A multi-source fusion positioning method and device in a large-scale indoor scene

Abstract

The present invention provides a multi-source fusion positioning method and device in a large-scale indoor scene, including controlling a pre-built inertial odometer to collect three-dimensional positioning information; using a dynamic time warping algorithm to detect different types of indoor landmark points; extracting the motion characteristics of pedestrian multi-source indoor trajectories as input vectors, and extracting the reference error value of each step of the pedestrian multi-source indoor trajectory as an output vector to construct a multi-layer perception network; using an anti-error Kalman filter to fuse the three-dimensional positioning information, landmark point position information and multi-source fingerprints, eliminating abnormal observations, and obtaining updated three-dimensional positioning information. The present invention comprehensively considers the important factors that affect the accuracy of multi-source fusion positioning in a large-scale indoor space, models and evaluates the multi-source trajectory, and eliminates interference from the external environment; a unified anti-error Kalman filter is designed to combine different positioning sources, thereby achieving effective estimation of the three-dimensional position of pedestrians in a large-scale indoor space.

Description

Technical Field

[0001] The present invention relates to the field of positioning technology, and in particular to a multi-source fusion positioning method and device in a large-scale indoor scene. Background Art

[0002] Location-based services play an indispensable role in the application of smart terminals, and how to accurately obtain continuous and reliable location information of users in a large range of urban indoor spaces has become the key. However, in complex urban indoor environments, pedestrian movements are complex and diverse, and there is a lack of unified positioning sources like outdoor satellite positioning in indoor spaces, resulting in high-precision indoor positioning still being one of the difficult problems in the field of pedestrian positioning.

[0003] Therefore, the prior art has defects and needs to be improved and developed. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a multi-source fusion positioning method and device in a large-scale indoor scene in view of the above-mentioned defects of the prior art, aiming to solve the problem of poor accuracy in indoor positioning in the prior art.

[0005] The technical solution adopted by the present invention to solve the technical problem is as follows:

[0006] A multi-source fusion positioning method in a large-scale indoor scene, comprising:

[0007] Control the pre-built inertial odometer to collect three-dimensional positioning information;

[0008] Use dynamic time warping algorithm to detect different types of indoor landmarks and obtain the location information of the landmarks;

[0009] Extracting motion features of the pedestrian multi-source indoor trajectory as an input vector, and extracting a reference error value of each step of the pedestrian multi-source indoor trajectory as an output vector, constructing a multi-layer perception network, and adaptively evaluating the error of the pedestrian multi-source indoor trajectory according to the multi-layer perception network to obtain a multi-source fingerprint;

[0010] The three-dimensional positioning information, the landmark point position information and the multi-source fingerprint are fused and processed by using an anti-differential Kalman filter to eliminate abnormal observation values ​​and obtain updated three-dimensional positioning information.

[0011] In one implementation, the step of constructing the inertial odometer includes:

[0012] The data processing model for constructing accelerometer and gyroscope measurements is:

[0013] Among them, the represents the modeled output of the accelerometer, the represents the modeling output of the gyroscope, the a k represents the ideal output of the accelerometer, the w k represents the ideal output of the gyroscope, the Indicates the accelerometer zero bias, Indicates the gyroscope zero bias, is the rotation matrix, the g w is the gravity vector, n a is the Gaussian white noise measured by the accelerometer, the n w is the Gaussian white noise measured by the gyroscope;

[0014] The accelerometer bias and the gyroscope bias are modeled as a process of bias random walk obeying Gaussian distribution, which is expressed as:

[0015] Among them, the represents the random walk value of the accelerometer zero bias, Represents the random walk value of the gyroscope bias, which follows a Gaussian distribution.

[0016] In one implementation, the controlling a pre-built inertial odometer to collect three-dimensional positioning information includes:

[0017] The acceleration value at the current moment is collected by using the modeled accelerometer, and the angular velocity value at the current moment is collected by using the modeled gyroscope;

[0018] Performing pre-integration processing on the acceleration value and the angular velocity value to obtain a pre-integration result;

[0019] The calculation formula of the pre-integration result is:

[0020]

[0021]

[0022]

[0023] Among them, the represents the three-dimensional position information obtained by pre-integration, represents the three-dimensional velocity information obtained by pre-integration, represents the three-dimensional posture information obtained by pre-integration, is a rotation matrix, Ω is an antisymmetric matrix, k represents a timestamp, dk represents solving the integral according to the time interval, Indicates the three-dimensional posture information of the previous moment;

[0024] The pre-integration result is corrected according to the accelerometer bias and the gyroscope bias to calculate and obtain three-dimensional positioning information, wherein the three-dimensional positioning information includes: three-dimensional position information, three-dimensional velocity information and three-dimensional attitude information;

[0025] The calculation formula of the three-dimensional positioning information is:

[0026] Among them, the represents the corrected three-dimensional position information, represents the corrected three-dimensional velocity information, Represents the corrected three-dimensional posture information; Indicates the accelerometer zero bias, represents the gyroscope zero bias, and the δ represents the derivative; and represents the corresponding Jacobian matrix.

