A long-jump distance automatic measurement method and system based on a deep neural network
By using deep neural networks to automatically measure long jump distance, the problems of high dependence on professional personnel, insufficient real-time performance, and insufficient accuracy in existing technologies are solved, and efficient and accurate automatic measurement of long jump distance is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HEBEI UNIV OF TECH
- Filing Date
- 2024-05-24
- Publication Date
- 2026-08-04
AI Technical Summary
Existing automatic methods for measuring long jump distance require highly skilled measurement professionals, have poor real-time performance and accuracy, limited applicability, cannot generate effective data, have high computational complexity, and poor versatility.
A deep neural network-based approach is employed, utilizing techniques such as image segmentation, object detection, semantic segmentation, and perspective transformation to identify people and long jump mats, determine take-off and landing actions, calculate long jump distance, and improve recognition accuracy by leveraging convolutional neural networks and attention mechanisms.
It achieves precise and real-time long jump data processing, improves recognition accuracy, reduces reliance on professionals, has good applicability, is simple to calculate, and has high real-time performance and accuracy.
Smart Images
Figure CN118629088B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image processing technology, and in particular to an automatic method and system for measuring long jump distance based on deep neural networks. Background Technology
[0002] In long jump competitions and training, timely information on jump distance and whether a takeoff was foul is crucial for both athletes and referees. Traditionally, lime is sprinkled near the takeoff board in the sandpit, and footprints are manually observed on the lime to determine if a foul occurred. Long jump distance is usually measured using a measuring tape, requiring at least two people to complete the measurement.
[0003] Automated long jump measurement offers advantages over traditional manual methods. Automated methods can process and analyze data in real time, providing coaches and athletes with immediate feedback, which is crucial for adjusting training and competition strategies. Using automated systems reduces reliance on professional measurement personnel, lowers human resource costs, and minimizes measurement delays that may result from insufficient manpower. Furthermore, this method is not limited by environmental conditions; it can efficiently and accurately measure long jump distances both indoors and outdoors. However, existing automated measurement methods still have many shortcomings: they require highly skilled measurement professionals and cannot accurately and in real-time process and analyze data; their applicability is limited, with most existing methods requiring measurements in specific environments; their real-time performance and accuracy are poor; they can only generate long jump results data and cannot generate other effective data, resulting in high computational complexity and poor versatility. Summary of the Invention
[0004] This invention aims to at least solve one of the technical problems existing in related technologies. To this end, this invention provides an automatic long jump distance measurement method and system based on deep neural networks. By using deep neural networks to identify individuals and long jump mats, it achieves accurate and real-time processing and analysis of long jump data, improving the accuracy of identification. It also features simple deployment, high accuracy, good real-time performance, and good versatility.
[0005] This invention provides an automatic long jump distance measurement method based on a deep neural network, comprising:
[0006] S1: Use a deep learning-based image segmentation model to segment the original video images to obtain video images of people and long jump mats;
[0007] S2: Use an object detection model to perform human posture recognition on human video images to obtain images of the take-off motion and landing motion;
[0008] S3: Use a semantic segmentation model to segment the long jump mat video images to obtain the long jump mat location images;
[0009] S4: Use a semantic segmentation model to segment the take-off motion image to obtain the take-off point image; use a semantic segmentation model to segment the landing motion image to obtain the landing point image;
[0010] S5: Perform perspective transformation on the long jump mat positioning image to obtain the long jump mat recognition image; perform perspective transformation on the take-off point image to obtain the take-off point recognition image; perform perspective transformation on the landing point image to obtain the landing point recognition image.
[0011] S6: The long jump distance is obtained by inputting the start point recognition image, landing point recognition image, and long jump mat recognition image into the ratio calculation module.
[0012] The present invention provides an automatic long jump distance measurement method based on a deep neural network, which further includes step S2, using the coordinate change of the human ankle between frames as the basis for judging whether a jump or a landing has occurred, using the movement of the long jumper's ankle as the judgment information, and judging whether a jump or a landing action has occurred when the pixel change is greater than a set threshold.
