Dynamic indoor visible light positioning method based on spatio-temporal characteristic information

Abstract

The invention discloses a dynamic indoor visible light positioning method based on spatio-temporal characteristic information, and relates to the technical field of visible light communication.The method comprises the steps that firstly, a target motion model based on the spatio-temporal characteristic information is constructed to analyze spatio-temporal correlation in a received signal strength RSS sequence; the method comprises the following steps: extracting spatial texture feature information in a historical time period through a forward cyclic convolution module, predicting a change trend of the spatial texture feature information in a future time period in combination with a target motion model, and compensating the spatial texture feature information in the historical time period by using future observation information through a backward cyclic convolution module. Time patterns, space features and potential space-time correlation features are fully mined from space texture feature information of a received signal strength RSS sequence in a historical time period, and high-precision positioning of a moving target is achieved.

Description

Technical Field

[0001] The present invention relates to the field of visible light communication technology, and in particular to a dynamic indoor visible light positioning method, device, equipment and medium based on spatiotemporal feature information. Background Art

[0002] With the rapid development of wireless communication technology, the application of precise indoor positioning technology in multiple fields has gradually become a reality, such as smart warehouses, robot navigation, autonomous driving, smart parking and other emerging industries; Visible light positioning (VLP) as a new indoor positioning technology, uses light-emitting diode (LED) signals to locate and track the target, with the advantages of strong stability, low carbon and environmental protection, flexible coverage, etc. Compared with traditional positioning technologies such as Wi-Fi, Bluetooth and infrared, it provides higher accuracy and resolution.

[0003] However, the current dynamic positioning and posture tracking technology of visible light can achieve high positioning accuracy in static environments, but in dynamic environments, due to the influence of environmental factors (such as signal strength fluctuations, occlusion effects, and changes in illumination, etc.), the positioning accuracy and robustness of the system will be significantly reduced. The traditional method based on signal propagation model (SPM) relies on a preset and stable signal propagation model for positioning inference, but in complex dynamic environments, the signal propagation characteristics will change, resulting in increased positioning errors, especially under complex environmental conditions such as illumination, reflection, and multipath propagation. The accuracy of the SPM model is difficult to guarantee. The traditional fingerprint-based positioning method estimates the position by establishing a fingerprint database and matching the received signal with the fingerprint library without relying on an accurate SPM model. However, the FG method has high requirements for the static nature of the environment, such as the stability of the transmitting light source and the receiver parameters. In reality, due to factors such as human activities and changes in indoor temperature and humidity, the signal propagation environment will change dynamically, resulting in a decrease in the effectiveness of the fingerprint library, limiting its application in dynamic scenes. The traditional method based on photography positioning uses a camera to perceive the LED array projection, but this method has been affected by photography errors and the smear effect in dynamic scenes.

[0004] In response to such problems, some positioning methods based on machine learning have begun to attract attention in recent years. Although traditional deep learning methods, such as convolutional neural networks (CNNs), have achieved certain results in some static or near-static environments, when faced with rapid changes in dynamic environments, the signal propagation path and environmental changes caused by target movement will cause large fluctuations in signal strength, making it difficult to identify the time features in continuous moment observation data, resulting in poor positioning accuracy and unstable real-time performance. Therefore, in a dynamic environment, the current methods are difficult to fully explore the potential spatiotemporal correlation features in time series data, making it difficult for the positioning system to achieve high-precision positioning of moving targets. Summary of the invention

[0005] The embodiment of the present invention provides a dynamic indoor visible light positioning method based on spatiotemporal feature information, which can solve the problem in the prior art that the current methods are difficult to fully exploit the potential spatiotemporal correlation features in time series data, making it difficult for the positioning system to achieve high-precision positioning of moving targets.