[0027] In one implementation, the method of detecting different types of indoor landmark points using a dynamic time warping algorithm to obtain the location information of the landmark points includes:

[0028] The pre-built wireless ranging distance observation model is: L observed =L FTM +d N +d random ;

[0029] Among them, the L observed is the wireless distance value actually observed by the wireless base station, and the L FTM is the true distance value, the d N is the non-line-of-sight error, the d random is a random error;

[0030] The pre-built Bluetooth node distance observation model is:

[0031] Among them, the is the Bluetooth node distance value calculated according to the Bluetooth signal strength, is the corresponding signal strength value, μ is the scaling factor, is the reference signal strength value, and δ is the signal strength observation error;

[0032] Acquire the wireless ranging value at the current time and the starting time obtained according to the wireless ranging distance observation model and the Bluetooth node distance value at the current time and the starting time obtained according to the Bluetooth node distance observation model;

[0033] According to the wireless ranging value, a dynamic time warping algorithm is used to match and output the wireless base station position observation value;

[0034] According to the Bluetooth node distance value, a dynamic time warping algorithm is used to match and output the Bluetooth node position observation value;

[0035] The dynamic time warping algorithm is:

[0036] DTW(d refer ,d k )

[0037] = Dist(b n ,c m )+min[D(b n-1 ,c m ),D(b n ,c m-1 ),D(b n-1 ,c m-1 )];

[0038] Among them, the d refer is the reference sequence, the d k is the sequence value collected in real time, the b n Represents the value of each element in the reference sequence, the c m Represents the value of each element in the sequence collected in real time.

[0039] In one implementation, the method extracts the motion features of the pedestrian multi-source indoor trajectory as an input vector, extracts the reference error value of each step of the pedestrian multi-source indoor trajectory as an output vector, constructs a multi-layer perception network, and adaptively evaluates the error of the pedestrian multi-source indoor trajectory according to the multi-layer perception network to obtain a multi-source fingerprint, including:

[0040] Extract the step length information S of each step of the pedestrian included in the pedestrian multi-source indoor trajectory k and heading information

[0041] Extract the current accumulated step value η k ;

[0042] Extract the current cumulative heading change value

[0043] Extract the ratio of the current walking distance to the total distance

[0044] Extract the ratio of the current time to the total time ξ t (k) = T(k) / T total ;

[0045] Extract the ratio of the current cumulative number of steps to the total number of steps ξ s(k)=step(k) / step total ;

[0046] According to S k , η k , Δθ, ξ d , t and s Construct a multi-layer perception network and calculate the corresponding cost function;

[0047] The calculation formula of the cost function is: Wherein, α represents the network weight to be trained, β represents the network bias to be trained, m is the number of network nodes, g(x) is the network output, and η is the reference true value;

[0048] Use the stochastic gradient descent method to solve the optimal network parameter values ​​of the multi-layer perception network;

[0049] The solution model for the optimal network parameter value is:

[0050] Among them, α k represents the weight in iteration, β l Represents the bias in iteration, ν represents the iteration step value;

[0051] The solved multi-layer perception network is used to adaptively evaluate the error of the pedestrian multi-source indoor trajectory to obtain the multi-source fingerprint.

[0052] In one implementation, the three-dimensional positioning information, the landmark point position information and the crowd-source fingerprint are fused and processed by using a robust Kalman filter to remove abnormal observation values ​​to obtain updated three-dimensional positioning information, including:

[0053] Calculate the sensor distance value between the current time and the starting time according to the three-dimensional positioning information;

[0054] Taking the difference between the wireless distance measurement value and the sensor distance measurement value as the distance observation value;

[0055] Taking the wireless base station position observation value as the first position observation value, and taking the Bluetooth node position observation value as the second position observation value;

[0056] Calculating a variance inflation factor in a robust Kalman filter, and obtaining a covariance matrix according to the variance inflation factor, wherein the variance inflation factor is used to control the weights of the distance observation, the first position observation, and the second position observation;

[0057] Construct a robust Kalman filter state quantity prediction formula;

[0058] The distance observation quantity, the first position observation quantity and the second position observation quantity are updated according to a robust Kalman filter state quantity prediction formula to obtain a robust Kalman filter observation update equation;

[0059] Obtaining observation update values ​​corresponding to the distance observation value, the first position observation value, and the second position observation value according to a robust Kalman filter observation update equation;

[0060] Calculate the optimal state parameter that satisfies the minimum sum of the state prediction error and the observation residual according to the observation update value;

[0061] An optimal state update amount is calculated according to the optimal state parameters to obtain updated three-dimensional positioning information.

[0062] In one implementation, the calculation formula of the distance observation is:

[0063]

[0064] Among them, the δz m,range Represents the observed residual, the d MEMS,m Represents the sensor distance value, the d wifi,m Indicates the wireless distance value;

[0065] The first position observation is expressed as: Said represents the wireless base station position observation value, Represents three-dimensional position information, n wifi represents the observed noise of the wireless base station;

[0066] The second position observation is expressed as: Said Represents the Bluetooth node position observation value, Represents three-dimensional position information, n BLE represents the observation noise of the Bluetooth node;

[0067] The variance inflation factor is expressed as: Among them, the v i represents a distance observation, a first position observation or a second position observation, represents the covariance value corresponding to each observation, and c is the threshold value corresponding to each observation;

[0068] The robust Kalman filter state quantity prediction formula is expressed as:

[0069] Said is the predicted value of the state quantity, φ k,k-1 is the state update matrix, is the state quantity at the current moment;

[0070] The robust Kalman filter observation update equation is expressed as: A k is the observation equation, L k for each observation;

[0071] The optimal state parameters that satisfy the minimum sum of state prediction error and observation residual are calculated as follows: Said is the covariance matrix, the V k Update the value for each observation, α k is the robustness factor;

[0072] The optimal state update calculation formula is expressed as:

[0073]

[0074] The present invention also provides a multi-source fusion positioning device in a large-scale indoor scene, comprising:

[0075] The acquisition module is used to control the pre-built inertial odometer to collect three-dimensional positioning information;

[0076] A detection module is used to detect different types of indoor landmarks using a dynamic time warping algorithm to obtain the location information of the landmarks;

[0077] A construction module is used to extract the motion features of the pedestrian multi-source indoor trajectory as an input vector, and extract the reference error value of each step of the pedestrian multi-source indoor trajectory as an output vector, construct a multi-layer perception network, and adaptively evaluate the error of the pedestrian multi-source indoor trajectory according to the multi-layer perception network to obtain a multi-source fingerprint;

[0078] The fusion processing module is used to use the robust Kalman filter to fuse the three-dimensional positioning information, the landmark point position information and the multi-source fingerprint, eliminate abnormal observation values, and obtain updated three-dimensional positioning information.

