An Indoor Personnel Localization Method Based on Particle Filtering and Map Matching

Abstract

This invention relates to an indoor personnel positioning method based on particle filtering and map matching. The method includes the following steps: acquiring RSSI data from multiple Bluetooth beacons received at the point to be located; obtaining a preliminary positioning result based on the weighted centroid method; preliminarily correcting the preliminary positioning result using an improved particle filtering algorithm, wherein the improved particle filtering algorithm is implemented based on historical positioning data and the walking speed of people indoors; and further correcting the intermediate positioning result using indoor map matching technology based on Hidden Markov Models (HMM) to rationalize the personnel trajectory and obtain the final positioning result. Compared with existing technologies, this invention has advantages such as stability and high accuracy.

Description

Technical Field

[0001] This invention relates to the field of indoor positioning technology, and in particular to an indoor personnel positioning method based on particle filtering and map matching. Background Technology

[0002] The interior structures of modern buildings are becoming increasingly complex, leading to a surge in market demand for location services in indoor settings. In indoor environments, satellite signals are often blocked, rendering global navigation and positioning systems unusable. Therefore, indoor positioning technology is becoming a hot topic in both academic research and market applications. Indoor positioning technologies are generally based on computer vision, wireless communication, LED visible light technology, or geomagnetic matching. Wireless communication technologies include Wi-Fi positioning, ZigBee positioning, UWB positioning, RFID positioning, and Bluetooth Low Energy positioning, with Bluetooth Low Energy positioning technology and its industry experiencing particularly rapid growth.

[0003] Currently, Bluetooth Low Energy positioning technology can be divided into RSSI-based location fingerprinting technology and RSSI-based ranging technology.

[0004] like Figure 1 As shown, fingerprint-based localization relies on the differences in RSSI at different locations to establish a unique mapping relationship between RSSI and spatial location, and then uses a matching algorithm for localization. Building a fingerprint database requires collecting a large amount of RSSI data and continuously updating the database, resulting in a huge workload. Because the fingerprint database is closely integrated with the environment, the information acquired during collection is difficult to transfer to similar environments, leading to a large amount of repetitive work.

[0005] RSSI-based positioning methods utilize the distance-attenuation characteristic of RSSI to measure the distance from the target terminal to the wireless access point (AP). This method offers advantages such as strong practicality, scalability, and simple deployment and maintenance. The principle involves the target terminal calculating the distance to a pre-deployed Bluetooth beacon using a path loss model based on RSSI. Then, using the Bluetooth beacon as the centroid and the reciprocal of the distance from the target terminal to the beacon as the weight, a weighted centroid is calculated and used as the positioning result. However, in complex indoor environments, the inaccuracy of the path loss model and uncertain external interference can lead to less than ideal positioning results using the Bluetooth RSSI-based weighted centroid method. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the existing technology, such as the large amount of positioning work, and to provide a stable and highly accurate indoor personnel positioning method based on particle filtering and map matching.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] An indoor people localization method based on particle filtering and map matching, the method includes the following steps:

[0009] The RSSI data of multiple Bluetooth beacons received by the point to be located are acquired, and a preliminary positioning result is obtained based on the weighted centroid method.

[0010] The preliminary positioning result is initially corrected by an improved particle filter algorithm to obtain an intermediate corrected positioning result. The improved particle filter algorithm is based on historical positioning data and the walking speed of people indoors.

[0011] The indoor map matching technology based on HMM is used to further correct the intermediate positioning results, rationalize the personnel trajectory, and obtain the final positioning result.

[0012] Furthermore, the specific steps for obtaining the preliminary positioning results are as follows:

[0013] The received RSSI data is converted into distance data using a path loss model. Based on the distance data and the coordinates of each Bluetooth beacon, a weighted method is used to calculate the preliminary positioning result.

[0014] Furthermore, the weighting method is expressed by the following formula:

[0015]

[0016] Where x and y are the preliminary positioning results, representing the x and y coordinates of the point to be located, respectively. i y i The x and y coordinates of the i-th Bluetooth beacon are represented, N represents the number of Bluetooth beacons that received RSSI, and d i This represents the distance between the point to be located and the i-th Bluetooth beacon.

[0017] Furthermore, the path loss model is a logarithmic path loss model.

