A seamless indoor and outdoor positioning method for unmanned vehicles based on region recognition

Abstract

This invention discloses a seamless indoor and outdoor positioning method for unmanned vehicles based on region recognition, belonging to the field of autonomous driving. First, this invention designs a joint prediction model based on particle swarm optimization support vector machine (PSO-SVM) and hidden Markov model (HMM), using acquired GNSS environmental feature data as model input to obtain region recognition results. Then, this invention defines three region models (indoor model, indoor-outdoor transition zone model, and indoor model) to describe the vehicle's environmental location, and selects corresponding measurement information based on the region recognition results. Finally, the switching probabilities of the three environmental sub-models in the IMM algorithm are updated using region switching probabilities, thereby improving the accuracy and positioning precision of the seamless indoor and outdoor positioning environment for unmanned vehicles. Experimental verification shows that the method described in this invention, compared to a single sensor and the IMM_EKF algorithm, can better adapt to environmental changes, thus improving the accuracy and robustness of unmanned vehicle positioning.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving positioning, and specifically to a seamless indoor and outdoor positioning method for unmanned vehicles based on region recognition. Background Technology

[0002] In recent years, autonomous driving technology has developed rapidly, with high-precision vehicle positioning being a key aspect of the planning stage. Seamless indoor and outdoor positioning technology uses the autonomous vehicle terminal as a carrier to effectively integrate various sensor data sources, fusing and adapting different positioning techniques for different scenarios. Utilizing a seamless positioning framework in different indoor and outdoor environments, it aims to achieve seamless indoor and outdoor positioning for autonomous vehicles, meeting the high-precision positioning requirements under various conditions. Inertial Navigation Systems (INS) can provide relative attitude, velocity, and position estimates. A GNSS / INS integrated navigation system can effectively compensate for the short-term failures of Global Navigation Satellite Systems (GNSS).

[0003] However, in complex urban environments, GNSS is susceptible to interference from buildings, leading to signal attenuation and a significant decrease in positioning performance, sometimes even rendering positioning services unavailable. In indoor environments, the positioning capabilities of GNSS / INS integrated navigation systems are limited. One challenge of traditional seamless indoor / outdoor positioning frameworks is the inability of vehicles to accurately identify area boundaries during movement. Furthermore, the probability selection of sub-models during data fusion is not optimal, resulting in low positioning accuracy for unmanned vehicles both indoors and outdoors. Therefore, combining Ultra-Wide Band (UWB) with GNSS and Inertial Measurement Unit (IMU) to form a GNSS / IMU / UWB positioning system can effectively achieve seamless positioning for unmanned vehicles in complex urban environments. While PSO-SVM recognition possesses some pattern classification capabilities, it is prone to getting trapped in local optima when faced with complex and varied regional features, limiting its recognition accuracy and making it difficult to accurately adapt to subtle differences under various operating conditions, thus affecting overall recognition accuracy. HMM recognition relies excessively on state transition probability assumptions, making it poorly adaptable to sudden or abnormal situations. In complex real-world scenarios, it is prone to misjudgments and cannot reliably and efficiently complete high-precision region identification tasks. Furthermore, traditional IMM algorithms, due to the lack of reasonable measurement selection, introduce a large amount of redundant or noisy data, increasing the algorithm's data processing burden, interfering with the extraction of true and valid information, and ultimately significantly reducing positioning accuracy. Simultaneously, its lack of optimization for sub-model switching probabilities prevents it from dynamically adjusting model weights according to real-time scenarios, making it difficult to flexibly cope with changing environments. This results in lag in response when switching between different operating conditions, hindering the improvement of positioning accuracy. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a seamless indoor and outdoor positioning method for unmanned vehicles based on region recognition.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a seamless indoor and outdoor positioning method for unmanned vehicles based on region recognition, comprising the following steps:

[0006] S1. Construct a PSO-SVM / HMM joint prediction model, and collect GNSS environmental feature values ​​through unmanned vehicles, and input the feature values ​​into the joint prediction model for training.

[0007] GNSS environmental characteristics include: total number of visible satellites, mean signal-to-noise ratio, and environmental label corresponding to a known environmental area;

[0008] The joint prediction model constructed includes: PSO-SVM prediction model and HMM prediction model;

[0009] The steps to construct a joint prediction model are as follows:

[0010] S1.1 The optimal objective function is obtained through the PSO-SVM prediction model, as follows:

[0011] For the k-th SVM binary classifier, samples belonging to the k-th SVM binary classifier are defined as positive, and all other samples are defined as negative; the expression is as follows:

[0012]

[0013] The training objective of the model is to find a hyperplane that maximizes the margin between the hyperplane and the two classes of sample points.

