CN108053469A - Complicated dynamic scene human body three-dimensional method for reconstructing and device under various visual angles camera - Google Patents

Abstract

The invention discloses a method and device for three-dimensional reconstruction of a human body in a complex dynamic scene with a multi-view camera, wherein the method includes: shooting a target human body object from multiple viewpoints to obtain a multi-viewpoint two-dimensional image at the same moment; The model after network learning predicts the parts of the human body in the video sequence to obtain the human skeleton information of the human body; according to the skeleton information and through the virtual three-dimensional human body model to match the movements and general outlines of the characters in the image, to obtain the human body skeleton information Contour information: using the calibrated internal and external parameters of each viewpoint camera, combined with human skeleton information and human contour information, the Visual Hull method is used to carry out three-dimensional modeling of the human body. This method can combine camera calibration and other processes to perform three-dimensional reconstruction of human objects in video sequences, so as to achieve accurate segmentation of human objects and effectively improve the accuracy and reliability of reconstruction.

Description

Method and device for 3D reconstruction of human body in complex dynamic scenes under multi-view cameras

technical field

The invention relates to the technical field of computer vision, in particular to a method and device for three-dimensional reconstruction of a human body in complex dynamic scenes under multi-view cameras.

Background technique

In computer vision, 3D reconstruction is the process of reconstructing 3D information from single-view or multi-view images. Since the information of a single video is incomplete, 3D reconstruction needs to use prior knowledge, while multi-view 3D reconstruction can use more The information of the two-dimensional image of the viewpoint is used to reconstruct the three-dimensional model. However, most of the current 3D reconstruction algorithms do not use 2D information accurately and comprehensively enough, and the calculation process relies too much on information provided by external devices, such as depth information provided by depth cameras, or on the segmentation results of objects and backgrounds, etc. , resulting in a relatively rough reconstruction result.

Contents of the invention

The present invention aims to solve one of the technical problems in the related art at least to a certain extent.

Therefore, an object of the present invention is to propose a method for three-dimensional human body reconstruction in complex dynamic scenes under multi-view cameras, which can realize accurate segmentation of human objects and effectively improve the accuracy and reliability of reconstruction.

Another object of the present invention is to propose a device for three-dimensional human body reconstruction in complex dynamic scenes under multi-view cameras.

In order to achieve the above purpose, an embodiment of the present invention proposes a method for 3D human body reconstruction in complex dynamic scenes with multi-view cameras, including the following steps: shooting a target human body from multiple viewpoints to obtain multiple viewpoints at the same moment Two-dimensional images; use the deep network learning model to predict the various parts of the human body in the video sequence to obtain the human body skeleton information; according to the skeleton information and use the virtual three-dimensional human body model to match the actions and actions of the characters in the image rough outline to obtain the human body contour information of the human body; using the calibrated internal and external reference information of each viewpoint camera, combined with the human skeleton information and the human body contour information, the Visual Hull method is used to perform three-dimensional modeling of the human body.

The 3D human body reconstruction method in complex dynamic scenes under the multi-view camera of the embodiment of the present invention can use the 2D information provided by multiple viewpoints, use the method of deep learning to estimate the approximate position of the human body, and then use the virtual 3D human body model and graph cut The method jointly solves the problem that the human body object in the target scene is difficult to segment due to the complex background, and combines the camera calibration and other processes to perform 3D reconstruction of the human body object in the video sequence, so as to achieve accurate human object segmentation and effectively improve the accuracy and accuracy of reconstruction. reliability.

In addition, the method for three-dimensional human body reconstruction in complex dynamic scenes under multi-view cameras according to the above-mentioned embodiments of the present invention may also have the following additional technical features:

Furthermore, in an embodiment of the present invention, the multi-viewpoint shooting of the target human object to obtain multi-viewpoint two-dimensional images at the same time further includes: erecting cameras with multiple angles around the human body object , and align within the motion range of the human object, and ensure that the cameras at multiple angles are consistent.

Further, in one embodiment of the present invention, the acquiring the human body skeleton information further includes: using a deep convolutional neural network to predict the joint points of the human body, and combining multiple angles to obtain the joint points in the three-dimensional space The human skeleton information.