[0006] The embodiment of the present invention provides a dynamic indoor visible light positioning method based on spatiotemporal feature information, comprising the following steps:

[0007] When the moving target moves indoors, the received signal strength RSS sequence of the visible light signal received by the moving target is extracted;

[0008] The moving target moves regularly or randomly indoors, and the position coordinates and posture parameters of the moving target at time t are predicted to build a target motion model;

[0009] A bidirectional recurrent convolutional neural network Bi-RCNN including a forward recurrent convolution module and a backward recurrent convolution module is constructed; the forward recurrent convolution module is used to extract the spatial texture features in the RSS sequence of the received signal strength in the historical period, and the spatial texture features in the RSS sequence of the received signal strength in the future period are predicted by combining the target motion model; the backward recurrent convolution module is used to extract the spatial texture features in the RSS sequence of the received signal strength in the predicted future period, so as to correct the spatial texture features in the RSS sequence of the received signal strength in the historical period;

[0010] According to the spatial texture features in the RSS sequence of the received signal strength in the corrected historical period, and combined with the target motion model, the pose parameters of the moving target at the current moment are predicted to determine the position of the moving target at the current moment.

[0011] Preferably, the building of the target motion model includes:

[0012] Assume that the moving target performs regular motion or random motion indoors, and the regular motion or random motion is characterized as uniform motion or random motion respectively;

[0013] Assume the uniform motion speed is Assumptions in represents the absolute value of speed; let Represents the random motion vector at time t, which obeys the Gaussian random process and is expressed as in express The covariance matrix of the moving target is assumed to be Random changes; then the predicted position coordinates of the moving target at time t+1 It is expressed as:

[0014]

[0015] Where: w u and w s ∈[0,1] represent the weights of the uniform velocity component and the random component respectively, and w u +w s =1;

[0016] In the positioning of moving targets, the samples of the received signal strength RSS sequence depend on the posture parameters of the maneuvering target at time t, and the posture parameters are expressed as Use pose parameters Predict the time-varying position information of the target, and locate and track the moving target;

[0017] In the actual moving target positioning process, the received signal strength RSS of the moving target is continuously obtained to obtain the sample sequence of the received signal strength RSS The corresponding positioning tag is

[0018] Preferably, the modified spatial texture features in the received signal strength RSS sequence of the historical period include:

[0019] The forward recurrent convolution module and the backward recurrent convolution module both include a forget gate branch; the forward recurrent convolution module By integrating the hidden state information of the past moments And the current observation information generates a weight vector Backward recurrent convolution module Integrate future hidden state information and current observation information Forming weight

[0020] In the forward recurrent convolution module, the forget gate is used to store historical memory Empowerment and inheritance, and current moment observation information Fusion to get new features Existing in memory cells, expressed as:

[0021]

[0022] Among them: ⊙ represents the Hadamard product, the forget gate output weight and memory gate output

[0023] The memory cell state of the backward recurrent convolution module is expressed as:

[0024]

[0025] The forward recurrent convolution module uses the memory gate and the forget gate to extract the RSS observation sample of the received signal strength at the current moment. and historical hidden information and The backward recurrent convolution module uses the memory gate and the forget gate to extract the current received signal strength RSS observation sample. and future hidden information Stable spatial texture features.

[0026] Preferably, the predicting of the position and posture parameters of the moving target at the current moment includes:

[0027] The hidden state obtained by fusing the forward recurrent convolution module and the backward recurrent convolution module using the perception layer of the bidirectional recurrent convolutional neural network Bi-RCNN To determine the pose parameters of the moving target

[0028] make:

[0029]

[0030] For the perception layer input that fuses the hidden states of the forward recurrent convolution module and the backward recurrent convolution module, let:

[0031]

[0032] and Represent the weight and bias of the perception layer respectively, then the pose parameters of the moving target Estimated to be:

[0033]

[0034] Where: ReLU(·) represents the ReLU activation function.

[0035] Preferably, after the received signal strength RSS sequence is acquired, the received signal strength RSS sequence is intercepted and normalized using a sliding window method.

