Indoor multi-target visible light positioning method based on compressed sensing

Abstract

The invention discloses an indoor multi-target visible light positioning method based on compressed sensing, and relates to the technical field of visible light communication.The method comprises the steps that visible light signals are converted into sparse vectors, and autocorrelation signal vectors in the multiple visible light signals are aggregated to obtain the total power of the signals; when the compressed sensing algorithm CS measures indoor visible light transmission, an observation matrix is formed, measurement power is given to the observation matrix based on the total power, and a plurality of sparse vectors are recovered by using the measurement power; aggregating cross-correlation measurement vectors in the plurality of visible light signals, reconstructing sparse vectors to obtain a plurality of positioning sparse vectors, identifying positions of non-zero elements in the positioning sparse vectors, and determining positions of a plurality of targets; according to the process, sparsity and cross correlation in multi-target visible light signals are mined, a multi-target positioning problem in a complex environment is converted into a sparse recovery problem by using a compressed sensing (CS) method, and multi-target signals are distinguished and identified by using sparse weak signals, so that high-precision multi-target positioning is realized.

Description

Technical Field

[0001] The present invention relates to the field of visible light communication technology, and in particular to a method, device, equipment and medium for indoor multi-target visible light positioning based on compressed sensing. Background Art

[0002] With the rapid development of Industrial Internet of Things (IIoT) and smart manufacturing, indoor positioning systems (RTLS) play an increasingly important role in smart factories, automated warehousing and logistics management. Accurate multi-target positioning technology is essential for improving production efficiency, optimizing equipment scheduling and ensuring operational safety. However, existing indoor positioning technologies (such as methods based on RSS, TOA, and AOA) have problems such as insufficient positioning accuracy, high computational complexity and poor robustness in complex environments. Especially in scenarios where multiple targets are positioned simultaneously, existing methods often cannot meet the requirements of high precision and real-time performance.

[0003] At present, indoor positioning systems mainly rely on traditional methods based on received signal strength (RSS), arrival time (TOA) and arrival angle (AOA). These technologies can provide high accuracy in single-target positioning, but face significant challenges and limitations in scenarios where multiple targets are positioned simultaneously. These include: ① Decreased positioning accuracy due to environmental complexity. In complex indoor environments, traditional positioning methods are susceptible to interference from factors such as multipath effects, non-line-of-sight propagation (NLOS) and environmental noise. These problems cause distortion of received signal characteristics, which significantly reduces positioning accuracy. ② Limitations in multi-target scenarios. In the case of simultaneous positioning of multiple targets, existing methods usually exhibit high computational complexity and low efficiency. ③ Signal sparsity problem. Although compressed sensing (CS) technology has been The sparse characteristics of LED signals are introduced to achieve efficient positioning, but in complex environments, the sparseness assumption of signals may not be completely valid, thus affecting the positioning effect; ④ The influence of noise and interference: Background noise and signal interference in complex indoor environments will still significantly affect the positioning accuracy, and the existing methods lack effective robustness mechanisms to deal with these challenges; ⑤ Real-time and multi-target positioning accuracy: The existing VLP system has insufficient real-time processing capabilities in multi-target scenarios, and it is difficult to simultaneously achieve high-precision and low-latency positioning requirements in highly dynamic environments; ⑥ Insufficient technology integration: Although generative adversarial networks (GANs) as a deep learning method have shown significant advantages in image processing and signal recovery, how to effectively apply GAN technology to VLP systems is still in the initial exploration stage.

[0004] Therefore, in complex environments, the received signals of multiple targets are often affected by geometric structures and signal interference, resulting in distortion of signal characteristics of the received signals of multiple targets, accompanied by the problem of signal overlap. The current methods make it difficult to distinguish and identify the signals of multiple targets, making it difficult to achieve high-precision simultaneous positioning of multiple targets. Summary of the invention

[0005] The embodiment of the present invention provides an indoor multi-target visible light positioning method based on compressed sensing, which can solve the problem in the prior art that it is difficult to distinguish and identify the signals of multiple targets in the current method, thereby making it difficult to achieve high-precision simultaneous positioning of multiple targets.

[0006] The embodiment of the present invention provides a method for indoor multi-target visible light positioning based on compressed sensing, comprising the following steps:

[0007] Build an indoor multi-target visible light positioning VLP system, where multiple targets receive visible light signals simultaneously to form a visible light signal data set;

[0008] Convert multiple visible light signals into multiple sparse vectors, where the non-zero element position of the sparse vector corresponds to the position of the target; aggregate the autocorrelation signal vectors in the multiple visible light signals to obtain the total power of the multiple visible light signals; when the compressed sensing algorithm CS is used to measure the visible light transmission in the room, an observation matrix is ​​formed, and a measurement power is assigned to each element in the observation matrix based on the total power; use the measurement power assigned by the observation matrix to restore multiple sparse vectors to form multiple preliminary positioning sparse vectors;

[0009] According to the cooperative relationship between different emitting light sources in the system, the cross-correlation measurement vectors in multiple visible light signals are aggregated, and the preliminary positioning sparse vector is reconstructed according to the cross-correlation measurement vector to obtain multiple positioning sparse vectors;

[0010] The positions of the multiple targets are determined according to the positions of the non-zero elements in the multiple positioning sparse vectors.

