A high-precision indoor visible light positioning method for complex dynamic crowd environments

Abstract

A method for indoor high-precision visible light positioning in a complex dynamic crowd environment, which relates to the field of indoor visible light communication technology. In order to solve the problems existing in the prior art, the method: start the positioning system and perform signal filtering and dynamic modulation; perform preliminary position calculation based on the quadrilateral measurement method combined with the phase and intensity information of the optical signal; optimize the initial position of the receiving end using the QPSO algorithm; the infrared sensor collects the density and dynamic changes of the crowd in the area in real time and generates a heat map, and predicts the influence of the crowd flow on the light field based on the fluid mechanics eddy model combined with the RNN network, generates a crowd flow eddy model, optimizes the optical signal of the receiving end using the CVPA algorithm, and dynamically adjusts the signal characteristics through the eddy effect; FPGA accurately tracks and updates the position of the APD receiving end through the EKPF filter; adaptively adjusts the weights of QPSO, CVPA and EKPF; after multiple iterations, the FPGA obtains the final accurate position coordinates of the receiving end. Real-time positioning of high-density crowds is suitable.

Description

Technical Field

[0001] The present invention relates to the field of indoor visible light communication technology, and in particular to an indoor high-precision visible light positioning method for a complex dynamic crowd environment. Background Art

[0002] With the popularization of smart buildings, smart businesses and modern public facilities, the demand for indoor positioning technology is increasing. Especially in large shopping malls, exhibition halls, museums, airports and other public places, high-precision indoor positioning can significantly improve navigation experience and service efficiency.

[0003] Positioning technology based on visible light communication (VLC) has gradually become a research topic with great potential in recent years. Indoor visible light positioning mainly relies on a number of key technologies. Including light source modulation and coding, optical receivers, interference noise filtering, positioning algorithms and many other technologies. The visible light positioning system uses the visible light signals emitted by ordinary LED lights for positioning, and has the advantages of no electromagnetic interference and low signal attenuation. In particular, LED lighting has been widely used in public buildings. Therefore, integrating positioning functions on the basis of existing lighting systems does not affect the lighting effect, and can efficiently provide accurate positioning services. However, despite the obvious technical advantages of visible light positioning, it still faces many challenges in practical applications, especially in complex dynamic crowd environments where real-time dynamic positioning requirements are more stringent.

[0004] The Chinese patent publication number is "CN 116087877 A", and the patent name is "A point classification indoor visible light positioning method based on artificial intelligence algorithm". This method proposes a three-dimensional indoor visible light positioning method based on artificial intelligence algorithm. First, the positioning area is divided into multiple height planes, and the receiving points are evenly distributed on each plane. The power of the received LED light source is measured, and the height classification, label classification and power mapping models are trained using artificial intelligence algorithms. Secondly, after the receiving power data of the point to be measured is input into the model, the height value and link receiving power are obtained, and the positioning coordinates are calculated by the three-sided positioning method. However, this method cannot achieve dynamic, crowded, multi-terminal positioning of indoor visible light positioning, and still cannot meet the needs of stable and continuous positioning for scenes such as large shopping malls, exhibition halls, and airports. Summary of the invention

[0005] In order to solve the problem that the prior art cannot realize dynamic, crowded, and multi-terminal indoor visible light positioning, the present invention provides a high-precision indoor visible light positioning method for complex dynamic crowd environments, which can solve the dynamic interference problem in crowded environments, improve the accuracy and robustness of the positioning system, realize adaptive positioning for complex dynamic crowd environments, and meet the high-precision and low-latency positioning requirements in smart buildings and public places.

[0006] The technical solution of the present invention to solve the technical problem is as follows:

[0007] A high-precision visible light positioning method for indoor use in a complex dynamic crowd environment comprises the following steps:

[0008] S1, start the positioning system and perform signal filtering and dynamic modulation;

[0009] S2, performing preliminary position calculation based on the quadrilateral measurement method combined with the phase and intensity information of the optical signal;

[0010] S3, using quantum-behaved particle swarm optimization (QPSO) algorithm to optimize the initial position of the receiving end;

[0011] S4, infrared sensors collect the density and dynamic changes of people in the area in real time and generate a heat map. Then, based on the fluid mechanics vortex model combined with the recurrent neural network (RNN), the impact of crowd flow on the light field is predicted to generate a crowd flow vortex model;

[0012] S5. Optimize the optical signal at the receiving end using the newly proposed Catfish Vortex Particle Algorithm (CVPA), and dynamically adjust the signal characteristics through the vortex effect;

[0013] S6, Field Programmable Gate Array Circuit (FPGA) accurately tracks and updates the position of the avalanche photodiode (APD) receiving end through an extended Kalman particle filter (EKPF);

[0014] S7, adaptively adjust the weights of QPSO, CVPA and EKPF according to the current environment and positioning accuracy requirements;

[0015] S8, FPGA obtains the final accurate position coordinates of the receiving end after multiple iterations.

