Image screening method, device, equipment and storage medium

Abstract

This application proposes an image screening method, apparatus, device, and storage medium. The method may include: acquiring an input image; extracting key points of a target object in the input image; obtaining information about the key points; the key point information including at least one of the following: the position of the key points in the input image, the side length of the image region enclosed by the key points, and the confidence level of the key points; and screening the input image based on the key point information. This improves image screening efficiency.

Description

Technical Field

[0001] The present application relates to computer technology, and in particular to an image screening method, apparatus, device and storage medium. Background Art

[0002] In some scenarios, it's necessary to obtain an image containing a target object from a captured video stream in order to perform subsequent operations on the target object in the image. The target object can be set as needed. For example, in a license plate recognition scenario, the target object can be a license plate. In another example, in a face recognition scenario, the target object can be a face.

[0003] However, in real-world situations, the quality of captured images varies greatly due to factors such as the camera's shooting angle, the distance of the target object from the camera, and whether the target object is in motion. Therefore, it is necessary to filter out clear and complete images of the target object from the captured images.

[0004] In related technologies, when performing image screening, target objects in the input image are first detected using an object detection algorithm. Feature comparison or deep learning analysis of the target object's features is then performed to determine whether the input image is clear and complete. If the input image is clear and complete, the image is retained. However, this related technology, due to the computationally intensive object detection required, results in low image screening efficiency and a lack of lightweightness. Summary of the Invention

[0005] In view of this, the present application discloses at least one image screening method. The method may include: obtaining an input image; extracting key points of a target object in the input image; obtaining information about the key points; the key point information including at least one of the following: position information of the key points in the input image, side length information of the image region enclosed by the key points, and confidence information of the key points; and screening the input image based on the key point information.

[0006] In some embodiments, extracting the key points of the target object in the input image includes: processing the input image using a target object key point model to obtain a heat map corresponding to the input image; and determining the pixel point corresponding to the maximum pixel value in the heat map as the key point of the target object.

[0007] In some embodiments, the input image is processed using a target object key point model to obtain a heat map corresponding to the input image, including: using the target object key point model to perform multiple downsampling and multiple upsampling on the input image to obtain the heat map; wherein, in the multiple upsampling processes, the multi-level feature maps generated in the multiple downsampling processes are fused.

[0008] In some embodiments, obtaining the information of the key point includes: determining the position information and / or the side length information based on the coordinates of the key point in the input image; and determining the confidence information based on the pixel value of the key point in the heat map.

[0009] In some embodiments, the key point information includes confidence information of the key point, and the screening of the input image based on the key point information includes: determining a confidence mean according to the confidence of each key point; and screening the input image according to a comparison result of the confidence mean with a preset confidence threshold.

[0010] In some embodiments, the key point information includes side length information of the image area enclosed by the key point, the image area is a quadrilateral, and the side length information includes first side length information and second side length information of the quadrilateral; the filtering of the input image based on the key point information includes: filtering the input image according to the comparison result of the first side length indicated by the first side length information and the preset first side length threshold, and the comparison result of the second side length indicated by the second side length information and the preset second side length threshold.

[0011] In some embodiments, the key point information includes position information of the key point in the input image, and the filtering of the input image based on the key point information includes: determining the distance from the position of the key point in the input image to the image edge of the input image based on the position information; and filtering the input image based on a comparison result of the distance with a preset distance threshold.

[0012] In some embodiments, before screening the input image, the method includes: determining a quality score of the input image based on a comparison result, wherein the comparison result includes at least one of the following: a comparison result of the confidence mean with a preset confidence threshold; a comparison result of the first side length with a preset first side length threshold; a comparison result of the second side length with a preset second side length threshold; a comparison result of the distance with a preset distance threshold; screening the input image includes: screening the input image based on the quality score of the input image.

[0013] In some embodiments, the comparison result further includes: a comparison result of the side length of the input image and a preset image side length threshold.

[0014] In some embodiments, determining the quality score of the input image based on the comparison result includes: determining the quality score of the input image based on the confidence mean and the area of ​​the input image when the comparison result indicates one of the following situations: the side length of the input image reaches the image side length threshold; the confidence mean reaches the confidence threshold; the first side length reaches the first side length threshold; the second side length reaches the second side length threshold; the distance does not reach the distance threshold.

[0015] In some embodiments, extracting the key points of the target object in the input image includes: extracting the key points of the target object in the input image using a target object key point model generated based on a convolutional neural network; wherein the training method of the target object key point model includes: obtaining a sample image; the sample image includes annotation information for the key points of the target object; the annotation information is used to indicate the coordinate information of the key points of the target object in the sample image; using the sample image to perform multiple rounds of training on the target object key point model, wherein each round of training includes: inputting the sample image into the target object key point model to obtain prediction information for the key points; determining the loss information between the annotation information and the prediction information; and adjusting the parameters of the target object key point model based on the loss information.

[0016] The present application also proposes an image screening device, which includes: a first acquisition module for acquiring an input image; an extraction module for extracting key points of a target object in the input image; a second acquisition module for acquiring information of the key points; the information of the key points includes at least one of the following: position information of the key points in the input image, side length information of the image area enclosed by the key points, and confidence information of the key points; and a screening module for screening the input image based on the information of the key points.

[0017] The present application also proposes an electronic device, comprising: a processor; and a memory for storing processor-executable instructions; wherein the processor implements the image screening method shown in any of the aforementioned embodiments by running the executable instructions.

[0018] The present application also provides a computer-readable storage medium, wherein the storage medium stores a computer program, and the computer program is used to enable a processor to execute the image screening method shown in any of the aforementioned embodiments.

[0019] In the aforementioned scheme, information about the key points of the target object can be obtained. The key point information includes at least one of the following information that can be used to characterize whether the target object is clear and complete: the position of the key point in the input image, the side length of the image area enclosed by the key point, and the confidence level of the key point. The input image is then filtered based on the key point information, and a clear and complete image can be obtained by filtering out the key points that can characterize whether the target object is clear and complete. Compared with related technologies, this method does not rely on extremely time-consuming target detection algorithms and deep learning algorithms, greatly reducing the amount of computation required during image screening, improving image screening efficiency, and achieving lightweight results.

