Hand modeling method based on key point positioning

Abstract

The invention relates to the field of modeling, in particular to a hand modeling method based on key point positioning. According to the technical problems, a method for collecting and recognizing images of the hand of the human body through a set of image collecting devices is adopted, data of the hand of the human body is collected, on one hand, the images of the hand cannot be completely collected due to angle limitation, and on the other hand, when the hand moves intensely, the situation of detection errors occurs, and therefore the detection accuracy of the hand of the human body is improved. Human hand modeling errors are caused; according to the technical scheme, the hand modeling method based on key point positioning comprises the steps that feature recognition is conducted on a human hand contour image; according to the invention, multiple groups of image acquisition modules are arranged to simultaneously acquire and identify the key point images of the human hand, multi-angle acquisition can be performed on the key point images of the human hand, the integrity of human hand image acquisition is ensured, and the error of key point identification can be eliminated by eliminating outliers, so that the accuracy of key point identification is improved. And the accuracy of data acquisition and identification is ensured.

Description

Technical Field

[0001] The present invention relates to the field of modeling, and in particular to a hand modeling method based on key point positioning. Background Art

[0002] Hand modeling is a digital technology that creates and manipulates three-dimensional hand models using computer software and hardware; it can be used in a variety of application areas, such as medicine, engineering, and entertainment; through hand modeling, we can analyze hand movements and bone structure, as well as simulate hand movement and deformation; this technology helps to study the anatomical structure and function of the hand, and can also be applied to virtual reality, animation production, and game development; the process of hand modeling usually includes data acquisition, data processing, and model generation; first, we use sensors or cameras to capture the movement and shape of the hand; then, we process the collected data to remove noise and correct errors; finally Finally, we use modeling software to convert the processed data into a visual hand model; the development of hand modeling technology enables us to better understand and simulate the movement and morphological changes of the hand; when modeling the human hand through image data acquisition, a group of image acquisition devices are used to collect images of the human hand, and then image recognition and detection methods are used to realize the collection of data on the human hand. On the one hand, due to the angle limitation of the image acquisition device, the image of the human hand cannot be completely collected, resulting in modeling failure. On the other hand, when the human hand moves more violently, detection errors will occur, resulting in errors in modeling the human hand. Summary of the invention

[0003] In order to overcome the problem that when modeling the human hand through image data acquisition, a group of image acquisition devices are used to acquire images of the human hand, and then image recognition and detection are performed to realize the acquisition of data on the human hand. On the one hand, due to the angle limitation of the image acquisition device, the image of the human hand cannot be completely acquired, resulting in modeling failure. On the other hand, when the human hand moves more violently, detection errors will occur, resulting in incorrect modeling of the human hand.

[0004] The technical solution of the present invention is: a hand modeling method based on key point positioning, comprising the following steps:

[0005] S11: collecting images of human hands through image collection devices, wherein when collecting images of human hands, the image collection devices are arranged in multiple groups, and the image collection devices are arranged in different directions of the human hands respectively;

[0006] S12: Processing the human hand image acquired by the image acquisition device, segmenting the human hand image, removing the background image, and obtaining a human hand contour image;

[0007] S13: performing feature recognition on the human hand contour image to identify key point images in the human hand contour image, wherein the key point image is an image of manually set key point positions of the human hand;

[0008] S14: Establishing a three-dimensional coordinate system, and calculating the coordinates of the key points in the three-dimensional coordinate system according to the key point recognition results and the acquisition results of the image acquisition device;

[0009] S15: Building a human hand model according to the calculated coordinates of the key points in the three-dimensional coordinate system.

[0010] Preferably, when the image of the human hand is captured by the image capture device, the image capture device includes a binocular camera and a fill light, wherein the binocular camera and the fill light correspond one to one, and when the image of the human hand is captured by the image capture device, it also includes analyzing the light intensity of the image of the human hand captured by the binocular camera, and controlling the fill light according to the analysis result.