Further, in an embodiment of the present invention, the obtaining the human body contour information further includes: performing a background subtraction operation on the single-angle picture to obtain the position of the person, and using the joint points determined by the graph cut method as a template The foreground point of the figure, the area obtained after subtracting the character background is the area to be segmented, and other areas are segmented as the background point of the template to obtain the segmentation result; using smpl as the character model, the detected joint points are used as input to solve a global The smallest optimization problem is to match the initial three-dimensional human body model to the action of the character, so that the contour of the matched character model can be used as a supplement to the previous segmentation to obtain the human body contour information.

Further, in one embodiment of the present invention, the 3D modeling of the human body using the Visual Hull method further includes: calibrating the cameras at multiple angles, obtaining the internal and external parameter information of each camera to perform a spatial model of the scene; Using the Visual Hull method, using the human body outline information under multiple views, traverse each point in the space model to determine whether it belongs to a human body object, if the spatial point is projected into multiple two-dimensional plane views, it is in the human body object If it is within the outline, it is considered to belong to the 3D human body object, until every point in the space is traversed to obtain the final 3D human body model.

In order to achieve the above purpose, another embodiment of the present invention proposes a device for 3D human body reconstruction in complex dynamic scenes under multi-view cameras, including: an acquisition module, which is used to shoot target human objects from multiple viewpoints, so as to obtain The following multi-view two-dimensional image; the prediction and acquisition module is used to predict each part of the human body in the video sequence through the model learned by the deep network, so as to obtain the human body skeleton information of the human body; the matching acquisition module is used for according to the skeleton. information and use the virtual three-dimensional human body model to match the action and rough outline of the person in the image to obtain the human body outline information of the human body; the processing module is used to use the calibrated internal and external parameter information of each viewpoint camera and combine the human skeleton information and the contour information of the human body, the Visual Hull method is used to carry out three-dimensional modeling of the human body.

The 3D human body reconstruction device for complex dynamic scenes under the multi-view camera of the embodiment of the present invention can use the two-dimensional information provided by multiple viewpoints, use the method of deep learning to estimate the approximate position of the human body, and then use the virtual 3D human body model and graph cut The method jointly solves the problem that the human body object in the target scene is difficult to segment due to the complex background, and combines the camera calibration and other processes to perform 3D reconstruction of the human body object in the video sequence, so as to achieve accurate human object segmentation and effectively improve the accuracy and accuracy of reconstruction. reliability.

In addition, the device for three-dimensional human body reconstruction in complex dynamic scenes under multi-view cameras according to the above-mentioned embodiments of the present invention may also have the following additional technical features:

Further, in an embodiment of the present invention, the acquisition module is also used to set up cameras with multiple angles around the human body object, and align them within the range of motion of the human body object, and ensure that the cameras at multiple angles The cameras are consistent.

Further, in one embodiment of the present invention, the prediction acquisition module u is also used to predict the joint points of the human body by using a deep convolutional neural network, and combine multiple angles to obtain the joint points in the three-dimensional space Human skeleton information.

Further, in one embodiment of the present invention, the matching acquisition module is also used to perform background subtraction operation on the single-angle picture to obtain the position of the character, and use the joint points determined by the graph cut method as the foreground points of the template, The area obtained after subtracting the character background is the area to be segmented, and other areas are segmented as the background points of the template to obtain the segmentation result, and the smpl is used as the character model, and the detected joint points are used as input to solve a global minimum optimization The problem is to match the initial three-dimensional human body model to the action of the character, so that the contour of the matched character model can be used as a supplement to the previous segmentation to obtain the human body contour information.

Further, in one embodiment of the present invention, the processing module is also used to calibrate cameras with multiple angles, obtain the internal and external parameter information of each camera to carry out a spatial model of the scene, and use the Visual Hull method, Use the human body contour information under multiple views to traverse each point in the space model to determine whether it belongs to the human body object. If the spatial point is projected into multiple two-dimensional plane views and is within the contour of the human body object, it is considered to belong to the three-dimensional Human body objects, until each point in the space is traversed to obtain the final 3D human body model.

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.

