Three-dimensional human key point detection method based on RGB-D camera in crowd gathering point scene

Abstract

The present application relates to a three-dimensional human body key point detection method based on an RGB-D camera in a crowd gathering scene, which comprises the following steps: obtaining an RGB image of a target human body to obtain two-dimensional pixel coordinates of human body key points through a deep learning-based 2D human body posture estimation algorithm, obtaining a depth image of the target human body to obtain depth values corresponding to the human body key points, optimizing the two-dimensional pixel coordinates through an adaptive mean filter, and obtaining three-dimensional coordinates of the human body key points in a camera coordinate system through coordinate transformation. The present application can quickly and accurately detect and obtain three-dimensional human body key point coordinates in a crowd gathering scene, provide information support for rescue work, and help robots operate and perform rescue tasks more accurately.

Description

A 3D Human Keypoint Detection Method Based on RGB-D Camera in Scenes with Concentrated Flaws Technical Field

[0001] This invention relates to a method for detecting 3D human key points based on an RGB-D camera in scenarios with concentrated damage points, belonging to the fields of computer vision and human-computer interaction technology. Background Technology

[0002] Human keypoint detection, also known as human pose estimation, is increasingly in demand due to the rapid development of computer vision and human-computer interaction. Human pose estimation refers to the process of acquiring and analyzing human posture using computer vision technology, and it is crucial for applications such as action recognition, pose analysis, and motion tracking. Furthermore, applying 3D human keypoint detection methods to robotic systems that collect injury data allows for real-time acquisition of the injured person's pose information, obtaining the positional information of key human points such as the nose, wrists, and ankles. This information can support robotic rescue operations, such as approaching the injured person and attaching injury detection sensors like nose clips, wristbands, and anklebands, thereby improving the efficiency of rescue work.

[0003] Traditional human pose estimation methods primarily rely on the analysis of single or multiple RGB images. However, in complex environments, factors such as changes in lighting, occlusion, and viewpoint can pose challenges to accurate pose estimation. To address these issues, RGB-D camera technology has gained increasing attention in recent years. RGB-D cameras can simultaneously provide RGB images and depth information, enabling more accurate acquisition of the three-dimensional keypoint positions of the human body for pose estimation.

[0004] While the 2D pixel coordinates of human keypoints can be obtained using mature 2D human pose estimation algorithms, the final robot planning and control requires the 3D coordinates of these keypoints in the camera or world coordinate system. However, current 3D human pose estimation methods still face several challenges in injury cluster scenarios. First, the target injured persons in injury cluster scenarios are mostly in a lying position, but there are relatively few lying-positioned individuals in human pose datasets, resulting in poor detection performance of deep learning-based 3D human pose estimation algorithms for lying-positioned individuals. Second, injury cluster scenarios are often in outdoor environments, and changes in lighting can affect image quality, necessitating solutions to the robustness of algorithms in outdoor scenes. Furthermore, the real-time performance and computational efficiency of the algorithms must be considered, especially since rescue robots at injury cluster scenarios have limited computing resources, making it difficult for existing methods to meet real-time requirements.

[0005] Therefore, in response to the need for efficient 3D human body key point detection at injury points, this invention proposes a new technical solution that aims to overcome the limitations of existing methods and provide a fast and accurate 3D human body key point detection method. Summary of the Invention

[0006] This invention provides a method for detecting three-dimensional human key points based on an RGB-D camera in scenarios with concentrated injury points, aiming to at least solve one of the technical problems existing in the prior art.

[0007] The technical solution of this invention relates to a three-dimensional human body key point detection method, which includes the following steps:

[0008] S100. Obtain the RGB image of the target human body to obtain the two-dimensional pixel coordinates of the human body key points through a deep learning-based two-dimensional human pose estimation algorithm.

[0009] S200. Obtain a depth image of the target human body to obtain the depth values ​​corresponding to the key points of the human body.

[0010] S300. Optimize the two-dimensional pixel coordinates using an adaptive mean filter;

[0011] S400: Obtain the three-dimensional coordinates of key human body points in the camera coordinate system through coordinate transformation.

[0012] Furthermore, the deep learning-based two-dimensional human pose estimation algorithm is expressed as follows:

[0013] O=F(I RGB )

[0014]

[0015] O = {o1, o2, ..., o} J}

[0016]

[0017]

[0018]

[0019] In the formula, O represents the set of J key points detected from the input RGB image, and p j This represents the pixel coordinates of the j-th keypoint in the image plane coordinate system. This represents the confidence level of the j-th key point;

[0020] Among them, the RGB image I with width w and height h RGB This is the input to the two-dimensional human pose estimation algorithm;

[0021] Where p(u,v) represents the two-dimensional pixel coordinates of the human body key points in the image plane coordinate system.

