High-resolution three-dimensional human body imaging method based on UWB signal

Abstract

The invention discloses a high-resolution three-dimensional human body imaging method based on a UWB signal, and the method comprises the steps: S1, collecting a dynamic echo signal of the surface of a human body through a UWB MIMO radar system, and removing static background noise, thereby obtaining a UWB dynamic target signal; s2, performing back projection on the UWB dynamic target signal to a three-dimensional coordinate system to obtain a UWB signal point cloud; s3, the UWB signal point cloud is complemented through a time sequence model, and human body global features and human body local features are extracted; s4, according to the global features and the local features of the human body, extracting comprehensive feature parameters of the human body through a Transform network; and S5, mapping the human body comprehensive characteristic parameters to human body form and posture parameters, and generating a three-dimensional human body grid, and through end-to-end deep learning optimization, a high-precision and high-stability grid reconstruction effect of a three-dimensional human body under a shielding scene and a dynamic action is realized.

Description

Technical Field

[0001] The present application relates to the field of three-dimensional imaging, and in particular to a high-resolution three-dimensional human body imaging method based on UWB signals. Background Art

[0002] 3D human body reconstruction technology refers to the generation of high-precision 3D human body models by capturing and digitally processing human body data. This technology can not only show the geometric structure of the human body, but also capture subtle changes such as muscle dynamics and skin details. Its application areas cover medical diagnosis, behavior recognition, security monitoring, virtual reality and many other aspects.

[0003] At present, 3D human reconstruction technology mainly relies on methods based on vision and radio frequency signals. Vision-based methods usually rely on cameras and image processing algorithms to capture human data, and generate 3D human meshes through multi-view fusion or deep learning models. Such methods perform well in well-lit and unobstructed environments, and can accurately capture the posture, expression and dynamic changes of the human body. However, in low-light, occluded and complex environments, the performance of visual methods will be significantly limited. In addition, visual methods usually have high requirements for hardware and algorithms, and have high computational complexity. Methods based on radio frequency signals, especially WiFi and millimeter wave technology, have become another research hotspot for 3D human reconstruction in recent years. Radio frequency signals have the ability to penetrate obstacles and can perceive human posture in non-line-of-sight (NLoS) environments. These methods construct a 3D skeleton or mesh of the human body through signal reflections from the surface of the human body. However, these technologies also have significant shortcomings. For example, the priority support of WiFi signals for communication data transmission limits its resource allocation ability for human perception. In addition, the narrow spectrum bandwidth of WiFi signals leads to low spatial resolution, which means that it is not capable of capturing human details. In contrast, millimeter wave technology has a higher spectral bandwidth and is therefore able to achieve higher spatial resolution, which makes it excel in static conditions, especially in accurately capturing the outline and posture of the human body. However, the propagation characteristics of millimeter wave signals determine that it has a weak ability to penetrate obstacles. When the signal passes through furniture, wood boards or walls, its energy will be rapidly attenuated due to reflection, scattering and absorption. This signal attenuation not only weakens the perception ability, but also increases the recognition error of the target. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present invention provides a high-resolution three-dimensional human body imaging method based on UWB signals, which solves the problem of insufficient accuracy and adaptability of the existing three-dimensional human body reconstruction technology in complex environments.

[0005] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a high-resolution three-dimensional human body imaging method based on UWB signals, comprising: S1. The dynamic echo signal on the human body surface is collected by the UWB MIMO radar system, and the static background noise is removed to obtain the UWB dynamic target signal; S2, back-projecting the UWB dynamic target signal into a three-dimensional coordinate system to obtain a UWB signal point cloud; S3, completing the UWB signal point cloud through the time series model, and extracting the global features and local features of the human body; S4, extracting comprehensive feature parameters of the human body through the Transformer network according to the global and local features of the human body; S5. Mapping the comprehensive characteristic parameters of the human body to the human body shape and posture parameters to generate a three-dimensional human body mesh.

[0006] Further: In S1, the UWB dynamic target signal is calculated by the time difference method , whose expression is:

[0007] in, and The radar is t Moment and The dynamic echo signal of the human body surface received at all times, is the time interval between continuous sampling of the radar system.

[0008] Further: S2 includes: S21. Calculate the target distance by the time it takes for the UWB dynamic target signal to reach the MIMO radar system , whose expression is:

[0009] in, c is the speed of light, is the arrival time of the UWB dynamic target signal; S22, according to the target distance The UWB dynamic target signal is back-projected into a three-dimensional coordinate system using the angle information of the array to obtain the UWB signal point cloud, which is expressed as:

[0010] in, and is the angle information of the antenna array of the MIMO radar system, It is the first k The point cloud coordinates of the points.

[0011] Further: S3 includes: S31, dividing the UWB signal point cloud into multiple local areas through space segmentation and K-means clustering method; S32. Extract features of local areas using PointNet++ , and the features of each local area Aggregate to generate global features ; S33, through the time series model, the characteristics of the current time frame and the historical frame are combined to analyze the characteristics of the local area and global features Complete and obtain the global features of the human body and local features of the human body , whose expression is:

[0012] in, represents the global features of the human body in the previous time frame, Represents the local features of the human body in the previous time frame, represents the time series model for global features of the human body, Represents a time series model for local features of a human body.