[0020] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in one or more embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the description of the embodiments or related technologies. Obviously, the drawings described below are only some embodiments recorded in one or more embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0022] Figure 1 A schematic diagram of a method flow of an image screening method according to an embodiment of the present application;

[0023] Figure 2 A schematic diagram of a method flow for training a target object key point model according to an embodiment of the present application;

[0024] Figure 3 A schematic diagram of a flow chart of an image screening method based on image quality scores according to an embodiment of the present application;

[0025] Figure 4 This is a structural diagram of a license plate key point model shown in an embodiment of the present application;

[0026] Figure 5 A schematic diagram of a flow chart of an image screening method according to an embodiment of the present application;

[0027] Figure 6 This is a schematic structural diagram of an image screening device according to an embodiment of the present application;

[0028] Figure 7 A schematic diagram of the hardware structure of an electronic device shown in an embodiment of the present application. DETAILED DESCRIPTION

[0029] The following exemplary embodiments are described in detail, with examples illustrated in the accompanying drawings. When the following description refers to the drawings, identical numbers in different figures represent identical or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with the present application. Instead, they are merely examples of devices and methods consistent with certain aspects of the present application, as detailed in the appended claims.

[0030] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms "a", "the" and "the" used in this application and the appended claims are also intended to include plural forms, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items. It should also be understood that the word "if" used herein, depending on the context, can be interpreted as "at the time of" or "when" or "in response to determining".

[0031] This application aims to provide an image screening method. The method can obtain an input image; extract key points of a target object in the input image; obtain information about the key points; the key point information includes at least one of the following: position information of the key points in the input image, side length information of the image region enclosed by the key points, and confidence information of the key points; and screen the input image based on the key point information.

[0032] In the aforementioned scheme, information about the key points of the target object can be obtained. The key point information includes at least one of the following information that can be used to characterize whether the target object is clear and complete: the position of the key point in the input image, the side length of the image area enclosed by the key point, and the confidence level of the key point. The input image is then filtered based on the key point information, and a clear and complete image can be obtained by filtering out the key points that can characterize whether the target object is clear and complete. Compared with related technologies, this method does not rely on extremely time-consuming target detection algorithms and deep learning algorithms, greatly reducing the amount of computation required during image screening, improving image screening efficiency, and achieving lightweight results.

[0033] See Figure 1 , Figure 1 A schematic diagram of a method flow of an image screening method shown in an embodiment of the present application. Figure 1The image screening method can be applied to an electronic device. The electronic device can execute the screening method by being equipped with software logic corresponding to the screening method. The electronic device can be a laptop, a computer, a server, a mobile phone, a PDA (Personal Digital Assistant), etc. The type of the electronic device is not particularly limited in this application. The electronic device can also be a client device or a server device, which is not particularly limited here.

[0034] like Figure 1 As shown, the method may include S102 to S108. Unless otherwise specified, the present application does not particularly limit the execution order of these steps.

[0035] Among them, S102, obtains the input image.

[0036] The input image can be an image from a video stream captured by any image acquisition device, or an image obtained by preprocessing the image from the captured video stream. This application does not limit the type of image acquisition device. These image acquisition devices can be deployed in a field environment to capture the target object and generate an input image containing the target object.

[0037] The target object can be specified based on the business scenario requirements. For example, in a license plate detection scenario, the target object can be a license plate. In a pedestrian detection scenario, the target object can be a human body. In a face detection scenario, the target object can be a face.

[0038] After capturing a video stream of a target object, in some embodiments, each frame in the video stream can be used sequentially as the input image to filter images in the video stream. For example, in a license plate detection scenario, the image capture device is a camera deployed on a highway. The camera can capture a video stream of a vehicle's license plate. In this example, the images in the video stream can be used sequentially as the input image.

[0039] In some embodiments, the target object is an object attached to another attached object. For example, in a license plate detection scenario, the target object is a license plate, and the attached object is a vehicle. For another example, in a bus route detection scenario, the target object is a bus route sign, and the attached object is a bus.

[0040] In this example, the attached object in the video stream can be preprocessed first to obtain an image sequence set for the attached object, and then the images in the image sequence set can be used as the input images in sequence to screen the images in the image sequence set. For example, in a license plate detection scenario, the image acquisition device is a camera deployed on a highway. The camera can capture a video stream for the license plate of the vehicle. In this application, the images in the video stream can be preprocessed first. During the preprocessing, a target detection algorithm can be used to detect each vehicle that appears in the captured image to obtain an image set for each vehicle. Each image in the image set can then be used as the input image. Therefore, when multiple vehicles appear in the video stream, image screening can be performed for the license plate of each vehicle.

[0041] S104: Extract key points of the target object in the input image.

[0042] The key points may be points on the contour of the target object. For example, in a license plate detection scenario, the key points may be four key points on the contour of the license plate.

[0043] In some embodiments, in S104 , a target object key point model may be used to extract the key points.

[0044] The target object key point model can be trained by using image samples with key point information of the target object labeled. After training, the target object key point model can be used to extract key points in the input image.

[0045] For example, in a license plate detection scenario, the target object is the license plate, and the key points can be the four key points of the license plate. In this example, a target object key point model can be pre-trained using a set of image samples labeled with the four key points of the license plate. After training, the model can be used to extract key points from the input image to obtain key point information.

[0046] The following introduces a training method for the key point model of the target object.

[0047] The target object key point model can be generated based on a convolutional neural network.

[0048] The target object key point model may include a downsampling unit and an upsampling unit. The downsampling unit may be used to reduce the input image and enhance features to obtain a feature map. The upsampling unit may be used to amplify the feature map to obtain a heat map. Downsampling and upsampling are mutually inverse processes, and the downsampling and upsampling processes usually have several pairs of sampling feature maps with the same resolution. For example, downsampling can scale a 64*64 image three times to obtain sampling feature maps with three resolutions of 32*32, 16*16, and 8*8, and upsampling can amplify an 8*8 image three times to obtain sampling feature maps with three resolutions of 16*16, 32*32, and 64*64. Among them, sampling feature maps of the same resolution are a pair of corresponding sampling feature maps.

[0049] In some embodiments, in order to limit the computational complexity of the target object key point model, the size of the image input to the model may be limited to a preset size, for example, 64*64.

[0050] See Figure 2 , Figure 2 This is a flow chart of a method for training a target object key point model according to an embodiment of the present application. Figure 2 As shown, S202-S204 may be executed. S204 includes S2042-S2046. Unless otherwise specified, this application does not specifically limit the execution order of these steps.

[0051] S202: Acquire a sample image.

[0052] In this step, a preset number of sample images, such as tens of thousands of sample images, may be obtained.