Description of drawings

The above and/or additional aspects and advantages of the present invention will become apparent and easy to understand from the following description of the embodiments in conjunction with the accompanying drawings, wherein:

Fig. 1 is a flow chart of a method for three-dimensional reconstruction of a human body in a complex dynamic scene under a multi-view camera according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of a single-angle single-frame image obtained in a multi-view video sequence according to an embodiment of the present invention;

3 is a flow chart of a method for three-dimensional reconstruction of a human body in a complex dynamic scene under a multi-view camera according to a specific embodiment of the present invention;

FIG. 4 is a schematic diagram of a human skeleton prediction result of a deep convolutional neural network according to an embodiment of the present invention;

5 is a flow chart of a deep convolutional neural network according to one embodiment of the present invention;

Fig. 6 is a schematic diagram of a matching result of a three-dimensional virtual human body according to an embodiment of the present invention;

Fig. 7 is a template diagram in the graph cut algorithm according to one embodiment of the present invention;

Fig. 8 is a schematic diagram of the results after graph cutting according to an embodiment of the present invention;

Fig. 9 is a schematic diagram of a graph cut result plus a virtual human body result according to an embodiment of the present invention;

Fig. 10 is a schematic diagram of human body results after accurate segmentation again according to an embodiment of the present invention;

Fig. 11 is a schematic diagram of a three-dimensional human skeleton reconstruction result according to an embodiment of the present invention;

Fig. 12 is a schematic diagram of a three-dimensional human body reconstruction result according to an embodiment of the present invention;

Fig. 13 is a schematic structural diagram of a device for three-dimensional reconstruction of a human body in a complex dynamic scene under multi-view cameras according to an embodiment of the present invention.

Detailed ways

Embodiments of the present invention are described in detail below, examples of which are shown in the drawings, wherein the same or similar reference numerals designate the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the figures are exemplary and are intended to explain the present invention and should not be construed as limiting the present invention.

The method and device for 3D human body reconstruction of complex dynamic scenes under multi-view cameras according to the embodiments of the present invention will be described below with reference to the accompanying drawings. First, the 3D human body reconstruction of complex dynamic scenes under multi-view cameras according to the embodiments of the present invention will be described with reference to the accompanying drawings. rebuild method.

FIG. 1 is a flow chart of a method for three-dimensional reconstruction of a human body in a complex dynamic scene under multi-view cameras according to an embodiment of the present invention.

As shown in Figure 1, the method for 3D human body reconstruction in complex dynamic scenes under the multi-view camera includes the following steps:

In step S101, multi-viewpoint shooting is performed on a target human body object, so as to obtain multi-viewpoint two-dimensional images at the same moment.

It can be understood that, in the embodiment of the present invention, multi-viewpoint shooting can be performed on the target human object to obtain multi-viewpoint two-dimensional images at the same moment, wherein the acquired images are shown in FIG. 2 .

Furthermore, in one embodiment of the present invention, the multi-viewpoint shooting of the target human body object to obtain multi-viewpoint two-dimensional images at the same moment further includes: erecting cameras with multiple angles around the human body object, and Aim within the range of motion of the human subject and ensure that the cameras from multiple angles are consistent.

Specifically, the embodiment of the present invention can perform multi-viewpoint shooting on the target human object to obtain multi-viewpoint two-dimensional images at the same time. It is enough to align within the motion range of the human body object, and try to ensure that the cameras are consistent, including the same model and the same height, which will help to obtain better reconstruction results.

In step S102, each part of the human body in the video sequence is predicted by using the learned model of the deep network, so as to obtain human body skeleton information.

It can be understood that, as shown in FIG. 3 , in this embodiment of the present invention, a model learned by a deep network can be used to predict various parts of a human body in a video sequence, thereby obtaining skeleton information of a human body object. Wherein, the deep network may be a deep convolutional neural network, and the result of human skeleton prediction is shown in FIG. 4 .

For example, the method of deep convolutional neural network generally uses the response map of each part of the human body to express the spatial constraints between the parts, and the response map and feature map are transmitted in the network together as data. The network is divided into stages. There are supervised training at each stage to avoid the problem that the deep network is difficult to optimize. Because the same network is used to process input features and responses at multiple scales at the same time, it not only ensures accuracy, but also considers the long-distance relationship between various components. The main process of the algorithm is to calculate the response graph of each component at each scale, and for each component, accumulate the response graphs of all scales to obtain the total response graph, and finally find the corresponding maximum response graph on the total response graph of each component The point is the position of the component. Finally, the prediction result is obtained.

