Parking space number detection method, device and equipment

Abstract

The present application discloses a parking space number detection method, device and equipment, the method comprising: obtaining a parking lot image to be detected; inputting the parking lot image to be detected into a convolutional neural network model to detect the parking space number, and obtaining a parking space number detection result corresponding to the parking lot image to be detected. The present application improves the existing parking space number detection method which uses a traditional image processing method to detect the parking space number, and has the technical problems of slow detection speed and low accuracy.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a parking space number detection method, device and equipment. Background Art

[0002] At present, parking spaces in parking lots are usually equipped with corresponding parking space numbers to facilitate parking space management. Usually, parking space numbers are composed of a string of characters, such as A001. In the field of unmanned driving, unmanned vehicles need to identify parking space numbers and other information in parking lots to locate or navigate. The existing technology usually extracts features manually from the collected images, and then inputs the extracted features into the classifier for character recognition. This traditional image processing method has the technical problems of slow detection speed and low accuracy. Summary of the invention

[0003] The present application provides a parking space number detection method, device and equipment, which are used to improve the existing parking space number detection method that uses traditional image processing methods to detect parking space numbers, and has technical problems such as slow detection speed and low accuracy.

[0004] In view of this, the first aspect of the present application provides a parking space number detection method, comprising:

[0005] Obtain the parking lot image to be detected;

[0006] The parking lot image to be detected is input into a convolutional neural network model to perform parking space number detection, and a parking space number detection result corresponding to the parking lot image to be detected is obtained.

[0007] Optionally, the convolutional neural network model is a first network model;

[0008] Inputting the parking lot image to be detected into the first network model to detect the parking space number, and obtaining the parking space number detection result corresponding to the parking lot image to be detected, including:

[0009] The parking lot image to be detected is input into the first network model, so that the first network model sequentially performs region of interest extraction, parking space number area corner point detection, parking space number area rotation and parking space number character sequence prediction on the parking lot image to be detected, and outputs the parking space number detection result corresponding to the parking lot image to be detected.

[0010] Optionally, the first network model includes a first submodule, a second submodule, a third submodule and a fourth submodule connected in sequence;

[0011] The first network model sequentially extracts the region of interest, detects corner points in the parking space number region, rotates the parking space number region, and predicts the parking space number character sequence for the parking lot image to be detected, and outputs the parking space number detection result corresponding to the parking lot image to be detected, including:

[0012] The first submodule extracts the region of interest from the parking lot image to be detected to obtain a feature map of the region of interest;

[0013] The second submodule detects the positions of several corner points of the parking space number area in the feature map of the region of interest to obtain a feature map of the parking space number area;

[0014] The third submodule rotates the parking space number area characteristic map to obtain a standard parking space number area characteristic map;

[0015] The fourth submodule predicts a parking space number character sequence for the standard parking space number area feature map, and outputs a parking space number detection result corresponding to the parking lot image to be detected.

[0016] Optionally, the fourth submodule predicts a parking space number character sequence for the standard parking space number area feature map and outputs a parking space number detection result corresponding to the parking lot image to be detected, including:

[0017] The fourth submodule performs feature extraction and feature fusion on the standard parking space number area feature map, predicts the parking space number character sequence based on the fused features, and outputs the parking space number detection result corresponding to the parking lot image to be detected.

[0018] Optionally, the configuration process of the first network model is:

[0019] Acquire a first training image, where the first training image is annotated with coordinates of a plurality of corner points of a parking space number area and a parking space number character string category;

[0020] The first network is trained using the first training image to obtain the first network model.

[0021] Optionally, the convolutional neural network model is a second network model;

[0022] Inputting the parking lot image to be detected into the second network model to detect the parking space number, and obtaining the parking space number detection result corresponding to the parking lot image to be detected, including:

[0023] The parking lot image to be detected is input into the second network model, so that the second network model performs parking space number frame detection, character frame detection and character recognition on the parking lot image to be detected in parallel, and outputs the parking space number detection result corresponding to the parking lot image to be detected.