[0053] The sample image includes annotation information for key points of the target object; the annotation information is used to indicate coordinate information of the key points of the target object in the sample image.

[0054] The generation process of the sample image is as follows:

[0055] You can first obtain an image containing the target object.

[0056] Then, a heat map of the same size as the image is obtained. Initially, the pixel value of each pixel in the heat map is 0. Then, according to the coordinates of the key points of the target object, the pixel values ​​of the pixels at the corresponding coordinates in the heat map can be modified to 1 to obtain the annotated heat map (annotation information).

[0057] Then, a sample image is generated based on the annotated heat map and the image.

[0058] In some embodiments, it is necessary to extract multiple key points of the target object. When generating a sample image, a labeled heat map consistent with the number of key points can be maintained. Each labeled heat map corresponds to only one key point. Therefore, a key point can be determined based on each heat map. Compared with the method of predicting multiple key points using one heat map, the influence between multiple key points is reduced and the detection accuracy of the key points is improved.

[0059] S204: Perform multiple rounds of training on the target object key point model using the sample images.

[0060] Each round of training includes:

[0061] S2042: Input the sample image into the target object key point model to obtain prediction information for the key points.

[0062] In this step, the downsampling unit and the upsampling unit included in the target object key point model may be used to extract features from the sample image to obtain prediction information for the key points.

[0063] The prediction information can be a prediction heatmap of the same size as the annotation heatmap (annotation information). For example, in a license plate detection scenario, if the annotation information is a 4-channel 16*16 annotation heatmap, the prediction information is also a 4-channel 16*16 heatmap. This is necessary to perform key point extraction.

[0064] S2044: Determine loss information between the annotation information and the prediction information.

[0065] In this step, China may use a preset loss function to determine the loss information.

[0066] The preset loss function may be a loss function specified according to business requirements. This application does not limit the type of the preset loss function. For example, a cross entropy loss function may be used in this example.

[0067] The annotation information and the prediction information are input into the cross entropy loss function for calculation, so as to obtain loss information that can indicate the error between the annotation and the predicted key points.

[0068] S2046: Adjust parameters of the target object key point model based on the loss information.

[0069] In this step, the loss information can be used to calculate the descent gradient, and back propagation can be used to update the parameters of the key point model of the loss target object.

[0070] Through S202-S204, the target object key point model is trained for multiple rounds using a large number of sample images, so that the model can be equipped with the ability to extract key points, so that the model can be used to execute S104 to extract key points for the target object in the input image.

[0071] In some embodiments, in the process of extracting the key points using the target object key point model, the input image can be processed using the target object key point model to obtain a heat map corresponding to the input image; and then the pixel point corresponding to the maximum pixel value in the heat map is determined as the key point of the target object.

[0072] The heat map can characterize the possibility that each pixel is a key point through pixel values. In some embodiments, the input image is subjected to multiple downsampling and multiple upsampling by using the target object key point model to obtain the heat map. For example, the target object key point model can be used to perform some downsampling processing such as convolution and pooling on the input image to obtain a feature map, and then the feature map is subjected to some upsampling processing such as deconvolution to obtain the heat map. In some embodiments, during the multiple upsampling processes, the multi-level feature maps generated during the multiple downsampling processes can be fused. In this way, more features can be fused in the feature extraction (upsampling) process, the feature information contained in the heat map can be enhanced, the accuracy of key point prediction can be improved, and the image screening effect can be improved. This application does not limit the specific method of fusion. For example, fusion methods such as feature superposition, convolution, and splicing can be used.

[0073] The multi-level feature map refers to feature maps with different resolutions generated during the downsampling process. Different resolutions contain different feature quantities and feature levels. High-resolution feature maps contain low-level features, while low-resolution feature maps contain high-level features.

[0074] For example, in a license plate detection scenario, a 64*64 image can be scaled three times by the downsampling unit to obtain sampling feature maps with three resolutions: 32*32, 16*16, and 8*8. An 8*8 image can be scaled three times by the upsampling unit to obtain sampling feature maps with three resolutions: 16*16, 32*32, and 64*64. The 32*32 and 16*16 sampling feature maps obtained during the downsampling process can then be fused with the 16*16 and 32*32 sampling feature maps obtained during the upsampling process. This allows the resulting heat map to contain more features, enhance feature information, improve key point prediction accuracy, and thus enhance image screening effectiveness.

[0075] After obtaining the heat map, the pixel point corresponding to the maximum pixel value in the heat map may be selected as the key point of the target object.

[0076] It should be noted that if multiple key points need to be extracted in S104, the input image can be processed by the target object key point model to obtain the same number of heat maps as the number of key points, and then the pixel points corresponding to the maximum pixel value in each heat map are selected as key points to obtain the multiple key points.

[0077] For example, in a license plate detection scenario, the four key points of the license plate need to be detected. The input image is processed using the target object key point model to obtain four heat maps. The pixel corresponding to the maximum pixel value in each heat map is then a key point of the license plate.

[0078] S106: Acquire information of the key points.

[0079] The information of these key points may include at least one of the following: position information of the key points in the input image, side length information of the image area enclosed by the key points, and confidence information of the key points.

[0080] The information of these key points can be used to characterize whether the target object is clear and complete. For example, if the position information of the key points in the input image determines that the target object is in the middle of the input image, it can be said that the target object is complete. Conversely, if the target object is determined to be at the edge of the input image, it can be said that the target object may be incomplete. For another example, if the side length information of the image area enclosed by the key points determines that the side length of the image area is greater than a preset side length threshold, it can be determined that the target object in the image area is clear. Conversely, if the side length of the image area is determined to be less than the side length threshold, it can be determined that the target object in the image area is unclear. For another example, if the confidence information of the key points determines that the confidence of the key points is higher than a preset confidence threshold, it can be determined that the target object is clear. Conversely, if the confidence of the key points is lower than the preset confidence threshold, it can be determined that the target object is unclear.

[0081] In some embodiments, in S106, the position information and / or the side length information can be determined based on the coordinates of the key point in the input image; and the confidence information can be determined based on the pixel value of the key point in the heat map.

[0082] After the key points are obtained in S104 , the coordinates of the key points in the input image may be determined according to the coordinates of the key points in the heat map.

[0083] The coordinates themselves can represent the position information of the key point in the input image.

[0084] In some embodiments, S104 obtains a plurality of key points, which may enclose an image region in the input image. The image region contains the target object.