[0079] The present invention also provides a terminal, comprising: a memory, a processor, and a multi-source fusion positioning program for a large-scale indoor scene stored in the memory and runable on the processor, wherein the multi-source fusion positioning program for a large-scale indoor scene implements the steps of the multi-source fusion positioning method for a large-scale indoor scene as described above when executed by the processor.

[0080] The present invention also provides a computer-readable storage medium, which stores a computer program. The computer program can be executed to implement the steps of the multi-source fusion positioning method in a large-scale indoor scene as described above.

[0081] The present invention provides a multi-source fusion positioning method and device in a large-scale indoor scene, including: controlling a pre-built inertial odometer to collect three-dimensional positioning information; using a dynamic time warping algorithm to detect different types of indoor landmarks to obtain landmark point location information; extracting the motion characteristics of pedestrian multi-source indoor trajectories as input vectors, and extracting the reference error value of each step of the pedestrian multi-source indoor trajectory as an output vector, constructing a multi-layer perception network, and adaptively evaluating the error of the pedestrian multi-source indoor trajectory according to the multi-layer perception network to obtain a multi-source fingerprint; using an anti-error Kalman filter to fuse the three-dimensional positioning information, the landmark point location information and the multi-source fingerprint, eliminating abnormal observations, and obtaining updated three-dimensional positioning information. The present invention comprehensively considers the important factors that affect the accuracy of multi-source fusion positioning in a large-scale indoor space, and models and evaluates the multi-source trajectory to eliminate interference from the external environment; at the same time, a unified anti-error Kalman filter is designed to organically combine different positioning sources, and finally realizes effective estimation of the three-dimensional position of pedestrians in a large-scale indoor space. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 It is a flow chart of a preferred embodiment of the multi-source fusion positioning method in a large-scale indoor scene of the present invention.

[0083] Figure 2 It is a schematic diagram of landmark point detection.

[0084] Figure 3 It is a schematic diagram of the multi-layer perception network structure.

[0085] Figure 4 This is a comparison chart between the trajectory output by the inertial odometer and the trajectory of the traditional method.

[0086] Figure 5 This is a comparison chart of the positioning accuracy between the trajectory output by the inertial navigation odometer and the trajectory of the traditional method.

[0087] Figure 6 It is a diagram of the reference distribution and real-time distribution effect in the dynamic time warping algorithm.

[0088] Figure 7 This is a schematic diagram of the matching results through the dynamic time warping algorithm.

[0089] Figure 8 This is a comparison chart of trajectory effects estimated by different models.

[0090] Fig. 9 This is a comparison chart of trajectory positioning accuracy estimated by different models.

[0091] Fig.10 It is a functional principle block diagram of a preferred embodiment of the multi-source fusion positioning device in a large-scale indoor scene in the present invention.

[0092] Fig.11 It is a functional principle block diagram of the terminal in the present invention. DETAILED DESCRIPTION

[0093] In order to make the purpose, technical solution and advantages of the present invention clearer and more specific, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0094] At present, due to the complexity and diversity of urban indoor scenes, the existing multi-source fusion positioning technology based on smart terminals in large-scale urban indoor scenes has the following shortcomings: 1) The positioning method relying on the inertial sensor of the smart terminal has the divergent error accumulated over time; 2) There is a lack of effective methods to evaluate the accuracy and reliability of multi-source trajectories; 3) A single wireless signal source is difficult to cover all indoor scenes, and there is a lack of a unified multi-source fusion algorithm architecture.

[0095] In view of the above shortcomings, the present invention proposes a multi-source fusion positioning method for smart terminals in a large-scale indoor scene, so as to improve the multi-source fusion positioning accuracy and robustness of smart terminals in a large-scale urban indoor space, and use a multi-layer perception network to evaluate and model the multi-source trajectory errors.

[0096] In an embodiment of the present invention, the multi-source fusion positioning method in a large-scale indoor scene is applied to computers, smart phones, Internet of Things terminals and other devices. By collecting, processing, and fusing multi-source sensors and wireless signals and QR code data, the indoor trajectory of pedestrians is modeled. At the same time, a multi-layer perception network is used to adaptively evaluate the errors of the acquired multi-source trajectories, and then the multi-source data is fused using an anti-error Kalman filter, ultimately achieving high-precision measurement of pedestrian trajectories in a large-scale urban indoor space.

[0097] See also Figure 1 , Figure 1 Flowchart of the multi-source fusion positioning method in a large-scale indoor scene in the present invention. Figure 1 As shown, the multi-source fusion positioning method in a large-scale indoor scene according to an embodiment of the present invention includes the following steps:

[0098] Step S100: Control the pre-built inertial odometer to collect three-dimensional positioning information.

[0099] Specifically, the present invention proposes a new inertial odometer that combines an inertial navigation mechanical arrangement algorithm with a multi-level observation vector to reduce the influence of magnetic interference and external acceleration.