[0024] Optionally, the second network model performs parking space number frame detection, character frame detection and character recognition on the parking lot image to be detected in parallel, and outputs a parking space number detection result corresponding to the parking lot image to be detected, including:

[0025] The second network model performs feature extraction and multi-scale feature fusion on the parking lot image to be detected, and outputs a plurality of scale feature maps;

[0026] The second network model performs parking space number frame detection, character frame detection and character recognition on each of the scale feature maps in parallel, and outputs parking space number frame detection results, character frame detection results and character recognition detection results corresponding to each of the scale feature maps;

[0027] The second network model obtains the parking space number detection result corresponding to the parking lot image to be detected based on the parking space number frame detection results, character frame detection results and character recognition detection results corresponding to all the scale feature maps.

[0028] Optionally, the configuration process of the second network model is:

[0029] Acquire a second training image, wherein the second training image is annotated with the position of the parking space number and the position and category of each character in the parking space number;

[0030] The second network is trained using the second training image to obtain the second network model.

[0031] A second aspect of the present application provides a parking space number detection device, comprising:

[0032] An acquisition unit, used for acquiring an image of a parking lot to be detected;

[0033] The detection unit is used to input the parking lot image to be detected into a convolutional neural network model to detect the parking space number, and obtain a parking space number detection result corresponding to the parking lot image to be detected.

[0034] Optionally, the convolutional neural network model is a first network model, and the detection unit is specifically used to:

[0035] The parking lot image to be detected is input into the first network model, so that the first network model sequentially performs region of interest extraction, parking space number area corner point detection, parking space number area rotation and parking space number character sequence prediction on the parking lot image to be detected, and outputs the parking space number detection result corresponding to the parking lot image to be detected.

[0036] Optionally, the convolutional neural network model is a second network model, and the detection unit is specifically used for:

[0037] The parking lot image to be detected is input into the second network model, so that the second network model performs parking space number frame detection, character frame detection and character recognition on the parking lot image to be detected in parallel, and outputs the parking space number detection result corresponding to the parking lot image to be detected.

[0038] A third aspect of the present application provides a parking space number detection device, the device comprising a processor and a memory;

[0039] The memory is used to store program code and transmit the program code to the processor;

[0040] The processor is used to execute any one of the parking space number detection methods described in the first aspect according to the instructions in the program code.

[0041] It can be seen from the above technical solutions that this application has the following advantages:

[0042] The present application provides a parking space number detection method, comprising: obtaining an image of a parking lot to be detected; inputting the image of the parking lot to be detected into a convolutional neural network model for parking space number detection, and obtaining a parking space number detection result corresponding to the image of the parking lot to be detected. In the present application, the parking space number detection is performed end-to-end on the acquired image of the parking lot to be detected by a convolutional neural network model, and the detection efficiency is high; the feature extraction is performed by the powerful self-learning ability of the convolutional neural network model, and the detection accuracy is high, thereby improving the existing parking space number detection method that uses traditional image processing methods to detect parking space numbers, and the technical problems of slow detection speed and low accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0044] Figure 1 A schematic diagram of a flow chart of a first embodiment of a parking space number detection method provided by the present application;

[0045] Figure 2 A schematic diagram of a flow chart of a second embodiment of a parking space number detection method provided by the present application;

[0046] Figure 3 A schematic diagram of a flow chart of a third embodiment of a parking space number detection method provided in the present application;

[0047] Figure 4 A structural diagram of a first network model provided by the present application;

[0048] Figure 5 A structural diagram of a second network model provided by the present application;

[0049] Figure 6 A schematic diagram of the structure of a parking space number detection device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0050] The present application provides a parking space number detection method, device and equipment, which are used to improve the existing parking space number detection method that uses traditional image processing methods to detect parking space numbers, and has technical problems such as slow detection speed and low accuracy.

[0051] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0052] Method embodiment 1:

[0053] For easier understanding, see Figure 1 , an embodiment of a parking space number detection method provided by the present application includes:

[0054] Step 101: Acquire an image of a parking lot to be detected.

[0055] In an embodiment of the present application, an image of the parking lot to be detected may be collected by a camera in the parking lot, wherein the image of the parking lot to be detected may include one or more parking space numbers.

[0056] Step 102: Input the parking lot image to be detected into the convolutional neural network model to detect the parking space number, and obtain the parking space number detection result corresponding to the parking lot image to be detected.