[0014] The hyperplane expression is as follows:

[0015] (ω (k) ) T x i +b (k) =0;

[0016] In the formula, ω (k) b represents the normal vector of the hyperplane in the k-th SVM classifier; (k) Represents the intercept; x i Indicates the sample to be tested;

[0017] The expression for calculating the interval is as follows:

[0018] 2 / ||ω (k) ||;

[0019] For any sample point (x) i ,y i (k) The distance to the hyperplane is expressed as:

[0020]

[0021] Wherein, any sample point (x) i y i (k) )middle, The sample to be tested is x i The corresponding area status label category, Region state ∈ S, where S represents the set of region states; sample x to be tested i It is a second-order vector that contains two pieces of information: the total number of visible GNSS satellites and the mean signal-to-noise ratio.

[0022] The objective function to be optimized is expressed as:

[0023]

[0024] In the formula, n is the total number of samples; C represents the penalty parameter; ξ(k) Let ξ represent the slack variable. i (k) ≥0 is used to handle cases where samples are linearly inseparable;

[0025] Introduce Lagrange multipliers α into the objective function i (k) and α ii (k) Construct the Lagrange function, with the following expression:

[0026]

[0027] In the formula, α i (k) Represents the first Lagrange multiplier; α ii (k) Indicates the second Lagrange multiplier; y i (k) and y ii (k) They represent the samples to be tested, x and x respectively. i and the sample to be tested x ii The region status label category, where i, ii∈n, i≠ii; For the sample to be tested, x i Mapping in high-dimensional space; For the sample to be tested, x ii Mapping in high-dimensional space; It is a kernel function;

[0028] In the PSO algorithm, a kernel parameter γ is introduced and, together with a penalty parameter C, is used as the particle position vector input to the particle swarm optimization algorithm to solve the dual problem mentioned above, thus obtaining the optimal α. (k)* Thus, ω is obtained. (k) and b (k) ;

[0029] S1.2, Construct an HMM prediction model;

[0030] A hidden Markov model based on indoor and outdoor area identification is established, with the following expression:

[0031] λ=(S,O,A,B,π)

[0032] In the formula, S represents the set of regional states, including three cases: outdoor region 1, indoor-outdoor transition region 2, and indoor region 3, S∈(1,2,3); O={o1,o2} represents the observation vector composed of the total number of visible satellites and the mean signal-to-noise ratio; A represents the probability matrix of moving from the current location of the unmanned vehicle to the other two regions; B represents the probability of obtaining the observation value from the three hidden states (outdoor region 1, indoor-outdoor transition region 2, and indoor region 3) after obtaining a set of observation values ​​O; π represents the probability of the unmanned vehicle being in different regions at the initial moment.

[0033] S1.3 Determine the model weight information and construct the PSO-SVM / HMM joint prediction model;

[0034] Suppose the validation set has n samples and the true class labels are Y = {y1, y2, ..., y3}. i ,...,y n}, where y i If the region state ∈ S, then the classification result of PSO-SVM on the validation set is: in, The region state ∈ S; the classification result of the HMM on the validation set is: in, The region state ∈ S;

[0035] Define N k Let N be the number of true samples of class k in Y, and let k be the region state ∈ S. s,k N represents the number of samples correctly classified by PSO-SVM for the validation samples. h,k P represents the number of samples correctly classified by the HMM for the validation samples. S,K P represents the classification accuracy of PSO-SVM. H,K The classification accuracy of the Hidden Markov Model (HMM) is represented by P. K =P S,K +P H,K ;

[0036] Defined in joint classification prediction and , respectively, represent the classification prediction weight coefficients of PSO-SVM and HMM in k;

[0037] Classification predictions include cases where results are inconsistent in indoor areas and cases where results are consistent in indoor areas;

[0038] S2. GNSS information is used as an outdoor positioning source to access the unmanned vehicle. The vehicle pose is provided by the IMU. Four UWB base stations are deployed in an indoor rectangular area as indoor positioning sources. Three positioning models (outdoor model, indoor-outdoor transition model, and indoor model) are constructed and the vehicle positioning is initially calculated through the interactive multi-model algorithm (IMM).

[0039] GNSS information includes: the longitude and latitude of the unmanned vehicle;

[0040] S3. Input the GNSS environmental feature values ​​collected during the movement of the unmanned vehicle into the joint prediction model to identify and divide the unmanned vehicle's trajectory into regions and output the region transition probability. The steps are as follows:

[0041] S3.1. Divide the results of S2 into a training set and a test set in an 8:2 ratio; train the model using the training set and test the trained model using the test set.