[0013] Further: S4 includes: S41. Global features of the human body and local features of the human body Concatenate to input vector ; S42, input vector Input Transformer network to extract comprehensive feature parameters of human body , whose expression is:

[0014] in, Represents the Transformer network.

[0015] Further: S5 includes: S51, through the multi-layer perceptron network model, the fusion features Convert to morphological parameters and attitude parameters , whose expression is:

[0016] in, Represents a multi-layer perceptron network model; S52, according to the morphological parameters and attitude parameters , through the SMPL model 3D mesh vertices and joint positions , whose expression is:

[0017] in, represents the SMPL model; S53, according to the three-dimensional mesh vertex and joint positions , generate a 3D human body mesh.

[0018] Further: Loss function of generating 3D human body mesh L It is expressed as:

[0019] in, represents the mesh vertex position error, represents the joint position error, represents the total number of mesh vertices, represents the total number of joints in the human body model, Represents the first i The position of the vertices, Represents the first i The ground truth positions of the vertices, Represents the first j The position of the joints, Represents the first j The ground truth positions of the joints, represents the Euclidean norm.

[0020] The beneficial effects of the present invention are: 1. Achieve high-precision 3D human reconstruction in multi-obstacle environments, solving the problem of insufficient perception ability of existing methods under occlusion conditions; 2. Extract global and local features of the human body from sparse signal data to ensure delicate capture of dynamic behaviors; 3. Effective feature fusion is achieved to generate a complete and accurate human morphology and posture model. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 The figure is a flow chart of the high-resolution three-dimensional human body imaging method based on UWB signals. DETAILED DESCRIPTION

[0022] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0023] like Figure 1 As shown, in one embodiment of the present invention, a high-resolution three-dimensional human body imaging method based on UWB signals is provided, comprising: S1. The dynamic echo signal on the human body surface is collected by the UWB MIMO radar system, and the static background noise is removed to obtain the UWB dynamic target signal; S2, back-projecting the UWB dynamic target signal into a three-dimensional coordinate system to obtain a UWB signal point cloud; S3, completing the UWB signal point cloud through the time series model, and extracting the global features and local features of the human body; S4, extracting comprehensive feature parameters of the human body through the Transformer network according to the global and local features of the human body; S5. Mapping the comprehensive characteristic parameters of the human body to the human body shape and posture parameters to generate a three-dimensional human body mesh.

[0024] UWB signals have high resolution and strong penetration ability, and can penetrate obstacles to perceive the target human body, but their signals are mixed with a lot of static environmental noise, such as reflected signals from walls and furniture. These noises will interfere with the extraction of human features. For this reason, this application designs a preprocessing step based on the mobile target indication MTI algorithm, which removes static background noise and retains dynamic target signals to provide high-quality input for subsequent point cloud generation. Specifically, in S1, the UWB dynamic target signal is calculated by the time difference method. , whose expression is:

[0025] in, and The radar is t Moment and The dynamic echo signal of the human body surface received at all times, is the time interval between continuous sampling of the radar system.

[0026] The core data foundation of human body modeling is to transform the echo signal of the human body from two-dimensional data, i.e., distance and angle, to three-dimensional point cloud representation. However, the sparsity of UWB signals and the noise introduced by environmental reflections will affect the quality of the point cloud. To this end, a point cloud generation process based on the back-projection algorithm is designed to accurately map the signal to three-dimensional space. Specifically, S2 includes: S21. Calculate the target distance by the time it takes for the UWB dynamic target signal to reach the MIMO radar system , whose expression is:

[0027] in, c is the speed of light, is the arrival time of the UWB dynamic target signal; S22, according to the target distance The UWB dynamic target signal is back-projected into a three-dimensional coordinate system using the angle information of the array to obtain the UWB signal point cloud, which is expressed as:

[0028] in, and is the angle information of the antenna array of the MIMO radar system, It is the first k The point cloud coordinates of the points.

[0029] In particular, the dynamic filtering algorithm CFAR can be combined in the S2 process to improve the accuracy of the point cloud.

[0030] The point cloud generated by UWB signals is usually sparse and unevenly distributed. Especially in dynamic scenes, the point cloud of some body areas may be missing. In order to fully characterize the global structure and local details of the human body and make up for the signal missing problem, a feature extraction and dynamic completion method combining PointNet++ and LSTM is designed. Specifically, S3 includes: S31, dividing the UWB signal point cloud into multiple local areas through space segmentation and K-means clustering method; S32. Extract features of local areas using PointNet++ , and the features of each local area Aggregate to generate global features ; S33, through the time series model, the characteristics of the current time frame and the historical frame are combined to analyze the characteristics of the local area and global features Complete and obtain the global features of the human body and local features of the human body , whose expression is:

[0031] in, represents the global features of the human body in the previous time frame, Represents the local features of the human body in the previous time frame, represents the time series model for global features of the human body, Represents a time series model for local features of the human body. By combining the features of the current time frame with the features of the historical frames, it completes the missing parts of the point cloud caused by occlusion or sparsity, ensuring the integrity and accuracy of human body modeling.