[0016] Compared with the prior art, the present invention has the following beneficial effects:

[0017] 1. Improve positioning accuracy: The present invention combines the QPSO optimization algorithm, the CVPA algorithm and the EKPF filter to achieve multi-algorithm collaborative optimization in the positioning process, which can greatly improve the positioning accuracy in complex dynamic crowd environments and adapt to the real-time positioning needs of high-density crowds.

[0018] 2. Enhanced anti-interference capability: The present invention utilizes an adaptive weight fusion strategy to dynamically adjust the weights of each algorithm according to changes in crowd density and ambient light intensity, so that the system has stronger anti-interference capability in dynamic crowded environments and effectively reduces the impact of signal obstruction and environmental noise on positioning accuracy.

[0019] 3. Achieve efficient and fast convergence: The global search capability of QPSO enables the positioning system to quickly converge to a reasonable initial position in the initial stage, while the state prediction and real-time update functions of EKPF ensure the continuous optimization of the positioning process, ultimately improving the accuracy of the positioning results and meeting the real-time requirements.

[0020] 4. Taking into account normal lighting function: The present invention uses ordinary LED lamps as positioning light sources and sends positioning signals through high-frequency signal modulation, which does not affect the normal lighting effect, avoids the modification of the existing lighting system, and is convenient for large-scale deployment in public places.

[0021] 5. Adapt to complex dynamic environments: The present invention constructs a crowd vortex model to collect and analyze crowd dynamic distribution information in real time, predict the impact of crowd flow on optical signal propagation, and ensure that the positioning system maintains a stable and accurate working state in a complex crowd environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a schematic diagram of an indoor high-precision visible light positioning system for use in a complex dynamic crowd environment as described in the present invention.

[0023] Figure 2 This is the distribution diagram of the visible light positioning sending end LED and infrared sensor in a complex indoor environment described in the present invention.

[0024] Figure 3 The present invention is a flow chart of a method for indoor high-precision visible light positioning in a complex dynamic crowd environment.

[0025] Figure 4 It is a schematic diagram of the receiving end positioning optimization process based on the QPSO algorithm described in the present invention.

[0026] Figure 5 This is a schematic diagram of the dynamic crowd density capture and signal interference adjustment process based on infrared sensors and RNN according to the present invention.

[0027] Figure 6 This is the indoor crowd density detection thermal map described in the present invention.

[0028] Figure 7 This is the indoor crowd density detection thermal map after reprocessing and optimization as described in the present invention.

[0029] Figure 8This is a diagram of the indoor human flow vortex model described in the present invention.

[0030] Fig. 9 This is the RNN predicted crowd density map based on time series processing described in the present invention (measured in 10 time steps).

[0031] Fig.10 This is the receiving end positioning diagram (based on 10 receiving ends) of the dynamic changes of marker prediction in a complex dynamic crowd environment described in the present invention.

[0032] Fig.11 It is a schematic diagram of the dynamic positioning optimization process based on the catfish eddy current algorithm described in the present invention.

[0033] Fig.12 It is a schematic diagram of the positioning optimization process based on particle filter and extended Kalman filter described in the present invention.

[0034] Fig.13 It is a schematic diagram of the weight adaptive adjustment process based on environmental feedback described in the present invention.

[0035] Fig.14 It is a schematic diagram of the high-precision position determination process based on multiple iterative optimization and error correction according to the present invention.

[0036] Fig.15 Schematic diagram of the box-line distribution of the positioning error at different time steps described in the present invention (based on 10 time steps).

[0037] Fig.16 middle Fig.16 (a) Fig.16 (b) are schematic diagrams of the random motion and regular motion positioning performance of the single receiving end described in the present invention in a complex dynamic crowd environment. DETAILED DESCRIPTION

[0038] The present invention will be described in further detail below with reference to the accompanying drawings.

[0039] like Figure 1 As shown in the figure, an indoor high-precision visible light positioning system for complex dynamic crowd environments includes n LED lights, n signal modulation modules, m APDs and m FPGAs. In a larger indoor environment, the LED light sources are arranged in a rectangular array in equal proportions, and the infrared sensor is located at the center of a two-dimensional square formed by every four LED light sources. The LED light source is driven by the signal modulation module to emit a light signal, which is transmitted through the atmospheric channel and received by the receiving end formed by the APD and converted to photoelectric conversion for processing by the FPGA.

