Power distribution room complex sight distance human body wrist tracking method based on single commercial millimeter wave radar

Abstract

The invention provides a power distribution room complex sight distance human body wrist tracking method based on a single commercial millimeter wave radar, and belongs to the technical field of signal detection. The method comprises the following steps: firstly, starting a radar MIMO function, and reconstructing an original signal into a three-dimensional structure; then, through an LSE-BBSR method based on least square estimation, body-background components with strong interference are eliminated from the radar signals by using a pre-acquired pure body-background signal spectrum, and arm signals embedded with wrist position information are purified; then, inputting the purified signal into a CE-GCrossViT model integrated with a multi-scale convolution word coding module and a global cross attention mechanism, so as to estimate a two-dimensional coordinate of a wrist relative to a screen cabinet plane; and finally, smoothing the trajectory by using a Butterworth filter. According to the method, the technical problem that the wrist of the human body is difficult to track in a non-contact and high-precision manner in a complicated NLOS environment of a power distribution room is solved, visual equipment or wearing of a sensor is not needed, and excellent robustness and tracking precision are shown under LOS and NLOS conditions while privacy is protected.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of signal detection, and in particular to a human wrist tracking method under complex visual distance in a power distribution room based on a single commercial millimeter wave radar. BACKGROUND

[0002] Real-time operation monitoring of power distribution room electricians is the key to ensuring the safe operation of the power distribution room. However, in this environment, the dense screen cabinet will form a large amount of visual obstruction, resulting in complex non-line-of-sight conditions, making traditional single-video-based monitoring methods have great limitations, such as being unable to effectively capture the hand movements of the operator under non-line-of-sight (NLOS).

[0003] The method based on radio frequency signals uses the reflection of radio frequency signals on the human body for perception, which can protect personal privacy and is not affected by light, providing the possibility for non-contact monitoring. However, existing radio frequency-based technologies face severe challenges in this specific scenario of the power distribution room: first, the echo signal strength generated by large parts of the human body such as the torso is much higher than that of the wrist, causing the effective wrist signal to be overwhelmed; second, in the non-line-of-sight (NLOS) environment, the radar signal undergoes multiple reflections, the waveform is severely dispersed and loses structural features, making it difficult for algorithms based on traditional signal modeling or feature engineering to accurately estimate the position of the wrist.

[0004] Therefore, there is an urgent need in the art for a new method that can effectively overcome the interference of the complex visual distance environment in the power distribution room, especially one that can achieve high-precision, non-contact human wrist tracking under strong body-background noise and non-line-of-sight (NLOS) conditions. SUMMARY

[0005] The present application aims to provide a digital movie transmission device to solve the problems raised in the background art.

[0006] The present application is achieved by the following technical solutions: The present application proposes a human wrist tracking method under complex visual distance in a power distribution room based on a single commercial millimeter wave radar, which includes the following steps: Step S1: Radar signal preprocessing: turn on the MIMO function of the single commercial millimeter wave radar to obtain the vertical and horizontal information of the scene, and reconstruct the original signal output by the radar into a three-dimensional data structure of frame x antenna channel number x fast time point number; wherein each frame only contains one chirp; Step S2: Body-background signal elimination: based on the pre-acquired pure body-background signal atlas, use the LSE-BBSR method of least squares estimation to calculate the proportion of the pure body-background signal in the radar signal to be processed, and eliminate the body-background signal component in each radar channel, thereby obtaining the purified radar arm signal; Step S3: wrist position estimation: input the purified radar arm signal into the CE-GCrossViT model, which extracts local and global features of the signal through a multi-scale convolutional word encoding module and a global cross-attention mechanism, and outputs the two-dimensional spatial coordinate estimation value of the wrist relative to the cabinet plane; Step S4: trajectory filtering: filter the two-dimensional spatial coordinate estimation value output by step S3 using a Butterworth filter to filter out outliers and obtain the final high-precision wrist trajectory.

[0007] Further, in step S2, the process of obtaining the pure body-background signal pattern includes: Collecting radar signals of the subject in the state of no arm movement and body stillness; Reconstructing the collected signals into a three-dimensional structure of frame x antenna channel number x fast time point number; Performing time average processing on the three-dimensional structure along the frame dimension; Normalizing the time-averaged signal to obtain the body-background signal pattern.