[0036] The embodiment of the present invention further provides a dynamic indoor visible light positioning device based on spatiotemporal feature information, comprising:

[0037] The acquisition module extracts the received signal strength RSS sequence of the visible light signal received by the moving target when the moving target moves indoors;

[0038] Motion module: The moving target performs regular or random motion indoors, and predicts the position coordinates and posture parameters of the moving target at time t to build a target motion model;

[0039] The recognition module is used to construct a bidirectional recurrent convolution neural network Bi-RCNN including a forward recurrent convolution module and a backward recurrent convolution module; the forward recurrent convolution module is used to extract the spatial texture features in the RSS sequence of the received signal strength in the historical period, and the spatial texture features in the RSS sequence of the received signal strength in the future period are predicted in combination with the target motion model; the backward recurrent convolution module is used to extract the spatial texture features in the RSS sequence of the received signal strength in the predicted future period, so as to correct the spatial texture features in the RSS sequence of the received signal strength in the historical period;

[0040] The positioning module is used to determine the position of the moving target at the current moment by predicting the position parameters of the moving target at the current moment based on the spatial texture features in the RSS sequence of the received signal strength in the corrected historical period and combining the target motion model.

[0041] An embodiment of the present invention further provides an electronic device, including a memory and a processor;

[0042] The memory is used to store computer programs;

[0043] The processor is used to implement the steps of the dynamic indoor visible light positioning method based on spatiotemporal feature information when executing the computer program stored in the memory.

[0044] 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 dynamic indoor visible light positioning method based on spatiotemporal feature information as described above.

[0045] The embodiment of the present invention provides a dynamic indoor visible light positioning method based on spatiotemporal feature information. Compared with the prior art, the method has the following beneficial effects:

[0046] The invention constructs a target motion model by making a moving target perform regular or random motion indoors; uses a forward cyclic convolution module to extract spatial texture features in a RSS sequence of received signal strength in a historical period, and predicts spatial texture features in a RSS sequence of received signal strength in a future period in combination with the target motion model; uses a backward cyclic convolution module to extract spatial texture features in a predicted RSS sequence of received signal strength in a future period, so as to correct the spatial texture features in the RSS sequence of received signal strength in a historical period; determines the position of the moving target according to the spatial texture features in the corrected RSS sequence of received signal strength in a historical period; the process first constructs a target motion model based on spatiotemporal feature information to analyze the spatiotemporal correlation in the RSS sequence of received signal strength, extracts spatial texture feature information in a historical period by using a forward cyclic convolution module, and predicts the change trend of spatial texture feature information in a future period in combination with the target motion model, and uses a backward cyclic convolution module to compensate for spatial texture feature information in a historical period by using future observation information, so as to fully mine the time pattern and spatial features, as well as potential spatiotemporal correlation features, from the spatial texture feature information of the RSS sequence of received signal strength in a historical period, so as to realize high-precision positioning of the moving target.

[0047] In addition, the present invention enhances the response capability to signal strength fluctuations in dynamic environments by setting memory cells and forget gate mechanisms in the forward recurrent convolution module and the backward recurrent convolution module; at the same time, it further analyzes the temporal pattern and spatial characteristics in the received signal strength RSS sequence in combination with a three-dimensional convolutional network, effectively improving the positioning and posture tracking performance of dynamic targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A schematic diagram of an overall solution of a dynamic indoor visible light positioning method based on spatiotemporal feature information provided by an embodiment of the present invention;

[0049] Figure 2 A schematic diagram of a dynamic target visible light positioning system of a dynamic indoor visible light positioning method based on spatiotemporal feature information provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0050] 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.

[0051] See also Figure 1The embodiment of the present invention provides a dynamic indoor visible light positioning method based on spatiotemporal feature information; the existing visible light positioning (VLP) technology can be mainly divided into a signal propagation model (SPM)-based method, a fingerprint-based positioning method, a machine learning-based neural network method, and a photography-based positioning method.

[0052] Method based on signal propagation model (SPM); this method uses photodiode (PD) to receive visible light signals and infers the target position and attitude parameters based on the signal propagation model; although the positioning accuracy is high in theory, it is highly dependent on environmental conditions in practical applications; when the indoor environment changes (such as small-scale fading, scattering and reflection, etc.), the SPM model is difficult to accurately reflect the actual signal propagation characteristics, resulting in a significant decrease in positioning accuracy and limited adaptability in dynamic scenes.