[0057] Existing parking space number detection methods are usually divided into two steps: first, manual feature extraction is performed on the collected image; then, the extracted features are input into the classifier for parking space number recognition. However, the extracted manual features are manually set, and the accuracy of the final recognition result cannot be guaranteed. Based on this, the embodiment of the present application adopts a convolutional neural network model for parking space number detection. The convolutional neural network model has end-to-end characteristics and powerful self-learning capabilities. It can automatically extract features without excessive manual intervention, which helps to improve the detection efficiency and accuracy of parking space numbers. Specifically, the parking lot image to be detected is input into the convolutional neural network model, and the convolutional neural network model automatically extracts features from the parking lot image to be detected, and detects the parking space number based on the extracted features to obtain the parking space number detection result corresponding to the parking lot image to be detected. It can be understood that when the parking lot image to be detected contains multiple parking space numbers, multiple parking space numbers are output accordingly.

[0058] Furthermore, considering that the parking space number characters in the collected parking lot image to be detected may be blurred, the human eye may not be able to recognize the parking space number, so it is likely that the parking space number cannot be detected or is prone to detection errors when detected by the convolutional neural network model. Based on this, the embodiment of the present application can use an unsupervised or supervised method to train a convolutional neural network to obtain a convolutional neural network model, and the convolutional neural network model can be used to determine whether the parking space number characters in the parking lot image to be detected are blurred. After determining that the parking space number characters are blurred, the parking space number character sequence is no longer predicted for the parking space number, thereby ensuring the accuracy of the parking space number detection result.

[0059] In the embodiment of the present application, end-to-end parking space number detection is performed on the acquired parking lot image to be detected through a convolutional neural network model, and the detection efficiency is high; feature extraction is performed through the powerful self-learning ability of the convolutional neural network model, and the detection accuracy is high, thereby improving the existing parking space number detection method that uses traditional image processing methods to detect parking space numbers, and the technical problems of slow detection speed and low accuracy.

[0060] The above is Example 1 of a parking space number detection method provided by the present application. The following is Example 2 of a parking space number detection method provided by the present application. Example 2 is a specific description of the parking space number detection method when the convolutional neural network model in Example 1 is the first network model.

[0061] Method embodiment 2:

[0062] For easier understanding, see Figure 2 , an embodiment of a parking space number detection method provided by the present application includes:

[0063] Step 201: Acquire an image of a parking lot to be detected.

[0064] In an embodiment of the present application, an image of the parking lot to be detected may be collected by a camera in the parking lot, wherein the image of the parking lot to be detected may include one or more parking space numbers.

[0065] Step 202: input the parking lot image to be detected into the first network model, so that the first network model sequentially performs region of interest extraction, parking space number area corner point detection, parking space number area rotation and parking space number character sequence prediction on the parking lot image to be detected, and outputs the parking space number detection result corresponding to the parking lot image to be detected.

[0066] The first network model in the embodiment of the present application includes a first submodule, a second submodule, a third submodule and a fourth submodule, and the first submodule, the second submodule, the third submodule and the fourth submodule are connected in sequence.

[0067] The first submodule is used to extract the region of interest from the parking lot image to be detected, and obtain a region of interest feature map. The first submodule may be a ROI (region of interest) network model.

[0068] The second submodule detects several corner point positions of the parking space number area in the feature map of the region of interest to obtain a feature map of the parking space number area. Usually, the shape of the parking space number area is a matrix, but considering that the image of the parking lot to be detected collected by the camera will have a certain deformation, the parking space number area in the image of the parking lot to be detected will also be deformed. If only the two corner point positions of the parking space number area are detected, an incomplete parking space number area may be extracted. Therefore, the embodiment of the present application preferably detects the four corner point positions of the parking space number area through the second submodule, and then the parking space number area can be accurately determined to obtain a complete feature map of the parking space number area. Among them, the second submodule can be a KeypointNet (key point network) model, and the KeypointNet model can use a residual network.

[0069] Since the parking space number area in the collected parking lot image to be detected may be tilted, the embodiment of the present application rotates the parking space number area feature map through the third submodule to solve the problem of the parking space number area tilt, which helps to improve the accuracy of the parking space number detection result; after the third submodule rotates the parking space number area feature map, the size of the parking space number area feature map can be further converted to obtain a standard parking space number area feature map of a fixed size. Among them, the third submodule can be an RROI (rotated region of interest) pooling layer.

