the parking space number frame, and the character center point set; and fusing the center point position set of the parking space number frame and the character center point position set to obtain the embedded feature set; and receiving the character set and the embedded feature set output by the deep learning network.

[0007] In some implementations, the feature set also includes a character category set of the characters corresponding to the parking space number frame. Inputting multiple parking space number images to be detected into a preset deep learning network to obtain the corresponding feature set further includes: inputting multiple parking space number images to be detected into a preset deep learning network, enabling the deep learning network to identify the character category of the characters corresponding to the parking space number frame, and obtaining a character category set; receiving the character category set output by the deep learning network; and outputting the target string of the target parking space number frame, including: determining whether the character category of the target string of the target parking space number frame is a preset character category; and when the character category of the target string is determined to be a preset character category, outputting the target string of the target parking space number frame.

[0008] In some implementations, the feature set also includes a set of corner coordinates corresponding to the parking space number frame. Multiple images of parking space numbers to be detected are input into a preset deep learning network to obtain the corresponding feature set. The process further includes: inputting multiple images of parking space numbers to be detected into a preset deep learning network, causing the deep learning network to detect the corner coordinates corresponding to the parking space number frame, obtaining a set of corner coordinates; receiving the corner coordinate set output by the deep learning network; and outputting a target string for the target parking space number frame, including: determining whether the target corner coordinates of the target parking space number frame are within the coordinate range corresponding to the parking space number image to be detected; and when the target corner coordinates are determined to be within the coordinate range, outputting the target string corresponding to the target parking space number frame.

[0009] In some implementations, the feature set further includes a parking space number bounding box area set. Multiple images of parking space numbers to be detected are input into a preset deep learning network to obtain the corresponding feature set. This further includes: inputting multiple images of parking space numbers to be detected into the preset deep learning network, causing the deep learning network to detect the area of ​​the parking space number bounding boxes to obtain a parking space number bounding box area set; receiving the parking space number bounding box area set output by the deep learning network; and when the coordinates of a target corner point are determined to be within the coordinate range, outputting the target string corresponding to the target parking space number bounding box, including: when the target angle coordinates are determined to be within the coordinate range, determining whether the area of ​​the target parking space number bounding box is greater than an area threshold; and when the area of ​​the target parking space number bounding box is determined to be greater than or equal to the area threshold, outputting the target string corresponding to the target parking space number bounding box.

[0010] In some implementations, the feature set also includes a string length set corresponding to the parking space number frame. Inputting multiple parking space number images to be detected into a preset deep learning network to obtain the corresponding feature set further includes: inputting multiple parking space number images to be detected into a preset deep learning network, causing the deep learning network to detect the string length corresponding to the parking space number frame, obtaining a string length set; receiving the string length set output by the deep learning network; and outputting the target string of the target parking space number frame, including: determining whether the target string length of the target parking space number frame is a preset string length; and when the target string length is determined to be a preset string length, outputting the target string corresponding to the target parking space number frame.

[0011] In some implementations, the feature set also includes a set of character horizontal coordinates corresponding to the parking space number frame. Inputting multiple parking space number images to be detected into a preset deep learning network to obtain the corresponding feature set further includes: inputting multiple parking space number images to be detected into a preset deep learning network, causing the deep learning network to detect the character horizontal coordinates of the characters corresponding to the parking space number frame, obtaining a set of character horizontal coordinates; receiving the character horizontal coordinate set output by the deep learning network; and outputting the target string of the target parking space number frame, including: outputting the target string of the target parking space number frame in ascending order of character horizontal coordinates.

[0012] Secondly, embodiments of this application provide a parking space number detection device, including: an acquisition module, an input module, a clustering module, and an output module. The acquisition module is used to acquire multiple images of parking space numbers to be detected; the input module is used to input the multiple images of parking space numbers to be detected into a preset deep learning network to obtain corresponding feature sets, the feature sets including at least the character set corresponding to the parking space number frame, and an embedded feature set of the center point position of the parking space number frame and the center point position of the character; the clustering module is used to cluster the character set based on the embedded feature set to obtain multiple character clusters, each character cluster corresponding to a string of a parking space number frame; the output module is used to output the target string of the target parking space number frame.

[0013] Thirdly, embodiments of this application provide a vehicle, including: a memory; one or more processors coupled to the memory; and one or more application programs, wherein the one or more application programs are stored in the memory and configured to be executed by the one or more processors, and the one or more application programs are configured to perform the parking space number detection method as provided in the first aspect above.

[0014] Fourthly, embodiments of this application provide a computer-readable storage medium storing program code, which can be called by a processor to execute the parking space number detection method provided in the first aspect above.

[0015] The solution provided in this application acquires multiple images of parking space numbers to be detected and inputs them into a preset deep learning network to obtain corresponding feature sets. The feature sets include at least the character set corresponding to the parking space number frame and the embedded feature set of the center point position of the parking space number frame and the center point position of the character. Based on the embedded feature set, the character set is clustered to obtain multiple character clusters. Each character cluster corresponds to a string of a parking space number frame, and the target string of the target parking space number frame is output. This realizes the automatic detection of parking space numbers in the images to be detected through a deep learning network, without the need for manual operation by the user, thus improving the detection efficiency and accuracy of parking space numbers.

[0016] Furthermore, by clustering the character set based on the embedding features of the center point of the parking space number frame and the center point of the character, the corresponding parking space number can be generated. This can reduce the detection error of the deep learning network for parking space numbers and further improve the detection accuracy of parking space numbers. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This illustration shows a scenario diagram of the parking space number detection system provided in an embodiment of this application.

[0019] Figure 2 A schematic flowchart of a parking space number detection method provided in an embodiment of this application is shown.