[0022] Furthermore, the adaptive mean filter smooths the detection results and removes erroneous points based on the depth value of the depth image and the confidence level of human key points.

[0023] Furthermore, the filtering algorithm of the adaptive mean filter is expressed as follows:

[0024]

[0025]

[0026] 0 < threshold < 1

[0027]

[0028] 0 < w N-m <w N-m+1 <...<w N <1

[0029] In the formula, This represents the pixel coordinates of the j-th keypoint in the N-th frame of the image plane coordinate system. This represents the depth value of the j-th keypoint in the t-th frame of the depth image. The confidence level of the j-th keypoint in frame t is represented by m, where m represents the filter window size and δ is the keypoint value. t w indicates whether the detection data of the j-th person's key points in the t-th frame is valid. t This represents the weights that can be adjusted adaptively.

[0030] Furthermore, in step S400, the coordinate transformation converts the two-dimensional pixel coordinates of the human body key points into three-dimensional coordinates in the camera coordinate system based on the depth value of the depth image.

[0031] Furthermore, the two-dimensional pixel coordinates p of the key points of the human body j (u j ,v j ) to the three-dimensional coordinates in the camera coordinate system The coordinate transformation formula is as follows:

[0032]

[0033]

[0034]

[0035] In the formula, dj =D(u) j ,v j ) represents point p j (u j ,v j The depth value at () is where u0 and v0 are internal parameters of the camera.

[0036] Furthermore, step S400 includes the following steps:

[0037] S410. Establish the coordinate transformation equation from the physical coordinates p(x,y) of the human body key points in the image plane coordinate system to the pixel coordinates p(u,v) in the image plane coordinate system, which is expressed as follows:

[0038]

[0039] In the formula, d x This represents the effective distance between two adjacent pixels in the x-direction of a computer image. d y This represents the effective distance between two adjacent pixels in the y-direction of a computer image.

[0040] S420. Based on the camera pinhole model, establish three-dimensional camera coordinates p in the camera coordinate system. c (x c ,y c ,z c The coordinate transformation equation between the coordinates p(x,y) and the two-dimensional physical coordinates p(x,y) is expressed as follows:

[0041]

[0042] In the formula, f represents the effective focal length of the camera.

[0043] S430, Establish 3D camera coordinates p c (x c ,y c ,z c The coordinate transformation equation between the two-dimensional pixel coordinates p(u,v) and the two-dimensional pixel coordinates p(u,v) is expressed as follows:

[0044]

[0045]

[0046] In the formula, f x f y u0 and v0 are the camera's internal parameters.

[0047] S440, Establish from camera coordinate system p c (x c ,y c ,z c ) to world coordinate system p w (x w ,y w ,z w The coordinate transformation equation of () is expressed as follows:

[0048]

[0049] In the formula, R∈R 3×3 Let be a rotation matrix. It is a translation vector.

[0050] S440, Establish three-dimensional world coordinates p w (x w ,y w ,z w The coordinate transformation equation from the two-dimensional pixel coordinates p(u,v) to p(u,v) is expressed as follows:

[0051]

[0052]

[0053] In the formula, K1 is the internal parameter matrix and K2 is the external parameter matrix.

[0054] Furthermore, the key points of the human body include the head, arms, wrists, legs, and ankles.

[0055] The present invention also relates to a computer-readable storage medium having program instructions stored thereon, which, when executed by a processor, implement the above-described method.

[0056] The technical solution of the present invention also relates to a three-dimensional human body key point detection system, the system including a computer device, the computer device including the aforementioned computer-readable storage medium.