[0053] Fingerprint-based positioning method; this method estimates the position by establishing a fingerprint database and matching the received signal with the fingerprint library, without relying on an accurate SPM model; however, the FG method has high requirements on the static nature of the environment, such as the stability of the transmitting light source and receiver parameters; in reality, due to factors such as human activities and changes in indoor temperature and humidity, the signal propagation environment will change dynamically, resulting in a decrease in the effectiveness of the fingerprint library, limiting its application in dynamic scenarios.

[0054] Positioning methods based on neural networks; In recent years, convolutional neural networks (CNNs) and deep neural networks (DNNs) based on deep learning have been applied to the VLP field; These methods improve positioning accuracy through feature learning, but most methods ignore the time correlation of continuous time series data, resulting in insufficient positioning performance under high-speed moving targets and complex dynamic environments, especially in terms of real-time response capability and accuracy. There are significant bottlenecks.

[0055] Methods based on photography positioning; the camera-based visible light positioning (Image-based VLP) method uses a camera to sense the LED array projection, but this method is susceptible to photographic errors (such as focal length error, camera shake, LED camera vignetting, etc.) and the smear effect in dynamic scenes, which greatly limits its application in fast-moving targets and dynamic environments.

[0056] The present invention aims to solve the problem that the positioning accuracy of the existing visible light positioning (VLP) system is significantly reduced in a dynamic environment, especially the signal strength fluctuation caused by the signal propagation path and environmental changes caused by the movement of the target; the traditional positioning method based on the signal propagation model (SPM) and the fingerprint method (FG) has insufficient adaptability in a dynamic environment and is difficult to maintain high-precision positioning; in addition, the existing machine learning model has limited information mining and generalization capabilities in dynamic scenes, and it is difficult to effectively utilize the spatiotemporal correlation of target motion; the present invention proposes a bidirectional recurrent convolutional network model based on spatiotemporal correlation, which aims to achieve high-precision estimation of target position and posture in a dynamic environment by mining the temporal pattern and spatial features in the received signal strength (RSS) sequence. The technology of the present invention mainly includes:

[0057] Step 1: Received signal strength preprocessing.

[0058] The sliding window technology is used to intercept and normalize the RSS sequence to reduce the impact of noise and improve data quality, providing high-quality input data for subsequent model training and pose estimation.

[0059] Step 2: Spatiotemporal feature extraction.

[0060] A model based on spatiotemporal feature information is constructed to analyze the spatiotemporal correlation in the RSS sequence; the spatial texture information in the historical time period is extracted through the forward recurrent memory module, and the change trend in the future time period is predicted; the backward recurrent memory module uses future observation information to compensate for the historical texture information. The spatial texture is extracted by combining the convolutional network to enhance the model's ability to respond to signal intensity fluctuations.

[0061] Step 3: Dynamic target pose prediction.

[0062] By using the extracted spatiotemporal feature information, the model is used to accurately predict the target's position and trajectory, ensuring stable and reliable positioning and posture tracking in a dynamic environment.

[0063] The present invention breaks through the reliance of traditional methods on static models by effectively utilizing the spatiotemporal features in the RSS sequence, improves the adaptability of the visible light positioning system in complex dynamic environments, and has significant practical value and broad application prospects.

[0064] Specifically:

[0065] 1. Visible light dynamic positioning model.

[0066] 1. Target motion model.

[0067] like Figure 2As shown in the figure, the maneuvering target is assumed to move randomly indoors, and the general situation is considered; the movement of the maneuvering target contains two components: (predictable) regular movement and (unpredictable) random movement, which are respectively described by uniform motion and random motion; suppose the uniform motion speed is assumed in represents the absolute value of speed; let represents the random motion vector at time t, which obeys the Gaussian random process, that is, in express The covariance matrix of the target pose is Random changes; Based on the above assumptions, the position coordinates of the maneuvering target at time t+1 Can be characterized as:

[0068]

[0069] Where: w u and w s ∈[0,1] represent the weights of the uniform velocity component and the random component respectively, and w u +w s =1; Here, the Gaussian process is considered as the maximum entropy model without high-order statistical information, and the model has the strongest generalization ability and the lowest modeling mismatch risk.