[0011] Preferably, when analyzing the light intensity by the image of the human hand collected by the binocular camera and controlling the fill light according to the analysis result, the following steps are included:

[0012] S21: grayscale the collected image of the human hand, and calculate the average grayscale value of all pixels in the image of the human hand;

[0013] S22: Calculating the fill light power required by the fill light according to the average gray value of all pixels in the human hand image;

[0014] S23: adjusting the fill light power of the fill light according to the calculation result, so that the average gray value of all pixels in the human hand image is within a set threshold range.

[0015] Preferably, by setting a fill light, the binocular camera can be filled with light when the light intensity is insufficient, thereby ensuring the image acquisition quality and effect of the binocular camera, and by setting an image brightness detection algorithm, the required fill light power can be quickly and automatically calculated. On the one hand, the fill light effect can be effectively guaranteed, and the clarity and discernibility of the image acquisition can be guaranteed, and automatic power adjustment can be achieved. While ensuring energy saving, there is no need for manual control of the fill light, thereby increasing the efficiency of image acquisition.

[0016] Preferably, when processing the human hand image acquired by the image acquisition device, segmenting the human hand image, removing the background image, and obtaining the human hand contour image, the following steps are included:

[0017] S31: grayscale the human hand image obtained by the binocular camera to obtain a grayscale image of the human hand, wherein the principle formula of the image grayscale processing is: Y=0.299×R+0.587×G+0.144×B; wherein Y is the brightness of the grayscale image after conversion, that is, the grayscale value, and R, G and B are the values ​​of the three primary colors of red, green and blue of the color image before conversion respectively;

[0018] S32: using an image segmentation method based on an adaptive threshold, segmenting the background region in the human hand image and the human hand region image;

[0019] S33: Using a feature recognition algorithm based on color features, the background area image in the segmented image is detected, and the background area image is removed, and the remaining image is saved as a human hand area image.

[0020] Preferably, when performing feature recognition on a human hand contour image and identifying key point images in the human hand contour image, the identified key points include multiple groups of wrist position points, palm position points, back of hand position points, phalangeal joint position points and fingertip position points, among which the wrist position points and fingertip position points are required key points, and the palm position points and back of hand position points are automatically identified and selected according to the image angle.

[0021] Preferably, when performing feature recognition on the human hand contour image and identifying the key point image in the human hand contour image, the following steps are included:

[0022] S41: using a feature point detection algorithm based on a recurrent neural network to detect wrist position points in the human hand contour image, and marking the wrist position points after the detection is completed;

[0023] S42: Detecting the human hand contour image using a feature point detection algorithm based on a recurrent neural network to detect whether there is an image of a palm position point or a back of the hand position point in the human hand contour image;

[0024] S43: marking the detected palm position point or the back of the hand position point, and binding the wrist position point with the palm position point or the back of the hand position point;

[0025] S44: Detect the remaining key point positions using a feature point detection algorithm based on a recurrent neural network, and bind them to the palm position points or the back of the hand position points.

[0026] Preferably, when the remaining key point positions are detected by using a feature point detection algorithm based on a recurrent neural network and bound to the palm position points or the back of the hand position points, the following steps are included:

[0027] S51: Connect the wrist position point and the palm position point to establish the basic skeleton of the human hand;

[0028] S52: inferring the range of the location points of the phalangeal joints at the bases of the fingers according to the basic skeleton of the human hand;

[0029] S53: Detect the position points of the phalangeal joints at the base of the finger using a feature point detection algorithm based on a recurrent neural network, and compare them with the results in step S52 to determine whether the position points of the phalangeal joints at the base of the finger are reasonable;

[0030] S54: If the rationality meets the standard, proceed to the next step; if the rationality does not meet the standard, re-identification and detection are performed;

[0031] S55: Continue to identify and detect the rationality of the next joint point until all key point positions are identified and detected.

[0032] Preferably, when detecting key points of the human hand, the positions of the remaining key points are restricted by constructing a human skeleton with wrist position points and palm position points. This can prevent confusion in the recognition of similar key points when identifying key points using a key point recognition algorithm based on image features due to the human hand being in motion or other reasons, thereby causing modeling errors. This ensures the accuracy of key point detection and makes hand modeling more precise.