[0042] S3.2 Output the PSO-SVM classification prediction results;

[0043] For the sample to be tested x i Calculate the hyperplane f when the region state ∈ S of k. k (x)=(ω (k) ) T x i +b (k) The test sample is classified into f k The value of (x) represents the largest category, and the final classification result is defined as:

[0044] V * ={V1 * V2 * ,...V i * ,...,V n *}, V i * The region state ∈S;

[0045] Test sample x i After inputting the hyperplane, the result will be calculated. If the result is greater than 0, it is classified as positive; if it is less than 0, it is classified as negative. This can achieve binary classification in a single model. Therefore, the hyperplane is a decision function.

[0046] S3.3 Output HMM classification prediction results;

[0047] The hidden sequence q with the highest probability on the test set is obtained using the Viterbi algorithm. t * =ψ t+1 (qt+1 * ) where q (t+1) Represents the hidden sequence at the next time step; ψ t+1 This indicates that the backtracking pointer at time t+1 is used; the resulting hidden sequence is denoted as:

[0048] H * ={S1 * S2 * ,...,S i * ,...,S n *}, where S i * The region state ∈S;

[0049] S3.4 Output the PSO-SVM joint classification prediction results;

[0050] Obtain the prediction result V of PSO-SVM * HMM prediction H * Then, the obtained classifier weight information is combined. and The final joint classification result is denoted as:

[0051] S3.5, Output region transition probability;

[0052] The region transition matrix is ​​C T =[c ij ] 3×3 , where c ij Indicates the region q after calculation i * Transferred to other regions q after calculation j * Number of transitions: If T c,t * =q i * And T c,t+1 * =q j * Then c ij =c ij +1, where T c,t * Let represent the category label corresponding to time t; Indicates from q i * The total number of departures, represented by the region switching probability matrix, is as follows: Where, m ij The region state ∈S;

[0053] S4. Optimize the sub-model switching probability in the interactive multi-model algorithm based on the regional transition probability;

[0054] In the IMM algorithm, the model transition probability matrix is ​​defined as: Where, p ij If the region state ∈ S, then the transition probability of the updated model is: P m =ω p P+ω m M = [p m,ij ] 3×3 ; where ω p With ω m These represent the weighting coefficients for the region switching probability and the model transition probability, respectively.

[0055] S5. Based on the region identification and division results, reselect measurement information;

[0056] Specifically, the measurement selection is re-performed based on the area identification results. If the identification result is digital label 1, GNSS information is selected as the measurement; if the identification result is 2, GNSS / UWB information is selected as the measurement. When selecting measurements in the indoor-outdoor transition area, measurement weight information should be added to make better measurement selections, and adjustments should be made based on experience. If the identification result is 3, UWB information is selected as the measurement.

[0057] S6. Input the updated sub-model switching probability and measurement information into the extended Kalman filter of the interactive multi-model algorithm to perform vehicle fusion relocalization.

[0058] Specifically, the measurement information updated after region identification is used as the measurement information of the interactive multi-model, and the IMU measurement value is used as the state input. The outdoor model (ODR), semi-outdoor model (I-OTR), and indoor model (IDO) are introduced as regional positioning sub-models to describe the vehicle's motion state. The sub-model switching probability in the IMM algorithm is optimized according to the region transition probability, and the extended Kalman filter is used as the filter for the interactive multi-model, which is beneficial for filtering nonlinear motion models. This invention enables the vehicle to provide high-precision positioning during indoor and outdoor movement. Compared with traditional IMM positioning and single-model filtering, the positioning accuracy of this invention is improved.

[0059] The beneficial effects of this invention are:

[0060] This invention improves the positioning accuracy of unmanned vehicles in complex environments compared to traditional analysis by using a PSO-SVM / HMM joint prediction model and considering three cases: GNSS / IMU fusion (outdoor areas), GNSS / UWB / IMU fusion (indoor-outdoor transition areas), and UWB / IMU fusion (indoor areas).

[0061] This invention improves the accuracy of identifying the movement area of ​​unmanned vehicles by adjusting the weight coefficients of the model, which is superior to a single classification and prediction model.

