Unmanned aerial vehicle positioning method and system based on sensing fusion non-line-of-sight recognition

Abstract

The invention discloses an unmanned aerial vehicle positioning method and system based on sensing fusion non-line-of-sight identification. The method comprises the steps of obtaining communication data and RGB images of an unmanned aerial vehicle to-ground communication path; rGB image features and communication data features are extracted, and fusion features are obtained through fusion; inputting the fusion features into a trained neural network model, outputting a classification probability matrix, and obtaining classification results of the line-of-sight path and the non-line-of-sight path according to the classification probability matrix; if the classification result is a sight distance path, calculating a linear distance and a horizontal distance from the unmanned aerial vehicle to the ground target according to the positioning parameters to obtain positioning coordinates; and if the classification result is a non-line-of-sight path, correcting the positioning parameters, and obtaining positioning coordinates according to the corrected positioning parameters. According to the method, the LoS and NLoS paths can be accurately identified, and reliable technical support is provided for accurate positioning of the unmanned aerial vehicle in a ground communication scene.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technology, and in particular to a method and system for positioning an unmanned aerial vehicle (UAV) based on synaesthesia fusion and non-line-of-sight recognition. Background Art

[0002] Network positioning, as a key application of wireless communication networks, plays a vital role in navigation, military, commercial, and security rescue. Traditional network positioning is limited. Urban environments are complex, with dense buildings and heavy traffic. Traditional methods such as cellular network positioning and Wi-Fi positioning are affected by signal reflection, interference, and multipath effects, which reduce the precision of received signals and lead to poor positioning accuracy. Furthermore, base stations and Wi-Fi are deployed in fixed locations with limited coverage, making them difficult to meet the high-precision positioning requirements of urban environments. Compared with traditional network positioning methods, drone-assisted positioning relies on communication with ground equipment, measuring the distance between the drone and the ground equipment to determine the ground equipment's location, thereby achieving high-precision positioning. Therefore, drone positioning has greater advantages in urban environments.

[0003] UAV-to-ground communication can be divided into two environments: line-of-sight (LoS) and non-line-of-sight (NLoS). In LoS environments, signals propagate along a straight path, resulting in low propagation time and path loss, and relatively high positioning performance. However, in NLoS environments, signal propagation is blocked by obstacles, resulting in a longer propagation path, increased propagation time, and high path loss. This causes positioning technologies based on UAV-to-ground communication to overestimate range in NLoS environments, introducing positive bias and severely impacting positioning accuracy. Therefore, accurately classifying LoS and NLoS environments is a key step in designing high-precision positioning methods.

[0004] Existing methods for LoS and NLoS classification primarily fall into the following categories: one classifies LoS and NLoS by analyzing paths, scene maps, speed, and time; another class achieves this distinction by pre-designing channel characteristics; and yet another class distinguishes based on deterministic thresholds of channel characteristic parameters. However, all of these methods suffer from time-consuming or low classification accuracy. Furthermore, these studies largely focus on indoor and vehicle-to-vehicle (V2V) scenarios, lacking adaptability and specificity for the specific needs of drone-to-ground communications. With the rapid development of artificial intelligence (AI), which has demonstrated remarkable performance in solving classification problems, researchers have begun employing AI methods for LoS and NLoS classification, achieving some success. However, these studies, likewise, focus solely on indoor or vehicle-to-vehicle (V2V) scenarios, and their adaptability to drone-to-ground communications remains limited.

[0005] In addition, most of the existing LoS and NLoS classifications are based on communication data, and the obtained environmental information is limited. Since signal propagation is affected by physical factors such as buildings in the environment, it is difficult to comprehensively characterize the signal propagation environment only by communication data, which is not conducive to the accurate classification of LoS and NLoS, and further affects the accuracy of UAV positioning. Summary of the Invention

[0006] Aiming at the deficiencies of the existing technology, the present invention provides a UAV positioning method and system based on communication-sensing fusion non-line-of-sight recognition. By splicing and fusing RGB image features and communication data features and using a convolutional neural network for classification, the present invention can accurately identify LoS and NLoS paths, providing reliable technical support for precise positioning in the UAV-to-ground communication scenario.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] In the first aspect, the present invention provides a UAV positioning method based on communication-sensing fusion non-line-of-sight recognition, including:

[0009] Obtain the communication data and RGB images of the UAV-to-ground communication path;

[0010] Extract the RGB image features and communication data features, splice the RGB image features and the communication data features to obtain fusion features; input the fusion features into a trained neural network model, output a classification probability matrix, and obtain the classification results of the line-of-sight path and the non-line-of-sight path according to the classification probability matrix.

