Multi-source information fusion indoor navigation method based on HMM (Hidden Markov Model)

Abstract

The invention relates to the technical field of indoor positioning and navigation, in particular to an HMM (Hidden Markov Model)-based multi-source information fusion indoor navigation method, which is characterized in that data of different sensors are fused by using an HMM algorithm, specifically, states of pedestrians in moving are dispersed by using the HMM algorithm, and for each moving state of the pedestrians, each moving state of the pedestrians is subjected to multi-source information fusion; pedestrian pose information generated by the IMU sensor and the UWB device can be effectively fused; according to the method, only data of an IMU accelerometer and a gyroscope at adjacent moments are utilized, displacement and rotation angles calculated by the two sensors are corrected by utilizing discrete state position data of an HMM algorithm, and position data measured by a UWB device are fused, so that the positioning precision of the algorithm is improved; the method has the advantages that the positioning accuracy is improved, errors caused by signal shielding are reduced, positioning jitter and drift are reduced by fusing multi-source data, high precision is guaranteed, the Viterbi algorithm is used for path calculation, and the calculation efficiency of the algorithm is optimized.

Description

Technical Field

[0001] The present invention relates to the field of indoor positioning and navigation technology, and in particular to an HMM-based multi-source information fusion indoor navigation method. Background Art

[0002] In large indoor environments, such as hospitals, shopping malls, and factories, high-precision, low-latency navigation is required. Indoor positioning and navigation help locate and track people and equipment. In indoor environments, due to the low accuracy of GPS-based positioning methods, other sensor data is needed for indoor positioning and navigation. However, a single sensor or measuring device is often affected by its own or environmental factors. For example, the noise and accumulated errors of the inertial measurement unit (IMU) sensor, when running for a long time, the drift problem seriously affects the accuracy of positioning. Ultra-wideband (UWB) devices have high positioning accuracy and can achieve centimeter-level positioning accuracy. They are mainly positioned by measuring the arrival time of the signal. However, they are limited by the layout of the equipment, the coverage is limited, and there is still a certain degree of accuracy fluctuation in complex indoor environments. Therefore, in the field of indoor positioning and navigation, the data of multiple sensors or measuring devices are often required to be fused to ensure more accurate indoor positioning and navigation.

[0003] Under the existing technical background, the positioning technology based on IMU sensors and UWB devices has complementary advantages and disadvantages in indoor navigation. The advantage of IMU sensors is that they can provide high-frequency motion information and are not affected by complex indoor environmental factors. The disadvantage is that due to the need for iterative algorithms, errors continue to accumulate, and the long-term positioning effect is not good; UWB devices have the disadvantage of low update rate, but more accurate positioning, and the positioning accuracy can reach the centimeter level. Therefore, by fusing the data of IMU sensors and UWB devices, the high-precision characteristics of UWB devices and the advantages of IMU that they can provide short-term, high-frequency, and high-precision motion information can be utilized to achieve more stable and accurate indoor navigation. In order to make up for the shortcomings of a single technology, many studies have fused the data of UWB and IMU. However, how to effectively fuse multiple data sources, suppress noise, and reduce errors remains a technical challenge. Summary of the invention

[0004] The purpose of the present invention is to provide an indoor navigation method based on HMM multi-source information fusion, which uses a hidden Markov model to fuse the information source data obtained by IMU sensors and UWB devices for indoor positioning and navigation, and solves the problem that vulnerable groups or robots cannot autonomously identify directions and autonomously and passively locate indoors.

[0005] To achieve the above object, the present invention provides a multi-source information fusion indoor navigation method based on HMM, comprising the following steps:

[0006] Step 1: Grid the indoor map, establish the map coordinate system, generate the coordinates of the center point of the map grid, and mark the coordinates of the walls and obstacles;

[0007] Step 2: Synchronize timestamps and collect data through IMU sensors and UWB devices;

[0008] Step 3: Read the map data, the collected IMU sensor measurement data, and the UWB device measurement data, and perform missing value processing and preliminary calculation operations;

[0009] Step 4: Fuse the map data, IMU sensor measurement data, and UWB device measurement data through the HMM algorithm;

[0010] Step 5: Use the Viterbi algorithm to solve the fused data and output the optimal path.