[0008] Further, in step S2, the formula for calculating the proportion using the LSE-BBSR method is:

[0009] wherein, represents the radar signal to be processed, represents the signal of the corresponding channel in the body-background signal pattern, represents the proportion of the body-background signal in the signal to be processed.

[0010] Further, in step S2, the calculation of body-background signal cancellation in each radar channel is:

[0011] wherein, is the purified arm signal, is the original radar signal to be processed, is the body-background signal pattern.

[0012] Further, in step S3, the CE-GCrossViT model adopts a dual-branch architecture, including a large-scale branch and a small-scale branch, which are respectively used to capture global multi-path reflection features and local distance information, and feature fusion is performed through a cross-attention mechanism.

[0013] Further, in the step S3, the multi-scale convolutional word encoding module adopts three different size convolution kernels to process the input signal in parallel, including global convolution, large kernel convolution and point convolution, and the output features of each convolution path are spliced along the channel dimension to form a block embedding representation.

[0014] Further, in the step S3, the training label of the CE-GCrossViT model is the real two-dimensional space coordinates of the wrist relative to the cabinet plane, and the training data is the purified radar arm signal corresponding to the real two-dimensional space coordinates.

[0015] Further, in the step S4, the Butterworth filter is used to smooth the wrist position estimation sequence and filter out coordinate jump points caused by noise.

[0016] Further, the method can adapt to different user body orientations, standing positions, wrist movement speeds and directions under the line-of-sight (LOS) condition.

[0017] Further, under the non-line-of-sight (NLOS) condition, the method can realize NLOS interference-resistant wrist tracking when the human body position and orientation are fixed.

[0018] Compared with the prior art, the present application has the following beneficial effects: 1. Effective wrist tracking in a non-line-of-sight (NLOS) environment is realized: The CE-GCrossViT deep learning model can extract effective local and global features from the diffuse and unstructured NLOS echo signal, breaking through the bottleneck of the failure of traditional algorithms in the non-line-of-sight (NLOS) environment, and realizing wrist tracking in the non-line-of-sight (NLOS) environment of the power distribution room using a single commercial millimeter wave radar.

[0019] 2. The signal purification problem in a strong interference background is solved: The LSE-BBSR method proposed in the present application can quantitatively calculate and eliminate the body-background signal that is severely coupled with the wrist signal based on the least squares estimation principle, effectively purifying the low-intensity effective wrist signal, and laying a solid foundation for subsequent high-precision position estimation.

[0020] 3. High robustness and adaptability: Under the line-of-sight (LOS) condition, the method is not sensitive to different user body orientations, standing positions, wrist movement speeds and directions, and exhibits stable tracking performance. Under the non-line-of-sight (NLOS) condition, the system exhibits strong anti-interference ability when the human body position is fixed.

[0021] 4. Non-contact and privacy protection: The present application only uses a single commercial millimeter wave radar, does not require a camera, and does not require any sensors to be worn on the human body, thereby completely avoiding visual privacy leakage problems and better meeting the safety and ethical requirements of industrial scenes such as power distribution rooms. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only preferred embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This invention provides a schematic diagram of the process structure of a method for tracking human wrists in a power distribution room under complex line-of-sight conditions based on a single commercial millimeter-wave radar.

[0024] Figure 2 shows the operation scenario of the control panel in the non-line-of-sight (NLOS) interval of the user's power distribution room in an embodiment of the present invention. Figure 2(a) shows an aerial view of the operation inside the power distribution room, and Figure 2(b) shows a two-dimensional mathematical modeling diagram of the control panel.

[0025] Figure 3 The system flow architecture in this embodiment of the invention consists of three modules: a data collection module, a body-background signal elimination module, and a wrist position estimation module.

[0026] Figure 4 This is a schematic diagram of the body-background signal spectrum collected by the radar when the subject stands in the non-line-of-sight (NLOS) operation interval under test, without arm or body movement, in an embodiment of the present invention.