[0033] Preferably, when establishing a three-dimensional coordinate system, the coordinates of the key points in the three-dimensional coordinate system are calculated according to the key point recognition results and the acquisition results of the image acquisition device, including the following steps:

[0034] S61: Selecting an origin by random selection or by fixing a reference point to establish a three-dimensional coordinate system;

[0035] S62: Calculate the coordinates of each key point in the multiple groups of binocular cameras in the stereo coordinate system by performing depth detection on the images in the binocular cameras and according to the original positions of the binocular cameras;

[0036] S63: Comprehensively analyzing the coordinate results obtained by calculating the images in the multiple sets of binocular cameras, eliminating errors in the calculation process, and obtaining the comprehensive coordinates of each key point.

[0037] Preferably, when performing depth detection on the images in the binocular cameras and calculating the coordinates of each key point in the multiple groups of binocular cameras in the stereo coordinate system in conjunction with the original positions of the binocular cameras, the following steps are included:

[0038] S71: dividing the image captured by the binocular vision camera into a left-eye view and a right-eye view, and placing the pixel points corresponding to the key points of the left-eye view and the right-eye view into the same rectangular coordinate system; wherein the origin of the rectangular coordinate system corresponds to the pixel point at the same position in the left-eye view and the right-eye view;

[0039] S72: Calculate the disparity of the pixel point corresponding to each key point in the binocular vision camera according to the coordinates of the pixel points corresponding to all key points in the left eye view and the right eye view; wherein the disparity of the pixel point at each position in the binocular vision camera is calculated according to the following formula:

[0040] d=|x l -x r |,

[0041] Among them, d is the disparity of the pixel point at that position in the binocular vision camera, x l is the horizontal coordinate of the pixel point at this position in the left eye view, x r is the horizontal coordinate of the pixel point at that position in the right eye view;

[0042] S73: Calculate the depth of the pixel at the position in space according to the parallax; wherein the calculation formula is: Among them, f is the focal length of the binocular vision camera, b is the camera baseline distance;

[0043] S74: Determine the coordinates of the key points according to the original position of the binocular camera, the horizontal coordinate and the vertical coordinate of the pixel point corresponding to each key point, and the calculated depth in space.

[0044] Preferably, when comprehensively analyzing the coordinate results obtained by calculating the images in multiple sets of binocular cameras, eliminating the errors in the calculation process, and obtaining the comprehensive coordinates of each key point, the following steps are included:

[0045] S81: numbering each key point detected according to the constructed basic skeleton of the human hand, wherein key points at the same position have the same number in images captured by different binocular cameras;

[0046] S82: Counting the coordinates of key points with the same number in different binocular cameras, and calculating the average coordinates of each key point;

[0047] S83: removing the key point coordinates whose difference from the average coordinates exceeds a threshold, and retaining the remaining key point coordinates;

[0048] S84: Recalculate the average coordinates according to the coordinates of the remaining key points to obtain the comprehensive coordinates of each key point.

[0049] Preferably, by setting up multiple groups of binocular cameras, on the one hand, it can effectively prevent the situation that the complete image of the human hand cannot be fully captured due to the angle limitation of a single group of binocular cameras, resulting in the inability to accurately detect the posture of the human hand. On the other hand, the idea of ​​outlier elimination can be used to replace and delete the coordinates of key points that are incorrectly recognized to ensure the accuracy and reliability of key point positioning. By calculating the average coordinate method to calculate the comprehensive coordinates of the key points, the data errors generated during key point detection can be eliminated through the statistical averaging method to ensure the accuracy of key point positioning and increase the quality of human hand modeling.

[0050] Preferably, when establishing the human hand model according to the calculated coordinates of the key points in the three-dimensional coordinate system, the following steps are included:

[0051] S91: Fill each key point into the established three-dimensional coordinate system according to the calculated comprehensive coordinates;

[0052] S92: connecting the key points in order according to their numbers to form a human hand skeleton;

[0053] S93: Filling and optimizing the human hand skeleton according to the set data content to obtain a human hand model.