[0011] Optionally, the execution process of step 1 is specifically to obtain a two-dimensional plane map of the indoor environment, perform a gridding operation, take the average stride length of a person as 0.7m as the side length of a grid, and generate a gridded indoor map; use the edge intersection point in the lower left corner of the map as the origin of the coordinate system, establish a rectangular coordinate system, and generate the coordinates of each grid center point in this rectangular coordinate system; and for the walls and obstacles in the room, select their coordinates and mark them.

[0012] Optionally, during the execution of step 2, data measurement is started after the IMU sensor and the UWB device are timestamped and calibrated; for the IMU sensor, angular velocity data and acceleration data of the x-axis and y-axis need to be measured; for the UWB device, the TODA method is used to generate the coordinates of the positioning point in the map rectangular coordinate system.

[0013] Optionally, the execution process of step 3 includes the following steps:

[0014] Step 3.1: Read the map grid coordinate data, the angular velocity data measured by the IMU sensor, the acceleration data measured by the two-axis sensor, and the UWB device to generate the coordinate data of the positioning point in the map rectangular coordinate system using the TODA method;

[0015] Step 3.2: For missing values ​​of measured data, use the front and back value filling method to fill them;

[0016] Step 3.3: For preliminary calculations, the real-time acceleration of the walking person is calculated using the acceleration measured by the two-axis sensor, the angular velocity between two moments is measured by the gyroscope of the IMU sensor and the angle data of the corresponding positions at the two moments is calculated and generated, and the coordinate data of the positioning point in the map rectangular coordinate system is generated by the UWB device using the TODA method to calculate the real-time distance walked by the walking person.

[0017] Optionally, the fusion process of step 4 includes the following steps:

[0018] Step 4.1: Calculate the single state generation probability, that is, the probability of the observation value being generated under the specified state;

[0019] Step 4.2: Calculate the state transition probability, and describe the probability of transition between states corresponding to two adjacent timestamps;

[0020] Step 4.3: Calculate the probability that the state transition mode is affected by the difference between the displacement at two moments and the distance between two states measured by the IMU;

[0021] Step 4.4: Calculate the probability that the state transition mode is affected by the difference between the angles at two moments measured by UWB and the angles between two states.

[0022] Step 4.5: Calculate the probability that the state transition is affected by the difference between the angles measured by the IMU at two moments and the angles between the two states.

[0023] Optionally, the execution process of step 5 includes the following steps:

[0024] Step 5.1: Calculate the initial path probability;

[0025] Step 5.2: Optimal path calculation based on Viterbi algorithm.

[0026] The present invention provides an indoor navigation method based on HMM multi-source information fusion, which uses a hidden Markov model (HMM) algorithm to fuse data from different sensors. Specifically, the HMM algorithm is used to discretize the states of pedestrians in motion. For each movement state of the pedestrian, the pedestrian posture information respectively generated by an IMU sensor and a UWB device can be effectively fused. In the present invention, only the data of an IMU accelerometer and a gyroscope at adjacent moments are used, and the displacement and rotation angle calculated by the two sensors are corrected using the discrete state position data of the HMM algorithm, and the position data measured by the UWB device are fused to improve the positioning accuracy of the algorithm; and the error caused by signal shielding is reduced, and positioning jitter and drift are reduced by fusing multi-source data. In addition, while ensuring high accuracy, the present invention uses a Viterbi algorithm to perform path solution, optimizes the calculation efficiency of the algorithm, is suitable for devices with limited resources, such as mobile terminals and Internet of Things devices, and has strong adaptability in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0028] Figure 1 It is a detailed step flow chart of a multi-source information fusion indoor navigation method based on HMM of the present invention.

[0029] Figure 2 It is a schematic diagram of marking the coordinates of walls and obstacles in a specific embodiment of the present invention.

