A visible light positioning method and system based on deep learning and particle swarm

Abstract

A visible light positioning method and system based on deep learning and particle swarm, signal modulation emission and signal receiving, preprocessing process, including modulating each lamp in the basic positioning unit into a signal of different frequency, PD receives the light signal, separates through the frequency converter and the band pass filter, unifies into the reference lamp frequency, exchanges the modulation frequency of two lamps to perform the same operation; in the phase difference estimation network, the key sequence is extracted, the output vector is obtained through the convolution layer and the activation layer, the transposition and connection are performed, and then the phase of each lamp is output through the full connection layer, and the phase difference is calculated; according to the relationship between the distance and the phase difference, the distance between each lamp and the PD is obtained; the particle swarm position updating mode based on the motion state is adopted to make the trajectory more smooth. The application adopts the deep learning mode, realizes the phase difference estimation through the set network structure, and simultaneously proposes the particle filtering method based on the motion model to improve the positioning robustness.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of visible light positioning, in particular to a visible light positioning method and system based on deep learning and particle swarm. BACKGROUND

[0002] The most typical method in the current PD-based visible light positioning positioning scheme is to use a pre-calibrated ranging model and RSS to calculate the distance, and then use the intersection method for positioning. The main defects are:

[0003] a) The performance of the positioning method based on the ranging model depends on the accuracy of the model parameters;

[0004] b) The performance of the traditional phase estimation method based on IQ value is affected by the errors caused by the filter, time synchronization error, receiver noise, system error, etc.

[0005] c) The pure signal positioning result is often not robust enough, and the trajectory smoothness is not enough. SUMMARY

[0006] In order to solve the above problems existing in the prior art, the present application proposes a visible light positioning scheme based on deep learning and particle swarm.

[0007] In order to achieve the above purpose, the present application proposes a visible light positioning method based on deep learning and particle swarm, which includes the following processing,

[0008] Signal modulation and transmission and signal reception and preprocessing process, including modulating each lamp in the basic positioning unit into a signal of different frequency, photodiode PD receiving optical signal, using a band-pass filter to separate the signal of each lamp, corresponding through the frequency converter and the band-pass filter, the frequency of all lamp signals is unified into the frequency of the reference lamp, the modulation frequency of two lamps is exchanged, the same operation is performed, and the obtained signal sequence is input to the subsequent phase difference estimation network;

[0009] Phase difference estimation network processing, including key sequence extraction from each signal sequence in the phase difference estimation network, the extraction result is respectively subjected to convolution layer and activation layer to obtain corresponding output vector, the output vectors of all signal sequences are transposed and connected, and then subjected to full connection layer, outputting the phase of each lamp, and calculating the phase difference through the phase;

[0010] Distance measurement according to phase difference, including obtaining the distance between each lamp and the photodiode PD according to the relationship between distance and phase difference;

[0011] The particle swarm position updating method based on the motion state makes the trajectory smoother.

[0012] Furthermore, a basic positioning unit includes 5 LED lights, respectively denoted as LED1, LED2, LED3, LED4, LED5, and the frequency of each light is respectively denoted as f 1 ，f 2 ，f 3 ，f 4 ，f 5, LED1 is set as a reference light, and LED1 and LED2 are exchange modulation frequency lights.

[0013] Furthermore, the implementation of the signal modulation emission and the signal receiving and preprocessing process includes the following steps,

[0014] Step 1, set each light to modulate into a signal with different frequencies, wherein f 2 ，f 3 ，f 4 ，f 5 is f 1 times an odd number, and then emit;

[0015] Step 2, the PD receives the optical signal, and the received is a combined signal of multiple lights;

[0016] Step 3, use a band-pass filter to separate the signal of each light;

[0017] Step 4, pass each separated light signal through a frequency down-converter and a band-pass filter, so that the frequency of all light signals is unified into the frequency of the reference light, and output a signal sequence;

[0018] Step 5, exchange the modulation frequencies of LED1 and LED2, repeat Steps 2, 3, and 4, and output the signal sequence of the two lights;

[0019] Step 6, input the signal sequence output from Steps 4 and 5 into a phase difference estimation network to estimate the phase difference.