As shown in Figure 5, the height and width scale of the original image (ori image) in the deep convolutional neural network is 368*368, the number of image channels is 3, and the image passes through two branch convolution layers ( Convs), one obtains the score (Score1); the other branch passes through a convolutional layer and connects with Score1 (concat), and then obtains score2 after passing through the convolutional layer; the third to sixth stages, from the second stage branch The feature image (feature image) obtained during the convolution step goes through steps similar to the second stage, and finally outputs the result image. The center map is a gray-scale image of Gaussian distribution, and a small center map is obtained after downsampling.

Furthermore, in one embodiment of the present invention, obtaining the human body skeleton information further includes: using a deep convolutional neural network to predict the joint points of the human body, and combining multiple angles to obtain the human body skeleton in three-dimensional space information.

That is to say, the embodiment of the present invention utilizes a deep convolutional neural network (Deep Pose Machines) to detect joint points, so as to obtain skeleton information of a human body object.

In step S103, according to the skeleton information and the virtual three-dimensional human body model, the actions and the general outline of the person in the image are matched to obtain the human body outline information.

It can be understood that, as shown in Figure 3, the embodiment of the present invention can use the model after deep network learning to predict the various parts of the human body in the video sequence, and segment the outline information of the target human body object. The segmented person appears to be over-segmented or under-segmented, so after step S102, the embodiment of the present invention uses a virtual three-dimensional human body model to match the action and outline of the person in the image, wherein the matching result is shown in Figure 6; finally Perform precise subdivision, and finally achieve a satisfactory segmentation effect.

Furthermore, in one embodiment of the present invention, obtaining the human body contour information of the human body further includes: performing a background subtraction operation on the single-angle picture to obtain the position of the character, and using the joint points determined by the graph cut method as the front of the template Scenic spots, the area obtained after subtracting the character background is the area to be segmented, and other areas are segmented as the background points of the template to obtain the segmentation result; using smpl as the character model, the detected joint points are used as input, and a global minimum optimization problem is solved , so that the initial three-dimensional human body model is matched to the character's action, so that the matched character model contour is used as a supplement to the previous segmentation to obtain human body contour information.

It can be understood that in the embodiment of the present invention, the joint points can be used as the foreground points in the graph cut method to roughly segment the contour information of the target human object, wherein, as shown in FIG. 7 , the black in the figure is the background, and the white range is In the area to be segmented, the gray area in the middle is the foreground point, and the result after using the graph cut method is shown in Figure 8. This step will encounter the phenomenon of over-segmentation or under-segmentation of the segmented characters due to the complex background in the image. Therefore, after the above-mentioned stage, a virtual 3D human body model is used to match the movements and general outlines of the characters in the image. Among them, the graph cut method is used. The result after segmentation plus the virtual three-dimensional human body model is shown in Figure 9; finally, accurate subdivision is performed to achieve a satisfactory segmentation effect, and the human body result after accurate segmentation is shown in Figure 10. In addition, the advantage of this method is that it avoids the shortcomings of insufficient segmentation accuracy and over-segmentation in complex backgrounds from the perspective of traditional image segmentation, and can achieve accurate segmentation of people and objects while achieving positioning of the human body. This part is the core of 3D reconstruction and plays a key role in the reconstruction results.

Specifically, the embodiment of the present invention can use the joint points as the foreground points in the graph cut method to roughly segment the outline information of the target human object. position, and then apply the graph cut method, using the determined joint points as foreground points, the area obtained after subtracting the character background is the area to be segmented, and other areas are segmented as background points to obtain the segmentation result. Secondly, a virtual three-dimensional human body model is used to match the actions of the characters in the image. In the embodiment of the present invention, smpl is used as the character model, and the detected joint points are used as input to solve a global minimum optimization problem, so that the initial three-dimensional human body model can Match to the character action we need. Finally, the matched character model outline is used as a supplement to the previous segmentation.

In step S104, using the calibrated internal and external reference information of each viewpoint camera, combined with human skeleton information and human body contour information, the Visual Hull method is used to carry out three-dimensional modeling of the human body.