[0070] 2. Continuous RSS sequence observation model.

[0071] In the static observation model of the RSS sequence, the observation samples depend on the relative position and angle relationship between the LED and the PD; while in dynamic positioning, the RSS samples ultimately depend on the posture parameters of the maneuvering target at time t, which is recorded as Just RSS Sample Depends on maneuvering target positioning parameters It can be used to estimate the time-varying position information of the target, so as to locate and track it. The actual positioning process requires continuous sampling of the RSS observation data of the moving target to obtain the RSS sample sequence. The corresponding positioning tag is Since the propagation parameters in a dynamic environment change dynamically over time, the observation samples and its tags It also changes randomly.

[0072] 2. Specific design.

[0073] like Figure 1As shown in the figure, it is divided into two stages: offline training and online positioning. In the offline training stage, the network parameters are optimized to obtain the mapping relationship between the observation samples and the pose parameters of the maneuvering target. Subsequently, in the online positioning and tracking stage, the real-time visible light observation sample sequence is input into the trained network to output the pose parameter estimation of the moving target. Since the received signal strength (RSS) observation samples are affected by the random changes of environmental parameters and the target motion posture, their fluctuations show random characteristics. Therefore, the static parameters are transformed from non-time-varying to time-varying parameters. The proposed network will extract the stable temporal and spatial texture structure in the observation data, thereby achieving robust positioning and tracking results. The specific steps are as follows:

[0074] 1. Data preprocessing.

[0075] In the input data preprocessing stage, the received signal strength sequence and position information sequence are first intercepted through a long sliding window to form the kth input sequence of the model; next, these sequences are normalized to eliminate the scale differences between the data and ensure the effectiveness and stability of the model training; specifically, the received signal strength sequence and position information sequence are normalized by calculating their mean and standard deviation, respectively, to generate the normalized center and scale parameters.

[0076] In this process, the normalized center is used to adjust the central position of the data, while the normalized scale is used to control the distribution range of the data; finally, these normalized center and scale parameters are integrated into the joint parameters of the input data to provide the necessary input for the subsequent forward recurrent convolutional memory and backward recurrent convolutional memory modules, ensuring that the model can effectively extract spatiotemporal features when processing dynamic target positioning.

[0077] 2. Forget gate branch.

[0078] The forget gate branch is designed for the forward and backward recurrent memory modules; the forward module By integrating past hidden state information and current observation information to generate the weight vector The backward module Then the future hidden state information is integrated and current observation information Forming weight These weight coefficients are calculated through the fully connected layer and processed by the sigmoid activation function so that historical information can be effectively retained or discarded when updating memory cell information, thereby enhancing the model's ability to extract temporal and spatial features.

[0079] 3. Memory gate branch.

[0080] The memory gate uses a 3D convolutional neural network to enhance the texture feature extraction of the current sample. The memory gate of the forward loop module consists of multiple 3D convolutional layers and fully connected layers. After the input data passes through the 3D convolutional network to extract spatial texture features, it is input into the fully connected layer to extract spatiotemporal features. The forward memory gate input at the current moment It is composed of a stack of past hidden states and observation information; the 3D convolution output is rearranged into a vector form and combined with weight and bias calculations to obtain the memory gate output; the structure of the backward loop module is similar to the forward module, and the output is calculated through the corresponding input and convolution kernel parameters; this structure effectively improves the model's ability to integrate spatial and temporal features.

[0081] 4. Output gate branch.

[0082] The output gate branch is similar to the forget gate. After the hidden state information is spliced ​​with the RSS observation information, it passes through a single fully connected layer to obtain the output of the output gate. The output gate state is divided into the forward loop module output gate and the backward loop module output gate, which are defined as:

[0083]

[0084] 5. Update memory cells and hidden state cells.

[0085] The forward recurrent convolution module uses memory cells to store the spatiotemporal texture features of the observed samples extracted by the module, and cyclically inputs them into the processing process at the next moment; then, the positioning and tracking performance is enhanced by fusing the features of the observed data at different moments.