[0027] Figure 5 is a schematic diagram illustrating the process of eliminating the body-background signal of the original radar signal Figure 5(a) and obtaining the purified radar arm signal Figure 5(b) using the LSE-BBSR method and body-background signal spectrum in an embodiment of the present invention.

[0028] Figure 6 This is a schematic diagram of the model architecture of the wrist position estimation module in an embodiment of the present invention.

[0029] Figure 7 This is a schematic diagram of the multi-scale convolutional word encoder in the wrist position estimation module model in an embodiment of the present invention.

[0030] Figure 8 shows the wrist tracking performance under line-of-sight (LOS) and non-line-of-sight (NLOS) conditions in an embodiment of the present invention. Figure 8(a) shows the system performance under LOS, Figure 8(b) shows an overview of the system performance under LOS, Figure 8(c) shows the system performance under NLOS, and Figure 8(d) shows an overview of the system performance under NLOS. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of the present invention more apparent, exemplary embodiments according to the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely a part of the embodiments of the present invention, and not all of the embodiments of the present invention. It should be understood that the present invention is not limited to the exemplary embodiments described herein. Based on the embodiments of the present invention described herein, all other embodiments obtained by those skilled in the art without inventive effort should fall within the protection scope of the present invention.

[0032] In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention can be practiced without one or more of these details. In other instances, certain technical features well-known in the art have not been described in order to avoid obscuring the invention.

[0033] It should be understood that the invention can be embodied in various forms and should not be construed as being limited to the embodiments set forth herein. Rather, providing these embodiments will make the disclosure thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0034] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. When used herein, the singular forms “a,” “an,” and “the” are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the terms “comprising” and / or “including,” when used in this specification, identify the presence of the stated features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups. When used herein, the term “and / or” includes any and all combinations of the associated listed items.

[0035] To fully understand this invention, a detailed structure will be presented in the following description to illustrate the technical solution proposed by this invention. Optional embodiments of the invention are described in detail below; however, in addition to these detailed descriptions, the invention may have other embodiments.

[0036] See Figure 1 Figure 8 illustrates a method for tracking the human wrist in a power distribution room under complex line-of-sight conditions based on a single commercial millimeter-wave radar. The method includes the following steps: Step S1: Radar signal preprocessing: Enable the MIMO function of a single commercial millimeter-wave radar to obtain vertical and horizontal information of the scene, and reconstruct the original signal output by the radar into a three-dimensional data structure of frame × number of antenna channels × number of fast time points; where each frame contains only one chirp. Step S2: Body-background signal elimination: Based on the pre-acquired pure body-background signal spectrum, the LSE-BBSR method of least squares estimation is used to calculate the proportion of the pure body-background signal in the radar signal to be processed, and the body-background signal component is eliminated in each radar channel to obtain the purified radar arm signal. Step S3: Wrist position estimation: The purified radar arm signal is input into the CE-GCrossViT model. The model extracts local and global features of the signal through a multi-scale convolutional word encoding module and a global cross-attention mechanism, and outputs the estimated two-dimensional spatial coordinates of the wrist relative to the screen plane. Step S4: Trajectory Filtering: The two-dimensional spatial coordinate estimate output in step S3 is filtered using a Butterworth filter to remove outliers and obtain the final high-precision wrist trajectory.

[0037] In step S2, the process of acquiring the pure body-background signal spectrum includes: Radar signals were collected from subjects when they were not moving their arms and were at rest. The acquired signal is reconstructed into a three-dimensional structure of frame × number of antenna channels × number of fast time points; The three-dimensional structure is subjected to time averaging along the frame dimension; The time-averaged signal is normalized to obtain the body-background signal spectrum.

[0038] In step S2, the formula for calculating the specific gravity using the LSE-BBSR method is as follows:

[0039] in, This indicates the radar signal to be processed. This represents the signal of the corresponding channel in the body-background signal spectrum. This indicates the proportion of the body-background signal in the signal to be processed.

[0040] In step S2, the calculation for body-background signal cancellation in each radar channel is as follows:

[0041] in, For the purified arm signal, The raw radar signal to be processed. For body-background signal spectrum.