It can be understood that in the embodiment of the present invention, the calibrated internal and external parameter information of each viewpoint camera can be used in combination with the information of the segmented human body object, and the Visual Hull method can be used to perform three-dimensional modeling of the human body.

Further, in one embodiment of the present invention, adopting the Visual Hull method to carry out three-dimensional modeling of the human body further includes: calibrating the cameras of multiple angles, obtaining the internal and external parameter information of each camera to carry out the spatial model of the scene; using Visual Hull The Hull method uses the human body contour information under multiple views to traverse each point in the spatial model to determine whether it belongs to the human body object. If the spatial point is projected into multiple two-dimensional plane views and is within the contour of the human body object, it is considered to belong to the human body object. 3D human body object until each point in the space is traversed to obtain the final 3D human body model.

Specifically, firstly, the cameras of several viewing angles are calibrated, and the internal and external parameter information of each camera is obtained to make a spatial model of the scene. Then use the Visual Hull method to traverse each point in the space model to determine whether it belongs to the human body object by using the contour information of the human body object under multiple views. If it is inside the contour, it is considered that the point belongs to the 3D human body object, and the final 3D human body model is obtained until each point in the space is traversed. 12 shown.

According to the method for 3D human body reconstruction in complex dynamic scenes with multi-view cameras proposed in the embodiment of the present invention, it can use the 2D information provided by multiple viewpoints to estimate the approximate position of the human body using the method of deep learning, and then use the virtual 3D human body model and The graph cut method jointly solves the problem that human objects in the target scene are difficult to segment due to complex backgrounds, and combines camera calibration and other processes to perform 3D reconstruction of human objects in video sequences, so as to achieve accurate segmentation of human objects and effectively improve the accuracy of reconstruction. sex and reliability.

Next, a device for three-dimensional reconstruction of a human body in a complex dynamic scene under a multi-view camera proposed according to an embodiment of the present invention will be described with reference to the accompanying drawings.

Fig. 13 is a schematic structural diagram of a device for three-dimensional reconstruction of a human body in a complex dynamic scene under multi-view cameras according to an embodiment of the present invention.

As shown in FIG. 13 , the device 10 for 3D human body reconstruction in complex dynamic scenes under multi-view cameras includes: an acquisition module 100 , a prediction acquisition module 200 , a matching acquisition module 300 and a processing module 400 .

Wherein, the collection module 100 is used for shooting a target human object from multiple viewpoints, so as to obtain multi-viewpoint two-dimensional images at the same moment. The prediction acquisition module 200 is used to predict each part of the human body in the video sequence through the learned model of the deep network, so as to obtain human body skeleton information. The matching acquisition module 300 is used to match the action and rough outline of the person in the image according to the skeleton information and through the virtual three-dimensional human body model, so as to obtain the human body outline information of the human body. The processing module 400 is used to use the calibrated internal and external parameter information of each viewpoint camera, combined with human skeleton information and human body contour information, to perform three-dimensional human body modeling using the Visual Hull method. The device 10 of the embodiment of the present invention can perform three-dimensional reconstruction of human objects in video sequences in combination with processes such as camera calibration, so as to achieve accurate segmentation of human objects and effectively improve the accuracy and reliability of reconstruction.

Further, in an embodiment of the present invention, the acquisition module 100 is also used to set up cameras with multiple angles around the human body object, align them within the motion range of the human body object, and ensure that the cameras at multiple angles are consistent.

Further, in an embodiment of the present invention, the prediction acquisition module 200 is also used to predict the joint points of the human body by using the deep convolutional neural network, and combine multiple angles to obtain the human skeleton information in the three-dimensional space.

Further, in one embodiment of the present invention, the matching acquisition module 300 is also used to perform background subtraction operation on the single-angle picture to obtain the position of the character, and the joint points determined by the graph cut method are used as the foreground point of the template, and the character The area obtained after background subtraction is the area to be segmented, and other areas are segmented as the background points of the template to obtain the segmentation result, and the smpl is used as the character model, and the detected joint points are used as input to solve a global minimum optimization problem, so that The initial 3D human body model is matched to the character's actions, so that the matched character model outline can be used as a supplement to the previous segmentation to obtain human body outline information.

