Palm information detection method, device, electronic device and storage medium

Abstract

The present application discloses a palm information detection method and device, belonging to the field of biometric feature recognition technology. The method comprises: inputting a palm image into a pre-trained palm and key point detection model, detecting the palm in the palm image through the palm and key point detection model, obtaining a candidate positioning result of the detected palm, the candidate positioning result comprising: a positioning frame confidence score, positioning frame position information and key point information; screening the candidate positioning results according to the positioning frame confidence score to obtain a first positioning result; filtering the first positioning result for positioning frame overlap according to the positioning frame position information to obtain a second positioning result; screening the key point information in the second positioning result for credibility, and determining the detection result of the key points in the palm image. The method adopts a deep neural network to perform palm key point detection based on the learning of positioning frame information and key point information, and has stronger robustness for palm posture.

Description

Technical Field

[0001] The present application relates to the field of biometric identification technology, and in particular to a palm information detection method, device, electronic device, and computer-readable storage medium. Background Art

[0002] In recent years, biometric recognition technology has been widely used in the fields of access control, security, sign-in, etc. for identity authentication, such as face recognition. However, in the face recognition process, there is a defect of low security because face images are easy to obtain. Compared with face recognition, identity recognition based on palm prints and palm veins has higher security because it is difficult to collect palm information such as palm prints and palm veins and has high requirements for equipment. For example, palm vein information can only be obtained with the help of an infrared camera, so the feature is highly concealed. Due to the variability of palm postures, many palm information collection devices currently have strict restrictions on palm postures. Many palm information collection devices require users to place their palms on the collection device, which results in poor user experience.

[0003] It can be seen that the palm information detection method in the prior art still needs to be improved. Summary of the invention

[0004] The embodiments of the present application provide a palm information detection method and device, an electronic device, and a storage medium, which can improve the robustness of palm posture when collecting palm information.

[0005] In a first aspect, an embodiment of the present application provides a palm information detection method, comprising:

[0006] Inputting a palm image into a pre-trained palm and key point detection model, detecting a palm in the palm image by using the palm and key point detection model, and obtaining a candidate positioning result of the detected palm, wherein the candidate positioning result includes: a positioning box confidence score, positioning box position information, and key point information;

[0007] The candidate positioning results are screened according to the positioning box confidence score to obtain a first positioning result;

[0008] Performing a positioning frame overlap filtering on the first positioning result according to the positioning frame position information to obtain a second positioning result;

[0009] The key point information in the second positioning result is screened for credibility to determine the detection result of the key points in the palm image.

[0010] In a second aspect, an embodiment of the present application provides a palm information detection device, including:

[0011] A candidate positioning result acquisition module is used to input a palm image into a pre-trained palm and key point detection model, detect the palm in the palm image through the palm and key point detection model, and obtain a candidate positioning result of the detected palm, wherein the candidate positioning result includes: a positioning box confidence score, positioning box position information, and key point information;

[0012] A first positioning result screening module, used for screening the candidate positioning results according to the positioning box confidence score to obtain a first positioning result;

[0013] A second positioning result screening module is used to filter the first positioning result for positioning frame overlap according to the positioning frame position information to obtain a second positioning result;

[0014] The third positioning result screening module is used to screen the key point information in the second positioning result for credibility, and determine the detection result of the key point in the palm image.

[0015] In a third aspect, an embodiment of the present application further discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the palm information detection method described in the embodiment of the present application when executing the computer program.

[0016] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the palm information detection method disclosed in the embodiment of the present application are performed.

[0017] The palm information detection method disclosed in the embodiment of the present application inputs a palm image into a pre-trained palm and key point detection model, detects the palm in the palm image through the palm and key point detection model, obtains a candidate positioning result of the detected palm, and the candidate positioning result includes: a positioning frame confidence score, positioning frame position information and key point information; the candidate positioning result is screened according to the positioning frame confidence score to obtain a first positioning result; the first positioning result is filtered for positioning frame overlap according to the positioning frame position information to obtain a second positioning result; finally, the key point information in the second positioning result is screened for credibility to determine the detection result of the key point in the palm image. This method uses a deep neural network to perform palm key point detection based on the learning of positioning frame information and key point information, and has stronger robustness to palm gestures. In addition, only one neural network model needs to be trained and used, which occupies less computing resources and storage resources, and has higher efficiency in palm and key point detection.

[0018] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, 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 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.

[0020] Figure 1 is a flow chart of the palm information detection method disclosed in the embodiment of the present application;

[0021] Figure 2 It is a schematic diagram of the palm and key point detection model structure in the palm information detection method disclosed in the embodiment of the present application;

[0022] Figure 3 is a schematic diagram of a candidate positioning result in the palm information detection method disclosed in the embodiment of the present application;

[0023] Figure 4 It is a schematic diagram of defining positioning frame information in the palm information detection method disclosed in the embodiment of the present application;

[0024] Figure 5 is a schematic diagram of key point information definition in the palm information detection method disclosed in the embodiment of the present application;

[0025] Figure 6 It is a schematic diagram of the principle of a key point quality assessment method in the palm information detection method disclosed in the embodiment of the present application;

[0026] Figure 7 is a structural schematic diagram of a palm information detection device disclosed in an embodiment of the present application;

[0027] Figure 8 A block diagram schematically shows an electronic device for executing the method according to the present application; and

[0028] Fig. 9 A storage unit for holding or carrying a program code for implementing the method according to the present application is schematically shown. DETAILED DESCRIPTION

[0029] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are 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.

[0030] Below, we first analyze the defects of the palm information detection method commonly used in the prior art.

[0031] In the prior art, when palm information is detected in a non-deep learning mode, the method of locating the finger gap based on the contour curvature is easily affected by changes in different palm postures (such as open, closed, curled, rotated, tilted, etc.), resulting in greater restrictions on user use.