[0100] In one implementation, the step of constructing the inertial odometer includes:

[0101] The data processing model for constructing accelerometer and gyroscope measurements is:

[0102] Among them, the represents the modeled output of the accelerometer, the represents the modeling output of the gyroscope, the a k represents the ideal output of the accelerometer, the w k represents the ideal output of the gyroscope, the Indicates the accelerometer zero bias, Indicates the gyroscope zero bias, is the rotation matrix, the g w is the gravity vector, n a is the Gaussian white noise measured by the accelerometer, the n w is the Gaussian white noise measured by the gyroscope;

[0103] The accelerometer bias and the gyroscope bias are modeled as a process of bias random walk obeying Gaussian distribution, which is expressed as:

[0104] Among them, the represents the random walk value of the accelerometer zero bias, Represents the random walk value of the gyroscope bias, which follows a Gaussian distribution.

[0105] In one embodiment, the step S100 specifically includes:

[0106] The acceleration value at the current moment is collected by using the modeled accelerometer, and the angular velocity value at the current moment is collected by using the modeled gyroscope;

[0107] Performing pre-integration processing on the acceleration value and the angular velocity value to obtain a pre-integration result;

[0108] The calculation formula of the pre-integration result is:

[0109]

[0110] Among them, the represents the three-dimensional position information obtained by pre-integration, represents the three-dimensional velocity information obtained by pre-integration, represents the three-dimensional posture information obtained by pre-integration, is a rotation matrix, Ω is an antisymmetric matrix, k represents a timestamp, dk represents solving the integral according to the time interval, Indicates the three-dimensional posture information of the previous moment;

[0111] The pre-integration result is corrected according to the accelerometer bias and the gyroscope bias to calculate and obtain three-dimensional positioning information, wherein the three-dimensional positioning information includes: three-dimensional position information, three-dimensional velocity information and three-dimensional attitude information;

[0112] The calculation formula of the three-dimensional positioning information is:

[0113] Among them, the represents the corrected three-dimensional position information, represents the corrected three-dimensional velocity information, Represents the corrected three-dimensional posture information; Indicates the accelerometer zero bias, represents the gyroscope zero bias, and the δ represents the derivative; and represents the corresponding Jacobian matrix.

[0114] The step S100 is followed by: step S200, detecting different types of indoor landmark points using a dynamic time warping algorithm to obtain the location information of the landmark points.

[0115] Specifically, the dynamic time warping algorithm is used to detect different types of indoor landmark points, such as Figure 2 As shown, it includes wireless base stations, Bluetooth beacons, etc. The landmark point detection results are used to provide absolute position reference information for the inertial odometer.

[0116] In one implementation, step S200 specifically includes:

[0117] The pre-built wireless ranging distance observation model is: L observed =L FTM +d N +d random ;

[0118] Among them, the L observed is the wireless distance value actually observed by the wireless base station, and the L FTM is the true distance value, the d N is the non-line-of-sight error, the d random is a random error;

[0119] The pre-built Bluetooth node distance observation model is:

[0120] Among them, the is the Bluetooth node distance value calculated according to the Bluetooth signal strength, is the corresponding signal strength value, μ is the scaling factor, is the reference signal strength value, and δ is the signal strength observation error;

[0121] Acquire the wireless ranging value at the current time and the starting time obtained according to the wireless ranging distance observation model and the Bluetooth node distance value at the current time and the starting time obtained according to the Bluetooth node distance observation model;

[0122] According to the wireless ranging value, a dynamic time warping algorithm is used to match and output the wireless base station position observation value;

[0123] According to the Bluetooth node distance value, a dynamic time warping algorithm is used to match and output the Bluetooth node position observation value;

[0124] The dynamic time warping algorithm is:

[0125] DTW(d refer ,d k )

[0126] = Dist(b n ,c m )+min[D(b n-1 ,c m ),D(b n ,c m-1 ),D(b n-1 ,c m-1 )];

[0127] Among them, the d refer is the reference sequence, the d k is the sequence value collected in real time, the b n Represents the value of each element in the reference sequence, the c m Represents the value of each element in the sequence collected in real time.

[0128] The step S200 is followed by: step S300, extracting the motion features of the pedestrian's multi-source indoor trajectory as an input vector, and extracting the reference error value of each step of the pedestrian's multi-source indoor trajectory as an output vector, constructing a multi-layer perception network, and adaptively evaluating the error of the pedestrian's multi-source indoor trajectory according to the multi-layer perception network to obtain a multi-source fingerprint.

[0129] Specifically, the motion features of pedestrian multi-source indoor trajectories are extracted as input vectors, and the reference error value of each step is extracted as output vectors to construct a multi-layer perceptron to adaptively evaluate the error of multi-source trajectories. Figure 3 shown.

[0130] In one implementation, step S300 specifically includes:

[0131] Extract the step length information S of each step of the pedestrian included in the pedestrian multi-source indoor trajectory k and heading information

[0132] Extract the current accumulated step value η k ;

[0133] Extract the current cumulative heading change value

[0134] Extract the ratio of the current walking distance to the total distance

[0135] Extract the ratio of the current time to the total time ξ t (k) = T(k) / T total ;

[0136] Extract the ratio of the current cumulative number of steps to the total number of steps ξ s (k)=step(k) / step total ;

[0137] According to S k , η k , Δθ, ξ d , t and s Construct a multi-layer perception network and calculate the corresponding cost function;

[0138] The calculation formula of the cost function is: Wherein, α represents the network weight to be trained, β represents the network bias to be trained, m is the number of network nodes, g(x) is the network output, and η is the reference true value;

[0139] Use the stochastic gradient descent method to solve the optimal network parameter values ​​of the multi-layer perception network;

[0140] The solution model for the optimal network parameter value is:

[0141] Among them, α k represents the weight in iteration, β l Represents the bias in iteration, ν represents the iteration step value;

[0142] The solved multi-layer perception network is used to adaptively evaluate the error of the pedestrian multi-source indoor trajectory to obtain the multi-source fingerprint.