[0070] The fourth submodule extracts features from the standard parking space number area feature map, predicts the parking space number character sequence based on the extracted features, and outputs the parking space number detection result corresponding to the parking lot image to be detected. In order to further improve the accuracy of the parking space number character sequence prediction of the fourth submodule, the embodiment of the present application improves the fourth submodule by performing feature fusion on the features extracted by the fourth submodule to enhance the feature representation.

[0071] Specifically, the fourth submodule performs feature extraction and feature fusion on the standard parking space number area feature map, and predicts the parking space number character sequence based on the fused features, and outputs the parking space number detection result corresponding to the parking lot image to be detected. Among them, the fourth submodule can be a CRNN (Convolutional Recurrent Neural Network) model, and the CRNN model is composed of a CNN (Convolutional Neural Network) model and a RNN (Recurrent Neural Network) model. The embodiment of the present application further improves the CRNN model, mainly improving the CNN model in the CRNN model. Specifically, the features extracted by the CNN model are further fused, and feature fusion can be performed by channel splicing. The fused features are then input into the RNN model to predict the parking space number character sequence. The structure of the first network model in the embodiment of the present application can be referred to. Figure 4 Of course, other network structures are also possible, which will not be illustrated one by one here.

[0072] Furthermore, the configuration process of the first network model is: obtaining a first training image, the first training image is annotated with several corner point coordinates of the parking space number area and the parking space number string category; training the first network through the first training image to obtain the first network model.

[0073] In an embodiment of the present application, a parking lot image can be obtained through a parking lot camera, or a parking lot image can be obtained from a public data set, and then several corner point coordinates and parking space number string categories of the parking space number area of ​​the parking lot image are annotated to obtain a first training image, wherein the embodiment of the present application preferably annotates the 4 corner point coordinates of the parking space number area. The first training image used to train the first network is annotated with the 4 corner point coordinates of the parking space number area and the parking space number string category, so that the first network learns the 4 corner point position features of the parking space number area and the parking space number string category features. The specific training process of the first network is similar to the training process of the existing deep learning model, and will not be repeated here.

[0074] In the embodiment of the present application, the first submodule in the first network model is used to extract the region of interest of the parking lot image to be detected, so as to extract the key area in the parking lot image to be detected, thereby reducing redundant feature representation and improving detection speed and detection accuracy; the second submodule is used to detect the positions of the four corner points of the parking space number area in the feature map of the region of interest to obtain a complete feature map of the parking space number area; the third submodule is used to rotate the feature map of the parking space number area to solve the problem of the tilt of the parking space number area and improve the detection accuracy of the parking space number; the fourth submodule is used to extract and fuse features of the standard parking space number area feature map, and the parking space number character sequence is predicted based on the fused features, and the feature representation is improved by feature fusion, thereby improving the detection accuracy, thereby improving the existing parking space number detection method which uses traditional image processing methods to detect parking space numbers, and has the technical problems of slow detection speed and low accuracy.

[0075] The above is Example 2 of a parking space number detection method provided in the present application. The following is Example 3 of a parking space number detection method provided in the present application. Example 3 is a specific description of the parking space number detection method when the convolutional neural network model in Example 1 is the second network model.

[0076] Method embodiment three:

[0077] For easier understanding, see Figure 3 , an embodiment of a parking space number detection method provided by the present application includes:

[0078] Step 301: Acquire an image of a parking lot to be detected.

[0079] In an embodiment of the present application, an image of the parking lot to be detected may be collected by a camera in the parking lot, wherein the image of the parking lot to be detected may include one or more parking space numbers.

[0080] Step 302: input the parking lot image to be detected into the second network model, so that the second network model performs parking space number frame detection, character frame detection and character recognition on the parking lot image to be detected in parallel, and outputs the parking space number detection result corresponding to the parking lot image to be detected.

[0081] After the parking lot image to be detected is input into the second network model, the second network model extracts features from the parking lot image to obtain a feature map. The second network model performs parking space number frame detection, character frame detection, and character recognition on the feature map in parallel. According to the parking space number frame detection results, character frame detection results, and character recognition results, the parking space number string and the position, size, and category of a single character are automatically generated, and the parking space number detection result corresponding to the parking lot image to be detected is output. The structure of the second network model in the embodiment of the present application can be referred to. Figure 5, where the Backbone network is used to extract features from the input image, which can be a residual network (ResNet-50), a visual geometry group network (VGG) or other network models.