Further, in one embodiment of the present invention, the processing module 400 is also used to calibrate the cameras of multiple angles, obtain the internal and external parameter information of each camera to carry out a spatial model of the scene, and use the Visual Hull method to utilize multiple Human body contour information under the view, traversing each point in the space model to determine whether it belongs to the human body object, if the space point is projected into multiple two-dimensional plane views and all are within the contour of the human body object, it is considered to belong to the three-dimensional human body object until the traversal Each point in the space is completed to obtain the final 3D human body model.

It should be noted that the foregoing explanations of the embodiment of the method for 3D reconstruction of a human body in complex dynamic scenes with multi-view cameras are also applicable to the device for 3D reconstruction of human bodies in complex dynamic scenes with multi-view cameras in this embodiment, so details are not repeated here.

According to the embodiment of the present invention, the device for 3D human body reconstruction in complex dynamic scenes with multi-view cameras can use the 2D information provided by multiple viewpoints, use the method of deep learning to estimate the approximate position of the human body, and then use the virtual 3D human body model and The graph cut method jointly solves the problem that human objects in the target scene are difficult to segment due to complex backgrounds, and combines camera calibration and other processes to perform 3D reconstruction of human objects in video sequences, so as to achieve accurate segmentation of human objects and effectively improve the accuracy of reconstruction. sex and reliability.

In describing the present invention, it should be understood that the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", " Back", "Left", "Right", "Vertical", "Horizontal", "Top", "Bottom", "Inner", "Outer", "Clockwise", "Counterclockwise", "Axial", The orientation or positional relationship indicated by "radial", "circumferential", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying the referred device or element Must be in a particular orientation, be constructed in a particular orientation, and operate in a particular orientation, and therefore should not be construed as limiting the invention.

In addition, the terms "first" and "second" are used for descriptive purposes only, and cannot be interpreted as indicating or implying relative importance or implicitly specifying the quantity of indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include at least one of these features. In the description of the present invention, "plurality" means at least two, such as two, three, etc., unless otherwise specifically defined.

In the present invention, unless otherwise clearly specified and limited, terms such as "installation", "connection", "connection" and "fixation" should be understood in a broad sense, for example, it can be a fixed connection or a detachable connection , or integrated; it may be mechanically connected or electrically connected; it may be directly connected or indirectly connected through an intermediary, and it may be the internal communication of two components or the interaction relationship between two components, unless otherwise specified limit. Those of ordinary skill in the art can understand the specific meanings of the above terms in the present invention according to specific situations.

In the present invention, unless otherwise clearly specified and limited, the first feature may be in direct contact with the first feature or the first and second feature may be in direct contact with the second feature through an intermediary. touch. Moreover, "above", "above" and "above" the first feature on the second feature may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is higher in level than the second feature. "Below", "beneath" and "beneath" the first feature may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is less horizontally than the second feature.

In the description of this specification, descriptions referring to the terms "one embodiment", "some embodiments", "example", "specific examples", or "some examples" mean that specific features described in connection with the embodiment or example , structure, material or characteristic is included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the described specific features, structures, materials or characteristics may be combined in any suitable manner in any one or more embodiments or examples. In addition, those skilled in the art can combine and combine different embodiments or examples and features of different embodiments or examples described in this specification without conflicting with each other.

Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention, those skilled in the art can make the above-mentioned The embodiments are subject to changes, modifications, substitutions and variations.

Claims (10)