[0020] Figure 3 This paper illustrates another flowchart of the parking space number detection method provided in an embodiment of this application.

[0021] Figure 4 This illustration shows another flowchart of the parking space number detection method provided in the embodiments of this application.

[0022] Figure 5 This illustration shows another flowchart of the parking space number detection method provided in the embodiments of this application.

[0023] Figure 6 This paper illustrates another flowchart of the parking space number detection method provided in an embodiment of this application.

[0024] Figure 7 A structural block diagram of a parking space number detection device provided in an embodiment of this application is shown.

[0025] Figure 8A functional block diagram of a vehicle provided in an embodiment of this application is shown.

[0026] Figure 9 This application illustrates a computer-readable storage medium for storing or carrying program code of a parking space number detection method according to an embodiment of the present application. Detailed Implementation

[0027] 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 are only used to explain this application, and should not be construed as limiting this application.

[0028] The following disclosure provides many different implementations or examples for carrying out different structures of this application. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the scope of this application. Furthermore, reference numerals and / or reference letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various implementations and / or arrangements discussed.

[0029] Currently, parking spaces in parking lots are typically assigned corresponding parking space numbers for parking management purposes. These parking space numbers usually consist of a string of characters, such as A001. In the field of autonomous driving, self-driving vehicles need to identify parking space numbers for location or navigation in parking lots.

[0030] Existing technologies typically involve manually extracting features from the acquired parking space number images, and then inputting the extracted features into a classifier for character recognition to obtain the parking space number. However, this traditional image processing method suffers from technical problems such as slow detection speed and low accuracy.

[0031] To address the aforementioned problems, the inventors, after extensive research, have proposed the parking space number detection method, detection device, vehicle, and storage medium provided in this application. This method enables automatic detection of parking space numbers from images using deep learning networks, eliminating the need for manual user operation and improving both the efficiency and accuracy of parking space number detection.

[0032] Furthermore, by clustering the character set based on the embedding features of the center point of the parking space number frame and the center point of the character, the corresponding parking space number can be generated. This can reduce the detection error of the deep learning network for parking space numbers and further improve the detection accuracy of parking space numbers.

[0033] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0034] Please see Figure 1 The illustration shows an application scenario diagram of the parking space number detection system provided in the embodiments of this application, including parking space 100 and vehicle 200. Vehicle 200 can be used to collect parking space images including parking space 100, and can detect the parking space images to obtain the corresponding string of the parking space image, which is the parking space number.

[0035] In some embodiments, parking space 100 may include parking space frame 110, parking space number frame 120 and parking space number 130. Parking space number 130 is embedded in parking space number frame 120 and parking space number frame 120 is embedded in parking space frame 110. Parking space number 130 can be used to identify parking space 100 for the purpose of managing parking space 100.

[0036] In some embodiments, the vehicle 200 may include a frame 210, a control module 220, and an image acquisition module 230. Both the control module 220 and the image acquisition module 230 may be mounted on the frame 210, which may provide mounting support for the control module 220 and the image acquisition module 230.

[0037] In some implementations, the control module 220 can be communicatively connected to the image acquisition module 230. The control module 220 can control the image acquisition module 230 to acquire images of the parking space 100, obtain parking space number images including the parking space number 130, and detect the parking space number images to obtain the corresponding parking space number. The image acquisition module 230 can be a front-facing camera installed at the front of the vehicle, or a panoramic camera composed of multiple cameras installed at the front and sides of the vehicle, etc. The type of image acquisition module 230 is not limited here; it can be configured according to actual needs.

[0038] Please see Figure 2 This document illustrates a flowchart of a parking space number detection method according to an embodiment of this application. In specific embodiments, the parking space number detection method can be applied to, for example... Figure 1 The parking space number detection system shown below uses vehicle 200 as an example to analyze the following. Figure 2 The process shown is described in detail. The parking space number detection method may include the following steps S110 to S140.

[0039] Step S110: Obtain multiple images of parking space numbers to be detected.

[0040] In this embodiment, the vehicle may include a control module and an image acquisition module, and the control module may be communicatively connected to the image acquisition module. The control module may send an image acquisition command to the image acquisition module, and the image acquisition module, upon receiving and responding to the image acquisition command, may take pictures of the parking space to obtain multiple images of the parking space numbers to be detected, and send the multiple images of the parking space numbers to be detected to the control module. The control module receives the multiple images of the parking space numbers to be detected returned by the image acquisition module.

[0041] Multiple images of parking space numbers to be detected can correspond to one parking space or multiple parking spaces, etc. There is no limitation here, and the specific settings can be made according to actual needs.

[0042] In some implementations, the image acquisition module can be a front-facing camera, which can communicate with the control module. The control module can send image acquisition commands to the front-facing camera, which receives and responds to the commands, capturing images of the parking spaces to obtain multiple images of the parking space numbers to be detected. These images are then sent to the control module, which receives the images of the parking spaces returned by the front-facing camera.

[0043] In some implementations, the image acquisition module can be a panoramic camera, which can communicate with the control module. The control module can send image acquisition commands to the panoramic camera, which receives and responds to the commands, capturing images of the parking spaces to obtain multiple images of the parking space numbers to be detected. These images are then sent to the control module, which receives the images of the parking spaces returned by the panoramic camera.

[0044] Step S120: Input multiple images of parking space numbers to be detected into a preset deep learning network to obtain the corresponding feature set.

[0045] In this embodiment, after acquiring multiple images of parking space numbers to be detected, the control module can input these images into a preset deep learning network. The deep learning network receives and responds to the multiple images, extracts features from them, and outputs a corresponding feature set. The control module can receive the feature set output by the deep learning network. The feature set may include at least the character set corresponding to the parking space number frame, and an embedding set representing the center point positions of the parking space number frame and the character center points. The embedding can be used to characterize the relative relationship between the center point positions of the parking space number frame and the character center points.