[0042] In step S3, the CE-GCrossViT model adopts a dual-branch architecture, including a large-scale branch and a small-scale branch, which are used to capture global multipath reflection features and local distance information, respectively, and feature fusion is performed through a cross-attention mechanism.

[0043] In step S3, the multi-scale convolutional word encoding module uses three different sizes of convolutional kernels to process the input signal in parallel, including global convolution, large kernel convolution and point convolution. The output features of each convolutional path are concatenated along the channel dimension to form a block embedding representation.

[0044] In step S3, the training label of the CE-GCrossViT model is the real two-dimensional spatial coordinates of the wrist relative to the screen cabinet plane, and the training data is the purified radar arm signal corresponding to the real two-dimensional spatial coordinates.

[0045] In step S4, the Butterworth filter is used to smooth the wrist position estimation sequence and filter out coordinate jump points caused by noise.

[0046] The method can adapt to different user body orientations, standing positions, wrist movement speeds and directions under line-of-sight (LOS) conditions.

[0047] The method described above can achieve wrist tracking resistant to NLOS interference under non-line-of-sight (NLOS) conditions, when the human body position and orientation are fixed.

[0048] For example, this invention uses a single-chip commercial millimeter-wave radar such as the AWR1843BOOST, operating at a frequency of 76-81 GHz. The radar is deployed outside the operating bay of the power distribution room, facing or to the side of the target cabinet area.

[0049] Step S1: Radar signal preprocessing. Enable the radar's MIMO function to acquire vertical and horizontal information in the scene to expand the radar information dimension. The radar is configured to transmit only one chirp per frame. Therefore, this scheme does not rely on velocity information. The original ADC data output by the radar is initially reconstructed into a three-dimensional data structure of [number of frames × number of antenna channels × number of fast time sampling points].

[0050] Step S2: Body-background signal cancellation First, obtain a pure body-background signal spectrum. For example... Figure 4 As shown, the subject stands within the target operation interval with their arms hanging naturally and remaining still, and radar signals are collected in this state. ,in For frames, For the channel, For faster time points, along the frame dimension right The signal is then averaged, and then normalized for each channel (e.g., divided by its maximum value within that channel) to obtain the pure body-background signal spectrum. This atlas characterizes the fixed body and background reflection features in the environment without wrist movement. Secondly, body-background cancellation of dynamic signals is performed. Radar signals are collected when the subject performs an operation that generates arm movement. .for Data from each frame and each channel According to the formula Calculate its background spectrum signal with the corresponding channel. Fit coefficient This coefficient reflects the proportion of the body-background component in this frame of signal. Subsequently, signal cancellation is performed:

[0051] The purified radar arm signal was obtained. This process is as follows Figure 5(a) , 5(b) As shown, strong background interference is effectively suppressed, and the signal features related to arm movement are highlighted.

[0052] Step S3: Wrist position estimation. The refined radar arm signal obtained in step S2 is used for this step. As input to the CE-GCrossViT model. The architecture of this model is as follows. Figure 6 As shown. The core of the model lies in its feature extraction capability: Multi-scale convolutional word encoders (such as...) Figure 7 This module uses three different sizes of convolution kernels (e.g., global convolution, large kernel convolution, and point convolution) to embed and encode the input signal in parallel, which can capture local features of different granularities at the same time, providing rich low-level information for subsequent processing.

[0053] The model employs a dual-branch cross-attention mechanism: It includes a large-scale branch and a small-scale branch. The large-scale branch focuses on capturing global multipath reflection context information, while the small-scale branch concentrates on fine-grained local distance features. Through this cross-attention mechanism, the information from both branches can be fused bidirectionally, enabling the model to jointly utilize local physical quantities and global multipath features for decision-making. The model's output is the two-dimensional spatial coordinates (X, Z) of the wrist relative to a pre-defined screen plane. The model is trained using a large amount of labeled data, consisting of refined radar signals labeled with real wrist coordinates synchronously acquired through an optical motion capture system.

[0054] Step S4: Trajectory Filtering The wrist coordinates estimated frame by frame by the model may contain outliers or jitter due to noise interference. In this embodiment, a low-pass Butterworth filter is used to post-process the estimated coordinate sequence to filter out high-frequency noise and abnormal jump points, thereby outputting a smooth, stable and high-precision wrist movement trajectory.