[0062] This invention utilizes extended Kalman filtering as an interactive multi-model filter, which is beneficial for filtering nonlinear motion models. This invention enables vehicles to provide high-precision positioning during indoor and outdoor motion, and has good environmental adaptability and positioning accuracy stability. Attached Figure Description

[0063] Figure 1 This is a flowchart illustrating the steps of the present invention;

[0064] Figure 2 This is a technical framework diagram of the present invention;

[0065] Figure 3 This is a technical roadmap for solving the classification accuracy problem in this invention;

[0066] Figure 4 This is a schematic diagram comparing the region recognition accuracy of the present invention;

[0067] Figure 5 This is a schematic diagram comparing the positioning trajectory of the method of the present invention with that of a single sensor;

[0068] Figure 6 This is a schematic diagram comparing the single-point positioning error of the sampling point in the X direction of the present invention with that of a single sensor;

[0069] Figure 7 This is a schematic diagram comparing the single-point positioning error of the sampling point in the Y direction of the present invention with that of a single sensor;

[0070] Figure 8 This is a schematic diagram comparing the positioning trajectories of the method of the present invention with those of a single positioning model and the IMM_EKF algorithm;

[0071] Figure 9 This is a schematic diagram comparing the single-point positioning error of the sampling points in the X direction with that of the method of the present invention, a single positioning model, and the IMM_EKF algorithm.

[0072] Figure 10 This diagram illustrates a comparison of the single-point positioning error of the sampling points in the Y direction between the method of this invention and a single positioning model and the IMM_EKF algorithm. Detailed Implementation

[0073] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0074] like Figure 1 As shown, a seamless indoor and outdoor positioning method for unmanned vehicles based on region recognition includes the following steps:

[0075] S1. Construct a PSO-SVM / HMM joint prediction model, and collect GNSS environmental feature values ​​through unmanned vehicles, and input the feature values ​​into the joint prediction model for training.

[0076] GNSS environmental characteristics include: total number of visible satellites, mean signal-to-noise ratio, and environmental label corresponding to a known environmental area;

[0077] The joint prediction model constructed includes: PSO-SVM prediction model and HMM prediction model;

[0078] The steps to construct a joint prediction model are as follows:

[0079] S1.1 The optimal objective function is obtained through the PSO-SVM prediction model, as follows:

[0080] For the k-th SVM binary classifier, samples belonging to the k-th SVM binary classifier are defined as positive, and all other samples are defined as negative; the expression is as follows:

[0081]

[0082] The training objective of the model is to find a hyperplane that maximizes the margin between the hyperplane and the two classes of sample points.

[0083] The hyperplane expression is as follows:

[0084] (ω (k) ) T x i +b (k) =0;

[0085] In the formula, ω (k) b represents the normal vector of the hyperplane in the k-th SVM classifier; (k) Represents the intercept; x i Indicates the sample to be tested;

[0086] The expression for calculating the interval is as follows:

[0087] 2 / ||ω (k) ||;

[0088] For any sample point (x) i ,y i (k) The distance to the hyperplane is expressed as:

[0089]

[0090] Wherein, any sample point (x) i ,y i (k) )middle, The sample to be tested is x i The corresponding area status label category, Region state ∈ S, where S represents the set of region states; sample x to be tested i It is a second-order vector that contains two pieces of information: the total number of visible GNSS satellites and the mean signal-to-noise ratio.

[0091] The objective function to be optimized is expressed as:

[0092]

[0093] In this embodiment, n = 1406; C > 0 represents the penalty parameter, used to control the degree of penalty for misclassification, C ∈ (1 ~ 10), and in this embodiment, C = 1.8; ξ (k) Let ξ represent the slack variable. i (k) ≥0 is used to handle cases where samples are linearly inseparable;

[0094] Introduce Lagrange multipliers α into the objective function i (k) and α ii (k) Construct the Lagrange function, with the following expression:

[0095]

[0096] In the formula, α i (k) Represents the first Lagrange multiplier; α ii (k) Indicates the second Lagrange multiplier; y i (k) and y ii (k) They represent the samples to be tested, x and x respectively. i and the sample to be tested x ii The region status label category, where i, ii∈n, i≠ii; For the sample to be tested, x i Mapping in high-dimensional space; For the sample to be tested, xii Mapping in high-dimensional space; It is a kernel function; in this embodiment, it is a Gaussian kernel function. Specifically, it is... and The inner product;

[0097] In the PSO algorithm, a kernel parameter γ is introduced and, together with a penalty parameter C, is used as the particle position vector input to the particle swarm optimization algorithm to solve the dual problem mentioned above, thus obtaining the optimal α. (k)* Thus, ω is obtained. (k) and b (k) The kernel parameter γ is obtained directly using MATLAB's built-in tools.