[0011] Preferably, obtaining the total power of the plurality of visible light signals comprises:

[0012] Aggregate the received signal vectors of multiple visible light signals to form an autocorrelation signal vector y, which is expressed as:

[0013]

[0014] Where: ω represents the total background noise vector; θ represents the target positioning vector on the grid, θ=[θ 1 ,...,θ j ,...,θ N ] T is an indicator vector with N elements;

[0015] The observation matrix φ in the compressed sensing algorithm CS is expressed as:

[0016]

[0017] By aggregating the autocorrelation signal vector y of multiple visible light signals, the total power P of multiple visible light signals is obtained. rx , expressed as:

[0018]

[0019] Where: E{·} represents the statistical expectation operator; ⊙ represents the Hadamard product operator; (·) * represents the complex conjugate operator; represents the variance background noise; 1 M Represents an M-length vector of full values.

[0020] Preferably, the recovering of multiple sparse vectors to form multiple preliminary positioning sparse vectors includes:

[0021] When the compressed sensing algorithm CS measures the indoor visible light channel and propagation environment, it forms an observation matrix J, which is expressed as:

[0022]

[0023] Based on the total power, each element in the observation matrix J is assigned a measurement power, and the measurement power assigned by the observation matrix is ​​used to restore multiple sparse vectors to form multiple preliminary positioning sparse vectors.

[0024] Preferably, the reconstructing the preliminary positioning sparse vector includes:

[0025] According to the cooperative relationship between different emitting light sources in the indoor multi-target visible light positioning VLP system, the cross-correlation measurement vector P in multiple visible light signals is aggregated. corr , expressed as:

[0026]

[0027] in:(·) H represents the conjugate transpose operator; represents the Kronecker product operator; represents the Khatri-Rao product; I M Represents the identity matrix of size M×M; the cross-correlation measurement vector P corr is length-M 2 ;

[0028] Define an M(M+1) / 2×M 2 The selection matrix S is of size M×M; each row of the selection matrix S corresponds to P corrThere are M independent diagonal elements and M(M-1) / 2 upper diagonal elements. After the selection process is completed, a cross-correlation measurement model based on the compressed sensing algorithm CS is formulated, which is expressed as:

[0029]

[0030] Among them, the size of matrix ψ is m(m+1) / 2×N, which represents the observation matrix; the matrix ψ is expressed as:

[0031]

[0032] The number of independent linear equations used in the cross-correlation measurement model based on the compressed sensing algorithm CS increases from M to M(M+1) / 2, and the measurement data that can be used for sparse recovery based on the compressed sensing algorithm CS gradually increases;

[0033] A preliminary positioning sparse vector is reconstructed according to the cross-correlation measurement model to obtain multiple positioning sparse vectors.

[0034] Preferably, it also includes:

[0035] Using a generative adversarial network (GAN) to identify a received signal vector in a plurality of aggregated visible light signals, generating a positioning vector, and using the positioning vector to correct a plurality of positioning sparse vectors to obtain a plurality of final positioning sparse vectors; comprising the following steps:

[0036] The received signal vector y of the aggregated multiple visible light signals is used as the initial input of the generator of the generative adversarial network GAN to replace the random variable z in the generative adversarial network GAN, and two generators are used to gradually establish an explicit mapping relationship from the received signal vector y to the target positioning vector θ;

[0037] Generator G1(·) generates the corresponding potential channel matrix H according to the input y, and inputs the potential channel matrix H into the pre-trained generator G2(·) to generate a sparse positioning vector containing position information

[0038] A regularization term is added to the loss function of the generative adversarial network GAN to reduce the error between the generated vector and the true vector, forming a composite loss function Lclf, which is expressed as:

[0039]

[0040] in:

[0041] Generate a sparse positioning vector based on The multiple positioning sparse vectors are corrected to obtain multiple final positioning sparse vectors.

[0042] The embodiment of the present invention further provides an indoor multi-target visible light positioning device based on compressed sensing, comprising:

[0043] The acquisition module is used to build an indoor multi-target visible light positioning VLP system. Multiple targets receive visible light signals at the same time to form a visible light signal data set;

[0044] The recognition module is used to convert multiple visible light signals into multiple sparse vectors, where the non-zero element position of the sparse vector corresponds to the position of the target; aggregate the autocorrelation signal vectors in the multiple visible light signals to obtain the total power of the multiple visible light signals; when the compressed sensing algorithm CS is used to measure the visible light transmission in the room, an observation matrix is ​​formed, and a measurement power is assigned to each element in the observation matrix based on the total power; the measurement power assigned by the observation matrix is ​​used to restore the multiple sparse vectors to form multiple preliminary positioning sparse vectors;

[0045] The positioning module is used to aggregate the cross-correlation measurement vectors in multiple visible light signals according to the cooperative relationship between different emitting light sources in the system, and reconstruct the preliminary positioning sparse vector according to the cross-correlation measurement vector to obtain multiple positioning sparse vectors.

[0046] An embodiment of the present invention further provides an electronic device, including a memory and a processor;

[0047] The memory is used to store computer programs;

[0048] The processor is used to implement the steps of the indoor multi-target visible light positioning method based on compressed sensing when executing the computer program stored in the memory.

[0049] An embodiment of the present invention further provides a computer-readable storage medium for storing a computer program, which, when executed by a processor, implements the steps of the indoor multi-target visible light positioning method based on compressed sensing as described above.