[0013] The ankle features are calculated using a convolutional neural network. The expression for calculating the ankle features is as follows:
[0014]
[0015] in, It is a characteristic of the human ankle area. The x-coordinate of the center position of the convolution kernel. The vertical coordinate of the center position of the convolution kernel is . For relative to The offset, For relative to The offset, For the input image, For convolution kernel;
[0016] Pooling is performed on the features of the human ankle area to obtain pooled features. The calculation expression for the pooled features is as follows:
[0017]
[0018] in, For pooling features, For With the center coordinates and offsets respectively and The pooling region, For relative to The offset, For relative to The offset, Coordinates are Characteristics of the human ankle area;
[0019] The pooled human ankle features are then subjected to a nonlinear mapping. The resulting expression for the human ankle features is:
[0020]
[0021] in, The features of the human ankle area after nonlinear mapping. For activation function, A diagram showing the characteristics of the human ankle area. The corresponding pixels;
[0022]
[0023] This refers to the take-off or landing motion image output by the fully connected layer. For trainable weights of fully connected layers, This represents the bias coefficient for the fully connected layer.
[0024] The present invention provides an automatic long jump distance measurement method based on a deep neural network, which further includes using the Douglas-Pokal algorithm in step S3 to perform pixel-level precise positioning of the four endpoints of the long jump mat video image.
[0025] According to the present invention, an automatic long jump distance measurement method based on a deep neural network is provided, which further includes, in step S4, detecting the shoe pixel image of the human body in the take-off motion image or landing motion image by human body analysis, correcting the detected shoe pixel image by tensor processing and linear transformation, and determining the position of the outermost edge of the shoe by the corrected shoe pixel image and the overall position of the shoe, thereby obtaining the take-off point image or landing point image.
[0026] Attention features of the shoe pixel image are obtained through an attention mechanism. The expression for the attention features is:
[0027]
[0028] in, For attention features, For query, As key, For value, For activation function, For transpose operation, The dimension size of each attention subspace;
[0029] The key matrix for extracting attention features is obtained by tensor processing of the key matrix, resulting in a transformed key matrix. The expression for the transformed key matrix is as follows:
[0030]
[0031] in, This is the transformed key matrix. To change the tensor processing operations of the data dimension, Let be the horizontal dimension of the key matrix. Let be the vertical dimension of the key matrix. This refers to the number of image channels;
[0032] By performing a linear transformation on the transformed key matrix, the corrected key matrix is obtained. The expression for the corrected key matrix is as follows:
[0033]
[0034] in, This is the corrected key matrix. It is a linear function;
[0035] The shoe pixel image is corrected using the modified key matrix. This corrected image is then processed by a deep neural network to obtain either the takeoff point image or the landing point image. The expression for calculating the takeoff point image or the landing point image is as follows:
[0036]
[0037]
[0038] in, This is the final output tensor after processing by the deep neural network. It is a multilayer perceptron. For Gaussian error linear units, The kernel size is Convolution operation, The tensor after its shape has been reorganized. Let be the cumulative probability function of the first standard normal distribution. Let be the cumulative probability function of the second standard normal distribution. These are the pixels corresponding to the feature map of the human ankle area.
[0039] The present invention provides an automatic long jump distance measurement method based on a deep neural network, which further includes, in step S5, performing perspective transformation on the long jump mat positioning image, the take-off point image, and the landing point image respectively to obtain a long jump mat recognition image, a take-off point recognition image, and a landing point recognition image. The perspective transformation expression is:
[0040]
[0041] in, The x-coordinate of the image after perspective transformation. The vertical axis of the image after perspective transformation. Let these be the homogeneous coordinates of the image after perspective transformation. This is the perspective transformation matrix. The scaling factor for the x-coordinate of the original image region that has already undergone image segmentation. The scaling factor for the ordinate of the original image region that has already undergone image segmentation. The scaling factor is the homogeneous coordinates of the original image region that has already been segmented.
[0042] The present invention provides an automatic long jump distance measurement method based on a deep neural network, further comprising step S6, which further includes classifying and recognizing the take-off point recognition image, the landing point recognition image, and the long jump mat recognition image using a Softmax activation function to obtain the take-off action image, the landing action image, the take-off position image, the landing position image, and the line-stepping recognition image. The classification and recognition expression is as follows:
[0043]
[0044] in, The Softmax activation function is used. For the target class, For all classes, for Class confidence score, for Class confidence score, The x-coordinate of the center position of the convolution kernel. The vertical coordinate represents the center position of the convolution kernel.