[0032] Specifically, S4 includes: S41. Global features of the human body and local features of the human body Concatenate to input vector ; S42, input vector Input Transformer network to extract comprehensive feature parameters of human body , whose expression is:

[0033] in, Represents the Transformer network.

[0034] The ultimate goal of this application is to generate a high-precision three-dimensional human body mesh. The fused feature parameters need to be mapped to human body morphology and posture parameters. Specifically, S5 includes: S51, through the multi-layer perceptron network model, the fusion features Convert to morphological parameters and attitude parameters , whose expression is:

[0035] in, Represents a multi-layer perceptron network model; S52, according to the morphological parameters and attitude parameters , through the SMPL model 3D mesh vertices and joint positions , whose expression is:

[0036] in, represents the SMPL model; S53, according to the three-dimensional mesh vertex and joint positions , generate a 3D human body mesh.

[0037] Specifically, the loss function of the process of generating a 3D human body mesh is L It is expressed as:

[0038] in, represents the mesh vertex position error, represents the joint position error, represents the total number of mesh vertices, represents the total number of joints in the human body model, Represents the first i The position of the vertices, Represents the first i The ground truth positions of the vertices, Represents the first j The position of the joints, Represents the first j The ground truth positions of the joints, represents the Euclidean norm.

[0039] Through the above steps, this method can generate fine three-dimensional human body meshes in complex scenes, which not only overcomes the shortcomings of existing technologies in occluded scenes and dynamic capture, but also achieves high-precision and high-robustness modeling effects.

[0040] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A high-resolution three-dimensional human body imaging method based on UWB signals, characterized in that: include: S1. The dynamic echo signal on the human body surface is collected by the UWB MIMO radar system, and the static background noise is removed to obtain the UWB dynamic target signal; S2, back-projecting the UWB dynamic target signal into a three-dimensional coordinate system to obtain a UWB signal point cloud; S3, completing the UWB signal point cloud through the time series model, and extracting the global features and local features of the human body; S4, extracting comprehensive feature parameters of the human body through the Transformer network according to the global and local features of the human body; S5. Mapping the comprehensive characteristic parameters of the human body to the human body shape and posture parameters to generate a three-dimensional human body mesh.

2. The high-resolution three-dimensional human body imaging method based on UWB signals according to claim 1, characterized in that: In S1, the UWB dynamic target signal is calculated by the time difference method , whose expression is: in, and The radar is t Moment and The dynamic echo signal of the human body surface received at all times, is the time interval between continuous sampling of the radar system.

3. The high-resolution three-dimensional human body imaging method based on UWB signals according to claim 1, characterized in that S2 include: S21. Calculate the target distance by the time it takes for the UWB dynamic target signal to reach the MIMO radar system , whose expression is: in, c is the speed of light, is the arrival time of the UWB dynamic target signal; S22, according to the target distance The UWB dynamic target signal is back-projected into a three-dimensional coordinate system using the angle information of the array to obtain the UWB signal point cloud, which is expressed as: in, and is the angle information of the antenna array of the MIMO radar system, It is the first k The point cloud coordinates of the points.

4. The high-resolution three-dimensional human body imaging method based on UWB signals according to claim 1, characterized in that S3 include: S31, dividing the UWB signal point cloud into multiple local areas through space segmentation and K-means clustering method; S32. Extract features of local areas using PointNet++ , and the features of each local area Aggregate to generate global features ; S33, through the time series model, the characteristics of the current time frame and the historical frame are combined to analyze the characteristics of the local area and global features Complete and obtain the global features of the human body and local features of the human body , whose expression is: in, represents the global features of the human body in the previous time frame, Represents the local features of the human body in the previous time frame, represents the time series model for global features of the human body, Represents a time series model for local features of a human body.

5. The high-resolution three-dimensional human body imaging method based on UWB signals according to claim 4, characterized in that S4 include: S41. Global features of the human body and local features of the human body Concatenate to input vector ; S42, input vector Input Transformer network to extract comprehensive feature parameters of human body , whose expression is: in, Represents the Transformer network.

6. The high-resolution three-dimensional human body imaging method based on UWB signals according to claim 5, characterized in that S5 include: S51, through the multi-layer perceptron network model, the fusion features Convert to morphological parameters and attitude parameters , whose expression is: in, Represents a multi-layer perceptron network model; S52, according to the morphological parameters and attitude parameters , through the SMPL model 3D mesh vertices and joint positions , whose expression is: in, represents the SMPL model; S53, according to the three-dimensional mesh vertex and joint positions , generate a 3D human body mesh.

7. The high-resolution three-dimensional human body imaging method based on UWB signals according to any one of claims 1 to 6, characterized in that: Loss function for generating 3D human body mesh L It is expressed as: in, represents the mesh vertex position error, represents the joint position error, represents the total number of mesh vertices, represents the total number of joints in the human body model, Represents the first i The position of the vertices, Represents the first i The ground truth positions of the vertices, Represents the first j The position of the joints, Represents the first j The ground truth positions of the joints, represents the Euclidean norm.

Images