[0054] Beneficial effects of the present invention:

[0055] 1. Compared with the prior art, when modeling human hands by image data acquisition, a group of image acquisition devices are used to acquire images of human hands, and then image recognition and detection are performed to realize data acquisition of human hands. On the one hand, due to the angle limitation of the image acquisition device, the image of the human hand cannot be completely acquired, resulting in modeling failure. On the other hand, when the human hand moves more violently, detection errors may occur, resulting in errors in modeling of the human hand. The hand modeling method sets multiple groups of image acquisition modules to acquire and identify key point images of the human hand at the same time, and can acquire key point images of the human hand from multiple angles to ensure the integrity of the human hand image acquisition. It can also eliminate errors in key point recognition by eliminating outliers, thereby ensuring the accuracy and effectiveness of data acquisition and recognition.

[0056] 2. By setting up multiple groups of binocular cameras, on the one hand, it can effectively prevent the situation that the complete image of the human hand cannot be fully collected due to the angle limitation of a single group of binocular cameras, resulting in the inability to accurately detect the posture of the human hand. On the other hand, the coordinates of the key points with identification errors can be replaced and deleted through the idea of ​​outlier elimination to ensure the accuracy and reliability of key point positioning. The comprehensive coordinates of the key points can be calculated by calculating the average coordinate method, and the data errors generated during key point detection can be eliminated by the statistical averaging method to ensure the accuracy of key point positioning and increase the quality of human hand modeling.

[0057] 3. When detecting the key points of the human hand, the positions of the remaining key points are restricted by constructing the human skeleton with the wrist position points and the palm position points. This can prevent the confusion of similar key point recognition caused by the movement of the human hand or other reasons when the key points are recognized by the key point recognition algorithm based on image features, thereby preventing the occurrence of modeling errors, ensuring the accuracy of key point detection and making hand modeling more accurate;

[0058] 4. By setting the fill light, you can fill in the light for the binocular camera when the light intensity is insufficient, thereby ensuring the image acquisition quality and effect of the binocular camera. By setting the image brightness detection algorithm, you can quickly and automatically calculate the required fill light power. On the one hand, it can effectively ensure the fill light effect, ensure the clarity and discernibility of image acquisition, and realize automatic power adjustment. While ensuring energy saving, there is no need for manual control of the fill light, thereby increasing the efficiency of image acquisition. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 What is shown is a flow chart of the hand modeling method based on key point positioning of the present invention;

[0060] Figure 2 What is shown is a schematic diagram of a partial workflow of key point detection in the hand modeling method based on key point positioning of the present invention;

[0061] Figure 3 What is shown is a schematic diagram of the workflow of key point coordinate calculation in the hand modeling method based on key point positioning of the present invention. DETAILED DESCRIPTION

[0062] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0063] See also Figure 1 The present invention provides an embodiment: a hand modeling method based on key point positioning, comprising the following steps:

[0064] S11: collecting images of human hands through image collection devices, wherein when collecting images of human hands, the image collection devices are arranged in multiple groups, and the image collection devices are arranged in different directions of the human hands respectively;

[0065] S12: Processing the human hand image acquired by the image acquisition device, segmenting the human hand image, removing the background image, and obtaining a human hand contour image;

[0066] S13: performing feature recognition on the human hand contour image to identify key point images in the human hand contour image, wherein the key point image is an image of manually set key point positions of the human hand;

[0067] S14: Establishing a three-dimensional coordinate system, and calculating the coordinates of the key points in the three-dimensional coordinate system according to the key point recognition results and the acquisition results of the image acquisition device;

[0068] S15: Building a human hand model according to the calculated coordinates of the key points in the three-dimensional coordinate system.

[0069] Preferably, when the image of the human hand is captured by the image capture device, the image capture device includes a binocular camera and a fill light, wherein the binocular camera and the fill light correspond one to one, and when the image of the human hand is captured by the image capture device, it also includes analyzing the light intensity of the image of the human hand captured by the binocular camera, and controlling the fill light according to the analysis result.