[0082] Furthermore, considering that the relative positions of the same camera and each parking space in the parking lot are different, the parking space number characters in the collected parking lot images to be detected may be different in size, which will affect the detection accuracy of the parking space number. The embodiment of the present application improves the second network model by embedding Feature Pyramid Networks (FPN) in the second network model.

[0083] Specifically, the features extracted by the second network model are subjected to multi-scale feature fusion through FPN, and a plurality of scale feature maps are output; the second network model performs parking space number frame detection, character frame detection and character recognition on each scale feature map in parallel, and outputs the parking space number frame detection result, character frame detection result and character recognition detection result corresponding to each scale feature map; the second network model obtains the parking space number detection result corresponding to the parking lot image to be detected based on the parking space number frame detection results, character frame detection results and character recognition detection results corresponding to all scale feature maps.

[0084] In the embodiment of the present application, multi-scale feature fusion is performed on the extracted feature map through FPN to obtain scale feature maps of different sizes; and multi-scale prediction is then performed on the different scale feature maps to solve the problem of different sizes of parking space number characters.

[0085] Furthermore, the configuration process of the second network model is: obtaining a second training image, the second training image is annotated with the parking space number position, the position and category of each character in the parking space number; training the second network through the second training image to obtain the second network model.

[0086] In an embodiment of the present application, a parking lot image can be obtained through a parking lot camera or from a public dataset, and then the parking lot image is labeled with the parking space number position, the position of each character in the parking space number, and the category of each character to obtain a second training image; the second network is trained through the second training image, so that the second network learns the foreground and background attributes of the characters at each pixel position, the foreground and background attributes of the parking space number, the upper, lower, left, and right offsets of the character frame, the upper, lower, left, and right offsets of the parking space number frame, and the category of the parking space number character, and automatically generates the parking space number character string and the position, size, and category of a single character based on these features, and then outputs the final parking space number detection result.

[0087] In the embodiment of the present application, the parking space number detection is performed end-to-end on the parking lot image to be detected through the second network model, which is a one-stage network model, thereby improving the parking space number detection speed; the FPN in the second network model performs multi-scale feature fusion on the extracted feature map, and performs multi-scale prediction on the obtained multi-scale feature map, which can solve the problem of different character sizes, thereby improving the parking space number detection accuracy, and improving the existing parking space number detection method. The traditional image processing method is used for parking space number detection, which has the technical problems of slow detection speed and low accuracy.

[0088] Furthermore, the settings of the first network model and the second network model can be selected by those skilled in the art according to actual needs. For example, when less annotation information is needed for the training data used to train the model, and the problem of tilted parking space number recognition needs to be solved, the first network model can be selected, because the first network model only needs to annotate the coordinates of the four corner points of the parking space number area and the parking space number string category, while the second network model requires character-level annotation data, including the position, area, and category of the characters. When fast detection speed is required and the problem of different sizes of parking space number characters can be solved, the second network model can be selected, because the second network model is equivalent to a one-stage network model compared to the first network model, while the first network model is equivalent to a multi-stage network model, the detection speed of the second network model is relatively fast, and the second network model can solve the problem of different character sizes.

[0089] The above is Example 3 of a parking space number detection method provided by the present application, and the following is an embodiment of a parking space number detection device provided by the present application.

[0090] For easier understanding, please refer to Figure 6 , a parking space number detection device provided in an embodiment of the present application includes:

[0091] An acquisition unit 601 is used to acquire an image of a parking lot to be detected;

[0092] The detection unit 602 is used to input the parking lot image to be detected into the convolutional neural network model to detect the parking space number, and obtain the parking space number detection result corresponding to the parking lot image to be detected.

[0093] As a further improvement, the parking space number detection device in the embodiment of the present application also includes: a configuration unit 603, which is used to configure the convolutional neural network model.

[0094] In the embodiment of the present application, the parking space number detection device performs end-to-end parking space number detection on the acquired parking lot image to be detected through a convolutional neural network model, with high detection efficiency. Feature extraction is performed through the powerful self-learning ability of the convolutional neural network model, and the detection accuracy is high, thereby improving the existing parking space number detection method that uses traditional image processing methods to detect parking space numbers, and has the technical problems of slow detection speed and low accuracy.

[0095] The above is Example 1 of a parking space number detection device provided by the present application. The following is Example 2 of a parking space number detection device provided by the present application. Example 2 is a specific description of when the convolutional neural network model in Example 1 is the first network model.