[0032] The deep learning solution usually uses two deep neural networks. The first neural network performs palm positioning, and the implementation method is similar to other deep learning object positioning models. The model inputs the image to be detected, and the model outputs the positioning result, which contains the classification of the detected object and the positioning frame information. Then, the palm part of the image is cropped out using the classification and positioning frame information. The cropped image is sent to the second neural network for more accurate palm key point positioning. Based on this more accurate key point information, the ROI (Region of Interest) area coordinates are calculated, the palm image is cropped, and the ROI area image is obtained for subsequent feature extraction and recognition. Using two deep neural networks for palm information detection, the two deep neural networks occupy more memory space and require more running time. If multiple palms need to be located, the positioning time will increase several times.

[0033] In the embodiments of the present application, in order to improve the accuracy of palm key point detection and the robustness of palm posture, while shortening the palm key point positioning time and reducing the storage and computing resources consumed in locating the palm key points, a solution is proposed to use a single deep network to complete palm positioning and key point detection.

[0034] The specific implementation of the palm information detection method disclosed in the present application is described below with reference to the accompanying drawings.

[0035] The present application discloses a palm information detection method, such as Figure 1 As shown, the method includes: steps 110 to 140.

[0036] Step 110, inputting the palm image into a pre-trained palm and key point detection model, detecting the palm in the palm image by the palm and key point detection model, and obtaining a candidate positioning result of the detected palm, wherein the candidate positioning result includes: a positioning box confidence score, positioning box position information, and key point information;

[0037] Step 120, screening the candidate positioning results according to the positioning box confidence score to obtain a first positioning result;

[0038] Step 130, performing positioning frame overlap filtering on the first positioning result according to the positioning frame position information to obtain a second positioning result;

[0039] Step 140: Perform credibility screening on the key point information in the second positioning result to determine the detection result of the key points in the palm image.

[0040] The palm and key point detection model used in the embodiment of the present application can be constructed based on the positioning model of the lightweight backbone network. Figure 2 The palm and key point detection model includes: a backbone network 210, a first branch network 220, a second branch network 230 and a third branch network 240.

[0041] Optionally, the backbone network 210 may adopt a deep neural network such as resnet18 or MobileNet. The backbone network 210 may include a number of convolution blocks and connection layers, each of which may be composed of a single or multiple convolution layers, a normalization layer (such as a BN layer) and an activation function. The backbone network 210 may also include a downsampling layer for downsampling the hidden layer vectors output by one or more convolution blocks. The connection layer is used to connect the hidden layer vectors output by the convolution block and the downsampling layer to obtain a connection vector.

[0042] Optionally, the first branch network 220, the second branch network 230 and the third branch network 240 may include a fully connected layer, which is used to classify and map the connection vectors to obtain the positioning box confidence score, positioning box position information and key point information of the positioning box of the palm detected in the palm image.

[0043] The training method of the palm and key point detection model is described below.

[0044] In palm localization and key point detection applications, a single image captured from an image acquisition device or a frame of an image extracted from a video stream is first scaled to a specified size. For example, it is scaled to an image with a height of H and a width of W. After that, the scaled image is normalized to obtain a palm image suitable as input for the palm and key point detection model. For example, based on the pre-calculated values ​​m and v, The scaled image is normalized, and the image pixel value Img is normalized to a distribution between -1 and 1 or between 0 and 1. The values ​​of m and v can be obtained by statistically analyzing the training data of the palm and key point detection model.

[0045] Next, the palm image obtained after normalization is input into the pre-trained palm and key point detection model, and the palm and key point detection model performs feature extraction and mapping processing on the input palm image. The palm and key point detection model will output the following positioning results of the detected palm through the above three branch networks: positioning box confidence score, positioning box position information and key point information. Among them, the positioning box and key points detected by the palm and key point detection model are as follows: Figure 3 In the embodiment of the present application, the number of key points output by the palm and key point detection model can be configured to be 3 to 14.

[0046] In an embodiment of the present application, when a palm image includes multiple palms, the palm and key point detection model can output a positioning result for each palm.

[0047] Optionally, the palm and key point detection model output further includes: a positioning box confidence score (hereinafter referred to as “T cls ” represents the vector of N×C dimensions, the location information of the positioning frame (hereinafter referred to as “T box ” represents a vector of N×D dimensions, and the key point information is hereinafter represented by the symbol “T kp ” represents) is a vector of N×K×P dimensions. Among them, N represents the number of groups of positioning results output by the palm and key point detection model, which can be understood as each group of positioning results corresponding to a possible palm; C represents the positioning box confidence score corresponding to each group of positioning results; D represents the number of dimensions of the positioning box position information, and D is usually 4; K represents the number of key points, and in the embodiment of the present application, the value range of K is an integer in [3,14]; P represents the length of the key point coordinate information, and P is usually an integer multiple of 2. In the embodiment of the present application, the key point coordinate information is the probability distribution of the offset of the key point relative to the specified center position, for example, it can be: the probability of the key point offsetting each preset offset relative to the specified center position.

[0048] In some embodiments of the present application, when the positioning frame is a rectangle, D is 4, and the positioning frame position information T box Respectively represent the offset of the four sides of the palm positioning frame from the specified center position. Figure 4Taking the positioning schematic diagram shown as an example, the designated center position is (xc, yc), the distance between the upper edge of the positioning frame and the designated center position is T, the distance between the left edge of the positioning frame and the designated center position is L, the distance between the right edge of the positioning frame and the designated center position is R, and the distance between the lower edge of the positioning frame and the designated center position is B, then the positioning frame position information includes the offsets of T, L, B, and R. Further, according to the coordinates of the designated center position and the offsets T, L, B, and R of the edges of the positioning frame from the designated center position, the position coordinates of the positioning frame can be calculated.

[0049] In the embodiment of the present application, the key point information is the probability distribution of the offset of each key point relative to the specified center position. Figure 5 As shown, the horizontal coordinate offset of the key point relative to the specified center position can be recorded as dx, and the vertical coordinate offset can be recorded as dy. Among them, the value of P is an integer multiple of 2. In the embodiment of the present application, two z-dimensional vectors can be used to represent the key point coordinate information, that is, vectors with a length of z are used to represent the probability distribution of the horizontal and vertical coordinate offsets of the key point relative to the specified center position. At this time, the length of p is 2×z (where z is an integer greater than 1). By calculating the expectation of this probability distribution, the corresponding offset value can be obtained to determine the key point coordinates. Compared with the previous method of directly predicting the key point coordinate value. By predicting the distance offset probability distribution, and then calculating the distance from the key point to the specified center position according to the offset probability distribution, the predicted key point position can be made more accurate.