[0057] The beneficial effects of this invention are as follows:

[0058] This invention presents a 3D human keypoint detection method based on an RGB-D camera in injury-prone scenarios. This method enables rapid and accurate detection and acquisition of 3D human keypoint coordinates in such scenarios, providing information support for rescue operations and helping robots operate and execute rescue tasks more precisely. An adaptive mean filter is introduced to optimize the detection results. This filter automatically adjusts weights, eliminates erroneous points, and smooths the data, thereby reducing noise interference and improving the accuracy and stability of human keypoint detection. Compared with traditional fixed-weight filters, the adaptive mean filter designed in this invention can better adapt to different depth values, providing more reliable detection results. The method of this invention has high real-time performance and accuracy, enabling efficient 3D human keypoint detection using an RGB-D camera. By optimizing the 2D human pose estimation algorithm, data from the RGB-D camera can be processed in a short time, providing accurate human keypoint detection results in real time. This facilitates the robot's timely perception and response to changes in the posture and position of the injured, thus ensuring the stability of the control system. This invention enables robots to provide real-time coordinates of key points on the injured person's body, supporting robots in making rescue decisions and taking actions at the injury site, thereby improving rescue efficiency. Furthermore, by accurately sensing the location of key points on the injured person's body, it facilitates more precise operation of the robot, such as the operation of equipping the injured person with triage sensors. Attached Figure Description

[0059] Figure 1 is a basic flowchart of the three-dimensional human body key point detection method according to the present invention.

[0060] Figure 2 is a filtering effect diagram of the adaptive mean filter according to the present invention.

[0061] Figure 3 is a schematic diagram of the camera coordinate system according to the method of the present invention.

[0062] Figure 4 is a schematic diagram of the detection results according to the method of the present invention.

[0063] Figure 5 is a schematic diagram of the detection speed according to the method of the present invention. Detailed Implementation

[0064] The following will provide a clear and complete description of the concept, specific structure, and technical effects of the present invention in conjunction with the embodiments and accompanying drawings, so as to fully understand the purpose, solution, and effects of the present invention.

[0065] It should be noted that, unless otherwise specified, when a feature is referred to as "fixed" or "connected" to another feature, it can be directly fixed or connected to the other feature, or indirectly fixed or connected to the other feature. The singular forms "a," "described," and "the" used herein are also intended to include the plural forms, unless the context clearly indicates otherwise. Furthermore, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing particular embodiments only and not for limiting the invention. The term "and / or" as used herein includes any combination of one or more of the associated listed items.

[0066] It should be understood that although the terms first, second, third, etc., may be used in this disclosure to describe various elements, these elements should not be limited to these terms. These terms are only used to distinguish elements of the same type from one another. For example, a first element may also be referred to as a second element without departing from the scope of this disclosure, and similarly, a second element may also be referred to as a first element. Any and all instances or exemplary language (“e.g.,” “such as,” etc.) provided herein are intended only to better illustrate embodiments of the invention and, unless otherwise required, do not impose a limitation on the scope of the invention.

[0067] Referring to Figures 1 to 3, in some embodiments, the three-dimensional human key point detection method according to the present invention includes at least the following steps:

[0068] 100. Obtain the RGB image of the target human body to obtain the two-dimensional pixel coordinates of the human body key points through a deep learning-based two-dimensional human pose estimation algorithm;

[0069] S200. Obtain a depth image of the target human body to obtain the depth values ​​corresponding to the key points of the human body.

[0070] S300. Optimize the two-dimensional pixel coordinates using an adaptive mean filter;

[0071] S400: Obtain the three-dimensional coordinates of key human body points in the camera coordinate system through coordinate transformation.

[0072] This invention provides a method for detecting 3D human key points based on an RGB-D camera in the context of injury clusters, aiming to solve the problem that traditional methods struggle to quickly and accurately obtain 3D human key points in the complex environment of post-disaster injury clusters. The method described in this invention can quickly and accurately detect and obtain the coordinates of 3D human key points, including key points such as the head, wrists, and ankles, in injury cluster scenarios. This provides information support for rescue work and helps robots operate and perform rescue tasks more precisely. For example, it allows for adjustments to the robot's posture, pose, or movements to adapt to the rescue environment and effectively interact with the injured, improving the efficiency and safety of rescue operations.

[0073] Detailed Implementation of Step S100

[0074] This invention presents a method for detecting key points in a 3D human body based on an RGB-D camera. It utilizes an RGB-D camera to acquire data containing both color and depth information, and employs a deep learning algorithm to detect key points on the injured person's 3D body. This method acquires RGB (color) and depth images of the injured target's body using an RGB-D camera. The information from these images allows for precise identification and location of key points on the body, such as the head, arms, and legs, helping rescue personnel and rescue robots quickly understand the injured person's condition and thus improving the efficiency of rescue operations.

