UWB indoor positioning method suitable for sparse base station

Abstract

The invention discloses a UWB indoor positioning method suitable for a sparse base station. The method comprises the following steps: acquiring a distance and a time sequence between a UWB label and each known point base station under a sparse deployment condition; segmenting the time sequence ranging data by taking R as a step length, and calculating an average value of each segment of ranging data to obtain ranging data output by the average value; gaussian filtering processing is carried out on the output ranging data; inputting the UWB epoch state in the ranging data and the UWB tag coordinates of the point to be measured as state vectors into a Kalman filtering model, and updating the state vectors by using the ranging data after Gaussian filtering processing; and outputting the UWB tag coordinates of the to-be-detected point obtained by the last update as a positioning result. According to the method, quality control is carried out on signals of the sparsely deployed UWB base stations, and high-precision positioning can still be achieved under the conditions that the indoor environment is closed and the deployment density of the UWB base stations is low.

Description

Technical Field

[0001] The invention belongs to the technical field of wireless positioning, and in particular relates to a UWB indoor positioning method suitable for sparse base stations. Background Art

[0002] UWB can achieve high-precision positioning at the centimeter level or even sub-centimeter level, and is suitable for application scenarios with high accuracy requirements, but it has the disadvantage of high deployment cost. Studies have shown that under the existing positioning algorithm framework, the number of anchor nodes is positively correlated with positioning accuracy, that is, the higher the proportion of anchor nodes, the higher the positioning accuracy. However, increasing the number of anchor nodes means higher hardware costs and complex deployment requirements, which is not conducive to large-scale applications. In addition, the complexity of the indoor positioning environment also increases the difficulty of deploying the UWB system, especially the installation and coordinate measurement requirements of the anchor nodes, which are different from the traditional ground measurement methods. Therefore, it is often necessary to configure and calibrate separately for the specific environment, which further increases the difficulty of implementing the system. Therefore, how to achieve high-precision and stable indoor positioning when UWB is sparsely deployed has become an urgent problem to be solved.

[0003] Among the existing methods, the classic one is to fuse UWB with multi-source sensors to reduce the number of UWB base stations deployed, thereby reducing the deployment cost. However, this method requires the fusion analysis of signals from multiple sensors, and the signal processing process is complicated. Summary of the invention

[0004] The purpose of the present invention is to provide a positioning method suitable for a closed indoor environment with a small number of UWB base stations deployed, aiming to solve the problem that the deployment of multiple UWB base stations in a closed indoor environment will lead to high deployment costs of the positioning system.

[0005] The present invention is achieved through the following technical solutions: A UWB indoor positioning method applicable to sparse base stations comprises the following steps: Step S1. Obtaining UWB base station ranging data under sparse deployment conditions, the data including the distance between the UWB tag at the measured point and each known point base station and the time series; Step S2. Perform mean filtering on the time series ranging data obtained in step S1; segment the time series ranging data obtained in step S1 with a step length of R, calculate the average value of each segment of ranging data, and obtain ranging data outputted as the mean value; Used to reduce the amount of ranging data.

[0006] Step S3: construct a Gaussian fitting function and perform Gaussian filtering on the ranging data outputted from step S2; this is used to reduce the observation noise in the ranging data.

[0007] Step S4. Input the UWB epoch status and the UWB tag coordinates of the point to be measured in the ranging data obtained in Step S2 into the Kalman filter model, and update the state vector using the ranging data processed by Gaussian filtering in Step S3; Step S5. Output the UWB tag coordinates of the point to be measured obtained by the last update in Step S4 as the positioning result.

[0008] The state vector consists of the epoch status and the UWB tag coordinates of the point to be measured. The state vector obtained by the last update also includes the epoch status and the UWB tag coordinates of the point to be measured. The final positioning result only needs to output the UWB tag coordinates of the point to be measured.

[0009] Further, the sparse deployment condition in Step S1 is: an indoor environment where the average distance between UWB base stations is more than 30m or the number of deployed UWB base stations is less than 3.