[0011] Obtain positioning parameters. If the classification result is a line-of-sight path, calculate the straight-line distance and horizontal distance from the UAV to the ground target according to the positioning parameters to obtain the positioning coordinates; if the classification result is a non-line-of-sight path, correct the positioning parameters and obtain the positioning coordinates according to the corrected positioning parameters.

[0012] As a further technical solution, the communication data of the UAV-to-ground communication path is obtained by the ray tracing method, and the line-of-sight and non-line-of-sight category labels of each communication path are obtained according to the number of reflections, diffractions, and scatterings of the rays in the path. Specifically: the communication path without reflections, diffractions, and scatterings is a line-of-sight path, and the label is 1; the path with reflections, diffractions, and scatterings is a non-line-of-sight path, and the label is 0.

[0013] As a further technical solution, a convolutional neural network is used to extract features from the RGB image, including: inputting the RGB image into the convolutional neural network and outputting the RGB image features; wherein, the convolutional neural network includes two convolutional layers, a batch normalization layer, and a pooling layer.

[0014] As a further technical solution, a convolution module is used to extract features from the communication data. The convolution module is a convolutional layer, including: inputting the communication data into the convolution module and outputting communication data features; wherein, the communication data is divided into a real part and an imaginary part, and the number of channels of the convolutional layer in the convolution module is set to 2, corresponding to the real part and the imaginary part of the communication data.

[0015] As a further technical solution, the trained neural network model includes a fusion training module and a classifier module. The fusion features are first input into the fusion training module, and after output, enter the classifier module for classification prediction; wherein, the fusion training module includes two convolutional layers and a batch normalization layer; the classifier module consists of one convolutional layer, and is used to map the data output by the fusion training module into the final classification result.

[0016] As a further technical solution, if the classification result is a line-of-sight path, the straight-line distance from the UAV to the ground target is: d = cT; where c is the speed of light and T is the time from when the UAV emits a signal to when the signal is received on the ground; the horizontal distance from the UAV to the ground target is: d h = d cosθ; where θ is the pitch angle of the UAV; the positioning coordinates are:

[0017]

[0018] where x u 、y u are the coordinates of the UAV, is the azimuth angle of the UAV.

[0019] As a further technical solution, if the classification result is a non-line-of-sight path, the true distance is calculated through a linear error model as: D = d - Δd NLoS = cT - Δd NLoS ; where Δd NLoS = k·d + b; k and b represent fitting coefficients; the corrected positioning parameters include correcting the pitch angle and azimuth angle of the UAV, and the corrected pitch angle and azimuth angle are respectively:

[0020]

[0021] The horizontal projection distance from the UAV to the ground target calculated according to the corrected pitch angle is: The positioning coordinates calculated according to the corrected azimuth angle are:

[0022]

[0023] In a second aspect, the present invention provides a UAV positioning system based on communication-sensing fusion non-line-of-sight recognition, including the following modules:

[0024] A data acquisition module, configured to: acquire communication data and RGB images of the communication path between the unmanned aerial vehicle (UAV) and the ground.

[0025] An identification and classification module, configured to: extract RGB image features and communication data features, splice the RGB image features and the communication data features to obtain fused features; input the fused features into a trained neural network model, output a classification probability matrix, and obtain classification results of the line-of-sight (LoS) path and the non-line-of-sight (NLoS) path according to the classification probability matrix.

[0026] A positioning module, configured to: acquire positioning parameters. If the classification result is the LoS path, calculate the straight-line distance and the horizontal distance from the UAV to the ground target according to the positioning parameters to obtain positioning coordinates; if the classification result is the NLoS path, correct the positioning parameters, and obtain positioning coordinates according to the corrected positioning parameters.