[0046] Deep learning networks are end-to-end, one-step network models, such as Convolutional Neural Networks (CNNs) and / or Feature Pyramid Networks (FPNs). Deep learning networks possess end-to-end characteristics and powerful self-learning capabilities, enabling automatic feature extraction with minimal manual intervention. This avoids the inaccuracies in feature recognition caused by manually set features during manual feature extraction, thus improving the efficiency and accuracy of parking space number detection.

[0047] In some implementations, the vehicle may further include a local memory that can communicate with the control module. The local memory pre-stores a preset deep learning network. After acquiring multiple images of parking space numbers to be detected, the control module can invoke the deep learning network stored in the local memory and input the multiple images of the parking space numbers to be detected into the deep learning network. The deep learning network receives and responds to the multiple images of the parking space numbers to be detected, extracts features from the multiple images, and outputs a corresponding feature set. The control module can receive the feature set output by the deep learning network.

[0048] In some implementations, the parking space number detection system may further include a server. The server and the control module can communicate via a vehicle network. The server stores a preset deep learning network. After acquiring multiple images of parking space numbers to be detected, the control module can send these images to the server via the vehicle network. The server receives and responds to the multiple images, inputting them into the deep learning network. The deep learning network receives and responds to the images, extracts features from them, and outputs a corresponding feature set. The server then sends the feature set output by the deep learning network to the control module via the vehicle network. The control module can receive the feature set returned by the server.

[0049] Servers can be storage-type cloud servers, computing servers, and integrated cloud computing servers, etc. Vehicle-to-everything (V2X) networks are large-scale systems based on in-vehicle networks, inter-vehicle networks, and in-vehicle mobile internet. They operate according to agreed-upon communication protocols and data exchange standards, enabling wireless communication and information exchange between vehicles, between vehicles and infrastructure, between vehicles and people, and between vehicles and the internet. It is an integrated network capable of achieving intelligent traffic management, intelligent dynamic information services, and intelligent vehicle control. V2X networks can include Wireless Wide Area Networks (WWAN), Wireless Local Area Networks (WLAN), Wireless Metropolitan Area Networks (WMAN), and Wireless Personal Area Networks (WPAN), etc., without further limitation here.

[0050] In some implementations, the feature set may include the character set corresponding to the parking space number frame, and the embedding set of the center point position of the parking space number frame and the center point position of the character. The control module can input multiple parking space number images to be detected into a preset deep learning network. The deep learning network receives and responds to the multiple parking space number images to be detected, and detects the character set, the center point position of the parking space number frame, and the center point position of the character in the multiple parking space number images respectively, to obtain the character set, the center point position set, and the center point set of the character corresponding to the parking space number frame. The center point position set of the parking space number frame and the center point position set of the character corresponding to the parking space number frame are fused to obtain the corresponding embedding set, and the character set and the embedding set are output to the control module. The control module can receive the character set and the embedding set output by the deep learning network.

[0051] In some implementations, the feature set may also include a character category set of the characters corresponding to the parking space number frame. The control module can input multiple parking space number images to be detected into a preset deep learning network. The deep learning network receives and responds to the multiple parking space number images to be detected, identifies the character category of the characters corresponding to the parking space number frame, obtains the corresponding character category set, and outputs the character category set to the control module. The control module can receive the character category set output by the deep learning network.

[0052] In some implementations, the feature set may also include the set of corner coordinates corresponding to the parking space number frame. The control module can input multiple images of parking space numbers to be detected into a preset deep learning network. The deep learning network receives and responds to the multiple images of parking space numbers to be detected, detects the corner coordinates corresponding to the parking space number frame, obtains the corresponding set of corner coordinates, and outputs the set of corner coordinates to the control module. The control module can receive the set of corner coordinates output by the deep learning network.

[0053] In some implementations, the feature set may also include a set of parking space number bounding box areas. The control module can input multiple images of parking space numbers to be detected into a preset deep learning network. The deep learning network receives and responds to the multiple images of parking space numbers to be detected, detects the area of ​​the parking space number bounding box, obtains the corresponding set of parking space number bounding box areas, and outputs the set of parking space number bounding box areas to the control module. The control module can receive the set of parking space number bounding box areas output by the deep learning network.

[0054] In some implementations, the feature set may also include the string length set corresponding to the parking space number frame. The control module can input multiple parking space number images to be detected into a preset deep learning network. The deep learning network receives and responds to the multiple parking space number images to be detected, detects the string length corresponding to the parking space number frame, obtains the corresponding string length set, and outputs the string length set to the control module. The control module can receive the string length set output by the deep learning network.

[0055] In some implementations, the feature set may also include the set of character abscissas corresponding to the parking space number frame. The control module can input multiple parking space number images to be detected into a preset deep learning network. The deep learning network receives and responds to the multiple parking space number images to be detected, detects the character abscissas of the characters corresponding to the parking space number frame, obtains the corresponding character abscissa set, and outputs the character abscissa set to the control module. The control module can receive the character abscissa set output by the deep learning network.

[0056] Step S130: Based on the Embedding set, cluster the character set to obtain multiple character clusters.

[0057] In this embodiment, since Embedding can be used to characterize the relative relationship between the center point of the parking space number frame and the center point of the character, the control module can cluster the character set based on the Embedding set to obtain multiple character clusters. Each character cluster corresponds to a string of a parking space number frame, which can reduce the detection error of the deep learning network for parking space numbers and further improve the detection accuracy of parking space numbers.