[0010] Further, the average value of each segment of ranging data in Step S2 is as follows:

[0011] where n is the number of base stations, n = 1, 2, 3,...; k is the new epoch obtained after mean filter update, k = 0, 1, 2, 3...M; i is the epoch before mean filter, i = 1, 2, 3,...N; M is the total amount of data after mean filter, N is the total amount of data before mean filter, and the conversion relationship between M and N is: ; R represents the step size of mean filter, represents the ranging data between the nth known point UWB base station and the UWB tag of the point to be measured at the i-th epoch, is the average distance between the nth known point UWB base station and the UWB tag of the point to be measured at the k-th epoch after mean filter.

[0012] Further, Step S3 specifically includes: The Gaussian fitting function is: ; where, x represents a random variable conforming to the Gaussian distribution, represents the sample mean, represents the variance, and: ; where m is the harmonic coefficient, generally taking 2 or 3. By selecting different m to set different value ranges, ranging average values with different precisions are obtained, thereby adjusting the noise removal ability of the Gaussian fitting function to obtain different quality control effects.

[0013] N is the total amount of data before mean filtering. That is, the average value of each segment of ranging data before mean filtering. By taking the average value of each R ranging data as the ranging data of the next epoch of the new time unit, the ranging average value is the new observation data after mean filtering. Gaussian filtering removes the values ​​with large gross errors in the new ranging data obtained by mean filtering, so the result obtained by Gaussian filtering is the ranging average value after removing the gross errors.

[0014] Furthermore, step S4 specifically includes: Step S4-1, initially weighting the state vector and implementing updates between epochs through Kalman filtering; Step S4-2, construct statistics through data detection method, determine whether the ranging data exceeds the limit, eliminate the ranging data with large errors, and re-predict; Step S4-3, updating the weight matrix of state estimation by using the improved IGGIII weight function; Step S4-4, repeat S4-1 to S4-3 until the difference between the two adjustment results is less than the threshold value, and there is no over-limit distance measurement data during the data detection process, then end the process.

[0015] By combining the distance formula and Kalman filtering, the distance measurement data is used to predict the coordinates of the UWB tag of the next epoch. The epoch state can be obtained from the data list, and the state of the next epoch is directly used to overwrite the state of the previous epoch. The purpose of updating the epoch state is to help determine whether the data is available.

[0016] Furthermore, step S4-1 specifically includes: S4-1-1. The coordinates of the UWB tag of the test point at the initial epoch and the epoch states of the UWB base stations at each known point are used as the state vector of the initial k epochs, and the weight of the initial epoch state vector is set to 1 to obtain the covariance matrix of the initial epoch state vector , where k = 0, 1, 2, 3, …, M; S4-1-2. Calculate the prediction vector: ; in, is the prediction vector for the k+1 epoch, is the transfer matrix from epoch k to epoch k+1, is the state estimation vector of epoch k; S4-1-3. In Kalman filtering, the transfer matrix , state estimation Gaussian white noise The covariance matrix of , calculate the covariance matrix of the prediction vector : ; where the superscript T represents the transpose of the matrix, represents the weight matrix of the prediction vector ; S4-1-4. Take the ranging data of each base station in the (k + 1)-th epoch as the observation vector, and calculate the prediction residual vector and its covariance matrix : ; where is the observation matrix in the (k + 1)-th epoch, is the observation vector in the (k + 1)-th epoch, is the covariance matrix of the observation vector in the (k + 1)-th epoch.

[0017] Furthermore, step S4-2 is specifically as follows: Use the Baarda gross error detection method. Assume that there is only one gross error in the ranging data, and construct a statistic using the statistical hypothesis testing method to detect the gross error.

[0018] Furthermore, the original hypothesis of the Baarda data detection method is: the ranging data contains no gross error, then the expectation of the observation residual vector in this epoch is zero, that is , construct a w statistic that conforms to the normal distribution:

[0019] where represents the statistic in the (k + 1)-th epoch, represents the residual vector in the (k + 1)-th epoch. Conduct a normal test on equation (10). If , it means , represents the critical value of the rejection region, that is, there may be a gross error in the ranging data. Determine the threshold by setting the significance level and repeat the detection process to eliminate the gross error.