[0098] S1.2, Construct an HMM prediction model;

[0099] A hidden Markov model based on indoor and outdoor area identification is established, with the following expression:

[0100] λ=(S,O,A,B,π)

[0101] In the formula, S represents the set of regional states, including three cases: outdoor region 1, indoor-outdoor transition region 2, and indoor region 3, with values ​​shown in Table 1, S∈(1,2,3); O={o1,o2} represents the observation vector composed of the total number of visible satellites and the mean signal-to-noise ratio; A represents the probability matrix of moving from the current location of the unmanned vehicle to the other two regions, i.e., A=[a ij ] indicates starting from the current region s i To the other two s j The probability of a, where a ij =P(s) j,t+1 |s i,t ), where t represents the current time t, and t+1 represents the next time t. For any s i s j The region state ∈S satisfies The probability matrix is ​​a 3×3 square matrix with the following values: B represents the probability of obtaining the observed value from the three hidden states (outdoor region 1, indoor-outdoor transition region 2, and indoor region 3) after obtaining a set of observed values ​​O, i.e., B = [b i (o t )] indicates that the hidden state is s i Sequence O was observed at that time t The probability of b, where b i (o t )=p(o t |q t =s i B is a 3×2 square matrix with the following values: π represents the probability that the unmanned vehicle is in different areas at the initial moment, i.e., π = {π i} indicates that the initial time is s i The probability of, where π is a 3×1 matrix:

[0102] Table 1: Values ​​of S

[0103]

[0104] S1.3 Determine the model weight information and construct the PSO-SVM / HMM joint prediction model;

[0105] like Figure 3 As shown, there are n samples, and the true class labels are Y = {y1, y2, ..., y3}. i ,...,y n}, where y i If the region state ∈ S, then the classification result of PSO-SVM on the validation set is: in, The region state ∈ S; the classification result of the HMM on the validation set is: in, The region state ∈S;

[0106] Define N k Let N be the number of true samples of class k in Y, and let k be the region state ∈ S. s,k N represents the number of samples correctly classified by PSO-SVM for the validation samples. h,k P represents the number of samples correctly classified by the HMM for the validation samples. S,K P represents the classification accuracy of PSO-SVM. H,K The classification accuracy of the Hidden Markov Model (HMM) is represented by P. K =P S,K +P H,K ;

[0107] In this embodiment, N1 = 975 (975 outdoor area points), N2 = 365 (365 indoor-outdoor transition area points), and N3 = 66 (66 indoor point area points); N S,1 =875; N S,2 =365; N S,3 =60 (where the discrepancy between the total number of categories and the actual number of samples is due to N) S，K This refers to the number of correctly classified points, not the final classification result; for example, in this embodiment, there are 975 actual outdoor sample points, but the SVM classification only yields 875 correct outdoor results (the same applies to the HMM classification result); N h,1 =955; Nh,2 =343; N h,3 =60; the calculated P s,1 =0.90; P s,2 =1; P s,3 =0.90; P h,1 =0.98; P h,2 =0.94; P h,3 =0.90; P k,1 =1.88; P K,2 =1.94; P K,3 =1.8;

[0108] Defined in joint classification prediction and , respectively, represent the classification prediction weight coefficients of PSO-SVM and HMM in k;

[0109] Classification predictions include cases where results are inconsistent in indoor areas and cases where results are consistent in indoor areas;

[0110] In cases of inconsistency, the following situations exist: When the SVM classification result is outdoor and the HMM result is semi-outdoor, the distinction is made based on classification accuracy to determine the final result. In this embodiment, matching is performed from the time series. There are 975 data points in the real outdoor area, but the SVM correctly classifies only 875 points with a classification accuracy of 0.90, while the HMM correctly classifies 955 points with a classification accuracy of 0.98. When all points are classified as outdoor areas, there are 80 points with inconsistent classifications. In this case, weighting coefficients are introduced, and the calculated weighting coefficient for PSO-SVM in the outdoor area is 0.47, and the weighting coefficient for HMM is 0.53. In the case of the indoor-outdoor transition area, there are 22 points with inconsistent classifications. Weighting coefficients are introduced, and the calculated weighting coefficient for PSO-SVM in the transition area is 0.52, and the weighting coefficient for HMM is 0.48.

[0111] The results are consistent in the indoor region, and there is no need to introduce weight coefficients for both classifiers. The weight coefficients for PSO-SVM are expressed as follows: The HMM weighting coefficients are expressed as follows: Region identification results are as follows Figure 4 As shown;

[0112] S2, such as Figure 2As shown, GNSS information is used as an outdoor positioning source to access the unmanned vehicle, the vehicle pose is provided by the IMU, and four UWB base stations are deployed in an indoor rectangular area as indoor positioning sources. Three positioning models (outdoor model, indoor-outdoor transition model, and indoor model) are constructed, and the vehicle positioning is initially calculated by the interactive multi-model algorithm (IMM).