[0050] In an embodiment of the present application, each group of positioning results corresponds to a specified center position (xc, yc), wherein the specified center position is predetermined based on a mapping relationship between hidden layer features corresponding to the positioning box in the group of positioning results (i.e., hidden layer features corresponding to the group of positioning results) and the image area of ​​the palm image.

[0051] Taking the example that the backbone network of the palm and key point detection model includes two convolutional block branches, the two convolutional block branches converge at the connection layer, and the connection layer splices the hidden layer feature maps of different scales output by each convolutional block branch. The scale of the hidden layer feature map output by each convolutional block branch corresponds to the downsampling scale of the palm image input to the palm and key point detection model by the corresponding convolutional block branch. Figure 2Taking the model structure shown as an example, the convolution block ConvBlock is divided into two branches, and the two branches converge at the connection layer cat, and the two branches of the convolution block ConvBlock output tensors of different sizes respectively. When the second branch network 230 performs feature mapping based on the hidden layer vector obtained by converging at cat to obtain the coordinates of the positioning box, the position of the target object on the input tensor is related to the horizontal and vertical coordinates of the position of the target object on the image input to the model, which is determined by the characteristics of the convolutional neural network. In an embodiment of the present application, the specified center position on the palm image input to the model corresponding to each tensor output by the model can be predetermined based on the coordinate association relationship between the tensor output by the model and the input image.

[0052] The specific implementation method of determining the designated center position corresponding to each tensor is prior art and will not be repeated in the embodiments of this application. Further, the designated center position corresponding to each tensor can be used as the designated center position corresponding to the positioning result obtained based on the tensor.

[0053] As mentioned above, when the palm image includes multiple palms, the palm and key point detection model can output a set of positioning results for each palm. In the embodiment of the present application, the i-th group of positioning results can be represented as result[i], and the i-th group of positioning results result[i] includes: the positioning box confidence score of the palm positioning box, for example, denoted as "T cls [i]”, the positioning frame position information of the palm positioning frame, for example, “T box [i]” and key point information such as “T kp [i]”.

[0054] In order to provide more accurate key point information for subsequent palm recognition, in the embodiment of the present application, it is necessary to further filter and screen the detection results output by the palm and key point detection model, and only use the detection results with high confidence and high quality as the final detection results.

[0055] First, the positioning results are filtered according to the confidence of the positioning box.

[0056] Optionally, in the aforementioned step 120, the candidate positioning results are screened according to the positioning box confidence score to obtain the first positioning result, including: taking the candidate positioning result whose positioning box confidence score is greater than a preset score threshold as the first positioning result. The preset score threshold is determined based on test data. In an embodiment of the present application, the confidence comprehensive score may be greater than the preset score threshold only when the IoU score of the candidate positioning result and the positioning box confidence score are both relatively high.

[0057] In some embodiments of the present application, when the confidence comprehensive score of a group of candidate positioning results is greater than a preset score threshold, it can be considered that the credibility of the group of candidate positioning results meets the requirements and is retained; when the confidence comprehensive score of a group of candidate positioning results is less than or equal to the preset score threshold, it can be considered that the credibility of the group of candidate positioning results does not meet the requirements and is abandoned. In this way, the retained candidate positioning results are used as the first positioning results and continue to be filtered in subsequent steps.

[0058] Optionally, the positioning frame position information includes: the offset of each edge of the positioning frame from the specified center position. Accordingly, in the aforementioned step 130, the first positioning result is subjected to positioning frame overlap filtering according to the positioning frame position information to obtain a second positioning result, including: calculating the position coordinates of the positioning frame in the first positioning result according to the specified center position and the offset; and performing overlap filtering on the positioning frame according to the position coordinates of each positioning frame to obtain a second positioning result. The specified center position is predetermined based on the mapping relationship between the hidden layer features corresponding to the positioning frame and the image area of ​​the palm image.

[0059] As mentioned above, in the embodiment of the present application, each group of candidate positioning results has a fixed center information, that is, a specified center position, such as Figure 4 The position represented by the coordinates (xc, yc) in . In the embodiment of the present application, the positioning frame position information output by the palm and key point detection model is not the position coordinates of the positioning frame, but the offset of each edge of the positioning frame from the specified center position. Therefore, before performing overlap filtering on the positioning frame, firstly, according to the positioning frame position information T box and the specified center position (xc, yc), calculate the position coordinates of the positioning box in each first positioning result, for example, recorded as (X 1 , Y 1 , X 2 , Y 2 ).

[0060] In some embodiments of the present application, the position coordinates of the positioning frame may be calculated according to the positioning frame position information and the designated center position using the following formula.

[0061]

[0062]

[0063]

[0064]

[0065] Among them, T box[i][1] represents the offset corresponding to the horizontal coordinates of the upper left corner and lower left corner of the i-th positioning box, T box [i][0] represents the offset of the upper left and upper right ordinates of the i-th positioning frame. box [i][3] represents the offset corresponding to the horizontal coordinates of the upper right corner and lower right corner of the i-th positioning box, T box [i][2] represents the offsets of the lower left and lower right ordinates of the i-th positioning box. By predicting the offsets of the sides of the positioning box to the specified center position, and calculating the position coordinates of the positioning box based on the predicted offsets and the specified center position, the anchor is eliminated, so that there is no need to introduce multiple hyperparameters related to the anchor size during the model training phase, reducing the difficulty of adjusting model parameters during the training phase.

[0066] Afterwards, the positioning frames in each first positioning result are subjected to overlap filtering according to the position coordinates of the positioning frames in each first positioning result.