1. A complex dynamic scene human body three-dimensional reconstruction method under a multi-view camera, is characterized in that, comprises the following steps: Multi-viewpoint shooting of the target human object to obtain multi-viewpoint two-dimensional images at the same moment; Predict the parts of the human body in the video sequence through the deep network learning model to obtain the human body skeleton information; According to the skeleton information and through the virtual three-dimensional human body model to match the action and the general outline of the person in the image, to obtain the human body outline information of the human body; and Using the calibrated internal and external reference information of each viewpoint camera, combined with the human skeleton information and the human body contour information, the Visual Hull method is used to carry out three-dimensional modeling of the human body. 2. the complex dynamic scene human body three-dimensional reconstruction method under the multi-view camera according to claim 1, is characterized in that, described target human body object is carried out multi-viewpoint shooting, to obtain the multi-viewpoint two-dimensional image under the same moment, Further includes: Cameras with multiple angles are set up around the human body object and aligned within the motion range of the human body object, and the cameras at multiple angles are guaranteed to be consistent. 3. the complex dynamic scene human body three-dimensional reconstruction method under the multi-view camera according to claim 1, is characterized in that, described obtaining the human body skeleton information of human body, further comprises: The joint points of the human body are predicted by using the deep convolutional neural network, and the skeleton information of the human body in three-dimensional space is obtained by combining multiple angles. 4. the complex dynamic scene human body three-dimensional reconstruction method under the multi-view camera according to claim 3, is characterized in that, described obtaining the human body contour information of human body, further comprises: The background subtraction operation is performed on the single-angle picture to obtain the position of the character, and the joint points determined by the graph cut method are used as the foreground point of the template, the area obtained after subtracting the background of the character is the area to be segmented, and other areas are used as the background point of the template. Segmentation to obtain segmentation results; Using smpl as a character model, the detected joint points are used as input, and a global minimum optimization problem is solved, so that the initial 3D human body model is matched to the character's action, so that the matched character model outline is used as a supplement to the previous segmentation, Obtain the human body contour information. 5. the complex dynamic scene human body three-dimensional reconstruction method under the multi-view camera according to claim 2, is characterized in that, described adopting Visual Hull method to carry out human body three-dimensional modeling, further comprises: Calibrate the cameras at multiple angles, obtain the internal and external parameters of each camera, and make a spatial model of the scene; Using the Visual Hull method, using the human body outline information under multiple views, traverse each point in the space model to determine whether it belongs to a human body object, if the spatial point is projected into multiple two-dimensional plane views, it is in the human body object If it is within the outline, it is considered to belong to the 3D human body object, until every point in the space is traversed to obtain the final 3D human body model. 6. A device for three-dimensional reconstruction of a human body in a complex dynamic scene under a multi-view camera, characterized in that it comprises: The acquisition module is used to perform multi-viewpoint shooting on the target human object, so as to obtain multi-viewpoint two-dimensional images at the same moment; The prediction acquisition module is used to predict each part of the human body in the video sequence through the model after deep network learning, so as to obtain the human body skeleton information of the human body; The matching acquisition module is used to match the action and the general outline of the person in the image according to the skeleton information and through the virtual three-dimensional human body model, so as to obtain the human body outline information of the human body; and The processing module is used to use the calibrated internal and external reference information of each viewpoint camera and combine the human skeleton information and the human body contour information to perform three-dimensional modeling of the human body by using the Visual Hull method. 7. The device for three-dimensional human body reconstruction of complex dynamic scenes under multi-view cameras according to claim 6, wherein the acquisition module is also used to set up cameras with multiple angles around the human body object, and aim at the human body object Within the range of motion of the object, and ensure that the cameras from multiple angles are consistent. 8. The device for three-dimensional human body reconstruction in complex dynamic scenes under multi-view cameras according to claim 6, wherein the prediction acquisition module is also used to predict the joint points of the human body by using a deep convolutional neural network, and in combination with The human skeleton information in three-dimensional space is obtained from multiple angles. 9. The complex dynamic scene human body three-dimensional reconstruction device under the multi-view camera according to claim 8, characterized in that, the matching acquisition module is also used to perform background subtraction operations on single-angle pictures to obtain the position of the person, and The joint points determined by the graph cut method are used as the foreground point of the template, the area obtained after subtracting the background of the character is the area to be segmented, and other areas are segmented as the background point of the template to obtain the segmentation result, and the smpl is used as the character model to detect The joint points of the algorithm are used as input to solve a global minimum optimization problem, so that the initial 3D human body model is matched to the character's actions, so that the matched character model contour is used as a supplement to the previous segmentation to obtain the human body contour information. 10. The device for three-dimensional human body reconstruction of complex dynamic scenes under multi-view cameras according to claim 7, wherein the processing module is also used for calibrating cameras with multiple angles, and obtaining internal and external parameter information of each camera Carry out a spatial model of the scene, and use the Visual Hull method to use the human body contour information under multiple views to traverse each point in the spatial model to determine whether it belongs to a human body object. If the spatial point is projected to multiple two-dimensional If it is within the contour of the human body object in the plane view, it is considered to belong to the 3D human body object until every point in the space is traversed to obtain the final 3D human body model.