[0113] GNSS information includes: the longitude and latitude of the unmanned vehicle;

[0114] S3. Input the GNSS environmental feature values ​​collected during the movement of the unmanned vehicle into the joint prediction model to identify and divide the unmanned vehicle's trajectory into regions and output the region transition probability. The steps are as follows:

[0115] S3.1. Divide the results of S2 into a training set and a test set in an 8:2 ratio; train the model using the training set and test the trained model using the test set.

[0116] S3.2 Output the PSO-SVM classification prediction results;

[0117] For the sample to be tested x i Calculate the hyperplane f when the region state ∈ S of k. k (x)=(ω (k) ) T x i +b (k) The test sample is classified into f k The value of (x) represents the largest category, and the final classification result is defined as:

[0118] V * ={V1 * V2 * ,...V i * ,...,V n *}, V i * The region state ∈S;

[0119] Test sample x i After inputting the hyperplane, the result will be calculated. If the result is greater than 0, it is classified as positive; if it is less than 0, it is classified as negative. This can achieve binary classification in a single model. Therefore, the hyperplane is a decision function.

[0120] S3.3 Output HMM classification prediction results;

[0121] The hidden sequence q with the highest probability on the test set is obtained using the Viterbi algorithm. t * =ψ t+1 (qt+1 * ) where q (t+1) Represents the hidden sequence at the next time step; ψ t+1 This indicates that the backtracking pointer at time t+1 is used; the resulting hidden sequence is denoted as:

[0122] H * ={S1 * S2 * ,...,S i * ,...,S n *}, where S i * The region state ∈S;

[0123] S3.4 Output the PSO-SVM joint classification prediction results;

[0124] Obtain the prediction result V of PSO-SVM * HMM prediction H * Then, the obtained classifier weight information is combined. and The final joint classification result is denoted as:

[0125] S3.5, Output region transition probability;

[0126] The region transition matrix is ​​C T =[c ij ] 3×3 , where c ij Indicates the region q after calculation i * Transferred to other regions q after calculation j * Number of transitions: If T c,t * =q i * And T c,t+1 * =q j * Then c ij =c ij +1, where T c,t * Let represent the category label corresponding to time t; Indicates from q i * The total number of departures, represented by the region switching probability matrix, is as follows: Where, m ij The region state ∈S;

[0127] S4. Optimize the sub-model switching probability in the IMM algorithm based on the regional transition probability;

[0128] In the IMM algorithm, the model transition probability matrix is ​​defined as: P = [p ij ] 3×3 , where p ij If the region state ∈ S, then the transition probability of the updated model is: P m =ω p P+ω m M = [p m,ij ] 3×3 ; where ω p With ω m These represent the weighting coefficients for the region switching probability and the model transition probability, respectively. In this embodiment, the weighting coefficients obtained are: region switching probability weight: 0.5 and IMM sub-model transition probability weight: 0.5.

[0129] S5. Based on the region identification and division results, reselect measurement information;

[0130] Specifically, the measurement selection is re-performed based on the area identification results. If the identification result is digital label 1, GNSS information is selected as the measurement; if the identification result is 2, GNSS / UWB information is selected as the measurement. When selecting measurements in the transition area between indoor and outdoor environments, measurement weight information should be added to better select the measurement, and adjustments should be made based on experience. In this embodiment, there are a total of 365 points assigned label 2. Based on experience, the weight allocation for the first 30% of points is determined to be 0.75*GNSS information + 0.25*UWB information; for the middle 40% of points, it is 0.5*GNSS information + 0.5*UWB information; and for the last 30% of points, it is 0.35*GNSS information + 0.65*UWB information. If the identification result is 3, UWB information is selected as the measurement.

[0131] S6. Input the updated sub-model switching probability and measurement information into the extended Kalman filter of the IMM algorithm to perform vehicle fusion relocalization.