[0067] Optionally, the positioning frames are filtered for overlap according to the position coordinates of each positioning frame to obtain a second positioning result, including: calculating the degree of overlap of the positioning frames according to the position coordinates of each positioning frame, and determining the overlapping positioning frames according to the degree of overlap; and generating a second positioning result according to the first positioning result corresponding to a positioning frame in the overlapping positioning frames.

[0068] In some embodiments of the present application, NMS (non maximum suppression) can be used to filter out overlapping positioning frames. NMS filtering is a routine process. The positioning detection model will output a large number of positioning results, many of which are likely to describe the same object. Therefore, it is necessary to calculate the overlap (IOU, Intersection Over Union) of different positioning results and only retain one of the multiple overlapping results.

[0069] For example, during model training and actual calculation, the overlap threshold of NMS can be set to 0.3 or 0.5, that is, positioning boxes with an overlap greater than 0.3 or 0.5 can be passed.

[0070] After filtering out the positioning frames whose overlaps meet the preset filtering conditions, the remaining positioning frames constitute the second positioning result. In this way, each positioning frame of the second positioning result is a non-overlapping positioning frame. For example, the high-reliability positioning result of an independent palm is retained.

[0071] After filtering in the above steps, one or more independent palm positioning information with a confidence level higher than a preset threshold can be obtained, that is, one or more groups of second positioning results. Next, key point credibility evaluation is further performed on each group of second positioning results, and the positioning results are filtered again based on the key point dimension.

[0072] In the aforementioned step 140, the key point information in the second positioning result is screened for credibility to determine the detection result of the key points in the palm image, including: for each key point in the second positioning result, the coordinates of the key point in the palm image are calculated according to the specified center position and the offset probability distribution information; the coordinates of the key point in the palm image are quality evaluated to obtain the key point information evaluation result; and the detection result of the key point in the palm image is determined according to the key point information evaluation result.

[0073] As mentioned above, in the embodiment of the present application, each group of positioning results has a fixed center information, that is, a specified center position, such as Figure 5 The position represented by the coordinates (xc, yc) in . In the embodiment of the present application, the key point information output by the palm and key point detection model is not the position coordinates of the key point. The key point information in the embodiment of the present application includes: the probability distribution information of the offset of the key point relative to the specified center position. Among them, the offset step size is preferably 0.5, 1, 2. As mentioned above, the key point coordinate information of each key point is represented by a vector of length P. For example, the key point coordinate information of the key point can be expressed as [dy, dx], wherein dy and dx are z-dimensional vectors, and the key point coordinate information is obtained by splicing dy and dx.

[0074] In an embodiment of the present application, when filtering positioning results based on key points, it is first necessary to calculate the coordinates of the key point in the palm image according to the designated center position and the probability distribution information of the offset in the key point information. The designated center position is the designated center position corresponding to the second positioning result of the group. The designated center position is pre-calculated, and the calculation method of the designated center position is described above and will not be repeated here.

[0075] In some embodiments of the present application, the coordinates of the key points in the palm image are calculated according to the designated center position and the offset probability distribution information, including: calculating the expected value of the offset probability distribution according to the offset probability distribution information; and calculating the coordinates of the key points in the palm image according to the designated center position and the expected value. For example, the position coordinates of each key point can be calculated according to the key point information and the designated center position by the following formula.

[0076]

[0077]

[0078] Among them, when the offset step size is 1 and non-zero symmetry, the offset value can be Δx=1, 2, 3, etc., Δy=1, 2, 3, etc. In some other embodiments of the present application, the step size can also be set to zero symmetry. For example, in the case of zero symmetry, when the step size is 0.5, the offset value can be: Δx=etc, -1, -0.5, 0, 0.5, 1, etc.

[0079] Where Δx and Δy represent the offset of the horizontal and vertical coordinates relative to the specified center position, respectively; n represents the positioning result number, k represents the key point number, P represents the length of the key point coordinate information, j represents the number of the offset probability distribution information, and T kp [n][k][j] indicates that the kth key point in the nth group of positioning results is offset by Δy relative to the specified center position (xc, yc) j and Δx j (i.e. the probability value at the jth offset).

[0080] In the embodiment of the present application, the length of the probability distribution of the offset can be preferably set to 9, 11, 13, 15, 17, 19 according to different scenarios. The probability distribution step size can be preferably set to 0.5, 1, 2 according to the situation. The offset can be zero symmetric or non-zero symmetric.

[0081] By calculating the expectation of the probability distribution of the coordinate offset, the distance value of the corresponding specified coordinate position is obtained. Compared with the traditional method of directly predicting the distance value, a more stable coordinate value can be obtained.

[0082] By predicting the probability distribution of the offset from the key point to the specified center position, and calculating the position coordinates of the key point based on the predicted offset probability distribution and the specified center position, the problem of large deviation in the prediction results of object detection without anchor (i.e. anchor box) is improved. At the same time, the anchor is cancelled, so that there is no need to introduce multiple hyperparameters related to the anchor size in the model training stage, reducing the difficulty of adjusting the model parameters in the training stage.

[0083] According to the above method, the positioning frame and key point coordinates of each screened and retained palm can be calculated respectively. Compared with the prior art that needs to use two neural networks to determine the positioning frame and key points respectively, the palm detection efficiency is significantly improved.

[0084] Next, the coordinates of the key points in the palm image are evaluated for quality to obtain key point information evaluation results, which are used to filter out poor quality positioning results based on the positioning point information to prevent inaccurate positioning results from affecting subsequent palm recognition applications.

[0085] Optionally, the quality assessment of the coordinates of the key points in the palm image to obtain the key point information assessment result includes: performing quality assessment on the key points in each second positioning result according to the positional relationship of the key points in the palm image to obtain the key point information assessment result; and / or performing quality assessment on the key points in each second positioning result according to the relative position of the key points and the positioning frame corresponding to the key points to obtain the key point information assessment result. The positioning frame corresponding to the key point is a positioning frame in a group of second positioning results with the key point.