[0040] like Figure 2 Shown is the distribution diagram of visible light positioning transmitter LED and infrared sensor in a complex indoor environment.

[0041] like Figure 3 As shown, the overall steps of a high-precision visible light positioning method for indoor complex dynamic crowd environments are as follows:

[0042] S1. Start the positioning system and perform signal filtering and dynamic modulation: In a larger indoor environment, the LED array light source at the transmitting end is arranged in a rectangular array in equal proportions, and the infrared sensor is at the center of a two-dimensional square formed by every 4 LED light sources. The LED array light source is driven by the signal modulation module to emit a light signal, which is received by the receiving end composed of the APD after propagation through the atmospheric channel and photoelectrically converted to the FPGA for processing. Subsequently, the filter algorithm built into the FPGA performs preliminary filtering on the received light signal to remove environmental noise and improve signal clarity. The signal modulation module dynamically adjusts the modulation frequency and signal strength according to changes in ambient light intensity and crowd density;

[0043] S2. Preliminary position calculation: FPGA uses the phase and intensity information of the filtered optical signal, combined with the light propagation model and based on the four-sided measurement method to calculate the distance between the receiving end and the four LED light sources with the best optical signal indicators and unique identification IDs;

[0044] S3, Position Optimization Search: FPGA uses the QPSO algorithm to further search and optimize the initial position of the receiving end;

[0045] S4. Prediction of crowd flow impact: Infrared sensors collect crowd density and dynamic changes in the area in real time and generate heat maps. Then, based on the fluid mechanics vortex model combined with the RNN network, the impact of crowd flow on the light field is predicted to generate a crowd flow vortex model.

[0046] S5, Dynamic Signal Optimization: FPGA uses the CVPA algorithm to optimize the optical signal at the receiving end and dynamically adjusts the signal characteristics through the eddy current effect;

[0047] S6, Position tracking adjustment: FPGA accurately tracks and updates the position of the APD receiver through the EKPF filter;

[0048] S7, adaptive weight adjustment: FPGA adaptively adjusts the weights of QPSO, CVPA and EKPF according to the current environment and positioning accuracy requirements;

[0049] S8, precise positioning output: After multiple iterations, FPGA obtains the final precise position coordinates of the receiving end.

[0050] In S1, the LED light source in the system integrates a communication module while maintaining normal lighting functions. It is controlled by a signal modulation module and sends out positioning signals with high-frequency signal modulation. The signal modulation module uses PWM modulation technology to embed the positioning signal in a high-frequency range that is imperceptible to the human eye (usually higher than 1 kHz). PWM modulation can be expressed as:

[0051] I(t)=I base +A·sin(2πf mod t+φ)

[0052] Where: I base is the basic light intensity that meets the lighting requirements; A is the amplitude of the modulation signal; f mod is the modulation frequency, which is set above the threshold that the human eye can perceive; t represents the time variable, which is used to describe the change of the signal over time; φ is the phase of the positioning signal, which is used for the unique encoding of the signal.

[0053] The signal modulation module will dynamically adjust the modulation frequency and signal strength according to the environmental feedback. Assume that the ambient light intensity is L env , the population density is D crowd , then the modulation frequency and intensity can be expressed as:

[0054] f mod =f0+k1·L env +k2·D crowd

[0055] A=A0+k3·D crowd

[0056] Among them: f0 and A0 are the basic modulation frequency and signal strength; k1, k2, k3 are adjustment coefficients used to adapt to environmental changes.

[0057] The signal of each LED light source uses phase encoding generated by PRNG. The phase encoding formula is expressed as:

[0058] φ(t)=φ0+PRNG(t)

[0059] Where: φ0 is the basic phase offset; PRNG(t) is the pseudo-randomly generated phase change.

[0060] In a complex and dynamic crowd environment, the FPGA system filters the optical signal immediately after receiving it. The filter has adaptive characteristics and can better adapt to changes in environmental noise. The center frequency and bandwidth of the filter will be adjusted according to the real-time L env and D crowd Make dynamic adjustments. The adjustment formula is as follows:

[0061] f center =fmod +k·L env +m·D crowd

[0062] Where: f center is the center frequency of the filter after dynamic adjustment; f mod is the initial modulation frequency; k and m are the adjustment coefficients.