[0045] The present invention provides an automatic long jump distance measurement method based on a deep neural network, further comprising step S6, which further includes calculating the effective distance using the abscissa of the pixels at the toe of the take-off position image and the abscissa of the pixels at the heel of the landing position image; calculating the long jump distance by using the ratio of the effective distance to the overall abscissa of the long jump mat recognition image and the mapping between the pixels of the long jump mat and the actual distance; the long jump distance expression is:
[0046]
[0047] in, For long jump distance, For effective distance, To identify the overall distance in the image of the long jump mat, It is a specific constant.
[0048] This invention also provides a long jump distance estimation system based on a deep neural network, applied to an automatic long jump distance measurement method based on a deep neural network, comprising:
[0049] The module for locating people and long jump mats is used to segment images of people and long jump mats.
[0050] Take-off and landing recognition modules: used to determine take-off and landing actions;
[0051] Semantic segmentation module: used to obtain the long jump mat positioning image, take-off point image, and landing point image;
[0052] Perspective Transformation Module: Used to perform perspective transformation on the long jump mat positioning image, take-off point image, and landing point image;
[0053] Classification and recognition module: used to classify and recognize images of the take-off point, landing point, and long jump mat;
[0054] Calculation ratio module: Used to calculate long jump distance.
[0055] The above-described one or more technical solutions in the embodiments of the present invention have at least one of the following technical effects:
[0056] This invention provides an automatic long jump distance measurement method and system based on deep neural networks. By using deep neural networks to identify people and long jump mats, it achieves accurate and real-time processing and analysis of long jump data, improving the accuracy of identification. It also features simple deployment, high accuracy, good real-time performance, and good versatility.
[0057] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description
[0058] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0059] Figure 1 This is a flowchart illustrating an automatic long jump distance measurement method based on a deep neural network provided by the present invention.
[0060] Figure 2This is a schematic diagram of an automatic long jump distance measurement system based on a deep neural network, provided by the present invention.
[0061] Figure label:
[0062] 201. Locating the person and the long jump mat; 202. Taking-off and landing recognition modules; 203. Semantic segmentation module; 204. Perspective transformation module; 205. Classification and recognition module; 206. Calculating proportions module. Detailed Implementation
[0063] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention. The following embodiments are used to illustrate this invention but cannot be used to limit the scope of this invention.
[0064] The following is combined with Figures 1 to 2 This invention describes an automatic long jump distance measurement method and system based on a deep neural network.
[0065] like Figure 1 As shown, this invention provides an automatic long jump distance measurement method based on a deep neural network, comprising:
[0066] S1: Use a deep learning-based image segmentation model (Segment Anything Model, SAM) to segment the original video image to obtain the long jump mat video image and the person video image;
[0067] Video images of the long jumper and the long jump mat are captured using a wide-angle lens. The video images of the long jumper and the long jump mat containing the jumper are then input into the SAM model. The SAM algorithm is used to segment the long jump mat video image and the task video image to obtain the video images of the jumper and the long jump mat.
[0068] S2: Use the object detection model (YOLOv8) to perform human posture recognition on human video images to obtain images of the take-off motion and landing motion;
[0069] The target detection algorithm is used to locate the long jumper. Non-long jumpers are excluded based on the human body location. The entire long jump process of the long jumper is identified based on the key points of the human body and the long jump movement pattern. The take-off frame and landing frame of the long jump process are obtained by replaying and comparing the action.
[0070] By using the coordinate changes of the human ankle between frames as the basis for determining whether a jump or landing has occurred, and using the movement of the long jumper's ankle as the judgment information, it is determined whether a jump or landing action has occurred when the pixel change is greater than a set threshold.
[0071] The ankle features are calculated using a convolutional neural network. The expression for calculating the ankle features is as follows:
[0072]
[0073] in, It is a characteristic of the human ankle area. The x-coordinate of the center position of the convolution kernel. The vertical coordinate of the center position of the convolution kernel is . For relative to The offset, For relative to The offset, For the input image, For convolution kernel;
[0074] Pooling is performed on the features of the human ankle area to obtain pooled features. The calculation expression for the pooled features is as follows:
[0075]
[0076] in, For pooling features, For With the center coordinates and offsets respectively and The pooling region, For relative to The offset, For relative to The offset, Coordinates are Characteristics of the human ankle area;
[0077] The fit of human ankle features is improved by applying a non-linear mapping. The expression for human ankle features after non-linear mapping is as follows:
[0078]
[0079] in, The features of the human ankle area after nonlinear mapping. For activation function, A diagram showing the characteristics of the human ankle area. The corresponding pixels;
[0080]
[0081] This refers to the take-off or landing motion image output by the fully connected layer. For trainable weights of fully connected layers, This represents the bias coefficient for the fully connected layer.