[0027] One or more technical solutions of the present invention have the following beneficial effects:

[0028] (1) Existing LoS and NLoS identification and classification methods have problems of long time consumption, low accuracy, and insufficient adaptability. The present invention extracts features from RGB images and communication data by using a convolutional neural network. The RGB image features and the channel impulse response (CIR) features are directly spliced and fused on the channel, and the fused data is input into the convolutional neural network again for classification to accurately identify the LoS and NLoS paths. The present invention can classify the LoS and NLoS of the communication path between the UAV and the ground in real time with high accuracy, providing a reliable technical support for precise positioning in the UAV-to-ground communication scenario.

[0029] (2) Traditional positioning technologies have large positioning errors in urban scenarios and NLoS environments. The present invention adopts different positioning algorithms according to the LoS and NLoS classification results. In the LoS path, direct high-precision positioning is performed by using the acquired positioning parameters; in the NLoS path, the error introduced by the NLoS path is effectively eliminated by correcting the positioning parameters, improving the positioning accuracy of the coordinates of the ground target object, and solving the problem of inaccurate positioning in complex scenarios. Description of the Drawings

[0030] The specification drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention.

[0031] Figure 1 It is a flowchart of the UAV positioning method of the present invention. Detailed Embodiments

[0032] It should be noted that the following detailed description is illustrative and aims to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention pertains.

[0033] Embodiment 1

[0034] In this embodiment, a UAV positioning method based on cross-sensing fusion non-line-of-sight recognition is provided. As Figure 1 shown, the steps of the specific method are as follows:

[0035] S1: Obtain the communication data and RGB images of the UAV's communication path to the ground;

[0036] S2: Extract the RGB image features and communication data features, splice the RGB image features and communication data features to obtain fusion features; input the fusion features into the trained neural network model, output the classification probability matrix, and obtain the classification results of the line-of-sight path and non-line-of-sight path according to the classification probability matrix;

[0037] S3: Obtain the positioning parameters. If the classification result is a line-of-sight path, calculate the straight-line distance and horizontal distance from the UAV to the ground target according to the positioning parameters to obtain the positioning coordinates; if the classification result is a non-line-of-sight path, correct the positioning parameters and obtain the positioning coordinates according to the corrected positioning parameters.

[0038] In step S1, it also includes obtaining the LoS and NLoS class labels of each communication path. Among them, the communication data is the channel CIR data collected by the electromagnetic simulation software Wireless InSite based on the ray tracing method, and the LoS and NLoS class labels of the path are obtained through the number of reflections, diffractions, and scatterings of the rays in the path. Specifically: the communication path without reflections, diffractions, and scatterings is the line-of-sight path, and the label is 1; the path with reflections, diffractions, and scatterings is the non-line-of-sight path, and the label is 0.

[0039] The method for obtaining the RGB image is to collect the RGB image data of the UAV's communication to the ground through the sensing simulation software AirSim. Since the flight altitude of the UAV can be adjusted, the range of the RGB image with a size of 1080×1080 can cover the entire receiving antenna grid with a size of 41×41.

[0040] In step S2, the original size of the RGB image is 1080×1080, the size of the receiving antenna grid is 41×41, and the dimension of the channel CIR data matrix is 41×41. There is a difference between the original size of the RGB image and the size of the receiving antenna grid. Therefore, it is necessary to extract the features of the RGB image first to avoid data redundancy. The specific method for extracting the RGB image features is:

[0041] The RGB image is subjected to feature extraction using a convolutional neural network. The specific process is as follows: The RGB image is input into the convolutional neural network, and RGB image features are output. Among them, the convolutional neural network includes two convolutional layers (Conv2d), a batch normalization layer (BatchNorm2d), and a pooling layer (MaxPool2d). The batch normalization layer performs batch normalization on the output channel data of the convolutional layer, making the data distribution stable and the training gradient more stable, and accelerating convergence. Among them, there is a batch normalization layer, an activation function, and a pooling layer behind each convolutional layer. Specifically: the number of input channels of the first convolutional layer in the two convolutional layers is 3, the convolutional kernel size is 3×3, the number of output channels is 64, the activation function is ReLU, and the dimension of the max pooling layer is 2×2; the number of input channels of the second convolutional layer is 64, the convolutional kernel size is 3×3, the number of output channels is 64, the activation function is ReLU, and the dimension of the max pooling layer is 2×2; the ReLU function is: f(x) = max(0, x), where x is the input of the ReLU function.