[0018] Furthermore, the preliminary correction of the initial positioning result using the improved particle filter algorithm specifically includes the following steps:

[0019] An initial particle population is obtained based on the preliminary positioning results;

[0020] The state of each particle is updated based on system state noise and changes in the walking speed of personnel.

[0021] Update particle weights and, in conjunction with the indoor map, reset the weights of particles located in inaccessible areas to zero.

[0022] Intermediate correction positioning results are obtained through iterative calculation.

[0023] Furthermore, the initial particle population is generated using a Gaussian distribution, the mean of which is the result of preliminary positioning, and the variance is the error of the weighted centroid method.

[0024] Furthermore, the system state noise and the changes in personnel walking speed conform to a normal distribution.

[0025] Furthermore, the HMM-based indoor map matching technology further refines the intermediate corrected positioning results as follows:

[0026] Feasible paths are obtained by combining indoor map abstraction and represented by grid points;

[0027] In the HMM decoding algorithm, the intermediate corrected positioning result is used as the observation value, the grid point is used as the hidden state, and the positioning result is corrected using a map matching method.

[0028] Furthermore, in the HMM decoding algorithm, a dynamic observation sequence and state sequence of length L are maintained.

[0029] Furthermore, the Bluetooth beacon is a Bluetooth Low Energy node.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] 1. This invention features layered processing, including weighted centroid localization, improved particle filter correction, and HMM-based indoor map matching technology correction. This effectively combines the advantages of various algorithms while also leaving room for future algorithm upgrades.

[0032] 2. This invention utilizes particle filtering to fuse historical positioning trajectories and indoor personnel activity patterns with weighted centroid positioning, which can narrow down the possible locations of personnel and obtain stable and highly accurate positioning results.

[0033] 3. This invention focuses on indoor personnel positioning, and improves the positioning effect by introducing prior knowledge of indoor maps and indoor personnel walking speed.

[0034] 4. During the positioning system deployment phase, the method proposed in this invention only requires the deployment of low-power Bluetooth nodes in the area to be located. After deployment, there is no additional signal acquisition work, which can save a lot of manpower costs and facilitate the rapid launch of the system.

[0035] 5. Real-world testing shows that this method has high accuracy and robustness, with performance improvements of 59% and 43% respectively compared to weighted centroid positioning and Kalman filter-based Bluetooth indoor positioning methods. Attached Figure Description

[0036] Figure 1 A flowchart for existing fingerprint localization;

[0037] Figure 2 This is a schematic diagram of the structure of the present invention;

[0038] Figure 3 This is a schematic diagram of Bluetooth node deployment in the embodiment. Detailed Implementation

[0039] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0040] like Figure 2 As shown, this embodiment provides an indoor personnel positioning method based on particle filtering and map matching. This method achieves accurate indoor positioning and obtains the location of personnel through three progressively layered steps: a weighted positioning layer, a particle filter correction layer, and an indoor map fusion layer. The method includes the following steps:

[0041] The RSSI data of multiple Bluetooth beacons received by the point to be located are acquired, and a preliminary positioning result is obtained based on the weighted centroid method.

[0042] The preliminary positioning result is initially corrected by an improved particle filtering algorithm to obtain an intermediate corrected positioning result. The improved particle filtering algorithm is based on historical positioning data and indoor walking speed. It combines particle positions with indoor maps, resets the weights of particles with unreasonable positions to zero, and generates new particles through resampling to make the positioning trajectory smoother.

[0043] The intermediate positioning results are further refined using indoor map matching technology based on HMM (Hidden Markov Model) to rationalize personnel trajectories and obtain the final positioning result. Specifically, for personnel positioning in indoor scenes, the characteristics of slow walking speed and subjective initiative of people are utilized. In conjunction with the indoor map, regular building structures such as corridors are abstracted into feasible paths, and grid points are used to represent feasible paths. Then, by combining map information, the positioning results are refined using map matching methods to rationalize personnel trajectories. The results after particle filtering are used as observation values, and the grid points representing feasible paths are used as hidden states. At the same time, in order to solve the problem of the computational cost of HMM increasing over time, this method uses a fixed-length observation sequence and state sequence, and the decoding algorithm only backtracks a fixed number of times.

[0044] The specific processing steps of each layer of the above method are explained below.