[0096] Device embodiment 2:

[0097] The present application provides a parking space number detection device, comprising:

[0098] An acquisition unit 601 is used to acquire an image of a parking lot to be detected;

[0099] The detection unit 602 is used to input the parking lot image to be detected into the convolutional neural network model to detect the parking space number, and obtain the parking space number detection result corresponding to the parking lot image to be detected.

[0100] As a further improvement, the convolutional neural network model is the first network model, and the detection unit 602 is specifically used for:

[0101] The parking lot image to be detected is input into the first network model, so that the first network model sequentially extracts the region of interest, detects the corner points of the parking space number area, rotates the parking space number area and predicts the parking space number character sequence for the parking lot image to be detected, and outputs the parking space number detection result corresponding to the parking lot image to be detected.

[0102] As a further improvement, the first network model includes a first submodule, a second submodule, a third submodule and a fourth submodule;

[0103] The detection unit 602 is specifically used for:

[0104] The parking lot image to be detected is input into the first network model for parking space number detection, so that the first submodule extracts the region of interest of the parking lot image to obtain a feature map of the region of interest, the second submodule detects several corner points of the parking space number region in the feature map of the region of interest to obtain a parking space number region feature map, the third submodule rotates the parking space number region feature map to obtain a standard parking space number region feature map, the fourth submodule predicts the parking space number character sequence of the standard parking space number region feature map, and outputs the parking space number detection result corresponding to the parking lot image to be detected.

[0105] As a further improvement, the convolutional neural network model is the first network model, and the configuration unit 603 is specifically used for:

[0106] The acquisition unit acquires a first training image, where the first training image is annotated with coordinates of a plurality of corner points of a parking space number area and a parking space number character string category;

[0107] The first network is trained using the first training image to obtain a first network model.

[0108] Usually, the parking space number area is in the shape of a matrix, but considering that the parking lot image to be detected collected by the camera will have a certain deformation, the parking space number area in the parking lot image to be detected will also be deformed. If only the two corner points of the parking space number area are detected, an incomplete parking space number area may be extracted. Therefore, the embodiment of the present application preferably detects the four corner points of the parking space number area, so that the parking space number area can be accurately determined and a complete parking space number area feature map can be obtained.

[0109] In the embodiment of the present application, the parking space number detection device extracts the region of interest of the parking lot image to be detected through the first submodule in the first network model, and extracts the key areas in the parking lot image to be detected, which can reduce redundant feature representation and improve detection speed and detection accuracy; detects the four corner point positions of the parking space number area in the feature map of the region of interest through the second submodule to obtain a complete parking space number area feature map; rotates the parking space number area feature map through the third submodule to solve the problem of tilt of the parking space number area and improve the parking space number detection accuracy; extracts and fuses features of the standard parking space number area feature map through the fourth submodule, predicts the parking space number character sequence based on the fused features, improves feature representation through feature fusion, and then improves detection accuracy, thereby improving the existing parking space number detection method that uses traditional image processing methods to detect parking space numbers, and has the technical problems of slow detection speed and low accuracy.

[0110] The above is Example 2 of a parking space number detection device provided by the present application. The following is Example 3 of a parking space number detection device provided by the present application. Example 3 is a specific description of when the convolutional neural network model in Example 1 is the second network model.

[0111] Device embodiment three:

[0112] The present application provides a parking space number detection device, comprising:

[0113] An acquisition unit 601 is used to acquire an image of a parking lot to be detected;

[0114] The detection unit 602 is used to input the parking lot image to be detected into the convolutional neural network model to detect the parking space number, and obtain the parking space number detection result corresponding to the parking lot image to be detected.

[0115] As a further improvement, the convolutional neural network model is the second network model, and the detection unit 602 is specifically used for:

[0116] The parking lot image to be detected is input into the second network model, so that the second network model performs parking space number frame detection, character frame detection and character recognition on the parking lot image to be detected in parallel, and outputs the parking space number detection result corresponding to the parking lot image to be detected.