[0030] Figure 3 It is a flowchart of an algorithm for determining whether there is a wall-penetrating state transition in a specific embodiment of the present invention. DETAILED DESCRIPTION

[0031] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0032] The present invention provides a multi-source information fusion indoor navigation method based on HMM, comprising the following steps:

[0033] S1: Grid the indoor map, establish the map coordinate system, generate the coordinates of the center point of the map grid, and mark the coordinates of the walls and obstacles;

[0034] S2: synchronize timestamps and collect data through IMU sensors and UWB devices;

[0035] S3: Read the map data, the collected IMU sensor measurement data, and the UWB device measurement data, and perform missing value processing and preliminary calculation operations;

[0036] S4: Fusing the map data, IMU sensor measurement data, and UWB device measurement data through the HMM algorithm;

[0037] S5: Use the Viterbi algorithm to solve the fused data and output the optimal path.

[0038] The following is a further explanation combined with the implementation steps:

[0039] In step S1, the indoor map is gridded to generate the coordinates of the center point of the map grid, and the coordinates of obstacles such as walls are marked, including:

[0040] Get a 2D map of the indoor environment, perform a gridding operation on it, take the average stride length of a person, 0.7m, as the side length of a grid, and generate a gridded indoor map. Take the edge intersection point in the lower left corner of the map as the origin of the coordinate system, establish a rectangular coordinate system, and generate the coordinates of each grid center point in this rectangular coordinate system. In addition, for obstacles such as walls in the room, select their coordinates and mark them.

[0041] In step S2, data collection by using the IMU sensor and the UWB device includes:

[0042] After the IMU sensor and UWB device are time-stamped and calibrated, data measurement begins. For the IMU sensor, the angular velocity data and the acceleration data of the x-axis and y-axis need to be measured. For the UWB device, the TODA method is used to generate the coordinates of the positioning point in the map rectangular coordinate system.

[0043] In step S3, the map data, the collected IMU sensor measurement data, and the UWB device measurement data are read, and missing value processing and preliminary calculation operations are performed, including:

[0044] The map grid coordinate data, the angular velocity data measured by the IMU sensor, the acceleration data measured by the two-axis sensor, and the UWB device are read using the TODA method to generate the coordinate data of the positioning point in the map rectangular coordinate system. For the missing values ​​of the measured data, the front and back value filling method is used to complete them. For the preliminary calculation, the real-time acceleration of the walking person is calculated by the acceleration measured by the two-axis sensor, the angular velocity between two moments is measured by the gyroscope of the IMU sensor, and the angle data of the corresponding positions at the two moments is calculated and generated. The coordinate data of the positioning point in the map rectangular coordinate system is generated by the UWB device using the TODA method to calculate the real-time distance walked by the walking person.

[0045] The fusing of the map data, the IMU sensor measurement data and the UWB device measurement data by the HMM algorithm in step S4 includes:

[0046] (1) Calculate the probability of single state generation

[0047] First, the single-state generation probability, also known as the emission probability, is calculated, which refers to the probability of the observation value being generated under a specified state. In this algorithm, a circle is drawn with the coordinates of the UWB positioning point as the center and twice the grid distance as the radius. All grids containing the center point in the circle are specified as the state at time t, and the generation probability of the observation value of the UWB positioning point at time t is calculated. This probability is considered to conform to the normal distribution and is related to the distance from the UWB positioning point to the center point of the state grid, that is:

[0048]

[0049] Among them, S t (x s ,y s ) is a state sequence, including the coordinates of the center point of each state, U t (x u ,y u ) is the observation sequence, which contains the coordinates of each UWB positioning point. is the standard deviation of UWB measurement, l u It is the Euclidean distance from the observation value coordinate to the state value coordinate measured by UWB, that is:

[0050]