[0050] The embodiment of the present invention provides a method for indoor multi-target visible light positioning based on compressed sensing. Compared with the prior art, the method has the following beneficial effects:

[0051] The present invention converts multiple visible light signals into multiple sparse vectors, aggregates the autocorrelation signal vectors in the multiple visible light signals to obtain the total power of the multiple visible light signals; when the compressed sensing algorithm CS measures the visible light transmission in the room, an observation matrix is ​​formed, and a measurement power is assigned to each element in the observation matrix based on the total power; the measurement power assigned by the observation matrix is ​​used to restore multiple sparse vectors; the cross-correlation measurement vectors in the multiple visible light signals are aggregated, and the sparse vectors are reconstructed according to the cross-correlation measurement vectors to obtain multiple positioning sparse vectors, thereby identifying the positions of non-zero elements in the positioning sparse vectors and determining the positions of multiple targets; the process converts multiple visible light signals into multiple sparse vectors, aggregates the total power of the autocorrelation signal vectors in the multiple visible light signals to form the measurement power of the observation matrix to restore the sparse vector, and aggregates the cross-correlation measurement vectors in the multiple visible light signals, and reconstructs the sparse vector according to the cross-correlation measurement vector. The process described above converts the multi-target positioning problem in a complex environment into a sparse recovery problem by exploiting the sparsity and cross-correlation in the multi-target visible light signals, and the sparse weak signals can be used to distinguish and identify the signals of the multiple targets, so as to achieve high-precision multi-target positioning.

[0052] In addition, the present invention utilizes a generative adversarial network (GAN) to identify the positioning vector within the received signal vector to correct multiple positioning sparse vectors; the generative adversarial network (GAN) enables the generator to directly extract sparse location information from the received signal through an adversarial training mechanism while suppressing background noise and interference signals; this adversarial learning framework not only improves the robustness of the positioning system, but also significantly enhances its adaptability in complex dynamic environments, thereby further optimizing the sparse recovery process of compressed sensing. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A schematic diagram of the overall process of a method for indoor multi-target visible light positioning based on compressed sensing provided by an embodiment of the present invention;

[0054] Figure 2 A schematic diagram of a typical multi-target VLP system of an indoor multi-target visible light positioning method based on compressed sensing provided by an embodiment of the present invention;

[0055] Figure 3 A schematic diagram of a visible light position estimation technology based on collaborative compressed sensing for an indoor multi-target visible light positioning method based on compressed sensing provided by an embodiment of the present invention;

[0056] Figure 4 A schematic diagram of the CS-GAN technology of a compressed sensing-based indoor multi-target visible light positioning method provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0057] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention, so the present invention is not limited by the specific embodiments disclosed below.

[0058] See also Figure 1 The embodiment of the present invention provides an indoor multi-target visible light positioning method based on compressed sensing; in view of the shortcomings of the existing visible light positioning (VLP) technology in multi-target positioning in terms of accuracy, noise resistance and computational efficiency, especially in complex environments affected by geometric structures and signal interference, the existing methods are difficult to achieve high-precision multi-target simultaneous positioning; the current technology cannot guarantee centimeter-level positioning accuracy in environments with dense targets or high noise, and exhibits poor robustness and efficiency in applications with high real-time requirements. Specifically, it is embodied in:

[0059] Decreased positioning accuracy due to environmental complexity: In complex indoor environments, traditional positioning methods are susceptible to interference from factors such as multipath effects, non-line-of-sight propagation (NLOS), and environmental noise. These problems cause distortion of received signal characteristics, which significantly reduces positioning accuracy. In addition, target movement and signal changes in dynamic environments make it difficult for traditional methods to quickly adapt and adjust, limiting their robustness and reliability in practical applications.

[0060] Limitations in multi-target scenarios: In the case of simultaneous positioning of multiple targets, existing methods usually exhibit high computational complexity and low efficiency. For example, the RSS-based positioning method cannot effectively distinguish the signal characteristics of each target in a multi-target scenario, especially in an environment with overlapping signals and high noise, where its computational complexity is further increased, and positioning accuracy and real-time performance are significantly reduced. Even if the RF fingerprint method is used, it is easy to cause identification confusion and misjudgment due to similar signal characteristics or device heterogeneity, which makes it difficult to meet the needs of actual industrial scenarios.

[0061] Signal sparsity problem: Although compressed sensing (CS) technology has been introduced to utilize the sparse characteristics of LED signals to achieve efficient positioning, in complex environments, the signal sparsity assumption may not be completely valid, thus affecting the positioning effect.

[0062] Impact of noise and interference: Background noise and signal interference in complex indoor environments will still significantly affect positioning accuracy, and existing methods lack effective robustness mechanisms to deal with these challenges.

[0063] Real-time and multi-target positioning accuracy: The existing VLP system has insufficient real-time processing capabilities in multi-target scenarios, and it is difficult to simultaneously achieve high-precision and low-latency positioning requirements in highly dynamic environments.

[0064] Insufficient integration of new technologies: Although generative adversarial networks (GANs) as a deep learning method have shown significant advantages in image processing and signal recovery, how to effectively apply GAN technology to VLP systems is still in the initial exploration stage. Specifically, although the adversarial training framework of GAN can enhance anti-interference capabilities, its computational complexity is high, and its adaptability and stability in positioning tasks have not been fully verified. In addition, the potential of the collaborative optimization of GAN and CS technology in multi-target rapid positioning has not been fully explored, making it difficult for existing systems to achieve significant performance breakthroughs.