[0058] Clustering refers to the process of grouping characters corresponding to multiple embeddings into a single character cluster. A character cluster generated by clustering can be a set of characters. Embeddings corresponding to characters within the same character cluster are similar, while embeddings corresponding to characters within different character clusters are dissimilar. Therefore, when multiple character clusters are obtained through clustering, each character cluster can correspond to the string of a parking space number box.

[0059] It should be noted that the control module can cluster multiple characters using a clustering algorithm without a preset number of clusters, thus avoiding the need for manual limitation on the number of characters. In reality, the number of characters corresponding to multiple parking space number images to be detected is unpredictable.

[0060] In some implementations, the control module can cluster the character set using a density clustering algorithm, eliminating the need to manually determine the number of clusters and enabling the discovery of clusters of arbitrary shapes, effectively handling data noise. Specifically, the clustering results can be controlled by adjusting the embedding parameter and density threshold in density clustering, allowing the algorithm to group two characters with close embeddings but belonging to different parking space number frames into different data clusters.

[0061] In some implementations, each character cluster may include multiple sub-character clusters, and each sub-character cluster corresponds to a target character. The control module may perform sub-clustering on the multiple characters included in each character cluster based on the embedding set to obtain multiple sub-character clusters corresponding to each character cluster, and determine the corresponding string based on the multiple sub-character clusters corresponding to the same parking space number frame.

[0062] Sub-clustering refers to the process of grouping characters corresponding to multiple embeddings into a sub-character cluster. A sub-character cluster generated by sub-clustering can be a set of sub-characters. Sub-characters within the same sub-character cluster have similar embeddings, while sub-characters within different sub-character clusters have different embeddings. Therefore, when sub-clustering yields multiple sub-character clusters, each sub-character cluster can correspond to one character.

[0063] Step S140: Output the target string for the target parking space number box.

[0064] In this embodiment, the control module clusters the character sets corresponding to the Embedding set based on the Embedding set to obtain multiple character clusters, and then outputs the target string of the target parking space number frame, which is the parking space number of the target parking space number frame. This realizes the automatic detection of the parking space number of the image to be detected through a deep learning network, without the need for manual operation by the user, thus improving the detection efficiency and accuracy of the parking space number.

[0065] In some implementations, after the control module clusters the character sets corresponding to the Embedding set based on the Embedding set to obtain multiple character clusters, it can output the target character of the target parking space number frame according to the preset output rules. The preset output rules may include at least one of the following: character category rules, corner coordinate rules, parking space number frame area rules, string length rules, character order rules, etc., which can ensure the accuracy of the output parking space number.

[0066] The solution provided in this application acquires multiple images of parking space numbers to be detected and inputs them into a preset deep learning network to obtain corresponding feature sets. The feature sets include at least the character set corresponding to the parking space number frame and the embedding set of the center point of the parking space number frame and the center point of the character. Based on the embedding set, the character set is clustered to obtain multiple character clusters. Each character cluster corresponds to a string of a parking space number frame, and the target string of the target parking space number frame is output. This realizes the automatic detection of parking space numbers in the images to be detected through a deep learning network, without the need for manual operation by the user, thus improving the detection efficiency and accuracy of parking space numbers.

[0067] Furthermore, by clustering the character set based on the embedding of the center point of the parking space number frame and the center point of the character, the corresponding parking space number is generated. This reduces the detection error of the deep learning network for parking space numbers and further improves the detection accuracy of parking space numbers.

[0068] Please see Figure 3 This illustrates a flowchart of a parking space number detection method provided in another embodiment of this application. In a specific embodiment, the parking space number detection method can be applied to, for example... Figure 1 The parking space number detection system shown below uses vehicle 200 as an example to analyze the following. Figure 3 The process shown is described in detail. The parking space number detection method may include the following steps S210 to S250.

[0069] Step S210: Obtain multiple images of parking space numbers to be detected.

[0070] Step S220: Input multiple images of parking space numbers to be detected into a preset deep learning network to obtain a character set, an embedding set, and a character category set.

[0071] Step S230: Based on the Embedding set, cluster the character set to obtain multiple character clusters.

[0072] In this embodiment, steps S210, S220 and S230 can be referred to the corresponding steps in the previous embodiments, and will not be repeated here.

[0073] Step S240: Determine whether the character category of the target string in the target parking space number frame is a preset character category.

[0074] In this embodiment, the control module can match the character categories of the target string in the target parking space number frame with preset character categories to obtain multiple character matching scores. Based on these multiple character matching scores and a first matching score threshold, the module can determine whether the character category of the target string is a preset character category. When all multiple character matching scores are greater than or equal to the first matching score threshold, the character category of the target string is determined to be a preset character category; when any of the multiple character matching scores is less than the first matching score threshold, the character category of the target string is determined not to be a preset character category.

[0075] Step S250: When the character category of the target string is determined to be the preset character category, output the target string of the target parking space number box.

[0076] In this embodiment, when the control module determines that the character category of the target string is a preset character category, it can output the target string of the target parking space number box, which can avoid outputting unqualified parking space numbers and ensure the accuracy of the output parking space numbers.

[0077] The solution provided in this implementation acquires multiple images of parking space numbers to be detected and inputs them into a preset deep learning network to obtain a character set, an embedding set, and a character category set. Based on the embedding set, the character set is clustered to obtain multiple character clusters. The solution then determines whether the character category of the target string of the target parking space number bounding box is a preset character category. If the character category of the target string is determined to be a preset character category, the target string is output. This avoids outputting incorrect parking space numbers and ensures the accuracy of the output parking space numbers.