[0051] (2) Calculate state transition probability

[0052] The state transition probability of HMM describes the probability of transition between states corresponding to two adjacent timestamps. Assuming that there are a certain number of states at time t, then there are also a certain number of states at time t-1. Assuming that there are i states at time t-1 and j states at time t, then there are i*j ways of state transition from time t-1 to time t. Correspondingly, there is a probability for each state transition method, which is the state transition probability. Therefore, for time t-1 to time t, there is an i*j state transition probability matrix. For the probability of each state transition method, it is assumed to obey the normal distribution and is related to the change in the yaw angle between the two moments measured by the IMU and the difference in the angle between the two states. Since the IMU calculates the yaw angle by measuring the angular velocity with a gyroscope, the yaw angle between the two moments is the angular velocity measured at time t-1 multiplied by the sampling time, that is:

[0053]

[0054] in, is the observed value of IMU angular velocity at time t-1, S t (x t ,y t ) is the current state, S t-1 (x t-1 ,y t-1 ) is the previous state, which contains the position coordinates of the two states. is the standard deviation of the IMU measured angle, S θ For S t , S t-1 The angle between the two states is calculated as:

[0055]

[0056] In addition, it is believed that the probability of each state transition mode is also related to the difference between the displacement at two moments measured by the IMU and the distance between the two states. Since the IMU calculates the walking distance by measuring the acceleration of three axes, and the algorithm only needs to calculate the distance of the two-dimensional plane, only the acceleration data measured by two axes can be used. The acceleration data of these two axes are vector-added and modulo-calculated to obtain the real-time acceleration of the experimenter's actual walking. Multiplying this actual acceleration by the square of the sampling time can obtain the distance measured by the IMU. The formula is:

[0057]

[0058] Among them, a I is the walking acceleration of the experimenter measured by IMU, a x 、a yis the acceleration measured by the x-axis and y-axis of the IMU. According to the velocity formula, the moving distance of the sampling time measured by the IMU can be obtained:

[0059] I l =a I ·T 2 (7)

[0060] Among them, I l is the moving distance at the sampling time measured by IMU. Finally, the state transition probability related to the distance measured by IMU and the distance difference between the states can be obtained. The formula is

[0061]

[0062] in, is the observed value of IMU acceleration at time t-1, including a x 、a y , that is, the acceleration measured by the two axes of the IMU, σ l is the standard deviation of the IMU measured distance, S l is the Euclidean distance between two states, and the calculation formula is,

[0063]

[0064] Finally, it is also believed that the probability of each state transition mode is related to the difference between the angle measured by UWB at two moments and the angle between the two states. By performing an inverse tangent calculation on the coordinates of the two positions measured by UWB, the angle between the two positions can be obtained. Similarly, performing an inverse tangent calculation on the coordinates of the two corresponding states can also obtain the angle between the two states.

[0065]

[0066] in, is the angle between the observations obtained by UWB measurement, is the standard deviation of the UWB measurement angle, S θ For S t , S t-1 The angle between two states.

[0067] Step S5 uses the Viterbi algorithm to solve the fused data and output the optimal path including:

[0068] (1) Initialize the initial path probability

[0069] In addition to the single-state generation probability matrix and the state transition probability matrix, the parameters of HMM also include the initial path probability. The initial path probability refers to the path probability at the initial moment. The role of the path probability is to calculate the time t in the Viterbi algorithm and calculate the probability of the most likely state by obtaining the maximum path probability. Since the most likely path probability is calculated recursively, it is only necessary to calculate the path probability at the moment t=1, that is, the initial moment, which is the initial path probability. The most likely path probability at subsequent moments can be continuously calculated through the initial path probability. When initializing the initial path probability, the initial probability of the state grid is needed. It is set that each state grid may be the initial state, so the initial probabilities of the state grids are all set to 1, that is

[0070] π(P s |t=1)=1 (12)

[0071] Since there is no state transition in the initial path probability, the initial path probability is considered to be equal to the product of the initial probability and the single state generation probability, that is,

[0072] λ1=π(P s |t=1)·f(S1|U1) (13)