[0082] S3: Use a semantic segmentation model to segment the long jump mat video image, detect all pixels involved in the long jump mat, detect the contour of the long jump mat through contour approximation, determine the coordinates of the four points of the long jump mat (upper left, upper right, lower right, and lower left) based on the folding ratio of the long jump mat contour, and use the Douglas-Peucker algorithm to perform pixel-level precise localization of the four endpoints of the long jump mat video image to obtain the long jump mat localization image.
[0083] S4: Use a semantic segmentation model to segment the take-off motion image to obtain the take-off point image; use a semantic segmentation model to segment the landing motion image to obtain the landing point image;
[0084] The semantic segmentation model uses the SegFormer algorithm to identify take-off and landing motion images. It detects the shoe pixel image of the human body in the take-off or landing motion image through human body parsing. The detected shoe pixel image is corrected through tensor processing and linear transformation. The position of the outermost edge of the shoe is determined by the shoe pixel image and the overall position of the shoe. The pixel at the very front of the toe is taken as the take-off point. When the landing frame is identified, the last pixel of the heel is taken as the landing point, thus obtaining the take-off point image or the landing point image.
[0085] The key matrix of the shoe pixel image is obtained through an attention mechanism. The expression for the key matrix is:
[0086]
[0087] in, For attention mechanisms, For query, As key, For value, For activation function, For transpose operation, The dimension size of each attention subspace;
[0088] By processing the key matrix as a tensor, the transformed key matrix is obtained. The expression of the transformed key matrix is as follows:
[0089]
[0090] in, This is the transformed key matrix. To change the tensor processing operations of the data dimension, Let be the horizontal dimension of the key matrix. Let be the vertical dimension of the key matrix. This refers to the number of image channels;
[0091] By performing a linear transformation on the transformed key matrix, the corrected key matrix is obtained. The expression for the corrected key matrix is as follows:
[0092]
[0093] in, This is the corrected key matrix. It is a linear function;
[0094] By processing the corrected shoe pixel image through a deep neural network, the take-off point image or landing point image is obtained; the calculation expression for the take-off point image or landing point image is:
[0095]
[0096]
[0097] in, This is the final output tensor after processing by the deep neural network. It is a multilayer perceptron. For Gaussian error linear units, The kernel size is Convolution operation, The tensor after its shape has been reorganized. Let be the cumulative probability function of the first standard normal distribution. Let be the cumulative probability function of the second standard normal distribution. These are the pixels corresponding to the feature map of the human ankle area.
[0098] S5: Perform perspective transformation on the long jump mat positioning image to obtain the long jump mat recognition image; perform perspective transformation on the take-off point image to obtain the take-off point recognition image; perform perspective transformation on the landing point image to obtain the landing point recognition image.
[0099] The perspective transformation expressions for the long jump mat positioning image, take-off point image, and landing point image are as follows:
[0100]
[0101] in, The x-coordinate of the image after perspective transformation. The vertical axis of the image after perspective transformation. Let these be the homogeneous coordinates of the image after perspective transformation. This is the perspective transformation matrix. The scaling factor for the x-coordinate of the original image region that has already undergone image segmentation. The scaling factor for the ordinate of the original image region that has already undergone image segmentation. The scaling factor is the homogeneous coordinates of the original image region that has already been segmented.
[0102] S6: The long jump distance is obtained by inputting the start point recognition image, landing point recognition image, and long jump mat recognition image into the ratio calculation module.
[0103] The softmax activation function is used to classify and recognize the take-off point recognition image, landing point recognition image, and long jump mat recognition image, obtaining the take-off action image, landing action image, take-off position image, landing position image, and line-stepping recognition image. The classification and recognition expression is:
[0104]
[0105] in, The Softmax activation function is used. For the target class, For all, for Class confidence score, for Class confidence score, The x-coordinate of the center position of the convolution kernel. The vertical coordinate represents the center position of the convolution kernel.
[0106] The image of the take-off position is determined by identifying the image of the stepping line.