[0143] The step S300 is followed by: step S400, using an anti-error Kalman filter to fuse the three-dimensional positioning information, the landmark point position information and the multi-source fingerprint, eliminating abnormal observations, and obtaining updated three-dimensional positioning information.

[0144] Specifically, the robust Kalman filter is used to fuse the inertial navigation odometer, wireless ranging, multi-source fingerprint, and landmark point detection results, and adaptively remove abnormal observations, ultimately achieving an organic combination of different positioning sources and high-precision positioning in a wide range of scenarios.

[0145] In one implementation, step S400 specifically includes:

[0146] Calculate the sensor distance value between the current time and the starting time according to the three-dimensional positioning information;

[0147] On the basis of using the inertial odometer as the state quantity, taking the difference between the wireless distance measurement value and the sensor distance measurement value as the distance observation quantity;

[0148] On the basis of using the inertial odometer as the state quantity, the position of the wireless base station detected by the landmark point detection algorithm is used as the first position observation quantity;

[0149] On the basis of using the inertial odometer as the state quantity, the position of the Bluetooth base station detected by the landmark point detection algorithm is used as the second position observation quantity;

[0150] Design the variance inflation factor in the robust Kalman filter to control the weight of the observations;

[0151] Construct a robust Kalman filter state quantity prediction formula;

[0152] The distance observation quantity, the first position observation quantity and the second position observation quantity are updated according to a robust Kalman filter state quantity prediction formula to obtain a robust Kalman filter observation update equation;

[0153] Obtaining observation update values ​​corresponding to the distance observation value, the first position observation value, and the second position observation value according to a robust Kalman filter observation update equation;

[0154] Calculate the optimal state parameter that satisfies the minimum sum of the state prediction error and the observation residual according to the observation update value;

[0155] An optimal state update amount is calculated according to the optimal state parameters to obtain updated three-dimensional positioning information.

[0156] In one embodiment, the calculation formula of the distance observation is:

[0157] Among them, the δz m,range Represents the observed residual, the d MEMS,m Represents the sensor distance value, the d wifi,m Indicates the wireless distance value;

[0158] The first position observation is expressed as: Said represents the wireless base station position observation value, Represents three-dimensional position information, n wifi represents the observed noise of the wireless base station;

[0159] The second position observation is expressed as: Said Represents the Bluetooth node position observation value, Represents three-dimensional position information, n BLE represents the observation noise of the Bluetooth node;

[0160] The variance inflation factor is expressed as: Among them, the v i represents a distance observation, a first position observation or a second position observation, represents the covariance value corresponding to each observation, and c is the threshold value corresponding to each observation;

[0161] The robust Kalman filter state quantity prediction formula is expressed as:

[0162] Said is the predicted value of the state quantity, φ k,k-1 is the state update matrix, is the state quantity at the current moment;

[0163] The robust Kalman filter observation update equation is expressed as: A k is the observation equation, L k for each observation;

[0164] The optimal state parameters that satisfy the minimum sum of state prediction error and observation residual are calculated as follows: Said is the covariance matrix, the V k Update the value for each observation, α k is the robustness factor;

[0165] The optimal state update calculation formula is expressed as:

[0166]

[0167] Compared with the existing multi-source fusion positioning technology in large-scale urban indoor spaces, the embodiments of the present invention comprehensively consider the processing, evaluation and fusion of multi-source trajectory data, the organic combination of different positioning sources, the elimination of cumulative positioning errors and other issues:

[0168] 1) A new inertial odometer is proposed, which combines the inertial navigation mechanical arrangement algorithm with a multi-level observation vector to reduce the influence of magnetic interference and external acceleration;

[0169] 2) Use dynamic time warping algorithms to detect different types of indoor landmarks, including wireless base stations, Bluetooth beacons, etc. The landmark detection results are used to provide absolute position reference information for the inertial odometer;

[0170] 3) Extract the motion features of pedestrian multi-source indoor trajectories as input vectors, and extract the reference error value of each step as output vectors to construct a multi-layer perceptron to adaptively evaluate the errors of multi-source trajectories;

[0171] 4) Use the robust Kalman filter to fuse the inertial navigation odometer, wireless ranging, multi-source fingerprint, and landmark point detection results, and adaptively remove abnormal observations, ultimately achieving an organic combination of different positioning sources and high-precision positioning in a wide range of scenarios.

[0172] Through the above four points, the embodiment of the present invention can achieve better multi-source fusion positioning results of pedestrian trajectories in large-scale indoor scenes.

[0173] By comparing the estimation results of the inertial navigation odometer proposed in the present invention and the traditional pedestrian dead reckoning algorithm, as shown in Figure 4 and Figure 5 As shown. It can be seen that the inertial navigation odometer proposed in this patent is ahead of the traditional algorithm in the accuracy index of positioning trajectory, and the positioning accuracy can reach 5.45 meters under the condition of 15-20 minutes of movement time and 75% cumulative error distribution.

[0174] Figure 6 A comparison of the real-time and ideal distributions of the collection using a landmark point detection algorithm based on dynamic time warping is shown. Figure 7 The similarity results calculated by using dynamic time warping for real-time landmark point detection are shown. From the figure, we can see that as the distance measurement value of the pedestrian to the landmark point changes, the calculated similarity result first decreases and then increases. When the similarity result is the lowest, it means that the result is closest to the landmark point. The landmark point position can be extracted by combining with the set threshold.