[0073] (2) Optimal path calculation based on Viterbi algorithm

[0074] The initial path probability has been mentioned before, that is, the path probability at time t = 1. Then, the path probability at time t = 2 is the path probability at time t = 1 multiplied by the state transition probability from time t = 1 to time t = 2 multiplied by the single state generation probability at time t = 2. Since the state transition probability is determined by three parameters, the product of these three state transition probabilities is used to obtain the final state transition probability from time t = 1 to time t = 2, and the above most likely path probability calculation operation at time t = 2 is performed, that is:

[0075]

[0076] Among them, λ2 is the path probability corresponding to all states at time t=2. After obtaining this probability, the most likely path probability at time t=2 with the largest value is used as the state probability at time t=2 to calculate the most likely path probability at the next moment, and the formula can be obtained:

[0077]

[0078] in, is the most likely state probability at time t = 2. The product of the path probability and state transition probability calculated at the previous moment and the single state generation probability of the current state is calculated iteratively until the last state. The calculation formula for the general most likely path probability can be obtained:

[0079]

[0080] After calculating the path probability of the last state, select the state corresponding to the maximum path probability as the last state of the output path, then backtrack to the previous state matrix, select the state corresponding to the maximum path probability as the second to last state of the output path, and keep backtracking until the maximum path probability λ is selected in the initial state matrix. s m The corresponding initial state is used as the initial state of the output path, and other states corresponding to the maximum path probability are selected as the states of the output path.

[0081] The detailed steps are as follows: Figure 1 shown.

[0082] Furthermore, the present invention is further described by specific embodiments, and the execution steps in the embodiments are similar to the execution process, which are as follows:

[0083] Step 1: Discretize the map grid

[0084] A two-dimensional plane map of the laboratory environment is obtained and gridded. The average step length of a person, 0.7 m, is taken as the side length of a grid. The edge intersection point at the lower right corner of the map is taken as the origin to establish a rectangular coordinate system. In this rectangular coordinate system, the coordinates of all grid center points can be obtained. The coordinate value sequence of these grid center points represents the state sequence in the HMM algorithm.

[0085] Step 2: Mark the wall and obstacle coordinates

[0086] like Figure 2 As shown in the figure, in the rectangular coordinate system of the map, the coordinates of the wall and the obstacle are marked to filter out the data that passes through the wall. When calculating the state transition probability, it is necessary to consider whether there is a grid where a wall or an obstacle is located between the two state grids. If there is a grid where a wall or an obstacle is located, the state transition probability should be 0. Secondly, the next state of the current state should not be a grid where a wall or an obstacle is located. Therefore, the single state generation probability of the grid where the wall and the obstacle are located should be 0.

[0087] Step 3: Determine whether there is a wall-penetrating state transition algorithm

[0088] As shown in the figure, first determine whether the current state grid is connected to the wall grid. If so, it means that it is possible to pass through the wall, and proceed to the next step. Determine the coordinates of the point that is half the distance between the previous state grid and the current grid, called the transfer midpoint coordinates, and the relationship with the wall grid coordinates. Since the grid is discrete and the human step length is limited, if the state transfer passes through the wall, such as Figure 3As shown in the figure, there are only two cases. The first case is that the transfer midpoint coordinates coincide with the coordinates of the wall grid midpoint, and the second case is that the transfer midpoint coordinates are the midpoint of the wall grid edge. Therefore, it is necessary to find the midpoint coordinates of all wall grids and the midpoint coordinates of the wall grid edge, and determine whether the transfer midpoint coordinates coincide with the coordinates of the wall grid midpoint, or whether the transfer midpoint coordinates are the midpoint of the wall grid edge. If so, it means that the state transfer passes through the wall. The state transfer probability needs to be set to 0 to avoid this situation.

[0089] Step 4: Calculate the probability of single state generation

[0090] Calculate the single state generation probability, also known as the emission probability, draw a circle with the UWB positioning point coordinates as the center and twice the grid distance as the radius, specify all grids containing the center point in the circle as the state at time t, and calculate the generation probability of the observation value of the UWB positioning point at time t. This probability is considered to conform to the normal distribution and is related to the distance from the UWB positioning point to the center point of the state grid.