[0117] As a further improvement, the detection unit 602 is specifically used for:

[0118] The parking lot image to be detected is input into the second network model, so that the second network model performs feature extraction and multi-scale feature fusion on the parking lot image to be detected, and outputs a plurality of scale feature maps. The second network model performs parking space number frame detection, character frame detection and character recognition on each scale feature map in parallel, and outputs the parking space number frame detection result, character frame detection result and character recognition detection result corresponding to each scale feature map. The second network model obtains the parking space number detection result corresponding to the parking lot image to be detected based on the parking space number frame detection results, character frame detection results and character recognition detection results corresponding to all scale feature maps.

[0119] As a further improvement, the convolutional neural network model is the second network model, and the configuration unit 603 is specifically used for:

[0120] Obtain a second training image, where the second training image is annotated with the location of the parking space number, and the location and category of each character in the parking space number;

[0121] The second network is trained using the second training image to obtain a second network model.

[0122] In the embodiment of the present application, the parking space number detection device performs end-to-end parking space number detection on the parking lot image to be detected through the second network model, which is a one-stage network model, thereby improving the parking space number detection speed; the extracted feature map is multi-scale fused by the FPN in the second network model, and the obtained multi-scale feature map is multi-scale predicted, which can solve the problem of different character sizes, thereby improving the parking space number detection accuracy, and improving the existing parking space number detection method. The traditional image processing method is used for parking space number detection, which has the technical problems of slow detection speed and low accuracy.

[0123] The embodiment of the present application also provides a parking space number detection device, the device comprising a processor and a memory;

[0124] The memory is used to store the program code and transmit the program code to the processor;

[0125] The processor is used to execute the parking space number detection method in the aforementioned method embodiment according to the instructions in the program code.

[0126] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0127] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0128] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0129] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0130] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for executing all or part of the steps of the method described in each embodiment of the present application through a computer device (which can be a personal computer, a server, or a network device, etc.). The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (full name in English: Read-Only Memory, English abbreviation: ROM), random access memory (full name in English: Random Access Memory, English abbreviation: RAM), disk or optical disk and other media that can store program codes.

[0131] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A parking space number detection method, characterized in that: include: Obtain the parking lot image to be detected; Input the parking lot image to be detected into a convolutional neural network model to detect the parking space number, and obtain the parking space number detection result corresponding to the parking lot image to be detected, wherein the convolutional neural network model includes a first network model and a second network model; the second network model includes a backbone network, and the backbone network is used to extract features from the input image, and the backbone network is a residual network or a visual geometry group network; a feature pyramid network is embedded in the second network model, and the feature pyramid network is used to perform multi-scale feature fusion on the features extracted by the second network model, and output a plurality of scale feature maps; When the training data for training the model needs less annotation information and the problem of tilted parking space number recognition needs to be solved, the first network model is selected; when the detection speed needs to be fast and the problem of different parking space number character sizes needs to be solved, the second network model is selected; When the convolutional neural network model is the first network model, the parking lot image to be detected is input into the first network model for parking space number detection, and a parking space number detection result corresponding to the parking lot image to be detected is obtained, including: Input the parking lot image to be detected into the first network model, so that the first network model sequentially performs region of interest extraction, parking space number area corner point detection, parking space number area rotation and parking space number character sequence prediction on the parking lot image to be detected, and outputs the parking space number detection result corresponding to the parking lot image to be detected; When the convolutional neural network model is the second network model, the parking lot image to be detected is input into the second network model for parking space number detection, and a parking space number detection result corresponding to the parking lot image to be detected is obtained, including: Input the parking lot image to be detected into the second network model, so that the second network model performs parking space number frame detection, character frame detection and character recognition on the parking lot image to be detected in parallel, automatically generates a parking space number string and the position, size and category of a single character according to the parking space number frame detection result, the character frame detection result and the character recognition result, and outputs the parking space number detection result corresponding to the parking lot image to be detected; The second network model performs parking space number frame detection, character frame detection and character recognition on the parking lot image to be detected in parallel, and outputs a parking space number detection result corresponding to the parking lot image to be detected, including: The second network model performs feature extraction and multi-scale feature fusion on the parking lot image to be detected, and outputs a plurality of scale feature maps; The second network model performs parking space number frame detection, character frame detection and character recognition on each of the scale feature maps in parallel, and outputs parking space number frame detection results, character frame detection results and character recognition detection results corresponding to each of the scale feature maps; The second network model obtains the parking space number detection result corresponding to the parking lot image to be detected based on the parking space number frame detection results, character frame detection results and character recognition detection results corresponding to all the scale feature maps.