[0055] Performance verification like Figure 8(a) , 8(b) As shown in 8(c) and 8(d), under line-of-sight (LOS) conditions, the system tracking trajectory closely matches the actual trajectory. Under non-line-of-sight (NLOS) conditions, despite the complexity of the actual trajectory, the system proposed in this invention can still achieve effective tracking, which is significantly better than traditional methods, demonstrating its excellent practicality and robustness in complex line-of-sight environments in power distribution rooms.

[0056] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for tracking human wrists in complex line-of-sight conditions in a power distribution room based on a single commercial millimeter-wave radar, characterized in that, The method includes the following steps: Step S1: Radar signal preprocessing: Enable the MIMO function of a single commercial millimeter-wave radar to obtain vertical and horizontal information of the scene, and reconstruct the original signal output by the radar into a three-dimensional data structure of frame × number of antenna channels × number of fast time points; where each frame contains only one chirp. Step S2: Body-background signal elimination: Based on the pre-acquired pure body-background signal spectrum, the LSE-BBSR method of least squares estimation is used to calculate the proportion of the pure body-background signal in the radar signal to be processed, and the body-background signal component is eliminated in each radar channel to obtain the purified radar arm signal. Step S3: Wrist position estimation: The purified radar arm signal is input into the CE-GCrossViT model. The model extracts local and global features of the signal through a multi-scale convolutional word encoding module and a global cross-attention mechanism, and outputs the estimated two-dimensional spatial coordinates of the wrist relative to the screen plane. Step S4: Trajectory Filtering: The two-dimensional spatial coordinate estimate output in Step S3 is filtered using a Butterworth filter to remove outliers and obtain the final high-precision wrist trajectory.

2. The method according to claim 1, characterized in that, In step S2, the process of acquiring the pure body-background signal spectrum includes: Radar signals were collected from subjects when they were not moving their arms and their bodies were still. The acquired signal is reconstructed into a three-dimensional structure of frame × number of antenna channels × number of fast time points; The three-dimensional structure is subjected to time averaging along the frame dimension; The time-averaged signal is normalized to obtain the body-background signal spectrum.

3. The method according to claim 2, characterized in that, In step S2, the formula for calculating the specific gravity using the LSE-BBSR method is as follows: in, This indicates the radar signal to be processed. This represents the signal of the corresponding channel in the body-background signal spectrum. This indicates the proportion of the body-background signal in the signal to be processed.

4. The method according to claim 3, characterized in that, In step S2, the calculation for body-background signal cancellation in each radar channel is as follows: in, For the purified arm signal, The raw radar signal to be processed. For body-background signal spectrum.

5. The method according to claim 1, characterized in that, In step S3, the CE-GCrossViT model adopts a dual-branch architecture, including a large-scale branch and a small-scale branch, which are used to capture global multipath reflection features and local distance information, respectively, and feature fusion is performed through a cross-attention mechanism.

6. The method according to claim 1 or 5, characterized in that, In step S3, the multi-scale convolutional word encoding module uses three different sizes of convolutional kernels to process the input signal in parallel, including global convolution, large kernel convolution and point convolution. The output features of each convolutional path are concatenated along the channel dimension to form a block embedding representation.

7. The method according to claim 1, characterized in that, In step S3, the training label of the CE-GCrossViT model is the real two-dimensional spatial coordinates of the wrist relative to the screen cabinet plane, and the training data is the purified radar arm signal corresponding to the real two-dimensional spatial coordinates.

8. The method according to claim 1, characterized in that, In step S4, the Butterworth filter is used to smooth the wrist position estimation sequence and filter out coordinate jump points caused by noise.

9. The method according to claim 1, characterized in that, The method can adapt to different user body orientations, standing positions, wrist movement speeds and directions under line-of-sight (LOS) conditions.

10. The method according to claim 1, characterized in that, The method described above can achieve wrist tracking resistant to NLOS interference under non-line-of-sight (NLOS) conditions, when the human body position and orientation are fixed.

Images