[0086] For example, you can first form a vector by placing two adjacent key points in a clockwise or counterclockwise order. Then, you can use the formula Calculate the vector angle, where and Respectively represent the vectors formed by two key points that are adjacent in sequence. Afterwards, by calculating the included angle, it is further determined whether there are key points with serious errors in the key points in each group of second positioning results. If so, an evaluation result of low credibility of key point information can be obtained. If not, an evaluation result of high credibility of key point information can be obtained.

[0087] In some embodiments of the present application, key points with serious errors may also be filtered out to obtain reliable key point information.

[0088] For another example, the confidence score of the key point can be calculated by calculating the relative position of the key point and the positioning frame, and the credibility evaluation result of the key point can be obtained based on the comparison result of the confidence score and the preset score threshold.

[0089] Optionally, the key points in each second positioning result are quality evaluated according to the relative position of the key point and the positioning box corresponding to the key point to obtain a key point information evaluation result, including: performing quality evaluation on the key points in the second positioning result according to the area intersection and union ratio of the first image area surrounded by the key point in the second positioning result and the second image area surrounded by the positioning box corresponding to the key point to obtain a key point information evaluation result.

[0090] In some embodiments of the present application, all key points in the second positioning result may be connected in sequence to obtain a polygon, and the area surrounded by the polygon is used as the first image area.

[0091] In some other embodiments of the present application, in order to reduce the amount of calculation and the complexity of calculation, some key points in the second positioning result can be selected to connect to obtain a polygon, and the area surrounded by the polygon is used as the first image area.

[0092] For example, at least four points on the top, bottom, left, and right of the key points in the second positioning result may be selected and connected in sequence to form a convex polygon, such as Figure 6 The convex polygon in the image is used, and the area surrounded by the convex polygon is used as the first image area 610. Figure 6 In FIG. 6 , the rectangular area 620 represents the second image area surrounded by the positioning frame.

[0093] After that, determine whether the edge of the convex polygon has an intersection with the positioning box. If there is an intersection, it means that the polygon and the rectangular box partially overlap. Calculate the coordinates of the intersection point and the area of ​​the overlapping part. For example, find the center point of the overlapping part, connect each vertex with the center point to form a small triangle, calculate the area of ​​the small triangle and sum it up, and get the area S1 of the convex polygon of the overlapping part, such as Figure 6 The area S2 of the large convex polygon (i.e., the first image area 610) is calculated. Then, the area S3 of the second image area 620 surrounded by the positioning frame is calculated. Finally, the area intersection and union ratio P-IOU of the first image area surrounded by the key point and the second image area surrounded by the positioning frame corresponding to the key point can be calculated according to the formula P-IOU=S1 / (S2∪S3).

[0094] If there is no intersection between the side of the convex polygon and the positioning box, it is further determined whether the convex polygon contains the positioning box, or the positioning box contains the convex polygon. If the above inclusion situation exists, the area intersection and union ratio P-IOU of the first image area surrounded by the key point and the second image area surrounded by the positioning box corresponding to the key point is calculated according to the formula P-IOU = (area of ​​convex polygon ∩area of ​​positioning box) / (area of ​​convex polygon ∪area of ​​positioning box).

[0095] If the edge of the convex polygon has no intersection with the positioning box and there is no mutual inclusion, then the area intersection-union ratio P-IOU=0.

[0096] Next, the area intersection and union ratio P-IOU of the first image area surrounded by the key point and the second image area surrounded by the positioning frame corresponding to the key point is used as the evaluation score of the key point to evaluate the quality of the key point. If the area intersection and union ratio P-IOU is zero, it means that the quality of the key point is very poor. The closer the area intersection and union ratio P-IOU is to the quality score threshold, the better the quality of the key point. Among them, the quality score threshold can be 1, or the area intersection and union ratio P-IOU value calculated based on the calibrated key point and positioning frame.

[0097] After quality assessment of the key points, the second positioning result with reliable key point information is retained as the final detection result of the palm image.

[0098] In order to make the palm information detection method disclosed in the embodiment of the present application clearer, the training method of the palm and key point detection model is illustrated below.

[0099] The palm and key point detection model is used to perform palm and key point detection on an input palm image, and output a predicted value of a positioning box confidence score, a predicted value of a positioning box position information, and a predicted value of key point information of the detected palm. The loss function of the palm and key point detection model is configured as follows: according to the positioning box confidence prediction loss, the positioning box position prediction loss, and the key point position prediction loss, the loss value of the palm and key point detection model is calculated, wherein the positioning box confidence prediction loss is obtained by multiplying the positioning box confidence score prediction value and the IoU score by cross entropy calculation, the positioning box position prediction loss is calculated by the target detection regression loss function, and the key point position prediction loss is obtained by weighted summing the probability expectation loss of the key point information offset and the probability distribution loss of the offset.

[0100] As mentioned above, the palm and key point detection model includes: a backbone network 210, a first branch network 220, a second branch network 230 and a third branch network 240, wherein:

[0101] The backbone network is used to extract features from the palm image input to the palm and key point detection model to obtain a hidden layer vector;

[0102] The first branch network is used to perform feature mapping on the hidden layer vector to obtain a predicted value of a confidence score of a positioning box corresponding to a palm detected in the palm image;

[0103] The second branch network is used to perform feature mapping on the hidden layer vector to obtain a predicted value of the positioning frame position information of the palm detected in the palm image;

[0104] The third branch network is used to perform feature mapping on the hidden layer vector to obtain a predicted value of key point information of the palm detected in the palm image.

[0105] During the training process, the loss function of the palm and key point detection model can be expressed as:

[0106] Loss = αLoss cls +βLoss box +δLoss keypoint ;

[0107] Among them, α, β and δ are hyperparameters, and their values ​​can be set as needed. For example, the values ​​of α, β and δ can be 2; Loss represents the model loss, Loss cls Represents the prediction loss of the positioning box confidence, Lossbox Represents the positioning box position prediction loss, Loss keypoint represents the key point position prediction loss.

[0108] Optionally, the positioning box confidence prediction loss Loss cls It can be calculated according to the following formula:

[0109] Loss cls = -(1-y)log(1-P score )-ylogP score ;

[0110] Among them, P score =T cls ×IoU,T cls represents the confidence prediction score of the positioning box, IoU represents the metric score of the accuracy of detecting the palm in the palm image, that is, the IoU score, and y represents the classification label, that is, whether the target is a corresponding classified object. For example, y=1 means that the target is a classified object, and y=0 means that the target is not a classified object.