[0070] Preferably, when analyzing the light intensity by the image of the human hand collected by the binocular camera and controlling the fill light according to the analysis result, the following steps are included:

[0071] S21: grayscale the collected image of the human hand, and calculate the average grayscale value of all pixels in the image of the human hand;

[0072] S22: Calculating the fill light power required by the fill light according to the average gray value of all pixels in the human hand image;

[0073] S23: adjusting the fill light power of the fill light according to the calculation result, so that the average gray value of all pixels in the human hand image is within a set threshold range.

[0074] Preferably, by setting a fill light, the binocular camera can be filled with light when the light intensity is insufficient, thereby ensuring the image acquisition quality and effect of the binocular camera, and by setting an image brightness detection algorithm, the required fill light power can be quickly and automatically calculated. On the one hand, the fill light effect can be effectively guaranteed, and the clarity and discernibility of the image acquisition can be guaranteed, and automatic power adjustment can be achieved. While ensuring energy saving, there is no need for manual control of the fill light, thereby increasing the efficiency of image acquisition.

[0075] Preferably, when processing the human hand image acquired by the image acquisition device, segmenting the human hand image, removing the background image, and obtaining the human hand contour image, the following steps are included:

[0076] S31: grayscale the human hand image obtained by the binocular camera to obtain a grayscale image of the human hand, wherein the principle formula of the image grayscale processing is: Y=0.299×R+0.587×G+0.144×B; wherein Y is the brightness of the grayscale image after conversion, that is, the grayscale value, and R, G and B are the values ​​of the three primary colors of red, green and blue of the color image before conversion respectively;

[0077] S32: using an image segmentation method based on an adaptive threshold, segmenting the background region in the human hand image and the human hand region image;

[0078] S33: Using a feature recognition algorithm based on color features, the background area image in the segmented image is detected, and the background area image is removed, and the remaining image is saved as a human hand area image.

[0079] Preferably, when performing feature recognition on a human hand contour image and identifying key point images in the human hand contour image, the identified key points include multiple groups of wrist position points, palm position points, back of hand position points, phalangeal joint position points and fingertip position points, among which the wrist position points and fingertip position points are required key points, and the palm position points and back of hand position points are automatically identified and selected according to the image angle.

[0080] See also Figure 2 In this embodiment, when performing feature recognition on a human hand contour image and identifying key point images in the human hand contour image, the following steps are included:

[0081] S41: using a feature point detection algorithm based on a recurrent neural network to detect wrist position points in the human hand contour image, and marking the wrist position points after the detection is completed;

[0082] S42: Detecting the human hand contour image using a feature point detection algorithm based on a recurrent neural network to detect whether there is an image of a palm position point or a back of the hand position point in the human hand contour image;

[0083] S43: marking the detected palm position point or the back of the hand position point, and binding the wrist position point with the palm position point or the back of the hand position point;

[0084] S44: Detect the remaining key point positions using a feature point detection algorithm based on a recurrent neural network, and bind them to the palm position points or the back of the hand position points.

[0085] Preferably, when the remaining key point positions are detected by using a feature point detection algorithm based on a recurrent neural network and bound to the palm position points or the back of the hand position points, the following steps are included:

[0086] S51: Connect the wrist position point and the palm position point to establish the basic skeleton of the human hand;

[0087] S52: inferring the range of the location points of the phalangeal joints at the bases of the fingers according to the basic skeleton of the human hand;

[0088] S53: Detect the position points of the phalangeal joints at the base of the finger using a feature point detection algorithm based on a recurrent neural network, and compare them with the results in step S52 to determine whether the position points of the phalangeal joints at the base of the finger are reasonable;

[0089] S54: If the rationality meets the standard, proceed to the next step; if the rationality does not meet the standard, re-identification and detection are performed;

[0090] S55: Continue to identify and detect the rationality of the next joint point until all key point positions are identified and detected.