[0042] Using the above convolutional neural network to extract features from the RGB image, RGB image features with 64 channels can be obtained. Subsequently, an added adaptive pooling layer adjusts the size of the features to 41×41, facilitating subsequent concatenation with communication data.

[0043] In step S2, a convolutional module is used to extract features from the communication data. The convolutional module is a convolutional layer. The specific process includes: The communication data is input into the convolutional module, and communication data features are output. Among them, since the communication data is divided into two parts, the real part and the imaginary part, that is, two channels, the number of channels of the convolutional layer in the convolutional module is set to 2, corresponding to the real part and the imaginary part of the communication data. The size of the convolutional kernel is 3×3, the number of output channels is 64, and the activation function is ReLU. Finally, 64-channel features with a size of 41×41 are output.

[0044] After extracting the RGB image features and the communication data features respectively, the RGB image features and the communication data features are concatenated to obtain fused features. Specifically: In the above process, the RGB image and the communication data are respectively extracted as feature data with 64 channels and a size of 41×41; subsequently, these two types of feature data are concatenated, and the number of channels is correspondingly increased to 128, thereby realizing the fusion of features and obtaining fused features.

[0045] In step S2, the fused features are input into the trained neural network model, and a classification probability matrix is output. The specific method is as follows:

[0046] First, the trained neural network model includes a fusion training module and a classifier module. Then, in the above steps, the features extracted from the RGB image and communication data are concatenated to form a fused feature with 128 channels and a size of 41×41. This fused feature is input into the fusion training module, and then data with 256 channels and a size of 41×41 is output. This data is input into the classifier module for classification prediction, and then a classification probability matrix with a size of 41×41 is output.

[0047] The above neural network model including the fusion training module and the classifier module can process 128-channel input data and perform effective feature fusion and classification prediction. The input dimension of the fusion training module is B×128×H×W (where B is the batch size, 128 represents the number of channels, and H and W are the spatial dimensions of the input). After the output data is input into the classifier, the classification probabilities of multiple communication channels are obtained. The specific network structure is as follows:

[0048] The fusion training module includes two convolutional layers and a batch normalization layer. The batch normalization layer normalizes the output channel data of the convolutional layer to make the data distribution stable, the training gradient more stable, and accelerate convergence. And there is a batch normalization layer, an activation function, and a pooling layer behind each convolutional layer. The number of input channels of the first convolutional layer in the fusion training module is 128, the convolutional kernel size is 3×3, the number of output channels is 128, and the activation function is ReLU. The number of input channels of the second convolutional layer in the fusion training module is 128, the convolutional kernel size is 3×3, the number of output channels is 256, and the activation function is ReLU. The ReLU function is: f(x) = max(0, x), where x is the input of the ReLU function.

[0049] The classifier module consists of a convolutional layer, which is used to map the data output by the fusion training module to the final classification result. The B×256×H′×W′ data output by the fusion training module is input into the classifier module, and then a value of 41×41 is output, and the class probability is output after passing through the Sigmoid function. The Sigmoid function is expressed as follows:

[0050] where x is the input of the Sigmoid function.

[0051] In step S2, after extracting features from the RGB image and communication data respectively, the RGB image features and communication data features are concatenated in the channel dimension to obtain fused features. Subsequently, the fused features are input into a pre-trained fusion training module and a classifier. Finally, a classification probability matrix with a size of 41×41 is output. According to the probability values in the classification probability matrix, the paths with probabilities greater than 0.5 are classified as LoS paths, and the paths with probabilities less than or equal to 0.5 are classified as NLoS paths, thereby realizing the LoS and NLoS identification and classification of communication paths.

[0052] Many positioning systems, such as the global positioning system, calculate distances and determine target positions based on measuring the one-way propagation time of signals between the transmitting antenna and the receiving antenna. Under LoS paths, signals can directly travel from the transmitting end to the receiving end. Therefore, the measurement errors of parameters such as Time of Arrival (TOA), Time Difference of Arrival (TDOA), and Angle of Arrival (AOA) are relatively small, and thus can be directly used for high-precision positioning. However, in dynamic scenarios, LoS paths will be restricted to NLoS paths due to the presence of obstacles between the transmitter and receiver. NLoS paths introduce positive biases in distance estimation for positioning, leading to inaccurate or incorrect positioning.