[0111] Among them, IoU (Intersection over Union) is a standard for measuring the accuracy of detecting corresponding objects in a specific data set. IoU is a simple measurement standard. As long as a task that obtains a predicted range (bounding boxes) in the output can be measured using the IoU indicator. The calculation method of the IoU score of the positioning box refers to the prior art and will not be repeated in the embodiments of this application.

[0112] During the palm and key point detection model training process, cross entropy is used as the loss function to calculate the positioning box confidence prediction loss. The score of the palm sample in the image is close to 1, and the score of the background is close to 0. The difference between this solution and the traditional solution is that when the positioning box confidence score is finally calculated, it will also be multiplied by a positioning result quality score as the positioning box confidence score during the training process. The positioning result quality score used here uses the IoU score. That is, the score P sent to the cross entropy loss function during the training process score = positioning box confidence score × IoU. From the loss function of the positioning box confidence, it can be seen that only when the positioning box confidence score and the IoU score are both high, the result score P score Only when T is close to 1, the prediction loss of the positioning box confidence will be relatively small, which will help improve the detection accuracy of the trained palm and key point detection model. That is, for the positioning result of an object, only when T cls When both P and IoU are close to 1, score It is close to 1, therefore, both the confidence and the positioning box results are constrained.

[0113] Optional, the positioning box position prediction loss Loss box GIoU loss (Generalized Intersection over Union Loss, random loss function for target detection) can be used for calculation. For example, the following formula is used to calculate the positioning box position prediction loss Loss box :

[0114] Loss box =GIoU(box,b′ox);

[0115] Among them, box represents the predicted value of the positioning box position, b′ox represents the true value of the labeled positioning box position, and box = [X1, Y1, X2, Y2].

[0116] Optionally, the key point position prediction loss Loss can be calculated by the following formula: keypoint :

[0117]

[0118] in, represents the mean square error loss of key point information, kp represents the predicted value of key point information, including the predicted value of key point position, Indicates the true value of the key point information, including the true value of the key point position; Loss d It represents the probability distribution loss of the offset of the key point position relative to the specified center position, τ is the weight value, and the value of τ is less than 1, for example, τ = 0.3.

[0119] Optional, the probability distribution loss of the offset of the key point position relative to the specified center position Loss d , can be calculated by the following formula:

[0120] Loss d =-((Δy j+1 -y)logS j +(y-Δy j )logS j+1 );

[0121] in, y j+1 and j represents the target stable value, such as the marked probability distribution stable value, and y represents the stable value of the model output; S j and S j+1 The probability distribution of the offset of the key point coordinates representing the same offset changes. The above offset probability distribution loss Loss d It can be seen from the calculation formula that the probability distribution loss Lossd The constraint can make the stable value select a value near the target stable value, so that for the key point coordinates with the same offset, the output offset probability distribution is more stable, which helps to improve the accuracy of key point detection.

[0122] During the palm and key point detection model training process, the overall loss value of the model is optimized, and the model parameters are continuously optimized to reduce the prediction loss of the positioning box confidence, the positioning box position prediction loss, and the key point position prediction loss, so as to complete the training of the palm and key point detection model. Afterwards, the backbone network, the first branch network, the second branch network, and the third branch network of the trained palm and key point detection model are migrated and deployed for palm and palm key point detection.

[0123] The palm information detection method disclosed in the embodiment of the present application inputs a palm image into a pre-trained palm and key point detection model, detects the palm in the palm image through the palm and key point detection model, obtains a candidate positioning result of the detected palm, and the candidate positioning result includes: a positioning frame confidence score, positioning frame position information and key point information; the candidate positioning result is screened according to the positioning frame confidence score to obtain a first positioning result; the first positioning result is filtered for positioning frame overlap according to the positioning frame position information to obtain a second positioning result; finally, the key point information in the second positioning result is screened for credibility to determine the detection result of the key point in the palm image. This method uses a deep neural network to perform palm key point detection based on the learning of positioning frame information and key point information, and has stronger robustness to palm gestures. In addition, only one neural network model needs to be trained and used, which occupies less computing resources and storage resources, and has higher efficiency in palm and key point detection.

[0124] The present application also discloses a palm information detection device. Figure 7 As shown, the device comprises:

[0125] The candidate positioning result acquisition module 710 is used to input the palm image into a pre-trained palm and key point detection model, detect the palm in the palm image through the palm and key point detection model, and obtain the candidate positioning result of the detected palm, wherein the candidate positioning result includes: a positioning box confidence score, positioning box position information and key point information;

[0126] A first positioning result screening module 720, configured to screen the candidate positioning results according to the positioning box confidence score to obtain a first positioning result;

[0127] A second positioning result filtering module 730 is used to perform positioning frame overlap filtering on the first positioning result according to the positioning frame position information to obtain a second positioning result;

[0128] The third positioning result screening module 740 is used to screen the key point information in the second positioning result for credibility, and determine the detection result of the key point in the palm image.

[0129] Optionally, the first positioning result screening module 720 is further configured to:

[0130] The candidate positioning result whose positioning box confidence score is greater than a preset score threshold is used as the first positioning result.

[0131] Optionally, the positioning frame position information includes: an offset of each side of the positioning frame from a specified center position, wherein the specified center position is predetermined based on a mapping relationship between a hidden layer feature corresponding to the positioning frame and an image area of ​​the palm image, and performing positioning frame overlap filtering on the first positioning result based on the positioning frame position information to obtain a second positioning result includes:

[0132] Calculating the position coordinates of the positioning frame in the first positioning result according to the specified center position and the offset;

[0133] The positioning frames are subjected to overlap filtering according to the position coordinates of each positioning frame to obtain a second positioning result.