2. The parking space number detection method according to claim 1, characterized in that: The first network model includes a first submodule, a second submodule, a third submodule and a fourth submodule connected in sequence; The first network model sequentially extracts the region of interest, detects corner points in the parking space number region, rotates the parking space number region, and predicts the parking space number character sequence for the parking lot image to be detected, and outputs the parking space number detection result corresponding to the parking lot image to be detected, including: The first submodule extracts the region of interest from the parking lot image to be detected to obtain a feature map of the region of interest; The second submodule detects the positions of several corner points of the parking space number area in the feature map of the region of interest to obtain a feature map of the parking space number area; The third submodule rotates the parking space number area characteristic map to obtain a standard parking space number area characteristic map; The fourth submodule predicts a parking space number character sequence for the standard parking space number area feature map, and outputs a parking space number detection result corresponding to the parking lot image to be detected.

3. The parking space number detection method according to claim 2, characterized in that: The fourth submodule predicts the parking space number character sequence for the standard parking space number area feature map and outputs the parking space number detection result corresponding to the parking lot image to be detected, including: The fourth submodule performs feature extraction and feature fusion on the standard parking space number area feature map, predicts the parking space number character sequence based on the fused features, and outputs the parking space number detection result corresponding to the parking lot image to be detected.

4. The parking space number detection method according to claim 1, characterized in that: The configuration process of the first network model is: Acquire a first training image, where the first training image is annotated with coordinates of a plurality of corner points of a parking space number area and a parking space number character string category; The first network is trained using the first training image to obtain the first network model.

5. The parking space number detection method according to claim 1, characterized in that: The configuration process of the second network model is: Acquire a second training image, wherein the second training image is annotated with the position of the parking space number and the position and category of each character in the parking space number; The second network is trained using the second training image to obtain the second network model.

6. A parking space number detection device, characterized in that: include: An acquisition unit, used for acquiring an image of a parking lot to be detected; A detection unit is used to input the parking lot image to be detected into a convolutional neural network model for parking space number detection, and obtain a parking space number detection result corresponding to the parking lot image to be detected, wherein the convolutional neural network model includes a first network model and a second network model; the second network model includes a backbone network, and the backbone network is used to extract features from the input image, and the backbone network is a residual network or a visual geometry group network; a feature pyramid network is embedded in the second network model, and the feature pyramid network is used to perform multi-scale feature fusion on the features extracted by the second network model, and output a plurality of scale feature maps; When the training data for training the model needs less annotation information and the problem of tilted parking space number recognition needs to be solved, the first network model is selected; when the detection speed needs to be fast and the problem of different parking space number character sizes needs to be solved, the second network model is selected; When the convolutional neural network model is a first network model, the detection unit is specifically used to input the parking lot image to be detected into the first network model, so that the first network model sequentially performs region of interest extraction, parking space number area corner point detection, parking space number area rotation and parking space number character sequence prediction on the parking lot image to be detected, and outputs a parking space number detection result corresponding to the parking lot image to be detected; When the convolutional neural network model is the second network model, the detection unit is specifically used for: Input the parking lot image to be detected into the second network model, so that the second network model performs parking space number frame detection, character frame detection and character recognition on the parking lot image to be detected in parallel, automatically generates a parking space number string and the position, size and category of a single character according to the parking space number frame detection result, the character frame detection result and the character recognition result, and outputs the parking space number detection result corresponding to the parking lot image to be detected; The second network model performs parking space number frame detection, character frame detection and character recognition on the parking lot image to be detected in parallel, and outputs a parking space number detection result corresponding to the parking lot image to be detected, including: The second network model performs feature extraction and multi-scale feature fusion on the parking lot image to be detected, and outputs a plurality of scale feature maps; The second network model performs parking space number frame detection, character frame detection and character recognition on each of the scale feature maps in parallel, and outputs parking space number frame detection results, character frame detection results and character recognition detection results corresponding to each of the scale feature maps; The second network model obtains the parking space number detection result corresponding to the parking lot image to be detected based on the parking space number frame detection results, character frame detection results and character recognition detection results corresponding to all the scale feature maps.

7. A parking space number detection device, characterized in that: The device comprises a processor and a memory; The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the parking space number detection method according to any one of claims 1-5 according to the instructions in the program code.

Images