[0085] After determining the coordinates of multiple key points in the input image, the length of a side of the image region can be determined by the position coordinates of two adjacent key points. Multiple side lengths of the image region can be determined by multiple pairs of adjacent key points.

[0086] For example, in a license plate detection scenario, four key points are detected on the license plate, and these four key points can enclose an image area that happens to be the license plate area. Three key points are selected from the four key points, and the lengths of the two sides of the image area, namely the width and height, are obtained using distance determination methods such as cosine distance, Euclidean distance, and Mahalanobis distance.

[0087] In S106, the confidence information may also be determined according to the pixel value of the key point in the heat map.

[0088] The confidence information can indicate the probability that the extracted key point is predicted as a key point. The higher the confidence, the higher the probability that the extracted key point is a key point of the target object. That is, the confidence information of the key point can indicate the credibility of the target object.

[0089] In some embodiments, the pixel values ​​of the key points in the heat map may be normalized (eg, processed by a sigmoid function) to obtain the confidence information.

[0090] S108: Filter the input image based on the information of the key points.

[0091] The following describes three situations in which the key point information is the position information of the key point in the input image, the side length information of the image area enclosed by the key point, and the confidence information of the key point.

[0092] In the first scenario, the key point information includes confidence information of the key point.

[0093] In the first case, in S108 , a confidence mean may be determined according to the confidence of each key point.

[0094] In some methods, S104 may extract multiple key points, and the confidence information of each key point may be obtained in S106. The confidence mean may then be obtained by averaging or weighted averaging.

[0095] After the confidence mean is obtained, the input image may be screened according to a comparison result between the confidence mean and a preset confidence threshold.

[0096] The confidence threshold is an empirical threshold. If the average confidence value of multiple key points reaches this confidence threshold, it can be confirmed that the extracted key points are indeed key points of the target object with high confidence, that is, it can be confirmed that the input image contains the target object with high confidence. The term "reach" as used in this application includes at least two situations: greater than and greater than or equal to. For example, A reaching B includes at least two situations: A>B and A≥B.

[0097] In some embodiments, in response to the confidence mean reaching the confidence threshold, it can be determined that the confidence level of the input image containing the target object is relatively high, that is, the possibility that the target object is clear and complete is relatively high, and the input image is retained.

[0098] In response to the confidence mean not reaching the confidence threshold, it is determined that the confidence of the input image containing the target object is relatively low, that is, the possibility that the target object is not clear and complete is relatively high, and the input image is deleted.

[0099] Therefore, in the first case, input images containing clear and complete target objects can be screened out based on the key point confidence.

[0100] For example, in the license plate detection scenario, the confidence average of the four license plate key points obtained in S106 is 0.3. The confidence threshold is 0.6. Since the confidence average of 0.3 is less than 0.6, it can be concluded that the extracted key points are not license plate key points, and it can also be concluded that the input image may contain a clear and complete target object, so the image is deleted.

[0101] In the second scenario, the key point information includes side length information of the image area enclosed by the key point. Assuming that the image area is a quadrilateral, the side length information includes first side length information and second side length information of the quadrilateral.

[0102] In the second case, in S108, the input image can be filtered based on the comparison result of the first side length indicated by the first side length information and the preset first side length threshold, and the comparison result of the second side length indicated by the second side length information and the preset second side length threshold.

[0103] In some embodiments, S104 may extract four key points, and S106 may be used to obtain two side length information of the quadrilateral, namely, first side length information and second side length information. The first side length information may indicate the first side length, and the second side length information may indicate the second side length.

[0104] The first and second side length thresholds are empirical thresholds. If the first side length reaches the first side length threshold and the second side length reaches the second side length threshold, it can be determined that the size of the image region meets the requirements, that is, the clarity of the target object within the image region meets the requirements, and that the input image contains a clear and complete target image, and the input image can be retained.

[0105] If the first side length does not reach the first side length threshold, or the second side length does not reach the second side length threshold, it may indicate that the size of the image area does not meet the requirements, that is, the clarity of the target object contained in the image area does not meet the standards, which means that the input image may not contain a clear and complete target image, and the input image can be deleted.

[0106] Therefore, in the second scenario, the input image containing a clear and complete target object can be screened out based on the side length information of the image area enclosed by the key points.

[0107] For example, in a license plate detection scenario, the coordinate information of four key points of the license plate is obtained through S106. Three of these key points are selected and, using the Euclidean distance method, the width (first side length) and height (second side length) of the license plate are determined. The first side length threshold is 13, and the second side length threshold is 4. If the width of the license plate reaches 13 and the height reaches 4, then the license plate contained in the input image meets the required clarity and can be retained, thus completing the image screening.

[0108] In a third scenario, the key point information includes position information of the key point in the input image.

[0109] In the third case, in S108 , the distance from the position of the key point in the input image to the image edge of the input image may be determined according to the position information.

[0110] In some embodiments, multiple key points may be obtained through S104, and position information of each key point may be obtained through S106. The position information indicates the position of the key point in the input image.

[0111] According to the position information of each key point, the distance between each key point and the image edge of the input image can be obtained.

[0112] After the distance is obtained, the input image may be screened according to a comparison result between the distance and a preset distance threshold.

[0113] The distance threshold is an empirical threshold. If the distance from each of the multiple key points to the image edge reaches the distance threshold, it can be said that the key point is far from the image edge, that is, the input image contains a clear and complete target object, and the input image can be retained.

[0114] If the distance from any key point among the multiple key points to the image edge does not reach the distance threshold, it can be said that the key point is close to the image edge, that is, the input image may not contain a clear and complete target object, and the input image can be deleted.

[0115] Therefore, in the third situation, the input image containing a clear and complete target object can be screened out based on the position information of the key point in the input image.

[0116] In some embodiments, the three situations can cooperate with each other to complete image selection. For example, if it is determined that the input image can be retained in at least one of the situations, the input image will be retained. If it is determined that the input image can be deleted in all three situations, the input image will be deleted.

[0117] According to the scheme described in S102-S108, information about the key points of the target object can be obtained. The key point information includes at least one of the following information that can be used to characterize whether the target object is clear and complete: the position information of the key point in the input image, the side length information of the image area enclosed by the key point, and the confidence information of the key point. The input image is then filtered based on the key point information, and a clear and complete image can be filtered out using this information about the key points that can characterize whether the target object is clear and complete. Compared with related technologies, this method does not rely on extremely time-consuming target detection algorithms and deep learning algorithms, greatly reducing the amount of computation required during the image screening process, improving image screening efficiency, and achieving the goal of lightweighting.