[0091] Step 5: Calculate state transition probability

[0092] For the probability of each state transition mode, it is assumed to obey the normal distribution and is related to the difference between the yaw angle change at two moments measured by the IMU and the angle between the two states. Since the IMU calculates the yaw angle by measuring the angular velocity with a gyroscope, the yaw angle between the two moments is the angular velocity measured at time t-1 multiplied by the sampling time.

[0093] In addition, it is believed that the probability of each state transition mode also conforms to the normal distribution related to the difference between the displacement at two moments measured by the IMU and the distance between the two states. Since the IMU calculates the walking distance by measuring the acceleration of three axes, and the algorithm only needs to calculate the distance of the two-dimensional plane, the acceleration data measured by two axes is used, and the acceleration data of the two axes are vector-added and modulo-calculated to obtain the real-time acceleration of the experimenter's actual walking. Multiplying this actual acceleration by the square of the sampling time can obtain the distance measured by the IMU.

[0094] Finally, it is also believed that the probability of each state transition mode conforms to the normal distribution related to the difference between the angle measured by UWB at two moments and the angle between the two states. By performing inverse tangent calculation on the coordinates of the two positions measured by UWB, the angle between the two positions can be obtained. Similarly, by performing inverse tangent calculation on the coordinates of the two corresponding states, the angle between the two states can also be obtained.

[0095] According to the classic formula of normal distribution and the calculated data parameters, the characteristic functions of the three state transition probabilities can be obtained. Based on these characteristic functions, the state transition probability matrix of each state can be calculated.

[0096] Step 6: Calculate initial path probabilities

[0097] The initial path probability refers to the path probability at the initial moment. The role of the path probability is to calculate the probability of the most likely state at time t in the Viterbi algorithm by obtaining the maximum path probability. Since the most likely path probability is calculated recursively, it is only necessary to calculate the path probability at time t=1, that is, the initial moment, which is called the initial path probability. The most likely path probability at subsequent moments can be calculated continuously through the initial path probability. When initializing the initial path probability, the initial probability of the state grid is needed. It is set that each state grid may be the initial state, and the initial probability of the state grid is all set to 1.

[0098] Step 7: Optimal path calculation based on Viterbi algorithm

[0099] Through the initial path probability, we can get the path probability of the state at the next moment. The path probability of the state at the next moment is equal to the maximum path probability of the state at the initial moment multiplied by the state transition probability multiplied by the single state generation probability of the state at the next moment. Through calculation, we can get the path probability of each state corresponding to the next moment, select the largest path probability, and use it to calculate the subsequent path probability. In this way, we continue to iterate until the path probability of the state at the last moment.

[0100] Select the maximum path probability of the state at the last moment, and get the state corresponding to this probability. This state is the state of the output path at the last moment. From this state, trace back to the previous state set and select the state corresponding to the maximum path probability. In this way, iterate continuously until the initial state. The states corresponding to the selected series of maximum path probabilities are the states corresponding to the output path.

[0101] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0102] (1) The problem of IMU cumulative error is solved. By introducing the HMM method, the short-term high-frequency data of the IMU can be combined with the high-precision position information of the UWB. The HMM can infer the optimal motion trajectory through the state sequence and use the absolute position of the UWB to correct the cumulative error in the IMU data. In this way, the drift problem of IMU positioning due to the accumulation of sensor noise and error is reduced. Over time, it becomes more and more serious, resulting in large positioning errors after long-term operation.

[0103] (2) The signal blocking and multipath effect problems are solved. By fusing the relative motion data of the IMU with the absolute position data of the UWB, the HMM can smooth out the abnormal values ​​in the UWB ranging (such as signal blocking and multipath effect), thereby more accurately estimating the motion state and position of the device. In this way, the problem of wireless positioning technologies such as UWB being easily affected by obstacles such as walls and furniture in complex indoor environments, which easily causes signal blocking and multipath effects, increases the ranging error, and thus affects the positioning accuracy, is reduced.