[0053] In step S3, positioning parameters are obtained, including the coordinates of the UAV, the azimuth angle of the UAV, the pitch angle of the UAV, and the time from when the UAV transmits a signal to when the signal is received on the ground, etc. According to the classification results obtained in step S2, different positioning algorithms are used for the two types of paths, and the specific method is as follows:

[0054] If the classification result is a line-of-sight path, first, calculate the straight-line distance from the UAV to the ground target:

[0055] d = cT;

[0056] where c is the speed of light and T is the time from when the UAV transmits a signal to when the signal is received on the ground.

[0057] Then calculate the horizontal distance from the UAV to the ground target as:

[0058] d h = d cosθ;

[0059] where θ is the pitch angle of the UAV.

[0060] Finally, calculate the ground coordinates of the target:

[0061]

[0062] Among them, x u and y u are the coordinates of the drone, and

[0063] is the azimuth angle of the drone. If the classification result is a non-line-of-sight path, the NLoS path is affected by obstacles, and the signal propagates through reflection, refraction, and diffraction. The signal propagation distance is not equal to the geometric distance, resulting in errors in the positioning parameters. Therefore, to correct the positioning parameters, the specific method is as follows:

[0064] First, calculate the true distance through a linear error model as:

[0065] D = d - Δd NLoS = cT - Δd NLoS ; where Δd NLoS = k·d + b; k and b represent fitting coefficients; correcting the positioning parameters includes correcting the pitch angle and azimuth angle of the drone. The corrected pitch angle and azimuth angle are respectively:

[0066]

[0067] The horizontal projection distance from the drone to the ground target is calculated according to the corrected pitch angle as: The positioning coordinates are calculated according to the corrected azimuth angle as:

[0068]

[0069] where Δd NLoS is a linear error model obtained by fitting actual measurement data, and Δθ NLoS and are angle errors predicted according to historical data through machine learning methods. By eliminating the errors generated by NLoS in the above manner, more accurate position coordinates can be obtained.

[0070] In summary, by fusing perception and communication data, using a neural network to classify LoS and NLoS paths, and then adopting corresponding positioning calculation methods for different paths, the error influence can be effectively reduced and the positioning accuracy can be improved.

[0071] Embodiment 2

[0072] In this embodiment, a drone positioning system based on communication-sensing fusion non-line-of-sight recognition is provided, including the following modules:

[0073] A data acquisition module, configured to: acquire communication data and RGB images of the communication path between the drone and the ground;

[0074] The recognition and classification module is configured to: extract RGB image features and communication data features, splice the RGB image features and the communication data features to obtain fused features; input the fused features into a trained neural network model, output a classification probability matrix, and obtain classification results of the LOS path and the NLOS path according to the classification probability matrix.

[0075] The positioning module is configured to: obtain positioning parameters. If the classification result is the LOS path, calculate the straight-line distance and the horizontal distance from the UAV to the ground target according to the positioning parameters to obtain positioning coordinates; if the classification result is the NLOS path, correct the positioning parameters and obtain positioning coordinates according to the corrected positioning parameters.

[0076] For those skilled in the art, various changes and modifications can be made to the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A UAV positioning method based on cross-sensing fusion non-line-of-sight recognition, characterized in that, Including: Obtaining communication data and RGB images of the communication path between the UAV and the ground; Extracting RGB image features and communication data features, splicing the RGB image features and the communication data features to obtain fused features; inputting the fused features into a trained neural network model, outputting a classification probability matrix, and obtaining classification results of the line-of-sight path and the non-line-of-sight path according to the classification probability matrix; Obtaining positioning parameters. If the classification result is the line-of-sight path, calculating the straight-line distance and the horizontal distance from the UAV to the ground target according to the positioning parameters to obtain positioning coordinates; if the classification result is the non-line-of-sight path, correcting the positioning parameters and obtaining positioning coordinates according to the corrected positioning parameters.