[0091] Preferably, when detecting key points of the human hand, the positions of the remaining key points are restricted by constructing a human skeleton with wrist position points and palm position points. This can prevent confusion in the recognition of similar key points when identifying key points using a key point recognition algorithm based on image features due to the human hand being in motion or other reasons, thereby causing modeling errors. This ensures the accuracy of key point detection and makes hand modeling more precise.

[0092] Preferably, when establishing a three-dimensional coordinate system, the coordinates of the key points in the three-dimensional coordinate system are calculated according to the key point recognition results and the acquisition results of the image acquisition device, including the following steps:

[0093] S61: Selecting an origin by random selection or by fixing a reference point to establish a three-dimensional coordinate system;

[0094] S62: Calculate the coordinates of each key point in the multiple groups of binocular cameras in the stereo coordinate system by performing depth detection on the images in the binocular cameras and according to the original positions of the binocular cameras;

[0095] S63: Comprehensively analyzing the coordinate results obtained by calculating the images in the multiple sets of binocular cameras, eliminating errors in the calculation process, and obtaining the comprehensive coordinates of each key point.

[0096] See also Figure 3 In this embodiment, when performing depth detection on the images in the binocular cameras and calculating the coordinates of each key point in the multiple groups of binocular cameras in the stereo coordinate system in conjunction with the original positions of the binocular cameras, the following steps are included:

[0097] S71: dividing the image captured by the binocular vision camera into a left-eye view and a right-eye view, and placing the pixel points corresponding to the key points of the left-eye view and the right-eye view into the same rectangular coordinate system; wherein the origin of the rectangular coordinate system corresponds to the pixel point at the same position in the left-eye view and the right-eye view;

[0098] S72: Calculate the disparity of the pixel point corresponding to each key point in the binocular vision camera according to the coordinates of the pixel points corresponding to all key points in the left eye view and the right eye view; wherein the disparity of the pixel point at each position in the binocular vision camera is calculated according to the following formula:

[0099] d=|x l -x r |,

[0100] Among them, d is the disparity of the pixel point at that position in the binocular vision camera, x l is the horizontal coordinate of the pixel point at this position in the left eye view, x r is the horizontal coordinate of the pixel point at that position in the right eye view;

[0101] S73: Calculate the depth of the pixel at the position in space according to the parallax; wherein the calculation formula is: Among them, f is the focal length of the binocular vision camera, b is the camera baseline distance;

[0102] S74: Determine the coordinates of the key points according to the original position of the binocular camera, the horizontal coordinate and the vertical coordinate of the pixel point corresponding to each key point, and the calculated depth in space.

[0103] Preferably, when comprehensively analyzing the coordinate results obtained by calculating the images in multiple sets of binocular cameras, eliminating the errors in the calculation process, and obtaining the comprehensive coordinates of each key point, the following steps are included:

[0104] S81: numbering each key point detected according to the constructed basic skeleton of the human hand, wherein key points at the same position have the same number in images captured by different binocular cameras;

[0105] S82: Counting the coordinates of key points with the same number in different binocular cameras, and calculating the average coordinates of each key point;

[0106] S83: removing the key point coordinates whose difference from the average coordinates exceeds a threshold, and retaining the remaining key point coordinates;

[0107] S84: Recalculate the average coordinates according to the coordinates of the remaining key points to obtain the comprehensive coordinates of each key point.

[0108] Preferably, by setting up multiple groups of binocular cameras, on the one hand, it can effectively prevent the situation that the complete image of the human hand cannot be fully captured due to the angle limitation of a single group of binocular cameras, resulting in the inability to accurately detect the posture of the human hand. On the other hand, the idea of ​​outlier elimination can be used to replace and delete the coordinates of key points that are incorrectly recognized to ensure the accuracy and reliability of key point positioning. By calculating the average coordinate method to calculate the comprehensive coordinates of the key points, the data errors generated during key point detection can be eliminated by the statistical averaging method to ensure the accuracy of key point positioning and increase the quality of human hand modeling.