[0086] First, in the forward loop module, the forget gate is used to store historical memory Empowerment and inheritance, and then with the current observation information Fusion to get new features Existing in memory cells, namely:

[0087]

[0088] Among them: ⊙ represents the Hadamard product, the forget gate output weight and memory gate output

[0089] Similarly, the memory cell state of the backward loop module is expressed as:

[0090]

[0091] The forward recurrent convolution module uses the memory gate and the forget gate to extract the current RSS observation sample. and historical hidden information and The backward loop module uses the memory gate and the forget gate to extract the actual RSS observation samples. and future hidden information Stable spatial texture features.

[0092] 6. Output pose trajectory estimation.

[0093] The hidden state obtained by fusing the forward and backward recurrent convolution modules using the perception layer To determine the pose parameters of the maneuvering target make:

[0094]

[0095] To fuse the forward and backward hidden states of the perception layer input, let:

[0096] and

[0097] and Represent the weight and bias of the perception layer respectively, then the moving target pose estimation is:

[0098]

[0099] Where: ReLU(·) represents the ReLU activation function. For simplicity, let

[0100] The present invention proposes a dynamic visible light positioning scheme based on spatiotemporal feature information to address the positioning accuracy problem caused by signal propagation path and environmental changes caused by target movement in existing visible light positioning systems; the scheme of the present invention intercepts and normalizes the received signal strength (RSS) sequence through a sliding window technology, laying a foundation for subsequent model training and pose estimation; in the model, by analyzing the spatiotemporal correlation in the RSS sequence, the forward recurrent memory module extracts the spatial texture information of historical moments to predict and correct the spatial texture information of future moments; and the backward recurrent memory module extracts the spatial texture information of future moments to compensate for the spatial texture information of historical moments; this design enhances the discriminative characteristics of moving targets; the bidirectional recurrent neural network significantly improves the performance of positioning and tracking by fusing the spatiotemporal texture features of previous and next moments, and enhances the response capability to dynamic signal fluctuations.

[0101] Aiming at the problem that in the existing visible light positioning (VLP) system, the signal propagation path and environment change due to the movement of the target, so that the static parameter model is difficult to maintain high-precision positioning in a dynamic environment, the present invention proposes a bidirectional recurrent convolutional network model based on spatiotemporal correlation, and the core steps of the model are as follows: first, the received signal strength (RSS) sequence is intercepted and normalized by using a sliding window technology to lay a foundation for subsequent model training and pose estimation; second, the spatiotemporal correlation in the RSS sequence is analyzed, the temporal pattern and spatial features in the data are identified, and the spatiotemporal texture in the RSS observation sequence is extracted by using a bidirectional convolution module, and the spatiotemporal texture feature information in the RSS sequence is extracted by introducing a memory cell structure, and the cached spatiotemporal information is reasonably inherited by using a forget gate mechanism to ensure the long-term memory of these features, thereby enhancing the model's response ability to dynamic signal fluctuations; finally, based on the extracted spatiotemporal correlation, the pose trajectory of the target is predicted to achieve stable and accurate estimation of the dynamic target.

[0102] The present invention assumes that the target terminal moves at a uniform speed or randomly in the room (obeying Gaussian random process), the spatial position obeys the random distribution function, and the target spatial direction information obeys the truncated Laplace distribution, and is based on the typical visible light positioning boundary conditions (5m×5m×3m, the emitting light source is 4×4 LEDs). 6 For spatial data points, when the moving speed is ≤15km / h, the CDF-90% accuracy is ≤10cm; the average yaw angle estimation error α error is ≤3°, and the time required for a positioning calculation is no more than 100ms.