[0075] Specifically, an RGB-D camera is connected to a computer to acquire a color image I with a height of h and a width of w. RGB A deep learning-based two-dimensional (2D) human pose estimation algorithm trained on the COCO dataset is used to detect RGB images and obtain the two-dimensional pixel coordinates of human key points in the image. In this embodiment of the invention, the input to the two-dimensional human pose estimation algorithm is a color image I with width w and height h. RGB The algorithm's output includes a set O of detected keypoint coordinates and confidence scores. The two-dimensional human pose estimation algorithm is expressed as:

[0076] O=F(I RGB (1)

[0077]

[0078] O = {o1, o2, ..., o} J} (3)

[0079]

[0080]

[0081]

[0082] In the formula, O represents the set of J keypoints detected from the input image, and p j This represents the pixel coordinates of the j-th keypoint in the image plane coordinate system. Let I represent the confidence level of the j-th keypoint. RGB The image represents a three-dimensional matrix (h, w, 3) of a red, green, and blue three-channel color image acquired by a camera. Here, h and w represent the height and width of the image, respectively. Each pixel is composed of the intensity values ​​of the three color channels (red, green, and blue), and these three values ​​determine the color of the pixel.

[0083] The method based on an RGB-D camera in this invention can be applied to fields such as human-computer interaction, posture analysis, and action recognition, and has broad application prospects for intelligent equipment, virtual reality, and medical rehabilitation.

[0084] Detailed Implementation of Step S200

[0085] The method of this invention connects an RGB-D camera to a computer to acquire a depth image of the injured target human body, and then obtains the depth value z corresponding to each human body key point from the depth image aligned with the color image. j =D(u) j ,v j It should be noted that due to the influence of environmental and sensor noise, the depth values ​​corresponding to some key points may be invalid and require further processing.

[0086] Detailed Implementation of Step S300

[0087] The method of this invention employs a designed adaptive mean filter to further process the obtained pixel coordinates p(u,v) of human key points in the image plane coordinate system. This addresses the problem of camera shake caused by robot movement affecting detection results when the RGB-D camera is mounted on a robot. The method smooths the detection results using a self-designed adaptive mean filter and removes invalid depth points or erroneous points with confidence levels below a threshold.

[0088] Specifically, Let represent the pixel coordinates of the j-th keypoint in the N-th frame of the image plane coordinate system. The adaptive mean filter is set as follows:

[0089]

[0090]

[0091] 0 < threshold < 1 (9)

[0092]

[0093] 0 < w N-m <w N-m+1 <...<w N <1 (11)

[0094] in, This represents the depth value of the j-th keypoint in the t-th frame of the depth image. Let m represent the confidence level of the j-th person's keypoint in frame t, and let δ represent the filter window size. t w indicates whether the detection data of the j-th person's key points in the t-th frame is valid. t This represents the weights that can be adjusted adaptively.

[0095] The method in this embodiment of the invention uses an adaptive mean filter to smooth the detection results and remove erroneous points. The adaptive mean filter is an extension of the weighted mean filter by introducing δ... t And adaptively adjust the weight w t This eliminates erroneous key points and improves the detection performance of the algorithm. Specifically, the effect of the adaptive mean filter is shown in Figure 2. The adaptive mean filter can effectively improve the detection performance. Even if the nose is not a target key point in the image, its filtering algorithm can still use the detection results of the previous few frames to achieve effective estimation.

[0096] Detailed Implementation of Step S400

[0097] The method in this invention obtains the three-dimensional coordinates of key human body points in the camera coordinate system through coordinate transformation, such as the head, wrist, and ankle, providing necessary conditions and references for robot rescue operations. See Figure 3, p w (x w ,y w ,z w p represents a point in the world coordinate system. c (x c ,y c ,z c p(x,y) represents the 3D camera coordinates of the point in the camera coordinate system, p(x,y) represents the 2D physical coordinates of the point in the image plane coordinate system, and p(u,v) represents the 2D pixel coordinates of the point in the image plane coordinate system.

[0098] Specifically, d x d represents the effective distance (mm / pixel) between two adjacent pixels in the horizontal (x-direction) of a computer image. yLet p(x,y) represent the effective distance (mm / pixel) between two adjacent pixels in the vertical direction (y-direction) of a computer image. The coordinate transformation from two-dimensional physical coordinates p(x,y) to two-dimensional pixel coordinates p(u,v) in the image plane coordinate system is obtained by the following formula:

[0099]

[0100] Based on the camera pinhole model, establish the 3D camera coordinate system p. c (x c ,y c ,z c The coordinate transformation equation between the coordinates p(x,y) and the two-dimensional physical coordinates p(x,y) is expressed as follows:

[0101]

[0102] In the formula, f represents the effective focal length of the camera.