[0132] Specifically, the updated measurement information after region identification is used as the measurement information for the interactive multi-model (IMM) system. IMU measurements are used as the state input. An outdoor model (ODR), a semi-outdoor model (I-OTR), and an indoor model (IDO) are introduced as regional positioning sub-models to describe the vehicle's motion state. The sub-model switching probability in the IMM algorithm is optimized based on the region transition probability, and an extended Kalman filter is used as the filter for the interactive multi-model system, which is beneficial for filtering nonlinear motion models. This invention enables the vehicle to provide high-precision positioning during both indoor and outdoor motion. Compared with traditional IMM positioning and single-model filtering, the positioning accuracy of this invention is improved. The positioning trajectory compared to a single sensor is shown below. Figure 5 As shown in the figure, the error comparison diagram between the present invention and a single sensor in the X direction is as follows. Figure 6 As shown in the figure, the error comparison diagram in the Y direction is as follows: Figure 7 As shown; the present invention compares the traditional IMM positioning with the filtered positioning trajectory of a single model, as shown in the figure. Figure 8 As shown in the figure, the error comparison diagram in the X direction between the present invention, the IMM algorithm, and single-model filtering is as follows. Figure 9 As shown in the figure, the error comparison diagram in the Y direction is as follows: Figure 10 As shown. Furthermore, this invention incorporates environmental recognition, enabling it to better adapt to complex environments.

Claims

1. A seamless indoor and outdoor positioning method for unmanned vehicles based on region recognition, characterized in that, Includes the following steps: S1. Construct a PSO-SVM / HMM joint prediction model, and collect GNSS environmental feature values ​​through unmanned vehicles, and input the GNSS environmental feature values ​​into the joint prediction model for training. The GNSS environmental characteristics include: total number of visible satellites, average signal-to-noise ratio, and environmental labels corresponding to known environmental areas; S2. GNSS information is used as an outdoor positioning source to access the unmanned vehicle. The vehicle pose is provided by the IMU. Four UWB base stations are deployed in an indoor rectangular area as indoor positioning sources. Three positioning models are constructed and the vehicle positioning is initially calculated through an interactive multi-model algorithm. The GNSS information includes: the longitude and latitude of the unmanned vehicle; The three positioning models include: outdoor model, indoor-outdoor transition model, and indoor model; S3. Input the GNSS environmental feature values ​​collected during the movement of the unmanned vehicle into the joint prediction model, identify and divide the unmanned vehicle's trajectory into regions, and output the region transfer probability. S4. Optimize the sub-model switching probability in the interactive multi-model algorithm based on the regional transition probability; S5. Based on the region identification and division results, reselect measurement information; S6. Input the updated sub-model switching probability and measurement information into the extended Kalman filter of the interactive multi-model algorithm to perform vehicle fusion relocalization and obtain the final result.

2. The seamless indoor and outdoor positioning method for unmanned vehicles based on region recognition according to claim 1, characterized in that: The steps for constructing the PSO-SVM / HMM joint prediction model are as follows: S1.1 The optimal objective function is obtained through the PSO-SVM prediction model, as follows: The objective function to be optimized is expressed as: In the formula, n is the total number of samples; C represents the penalty parameter; ξ (k) Represents slack variables. Used to handle cases where samples are linearly inseparable; ω (k) b represents the normal vector of the hyperplane in the k-th SVM classifier; (k) Represents the intercept; x i Let x represent the sample to be tested; where any sample point (x) i ,y i (k) )middle, The sample to be tested is x i The corresponding area status label category, Region state ∈ S, where S represents the set of region states; sample x to be tested i It is a second-order vector that contains two pieces of information: the total number of visible GNSS satellites and the mean signal-to-noise ratio. Introduce Lagrange multipliers α into the objective function i (k) and α ii (k) Construct the Lagrange function, with the following expression: In the formula, α i (k) Represents the first Lagrange multiplier; α ii (k) Indicates the second Lagrange multiplier; y i (k) and y ii (k) They represent the samples to be tested, x and x respectively. i and the sample to be tested x ii The region status label category, where i, ii∈n, i≠ii; For the sample to be tested, x i Mapping in high-dimensional space; For the sample to be tested, x ii Mapping in high-dimensional space; It is a kernel function; In the PSO algorithm, a kernel parameter γ is introduced and, together with a penalty parameter C, is used as the particle position vector input to the particle swarm optimization algorithm to solve the dual problem and obtain the optimal α. (k)* Thus, ω is obtained. (k) and b (k) ; S1.2, Construct an HMM prediction model; A hidden Markov model based on indoor and outdoor area identification is established, with the following expression: λ=(S,O,A,B,π) In the formula, S represents the set of regional states, including three cases: outdoor region 1, indoor-outdoor transition region 2, and indoor region 3, S∈(1,2,3); O={o1,o2} represents the observation vector composed of the total number of visible satellites and the mean signal-to-noise ratio; A represents the probability matrix of moving from the current location of the unmanned vehicle to the other two regions; B represents the probability of obtaining the observation value from the three hidden states after obtaining a set of observation values ​​O, including: outdoor region 1, indoor-outdoor transition region 2, and indoor region 3; π represents the probability of the unmanned vehicle being in different regions at the initial moment; S1.3 Determine the model weight information and construct the PSO-SVM / HMM joint prediction model; There are n samples, and the true class labels are Y = {y1, y2, ..., y3}. i ,...,y n }, where y i If the region state ∈ S, then the classification result of PSO-SVM on the validation set is: in, The region state ∈ S; the classification result of the HMM on the validation set is: in, The region state ∈ S; Define N k Let N be the number of true samples of class k in Y, and let k be the region state ∈ S. s,k N represents the number of samples correctly classified by PSO-SVM for the validation samples. h,k P represents the number of samples correctly classified by the HMM for the validation samples. S,K P represents the classification accuracy of PSO-SVM. H,K The classification accuracy of the Hidden Markov Model (HMM) is represented by P. K =P S,K +P H,K ; Defined in joint classification prediction and , respectively, represent the classification prediction weight coefficients of PSO-SVM and HMM in k; The classification predictions include cases where the results are inconsistent in the indoor area and cases where the results are consistent in the indoor area.