[0020] Furthermore, the specific method of step S4-3 is: Utilize the prediction residual vector obtained through step S4-2, and update the covariance matrix of the prediction residual vector multiple times through the improved IGGⅢ weight selection and iteration method, and update the weight matrix of the prediction residual vector for this model; The weight function model is: ; where is the harmonic coefficient, The value is 2.0 or 3.0, which is adjusted according to the actual situation; The value is , representing the constant term of the error equation; representing the standardized residual, , being the unit weight variance factor; being the weight matrix of the updated predicted residual vector; Through the function calculates the median value to obtain the unit weight variance factor : ; wherein, the function of the median( ) function is to take the median value; According to the weight matrix of the updated predicted residual vector, its covariance matrix is transformed as follows: ; After determining the covariance matrix of the predicted residual vector, calculate the Kalman gain matrix : ; Solve for the new state estimation vector : ; Calculate the covariance matrix of the new state : ; wherein, represents the identity matrix, is the predicted vector at epoch k, represents the covariance matrix of representing the covariance matrix of the observation vector at epoch k; Update k to k + 1, and perform the prediction for the next epoch through formula (9) until the update of all epochs is completed.

[0021] The beneficial effects of the present invention are: The present invention first segments the ranging data by mean filtering, then uses Gaussian filtering to weaken the noise in the ranging data, then uses the weighted iteration method based on the residual value to determine the weight, and finally uses the Kalman filter considering the UWB base station information to calculate the coordinates of the UWB tag at the point to be measured; this method performs quality control on the signals of sparsely deployed UWB base stations, enabling high-precision and stable indoor positioning even in a closed indoor environment with a low deployment density of UWB base stations. Compared with the solution of simply increasing the number of UWB base station deployments, the present invention can reduce the deployment cost of the UWB positioning system and achieve stable indoor positioning. The present invention only needs to process the signals of known-point UWB base stations, can use a unified standard to process sensor signals, and simplifies the signal processing process. Description of the Drawings

[0022] Figure 1 A UWB indoor positioning method applicable to sparse base stations; Figure 2 A comparison chart of the variation of the positioning error of this method and the KF positioning algorithm in the x direction; Figure 3 A comparison chart of the variation of the positioning error of this method and the KF positioning algorithm in the y direction; Figure 4 A comparison chart of the variation of the positioning error of this method and the KF positioning algorithm in the z direction. Detailed Embodiment

[0023] In order to make the purpose, technical solution and advantages of the present invention clearer, the following will combine the drawings to briefly describe the present invention.

[0024] The idea of this method is to reduce the sampling rate by mean filtering, then use Gaussian filtering to fit the data, preliminarily eliminate the data with poor quality, then input the epoch state of the sparse base station and the coordinates of the UWB tag at the point to be measured into the Kalman filter equation, and in the update process between epochs, use the weighted iteration method and the data detection method to eliminate the gross error. In this method, in the update and prediction process of the Kalman filter between epochs, the weighted iteration method and the data detection method are used to ensure high-quality data, so as to achieve better filtering effect. Refer to Figure 1 , a UWB indoor positioning method applicable to sparse base stations, specifically includes the following steps: Step S1. In this embodiment, an experiment is carried out in an indoor environment of about 30m * 100m. A LinkTrack UWB is arranged at each of the four corners, and UWB tags to be measured are deployed at any position in this environment. The control console is connected to the UWB tags to be measured to receive the ranging signals of the known UWB base stations. Signal calibration is performed, and finally, the UWB tags to be measured are started to receive the ranging signals of the known UWB base stations. The four known UWB base stations fix the local coordinate system through one-key calibration. After determining the coordinate system, the coordinates of each known point in the local coordinate system are determined. In this mode, the UWB base station measures the distance through the TOA ranging method, and each base station is time-synchronized and is the local time of the system. Finally, the exported data includes the coordinates of each known UWB base station, the distance between the known UWB base station and the UWB tag to be measured, and the local system time corresponding to the node when this frame is output.

[0025] Step S2. The UWB base station broadcasts 50 ranging signals per second, and mean filtering is used to reduce the sampling rate. Taking 50 epochs as a group, the mean value of the distance data between the known point and the UWB tag to be measured is calculated according to formula (1) and output as the processing result.

[0026] Using mean filtering, the average value of 50 data within 1 second is output as one epoch, reducing the sampling rate of UWB and initially improving the stability of the UWB ranging signal.

[0027] Step S3. Gaussian fitting is used to weaken the observation noise in the measurement of the UWB base station. A Gaussian fitting function based on the UWB ranging signal is established to perform further quality control on the ranging signal processed by mean filtering in Step S2.