[0020] Furthermore, in Step 1, f 2 =3f 1 ， f 3 =5f 1 ， f 4 =7f 1 ， f 5 =9f 1.

[0021] Furthermore, the relationship between the distance and the phase difference is as follows,

[0022]

[0023] In the formula, denotes the speed of light, denotes the phase difference between LED1 and LED2, represents the phase difference between LED2 and LED1, represents the phase difference between LED1 and LED3, represents the phase difference between LED1 and LED4, represents the phase difference between LED1 and LED5, respectively represent the distances from LED1, LED2, LED3, LED4 and LED5 to PD.

[0024] Further, in the particle swarm position updating method based on the motion state, firstly, the initial positioning result of the photodiode (PD) is obtained by using the least square method; secondly, the weight is constructed by using the distance observation, and the particle swarm position is updated in combination with the motion state of the terminal; finally, the position of the terminal is updated by using the particle swarm position updating result.

[0025] On the other hand, the application provides a visible light positioning system based on deep learning and particle swarm, which is used to realize the visible light positioning method based on deep learning and particle swarm.

[0026] Further, the application comprises a processor and a memory, the memory is used to store program instructions, and the processor is used to call the stored instructions in the memory to execute the visible light positioning method based on deep learning and particle swarm.

[0027] Alternatively, the application comprises a readable storage medium, and the readable storage medium stores a computer program, and the computer program is executed to realize the visible light positioning method based on deep learning and particle swarm.

[0028] The application has the following advantages:

[0029] 1. The TDOA-based scheme proposed in the application reduces the dependence on empirical parameters compared with the traditional ranging model.

[0030] 2. The application proposes a CNN network for estimating the phase difference of LEDs, and the network is superior to the traditional IQ value-based method in terms of the accuracy and robustness of phase difference estimation.

[0031] 3. The application proposes a particle swarm position updating algorithm based on the motion state, and the parameters of the position updater can be adaptively adjusted according to the motion state, which improves the positioning accuracy and robustness and makes the trajectory smoother.

[0032] The application scheme is simple and convenient to implement, has strong practicability, solves the problems of low practicability and inconvenience in actual application existing in the related art, and has important market value. ​​​​BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1 The basic positioning unit base station layout diagram for the embodiment of the present application is shown in the figure.

[0034] Figure 2 The data processing flowchart for the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0035] The technical solutions of the present application are specifically described below in combination with the drawings and embodiments.

[0036] The performance of a visible light positioning system based on a ranging model is greatly dependent on the accuracy of the ranging model parameters, and the use of a TDOA mode can avoid the dependence on empirical parameters. The present application discloses a visible light positioning scheme based on TDOA, and considering that the phase difference calculation process will be affected by time synchronization errors, filter errors, receiver noise, etc., the present application intends to use a deep learning method to achieve phase difference estimation by designing a special network structure. At the same time, in order to improve the positioning robustness, a particle filtering method based on a motion model is proposed.

[0037] Related term explanation

[0038] Photodiode (PD)

[0039] Time difference of arrival (TDOA)

[0040] Visible light positioning (VLP)

[0041] Received signal strength (RSS)

[0042] Conventional neural network (CNN)

[0043] Down converter (DC)

[0044] Band-pass filter (BPF)

[0045] The visible light positioning method based on deep learning and particle swarm provided in the embodiment of the present application comprises the following steps:

[0046] a. Signal modulation transmission, and signal receiving and preprocessing process (see Figure 1, for example, a basic positioning unit usually includes 5 LED lights, which are LED1, LED2, LED3, LED4, and LED5. LED1 is set as the reference light, LED1 and LED2 are exchange modulation frequency lights, and the frequency of each light is f 1 ，f 2 ，f 3 ，f 4 ，f 5

[0047] Step 1: Each light is modulated into a signal of different frequency (where the modulation signal satisfies that the modulation frequency of each light is an odd multiple of the reference light, preferably f 2 =3f 1 ， f 3 =5f 1 ， f 4 =7f 1 ， f 5 =9f 1 ), and is emitted.