3. The seamless indoor and outdoor positioning method for unmanned vehicles based on region recognition according to claim 1, characterized in that: The steps for inputting GNSS environmental feature values ​​collected during the movement of the unmanned vehicle into the joint prediction model, identifying and dividing the unmanned vehicle's trajectory into regions, and outputting the region transition probability are as follows: S3.

1. Divide the results of S2 into a training set and a test set in an 8:2 ratio; train the model using the training set and test the trained model using the test set. S3.2 Output the PSO-SVM classification prediction results; For the sample to be tested x i Calculate the hyperplane f when the region state ∈ S of k. k (x)=(ω (k) ) T x i +b (k) The test sample is classified into f k The value of (x) represents the largest category, and the final classification result is defined as: V * ={V1 * V2 * ,...V i * ,...,V n * }, V i * The region state ∈S; Test sample x i After inputting the hyperplane, the result will be calculated. If the result is greater than 0, it is classified as positive; if it is less than 0, it is classified as negative. This can achieve binary classification in a single model. Therefore, the hyperplane is a decision function. S3.3 Output HMM classification prediction results; The hidden sequence q with the highest probability on the test set is obtained using the Viterbi algorithm. t * =ψ t+1 (q t+1 * ) where q (t+1) Represents the hidden sequence at the next time step; ψ t+1 This indicates that the backtracking pointer at time t+1 is used; the resulting hidden sequence is denoted as: H * ={S1 * S2 * ,...,S i * ,...,S n * }, where S i * The region state ∈S; S3.4 Output the PSO-SVM joint classification prediction results; Obtain the prediction result V of PSO-SVM * HMM prediction H * Then, the obtained classifier weight information is combined. and The final joint classification result is denoted as: S3.5, Output region transition probability; The region transition matrix is ​​C T =[c ij ] 3×3 , where c ij Indicates the region q after calculation i * Transferred to other regions q after calculation j * Number of transitions: If T c,t * =q i * And T c,t+1 * =q j * Then c ij =c ij +1, where T c,t * Let represent the category label corresponding to time t; Indicates from q i * The total number of departures, represented by the region switching probability matrix, is as follows: Where, m ij The region state ∈ S.

4. The seamless indoor and outdoor positioning method for unmanned vehicles based on region recognition according to claim 1, characterized in that: In the optimization of sub-model switching probabilities in the interactive multi-model algorithm based on regional transition probabilities, the model transition probability matrix in the interactive multi-model algorithm is defined as follows: Where, p ij If the region state ∈ S, then the transition probability of the updated model is: P m =ω p P+ω m M = [p m,ij ] 3×3 ; where ω p With ω m These represent the weighting coefficients for the region switching probability and the model transition probability, respectively.

5. The seamless indoor and outdoor positioning method for unmanned vehicles based on region recognition according to claim 1, characterized in that: Based on the area identification and division results, the measurement information is reselected. If the identification result is 1, GNSS information is selected as the measurement; if the identification result is 2, GNSS / UWB information is selected as the measurement. When selecting measurements in the indoor-outdoor transition area, measurement weight information should be added to make better measurement selection, and adjustments should be made based on experience. If the identification result is 3, UWB information is selected as the measurement. UWB refers to ultra-wideband obtained through UWB base stations.

Images