[0134] Optionally, the key point information includes: probability distribution information of the offset of the key point relative to the specified center position, and the third positioning result screening module 740 includes:

[0135] For each key point in each second positioning result, calculating the coordinates of the key point in the palm image according to the specified center position and the offset probability distribution information;

[0136] Performing quality assessment on the coordinates of the key points in the palm image to obtain key point information assessment results;

[0137] According to the key point information evaluation result, the key point detection result in the palm image is determined.

[0138] Optionally, performing quality assessment on the coordinates of the key points in the palm image to obtain key point information assessment results includes:

[0139] According to the positional relationship of the key points in the palm image, respectively, quality evaluation is performed on the key points in each second positioning result to obtain a key point information evaluation result; and / or,

[0140] According to the relative position of the key point and the positioning frame corresponding to the key point, quality assessment is performed on the key point in each second positioning result to obtain a key point information assessment result.

[0141] Optionally, the performing quality assessment on the key points in each second positioning result respectively according to the relative position of the key point and the positioning frame corresponding to the key point to obtain the key point information assessment result includes:

[0142] The key points in the second positioning result are quality evaluated based on the area intersection and union ratio of the first image area enclosed by the key points in the second positioning result and the second image area enclosed by the positioning frame corresponding to the key points to obtain a key point information evaluation result.

[0143] Optionally, the palm and key point detection model is used to perform palm and key point detection on an input palm image, and output a positioning box confidence score prediction value, a positioning box position information prediction value and a key point information prediction value of the detected palm; the loss function of the palm and key point detection model is configured as: according to the positioning box confidence prediction loss, the positioning box position prediction loss and the key point position prediction loss, the loss value of the palm and key point detection model is calculated, wherein the positioning box confidence prediction loss is obtained by multiplying the positioning box confidence score prediction value and the IoU score by cross entropy calculation, the positioning box position prediction loss is calculated by a target detection random loss function, and the key point position prediction loss is obtained by weighted summing the probability expectation loss of the key point information offset and the probability distribution loss of the offset.

[0144] The palm information detection device disclosed in the embodiment of the present application is used to implement the palm information detection method described in the embodiment of the present application. The specific implementation methods of each module of the device will not be repeated here, and reference can be made to the specific implementation methods of the corresponding steps in the method embodiment.

[0145] The palm information detection device disclosed in the embodiment of the present application inputs a palm image into a pre-trained palm and key point detection model, detects the palm in the palm image through the palm and key point detection model, obtains a candidate positioning result of the detected palm, and the candidate positioning result includes: a positioning frame confidence score, positioning frame position information and key point information; the candidate positioning result is screened according to the positioning frame confidence score to obtain a first positioning result; the first positioning result is filtered for positioning frame overlap according to the positioning frame position information to obtain a second positioning result; finally, the key point information in the second positioning result is screened for credibility to determine the detection result of the key point in the palm image. This method uses a deep neural network to perform palm key point detection based on the learning of positioning frame information and key point information, and has stronger robustness to palm gestures. In addition, only one neural network model needs to be trained and used, which occupies less computing resources and storage resources, and has higher efficiency in palm and key point detection.

[0146] Each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0147] The above is a detailed introduction to a palm information detection method and device provided by the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and a core idea. At the same time, for general technicians in this field, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

[0148] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative effort.

[0149] The various component embodiments of the present application can be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It should be understood by those skilled in the art that a microprocessor or digital signal processor (DSP) can be used in practice to implement some or all functions of some or all components in the electronic device according to the embodiment of the present application. The present application can also be implemented as a device or apparatus program (e.g., computer program and computer program product) for executing part or all of the methods described herein. Such a program implementing the present application can be stored on a computer-readable medium, or can have the form of one or more signals. Such a signal can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0150] For example, Figure 8 An electronic device that can implement the method according to the present application is shown. The electronic device can be a PC, a mobile terminal, a personal digital assistant, a tablet computer, etc. The electronic device traditionally includes a processor 810 and a memory 820 and a program code 830 stored on the memory 820 and can be run on the processor 810, and the processor 810 implements the method described in the above embodiment when executing the program code 830. The memory 820 can be a computer program product or a computer-readable medium. The memory 820 can be an electronic memory such as a flash memory, an EEPROM (electrically erasable programmable read-only memory), an EPROM, a hard disk or a ROM. The memory 820 has a storage space 8201 for the program code 830 of the computer program for executing any method step in the above method. For example, the storage space 8201 for the program code 830 can include various computer programs for implementing the various steps in the above method respectively. The program code 830 is a computer-readable code. These computer programs can be read from one or more computer program products or written into the one or more computer program products. These computer program products include program code carriers such as hard disks, compact disks (CDs), memory cards or floppy disks. The computer program includes a computer readable code, and when the computer readable code is executed on an electronic device, the electronic device is caused to execute the method according to the above embodiment.

[0151] The embodiment of the present application further discloses a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the palm information detection method as described in the embodiment of the present application are implemented.

[0152] Such a computer program product may be a computer-readable storage medium, which may have a computer-readable storage medium having a Figure 8The memory 820 in the electronic device shown in the figure is similar to the storage segment, storage space, etc. The program code can be compressed and stored in the computer-readable storage medium in an appropriate form. The computer-readable storage medium is generally as shown in the reference Fig. 9 The portable or fixed storage unit. Generally, the storage unit includes computer readable code 830', which is a code read by a processor. When the code is executed by the processor, each step in the method described above is implemented.

[0153] The term "one embodiment", "embodiment" or "one or more embodiments" herein means that a particular feature, structure or characteristic described in conjunction with the embodiment is included in at least one embodiment of the present application. In addition, please note that the examples of the term "in one embodiment" here do not necessarily all refer to the same embodiment.

[0154] In the description provided herein, a large number of specific details are described. However, it is understood that the embodiments of the present application can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.

[0155] In the claims, any reference signs placed between brackets shall not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present application may be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third etc. does not indicate any order. These words may be interpreted as names.