[0065] The present invention designs a multi-target collaborative VLP framework, which uses the existing indoor lighting system as the positioning signal source, and realizes high-precision positioning of multiple indoor targets by multi-point reception of LED signals combined with compressed sensing theory. Under this framework, the present invention transforms the positioning problem into a sparse recovery problem, and optimizes the recovery of sparse matrices through the CS algorithm, thereby improving the positioning accuracy, especially in complex indoor environments.

[0066] Secondly, in order to solve the problem that the positioning accuracy is affected by noise and signal distortion in the prior art, the present invention introduces the generative adversarial network (GAN) technology; by training the GAN model, we can significantly improve the sparse feature recovery accuracy of the signal and enhance the anti-interference ability of the positioning system in a low signal-to-noise ratio environment; the introduction of GAN can not only optimize the accuracy of sparse recovery, but also accurately identify and locate the positions of multiple targets in dynamic and complex environments, thereby improving the stability and reliability of the system.

[0067] Finally, the present invention further introduces a collaborative positioning mechanism between targets by implementing the system in an indoor environment; this mechanism ensures the real-time positioning of multiple targets, which not only improves the positioning accuracy, but also optimizes the real-time response capability of the system; through the technical integration of compressed sensing and generative adversarial networks, the present invention can effectively overcome the limitations of traditional VLP systems in multi-target positioning, and provide a fast, accurate and stable indoor positioning solution; the application of this technology provides strong support for the real-time positioning of multiple targets in the fields of intelligent manufacturing, automated warehousing, and industrial Internet of Things.

[0068] The present invention proposes a multi-target fast visible light positioning method based on the combination of compressed sensing (CS) and generative adversarial network (GAN), and innovatively improves the problems of insufficient multi-target positioning accuracy, anti-interference ability and real-time performance of the prior art in complex indoor environments; the technical solution of the present invention includes the design of a multi-target collaborative VLP framework, a sparse recovery algorithm based on CS, a collaborative mechanism among multiple targets, and the application of GAN in optimizing signal sparsity characteristics and enhancing anti-interference ability; by effectively integrating CS and GAN technologies, the present invention significantly improves the positioning accuracy and efficiency of the system in a low signal-to-noise ratio and high interference environment, and optimizes the collaboration among targets to achieve real-time dynamic positioning.

[0069] Specifically:

[0070] 1. Multi-target visible light positioning under the collaborative compressed sensing framework.

[0071] 1. Introduce compressed sensing theory into visible light multi-target positioning.

[0072] In a typical downlink visible light communication (VLP) system, such as Figure 2 As shown in the figure, a multi-objective VLP framework based on compressed sensing (CS) is introduced and defined as CSM-VLP.

[0073] Specifically, a generalized indoor VLP scenario is considered here, and it is assumed that the number of target terminals is K, where K is much smaller than the number of grids N (i.e., K<<N); the pilot signal of the i-th LED sent from all K target terminals can be aggregated through cooperation between targets (such as wireless communication links); after aggregating the K terminal signals, it is defined by the following formula:

[0074]

[0075] Where: i represents the sum of the background noise at the terminal of the transmission link from the i-th to the K-th target.

[0076] Then, a total reception is formulated by combining all the aggregated received signal vectors ysignals Corresponding to all M LEDs, y can be rewritten into matrix form as:

[0077]

[0078] Where: ω represents the total background noise vector.

[0079] The multi-target visible light communication (VLP) framework based on compressed sensing (CS) can be expressed as the above formula; this formula takes advantage of the sparsity of the target location in the indoor area relative to the entire grid point, and the specific form is as follows:

[0080]

[0081] Among them, the target positioning vector on the grid θ=[θ 1 ,...,θ j ,...,θ N ] T Is an indicator vector with N elements, each element corresponds to an area in the grid, and the value of each element indicates whether there is a target at the corresponding grid point; therefore, most elements are mostly zero, except for the elements corresponding to the K target positions to be located, which are 1; since the proportion of non-zero elements in the vector θ is much smaller than the length of the vector, θ is a sparse vector and can be recovered by sparse recovery methods. In the measurement model based on compressed sensing, the observation matrix φ is defined as:

[0082]

[0083] In order to improve the model based on compressed sensing (CS) and make it more universal in different scenarios, the present invention obtains the total received power by aggregating the autocorrelation signal vector y of the received signal, which is expressed as the received power P rx , expressed as:

[0084]

[0085] Where: E{·} represents the statistical expectation operator; ⊙ represents the Hadamard product operator; (·) * represents the complex conjugate operator; represents the variance background noise; 1 M Represents an M-length vector of full values.

[0086] Then, an improved power measurement model based on compressed sensing (CS) is established, in which the observation matrix J can also be regarded as a database of indoor visible light channels and propagation environments for positioning, after the following processing:

[0087]

[0088] It is observed that the elements in J are related to the channel gain. In the positioning stage, an improved power measurement model based on compressed sensing (CS) is obtained through a downlink VLP system with inter-target cooperation; it can be seen that this is actually a sparse recovery problem, which aims to measure the power through a given observation matrix J and recover the online grid target positioning sparse vector θ; the grid points corresponding to the positions of non-zero elements are the estimated target terminal positions; the sparse recovery problem can be solved using l 1 -norm minimization methods, or greedy algorithms based on compressed sensing, such as Orthogonal Matching Pursuit (OMP).