[0045] (1) Weighted centroid positioning

[0046] Wireless signal strength decreases with increasing propagation distance. There is a mapping relationship between signal strength and distance, which can be described by the logarithmic path loss model, as shown in equation (1):

[0047]

[0048] In the weighted centroid localization method, to simplify equation (1), we can assume that the noise power is 0 dBm, i.e., X δ Always equal to 0, when the reference distance d0 is 1m, it can be expressed as

[0049]

[0050] The weighted centroid positioning method treats Bluetooth beacons as unit mass points and considers the influence of the distance between the Bluetooth beacons and the target terminal. Different weights are assigned to each beacon based on the distance from the target terminal to the Bluetooth beacon. The specific method is as follows: First, the RSSI received by the target terminal is converted into a distance d using a path loss model. The influence of the beacon on the target terminal is inversely proportional to the distance, i.e., the reciprocal of the distance is used as the weight of the Bluetooth beacon. Finally, the weighted average method is used to solve for the coordinates of the target terminal. The weighted centroid positioning method can be described by equation (3).

[0051]

[0052] Where x and y are the horizontal and vertical coordinates of the terminal to be located, respectively. i y i Let d represent the x and y coordinates of the i-th beacon, N represent the number of RSSI beacons received by the terminal to be located, and d i The distance between the terminal to be located and the i-th beacon can be obtained from the path loss model.

[0053] (2) Improved particle filter correction

[0054] In weighted positioning based on BLE (Bluetooth Low Energy) technology, the path loss model is affected by the complex indoor environment, leading to inaccurate, unstable, and jump-like positioning results. Although filtering the RSSI data itself can improve positioning accuracy, the effect is not significant. Indoor walking speed is typically 0–1.5 m / s and is a continuous process; at a positioning frequency of 5 Hz, the distance a person moves in 0.2 seconds is typically 0–0.3 m. Therefore, by using historical positioning data and indoor walking speed data to improve the particle filter algorithm, the possible locations of people can be narrowed down.

[0055] In practical applications, the particle filter algorithm mainly includes five steps: initialization, prediction sampling, importance sampling, resampling, and result calculation. The particle filter algorithm performs each step sequentially.

[0056] During the initialization phase, particles need to be generated using a probability distribution, and an initial system state needs to be assigned. To smooth the positioning results using historical positioning information, the state space of particle i at time k is:

[0057]

[0058] in Let x and y represent the x and y coordinates of the particle at time k, respectively. This represents the coordinates of the particle at time k-1. This represents the velocity of the particle at time k. The initial position of the particle is generated using a Gaussian distribution, with its mean obtained by weighted centroid localization and its variance representing the error of the weighted centroid localization algorithm. The initial weight ω0 of each particle is 1 / n.

[0059] During the prediction sampling phase, each particle needs to predict its state at time k based on its state at time k-1 using equation (5). This phase moves the particle from its old state to a new state, thus updating the particle's state. Let the time interval from time k-1 to time k be dt, and the particle at time k can be obtained using equation (5):

[0060]

[0061] Q x Q y Q represents the noise of the system state. v This indicates the change in the walking speed of people.

[0062] During the importance sampling phase, the particle weights generated in the prediction sampling phase need to be updated based on the observed values. Combined with the indoor map, the generated particles are rationally judged, and the weights of particles located in inaccessible areas are reset to zero, thus correcting their orientation. During indoor activities, the speed of people is typically between 0m and 1.5m, and the particle weights are adjusted using the factor expressed in equation (7). The updated weights... for

[0063]

[0064]

[0065]

[0066] Where, γ i for (zx k ,zy k) represents the observation coordinates at time k, given by the weighted centroid positioning method.

[0067] Based on the improved particle algorithm described above, intermediate correction positioning results are obtained through iterative calculation.