[0156] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. 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 palm information detection method, characterized in that: The method comprises: Inputting a palm image into a pre-trained palm and key point detection model, detecting a palm in the palm image by using the palm and key point detection model, and obtaining a candidate positioning result of the detected palm, wherein the candidate positioning result includes: a positioning box confidence score, positioning box position information, and key point information; The candidate positioning results are screened according to the positioning box confidence score to obtain a first positioning result; Performing a positioning frame overlap filtering on the first positioning result according to the positioning frame position information to obtain a second positioning result; Performing credibility screening on the key point information in the second positioning result to determine a detection result of the key point in the palm image; Wherein, the key point information includes: the probability distribution information of the offset of the key point relative to the specified center position, the credibility screening of the key point information in the second positioning result, and determining the detection result of the key point in the palm image, including: for each key point in each second positioning result, calculating the coordinates of the key point in the palm image according to the specified center position and the probability distribution information of the offset; performing quality assessment on the coordinates of the key point in the palm image to obtain the key point information assessment result; and determining the detection result of the key point in the palm image according to the key point information assessment result; The step of calculating the coordinates of the key point in the palm image according to the designated center position and the offset probability distribution information comprises: calculating the expected value of the offset probability distribution according to the offset probability distribution information; and calculating the coordinates of the key point in the palm image according to the designated center position and the expected value by the following formula; Wherein, (xc, yc) is the designated center position, is the coordinate of the key point in the palm image, and They represent the offset of the horizontal and vertical coordinates relative to the specified center position respectively; n represents the positioning result number, k represents the key point number, P represents the length of the key point coordinate information, and j represents the number of the probability distribution information of the offset. Indicates the offset of the kth key point in the nth group of positioning results relative to the specified center position (xc, yc) and The probability value when .

2. The method according to claim 1, characterized in that The screening of the candidate positioning results according to the positioning box confidence score to obtain a first positioning result includes: The candidate positioning result whose positioning box confidence score is greater than a preset score threshold is used as the first positioning result.

3. The method according to claim 1, characterized in that: The positioning frame position information includes: an offset of each side of the positioning frame from a specified center position, wherein the specified center position is predetermined according to a mapping relationship between a hidden layer feature corresponding to the positioning frame and an image area of ​​the palm image, and the first positioning result is subjected to positioning frame overlap filtering according to the positioning frame position information to obtain a second positioning result, including: Calculating the position coordinates of the positioning frame in the first positioning result according to the specified center position and the offset; The positioning frames are subjected to overlap filtering according to the position coordinates of each positioning frame to obtain a second positioning result.

4. The method according to claim 1, characterized in that The step of performing quality assessment on the coordinates of the key points in the palm image to obtain key point information assessment results includes: According to the positional relationship of the key points in the palm image, respectively, quality evaluation is performed on the key points in each second positioning result to obtain a key point information evaluation result; and / or, According to the relative position of the key point and the positioning frame corresponding to the key point, quality assessment is performed on the key point in each second positioning result to obtain a key point information assessment result.

5. The method according to claim 4, characterized in that The step of performing quality assessment on the key points in each second positioning result according to the relative position of the key point and the positioning frame corresponding to the key point to obtain a key point information assessment result includes: The key points in the second positioning result are quality evaluated based on the area intersection and union ratio of the first image area enclosed by the key points in the second positioning result and the second image area enclosed by the positioning frame corresponding to the key points to obtain a key point information evaluation result.

6. The method according to any one of claims 1 to 5, characterized in that: The palm and key point detection model is used to perform palm and key point detection on the input palm image, and output a positioning frame confidence score prediction value, a positioning frame position information prediction value, and a key point information prediction value of the detected palm; The loss function of the palm and key point detection model is configured as follows: according to the positioning box confidence prediction loss, the positioning box position prediction loss and the key point position prediction loss, the loss value of the palm and key point detection model is calculated, wherein the positioning box confidence prediction loss is obtained by multiplying the positioning box confidence score prediction value and the IoU score and then calculating through cross entropy, the positioning box position prediction loss is calculated by the target detection regression loss function, and the key point position prediction loss is obtained by weighted summing the probability expectation loss of the key point information offset and the probability distribution loss of the offset.

7. A palm information detection device, characterized in that: The device comprises: A candidate positioning result acquisition module is used to input a palm image into a pre-trained palm and key point detection model, detect the palm in the palm image through the palm and key point detection model, and obtain a candidate positioning result of the detected palm, wherein the candidate positioning result includes: a positioning box confidence score, positioning box position information, and key point information; A first positioning result screening module, used for screening the candidate positioning results according to the positioning box confidence score to obtain a first positioning result; A second positioning result screening module is used to filter the first positioning result for positioning frame overlap according to the positioning frame position information to obtain a second positioning result; A third positioning result screening module, used for screening the key point information in the second positioning result for credibility, and determining the detection result of the key point in the palm image; Wherein, the key point information includes: the probability distribution information of the offset of the key point relative to the specified center position, the credibility screening of the key point information in the second positioning result, and determining the detection result of the key point in the palm image, including: for each key point in each second positioning result, calculating the coordinates of the key point in the palm image according to the specified center position and the probability distribution information of the offset; performing quality assessment on the coordinates of the key point in the palm image to obtain the key point information assessment result; and determining the detection result of the key point in the palm image according to the key point information assessment result; The step of calculating the coordinates of the key point in the palm image according to the designated center position and the offset probability distribution information comprises: calculating the expected value of the offset probability distribution according to the offset probability distribution information; and calculating the coordinates of the key point in the palm image according to the designated center position and the expected value by the following formula; Wherein, (xc, yc) is the designated center position, is the coordinate of the key point in the palm image, and They represent the offset of the horizontal and vertical coordinates relative to the specified center position respectively; n represents the positioning result number, k represents the key point number, P represents the length of the key point coordinate information, and j represents the number of the probability distribution information of the offset. Indicates the offset of the kth key point in the nth group of positioning results relative to the specified center position (xc, yc) and The probability value when .

8. An electronic device comprising a memory, a processor, and a program code stored in the memory and executable on the processor, wherein: When the processor executes the program code, the palm information detection method according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium having a program code stored thereon, characterized in that: When the program code is executed by a processor, the steps of the palm information detection method according to any one of claims 1 to 6 are implemented.

Images