[0175] The final positioning trajectory effect of different positioning source combinations is as follows: Figure 8As shown in the figure, it can be seen that the positioning method using multi-source fusion has significantly improved the positioning accuracy and robustness compared to a single indoor positioning source, and can effectively achieve accurate positioning in three-dimensional indoor scenes. Fig. 9 It can be seen that using a single inertial navigation odometer method, the accuracy can reach 10.94 meters under a 75% cumulative error distribution; using a combination of inertial navigation odometer, QR code, and Bluetooth, the accuracy can reach 4.41 meters under a 75% cumulative error distribution; using a combination of inertial navigation odometer, QR code, Bluetooth, and wireless fingerprint, the accuracy can reach 3.55 meters under a 75% cumulative error distribution; using a combination of inertial navigation odometer, QR code, Bluetooth, wireless fingerprint, and wireless ranging, the accuracy can reach 1.66 meters under a 75% cumulative error distribution, which significantly improves positioning accuracy.

[0176] Furthermore, if Fig.10 As shown, based on the above-mentioned multi-source fusion positioning method in a large-scale indoor scene, the present invention also provides a multi-source fusion positioning device in a large-scale indoor scene, including:

[0177] The acquisition module 100 is used to control the pre-built inertial odometer to acquire three-dimensional positioning information;

[0178] The detection module 200 is used to detect different types of indoor landmark points using a dynamic time warping algorithm to obtain the location information of the landmark points;

[0179] A construction module 300 is used to extract the motion features of the pedestrian multi-source indoor trajectory as an input vector, and extract the reference error value of each step of the pedestrian multi-source indoor trajectory as an output vector, construct a multi-layer perception network, and adaptively evaluate the error of the pedestrian multi-source indoor trajectory according to the multi-layer perception network to obtain a multi-source fingerprint;

[0180] The fusion processing module 400 is used to use the robust Kalman filter to fuse the three-dimensional positioning information, the landmark point position information and the multi-source fingerprint, eliminate abnormal observations, and obtain updated three-dimensional positioning information.

[0181] like Fig.11 As shown, the present invention also provides a terminal, characterized in that it includes: a memory 20, a processor 10, and a multi-source fusion positioning program 30 for a large-scale indoor scene stored in the memory 20 and executable on the processor 10, wherein the multi-source fusion positioning program 30 for a large-scale indoor scene implements the steps of the multi-source fusion positioning method for a large-scale indoor scene as described above when executed by the processor 10.

[0182] The present invention also provides a computer-readable storage medium, which stores a computer program. The computer program can be executed to implement the steps of the multi-source fusion positioning method in a large-scale indoor scene as described above.

[0183] In summary, the present invention discloses a multi-source fusion positioning method and device in a large-scale indoor scene, including: controlling a pre-built inertial odometer to collect three-dimensional positioning information; using a dynamic time warping algorithm to detect different types of indoor landmarks to obtain landmark point location information; extracting the motion characteristics of pedestrian multi-source indoor trajectories as input vectors, and extracting the reference error value of each step of the pedestrian multi-source indoor trajectory as an output vector, constructing a multi-layer perception network, and adaptively evaluating the error of the pedestrian multi-source indoor trajectory according to the multi-layer perception network to obtain a multi-source fingerprint; using an anti-error Kalman filter to fuse the three-dimensional positioning information, the landmark point location information and the multi-source fingerprint, eliminating abnormal observations, and obtaining updated three-dimensional positioning information. The present invention comprehensively considers the important factors that affect the accuracy of multi-source fusion positioning in a large-scale indoor space, and models and evaluates the multi-source trajectory to eliminate interference from the external environment; at the same time, a unified anti-error Kalman filter is designed to organically combine different positioning sources, and finally realizes effective estimation of the three-dimensional position of pedestrians in a large-scale indoor space.

[0184] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A multi-source fusion positioning method in a large-scale indoor scene, characterized in that: include: Control the pre-built inertial odometer to collect three-dimensional positioning information; Use dynamic time warping algorithm to detect different types of indoor landmarks and obtain the location information of the landmarks; Extracting motion features of the pedestrian multi-source indoor trajectory as an input vector, and extracting a reference error value of each step of the pedestrian multi-source indoor trajectory as an output vector, constructing a multi-layer perception network, and adaptively evaluating the error of the pedestrian multi-source indoor trajectory according to the multi-layer perception network to obtain a multi-source fingerprint; Using robust Kalman filtering to fuse the three-dimensional positioning information, the landmark point position information and the multi-source fingerprint, eliminating abnormal observation values, and obtaining updated three-dimensional positioning information; The three-dimensional positioning information, the landmark point position information and the crowd-source fingerprint are fused and processed by using a robust Kalman filter to remove abnormal observation values ​​to obtain updated three-dimensional positioning information, including: Calculate the sensor distance value between the current time and the starting time according to the three-dimensional positioning information; Acquire wireless ranging values ​​at the current time and the starting time obtained according to a pre-constructed wireless ranging distance observation model, and use the difference between the wireless ranging value and the sensor ranging value as a distance observation value; Obtaining a wireless base station position observation value and a Bluetooth node position observation value, taking the wireless base station position observation value as a first position observation value, and taking the Bluetooth node position observation value as a second position observation value; wherein the wireless base station position observation value is matched and outputted according to the wireless ranging value using a dynamic time warping algorithm; the Bluetooth node position observation value is matched and outputted according to the Bluetooth node distance value at the current time and the starting time after obtaining the Bluetooth node distance value at the pre-built Bluetooth node distance observation model; Calculating a variance inflation factor in a robust Kalman filter, and obtaining a covariance matrix according to the variance inflation factor, wherein the variance inflation factor is used to control the weights of the distance observation, the first position observation, and the second position observation; Construct a robust Kalman filter state quantity prediction formula; The distance observation quantity, the first position observation quantity and the second position observation quantity are updated according to a robust Kalman filter state quantity prediction formula to obtain a robust Kalman filter observation update equation; Obtaining observation update values ​​corresponding to the distance observation value, the first position observation value, and the second position observation value according to a robust Kalman filter observation update equation; Calculate the optimal state parameter that satisfies the minimum sum of the state prediction error and the observation residual according to the observation update value; Calculating an optimal state update amount according to the optimal state parameter to obtain updated three-dimensional positioning information; The calculation formula of the distance observation is: Among them, the δz m，range Represents the observed residual, the d MEMS，m Represents the sensor distance value, the d wifi，m Indicates the wireless distance value; The first position observation is expressed as: Said represents the wireless base station position observation value, Represents three-dimensional position information, n wifi represents the observed noise of the wireless base station; The second position observation is expressed as: Said Represents the Bluetooth node position observation value, Represents three-dimensional position information, n BLE represents the observation noise of the Bluetooth node; The variance inflation factor is expressed as: Among them, the v i represents a distance observation, a first position observation or a second position observation, represents the covariance value corresponding to each observation, and c is the threshold value corresponding to each observation; The robust Kalman filter state quantity prediction formula is expressed as: Said is the predicted value of the state quantity, φ k,k-1 is the state update matrix, is the state quantity at the current moment; The robust Kalman filter observation update equation is expressed as: A k is the observation equation, L k is each observation; The optimal state parameters that satisfy the minimum sum of state prediction error and observation residual are calculated as follows: Said is the covariance matrix, the V k Update the value for each observation, α k is the robustness factor; The optimal state update calculation formula is expressed as:

2. The multi-source fusion positioning method in a large-scale indoor scene according to claim 1 is characterized in that: The steps of constructing the inertial odometer include: The data processing model for constructing accelerometer and gyroscope measurements is: Among them, the represents the modeled output of the accelerometer, the represents the modeling output of the gyroscope, the a k represents the ideal output of the accelerometer, the w k represents the ideal output of the gyroscope, the Indicates the accelerometer zero bias, Indicates the gyroscope zero bias, is the rotation matrix, the g w is the gravity vector, n a is the Gaussian white noise measured by the accelerometer, the n w is the Gaussian white noise measured by the gyroscope; The accelerometer bias and the gyroscope bias are modeled as a process of bias random walk obeying Gaussian distribution, which is expressed as: Among them, the represents the random walk value of the accelerometer zero bias, Represents the random walk value of the gyroscope bias, which follows a Gaussian distribution.

3. The multi-source fusion positioning method in a large-scale indoor scene according to claim 2 is characterized in that: The controlling of the pre-built inertial odometer to collect three-dimensional positioning information includes: The acceleration value at the current moment is collected by using the modeled accelerometer, and the angular velocity value at the current moment is collected by using the modeled gyroscope; Performing pre-integration processing on the acceleration value and the angular velocity value to obtain a pre-integration result; The calculation formula of the pre-integration result is: Among them, the represents the three-dimensional position information obtained by pre-integration, represents the three-dimensional velocity information obtained by pre-integration, represents the three-dimensional posture information obtained by pre-integration, is a rotation matrix, Ω is an antisymmetric matrix, k represents a timestamp, dk represents solving the integral according to the time interval, Indicates the three-dimensional posture information of the previous moment; The pre-integration result is corrected according to the accelerometer bias and the gyroscope bias to calculate and obtain three-dimensional positioning information, wherein the three-dimensional positioning information includes: three-dimensional position information, three-dimensional velocity information and three-dimensional attitude information; The calculation formula of the three-dimensional positioning information is: Among them, the represents the corrected three-dimensional position information, represents the corrected three-dimensional velocity information, Represents the corrected three-dimensional posture information; Indicates the accelerometer zero bias, represents the gyroscope zero bias, and the δ represents the derivative; and represents the corresponding Jacobian matrix.

4. The multi-source fusion positioning method in a large-scale indoor scene according to claim 3 is characterized in that: The dynamic time warping algorithm is used to detect different types of indoor landmark points to obtain the location information of the landmark points, including: The pre-built wireless ranging distance observation model is: L observed =L FTM +d N +d random ; Among them, the L observed is the wireless distance value actually observed by the wireless base station, and the L FTM is the true distance value, the d N is the non-line-of-sight error, the d random is a random error; The pre-built Bluetooth node distance observation model is: Among them, the is the Bluetooth node distance value calculated according to the Bluetooth signal strength, is the corresponding signal strength value, μ is the scaling factor, is the reference signal strength value, and δ is the signal strength observation error; Acquire the wireless ranging value at the current time and the starting time obtained according to the wireless ranging distance observation model and the Bluetooth node distance value at the current time and the starting time obtained according to the Bluetooth node distance observation model; According to the wireless ranging value, a dynamic time warping algorithm is used to match and output the wireless base station position observation value; According to the Bluetooth node distance value, a dynamic time warping algorithm is used to match and output the Bluetooth node position observation value; The dynamic time warping algorithm is: DTW(d refer ,is k ) =Dist(b n ,c m )+min[D(b n-1 ,c m ),D(b n ,c m-1 ),D(b n-1 ,c m-1 )]; Among them, the d refer is the reference sequence, the d k is the sequence value collected in real time, the b n Represents the value of each element in the reference sequence, the c m Represents the value of each element in the sequence collected in real time.