[0048] Step 2: The PD receives the light signal, which is a combined signal of multiple lights.

[0049] Step 3: Use a bandpass filter to separate each light signal.

[0050] Step 4: Pass each separated light signal through a frequency downconverter and a bandpass filter, so that the frequency of all light signals is unified to the frequency of the reference light. The processed signal sequence of 5 lights can be output.

[0051] Step 5: Exchange the modulation frequencies of LED1 and LED2, repeat steps 2, 3, and 4, and output the signal sequence of these two lights after repeating.

[0052] Step 6: Input the 7 signal sequences output from Step 4 and Step 5 into the subsequent phase difference estimation network to estimate the phase difference. These signal sequences are denoted as f 1 f 2 f 2 f 1 f 3 f 4 f 5 . Among them, LED1 and LED2 are first modulated into f 1 ，f 2 , and then the modulation frequencies of the two lights are exchanged, becoming f 2 ，f 1 .

[0053] Referring to Figure 2The above steps, Steps 1-6, constitute the signal reception and preprocessing part:

[0054] The photodiode (PD) receives the signal, which first passes through the bandpass filter (BPF). i Separate the first i The signals of each lamp (i=1,2,3,4,5), the cutoff frequency of the bandpass filter [ f i -width, f i +width The center frequency is f i (with the first) i Taking a lamp as an example, width It is half the filter bandwidth, as set in this patent. f 1 / 2).

[0055] DC i ( i =1,2,3,4,5) represents the downconverter through which the i-th signal passes (mainly to achieve frequency shifting and ultimately frequency unification), so that the frequencies of all signals are eventually unified to 1. f 1 / 2 Even after down-conversion, other frequency components still exist, so a bandpass filter (BPF) is needed to unify the signals from each lamp. f 1 / 2 The signal is separated, and the cutoff frequency of this BPF is [ f 1 / 2-width, f 1 / 2+width The center frequency is f 1 / 2 The top part shows the operation after LED1 and LED2 have their modulation frequencies swapped (LED1's modulation frequency becomes...). f 2. The modulation frequency of LED2 becomes f 1) Same as above, except the center frequency of the bandpass filter has changed (i.e., BPF). i (i=1,2), the corresponding downconverter also changed (i.e., DC). i ', i =1,2), and finally passed through the bandpass filter BPF' (cutoff frequency [ f 2 / 2-width, f 2 / 2+width The center frequency is f 2 / 2 ).

[0056] b) See also Figure 2 The processing flow of the phase difference estimation network:

[0057] Step1: After the received optical signal passes through the down converter and the band-pass filter, the signals of each lamp are separated, and the vector composed of all lamp signal sequences is taken as the input of the network (i.e. the last output of 7 signal sequences in step a). Since the amplitude of the signal vector is very small, it is multiplied by a fixed ratio first. In specific implementation, the signal amplitude is greater than 1 after multiplication, and the signal amplitude is specifically set.

[0058] Step2: Considering the Gibbs phenomenon, the signal sequence will be attenuated initially, and part of the sequence will be extracted from the middle of each signal sequence, i.e. key sequence extraction, which can not only avoid the influence of signal attenuation, but also realize data dimension reduction.

[0059] Step3: The extraction results of each signal sequence are respectively subjected to a convolution layer, and the convolution kernel is required to be non-intersecting with its adjacent kernel. This is mainly to consider that more efforts should be put on finding the relationship between signal vectors rather than the relationship between elements in the vector.