[0104] (3) Solved the asynchrony and noise processing problems of multi-sensor data fusion. Since the data acquisition frequencies between sensors such as IMU and UWB are different and each is interfered by different noise sources, the data is inconsistent during direct fusion, making it difficult to accurately obtain the device position. By modeling the timing dependency and noise model between IMU and UWB, HMM can reasonably process the asynchronous data of the two sensors, automatically filter the noise data, and select the most likely state sequence.

[0105] What is disclosed above is only a preferred embodiment of the present invention, and it certainly cannot be used to limit the scope of rights of the present invention. Ordinary technicians in this field can understand that all or part of the processes of the above embodiment and equivalent changes made according to the claims of the present invention still fall within the scope of the invention.

Claims

1. A multi-source information fusion indoor navigation method based on HMM, characterized in that: The following steps are involved: Step 1: Grid the indoor map, establish the map coordinate system, generate the coordinates of the center point of the map grid, and mark the coordinates of the walls and obstacles; Step 2: Synchronize timestamps and collect data through IMU sensors and UWB devices; Step 3: Read the map data, the collected IMU sensor measurement data, and the UWB device measurement data, and perform missing value processing and preliminary calculation operations; Step 4: Fuse the map data, IMU sensor measurement data, and UWB device measurement data through the HMM algorithm; Step 5: Use the Viterbi algorithm to solve the fused data and output the optimal path.

2. The HMM-based multi-source information fusion indoor navigation method according to claim 1, characterized in that: The execution process of step 1 is specifically to obtain a two-dimensional plane map of the indoor environment, perform a gridding operation, take the average stride length of a person as 0.7m as the side length of a grid, and generate a gridded indoor map; use the edge intersection point in the lower left corner of the map as the origin of the coordinate system, establish a rectangular coordinate system, and generate the coordinates of each grid center point in this rectangular coordinate system; and for the walls and obstacles in the room, select their coordinates and mark them.

3. The HMM-based multi-source information fusion indoor navigation method as claimed in claim 2, characterized in that: During the execution of step 2, the IMU sensor and UWB device are timestamped and calibrated before data measurement begins; for the IMU sensor, the angular velocity data and the acceleration data of the x-axis and y-axis need to be measured; for the UWB device, the TODA method is used to generate the coordinates of the positioning point in the map rectangular coordinate system.

4. The HMM-based multi-source information fusion indoor navigation method as claimed in claim 3, characterized in that: The execution process of step 3 includes the following steps: Step 3.1: Read the map grid coordinate data, the angular velocity data measured by the IMU sensor, the acceleration data measured by the two-axis sensor, and the UWB device to generate the coordinate data of the positioning point in the map rectangular coordinate system using the TODA method; Step 3.2: For missing values ​​of measured data, use the front and back value filling method to fill them; Step 3.3: For preliminary calculations, the real-time acceleration of the walking person is calculated using the acceleration measured by the two-axis sensor, the angular velocity between two moments is measured by the gyroscope of the IMU sensor to calculate the angle data of the corresponding positions at the two moments, and the coordinate data of the positioning point in the rectangular coordinate system of the map is generated by the UWB device using the TODA method to calculate the real-time distance walked by the walking person.

5. The HMM-based multi-source information fusion indoor navigation method as claimed in claim 4, characterized in that: The fusion process of step 4 includes the following steps: Step 4.1: Calculate the single state generation probability, that is, the probability of the observation value being generated under the specified state; Step 4.2: Calculate the state transition probability, and describe the probability of transition between states corresponding to two adjacent timestamps; Step 4.3: Calculate the probability that the state transition mode is affected by the difference between the displacement at two moments and the distance between two states measured by the IMU; Step 4.4: Calculate the probability that the state transition mode is affected by the difference between the angles at two moments measured by UWB and the angles between the two states; Step 4.5: Calculate the probability that the state transition is affected by the difference between the angles measured by the IMU at two moments and the angles between the two states.

6. The HMM-based multi-source information fusion indoor navigation method according to claim 5, characterized in that: The execution process of step 5 includes the following steps: Step 5.1: Calculate the initial path probability; Step 5.2: Optimal path calculation based on Viterbi algorithm.

Images