[0109] Preferably, when establishing the human hand model according to the calculated coordinates of the key points in the three-dimensional coordinate system, the following steps are included:

[0110] S91: Fill each key point into the established three-dimensional coordinate system according to the calculated comprehensive coordinates;

[0111] S92: connecting the key points in order according to their numbers to form a human hand skeleton;

[0112] S93: Filling and optimizing the human hand skeleton according to the set data content to obtain a human hand model.

[0113] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of those skilled in the art without departing from the spirit of the present invention.

Claims

1. A hand modeling method based on key point positioning; characterized in that: The following steps are included: S11: collecting images of human hands through image collection devices, wherein when collecting images of human hands, the image collection devices are arranged in multiple groups, and the image collection devices are arranged in different directions of the human hands respectively; S12: Processing the human hand image acquired by the image acquisition device, segmenting the human hand image, removing the background image, and obtaining a human hand contour image; S13: performing feature recognition on the human hand contour image to identify key point images in the human hand contour image, wherein the key point image is an image of manually set key point positions of the human hand; S14: Establishing a three-dimensional coordinate system, and calculating the coordinates of the key points in the three-dimensional coordinate system according to the key point recognition results and the acquisition results of the image acquisition device; S15: Building a human hand model according to the calculated coordinates of the key points in the three-dimensional coordinate system.

2. The hand modeling method based on key point positioning according to claim 1, characterized in that: When the image of the human hand is captured by the image capture device, the image capture device includes a binocular camera and a fill light, wherein the binocular camera corresponds to the fill light one-to-one, and when the image of the human hand is captured by the image capture device, it also includes analyzing the light intensity of the image of the human hand captured by the binocular camera, and controlling the fill light according to the analysis result.

3. The hand modeling method based on key point positioning according to claim 2 is characterized in that: When analyzing the light intensity by using the image of the human hand collected by the binocular camera and controlling the fill light according to the analysis result, the following steps are included: S21: grayscale the collected image of the human hand, and calculate the average grayscale value of all pixels in the image of the human hand; S22: Calculating the fill light power required by the fill light according to the average gray value of all pixels in the human hand image; S23: adjusting the fill light power of the fill light according to the calculation result, so that the average gray value of all pixels in the human hand image is within a set threshold range.

4. A hand modeling method based on key point positioning according to claim 3, characterized in that:. When processing the human hand image collected by the image collection device, segmenting the human hand image, removing the background image, and obtaining the human hand contour image, the following steps are included: S31: grayscale the human hand image obtained by the binocular camera to obtain a grayscale image of the human hand, wherein the principle formula of the image grayscale processing is: Y=0.299×R+0.587×G+0.144×B; wherein Y is the brightness of the grayscale image after conversion, that is, the grayscale value, and R, G and B are the values ​​of the three primary colors of red, green and blue of the color image before conversion respectively; S32: using an image segmentation method based on an adaptive threshold, segmenting the background region in the human hand image and the human hand region image; S33: Using a feature recognition algorithm based on color features, the background area image in the segmented image is detected, and the background area image is removed, and the remaining image is saved as a human hand area image.

5. The hand modeling method based on key point positioning according to claim 4 is characterized in that: When performing feature recognition on a human hand contour image and identifying key point images in the human hand contour image, the identified key points include multiple groups of wrist position points, palm position points, back of hand position points, phalangeal joint position points and fingertip position points, among which the wrist position points and fingertip position points are required key points, and the palm position points and back of hand position points are automatically identified and selected according to the image angle.

6. The hand modeling method based on key point positioning according to claim 5, characterized in that: When performing feature recognition on a human hand contour image and identifying key point images in the human hand contour image, the following steps are included: S41: using a feature point detection algorithm based on a recurrent neural network to detect wrist position points in the human hand contour image, and marking the wrist position points after the detection is completed; S42: Detecting the human hand contour image using a feature point detection algorithm based on a recurrent neural network to detect whether there is an image of a palm position point or a back of the hand position point in the human hand contour image; S43: marking the detected palm position point or the back of the hand position point, and binding the wrist position point with the palm position point or the back of the hand position point; S44: Detect the remaining key point positions using a feature point detection algorithm based on a recurrent neural network, and bind them to the palm position points or the back of the hand position points.