[0078] Please see Figure 4 This document illustrates a flowchart of a parking space number detection method provided in another embodiment of this application. In a specific embodiment, the parking space number detection method can be applied to, for example... Figure 1 The parking space number detection system shown below uses vehicle 200 as an example to analyze the following. Figure 4 The process shown is described in detail. The parking space number detection method may include the following steps S310 to S350.

[0079] Step S310: Obtain multiple images of parking space numbers to be detected.

[0080] Step S320: Input multiple images of parking space numbers to be detected into a preset deep learning network to obtain a character set, an embedding set, and a corner coordinate set.

[0081] Step S330: Based on the Embedding set, cluster the character set to obtain multiple character clusters.

[0082] In this embodiment, steps S310, S320 and S330 can be referred to the corresponding steps in the previous embodiments, and will not be repeated here.

[0083] Step S340: Determine whether the coordinates of the target corner point of the target parking space number frame are within the coordinate range corresponding to the parking space number image to be detected.

[0084] In this embodiment, the control module can compare the coordinates of the target corner point of the target parking space number frame with the coordinate range corresponding to the parking space number image to be detected. When the coordinates of the target corner point are greater than the minimum coordinate of the coordinate range and less than the maximum coordinate of the coordinate range, it is determined that the coordinates of the target corner point are within the coordinate range corresponding to the parking space number image to be detected. Otherwise, it is determined that the coordinates of the target corner point are not within the coordinate range corresponding to the parking space number image to be detected.

[0085] Step S350: When the coordinates of the target corner point are determined to be within the coordinate range, output the target string of the target parking space number box.

[0086] In this embodiment, when the control module determines that the target corner point coordinates are within the coordinate range, it can output the target string of the target parking space number frame. This avoids the output parking space number being incomplete when the target corner point coordinates are outside the corresponding parking space number image, thus ensuring the accuracy of the output parking space number.

[0087] In some implementations, the feature set may also include a parking space number frame area set. When the control module determines that the target corner point coordinates are within the coordinate range, it can determine whether the target parking space number frame area is greater than the area threshold. When it is determined that the target parking space number frame area is greater than or equal to the area threshold, the target string of the target parking space number frame is output. This can prevent the detected parking space number from being too far away from the vehicle and ensure that the detected parking space number is close to the vehicle.

[0088] The control module can calculate the area difference between the target parking space number frame area and the area threshold. When the area difference is greater than or equal to zero, the target parking space number frame area is determined to be greater than or equal to the area threshold. When the area difference is less than zero, the target parking space number frame area is less than the area threshold.

[0089] The solution provided in this implementation acquires multiple images of parking space numbers to be detected and inputs them into a preset deep learning network to obtain a character set, an embedding set, and a corner coordinate set. Based on the embedding set, the character set is clustered to obtain multiple character clusters. The solution then determines whether the target corner coordinates of the target parking space number frame are within the coordinate range corresponding to the parking space number image to be detected. When the target corner coordinates are determined to be within the coordinate range, the target string of the target parking space number frame is output. This avoids outputting unqualified parking space numbers and ensures the accuracy of the output parking space numbers.

[0090] Please see Figure 5 This document illustrates a flowchart of a parking space number detection method according to another embodiment of this application. In a specific embodiment, the parking space number detection method can be applied to, for example... Figure 1 The parking space number detection system shown below uses vehicle 200 as an example to analyze the following. Figure 5 The process shown is described in detail. The parking space number detection method may include the following steps S410 to S450.

[0091] Step S410: Obtain multiple images of parking space numbers to be detected.

[0092] Step S420: Input multiple images of parking space numbers to be detected into a preset deep learning network to obtain character set, embedding set and string length set.

[0093] Step S430: Based on the Embedding set, cluster the character set to obtain multiple character clusters.

[0094] In this embodiment, steps S410, S420 and S430 can be referred to the corresponding steps in the previous embodiments, and will not be repeated here.

[0095] Step S440: Determine whether the target string length of the target parking space number frame is the preset string length.

[0096] In this embodiment, the control module can match the target string length of the target parking space number frame with a preset string length to obtain a string length matching degree. Based on the string length matching degree and a second matching degree threshold, it can determine whether the target string length of the target parking space number frame is the preset string length. When the string length matching degree is greater than or equal to the second matching degree threshold, the target string length of the target parking space number frame is determined to be the preset string length; when the string length matching degree is less than the second matching degree threshold, the target string length of the target parking space number frame is determined not to be the preset string length.

[0097] Step S450: When the target string length is determined to be the preset string length, output the target string of the target parking space number box.

[0098] In this embodiment, when the control module determines that the target string length is the preset string length, it can output the target string of the target parking space number box, which can avoid the output string being too short and thus ensure the accuracy of the output parking space number.

[0099] The solution provided in this implementation acquires multiple images of parking space numbers to be detected and inputs them into a preset deep learning network to obtain a character set, an embedding set, and a string length set. Based on the embedding set, the character set is clustered to obtain multiple character clusters. The solution also determines whether the target string length of the target parking space number bounding box is a preset string length. If the target string length is determined to be a preset string length, the target string of the target parking space number bounding box is output. This avoids the output string being too short, which would lead to inaccurate parking space numbers, and ensures the accuracy of the output parking space numbers.

[0100] Please see Figure 6 The document illustrates a flowchart of a parking space number detection method according to another embodiment of this application. In a specific embodiment, the parking space number detection method can be applied to, for example... Figure 1 The parking space number detection system shown below uses vehicle 200 as an example to analyze the following. Figure 6 The process shown is described in detail. The parking space number detection method may include the following steps S510 to S540.

[0101] Step S510: Obtain multiple images of parking space numbers to be detected.

[0102] Step S520: Input multiple images of parking space numbers to be detected into a preset deep learning network to obtain a character set, an embedding set, and a character x-coordinate set.