[0063] In S2, the filtered optical signal will use the four-sided measurement method according to the light propagation model to calculate the distance between the receiving end and the four LED light sources with the best optical signal indicators and unique identification IDs based on the phase and intensity information of the received signal. Let the coordinates of the receiving end be (x, y), and the coordinates of the four light sources that can be distinguished by unique IDs are (x1, y1), (x2, y2), (x3, y3) and (x4, y4), and the corresponding distances are d1, d2, d3, d4. Then the initial position of the receiving end can be obtained by solving the following set of equations:

[0064]

[0065] Assume that N measurements are made and the distance value measured is d i (i=1,2,…,N), the final distance calculation result is:

[0066]

[0067] like Figure 4 As shown in S3, the optimization process of QPSO includes the following steps:

[0068] S31, Initialization: Randomly initialize the particle swarm in the search space and set the initial position coordinates (x i ,y i ) and fitness value;

[0069] S32, calculating fitness: for each particle, calculating its fitness value, the fitness function is used to evaluate the proximity between the current position of each particle and the target position;

[0070] S33, position update: Update the position of each particle according to the quantum behavior model. The particle position update formula is as follows:

[0071]

[0072] Where: x i (t+1) is the new position of the i-th particle in the t+1 generation; P i is the best historical position of the i-th particle; M iis the current global optimal position; β is the shrink-expansion coefficient, which is used to control the convergence speed of the search; u is a uniform random number in the range of [0, 1], which is used to introduce randomness.

[0073] S34, global optimal update: in each iteration, the global optimal position M is updated by comparing the fitness values;

[0074] S35, convergence judgment: when the preset convergence conditions are met (such as the maximum number of iterations or the fitness reaches a certain threshold), the iteration is stopped, and the global optimal position at this time is the final optimization positioning result.

[0075] like Figure 5 As shown in Figure 4, in S4, the infrared sensor is placed at a high position such as the indoor ceiling, which can capture the crowd density and dynamic changes in the area in real time. The original signal received by the infrared sensor is converted into numerical crowd density data through preprocessing to generate a thermal map before and after processing. Figure 6 , Figure 7 As shown. Each heat map grid unit H i，j Represents the crowd density on the regional grid (i, j), and the specific calculation formula is:

[0076]

[0077] Where: f(d k , T k ) represents the thermal radiation intensity of the kth target, based on the distance d k and temperature T k ; n is the number of detected targets in the current area.

[0078] The system uses a vortex model based on fluid mechanics to describe the impact of crowd flow on the light field. The specific steps include:

[0079] S41. Establish vortex model: treat crowd flow as fluid, and describe its flow characteristics in space through the vortex model in fluid mechanics. According to vortex theory, densely populated areas will form "high-pressure areas", while sparsely populated areas will form "low-pressure areas", producing crowd vortex effects (such as Figure 8 shown);

[0080] S42, RNN-based prediction and dynamic update: Fig. 9 As shown in the figure, the system combines the RNN network in deep learning to predict the vortex model of human flow, takes 10 time steps, predicts the density of human flow according to the time series, and displays it intuitively through different colors. Fig.10 As shown in the figure, the RNN network is used to process the predicted crowd density again to obtain the receiver positioning map with the dynamic change of the predicted mark. Assume that the output of the crowd vortex model is V flow (t), then:

[0081] V flow (t)=RNN(H t-1 , H t-2 , ..., H t-n )

[0082] Where: H t-1 , H t-2 , ..., H t-n Represents the heat map data of the first n time steps; V flow (t) represents the predicted value of the passenger flow velocity field at the current moment.

[0083] like Fig.11 As shown, in S5, the newly proposed CVPA algorithm is used to dynamically optimize the optical signal at the receiving end. The optimization process of CVPA includes the following steps:

[0084] S51, Initialization: Randomly generate a particle swarm in the search space, and set the initial position according to the receiver position optimized by the QPSO algorithm. Assume that each particle represents a possible position of the receiver, and set its initial speed and position;

[0085] S52, eddy current disturbance model: Based on the crowd eddy current model generated in S4, a "eddy current zone" is created in the high crowd density area, and the phase and intensity of the optical signal are dynamically adjusted through the disturbance coefficient to adapt the signal to the current crowd environment. The eddy current disturbance formula is:

[0086] V disturb (t) = α·sin(θ t )·D crowd

[0087] Where: V disturb (t) is the signal disturbance intensity at the current moment; α is the disturbance coefficient, which controls the eddy current intensity; θ t is the phase angle of the disturbance, which fluctuates randomly to simulate the catfish vortex behavior; D crowd is the crowd density, obtained from the crowd flow heat map.