[0028] This method is based on the average value of the UWB distance data, and the 2-fold standard deviation of the data is set as the boundary, that is, the value in formula (5) is 2.

[0029] Step S4. Since there are cases where there is no ranging data for some epochs in the UWB base station under sparse deployment conditions, therefore, through Kalman filtering considering the UWB base station information, the UWB epoch state and the coordinates of the UWB tag to be measured are input into the Kalman filter as the state vector, and the state vector is updated using the ranging data in Step S3.

[0030] The specific implementation steps include: Step S4-1. After the prediction stage of Kalman filtering, the predicted state vector and the covariance matrix of the state vector are calculated, specifically including: S4-1-1. The coordinates of the UWB tag to be measured at the initial epoch and the information of each known UWB base station are used as the state vector at the initial k epoch :

[0031] Among them, is the coordinate vector of the UWB tag at the k-th epoch to be measured, is the epoch status vector of each known UWB base station at the k-th epoch. The coordinate of the UWB tag at the initial epoch to be measured is , and the initial vector of the signal of the known UWB base station is .

[0032]

[0033] Among them, x k , y k , z k is the coordinate of the UWB tag at the k-th epoch to be measured.

[0034] In this method, four known UWB base stations are used, so there are:

[0035] Among them, are the epoch statuses of the signals of the four known base stations at the k-th epoch, respectively. The corresponding value is 1 when the signal is available, and 0 when the signal is unavailable.

[0036] S4-1-2. Calculate the prediction vector according to formula (6). This method is carried out in a static state, and is set as the identity matrix , so there is:

[0037] S4-1-3. Calculate the covariance matrix of the state estimation vector according to formula (7). At the initial state, the weight of the state vector is initially set to 1:

[0038] Among them, is the inverse matrix of the weight matrix of the prediction vector covariance matrix, is the weight matrix of the prediction vector ； S4-1-4. The coordinates of each known UWB base station:

[0039] Matrix is the coordinate of the n-th known UWB base station.

[0040] Among them, is the coordinate of the n-th known UWB base station; Calculate the predicted distance between each known UWB base station and the UWB tag to be measured:

[0041] Wherein, is the predicted distance between the nth known UWB base station and the UWB tag to be measured at the kth epoch; Calculate the observation matrix at the (k + 1)th epoch :

[0042] is the conversion matrix of the epoch state, which is calculated from the difference between the coordinates of the UWB tag to be measured at the kth epoch and the coordinates of each known UWB base station and the predicted distance:

[0043] Wherein, is the observation vector weight matrix of, are the coordinates of the 1st, 2nd, 3rd, and 4th known UWB base stations respectively, are the predicted distances between the 1st, 2nd, 3rd, and 4th known UWB base stations and the UWB tag to be measured respectively; Observation vector There is in:

[0044] Wherein, are the average distances between the 1st UWB base station, the 2nd UWB base station, the 3rd UWB base station, and the 4th UWB base station and the UWB tag to be measured at the kth epoch after mean filtering respectively.

[0045] Calculate the covariance matrix of the observation vector at the (k + 1)th epoch through formula (9).

[0046] Step S4-2, based on the covariance matrix of the processed observation state vector, use the data snooping method to detect the gross error ranging data and delete the ranging data, and let k = k + 1, and repeat the above steps using the ranging data of the next epoch until the statistic does not exceed the limit.

[0047] The present invention uses the Baarda gross error detection method, assuming that there is only one gross error in the ranging data, and uses the statistical hypothesis testing method to construct w measurement to detect the gross error.

[0048] Repeat the data detection and the prediction stage of the Kalman filter to eliminate the errors in the ranging data. It can be considered that the ranging data of this epoch does not contain errors, and at this time, the repetition of this process ends.

[0049] Step S4-3, according to formulas (11) to (16), update the covariance matrix of the prediction residual vector multiple times through the improved IGGⅢ weight selection and iteration method. This model directly updates the weight matrix of the prediction residual vector.