[0068] (3) Indoor map fusion based on HMM

[0069] In indoor environments, people's need for location services is often for navigation, and since people possess a certain degree of initiative, this navigation task can be completed without relying on high-precision positioning information. Reaching a destination often requires traversing numerous regular building structures such as corridors, passageways, and staircases; these structures can be analogized to roads, serving as feasible paths for people. This method utilizes an indoor map to pre-rasterize feasible paths for people indoors, and then corrects the positioning trajectory using Hidden Markov Model (HMM)-based indoor map matching technology. In personnel positioning, the current position can be considered as the hidden state S of the HMM. n Only compared to the hidden state S of the previous moment n-1 Relatedly, the indoor positioning trajectory tracking problem can be transformed into solving the hidden state S given the Hidden Mirror (HMM) parameters and the observation sequence V, where the observation sequence is obtained using a particle filter algorithm. To incorporate map information into the HMM, the hidden state S is represented by rasterized feasible paths. Each rasterized point a... i Let S represent the possible hidden states of an Hidden MM, S∈{a1,...,a...} n}

[0070] The state transition matrix is ​​A = {a ij}, a ij The person is represented by grid point a i Move to grid point a j The probability of this. In indoor environments, people's walking speed is typically between 0m and 1.5m, therefore, from grid point a... i Move to grid point a j That is, from state a i Move to a j The probability of a i a j The distance between them is related, and this method uses the following formula to describe it:

[0071]

[0072] Where (x) i ,y i ) represents grid point a i The coordinates, (x j ,y j ) represents grid point a j The coordinates of c are such that the transition probability satisfies the equation The constant.

[0073] To obtain the emission probability, the space is discretized and gridded, with each grid point g... i This represents an observation state. The localization results of the particle filter layer are compared with those of grid point g. i Sort by Euclidean distance, with the grid point numbered v as the smallest distance. i Instead, it is converted into discrete observations, as shown in equation (10).

[0074]

[0075] Among them (g) ix ,g iy (p) represents the coordinates of the i-th grid point. x ,p y Let ) represent the coordinates of a continuous observation p, and G be the total number of grid points. Therefore, the emission probability p(v) is... i |a j ), the j-th row and i-th column of the emission matrix B ji Represented as

[0076]

[0077] in Indicates the vth i The coordinates of each grid point, (a jx ,a jy ) represents a grid point a j The coordinates of c are such that A constant that holds true.

[0078] The initial state π of an Hidden Markov Model (HMM) can determine the model's accuracy to a certain extent. To ensure the initial state matches the actual localization situation and accelerate algorithm convergence, the initial state can be calculated after obtaining the initial localization result p0 in the particle filter layer.

[0079]

[0080] Where c is such that The constant that holds true, π n Indicates the initial state as a n The probability that a person is located at grid point a n The probability, initial state Π={π1,...,π n}

[0081] The goal of the HMM decoding algorithm is to obtain the system state, i.e., the hidden state of the HMM, from the observation sequence. Meanwhile, to reduce computational load and ensure real-time positioning, the improved decoding algorithm in this method maintains a dynamic observation sequence and state sequence of length L, instead of using the entire observation sequence.

[0082] Assuming that at time t=1, the observation sequence V={v0,v1}, according to equation (12), we can obtain

[0083]

[0084] but At this point, the state sequence is S = {s0, s1}.

[0085] When t ≤ L-1, the observation sequence V = {v0,...,v} t The state sequence is solved using equation (14).

[0086]

[0087] at this time State sequence S = {s0,...,s} t Equation (14) requires recursive solution, with a time complexity of O(n*L*L), and the computational complexity increases with time.

[0088] When L-1 < t, the number of observations has exceeded L. To reduce the computational load, this method takes the system state at time t-L+1 as the defined system state, and sets the observation sequence as V = {v...} t-L+1 ,...,v t This avoids excessively long time backtracking, i.e., at this point...

[0089]

[0090] in Represents the system state at time t-L+1 to state a j The probability of . At this time, the system state s. t =ψ t The state sequence S = {s} t-L+1 ,...,s t}

[0091] The above decoding algorithm does not backtrack all observation sequences. Instead, it maintains an observation sequence and a state sequence of length L. It uses the system state at time t-L+1 as the determined system state, rather than inferring the system state from the initial state, thus reducing the amount of computation while ensuring accuracy.

[0092] In summary, the specific steps of this method include:

[0093] Step 1: Deploy Bluetooth nodes, such as Figure 3 As shown, the scanning and transmission frequencies are set to 5Hz.

[0094] Step 2: According to equation (3), perform weighted positioning and output the preliminary positioning results.

[0095] Step 3: Connect the output of the weighted localization layer to the particle filter correction layer.