7. The hand modeling method based on key point positioning according to claim 6, characterized in that: When the remaining key point positions are detected using a feature point detection algorithm based on a recurrent neural network and bound to the palm position points or the back of the hand position points, the following steps are included: S51: Connect the wrist position point and the palm position point to establish the basic skeleton of the human hand; S52: inferring the range of the location points of the phalangeal joints at the bases of the fingers according to the basic skeleton of the human hand; S53: Detect the position points of the phalangeal joints at the base of the finger using a feature point detection algorithm based on a recurrent neural network, and compare them with the results in step S52 to determine whether the position points of the phalangeal joints at the base of the finger are reasonable; S54: If the rationality meets the standard, proceed to the next step; if the rationality does not meet the standard, re-identification and detection are performed; S55: Continue to identify and detect the rationality of the next joint point until all key point positions are identified and detected.

8. The hand modeling method based on key point positioning according to claim 7, characterized in that: When establishing a three-dimensional coordinate system, the coordinates of the key points in the three-dimensional coordinate system are calculated according to the key point recognition results and the acquisition results of the image acquisition device, including the following steps: S61: Selecting an origin by random selection or by fixing a reference point to establish a three-dimensional coordinate system; S62: Calculate the coordinates of each key point in the multiple groups of binocular cameras in the stereo coordinate system by performing depth detection on the images in the binocular cameras and according to the original positions of the binocular cameras; S63: Comprehensively analyzing the coordinate results obtained by calculating the images in the multiple sets of binocular cameras, eliminating errors in the calculation process, and obtaining the comprehensive coordinates of each key point.

9. The hand modeling method based on key point positioning according to claim 8, characterized in that: When performing depth detection on the images in the binocular cameras and calculating the coordinates of each key point in the stereo coordinate system of multiple groups of binocular cameras in conjunction with the original positions of the binocular cameras, the following steps are included: S71: dividing the image captured by the binocular vision camera into a left-eye view and a right-eye view, and placing the pixel points corresponding to the key points of the left-eye view and the right-eye view into the same rectangular coordinate system; wherein the origin of the rectangular coordinate system corresponds to the pixel point at the same position in the left-eye view and the right-eye view; S72: Calculate the disparity of the pixel point corresponding to each key point in the binocular vision camera according to the coordinates of the pixel points corresponding to all key points in the left eye view and the right eye view; wherein the disparity of the pixel point at each position in the binocular vision camera is calculated according to the following formula: d=|x l -x r |, Among them, d is the disparity of the pixel point at that position in the binocular vision camera, x l is the horizontal coordinate of the pixel point at this position in the left eye view, x r is the horizontal coordinate of the pixel point at that position in the right eye view; S73: Calculate the depth of the pixel at the position in space according to the parallax; wherein the calculation formula is: Among them, f is the focal length of the binocular vision camera, b is the camera baseline distance; S74: Determine the coordinates of the key points according to the original position of the binocular camera, the horizontal coordinate and the vertical coordinate of the pixel point corresponding to each key point, and the calculated depth in space.

10. The hand modeling method based on key point positioning according to claim 9, characterized in that: When comprehensively analyzing the coordinate results obtained from the images calculated by multiple sets of binocular cameras, eliminating the errors in the calculation process, and obtaining the comprehensive coordinates of each key point, the following steps are included: S81: numbering each key point detected according to the constructed basic skeleton of the human hand, wherein key points at the same position have the same number in images captured by different binocular cameras; S82: Counting the coordinates of key points with the same number in different binocular cameras, and calculating the average coordinates of each key point; S83: removing the key point coordinates whose difference from the average coordinates exceeds a threshold, and retaining the remaining key point coordinates; S84: Recalculate the average coordinates according to the coordinates of the remaining key points to obtain the comprehensive coordinates of each key point.

Images