[0103] 3D camera coordinates p c (x c ,y c ,z c The coordinate transformation relationship between the two-dimensional pixel coordinates p(u,v) and the two-dimensional pixel coordinates p(u,v) is as follows:

[0104]

[0105]

[0106] In the formula, f x f y u0 and v0 are the internal parameters of the camera.

[0107] The coordinate transformation from the camera coordinate system to the world coordinate system is a rigid body transformation, which is represented as follows:

[0108]

[0109] In the formula, R∈R 3×3 Let be a rotation matrix. This is the translation vector between the world coordinate system and the camera coordinate system.

[0110] Finally, the coordinate transformation relationship between three-dimensional world coordinates and two-dimensional pixel coordinates is given by the following formula:

[0111]

[0112]

[0113] In the formula, K1 is the internal parameter matrix and K2 is the external parameter matrix.

[0114] The method in this embodiment of the invention uses depth images to perform coordinate transformation, obtaining the three-dimensional coordinates of key points on the human body. Specifically, d j =D(u) j ,v j ) represents point p j (u j ,v j The depth value at (). (From p) j (u j ,v j )arrive The coordinate transformation formula is expressed as follows:

[0115]

[0116]

[0117]

[0118] In the formula, u0 and v0 are both internal parameters of the camera.

[0119] It should be noted that the coordinate p w (x w ,y w ,z w It is related to the choice of the world coordinate system, which is easily obtained in actual robot systems.

[0120] This invention presents a practical test of a 3D human keypoint detection method based on an RGB-D camera in a scenario with concentrated injury points. Referring to Figure 4, the 3D human keypoint detection results are shown; the 3D coordinates of the human keypoints in the camera coordinate system are represented by red text, and the confidence level of the human keypoints is represented by green text. Referring to Figure 5, the 3D human keypoint detection speed is shown; the average detection speed is approximately 17 FPS, which meets the requirements of the robot control system. This invention, in a scenario with concentrated injury points, combines color images and depth information from an RGB-D camera to achieve the detection and localization of human keypoints. It can quickly and accurately obtain the 3D coordinates of key points such as the head, wrists, and ankles, and its application in rescue robot systems can improve rescue efficiency.

[0121] The three-dimensional human key point detection results of the method in this invention provide necessary information support for rescue robots. The robot can utilize the three-dimensional human key point detection results to conduct effective rescue operations at disaster sites. First, by analyzing the posture information of the injured, the robot can assess their condition. Second, based on the key point information, the robot can attach triage sensors such as nose clips, wristbands, and ankle bracelets to the injured to monitor their vital signs and physical condition. Furthermore, based on the key point detection results, the robot can also plan the optimal movement path, improving the efficiency and safety of rescue work.

[0122] It should be understood that the method steps in the embodiments of the present invention can be implemented or carried out by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable storage medium. The method can use standard programming techniques. Each program can be implemented in a high-level procedural or object-oriented programming language to communicate with the computer system. However, if necessary, the program can be implemented in assembly or machine language. In any case, the language can be a compiled or interpreted language. Furthermore, for this purpose, the program can run on a programmed application-specific integrated circuit (ASIC).

[0123] Furthermore, the procedures described herein may be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by the context. The procedures described herein (or variations and / or combinations thereof) may be executed under the control of one or more computer systems configured with executable instructions, and may be implemented by hardware or a combination thereof as code (e.g., executable instructions, one or more computer programs, or one or more applications) that commonly executes on one or more processors. The computer program comprises a plurality of instructions executable by one or more processors.

[0124] Furthermore, the method can be implemented in any suitable type of computing platform, including but not limited to personal computers, minicomputers, mainframes, workstations, networked or distributed computing environments, standalone or integrated computer platforms, or in communication with charged particle tools or other imaging devices, etc. Aspects of the invention can be implemented as machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard disk, optical read and / or write storage medium, RSM, ROM, etc., such that it can be read by a programmable computer, and when the storage medium or device is read by the computer, it can be used to configure and operate the computer to perform the processes described herein. Furthermore, the machine-readable code, or portions thereof, can be transmitted via wired or wireless networks. The invention described herein includes these and other different types of non-transitory computer-readable storage media when such media comprises instructions or programs that implement the steps described above in conjunction with a microprocessor or other data processor. When programmed according to the methods and techniques described in the invention, the invention may also include the computer itself.