[0096] Step 4: Q-factor correction layer for particle filtering x Q y Q v The localization results of the particle filter correction layer are obtained by setting the distribution to a normal distribution with a mean of 0.2 and a variance of 1.

[0097] Step 5: Connect the output of the particle filter correction layer to the indoor map fusion layer using HMM.

[0098] Step 6: State transition matrix parameter σ of the indoor map fusion layer of HMM A =0.3, emission matrix parameter σ B =1, queue length L=60.

[0099] Step 7: The positioning results of the indoor map fusion layer are used as the final output.

[0100] If the above methods are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0101] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. An indoor personnel positioning method based on particle filtering and map matching, characterized in that, The method includes the following steps: The RSSI data of multiple Bluetooth beacons received by the point to be located are acquired, and a preliminary positioning result is obtained based on the weighted centroid method. The preliminary positioning result is initially corrected by an improved particle filter algorithm to obtain an intermediate corrected positioning result. The improved particle filter algorithm is based on historical positioning data and the walking speed of people indoors. The intermediate positioning results are further corrected using indoor map matching technology based on HMM to rationalize personnel trajectories and obtain the final positioning results. The preliminary correction of the initial positioning result using the improved particle filter algorithm specifically includes the following steps: An initial particle population is obtained based on the preliminary positioning results. The initial particle population is generated using a Gaussian distribution, the mean of which is the preliminary positioning results, and the variance is the error of the weighted centroid method. The state of each particle is updated based on system state noise and changes in the walking speed of personnel. Update particle weights and, in conjunction with the indoor map, reset the weights of particles located in inaccessible areas to zero. Iterative calculations yield intermediate corrected positioning results; In the prediction sampling phase, each particle predicts its state at time k based on its state at time k-1. Let the time interval from time k-1 to time k be dt. The particle at time k is obtained using the following formula: Q x Q y Q represents the noise of the system state. v Indicates the change in walking speed. Let x and y represent the x and y coordinates of the particle at time k, respectively. This represents the coordinates of the particle at time k-1. This represents the velocity of the particle at time k; During the importance sampling phase, the particle weights generated in the prediction sampling phase are updated based on the observed values. Combined with the indoor map, the generated particles are rationally assessed, and the weights of particles located in inaccessible areas are reset to zero, with orientation correction performed. The updated weights... for: in, (zx k ,zy k ) represents the observation coordinates at time k, given by the weighted centroid positioning method.

2. The indoor personnel positioning method based on particle filtering and map matching according to claim 1, characterized in that, The specific steps for obtaining the preliminary positioning results are as follows: The received RSSI data is converted into distance data using a path loss model. Based on the distance data and the coordinates of each Bluetooth beacon, a weighted method is used to calculate the preliminary positioning result.

3. The indoor personnel positioning method based on particle filtering and map matching according to claim 2, characterized in that, The weighting method is expressed by the following formula: Where x and y are the preliminary positioning results, representing the x and y coordinates of the point to be located, respectively. i y i The x and y coordinates of the i-th Bluetooth beacon are represented, N represents the number of Bluetooth beacons that received RSSI, and d i This represents the distance between the point to be located and the i-th Bluetooth beacon.

4. The indoor personnel positioning method based on particle filtering and map matching according to claim 2, characterized in that, The path loss model is a logarithmic path loss model.

5. The indoor personnel positioning method based on particle filtering and map matching according to claim 1, characterized in that, The system state noise and the changes in personnel walking speed conform to a normal distribution.

6. The indoor personnel positioning method based on particle filtering and map matching according to claim 1, characterized in that, The HMM-based indoor map matching technology further refines the intermediate corrected positioning results as follows: Feasible paths are obtained by combining indoor map abstraction and represented by grid points; In the HMM decoding algorithm, the intermediate corrected positioning result is used as the observation value, the grid point is used as the hidden state, and the positioning result is corrected using a map matching method.

7. The indoor personnel positioning method based on particle filtering and map matching according to claim 6, characterized in that, In the HMM decoding algorithm, a dynamic observation sequence and state sequence of length L are maintained.

8. The indoor personnel positioning method based on particle filtering and map matching according to claim 1, characterized in that, The Bluetooth beacon is a Bluetooth Low Energy node.

Images