[0050] To sum up, in a specific embodiment, first, four node UWB base stations are deployed in the target indoor environment in a sparse network layout manner. After calibration, the local coordinate system is determined, and the known point UWB base station is started to receive ranging information. Secondly, mean filtering is used to output filtered data with 1 s as an epoch, improving the stability of the ranging signal and reducing the sampling rate. Then, Gaussian filtering is used to weaken the noise in the ranging process, further improving the stability of positioning. Since there are cases where there is no ranging data for some epochs under sparse deployment conditions, the Kalman filter considering UWB base station information is used to input the UWB epoch state and the coordinates of the UWB tag at the point to be measured as the state vector into the Kalman filter, and the state vector is updated using the ranging data. Then, the weights of the state vector are determined for the first time, and the Kalman filter is used to achieve the update between epochs. Next, the statistic constructed by the data detection method is used to detect whether the ranging data exceeds the limit, the ranging data with large errors is removed, and the ranging data of the next epoch is used for re-prediction. Then, the weight matrix of the state estimation is updated through the improved IGGⅢ weight function. The above process is repeated until there is no ranging data exceeding the limit in the data detection process, and the coordinates of the UWB tag at the point to be measured can be obtained. So far, under sparse deployment conditions, a quality control method of filtering and weight selection and iteration is added to the UWB ranging signal, the coordinates of the UWB tag at the point to be measured are solved using the Kalman filter, and robust estimation is performed using the data detection method. This method can reduce the deployment cost of UWB in the indoor environment. In the case of deploying four base stations, it can well ensure the accuracy, stability and reliability of positioning.

[0051] To verify the positioning performance of the present invention in the actual sparse base station application scenario, the positioning performance of the positioning algorithm of the present invention is compared with that based on the traditional Kalman filter KF. The experimental results are as Figure 2 、 Figure 3 and Figure 4 shown. Among them, Figure 2 、 Figure 3 、 Figure 4 are respectively the comparison diagrams of the positioning error changes of the positioning algorithm of the present invention and the KF positioning algorithm in the x-direction, y-direction and z-direction. From Figure 2 、 Figure 3 、 Figure 4It can be seen that in the x-direction, y-direction, especially the z-direction, the positioning error of the algorithm of the present invention is smaller than that of the KF algorithm, and the absolute error reduced by the algorithm of the present invention is obvious. Compared with the KF algorithm, the positioning accuracy of the algorithm of the present invention is improved by 45.6% in the x-direction, 32.56% in the y-direction, and 23.9% in the z-direction. In summary, the present invention improves the positioning accuracy in a sparse UWB base station environment and has good robustness and practicability.

Claims

1. A UWB indoor positioning method suitable for sparse base stations, characterized in that: The method comprises the following steps: Step S1. Obtaining UWB base station ranging data under sparse deployment conditions, the data including the distance between the UWB tag at the measured point and each known point UWB base station and the time series; Step S2. Perform mean filtering on the time series ranging data obtained in step S1; segment the time series ranging data obtained in step S1 with a step length of R, calculate the average value of each segment of ranging data, and obtain ranging data outputted as the mean value; Step S3. Construct a Gaussian fitting function and perform Gaussian filtering on the ranging data outputted from step S2; Step S4. The UWB epoch state and the coordinates of the UWB tag of the measured point in the ranging data obtained in step S2 are input into the Kalman filter model as a state vector, and the ranging data processed by Gaussian filtering in step S3 is used to update the state vector; Step S5, outputting the UWB tag coordinates of the measured point obtained by the last update in step S4 as the positioning result.

2. The UWB indoor positioning method applicable to sparse base stations according to claim 1, characterized in that: The sparse deployment condition in step S1 is: an indoor environment where the average distance between UWB base stations is more than 30 m or the number of UWB base stations deployed is less than 3.

3. The UWB indoor positioning method applicable to sparse base stations according to claim 1, characterized in that: The average value of each segment of ranging data in step S2 As shown below: ; Where n is the number of base stations, = 1, 2, 3, ...; k is the new epoch obtained after the mean filter update, k = 0, 1, 2, 3 ... M; i is the epoch before the mean filter, i = 1, 2, 3, ... N; M is the total amount of data after the mean filter, N is the total amount of data before the mean filter, and the conversion relationship between M and N is: ; R represents the step size of the mean filter, It represents the distance measurement data between the nth known point UWB base station and the UWB tag at the measured point at the i-th epoch, It is the average distance between the nth known point UWB base station and the UWB tag to be measured at the kth epoch after mean filtering.