[0060] Step4: After Step3, the corresponding results of each signal sequence are respectively subjected to an activation layer to obtain the corresponding output vector, and the Relu function is preferably used. For example Figure 2 The function image represented by the XY coordinate axis in the above formula is the activation function, which represents the activation layer.

[0061] Step5: The output vectors of all signal sequences obtained in Step4 are transposed and connected, and then subjected to a fully connected layer.

[0062] Step6: Finally, the phases of each LED are output, and the phase difference is calculated through the phase.

[0063] c). Distance measurement according to phase difference

[0064] The distance and the phase difference satisfy the following relationship, and the distance between each lamp and the PD can be obtained by the following formula :

[0065]

[0066] In the formula, c represents the speed of light, φ1,2 represents the phase difference between LED1 and LED2, φ2,1 represents the phase difference between LED2 and LED1. Similarly, φ1,3 represents the phase difference between LED1 and LED3, φ1,4 represents the phase difference between LED1 and LED4, φ1,5 represents the phase difference between LED1 and LED5. 、 、 、 、 , These represent the distances from LED1, LED2, LED3, LED4, and LED5 to the PD, respectively.

[0067] d) Particle swarm position update process

[0068] A particle swarm optimization (PSO) position update method based on motion state is used to improve positioning accuracy and robustness, resulting in smoother trajectories. The detailed process of PSO position update is described below; the distance obtained in step c is used to update the particle weights in equation (3):

[0069] First, the initial localization results of the PD are obtained using the least squares method. :

[0070]

[0071] In the formula, the matrix ,matrix ,variable ,in For the first i The coordinates of each LED ( i =2,3,4,5). For the location of PD, m The number of LEDs ( m =5).

[0072] The position updater parameters are set according to the motion state, which is divided into moving and stationary states. Then, the positions and velocities of the particles and their weights are initialized, and the weights are normalized after calculation. The following formula is the first update after the initial localization.

[0073]

[0074] In the formula , and Represents the k-th (k=1) positioning time. l The coordinates, velocity, and weight of each particle. and These represent random errors in coordinates and velocity, respectively. It is a weighting function.

[0075] The next step is particle update, and the update equation is as follows:

[0076]

[0077] In the formula , and Representing the k +1st positioning time l The coordinates, velocity, and weight of each particle. are the scaling factors for position and velocity, respectively, which are parameters adjusted according to the motion state. and are the random errors for position and velocity, respectively. is the time interval between two updates.

[0078] Next, resampling is performed, which mainly aims to delete some particles with low weights and add some random particles. Then, updating and normalization are performed again. Finally, the kth PD positioning result is obtained. k kth PD positioning result which can be obtained by the following formula.

[0079]

[0080] wherein, are the x-axis and y-axis coordinates of all particles at the kth update, is the mean function.

[0081] The initial positioning is obtained by least squares calculation, and the x-axis and y-axis coordinates of all particles at the 1st, 2nd, 3rd, … update are obtained in the above manner (when k = 1, according to formula (3), and otherwise according to formula (4)).

[0082] In the implementation, the method proposed in the technical scheme of the present application can be realized by computer software technology to automatically run the process, and the system device of the method, such as a computer readable storage medium storing the corresponding computer program of the technical scheme of the present application and a computer device including the running of the corresponding computer program, should also be within the protection scope of the present application.

[0083] In some possible embodiments, a visible light positioning system based on deep learning and particle swarm is provided, which includes a processor and a memory, the memory is used to store program instructions, and the processor is used to call the stored instructions in the memory to execute a visible light positioning method based on deep learning and particle swarm as described above.

[0084] In some possible embodiments, a visible light positioning system based on deep learning and particle swarm is provided, which includes a readable storage medium, and the readable storage medium stores a computer program, and the computer program is executed to realize a visible light positioning method based on deep learning and particle swarm as described above.