[0089] In fact, the accuracy of the proposed scheme is closely related to the step size of the grid points. A denser grid segmentation will result in a smaller quantization error of the position. Properly increasing the number of grid points can improve the positioning resolution. In addition, increasing N will also make θ a sparser vector, which is easier to recover through a compressed sensing (CS)-based algorithm. However, a smaller grid step size does not necessarily lead to better compressed sensing-based positioning performance. According to compressed sensing theory, the sparse recovery problem can only guarantee the recovery of a positive constant when the condition M ≥ μKlog (N / K) is met. Therefore, the values ​​of M and N are selected according to the actual situation.

[0090] 2. CS-GAN theory.

[0091] Compressed sensing (CS) is a method for reconstructing signals from a small amount of measurement data, and is particularly suitable for sparse signals. Its basic principle is to use the sparsity of the signal and an appropriate measurement matrix to restore the complete signal from a small number of linear measurements. Usually, this involves solving an optimization problem to find the signal that best fits the measurement results and satisfies the sparsity constraints.

[0092] In recent years, generative adversarial networks (GANs) have been introduced into the compressed sensing framework due to their powerful capabilities in generative modeling and data distribution learning. GANs consist of two neural networks, a generator and a discriminator. The generator aims to generate samples similar to the real data distribution, while the discriminator is responsible for distinguishing between real samples and generated samples. Through this adversarial training, GANs are able to learn complex distributions and generate high-quality data.

[0093] CS-GAN combines the advantages of compressed sensing and generative adversarial networks, aiming to improve the accuracy and robustness of signal reconstruction; this method relies on the sparsity of the signal, and the generator learns the mapping of the input data to generate outputs corresponding to the sparse signal, achieving efficient reconstruction; GAN's adversarial training mechanism enables the generator to continuously improve during the training process, and can better simulate the distribution of real signals, thereby overcoming the limitations of traditional CS algorithms under harsh conditions; in addition, the introduction of a composite loss function helps to optimize the adversarial loss and the reconstruction loss at the same time, ensuring the high quality and accuracy of the generated signal. CS-GAN is particularly suitable for tasks such as multi-target positioning, which requires recovering high-dimensional information from sparse observations and can effectively combine the spatial and temporal characteristics of the signal to adapt to various complex application scenarios.

[0094] 2. Specific plan design.

[0095] 1. Multi-target visible light positioning based on enhanced collaborative compressed sensing.

[0096] Although the CSM-VLP scheme has exploited the inter-target information to formulate aggregated measurement data, the potential geometric information of the indoor environment and the channel corresponding to the information of different anchors, i.e., LEDs, has not been fully utilized to enhance the robustness of localization; therefore, the present invention further designs a cooperative CSMVLP (CoCSM-VLP) scheme through cross-correlation, such as Figure 3 As shown, the inter-anchor collaboration is exploited between the corresponding aggregated received signals to achieve better multi-target joint localization performance.

[0097] Specifically, the cross-correlation P is calculated by the following formula: corr To obtain the aggregated received signal provided by the inter-anchor correlation:

[0098]

[0099] in:(·) H Represents the conjugate transpose operator.

[0100] Then, both sides of the above equation are vectorized to derive the cross-correlation measurement vector. The calculation process is expressed as:

[0101]

[0102] in: represents the Kronecker product operator; represents the Khatri-Rao product; I M Represents the identity matrix of size M×M; the cross-correlation measurement vector P corr is length-M 2 .

[0103] In fact, the problem given by the above formula is a system of linear equations, in which the independent mathematical equations have a direct impact on the sparse recovery performance based on compressed sensing (CS); this is actually determined by the CS theory based on the coherence of the measurement matrix; since the cross-correlation matrix P corr is symmetric, then we can define a M(M+1) / 2×M 2 The selection matrix S is of size M×M; each row of the selection matrix corresponds to P corr M independent diagonal elements and M(M-1) / 2 upper diagonal elements; after the selection process is completed, the CS-based cross-correlation measure can be formulated to satisfy the given model. The model is expressed as:

[0104]

[0105] Among them, the size of the matrix ψ m(m+1) / 2×N represents the observation matrix, that is, the fingerprint database CoCSM-VLP scheme based on the cross-correlation measurement model based on compressed sensing in the above formula; compared with the observation matrix J, the number of rows of the matrix ψ is much higher, which means that more independent equations can be used to solve the support; that is, more independent equations can be used to determine the position of the unknown sparse non-zero elements. The location vector θ; all elements in the matrix ψ are also related to the channel gain {h ij} related, so through the advance indoor channel estimation, the information required to construct the matrix ψ can be collected, which is given; the matrix ψ is expressed as:

[0106]

[0107] The number of independent linear equations available for the cross-correlation measurement model based on compressed sensing increases from M to M(M+1) / 2 compared to the measurement model; therefore, the measurement data available for sparse recovery based on compressed sensing gradually increases, which will directly benefit the multi-target localization based on N. Since the amount of measurement data in the cross-correlation measurement model, i.e., M(M+1) / 2, is much larger than M, it is easier to meet the requirements; this will improve the accuracy and reliability of the proposed scheme, especially under harsh conditions with a large number of unknowns, such as a large number of targets (large K), dense grid networks (large N), and strong background noise (low signal-to-noise ratio).