[0118] In some embodiments, before executing S104, the method can also use the side length of the input image to perform preliminary screening of the input image, filter out input images that do not meet the preset image side length threshold, and retain input images that meet the image side length threshold, thereby reducing the number of input images processed in S104-S108 and improving image screening efficiency.

[0119] In some methods, side length information of the input image may be obtained first, where the side length information may indicate the side length of the input image. In general, the input image is a rectangular image, that is, the side length of the input image may include width and height.

[0120] Then, based on the comparison result of the side length of the input image and the preset image side length threshold, it can be determined whether the size of the input image meets the standard, and if it meets the standard, the steps S104-S108 are continued; if it does not meet the standard, the input image is deleted.

[0121] The image side length threshold is an empirical threshold, which can generally include a first preset threshold corresponding to the height and a second preset threshold corresponding to the width.

[0122] If the height of the input image reaches the first preset threshold and the width reaches the second preset threshold, it can be determined that the size of the input image meets the standard, that is, it can be said that the clarity of the input image meets the standard, and steps S104-S108 can be continued.

[0123] If the height of the input image does not reach the first preset threshold and / or the width does not reach the second preset threshold, it can be determined that the size of the input image does not meet the requirements, that is, the clarity of the input image does not meet the requirements, and the input image can be deleted. This can reduce the number of input images processed in S104-S108 and improve image screening efficiency.

[0124] In some embodiments, the image quality of the input image can be quantified, which, on the one hand, facilitates intuitive understanding of the image quality of the input image; on the other hand, the image quality score can be used to more conveniently perform image screening.

[0125] Before screening the input image in S108,

[0126] See,3, Figure 3 A flowchart of an image screening method based on image quality score is shown in an embodiment of the present application. Figure 3 The illustrated method is a supplementary explanation of S108. Figure 3 As shown, the method may include S302 to S304. Unless otherwise specified, the present application does not limit the execution order of these steps.

[0127] S302: Determine a quality score of the input image according to the comparison result.

[0128] The comparison result includes at least one of the following: the comparison result of the confidence mean and the preset confidence threshold; the comparison result of the first side length and the preset first side length threshold; the comparison result of the second side length and the preset second side length threshold; the comparison result of the distance and the preset distance threshold; the comparison result of the side length of the input image and the preset image side length threshold.

[0129] In some embodiments, when the comparison result indicates one of the following situations, the quality score of the input image is determined according to the confidence mean and the area of ​​the input image:

[0130] The side length of the input image reaches the image side length threshold;

[0131] The confidence mean reaches the confidence threshold;

[0132] The first side length reaches the first side length threshold;

[0133] The second side length reaches the second side length threshold;

[0134] The distance does not reach the distance threshold.

[0135] In some embodiments, a preset formula for determining the image quality score can be pre-set. This preset formula may include a first parameter and a second parameter, and the goal of this preset formula is to ensure a positive correlation between the image quality score and the first and second parameters. That is, the larger the values ​​corresponding to the first and / or second parameters, the larger the calculated image quality score. The preset formula only needs to satisfy this positive correlation. This application does not specifically limit the form of the preset formula.

[0136] In some instances, the preset formula may include score = min(a × b ÷ c, d). Here, score indicates the image quality score of the input image. a indicates the first parameter; b is the second parameter; and c and d are preset constants. In some embodiments, to reduce computational complexity, the image quality score may be normalized. In a normalized scenario, c may be 1,000,000, and d may be 1.

[0137] In this step, the confidence mean value may be used as the first parameter, and the area of ​​the input image may be used as the second parameter, and the quality score of the input image may be obtained using the preset formula.

[0138] In the following situations, it can be determined that the input image does not meet the requirements, that is, the input image may not contain a clear and complete target object, and a lower quality score can be assigned to this type of input image that does not meet the requirements:

[0139] The side length of the input image does not reach the image side length threshold;

[0140] The confidence mean does not reach the confidence threshold;

[0141] The first side length does not reach the first side length threshold;

[0142] The second side length does not reach the second side length threshold;

[0143] The distance reaches the distance threshold.

[0144] In some embodiments, a in the preset formula may be set to a relatively small preset value (eg, 0.1, 0.01, etc.), thereby assigning a relatively low quality score to input images that do not meet the requirements.

[0145] After obtaining the quality score of the input image, in S304 , the input image may be screened based on the quality score of the input image.

[0146] For example, all input images can be sorted according to their image quality scores, and then the top N input images with the highest scores can be selected as the final input images to complete the image screening. N is a preset positive integer. When N is 1, the input image with the highest image quality score is selected.

[0147] Through S302-S304, on the one hand, it is easy to intuitively understand the image quality of the input image; on the other hand, the image quality score can be used to more conveniently screen images.

[0148] The following describes an example using a license plate detection scenario.

[0149] In this scenario, cameras can be installed on the highway. These cameras can capture video streams of passing vehicles. These cameras can pre-process the images in the captured video streams according to the vehicle, generating a vehicle-specific image sequence. These image sequences are then sent to a processing device for image filtering.

[0150] The processing device may include a license plate key point model. The model may be trained using the method shown in S202-S204 and then deployed. The model may extract key points of the four key point coordinates of the license plate.

[0151] See Figure 4 , Figure 4 This is a structural diagram of a license plate key point model shown in an embodiment of the present application. Figure 4 As shown, the license plate key point model 400 may include a downsampling unit 410 and an upsampling unit 420 .

[0152] The model takes a 64*64 image as input and outputs a 4-channel 64*64 heatmap. The downsampling unit 410 may include layers such as convolutional layers, pooling layers, residual network layers, and a self-attention mechanism. The downsampling unit is used to scale the 64*64 image three times to obtain sampled feature maps with three resolutions: 32*32, 16*16, and 8*8.

[0153] The upsampling unit 420 may include upsampling layers such as a deconvolution layer and an interpolation layer. It is used to amplify the 8*8 image three times to obtain sampling feature maps of three resolutions: 16*16, 32*32, and 64*64. The 32*32 and 16*16 sampling feature maps obtained during the downsampling process can be fused with the 16*16 and 32*32 sampling feature maps obtained during the upsampling process, so that the final heat map contains more features, enhances feature information, improves the accuracy of license plate key point prediction, and thus improves the image screening effect. The 64*64 sampling feature map obtained by upsampling is a 4-channel heat map. The license plate key point can be obtained according to the point corresponding to the maximum pixel value in the heat map of each channel.