[0103] The present invention proposes a method combining a bidirectional loop structure with a three-dimensional convolutional network to respectively extract the time features and spatial texture features in the received signal strength (RSS) data; by analyzing the spatiotemporal correlation in the RSS sequence, the forward loop memory module extracts the spatial texture information of the historical moments to predict and correct the spatial texture information of the future moments; and the backward loop memory module extracts the spatial texture information of the future moments to compensate for the spatial texture information of the historical moments; in a complex environment, the present invention enhances the response capability to dynamic signal fluctuations and realizes stable and accurate posture trajectory prediction of dynamic targets; compared with the traditional dynamic VLP positioning method based on the signal propagation model (SPM), under the same environmental conditions and data points, the CDF-90% accuracy is improved by 50%.

[0104] Creative Aspects:

[0105] The present invention innovatively introduces a bidirectional recurrent convolutional neural network based on spatiotemporal correlation, combined with a forward recurrent memory module and a backward recurrent memory module, to respectively extract spatial texture information of historical and future moments, and enhances the response capability to signal strength fluctuations in a dynamic environment through a memory cell and a forget gate mechanism; at the same time, a three-dimensional convolutional network is combined to further analyze the temporal pattern and spatial features in the received signal strength (RSS) sequence, effectively improving the positioning and posture tracking performance of dynamic targets; the method of the present invention is significantly superior to traditional positioning methods based on static parameter models or signal propagation models (SPM), solves the problems of insufficient positioning accuracy and robustness in dynamic environments, and realizes accurate prediction of posture trajectories in dynamic environments.

[0106] Novelty:

[0107] The present invention proposes a new dynamic visible light positioning method, which incorporates the spatiotemporal correlation of RSS sequences into model design, breaking through the bottleneck of insufficient adaptability of traditional static fingerprint methods and ranging models to dynamic environments; through the combination of a bidirectional cyclic structure and a three-dimensional convolutional network, the method of the present invention can deeply mine the spatiotemporal texture information in the RSS sequence, and store and update features in a dynamic environment through a memory unit, effectively overcoming the problem of insufficient generalization ability of existing machine learning models in dynamic scenes; experimental results show that in typical dynamic scenes, the CDF-90% positioning accuracy of the method of the present invention is improved by 50% compared with traditional methods, and the average yaw angle estimation error is controlled within 3°, which has strong technological leadership and scientific innovation.

[0108] Practicality:

[0109] The present invention has broad application prospects, and is particularly suitable for scenarios requiring high-precision dynamic positioning and posture tracking, such as smart warehousing, industrial automation, unmanned navigation and other fields; the present invention adopts dynamic positioning technology based on spatiotemporal feature information, which can quickly and stably realize the positioning and posture estimation of the target in a complex dynamic environment; by improving the system's adaptability to dynamic signal fluctuations, it meets the actual needs of high dynamics and high precision; in addition, the present invention has the advantage of low computing cost, the positioning calculation time does not exceed 100ms, and it can stably achieve an average positioning error of ≤10cm under the condition of a moving speed of ≤15km / h, providing a practical solution for positioning needs in dynamic scenes.

[0110] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A dynamic indoor visible light positioning method based on spatiotemporal feature information, characterized in that: The following steps are involved: When the moving target moves indoors, the received signal strength RSS sequence of the visible light signal received by the moving target is extracted; The moving target moves regularly or randomly indoors, and the position coordinates and posture parameters of the moving target at time t are predicted to build a target motion model; Construct a bidirectional recurrent convolutional neural network Bi-RCNN including a forward recurrent convolution module and a backward recurrent convolution module; The forward recurrent convolution module is used to extract the spatial texture features in the RSS sequence of the received signal strength in the historical period, and combined with the target motion model, the spatial texture features in the RSS sequence of the received signal strength in the future period are predicted; The spatial texture features in the RSS sequence of the received signal strength in the predicted future period are extracted by using the backward circular convolution module to correct the spatial texture features in the RSS sequence of the received signal strength in the historical period; According to the spatial texture features in the RSS sequence of the received signal strength in the corrected historical period, and combined with the target motion model, the pose parameters of the moving target at the current moment are predicted to determine the position of the moving target at the current moment.