[0088] S53, dynamic signal adjustment: According to the eddy current disturbance model, adjust the optical signal received by the receiving end. The adjustment process includes correcting the phase and intensity of the optical signal to cope with the disturbance of human flow in the dynamic environment. The update formula of this step is:

[0089] I optimized (t) = I received (t)+V disturb (t)

[0090] Where: I optimized (t) is the optical signal strength after optimization at the current moment; Ireceived (t) is the original optical signal received by the receiving end.

[0091] S54. Catfish behavior guidance: Select "catfish" particles from the particle swarm, which have a higher fitness value (particles close to the target position). Catfish particles guide other particles to converge to the global optimal position by simulating eddy current disturbances to avoid falling into the local optimum. The update formula of catfish guidance behavior is:

[0092] x i (t+1)=x i (t)+c disturb ·(x catfish (t)-x i (t))

[0093] Where: x i (t+1) is the new position of the ith particle in the t+1 generation; x catfish (t) is the current position of the catfish particle; c disturb is the eddy current disturbance coefficient.

[0094] S55, convergence judgment: after multiple iterations, when the overall fitness value of the particle swarm no longer changes significantly, it is judged to be converged. At this time, the global optimal position is the optimized positioning result.

[0095] like Fig.12 As shown in S6, the optimization process of EKPF includes the following steps:

[0096] S61. Initial particle sampling: Based on the initial position of the receiver optimized by QPSO and CVPA, particle sampling is performed with this position as the center. Assume that the number of particles is N, and each particle x i Represents a possible location of the receiver and assigns an initial weight wi =1 / N;

[0097] S62, state prediction: predict the receiving end position through the extended Kalman filter. The state transition model can be expressed as:

[0098] x t+1 =f(x t )+w t

[0099] Where: x t represents the position state of the receiving end at time t; f(x t ) represents the state transfer function, which is used to describe the change of the receiving end position over time; w t is the process noise, which is assumed to conform to the zero-mean Gaussian distribution.

[0100] S63, measurement update: The receiving end measures and updates the position state according to the real-time received optical signal and environmental data. The measurement equation is:

[0101] z t =h(x t )+v t

[0102] Where: z t is the measured value at time t (including optical signal and environmental data); h(x t ) is the measurement function, which describes the relationship between the actual observed receiving end position and the true position; v t To measure the noise, we assume a zero-mean Gaussian distribution.

[0103] S64, particle weight update: update the weight of each particle according to the observed data. The weight update formula for each particle is:

[0104]

[0105] in: is the updated weight of the ith particle at time t+1; z t is the observed data; h(x i ) is particle x i The measured value.

[0106] S65, Resampling: Resample according to the particle weight distribution, retain the particles with higher weights and discard the particles with lower weights, so as to generate a new particle swarm. The resampling process can avoid the problem of particle degradation caused by excessive concentration of particle weights;

[0107] S66, state estimation: The final receiving end position is obtained by weighted average method, the formula is:

[0108]

[0109] in: is the estimated position at time t+1; is the position of the i-th particle after resampling.

[0110] like Fig.13 As shown in S7, the adaptive weight fusion strategy ensures that each algorithm contributes the greatest effect in its optimal stage. The core is to dynamically adjust the weight of each algorithm according to different environmental conditions (such as crowd density, signal quality, etc.) and the staged requirements of the algorithm, which includes the following steps:

[0111] S71. Environmental feedback collection: The system uses real-time collected environmental data (such as crowd density, light signal intensity, noise level, etc. generated by heat maps) to determine the complexity of the current environment. Assume that the crowd density is D crowd and signal quality as SNR, the environmental state can be expressed as:

[0112] E=f(D crowd ,SNR)

[0113] Where E is the environmental complexity index, and f is the functional relationship of weight distribution.

[0114] S72. Weight calculation: Calculate the weights of QPSO, CVPA and EKPF according to the environment complexity E. Suppose the weights of each algorithm are w QPSO 、w CVPA and w EKPF , then the total weight W satisfies the following constraints:

[0115] w QPSO +w CVPA +w EKPF =1

[0116] S73, weight allocation function: The system uses a hybrid function of exponential decay and growth to ensure adaptive adjustment of weights in different environments. The weight allocation formula is:

[0117]

[0118] w EKPF =1-w QPSO -w CVPA

[0119] Among them: α and β are adjustment parameters, which are set according to different environmental characteristics; w QPSO 、w CVPA 、w EKPF The value of floats in the interval [0, 1].