[0154] In this example, the preset formula score=min(a×b÷1000000,1) can be used to calculate the image quality score of the input image.

[0155] See Figure 5 , Figure 5 The following is a flow chart of an image screening method according to an embodiment of the present application. Unless otherwise specified, the present application does not specifically limit the execution order of these steps.

[0156] like Figure 5 The processing device may execute S501-S511.

[0157] After receiving the image sequence for the vehicle, a frame of image may be selected from the image sequence as an input image, and S501 is executed to obtain the height and width of the input image.

[0158] S502 , determining whether the height of the input image reaches 222 (a first preset threshold value) and whether the width reaches 220 (a second preset threshold value).

[0159] If the height does not reach 222 and / or the width does not reach 220, S503 can be executed, and a in the aforementioned preset formula is set to 0.1 (the preset keypoint confidence level), and b is set to the area of ​​the input image (i.e., height * width). The image quality score of the input image is then calculated according to the preset formula. Since the value of 0.1 is relatively small, the image quality score of the input image is relatively low, which is equivalent to filtering out these input images.

[0160] If the height reaches 222 and the width reaches 220, it can be determined that the input image clarity meets the requirements. S504 can be executed to extract four license plate key points using the aforementioned license plate key point model and obtain key point information corresponding to the four license plate key points. The key point information includes the position of the key points in the input image, the side length of the image area enclosed by the key points, and the confidence level of the key points. The methods for extracting key points and obtaining key point information in this step can be referred to the aforementioned embodiments and will not be described in detail here.

[0161] Then, step S605 may be executed to calculate the mean confidence value of the four key points of the license plate, and the width and height of the license plate may be calculated using the coordinates of the four key points.

[0162] S506: Determine whether the confidence mean is greater than 0.8. This step can be referred to the aforementioned method for screening images using the confidence mean, and will not be described in detail here.

[0163] S507: Determine whether the width of the license plate reaches 13 (first side length threshold) and whether the width reaches 4 (second side length threshold). This step can refer to the aforementioned method of filtering images using the side length information of the image area enclosed by the key points, and will not be described in detail here.

[0164] S508: Determine whether the distance between the positions of the four key points of the license plate and the edge of the image reaches 3. This step can refer to the aforementioned method of filtering images using the position information of the key points in the input image, and will not be described in detail here.

[0165] If the judgment result of any of S506 - S508 is negative, it means that the license plate may not be clear and complete, and S503 may be executed to obtain a relatively low image quality score of the input image.

[0166] If the results of all three judgments in S506-S508 are yes, it means that the license plate is likely clear and complete. S509 can be executed. In the above preset formula, a is set to the average confidence value of the four key points, and b is set to the area of ​​the input image (i.e., height * width). Then, the real image quality score of the input image is calculated according to the preset formula. This image quality score can quantify the quality of the input image and facilitate image screening.

[0167] After determining the image quality score of the input image through S503 or S509 , S510 may be executed to determine whether there are any remaining images in the acquired image sequence.

[0168] If there are any remaining images, an input image can be selected from the remaining images and the process continues with S501-S510. If there are no remaining images in the image sequence, S511 can be executed to select the input image with the highest image quality score in the image sequence. This allows for a strategy to analyze license plate features based on key point information, eliminating the need for time-consuming license plate detection and deep learning algorithms. This significantly reduces the computational effort involved in image screening and improves image screening efficiency.

[0169] In some embodiments, the filtered input image can be used for license plate recognition. Since the clarity of the filtered input image, the size (clarity) of the license plate it contains, its position in the image (completeness), and the confidence (credibility) all meet the standards, the accuracy of license plate recognition can be improved.

[0170] Corresponding to any of the above embodiments, the present application also proposes an image screening device.

[0171] See Figure 6 , Figure 6 This is a schematic diagram of the structure of an image screening device shown in an embodiment of the present application. Figure 6 As shown, the image screening device 600 may include:

[0172] A first acquisition module 610 is used to acquire an input image;

[0173] An extraction module 620 is configured to extract key points of a target object in the input image;

[0174] A second acquisition module 630 is configured to acquire information of the key point; the information of the key point includes at least one of the following: position information of the key point in the input image, side length information of the image area enclosed by the key point, and confidence information of the key point;

[0175] The screening module 640 is configured to screen the input image based on the information of the key points.

[0176] In some embodiments, the extraction module 620 is further configured to:

[0177] Processing the input image using a target object key point model to obtain a heat map corresponding to the input image;

[0178] The pixel point corresponding to the maximum pixel value in the heat map is determined as the key point of the target object.

[0179] In some embodiments, the extraction module 620 is further configured to:

[0180] The input image is downsampled and upsampled multiple times using a target object key point model to obtain the heat map; wherein, in the multiple upsampling processes, the multi-level feature maps generated in the multiple downsampling processes are fused.

[0181] In some embodiments, the second acquisition module 630 is further configured to:

[0182] determining the position information and / or the side length information according to the coordinates of the key point in the input image;

[0183] The confidence information is determined according to the pixel value of the key point in the heat map.

[0184] In some embodiments, the key point information includes confidence information of the key point, and the screening module 640 is further configured to:

[0185] Determining a confidence mean according to the confidence of each key point;

[0186] The input image is screened according to a comparison result between the confidence mean and a preset confidence threshold.

[0187] In some embodiments, the key point information includes side length information of an image area enclosed by the key point, the image area is a quadrilateral, and the side length information includes first side length information and second side length information of the quadrilateral;

[0188] The screening module 640 is further configured to:

[0189] The input image is screened according to a comparison result between a first side length indicated by the first side length information and a preset first side length threshold, and a comparison result between a second side length indicated by the second side length information and a preset second side length threshold.

[0190] In some embodiments, the key point information includes position information of the key point in the input image, and the screening module 640 is further configured to:

[0191] Determining, based on the position information, a distance from a position of the key point in the input image to an image edge of the input image;

[0192] The input image is screened according to a comparison result between the distance and a preset distance threshold.

[0193] In some embodiments, before screening the input image, the apparatus 600 further includes:

[0194] a determination module, configured to determine a quality score of the input image based on a comparison result, wherein the comparison result includes at least one of the following: a comparison result of the confidence mean value and a preset confidence threshold; a comparison result of the first side length and a preset first side length threshold; a comparison result of the second side length and a preset second side length threshold; a comparison result of the distance and a preset distance threshold;

[0195] The screening module 640 is further configured to:

[0196] The input images are screened based on the quality scores of the input images.