[0103] Step S530: Based on the Embedding set, cluster the character set to obtain multiple character clusters.

[0104] In this embodiment, steps S510, S520 and S530 can be referred to the corresponding steps in the previous embodiments, and will not be repeated here.

[0105] Step S540: Output the target string of the target parking space number box in ascending order of the character's horizontal coordinate.

[0106] In this embodiment, the control module can arrange the target string of the target parking space number frame in ascending order according to the corresponding character horizontal coordinates to obtain the target sequence string, and output the target sequence string of the target parking space number frame. This can avoid the incorrect output parking space number due to the disordered character order of the output parking space number, and ensure the accuracy of the output parking space number.

[0107] The solution provided in this implementation acquires multiple images of parking space numbers to be detected and inputs them into a preset deep learning network to obtain a character set, an embedding set, and a character x-coordinate set. Based on the embedding set, the character set is clustered to obtain multiple character clusters. The target string of the target parking space number box is output in ascending order of character x-coordinates. This avoids incorrect output parking space numbers due to disordered character order, thus ensuring the accuracy of the output parking space numbers.

[0108] Please see Figure 7 This illustrates a parking space number detection device 600 provided in one embodiment of this application. The parking space number detection device 600 can be applied to applications such as... Figure 1 The parking space number detection system shown below uses vehicle 200 as an example to analyze the following. Figure 8 The parking space number detection device 600 shown will be described in detail. The parking space number detection device 600 may include an acquisition module 610, an input module 620, a clustering module 630, and an output module 640.

[0109] The acquisition module 610 can be used to acquire multiple images of parking space numbers to be detected; the input module 620 can be used to input multiple images of parking space numbers to be detected into a preset deep learning network to obtain the corresponding feature set. The feature set includes at least the character set corresponding to the parking space number frame, and the embedding set of the center point position of the parking space number frame and the center point position of the character; the clustering module 630 can be used to cluster the character set based on the embedding set to obtain multiple character clusters, and each character cluster corresponds to the string of a parking space number frame; the output module 640 can be used to output the target string of the target parking space number frame.

[0110] In some implementations, the feature set may include the character set corresponding to the parking space number frame, and the embedding set of the center point position of the parking space number frame and the center point position of the character. The input module 620 may include a first input unit and a first receiving unit.

[0111] The first input unit can be used to input multiple parking space number images to be detected into a preset deep learning network, so that the deep learning network can detect the characters corresponding to the parking space number frame, the center point position of the parking space number frame, and the center point position of the characters respectively, to obtain the character set, the center point position set of the parking space number frame, and the center point set of the characters corresponding to the parking space number frame, and then fuse the center point position set of the parking space number frame and the center point position set of the characters to obtain the embedding set; the first receiving unit can be used to receive the character set and the embedding set output by the deep learning network.

[0112] In some implementations, the feature set may also include a character category set of the characters corresponding to the parking space number frame, and the input module 620 may also include a second input unit and a second receiving unit.

[0113] The second input unit can be used to input multiple images of parking space numbers to be detected into a preset deep learning network, so that the deep learning network can identify the character category of the characters corresponding to the parking space number frame and obtain a character category set; the second receiving unit can be used to receive the character category set output by the deep learning network.

[0114] In some implementations, the output module 640 may include a first determining unit and a first output unit.

[0115] The first determining unit can be used to determine whether the character category of the target string of the target parking space number frame is a preset character category; the first output unit can be used to output the target string of the target parking space number frame when the character category of the target string is determined to be a preset character category.

[0116] In some implementations, the feature set may also include the set of corner coordinates corresponding to the parking space number frame, and the input module 620 may also include a third input unit and a third receiving unit.

[0117] The third input unit can be used to input multiple images of parking space numbers to be detected into a preset deep learning network, so that the deep learning network can detect the corner coordinates corresponding to the parking space number frame and obtain a set of corner coordinates; the third receiving unit can be used to receive the set of corner coordinates output by the deep learning network.

[0118] In some implementations, the output module 640 may further include a second determining unit and a second output unit.

[0119] The second determining unit can be used to determine whether the coordinates of the target corner point of the target parking space number frame are within the coordinate range corresponding to the parking space number image to be detected; the second output unit can be used to output the target string of the target parking space number frame when it is determined that the target corner point coordinates are within the coordinate range.

[0120] In some implementations, the feature set may also include a parking space number frame area set, and the input module 620 may also include a fourth input unit and a fourth receiving unit.

[0121] The fourth input unit can be used to input multiple images of parking space numbers to be detected into a preset deep learning network, so that the deep learning network can detect the area of ​​the parking space number frame and obtain a set of parking space number frame areas; the fourth receiving unit can be used to receive the set of parking space number frame areas output by the deep learning network.

[0122] In some implementations, the second output unit may include a determining subunit and an output subunit.

[0123] The determination sub-unit can be used to determine whether the area of ​​the target parking space number frame is greater than the area threshold when the target angle coordinates are within the coordinate range; the output sub-unit can be used to output the target string of the target parking space number frame when the area of ​​the target parking space number frame is greater than or equal to the area threshold.

[0124] In some implementations, the feature set may also include the string length set corresponding to the parking space number frame, and the input module 620 may also include a fifth input unit and a fifth receiving unit.

[0125] The fifth input unit can be used to input multiple parking space number images to be detected into a preset deep learning network, so that the deep learning network can detect the string length corresponding to the parking space number frame and obtain a string length set; the fifth receiving unit can be used to receive the string length set output by the deep learning network.

[0126] In some implementations, the output module 640 may further include a third determining unit and a third output unit.

[0127] The third determining unit can be used to determine whether the target string length of the target parking space number frame is the preset string length; the third output unit can be used to output the target string of the target parking space number frame when the target string length is determined to be the preset string length.