[0108] After this, the sparse localization vector can be reconstructed via an efficient sparse recovery problem, similar to CSM-VLP; this can be done by using l 1 -norm minimization method or greedy algorithm based on compressed sensing, such as Orthogonal Matching Pursuit (OMP) implementation.

[0109] 2. Introduce the GAN network.

[0110] Based on the enhanced collaborative compressed sensing anchor multi-target visible light positioning cooperation, a GAN-based sparse learning framework is further constructed to recover the vector θ. First, the aggregated received signal vector y is directly used as the initial input of the generator network to replace the random variable z in the traditional GAN, and two generators are used to gradually establish an explicit mapping relationship from y to θ. The generator G1(·) generates the corresponding latent channel matrix H according to the input y. Then the latent channel matrix H is used as the input to the pre-trained generator G2(·), whose purpose is to generate the corresponding sparse positioning vector containing position information according to the input channel information. In addition, the present invention adds a regularization term on the basis of the traditional loss function to reduce the error between the generated vector and the true vector, and designs a composite loss function Lclf, which is defined as:

[0111]

[0112] in:

[0113] like Figure 4 The figure shows the processing diagram after the introduction of CS and GAN. The GAN-MVLP scheme proposed in this invention mainly consists of two stages: training stage and inference stage. First, the discriminator network is trained to correctly distinguish the real positioning vector θ from the generated positioning vector θ to the greatest extent possible. Finally, the simultaneous positioning results of multiple targets are obtained through sparse recovery.

[0114] This paper proposes a multi-target fast visible light positioning solution based on a collaborative compressed sensing framework (CSM), aiming to solve the problem of high time complexity of multi-target simultaneous positioning in existing visible light positioning systems. This solution exploits the sparsity and cross-correlation in multi-target positioning, and uses compressed sensing technology to transform the multi-target positioning problem into a sparse recovery problem, significantly reducing the complexity of positioning calculations; at the same time, combined with the compressed sensing-generative adversarial network (CS-GAN) model, the efficiency and accuracy of multi-target simultaneous positioning are further optimized; by training the generative adversarial network, the model can quickly map from the aggregated received signal vector to the sparse positioning vector, ensuring efficient sparse recovery and positioning accuracy improvement in multi-target scenarios.

[0115] Aiming at the problem of high time complexity of multi-target simultaneous positioning in the existing visible light positioning (VLP) system, the present invention proposes a multi-target fast visible light position estimation scheme based on a collaborative compressed sensing framework. The core steps are as follows: Firstly, by fully exploiting the inherent sparsity of a small number of target positions relative to all grid points in the indoor environment, a visible light position estimation technology based on collaborative compressed sensing (CSM) is proposed; the technology aims to effectively solve the sparsity and cross-correlation problems of multi-targets in visible light positioning, and transform the multi-target positioning problem into a sparse recovery problem; secondly, a compressed sensing-generative adversarial network model (CS-GAN) is used as a research entry point to solve the high time complexity problem existing in multi-target simultaneous positioning; the core idea of ​​the model is to utilize the sparsity of the positioning targets and the cross-correlation between the received signals under the compressed sensing framework, and realize sparse recovery with low time complexity by solving underdetermined equations; finally, the CS-GAN is trained so that the generator network can directly learn the mapping from the aggregated received signal vector to the sparse positioning vector; by effectively integrating the collaborative information of multiple targets, the multi-target positioning accuracy and computational efficiency are improved, thereby reducing the time complexity required for multi-target simultaneous positioning.

[0116] The present invention assumes that the spatial position of the target terminal obeys a random distribution function, the spatial direction information of the target obeys a truncated Laplace distribution, and the typical visible light positioning boundary conditions are used as a basis (5m×5m×3m, and the emitting light source is 4×4 LEDs); combined with the proposed multi-target fast visible light position estimation technology based on the collaborative compressed sensing framework, CS-GAN is introduced to transform the multi-target positioning problem into a sparse recovery problem, and GAN is further used to learn the mapping from the aggregated received signal vector to the sparse positioning vector, so as to achieve the purpose of reducing the time complexity required for simultaneous positioning of multiple targets; thus, when the plane grid points are divided into 20cm×20cm and the number of targets is ≤5, the average position estimation error AE (Average Error) is ≤8cm, and the total calculation time of multiple targets is no more than 100ms.

[0117] The present invention aims to solve the problem of high time complexity of existing visible light positioning systems when simultaneously positioning multiple targets. By exploiting the sparsity between target positions and indoor environment grid points, the method of the present invention effectively transforms the multi-target positioning task into a sparse recovery problem. The present invention adopts a compressed sensing-generative adversarial network model (CS-GAN) to achieve efficient mapping of aggregated received signal vectors to sparse positioning vectors, significantly improving positioning accuracy and computational efficiency. Compared with the typical compressed sensing method Block-OMP algorithm multi-target positioning, which takes 490ms, the calculation time of the proposed algorithm is shortened by 59.18%.

[0118] In terms of improving positioning accuracy, the present invention extracts sparse signals and restores target positions through compressed sensing technology, effectively improving positioning accuracy, especially in noisy and interfering environments; in terms of enhancing system robustness, the generative adversarial network (GAN) optimizes signal processing through adversarial training, significantly improves anti-interference ability, and ensures stable positioning performance in complex environments; in terms of optimizing computational efficiency, multi-target positioning is converted into a sparse recovery problem, reducing computational complexity and improving the real-time response capability of the system; in terms of adapting to complex environments, based on visible light positioning (VLP), radio frequency band signal attenuation and multipath interference are avoided, which is particularly suitable for complex indoor environments and provides high-precision, low-interference positioning solutions; the present invention combines compressed sensing with generative adversarial networks, optimizes signal processing through sparse recovery and adversarial training, and significantly improves the accuracy, efficiency and robustness of multi-target positioning; it not only provides new multi-target positioning solutions for smart manufacturing, industrial Internet of Things and other fields, but also promotes the development of smart factories and automated production systems.