[0197] In some embodiments, the comparison result further includes: a comparison result of the side length of the input image and a preset image side length threshold.

[0198] In some embodiments, the determining module is further configured to:

[0199] In a case where the comparison result indicates one of the following situations, determining a quality score of the input image according to the confidence mean and the area of ​​the input image:

[0200] The side length of the input image reaches the image side length threshold;

[0201] The confidence mean reaches the confidence threshold;

[0202] The first side length reaches the first side length threshold;

[0203] The second side length reaches the second side length threshold;

[0204] The distance does not reach the distance threshold.

[0205] In some embodiments, the extraction module 620 is further configured to:

[0206] Extracting key points of the target object in the input image using a target object key point model generated based on a convolutional neural network;

[0207] The apparatus 600 further includes a training module for:

[0208] Acquire a sample image; the sample image includes annotation information for key points of the target object; the annotation information is used to indicate coordinate information of the key points of the target object in the sample image;

[0209] The target object key point model is trained for multiple rounds using the sample images, wherein each round of training includes:

[0210] Inputting the sample image into the target object key point model to obtain prediction information for the key points;

[0211] determining loss information between the labeled information and the predicted information;

[0212] Parameters of the target object keypoint model are adjusted based on the loss information.

[0213] In the aforementioned scheme, information about the key points of the target object can be obtained. The key point information includes at least one of the following information that can be used to characterize whether the target object is clear and complete: the position of the key point in the input image, the side length of the image area enclosed by the key point, and the confidence level of the key point. The input image is then filtered based on the key point information, and a clear and complete image can be obtained by filtering out the key points that can characterize whether the target object is clear and complete. Compared with related technologies, this method does not rely on extremely time-consuming target detection algorithms and deep learning algorithms, greatly reducing the amount of computation required during image screening, improving image screening efficiency, and achieving lightweight results.

[0214] The embodiment of the image screening device shown in this application can be applied to electronic devices. Accordingly, this application discloses an electronic device, which may include: a processor.

[0215] Memory used to store processor-executable instructions.

[0216] The processor is configured to call the executable instructions stored in the memory to implement the image screening method shown in any of the aforementioned embodiments.

[0217] See Figure 7 , Figure 7 A schematic diagram of the hardware structure of an electronic device shown in an embodiment of the present application.

[0218] like Figure 7 As shown, the electronic device may include a processor for executing instructions, a network interface for network connection, a memory for storing operating data for the processor, and a non-volatile memory for storing instructions corresponding to the image screening device.

[0219] The embodiment of the image screening device can be implemented by software, hardware or a combination of software and hardware. Taking software implementation as an example, as a device in a logical sense, it is formed by the processor of the electronic device in which it is located reading the corresponding computer program instructions in the non-volatile memory into the memory and running it. From the hardware level, in addition to Figure 7In addition to the processor, memory, network interface, and non-volatile memory shown, the electronic device in which the device is located in the embodiment may also include other hardware according to the actual function of the electronic device, which will not be described in detail.

[0220] It is understandable that, in order to increase the processing speed, the corresponding instructions of the image screening device can also be directly stored in the memory, which is not limited here.

[0221] The present application provides a computer-readable storage medium storing a computer program. The computer program can be used to enable a processor to execute the image screening method shown in any of the aforementioned embodiments.

[0222] It will be understood by those skilled in the art that one or more embodiments of the present application may be provided as a method, system, or computer program product. Therefore, one or more embodiments of the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, one or more embodiments of the present application may take the form of a computer program product implemented on one or more computer-usable storage media (which may include but are not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0223] The term "and / or" in the present application means at least one of the two. For example, "A and / or B" may include three options: A, B, and "A and B".

[0224] The various embodiments in this application are described in a progressive manner. Similar portions between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the data processing device embodiment is generally similar to the method embodiment, so its description is relatively simple. For relevant portions, refer to the description of the method embodiment.

[0225] The foregoing description describes specific embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0226] Embodiments of the subject matter and functional operations described in this application may be implemented in the following: digital electronic circuits, tangibly embodied computer software or firmware, computer hardware that may include the structures disclosed in this application and their structural equivalents, or a combination of one or more of them. Embodiments of the subject matter described in this application may be implemented as one or more computer programs, i.e., one or more modules of computer program instructions encoded on a tangible, non-transitory program carrier to be executed by a data processing device or to control the operation of a data processing device. Alternatively or additionally, the program instructions may be encoded on an artificially generated propagation signal, such as a machine-generated electrical, optical, or electromagnetic signal, which is generated to encode information and transmit it to a suitable receiver device for execution by a data processing device. The computer storage medium may be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or a combination of one or more of them.

[0227] The processes and logic flows described herein can be performed by one or more programmable computers executing one or more computer programs to perform the corresponding functions by operating on input data and generating output. The processes and logic flows can also be performed by, and apparatus can be implemented as, special purpose logic circuitry, such as an FPGA (field programmable gate array) or an ASIC (application-specific integrated circuit).

[0228] A computer suitable for executing a computer program may include, for example, a general-purpose and / or special-purpose microprocessor, or any other type of central processing unit. Typically, the central processing unit will receive instructions and data from a read-only memory and / or random access memory. The basic components of a computer may include a central processing unit for implementing or executing instructions and one or more memory devices for storing instructions and data. Typically, a computer will also include one or more large-capacity storage devices for storing data, such as magnetic disks, magneto-optical disks, or optical disks, or the computer will be operably coupled to such large-capacity storage devices to receive data from them or to transmit data to them, or both. However, a computer does not necessarily have such a device. In addition, a computer may be embedded in another device, such as a mobile phone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a global positioning system (GPS) receiver, or a portable storage device such as a universal serial bus (USB) flash drive, to name a few.

[0229] Computer-readable media suitable for storing computer program instructions and data may include all forms of non-volatile memory, media, and storage devices, such as semiconductor memory devices (e.g., EPROM, EEPROM, and flash memory devices), magnetic disks (e.g., internal hard disks or removable disks), magneto-optical disks, and CD ROM and DVD-ROM disks. The processor and memory may be supplemented by, or incorporated in, special purpose logic circuitry.