[0120] S74, Dynamically update weights: As environmental data changes in real time, the system continuously adjusts the weights of each algorithm to ensure that the current algorithm combination can adapt to positioning requirements in complex environments. At the end of each iteration, the weights are recalculated and the contribution of each algorithm is adjusted.

[0121] like Fig.14 As shown, in S8, the iterative process of calculating the final position is:

[0122] S81, initial position setting: In S3, the QPSO provides the preliminary position coordinates of the receiving end, and gradually optimizes them in the subsequent steps (S4 to S7) to obtain gradually accurate position coordinates;

[0123] S82, multiple iteration optimization: Through the iterative combination of the three algorithms of QPSO, CVPA, and EKPF, the system gradually converges to the optimal position. In each iteration, the system adjusts the weights of each algorithm based on the latest environmental feedback and dynamically updates the receiving end position until the convergence condition is met. The convergence condition can be set as:

[0124] A. The position change is less than the threshold: Assume that the position change of the receiving end after each iteration is Δ x , when Δ x <∈, where ∈ is the preset convergence threshold, the system determines that it has converged.

[0125] B. Weight stability: When the change in the adaptive weight after adjustment is less than the set value, it means that the system has reached a stable state under the current environment.

[0126] S83, position coordinate output: After the convergence condition is met, the system takes the coordinates of the receiving end at the current moment as the final positioning result and outputs it as the precise position of the receiving end. Suppose the final position coordinates are (x final ,y final , z final ),but:

[0127]

[0128] Where (x n ,y n , z n ) represents the receiving end coordinates at the nth iteration.

[0129] S84, Error correction and accuracy confirmation: The system will perform error correction to ensure positioning accuracy before the final output position. Based on the historical data of the previous iterations and the current environmental feedback, the system calculates the positioning error and corrects the final result according to the error correction coefficient. The error correction formula is:

[0130]

[0131] Where: Δ error is the average positioning error; N is the position data of the most recent N iterations, which is used to calculate the correction of the final position.

[0132] S85, accuracy assurance measures: If the final positioning error is greater than the preset accuracy threshold δ, the system will readjust the algorithm weights and iterate again to further improve the positioning accuracy. crror <δ, the positioning result is output, such as Fig.15 The following is a schematic diagram of the simulation results of the box line distribution of positioning error at different time steps (based on 10 time steps). The following is a schematic diagram of the simulation results of the random motion and regular motion positioning performance of a single receiver in a complex dynamic crowd environment. Fig.16 (a) Fig.16 (b) as shown.

Claims

1. A high-precision visible light positioning method for indoor use in complex dynamic crowd environments, characterized in that: The method comprises the following steps: S1, start the positioning system and perform signal filtering and dynamic modulation; S2, performing preliminary position calculation based on the quadrilateral measurement method combined with the phase and intensity information of the optical signal; S3, using QPSO algorithm to optimize the initial position of the receiving end; S4, infrared sensors collect the density and dynamic changes of people in the area in real time and generate a heat map. Then, based on the fluid mechanics vortex model combined with the RNN network, the impact of crowd flow on the light field is predicted to generate a crowd flow vortex model; S5. Optimize the optical signal at the receiving end using the CVPA algorithm and dynamically adjust the signal characteristics through the eddy current effect; S6, FPGA accurately tracks and updates the position of the APD receiver through the EKPF filter; S7, adaptively adjust the weights of QPSO, CVPA and EKPF according to the current environment and positioning accuracy requirements; S8, FPGA obtains the final accurate position coordinates of the receiving end after multiple iterations.

2. The indoor high-precision visible light positioning method for complex dynamic crowd environment according to claim 1, characterized in that: In the step S1, in a relatively large indoor environment, the LED array light source at the transmitting end is arranged in a rectangular array in equal proportions, and the infrared sensor is located at the two-dimensional center of a square formed by every four LED light sources; the LED array light source is driven by the signal modulation module to emit a light signal, which is received by the receiving end formed by the APD after propagation through the atmospheric channel and photoelectrically converted to the FPGA for processing; then the filter algorithm built into the FPGA performs preliminary filtering processing on the received light signal to remove environmental noise and improve signal clarity, and the signal modulation module dynamically adjusts the modulation frequency and signal strength according to changes in ambient light intensity and crowd density; The signal modulation module adopts PWM modulation technology, which is expressed as: I(t)=I base +A·sin(2πf mod t+φ) Where: I base is the basic light intensity that meets the lighting requirements; A is the amplitude of the modulation signal; f mod is the modulation frequency; t represents the time variable; φ is the phase of the positioning signal, which is used for the unique encoding of the signal; Assume that the ambient light intensity is L env , the population density is D crowd , then the modulation frequency and intensity can be expressed as: f mod =f0+k1·L env +k2·D crowd A=A0+k3·D crowd Among them: f0 and A0 are the basic modulation frequency and signal strength; k1, k2, k3 are the adjustment coefficients; The signal of each LED light source uses phase encoding generated by PRNG; the phase encoding formula is expressed as: φ(t)=φ0+PRNG(t) Where: φ0 is the basic phase offset; PRNG(t) is the pseudo-randomly generated phase change.