[0119] Creative Aspects:

[0120] The creativity of the present invention is mainly reflected in the redefinition and optimization of traditional positioning technology; by innovatively combining compressed sensing (CS) and generative adversarial network (GAN) technology, the present invention has pioneered a new design idea for a multi-target indoor visible light positioning (VLP) system; specifically, compressed sensing technology exploits signal sparsity to transform complex multi-target positioning problems into sparse signal recovery tasks, thereby achieving efficient processing from signal acquisition to positioning calculation; this technical advantage enables the system to guarantee high-precision and high-efficiency positioning performance even in scenarios where multiple targets are densely distributed or signal aliasing is severe.

[0121] On the other hand, the introduction of generative adversarial networks (GANs) further optimizes the sparse recovery process of compressed sensing. Through the adversarial training mechanism, GAN enables the generator to directly extract sparse location information from the received signal while suppressing background noise and interference signals. This adversarial learning framework not only improves the robustness of the positioning system, but also significantly enhances its adaptability in complex dynamic environments. In particular, it provides an innovative solution that can operate stably under conditions such as high noise, non-line-of-sight propagation (NLOS), and signal multipath effects that are easily limited by traditional VLP systems.

[0122] Novelty:

[0123] The novelty of the present invention lies in the dual aspects of technological innovation and application expansion. It is the first time that a generative adversarial network (GAN) is combined with compressed sensing (CS) and applied to an indoor multi-target visible light positioning (VLP) system. The present invention technically breaks the limitations of the traditional VLP system and proposes an efficient and stable solution for multi-target positioning scenarios in complex environments.

[0124] Traditional VLP systems show obvious deficiencies when faced with signal interference, environmental noise and densely distributed targets. For example, multipath interference will lead to a significant decrease in positioning accuracy, and the signal recovery process has a high computational complexity. However, the present invention uses compressed sensing technology and the sparse characteristics of LED signals to abstract the multi-target positioning problem into a sparse signal recovery task, significantly optimizing computational efficiency. On this basis, the introduction of GAN further improves the system's anti-interference ability and signal recovery quality. Through adversarial training of the generator and the discriminator, GAN can not only accurately restore the target position, but also dynamically adjust the system to adapt to complex environmental changes, such as non-line-of-sight propagation conditions or signal aliasing scenarios.

[0125] Practicality:

[0126] The practicality of the present invention is reflected in its broad application prospects in the field of multi-target positioning, especially its practical value in industrial manufacturing and Internet of Things-related scenarios; the multi-target visible light positioning method based on compressive sensing-GAN can break through the accuracy bottleneck of traditional positioning systems in complex environments and provide efficient and reliable positioning support for intelligent manufacturing, industrial Internet of Things, and automated production.

[0127] The compressive sensing technology greatly improves the computational efficiency of the system by reducing the signal acquisition amount and computational amount, and at the same time significantly reduces the response delay caused by high computational complexity in traditional methods; this feature enables the present invention to quickly complete the positioning task in a multi-target dense distribution or dynamic environment and meet the real-time requirements; in addition, the introduction of GAN further optimizes the system performance, and its powerful anti-interference ability enables the system to maintain stable high-precision positioning effects even in high-noise environments or complex interference conditions.

[0128] In terms of application scenarios, the present invention is particularly suitable for occasions with high requirements for high-precision positioning in industrial environments, such as intelligent robots, automated guided vehicles (AGVs), dynamic asset tracking, and automatic scheduling of workshop equipment, etc.; by using visible light as the positioning signal, the present invention effectively avoids the problems of limited wireless spectrum resources, signal congestion, and interference, and at the same time has the characteristics of energy conservation and environmental protection and can be seamlessly connected with intelligent lighting systems.

[0129] The above-described embodiments merely represent several implementation manners of the present invention, and the description thereof is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention patent shall be subject to the appended claims.

Claims

1. A method for indoor multi-target visible light positioning based on compressed sensing, characterized in that: The following steps are involved: Build an indoor multi-target visible light positioning VLP system, where multiple targets receive visible light signals simultaneously to form a visible light signal data set; Convert multiple visible light signals into multiple sparse vectors, where the non-zero element position of the sparse vector corresponds to the position of the target; aggregate the autocorrelation signal vectors in the multiple visible light signals to obtain the total power of the multiple visible light signals; when the compressed sensing algorithm CS is used to measure the visible light transmission in the room, an observation matrix is ​​formed, and a measurement power is assigned to each element in the observation matrix based on the total power; use the measurement power assigned by the observation matrix to restore multiple sparse vectors to form multiple preliminary positioning sparse vectors; According to the cooperative relationship between different emitting light sources in the system, the cross-correlation measurement vectors in multiple visible light signals are aggregated, and the preliminary positioning sparse vector is reconstructed according to the cross-correlation measurement vector to obtain multiple positioning sparse vectors; The positions of the multiple targets are determined according to the positions of the non-zero elements in the multiple positioning sparse vectors.