[0085] The specific embodiments described herein are merely illustrative of the spirit of the present application. Those skilled in the art of the present application can make various modifications or supplements to the described specific embodiments or replace them with similar ways, without deviating from the spirit of the present application or exceeding the scope defined by the appended claims.

Claims

1. A visible light positioning method based on deep learning and particle swarm, characterized in that: The method comprises the following processes, The signal modulation emission and the signal receiving and preprocessing process comprises modulating each lamp in the basic positioning unit into a different frequency signal, the photodiode PD receives the light signal, each lamp signal is separated by using a band-pass filter, the frequency of all lamp signals is unified into the frequency of the reference lamp by passing through a frequency down converter and a band-pass filter, the modulation frequencies of two lamps are exchanged, the same operation is performed, and the obtained signal sequence is input into a subsequent phase difference estimation network; The phase difference estimation network process comprises extracting a key sequence from each signal sequence in the phase difference estimation network, the extraction result is subjected to a convolution layer and an activation layer to obtain a corresponding output vector, the output vectors of all signal sequences are transposed and connected, and then subjected to a fully connected layer to output the phases of the lamps, and the phase difference is calculated through the phases; The distance is measured according to the phase difference, and the distance between each lamp and the photodiode PD is obtained according to the relationship between the distance and the phase difference; The particle swarm position updating method based on the motion state is adopted to make the trajectory more smooth. A basic positioning unit includes 5 LED lights, respectively denoted as LED1, LED2, LED3, LED4, LED5, and the frequencies of the respective lights are respectively denoted as f 1 ，f 2 ，f 3 ，f 4 ，f 5, LED1 is set as a reference light, and LED1 and LED2 are exchange modulation frequency lights; The implementation mode of the signal modulation emission and the signal receiving and preprocessing process comprises the following steps, Step 1, assume each lamp modulates a different frequency signal, where f 2 ，f 3 ，f 4 ，f 5 is f 1 odd multiple, then transmit; Step 2, the PD receives the light signal, and the received signal is a combined signal of multiple lamps; Step 3, each lamp signal is separated by using a band-pass filter; Step 4, each separated lamp signal is passed through a frequency down converter and a band-pass filter, the frequency of all lamp signals is unified into the frequency of the reference lamp, and a signal sequence is output; Step 5, the modulation frequencies of LED1 and LED2 are exchanged, and the steps of Step 2, 3 and 4 are repeated to output the signal sequence of the two lamps; Step 6, the signal sequences output in Step 4 and Step 5 are input into the phase difference estimation network to estimate the phase difference. 2.The visible light positioning method based on deep learning and particle swarm according to claim 1, wherein: In Step 1, set f 2 =3f 1 ， f 3 =5f 1 ， f 4 =7f 1 ， f 5 =9f 1. 3.The method of claim 2, wherein: The relationship between the distance and the phase difference is as follows, wherein denotes the speed of light, denotes the phase difference between LED1 and LED2, denotes the phase difference between LED2 and LED1, denotes the phase difference between LED1 and LED3, denotes the phase difference between LED1 and LED4, denotes the phase difference between LED1 and LED5, , , , , denote the distance from LED1, LED2, LED3, LED4, LED5 to PD, respectively. 4.The method of claim 1 or 2 or 3, wherein: In the particle swarm position updating method based on the motion state, firstly, the least square method is used to obtain an initial positioning result of the photodiode PD; secondly, a weight value is constructed by using a distance observation, and the particle swarm position is updated in combination with the motion state of the terminal; Finally, the terminal position is updated by using the particle swarm position updating result.

5. A visible light positioning system based on deep learning and particle swarm, characterized in that: The method comprises a processor and a memory, the memory is used for storing program instructions, and the processor is used for calling the storage instructions in the memory to execute the visible light positioning method based on deep learning and particle swarm.

6. A readable storage medium characterized by: The computer program is stored on the readable storage medium, and the computer program is executed to implement the visible light positioning method based on deep learning and particle swarm.

Images