[0125] A computer program can be applied to input data to perform the functions described herein, thereby transforming the input data to generate output data stored in non-volatile memory. The output information can also be applied to one or more output devices, such as a display. In a preferred embodiment of the invention, the transformed data represents physical and tangible objects, including specific visual depictions of physical and tangible objects generated on the display.

[0126] The above description is merely a preferred embodiment of the present invention. The present invention is not limited to the above-described embodiments. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention, as long as they achieve the technical effects of the present invention by the same means, should be included within the scope of protection of the present invention. Within the scope of protection of the present invention, the technical solutions and / or implementation methods can have various modifications and variations.

Claims

1. A method for detecting key points in a three-dimensional human body, characterized in that, The method includes the following steps: S100, acquiring an RGB image of the target human body to obtain the two-dimensional pixel coordinates of human body key points through a deep learning-based two-dimensional human pose estimation algorithm; S200, acquiring a depth image of the target human body to obtain the depth values ​​corresponding to the human body key points; S300, optimizing the two-dimensional pixel coordinates through an adaptive mean filter; S400, obtaining the three-dimensional coordinates of the human body key points in the camera coordinate system through coordinate transformation; wherein, the adaptive mean filter smooths the detection results and removes erroneous points based on the depth values ​​of the depth image and the confidence level of the human body key points; the filtering algorithm of the adaptive mean filter is expressed as follows: In the formula, This represents the pixel coordinates of the j-th keypoint in the N-th frame of the image plane coordinate system. This represents the depth value of the j-th keypoint in the t-th frame of the depth image. This represents the confidence level of the j-th keypoint in the t-th frame. Indicates the size of the filter window. This indicates whether the detection data of the j-th person's key points in the t-th frame is valid. This represents the weight of the t-th frame that can be adaptively adjusted.

2. The method according to claim 1, characterized in that, The deep learning-based two-dimensional human pose estimation algorithm is represented as follows: In the formula, This represents the set of J keypoints detected from the input RGB image. This represents the pixel coordinates of the j-th keypoint in the image plane coordinate system. This represents the confidence level of the j-th key point; This represents a three-channel (red, green, blue) color image acquired by a camera with dimensions (h, w, 3); where the width is w and the height is h. is the input to the two-dimensional human pose estimation algorithm; where, It represents the two-dimensional pixel coordinates of key points of the human body in the image plane coordinate system.

3. The method according to claim 1, characterized in that, In step S400, the coordinate transformation converts the two-dimensional pixel coordinates of the human body key points into three-dimensional coordinates in the camera coordinate system based on the depth value of the depth image.

4. The method according to claim 3, characterized in that, The two-dimensional pixel coordinates of key points of the human body To the three-dimensional coordinates in the camera coordinate system The coordinate transformation formula is as follows: In the formula, Point The depth value at that location, where 。 5. The method according to claim 4, characterized in that, Step S400 includes the following steps: S410, establishing the physical coordinates of the human body key points in the image plane coordinate system. to pixel coordinates The coordinate transformation equation is expressed as follows: In the formula, This represents the effective distance between two adjacent pixels in the x-direction of a computer image. This represents the effective distance between two adjacent pixels in the y-direction of a computer image; S420, based on the camera pinhole model, establishes three-dimensional camera coordinates in the camera coordinate system. With two-dimensional physical coordinates The coordinate transformation equation between them is expressed as follows: In the formula, f represents the effective focal length of the camera; S430, establish the three-dimensional camera coordinates. with two-dimensional pixel coordinates The coordinate transformation equation between them is expressed as follows: In the formula, ， ， ， For camera internal parameters; S440, establish the camera coordinate system To the world coordinate system The coordinate transformation equation is expressed as follows: In the formula, For rotation matrix, For translation vectors; S450, establish three-dimensional world coordinates. to two-dimensional pixel coordinates The coordinate transformation equation is expressed as follows: In the formula, For the internal parameter matrix, This is the external parameter matrix.

6. The method according to claim 1, characterized in that, The key points of the human body include the head, arms, wrists, legs, and ankles.

7. A computer-readable storage medium having stored thereon program instructions that, when executed by a processor, perform the method as described in any one of claims 1 to 6.

8. A three-dimensional human body key point detection system, characterized in that, include: A computer device, the computer device comprising the computer-readable storage medium according to claim 7.

Images