2. The method for indoor multi-target visible light positioning based on compressed sensing according to claim 1, characterized in that: The obtaining of the total power of the plurality of visible light signals comprises: Aggregate the received signal vectors of multiple visible light signals to form an autocorrelation signal vector y, which is expressed as: Where: ω represents the total background noise vector; θ represents the target positioning vector on the grid, θ=[θ1,...,θ j ,...,θ N ] T is an indicator vector with N elements; The observation matrix φ in the compressed sensing algorithm CS is expressed as: By aggregating the autocorrelation signal vector y of multiple visible light signals, the total power P of multiple visible light signals is obtained. rx , expressed as: Where: E{·} represents the statistical expectation operator; ⊙ represents the Hadamard product operator; (·) * represents the complex conjugate operator; represents the variance background noise; 1 M represents an M-length vector of full values.

3. The method for indoor multi-target visible light positioning based on compressed sensing according to claim 2, characterized in that: The recovering of the plurality of sparse vectors to form a plurality of preliminary positioning sparse vectors includes: When the compressed sensing algorithm CS measures the indoor visible light channel and propagation environment, it forms an observation matrix J, which is expressed as: Based on the total power, each element in the observation matrix J is assigned a measurement power, and the measurement power assigned by the observation matrix is ​​used to restore multiple sparse vectors to form multiple preliminary positioning sparse vectors.

4. The method for indoor multi-target visible light positioning based on compressed sensing according to claim 3, characterized in that: The reconstructing the preliminary positioning sparse vector includes: According to the cooperative relationship between different emitting light sources in the indoor multi-target visible light positioning VLP system, the cross-correlation measurement vector P in multiple visible light signals is aggregated. corr , expressed as: in:(·) H represents the conjugate transpose operator; represents the Kronecker product operator; represents the Khatri-Rao product; I M Represents the identity matrix of size M×M; the cross-correlation measurement vector P corr is length-M 2 ; Define an M(M+1) / 2×M 2 The selection matrix S is of size M×M; each row of the selection matrix S corresponds to P corr There are M independent diagonal elements and M(M-1) / 2 upper diagonal elements. After the selection process is completed, a cross-correlation measurement model based on the compressed sensing algorithm CS is formulated, which is expressed as: Among them, the size of matrix ψ is m(m+1) / 2×N, which represents the observation matrix; the matrix ψ is expressed as: The number of independent linear equations used in the cross-correlation measurement model based on the compressed sensing algorithm CS increases from M to M(M+1) / 2, and the measurement data that can be used for sparse recovery based on the compressed sensing algorithm CS gradually increases; A preliminary positioning sparse vector is reconstructed according to the cross-correlation measurement model to obtain multiple positioning sparse vectors.

5. The method for indoor multi-target visible light positioning based on compressed sensing according to claim 4, characterized in that: Also includes: A generative adversarial network (GAN) is used to identify a received signal vector in a plurality of aggregated visible light signals, generate a positioning vector, and use the positioning vector to correct a plurality of positioning sparse vectors to obtain a plurality of final positioning sparse vectors; The following steps are involved: The received signal vector y of the aggregated multiple visible light signals is used as the initial input of the generator of the generative adversarial network GAN to replace the random variable z in the generative adversarial network GAN, and two generators are used to gradually establish an explicit mapping relationship from the received signal vector y to the target positioning vector θ; Generator G1(·) generates the corresponding potential channel matrix H according to the input y, and inputs the potential channel matrix H into the pre-trained generator G2(·) to generate a sparse positioning vector containing position information A regularization term is added to the loss function of the generative adversarial network GAN to reduce the error between the generated vector and the true vector, forming a composite loss function Lclf, which is expressed as: in: Generate a sparse positioning vector based on The multiple positioning sparse vectors are corrected to obtain multiple final positioning sparse vectors.

6. An indoor multi-target visible light positioning device based on compressed sensing, characterized in that: include: The acquisition module is used to build an indoor multi-target visible light positioning VLP system. Multiple targets receive visible light signals at the same time to form a visible light signal data set; The recognition module is used to convert multiple visible light signals into multiple sparse vectors, where the non-zero element position of the sparse vector corresponds to the position of the target; aggregate the autocorrelation signal vectors in the multiple visible light signals to obtain the total power of the multiple visible light signals; when the compressed sensing algorithm CS is used to measure the visible light transmission in the room, an observation matrix is ​​formed, and a measurement power is assigned to each element in the observation matrix based on the total power; the measurement power assigned by the observation matrix is ​​used to restore the multiple sparse vectors to form multiple preliminary positioning sparse vectors; The positioning module is used to aggregate the cross-correlation measurement vectors in multiple visible light signals according to the cooperative relationship between different emitting light sources in the system, and reconstruct the preliminary positioning sparse vector according to the cross-correlation measurement vector to obtain multiple positioning sparse vectors.

7. An electronic device, characterized in that: include: Memory and processor; The memory is used to store computer programs; The processor is used to implement the steps of the indoor multi-target visible light positioning method based on compressed sensing as described in any one of claims 1 to 5 when executing the computer program stored in the memory.

8. A computer-readable storage medium, characterized in that: Used to store a computer program, which, when executed by a processor, implements the steps of a method for indoor multi-target visible light positioning based on compressed sensing as described in any one of claims 1 to 5.

Images