5. The multi-source fusion positioning method in a large-scale indoor scene according to claim 4 is characterized in that: The method extracts the motion features of the pedestrian multi-source indoor trajectory as an input vector, extracts the reference error value of each step of the pedestrian multi-source indoor trajectory as an output vector, constructs a multi-layer perception network, and adaptively evaluates the error of the pedestrian multi-source indoor trajectory according to the multi-layer perception network to obtain a multi-source fingerprint, including: Extract the step length information S of each step of the pedestrian included in the pedestrian multi-source indoor trajectory k and heading information Extract the current accumulated step value η k ; Extract the current cumulative heading change value Extract the ratio of the current walking distance to the total distance Extract the ratio of the current time to the total time ξ t (k) = T(k) / T total ; Extract the ratio of the current cumulative number of steps to the total number of steps ξ s (k)=step(k) / step total ; According to S k , η k , Δθ, ξ d , t and s Construct a multi-layer perception network and calculate the corresponding cost function; The calculation formula of the cost function is: Wherein, α represents the network weight to be trained, β represents the network bias to be trained, m is the number of network nodes, g(x) is the network output, and η is the reference true value; Use the stochastic gradient descent method to solve the optimal network parameter values ​​of the multi-layer perception network; The solution model for the optimal network parameter value is: Among them, α k represents the weight in iteration, β l Represents the bias in iteration, ν represents the iteration step value; The solved multi-layer perception network is used to adaptively evaluate the error of the pedestrian multi-source indoor trajectory to obtain the multi-source fingerprint.

6. A multi-source fusion positioning device in a large-scale indoor scene, characterized in that: include: The acquisition module is used to control the pre-built inertial odometer to collect three-dimensional positioning information; A detection module is used to detect different types of indoor landmarks using a dynamic time warping algorithm to obtain the location information of the landmarks; A construction module is used to extract the motion features of the pedestrian multi-source indoor trajectory as an input vector, and extract the reference error value of each step of the pedestrian multi-source indoor trajectory as an output vector, construct a multi-layer perception network, and adaptively evaluate the error of the pedestrian multi-source indoor trajectory according to the multi-layer perception network to obtain a multi-source fingerprint; A fusion processing module, used to use an anti-differential Kalman filter to fuse the three-dimensional positioning information, the landmark point position information and the multi-source fingerprint, eliminate abnormal observation values, and obtain updated three-dimensional positioning information; The three-dimensional positioning information, the landmark point position information and the crowd-source fingerprint are fused and processed by using a robust Kalman filter to remove abnormal observation values ​​to obtain updated three-dimensional positioning information, including: Calculate the sensor distance value between the current time and the starting time according to the three-dimensional positioning information; Acquire wireless ranging values ​​at the current time and the starting time obtained according to a pre-constructed wireless ranging distance observation model, and use the difference between the wireless ranging value and the sensor ranging value as a distance observation value; Obtaining a wireless base station position observation value and a Bluetooth node position observation value, taking the wireless base station position observation value as a first position observation value, and taking the Bluetooth node position observation value as a second position observation value; wherein the wireless base station position observation value is matched and outputted according to the wireless ranging value using a dynamic time warping algorithm; the Bluetooth node position observation value is matched and outputted according to the Bluetooth node distance value at the current time and the starting time after obtaining the Bluetooth node distance value at the pre-built Bluetooth node distance observation model; Calculating a variance inflation factor in a robust Kalman filter, and obtaining a covariance matrix according to the variance inflation factor, wherein the variance inflation factor is used to control the weights of the distance observation, the first position observation, and the second position observation; Construct a robust Kalman filter state quantity prediction formula; The distance observation quantity, the first position observation quantity and the second position observation quantity are updated according to a robust Kalman filter state quantity prediction formula to obtain a robust Kalman filter observation update equation; Obtaining observation update values ​​corresponding to the distance observation value, the first position observation value, and the second position observation value according to a robust Kalman filter observation update equation; Calculate the optimal state parameter that satisfies the minimum sum of the state prediction error and the observation residual according to the observation update value; Calculating an optimal state update amount according to the optimal state parameter to obtain updated three-dimensional positioning information; The calculation formula of the distance observation is: Among them, the δz m,range Represents the observed residual, the d MEMS,m Represents the sensor distance value, the d wifi,m Indicates the wireless distance value; The first position observation is expressed as: Said represents the wireless base station position observation value, Represents three-dimensional position information, n wifi represents the observed noise of the wireless base station; The second position observation is expressed as: Said Represents the Bluetooth node position observation value, Represents three-dimensional position information, n BLE represents the observation noise of the Bluetooth node; The variance inflation factor is expressed as: Among them, the v i represents a distance observation, a first position observation or a second position observation, represents the covariance value corresponding to each observation, and c is the threshold value corresponding to each observation; The robust Kalman filter state quantity prediction formula is expressed as: Said is the predicted value of the state quantity, φ k,k-1 is the state update matrix, is the state quantity at the current moment; The robust Kalman filter observation update equation is expressed as: A k is the observation equation, L k is each observation; The optimal state parameters that satisfy the minimum sum of state prediction error and observation residual are calculated as follows: Said is the covariance matrix, the V k Update the value for each observation, α k is the robustness factor; The optimal state update calculation formula is expressed as:

7. A terminal, characterized in that: include: A memory, a processor, and a multi-source fusion positioning program for a large-scale indoor scene stored in the memory and executable on the processor, wherein the multi-source fusion positioning program for a large-scale indoor scene, when executed by the processor, implements the steps of the multi-source fusion positioning method for a large-scale indoor scene as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program can be executed to implement the steps of the multi-source fusion positioning method in a large-scale indoor scene as described in any one of claims 1 to 5.

Images