4. The UWB indoor positioning method applicable to sparse base stations according to claim 1, characterized in that: Step S3 specifically includes: Gaussian fitting function for: ； in, x represents a random variable that conforms to a Gaussian distribution, represents the sample mean, represents the variance, and: ； Where m is the harmonic coefficient, which can be 2 or 3.

5. The UWB indoor positioning method applicable to sparse base stations according to claim 1, characterized in that: Step S4 specifically includes: Step S4-1, initially weighting the state vector and implementing updates between epochs through Kalman filtering; Step S4-2, construct statistics through data detection method, determine whether the ranging data exceeds the limit, eliminate the ranging data with large errors, and re-predict; Step S4-3, updating the weight matrix of state estimation by using the improved IGGIII weight function; Step S4-4, repeat S4-1 to S4-3 until the difference between the two adjustment results is less than the threshold value, and there is no over-limit distance measurement data during the data detection process, then end the process.

6. The UWB indoor positioning method applicable to sparse base stations according to claim 5, characterized in that: Step S4-1 specifically includes: S4-1-1. The coordinates of the UWB tag of the test point at the initial epoch and the epoch states of the UWB base stations at each known point are used as the state vector of the initial k epochs, and the weight of the initial epoch state vector is set to 1 to obtain the covariance matrix of the initial epoch state vector , where k = 0, 1, 2, 3, …, M; S4-1-2. Calculate the prediction vector: ； in, is the prediction vector for the k+1 epoch, is the transfer matrix from epoch k to epoch k+1, is the state estimation vector of epoch k; S4-1-3. In Kalman filtering, the transfer matrix , state estimation Gaussian white noise The covariance matrix of , calculate the covariance matrix of the prediction vector : ； Among them, the superscript T represents the transpose of a matrix, Represents the prediction vector 's power array; S4-1-4. Take the ranging data of each base station at the k+1 epoch as the observation vector and calculate the prediction residual vector of the k+1 epoch and its covariance matrix : ； in, is the observation matrix of the k+1 epoch, is the observation vector at epoch k+1, is the covariance matrix of the k+1 epoch observation vector.

7. The UWB indoor positioning method applicable to sparse base stations according to claim 5, characterized in that: Step S4-2 is specifically as follows: The Baarda gross error detection method is used, assuming that there is only one gross error in the ranging data, and the statistical hypothesis test method is used to construct statistics to detect the gross error.

8. The UWB indoor positioning method applicable to sparse base stations according to claim 7, characterized in that: The original hypothesis of the Baarda data detection method is: if the ranging data does not contain gross errors, then the expectation of the residual vector of the epoch observation is zero, that is, , construct a normal distribution w Statistics: ； in, represents the statistic of k+1 epoch, represents the residual vector of k+1 epoch, and a normality test is performed on equation (10). If , then it means , Represents the critical value of the rejection region, that is, there may be gross errors in the distance measurement data. By setting the significance level The threshold is determined and the detection process is repeated to eliminate gross errors.

9. The UWB indoor positioning method applicable to sparse base stations according to claim 6, characterized in that: The specific method of step S4-3 is: Using the prediction residual vector obtained in step S4-2 , the covariance matrix of the predicted residual vector is obtained by the improved IGGⅢ weighted iteration method After multiple updates, the model predicts the weight matrix of the residual vector Make updates; The weight function model is: ； in, is the harmonic coefficient, The value is 2.0 or 3.0, which can be adjusted according to the actual situation; The value is , represents the constant term of the error equation; represents the standardized residual, , is the unit weight variance factor; is the weight matrix of the updated prediction residual vector; pass Function Compute The median value is the unit weight variance factor : ； Among them, the function of median() is to get the median value; According to the updated prediction residual vector weight matrix , and transform it to get its covariance matrix: ； After determining the covariance matrix of the prediction residual vector, calculate the Kalman gain matrix : ； Solve for the new state estimate vector : ； Calculate the covariance matrix of the new state : ； in, represents the identity matrix, is the k-epoch prediction vector, express The covariance matrix of represents the covariance matrix of the k-epoch observation vector; Update k to k+1 and make a prediction for the next epoch using formula (9) until all epochs are updated.

Images