3. The indoor high-precision visible light positioning method for complex dynamic crowd environment according to claim 1, characterized in that: In step S2, the filtered optical signal will use the four-sided measurement method according to the light propagation model to calculate the distance between the receiving end and four LED light sources with the best optical signal indicators and unique identification IDs based on the phase and intensity information of the received signal.

4. The indoor high-precision visible light positioning method for complex dynamic crowd environment according to claim 1, characterized in that: In step S3, the optimization process of QPSO includes the following steps: S31, Initialization: Randomly initialize the particle swarm in the search space and set the initial position coordinates (x i ,y i ) and fitness value; S32, calculating fitness: for each particle, calculating its fitness value, the fitness function is used to evaluate the proximity between the current position of each particle and the target position; S33, position update: update the position of each particle according to the quantum behavior model; S34, global optimal update: in each iteration, the global optimal position M is updated by comparing the fitness values; S35, convergence judgment: When the preset convergence conditions are met, such as the maximum number of iterations or the fitness reaches a certain threshold, the iteration is stopped, and the global optimal position at this time is the final optimization positioning result.

5. The indoor high-precision visible light positioning method for complex dynamic crowd environment according to claim 1, characterized in that: In step S4, the infrared sensor is arranged at a high position such as the indoor ceiling, so as to capture the crowd density and dynamic changes in the area in real time; the original signal received by the infrared sensor is converted into numerical crowd density data through preprocessing, and a heat map is generated; each heat map grid unit H i,j Represents the crowd density on the regional grid (i, j), and the calculation formula is: Where: f(d k ,T k ) represents the thermal radiation intensity of the kth target, based on the distance d k and temperature T k ; n is the number of detected targets in the current area; The system uses an eddy current model based on fluid mechanics to describe the impact of crowd flow on the light field; the specific steps include: S41. Establish eddy flow model: treat crowd flow as fluid, and describe its flow characteristics in space through the eddy flow model in fluid mechanics; S42, RNN-based prediction and dynamic update: The system combines the RNN network in deep learning to predict the human flow vortex model; let the output of the human flow vortex model be V flow (t), then: V flow (t)=RNN(H t-1 ,H t-2 ,…,H t-n ) Where: H t-1 ,H t-2 ,…,H t-n Represents the heat map data of the first n time steps; V flow (t) represents the predicted value of the passenger flow velocity field at the current moment.

6. The indoor high-precision visible light positioning method for complex dynamic crowd environment according to claim 1, characterized in that: In step S5, the newly proposed CVPA algorithm is used to dynamically optimize the optical signal at the receiving end, including the following steps: S51, initialization: randomly generate a particle swarm in the search space, and set the initial position according to the receiving end position after the QPSO algorithm is optimized; assume that each particle represents a possible position of the receiving end, and set its initial speed and position; S52, eddy current disturbance model: Based on the crowd eddy current model generated in S4, a "eddy current zone" is created in the high crowd density area, and the phase and intensity of the optical signal are dynamically adjusted through the disturbance coefficient to adapt the signal to the current crowd environment; the eddy current disturbance formula is: V disturb (t)=α·sin(θ t )·D crowd Where: V disturb (t) is the signal disturbance intensity at the current moment; α is the disturbance coefficient, which controls the eddy current intensity; θ t is the phase angle of the disturbance, which fluctuates randomly to simulate the catfish vortex behavior; D crowd is the crowd density, obtained from the crowd flow heat map; S53, dynamic signal adjustment: according to the eddy current disturbance model, adjust the optical signal received by the receiving end; the update formula is: I optimized (t)=I received (t)+V disturb (t) Where: I optimized (t) is the optical signal strength after optimization at the current moment; I received (t) is the original optical signal received by the receiving end; S54, catfish behavior guidance: select "catfish" particles from the particle swarm, which must have a higher fitness value, that is, particles close to the target position; catfish particles guide other particles to converge to the global optimal position by simulating eddy current disturbances to avoid falling into the local optimum; the update formula of catfish guidance behavior is: x i (t+1)=x i (t)+c disturb ·(x catfish (t)-x i (t)) Where: x i (t+1) is the new position of the ith particle in the t+1 generation; x catfish (t) is the current position of the catfish particle; c disturb is the eddy current disturbance coefficient; S55, convergence judgment: after multiple iterations, when the overall fitness value of the particle swarm no longer changes significantly, it is judged to be converged; at this time, the global optimal position is the optimized positioning result.