[0128] In some implementations, the feature set also includes the character horizontal coordinate set corresponding to the parking space number frame, and the input module 620 may also include a sixth input unit and a sixth receiving unit.

[0129] The sixth input unit can be used to input multiple parking space number images to be detected into a preset deep learning network, so that the deep learning network can detect the horizontal coordinates of the characters corresponding to the parking space number frames and obtain a set of character horizontal coordinates; the sixth receiving unit can be used to receive the set of character horizontal coordinates output by the deep learning network.

[0130] In some implementations, the output module 640 may also include a fourth output unit.

[0131] The fourth output unit can be used to output the target string of the target parking space number box in ascending order of the character's horizontal coordinate.

[0132] The solution provided in this application acquires multiple images of parking space numbers to be detected and inputs them into a preset deep learning network to obtain corresponding feature sets. The feature sets include at least the character set corresponding to the parking space number frame and the embedding set of the center point of the parking space number frame and the center point of the character. Based on the embedding set, the character set is clustered to obtain multiple character clusters. Each character cluster corresponds to a string of a parking space number frame, and the target string corresponding to the target parking space number frame is output. This realizes the automatic detection of parking space numbers in the images to be detected through a deep learning network, without the need for manual operation by the user, thus improving the detection efficiency and accuracy of parking space numbers.

[0133] Furthermore, by clustering the character set based on the embedding of the center point of the parking space number frame and the center point of the character, the corresponding parking space number is generated. This reduces the detection error of the deep learning network for parking space numbers and further improves the detection accuracy of parking space numbers.

[0134] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For device embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to in the descriptions of the method embodiments. Any processing method described in the method embodiments can be implemented in the device embodiments through corresponding processing modules, and will not be elaborated upon further in the device embodiments.

[0135] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0136] Please see Figure 8 The diagram illustrates a functional block diagram of a vehicle 700 provided in another embodiment of this application. The vehicle 700 may include one or more components such as a memory 710, a processor 720, and one or more application programs. The one or more application programs may be stored in the memory 710 and configured to be executed by one or more processors 720. The one or more application programs are configured to perform the methods described in the foregoing method embodiments.

[0137] The memory 710 may include random access memory (RAM) or read-only memory. The memory 710 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 710 may include a program storage area and a data storage area. The program storage area may store instructions for implementing an operating system, instructions for implementing at least one function (such as acquiring an image of a parking space number to be detected, inputting the image to be detected into a deep learning network, obtaining a feature set, clustering an embedding set, obtaining a target character set, outputting a target string, detecting the center point position of the parking space number frame, detecting the center point position of a character, obtaining a set of center point positions of the parking space number frame, obtaining a set of center point positions of the character, fusing the set of center point positions of the parking space number frame and the set of center point positions of the character, receiving an embedding set, identifying the character category of a character, obtaining a set of character categories, determining the character category, detecting corner coordinates, obtaining a set of corner coordinates, determining the range of target angle coordinates, detecting the area of ​​the parking space number frame, obtaining a set of areas of the parking space number frame, determining the size of the area of ​​the parking space number frame, detecting the string length, obtaining a set of string lengths, determining the size of the target string length, detecting the horizontal coordinate of a character, and obtaining a set of horizontal coordinates of a character, etc.), and instructions for implementing the various method embodiments described below. The storage data area can also store data created by the vehicle 700 during use (such as parking space number images, deep learning networks, feature sets, embedding sets, target character sets, parking space number frames, target parking space number frames, target strings, parking space number frame center point positions, character center point positions, parking space number frame center point position sets, character center point sets, character category sets, preset character categories, corner coordinate sets, coordinate ranges, parking space number frame area sets, area thresholds, string length sets, preset string lengths, character horizontal coordinate sets, and character coordinate order).

[0138] The processor 720 may include one or more processing cores. The processor 720 connects to various parts of the vehicle 700 via various interfaces and lines, and performs various functions and processes data of the vehicle 700 by running or executing instructions, programs, code sets, or instruction sets stored in the memory 710, and by calling data stored in the memory 710. Optionally, the processor 720 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 720 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the displayed content; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 720 and may be implemented separately through a communication chip.

[0139] Please refer to Figure 9 This diagram illustrates a structural block diagram of a computer-readable storage medium provided in an embodiment of this application. The computer-readable storage medium 800 stores program code 810, which can be called by a processor to execute the methods described in the above method embodiments.

[0140] The computer-readable storage medium 800 may be an electronic memory such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, hard disk, or ROM. Optionally, the computer-readable storage medium 800 includes a non-transitory computer-readable storage medium. The computer-readable storage medium 800 has storage space for program code 810 that performs any of the method steps described above. This program code can be read from or written to one or more computer program products. The program code 810 may be compressed, for example, in a suitable form.

[0141] The solution provided in this application acquires multiple images of parking space numbers to be detected and inputs them into a preset deep learning network to obtain corresponding feature sets. The feature sets include at least the character set corresponding to the parking space number frame and the embedding set of the center point of the parking space number frame and the center point of the character. Based on the embedding set, the character set is clustered to obtain multiple character clusters. Each character cluster corresponds to a string of a parking space number frame, and the target string of the target parking space number frame is output. This realizes the automatic detection of parking space numbers in the images to be detected through a deep learning network, without the need for manual operation by the user, thus improving the detection efficiency and accuracy of parking space numbers.

[0142] Furthermore, by clustering the character set based on the embedding of the center point of the parking space number frame and the center point of the character, the corresponding parking space number is generated. This reduces the detection error of the deep learning network for parking space numbers and further improves the detection accuracy of parking space numbers.