2. The method for dynamic indoor visible light positioning based on spatiotemporal feature information according to claim 1, characterized in that: The target motion model is constructed, comprising: Assume that the moving target performs regular motion or random motion indoors, and the regular motion or random motion is characterized as uniform motion or random motion respectively; Assume the uniform motion speed is Assumptions in represents the absolute value of speed; let Represents the random motion vector at time t, which obeys the Gaussian random process and is expressed as in express The covariance matrix of the moving target is assumed to be Random changes; then the predicted position coordinates of the moving target at time t+1 It is expressed as: Where: w u and w s ∈[0,1] represent the weights of uniform velocity component and random component respectively, and w u +w s =1; In the positioning of moving targets, the samples of the received signal strength RSS sequence depend on the posture parameters of the maneuvering target at time t, and the posture parameters are expressed as Use pose parameters Predict the time-varying position information of the target, and locate and track the moving target; In the actual moving target positioning process, the received signal strength RSS of the moving target is continuously obtained to obtain the sample sequence of the received signal strength RSS The corresponding positioning tag is 3. The method for dynamic indoor visible light positioning based on spatiotemporal feature information according to claim 2, characterized in that: The spatial texture features in the modified historical period received signal strength RSS sequence include: The forward recurrent convolution module and the backward recurrent convolution module both include a forget gate branch; the forward recurrent convolution module By integrating the hidden state information of the past moments And the current observation information generates a weight vector Backward recurrent convolution module Integrate future hidden state information and current observation information Forming weight In the forward recurrent convolution module, the forget gate is used to store historical memory Empowerment and inheritance, and current moment observation information Fusion to get new features Existing in memory cells, expressed as: Among them: ⊙ represents the Hadamard product, the forget gate output weight and memory gate output The memory cell state of the backward recurrent convolution module is expressed as: The forward recurrent convolution module uses the memory gate and the forget gate to extract the RSS observation sample of the received signal strength at the current moment. and historical hidden information and The backward recurrent convolution module uses the memory gate and the forget gate to extract the current received signal strength RSS observation sample. and future hidden information Stable spatial texture features.

4. The method for dynamic indoor visible light positioning based on spatiotemporal feature information according to claim 3, characterized in that: The predicting of the position and posture parameters of the moving target at the current moment includes: The hidden state obtained by fusing the forward recurrent convolution module and the backward recurrent convolution module using the perception layer of the bidirectional recurrent convolutional neural network Bi-RCNN To determine the pose parameters of the moving target make: For the perception layer input that fuses the hidden states of the forward recurrent convolution module and the backward recurrent convolution module, let: and Represent the weight and bias of the perception layer respectively, then the pose parameters of the moving target Estimated to be: Where: ReLU(·) represents the ReLU activation function.

5. The method for dynamic indoor visible light positioning based on spatiotemporal feature information according to claim 1, characterized in that: After the received signal strength RSS sequence is obtained, the received signal strength RSS sequence is intercepted and normalized using a sliding window method.

6. A dynamic indoor visible light positioning device based on spatiotemporal feature information, characterized in that: include: The acquisition module extracts the received signal strength RSS sequence of the visible light signal received by the moving target when the moving target moves indoors; Motion module: The moving target performs regular or random motion indoors, and the position coordinates and posture parameters of the moving target at time t are predicted to build a target motion model; A recognition module is used to construct a bidirectional recurrent convolutional neural network Bi-RCNN including a forward recurrent convolution module and a backward recurrent convolution module; The forward recurrent convolution module is used to extract the spatial texture features in the RSS sequence of the received signal strength in the historical period, and combined with the target motion model, the spatial texture features in the RSS sequence of the received signal strength in the future period are predicted; The spatial texture features in the RSS sequence of the received signal strength in the predicted future period are extracted by using the backward circular convolution module to correct the spatial texture features in the RSS sequence of the received signal strength in the historical period; The positioning module is used to determine the position of the moving target at the current moment based on the spatial texture features in the RSS sequence of the received signal strength in the corrected historical period and the target motion model to predict the posture parameters of the moving target at the current moment.

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 a dynamic indoor visible light positioning method based on spatiotemporal feature information 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 dynamic indoor visible light positioning method based on spatiotemporal feature information as described in any one of claims 1 to 5.

Images