2. The UAV positioning method based on the non-line-of-sight recognition of the integration of communication and sensing according to claim 1, wherein, Obtaining the communication data of the communication path between the UAV and the ground through the ray tracing method, and obtaining the line-of-sight and non-line-of-sight category labels of each communication path according to the number of reflections, diffractions, and scatterings of the rays in the path. Specifically: the communication path without reflections, diffractions, and scatterings is the line-of-sight path with a label of 1, and the path with reflections, diffractions, and scatterings is the non-line-of-sight path with a label of 0.

3. The UAV positioning method based on the integrated perception and communication non-line-of-sight recognition according to claim 1, wherein, Using a convolutional neural network to extract features from the RGB image, including: inputting the RGB image into the convolutional neural network and outputting RGB image features; wherein, the convolutional neural network includes two convolutional layers, a batch normalization layer, and a pooling layer.

4. The UAV positioning method based on the integrated communication and sensing non-line-of-sight recognition according to claim 1, characterized in that Using a convolutional module to extract features from the communication data. The convolutional module is a convolutional layer, including: inputting the communication data into the convolutional module and outputting communication data features; wherein, the communication data is divided into a real part and an imaginary part, and the number of channels of the convolutional layer in the convolutional module is set to 2, corresponding to the real part and the imaginary part of the communication data.

5. The UAV positioning method based on the cross-sensing fusion non-line-of-sight recognition according to claim 1, characterized in that, The trained neural network model includes a fusion training module and a classifier module. The fused features are first input into the fusion training module, and after output, enter the classifier module for classification prediction; wherein, the fusion training module includes two convolutional layers and a batch normalization layer; the classifier module consists of a convolutional layer and is used to map the data output by the fusion training module to the final classification result.

6. The UAV positioning method based on cross-sensing fusion non-line-of-sight recognition according to claim 1, wherein, If the classification result is a line-of-sight path, the straight-line distance from the UAV to the ground target is: d = cT; where c is the speed of light and T is the time from when the UAV emits a signal to when the signal is received on the ground; the horizontal distance from the UAV to the ground target is: d h = dcosθ; where θ is the pitch angle of the UAV; the positioning coordinates are: Among them, x u , y u are the coordinates of the drone, and is the azimuth angle of the drone.

7. The UAV positioning method based on cross-sensing fusion non-line-of-sight recognition according to claim 1, wherein, If the classification result is a non-line-of-sight path, the true distance is calculated by the linear error model as: D = d - Δd NLoS = cT - Δd NLoS ; where, Δd NLoS = k·d + b; k and b represent fitting coefficients; the corrected positioning parameters include correcting the pitch angle and azimuth angle of the UAV, and the corrected pitch angle and azimuth angle are respectively: θ′ = θ - Δθ NLoS ; The horizontal projection distance from the UAV to the ground target calculated according to the corrected pitch angle is: d □, = Dcosθ′; The positioning coordinates calculated according to the corrected azimuth angle are:

8. An unmanned aerial vehicle positioning system based on cross-sensory fusion non-line-of-sight recognition, characterized in that, Including the following modules: A data acquisition module configured to: obtain communication data and RGB images of the communication path between the UAV and the ground; An identification and classification module configured to: extract RGB image features and communication data features, splice the RGB image features and the communication data features to obtain fused features; input the fused features into a trained neural network model, output a classification probability matrix, and obtain classification results of the line-of-sight path and the non-line-of-sight path according to the classification probability matrix; A positioning module configured to: obtain positioning parameters. If the classification result is the line-of-sight path, calculate the straight-line distance and the horizontal distance from the UAV to the ground target according to the positioning parameters to obtain positioning coordinates; if the classification result is the non-line-of-sight path, correct the positioning parameters and obtain positioning coordinates according to the corrected positioning parameters.

9. A computer-readable storage medium having a program stored thereon, characterized in that, When the program is executed by a processor, it implements the steps in a UAV positioning method based on communication-sensing fusion non-line-of-sight identification according to any one of claims 1-7.

10. An electronic device, comprising a memory, a processor, and a program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in a UAV positioning method based on integrated communication and sensing non-line-of-sight recognition as described in any one of claims 1-7.

Images