[0107] The effective distance is calculated using the x-coordinates of the pixels at the toes of the effective take-off position image and the x-coordinates of the pixels at the heel of the effective landing position image. The long jump distance is then calculated by using the ratio of the effective distance to the overall x-coordinate of the long jump mat recognition image and the mapping between the long jump mat pixels and the actual distance. The long jump distance expression is:
[0108]
[0109] in, For long jump distance, For effective distance, To identify the overall distance in the image of the long jump mat, It is a specific constant.
[0110] like Figure 2 As shown, an automatic long jump distance measurement system based on a deep neural network includes:
[0111] Module 201 for locating the person and the long jump mat: used to segment the person image and the long jump mat image;
[0112] Take-off and landing recognition module 202: used to determine take-off and landing actions;
[0113] Semantic segmentation module 203: used to obtain the long jump mat positioning image, the take-off point image, and the landing point image;
[0114] Perspective transformation module 204: used to perform perspective transformation on the long jump mat positioning image, take-off point image and landing point image;
[0115] Classification and recognition module 205: used to classify and recognize the starting point recognition image, landing point recognition image, and long jump mat recognition image;
[0116] Calculation Module 206: Used to calculate long jump distance.
[0117] Through the collaborative work of the above modules, the "Location of Person and Long Jump Mat" module 201 uses a deep learning-based image segmentation model to segment the original video image, obtaining the person video image and the long jump mat video image; the "Takeoff Recognition and Landing Recognition" module 202 uses a target detection model to perform human posture recognition on the person video image, obtaining the takeoff motion image and the landing motion image; the "Semantic Segmentation" module 203 uses a semantic segmentation model to segment the long jump mat video image, obtaining the long jump mat positioning image, and segments the takeoff motion image and the landing motion image, obtaining the takeoff point image and the landing point image. The perspective transformation module 204 performs perspective transformation on the long jump mat positioning image, take-off point image, and landing point image to obtain the long jump mat recognition image, take-off point recognition image, and landing point recognition image. The classification and recognition module 205 identifies the jump action image, landing action image, take-off position image, landing position image, and line recognition image, and determines whether the take-off position image is a valid image by using the line recognition image. The ratio calculation module 206 inputs the take-off point recognition image, landing point recognition image, and long jump mat recognition image into the ratio calculation module to obtain the long jump distance.
[0118] The beneficial effects of this invention are: This invention provides an automatic long jump distance measurement method and system based on deep neural networks, which identifies people and long jump mats through deep neural networks, realizes accurate and real-time processing and analysis of long jump data, improves the accuracy of identification, and also has the characteristics of simple deployment, high accuracy, good real-time performance and good versatility.
[0119] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. An automatic long jump distance measurement method based on deep neural networks, characterized in that, include: S1: Use a deep learning-based image segmentation model to segment the original video images to obtain video images of people and long jump mats; S2: Use an object detection model to perform human posture recognition on human video images to obtain images of the take-off motion and landing motion; S3: Use a semantic segmentation model to segment the long jump mat video images to obtain the long jump mat location images; S4: Use a semantic segmentation model to segment the take-off motion image to obtain the take-off point image; use a semantic segmentation model to segment the landing motion image to obtain the landing point image; The method of human body analysis is used to detect the shoe pixel image of a human body in a take-off motion image or a landing motion image; Attention features of the shoe pixel image are obtained through an attention mechanism. The key matrix of the attention features is extracted. The key matrix is processed by tensor to obtain a transformed key matrix. The transformed key matrix is linearly transformed to obtain a corrected key matrix. The shoe pixel image is corrected using the corrected key matrix. The corrected shoe pixel image is then processed by a deep neural network to determine the position of the outermost edge of the shoe and obtain the take-off point image or landing point image. S5: Perform perspective transformation on the long jump mat positioning image to obtain the long jump mat recognition image; perform perspective transformation on the take-off point image to obtain the take-off point recognition image; perform perspective transformation on the landing point image to obtain the landing point recognition image. S6: The long jump distance is obtained by inputting the start point recognition image, landing point recognition image, and long jump mat recognition image into the ratio calculation module.