7. The indoor high-precision visible light positioning method for complex dynamic crowd environment according to claim 1, characterized in that: In step S6, the optimization process of EKPF includes the following steps: S61, initial particle sampling: based on the initial position of the receiver optimized by QPSO and CVPA, particle sampling is performed with this position as the center; S62, state prediction: predicting the receiving end position through an extended Kalman filter; S63, measurement update: the receiving end measures and updates the position state according to the optical signal and environmental data received in real time; S64, particle weight update: update the weight of each particle according to the observed data; S65, Resampling: Resampling is performed according to the particle weight distribution, retaining particles with higher weights and discarding particles with lower weights, thereby generating a new particle swarm; the resampling process can avoid the problem of particle degradation caused by excessive concentration of particle weights; S66, state estimation: The final receiving end position is obtained by weighted averaging method.

8. The indoor high-precision visible light positioning method for complex dynamic crowd environment according to claim 1, characterized in that: In step S7, the adaptive weight fusion strategy ensures that each algorithm contributes the greatest effect in its optimal stage. The core of the strategy is to dynamically adjust the weight of each algorithm according to different environmental conditions and the stage requirements of the algorithm. The strategy includes the following steps: S71, Environmental feedback collection: The system uses real-time collected environmental data to determine the complexity of the current environment; assuming that the crowd density is D crowd and signal quality as SNR, the environmental state can be expressed as: E=f(D crowd (SNR) Where E is the environmental complexity index, and f is the functional relationship of weight distribution; S72. Weight calculation: According to the environment complexity E, calculate the weights of QPSO, CVPA and EKPF; let the weights of each algorithm be w QPSO 、w CVPA and w EKPF , then the total weight W satisfies the following constraints: In QPSO +in CVPA +in EKPF =1 S73, weight distribution function: The system adopts a mixed function of exponential decay and growth, and the weight distribution formula is: In EKPF =1-in QPSO -In CVPA Among them: α and β are adjustment parameters, which are set according to different environmental characteristics; w QPSO 、w CVPA 、w EKPF The value of floats in the interval [0,1]; S74, Dynamically update weights: As environmental data changes in real time, the system continuously adjusts the weights of each algorithm to ensure that the currently used algorithm combination can adapt to positioning requirements in complex environments; at the end of each iteration, the weights are recalculated and the contribution of each algorithm is adjusted.

9. The indoor high-precision visible light positioning method for complex dynamic crowd environment according to claim 1, characterized in that: In step S8, the iterative process of final position calculation is: S81, initial position setting: In S3, the QPSO provides the preliminary position coordinates of the receiving end, and gradually optimizes them in subsequent steps to obtain gradually accurate position coordinates; S82, multiple iteration optimization: Through the iterative combination of the three algorithms of QPSO, CVPA, and EKPF, the system gradually converges to the optimal position; during each iteration, the system adjusts the weights of each algorithm based on the latest environmental feedback and dynamically updates the receiving end position until the convergence condition is met; the convergence condition can be set as: A. The position change is less than the threshold: Assume that the position change of the receiving end after each iteration is Δ x , when Δ x <∈, where ∈ is the preset convergence threshold, the system determines that it has converged; B. Weight stability: When the change amplitude after adaptive weight adjustment is less than the set value, it means that the system has reached a stable state under the current environment; S83, position coordinate output: after the convergence condition is met, the system takes the coordinates of the receiving end at the current moment as the final positioning result, and outputs it as the precise position of the receiving end; S84, Error correction and accuracy confirmation: The system will perform error correction to ensure positioning accuracy before the final output position; Based on the historical data of the previous iterations and the current environmental feedback, the system calculates the positioning error and corrects the final result according to the error correction coefficient; the error correction formula is: Where: Δ error is the average positioning error; N is the position data of the most recent N iterations, which is used to calculate the correction of the final position; S85, accuracy assurance measures: If the final positioning error is greater than the preset accuracy threshold δ, the system will readjust the algorithm weights and iterate again to further improve the positioning accuracy; when the average error Δ error <δ, the positioning result is output.

Images