[0230] Although the present application includes many specific implementation details, these should not be interpreted as limiting the scope of any disclosure or the scope of protection claimed, but are mainly used to describe the features of the specific disclosed embodiments. Certain features described in multiple embodiments of the present application can also be implemented in combination in a single embodiment. On the other hand, the various features described in a single embodiment can also be implemented separately in multiple embodiments or implemented in any suitable sub-combination. In addition, although features can work in some combinations as described and even initially claim protection, one or more features from the claimed combination can be removed from the combination in some cases, and the claimed combination can point to a variation of a sub-combination or a sub-combination.

[0231] Similarly, although operations are depicted in a particular order in the accompanying drawings, this should not be understood as requiring that these operations be performed in the particular order shown or performed sequentially, or that all illustrated operations be performed to achieve the desired result. In some cases, multitasking and parallel processing may be advantageous. In addition, the separation of various system modules and components in the described embodiments should not be understood as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product, or packaged into multiple software products.

[0232] Thus, specific embodiments of the subject matter have been described. Other embodiments are within the scope of the following claims. In some cases, the actions recited in the claims can be performed in a different order and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the particular order shown or sequential sequence to achieve the desired results. In some implementations, multitasking and parallel processing may be advantageous.

[0233] The above are merely preferred embodiments of one or more embodiments of the present application and are not intended to limit one or more embodiments of the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of one or more embodiments of the present application shall be included in the scope of protection of one or more embodiments of the present application.

Claims

1. An image screening method, characterized in that: The method comprises: Get the input image; Extracting key points of a target object in the input image; Acquire information of the key point; the information of the key point includes at least one of the following information for characterizing whether the target object is clear and complete: position information of the key point in the input image, side length information of the image area enclosed by the key point, and confidence information of the key point; wherein, if it is determined through the position information of the key point in the input image that the target object is in the center of the input image, it indicates that the target object is complete; if it is determined through the side length information of the image area enclosed by the key point that the side length of the image area is greater than a preset side length threshold, it indicates that the target object is clear; if it is determined through the confidence information of the key point that the confidence of the key point is higher than a preset confidence threshold, it indicates that the target object is clear; The input image is filtered based on information of the key points.

2. The method according to claim 1, characterized in that The extracting key points of the target object in the input image includes: Processing the input image using a target object key point model to obtain a heat map corresponding to the input image; The pixel point corresponding to the maximum pixel value in the heat map is determined as the key point of the target object.

3. The method according to claim 2, characterized in that The input image is processed using a target object key point model to obtain a heat map corresponding to the input image, including: The input image is downsampled and upsampled multiple times using a target object key point model to obtain the heat map; wherein, in the multiple upsampling processes, the multi-level feature maps generated in the multiple downsampling processes are fused.

4. The method according to claim 2 or 3, characterized in that The obtaining of the key point information includes: determining the position information and / or the side length information according to the coordinates of the key point in the input image; The confidence information is determined according to the pixel value of the key point in the heat map.

5. The method according to any one of claims 1 to 4, characterized in that The key point information includes confidence information of the key point, and screening the input image based on the key point information includes: Determining a confidence mean according to the confidence of each key point; The input image is screened according to a comparison result between the confidence mean and a preset confidence threshold.

6. The method according to any one of claims 1 to 5, characterized in that The key point information includes side length information of the image area enclosed by the key point, the image area is a quadrilateral, and the side length information includes first side length information and second side length information of the quadrilateral; The screening of the input image based on the information of the key points includes: The input image is screened according to a comparison result between a first side length indicated by the first side length information and a preset first side length threshold, and a comparison result between a second side length indicated by the second side length information and a preset second side length threshold.

7. The method according to any one of claims 1 to 6, characterized in that The key point information includes position information of the key point in the input image, and filtering the input image based on the key point information includes: Determining, based on the position information, a distance from a position of the key point in the input image to an image edge of the input image; The input image is screened according to a comparison result between the distance and a preset distance threshold.

8. The method according to claim 7, characterized in that Before screening the input image, the method includes: determining a quality score of the input image according to a comparison result, wherein the comparison result includes at least one of the following: a comparison result of the confidence mean with a preset confidence threshold; a comparison result of the first side length with a preset first side length threshold; a comparison result of the second side length with a preset second side length threshold; a comparison result of the distance with a preset distance threshold; The screening of the input image comprises: The input images are screened based on the quality scores of the input images.

9. The method according to claim 8, characterized in that The comparison result also includes: a comparison result of the side length of the input image and a preset image side length threshold.

10. The method according to claim 9, characterized in that Determining the quality score of the input image according to the comparison result includes: In a case where the comparison result indicates one of the following situations, determining a quality score of the input image according to the confidence mean and the area of ​​the input image: The side length of the input image reaches the image side length threshold; The confidence mean reaches the confidence threshold; The first side length reaches the first side length threshold; The second side length reaches the second side length threshold; The distance does not reach the distance threshold.

11. The method according to any one of claims 1 to 10, characterized in that: The extracting key points of the target object in the input image includes: Extracting key points of the target object in the input image using a target object key point model generated based on a convolutional neural network; The training method of the target object key point model includes: Acquire a sample image; the sample image includes annotation information for key points of the target object; the annotation information is used to indicate coordinate information of the key points of the target object in the sample image; The target object key point model is trained for multiple rounds using the sample images, wherein each round of training includes: Inputting the sample image into the target object key point model to obtain prediction information for the key points; determining loss information between the labeled information and the predicted information; Parameters of the target object keypoint model are adjusted based on the loss information.

12. An image screening device, characterized in that: The device comprises: A first acquisition module is used to acquire an input image; An extraction module, configured to extract key points of a target object in the input image; A second acquisition module is configured to acquire information of the key points; the information of the key points includes at least one of the following information used to characterize whether the target object is clear and complete: position information of the key points in the input image, side length information of the image area enclosed by the key points, and confidence information of the key points; wherein, if the target object is determined to be in the center of the input image through the position information of the key points in the input image, it indicates that the target object is complete; if the side length information of the image area enclosed by the key points determines that the side length of the image area is greater than a preset side length threshold, it indicates that the target object is clear; if the confidence information of the key points determines that the confidence of the key points is higher than a preset confidence threshold, it indicates that the target object is clear; A screening module is used to screen the input image based on the information of the key points.

13. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor implements the image screening method according to any one of claims 1 to 11 by running the executable instructions.

14. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and the computer program is used to enable a processor to execute the image screening method according to any one of claims 1 to 11.

Images