2. The automatic long jump distance measurement method based on deep neural networks according to claim 1, characterized in that, In step S2, the coordinate changes of the human ankle between frames are used as the basis for determining whether a jump or a landing has occurred. The movement of the long jumper's ankle is used as the judgment information. When the pixel change is greater than the set threshold, it is determined that a jump or landing action has occurred. The ankle features are calculated using a convolutional neural network. The expression for calculating the ankle features is as follows: in, It is a characteristic of the human ankle area. The x-coordinate of the center position of the convolution kernel. The vertical coordinate of the center position of the convolution kernel is . For relative to The offset, For relative to The offset, For the input image, For convolution kernel; Pooling is performed on the features of the human ankle area to obtain pooled features. The calculation expression for the pooled features is as follows: in, For pooling features, For With the center coordinates and offsets respectively and The pooling region, For relative to The offset, For relative to The offset, Coordinates are Characteristics of the human ankle area; The pooled human ankle features are then subjected to a nonlinear mapping. The resulting expression for the human ankle features is: in, The features of the human ankle area after nonlinear mapping. To find the function with the maximum value, A diagram showing the characteristics of the human ankle area. The corresponding pixels; This refers to the feature vector of the take-off motion image or landing motion image output by the fully connected layer. For trainable weights of fully connected layers, This represents the bias coefficient for the fully connected layer.
3. The automatic long jump distance measurement method based on deep neural networks according to claim 1, characterized in that, In step S3, the Douglas-Pokal algorithm is used to perform pixel-level precise positioning of the four endpoints of the long jump mat video image.
4. The automatic long jump distance measurement method based on deep neural networks according to claim 1, characterized in that, In step S4, the expression for the attention feature is: in, For attention features, For query, As key, For value, For activation function, For transpose operation, The dimension size of each attention subspace; The expression for the transformed key matrix is: in, This is the transformed key matrix. To change the tensor processing operations of the data dimension, Let be the horizontal dimension of the key matrix. Let be the vertical dimension of the key matrix. This refers to the number of image channels; The corrected key matrix expression is: in, This is the corrected key matrix. It is a linear function; The expression for calculating the takeoff point image or landing point image is: in, This is the final output tensor after processing by the deep neural network. It is a multilayer perceptron. For Gaussian error linear units, The kernel size is Convolution operation, The tensor after its shape has been reorganized. Let be the cumulative probability function of the first standard normal distribution. Let be the cumulative probability function of the second standard normal distribution. These are the pixels corresponding to the feature map of the human ankle area.
5. The automatic long jump distance measurement method based on deep neural networks according to claim 1, characterized in that, In step S5, perspective transformation is performed on the long jump mat positioning image, the take-off point image, and the landing point image to obtain the long jump mat recognition image, the take-off point recognition image, and the landing point recognition image. The perspective transformation expression is: in, The x-coordinate of the image after perspective transformation. The vertical axis of the image after perspective transformation. Let these be the homogeneous coordinates of the image after perspective transformation. This is the perspective transformation matrix. The scaling factor for the x-coordinate of the original image region that has already undergone image segmentation. The scaling factor for the ordinate of the original image region that has already undergone image segmentation. The scaling factor is the homogeneous coordinates of the original image region that has already been segmented.
6. The automatic long jump distance measurement method based on deep neural networks according to claim 1, characterized in that, Step S6 further includes classifying and recognizing the take-off point recognition image, landing point recognition image, and long jump mat recognition image using the Softmax activation function to obtain the take-off action image, landing action image, take-off position image, landing position image, and line-stepping recognition image. The classification and recognition expression is: in, The Softmax activation function is used. For the target class, For all classes, for Class confidence score, for Class confidence score.
7. The automatic long jump distance measurement method based on a deep neural network according to claim 6, characterized in that, Step S6 further includes calculating the effective distance using the x-coordinate of the pixels at the toe of the take-off position image and the x-coordinate of the pixels at the heel of the landing position image. The long jump distance is then calculated by using the ratio of the effective distance to the overall x-coordinate of the long jump mat recognition image and the mapping between the long jump mat pixels and the actual distance. The long jump distance expression is: in, For long jump distance, For effective distance, To identify the overall distance in the image of the long jump mat, It is a specific constant.
8. A long jump distance estimation system based on a deep neural network, characterized in that, An automatic long jump distance measurement method based on a deep neural network as described in any one of claims 1 to 7, comprising: The module for locating people and long jump mats is used to segment images of people and long jump mats. Take-off and landing recognition modules: used to determine take-off and landing actions; Semantic segmentation module: used to obtain the long jump mat positioning image, take-off point image, and landing point image; Perspective Transformation Module: Used to perform perspective transformation on the long jump mat positioning image, take-off point image, and landing point image; Classification and recognition module: used to classify and recognize images of the take-off point, landing point, and long jump mat; Calculation ratio module: Used to calculate long jump distance.