[0143] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for detecting parking space numbers, characterized in that, include: Acquire multiple images of parking space numbers to be detected; Multiple images of parking space numbers to be detected are input into a preset deep learning network to obtain a corresponding feature set. The feature set includes at least the character set corresponding to the parking space number frame, and an embedded feature set of the center point position of the parking space number frame and the center point position of the character. The embedded feature set is used to characterize the relative relationship between the center point position of the parking space number frame and the center point position of the character. Based on the embedded feature set, the character set is clustered to obtain multiple character clusters, and each character cluster corresponds to a string of a parking space number frame; Output the target string for the target parking space number box.

2. The parking space number detection method according to claim 1, characterized in that, The feature set includes the character set corresponding to the parking space number frame, and the embedded feature set of the center point position of the parking space number frame and the center point position of the character. The step of inputting multiple images of the parking space numbers to be detected into a preset deep learning network to obtain the corresponding feature set includes: Multiple images of parking space numbers to be detected are input into a preset deep learning network, which detects the characters corresponding to the parking space number frame, the center point position of the parking space number frame, and the center point position of the characters, respectively, to obtain the character set, the center point position set of the parking space number frame, and the character center point set. The center point position set of the parking space number frame and the character center point position set are then fused to obtain the embedded feature set. Receive the character set and the embedded feature set output by the deep learning network.

3. The parking space number detection method according to claim 1, characterized in that, The feature set also includes a character category set of the characters corresponding to the parking space number frames. The step of inputting multiple images of the parking space numbers to be detected into a preset deep learning network to obtain the corresponding feature set further includes: Multiple images of the parking space numbers to be detected are input into a preset deep learning network, which enables the deep learning network to identify the character category of the characters corresponding to the parking space number frames, thereby obtaining a character category set; Receive the character category set output by the deep learning network; The target string for the output target parking space number box includes: Determine whether the character category of the target string in the target parking space number frame is a preset character category; When the character category of the target string is determined to be the preset character category, the target string of the target parking space number frame is output.

4. The parking space number detection method according to claim 1, characterized in that, The feature set also includes the set of corner coordinates corresponding to the parking space number frame. The step of inputting multiple images of the parking space numbers to be detected into a preset deep learning network to obtain the corresponding feature set further includes: Multiple images of the parking space numbers to be detected are input into a preset deep learning network, which detects the corner coordinates corresponding to the parking space number frames to obtain a set of corner coordinates. Receive the set of corner coordinates output by the deep learning network; The target string for the output target parking space number box includes: Determine whether the coordinates of the target corner point of the target parking space number frame are within the coordinate range corresponding to the image of the parking space number to be detected; When it is determined that the coordinates of the target corner point are within the range of the coordinates, the target string of the target parking space number frame is output.

5. The parking space number detection method according to claim 4, characterized in that, The feature set also includes a set of parking space number bounding box areas. The step of inputting multiple images of the parking space numbers to be detected into a preset deep learning network to obtain the corresponding feature set further includes: Multiple images of the parking space numbers to be detected are input into a preset deep learning network, which detects the area of ​​the parking space number frame to obtain a set of parking space number frame areas. Receive the parking space number frame area set output by the deep learning network; When it is determined that the coordinates of the target corner point are within the coordinate range, the target string of the target parking space number frame is output, including: When it is determined that the coordinates of the target corner point are within the range of the coordinates, it is determined whether the area of ​​the target parking space number frame is greater than the area threshold. When it is determined that the area of ​​the target parking space number frame is greater than or equal to the area threshold, the target string of the target parking space number frame is output.

6. The parking space number detection method according to claim 1, characterized in that, The feature set also includes the string length set corresponding to the parking space number frame. The step of inputting multiple images of the parking space numbers to be detected into a preset deep learning network to obtain the corresponding feature set further includes: Multiple images of the parking space numbers to be detected are input into a preset deep learning network, which detects the length of the string corresponding to the parking space number frame to obtain a string length set. Receive the string length set output by the deep learning network; The target string for the output target parking space number box includes: Determine if the target string length of the target parking space number frame is the preset string length; When the length of the target string is determined to be the preset string length, the target string of the target parking space number frame is output.

7. The parking space number detection method according to any one of claims 1 to 6, characterized in that, The feature set also includes the set of character horizontal coordinates corresponding to the parking space number frames. The step of inputting multiple images of the parking space numbers to be detected into a preset deep learning network to obtain the corresponding feature set further includes: Multiple images of parking space numbers to be detected are input into a preset deep learning network, which detects the horizontal coordinates of the characters corresponding to the parking space number frames to obtain a set of character horizontal coordinates. Receive the set of character x-coordinates output by the deep learning network; The target string for the output target parking space number box includes: Output the target string of the target parking space number frame in ascending order of the character's horizontal coordinate.

8. A parking space number detection device, characterized in that, include: The acquisition module is used to acquire multiple images of parking space numbers to be detected; The input module is used to input multiple images of the parking space numbers to be detected into a preset deep learning network to obtain a corresponding feature set. The feature set includes at least the character set corresponding to the parking space number frame, and an embedded feature set of the center point position of the parking space number frame and the center point position of the character. The embedded feature set is used to characterize the relative relationship between the center point position of the parking space number frame and the center point position of the character. The clustering module is used to cluster the character set based on the embedded feature set to obtain multiple character clusters, each character cluster corresponding to a string of a parking space number frame; The output module is used to output the target string of the target parking space number frame.

9. A vehicle, characterized in that, include: Memory; One or more processors are coupled to the memory; One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by one or more processors, the one or more applications being configured to perform the parking space number detection method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains program code, which can be called by a processor to execute the parking space number detection method as described in any one of claims 1 to 7.

Images