Non-line-of-sight identification method based on ultra-wideband ranging information clustering

Abstract

The invention discloses a non-line-of-sight identification method based on ultra-wideband ranging information clustering, which comprises the following steps: S1, taking one ultra-wideband signal transceiver as a label, taking the other ultra-wideband signal transceiver as a base station, and forming a scene between the label and the base station; s2, under a specific condition, enabling the label to collect a continuous ranging distance between the label and the base station in different scenes, and obtaining data sets corresponding to the different scenes; s3, selecting a value as m; substituting the data sets corresponding to different scenes and m into a distance set obtaining method to obtain a distance set; selecting a non-line-of-sight decision value according to the value range of the distance set; s4, obtaining a ranging data set; and S5, performing non-line-of-sight identification on the continuous distance measurement distance of the distance measurement data set according to the non-line-of-sight judgment value. According to the non-line-of-sight identification method provided by the invention, the non-line-of-sight identification accuracy and identification efficiency in the ultra-wideband signal are improved.

Description

Technical Field

[0001] The invention belongs to the field of ultra-wideband indoor ranging and positioning, and in particular relates to a non-line-of-sight recognition method based on ultra-wideband ranging information clustering. Background Art

[0002] With the rapid development of Internet of Things technology, smart mobile terminal technology and mobile computing technology, the demand for indoor high-precision positioning services in all walks of life is growing. Ultra Wide Band (UWB) technology has made significant progress in the field of wireless communications due to its high bandwidth, low power consumption and high-precision positioning. However, in practical applications, due to the influence of factors such as buildings and obstacles, the propagation of ultra-wideband signals in non-line-of-sight (NLOS) environments faces huge challenges. Traditional ultra-wideband positioning technology often has difficulty in accurately identifying and locating targets under non-line-of-sight conditions. Therefore, in order to improve the accuracy of indoor positioning, non-line-of-sight recognition becomes particularly critical.

[0003] At present, there are two main methods for identifying ultra-wideband non-line-of-sight errors. One is a recognition method based directly on measurement data, which uses the digital features of ranging data to detect non-line-of-sight conditions. This method makes judgments by analyzing the differences in propagation paths between line-of-sight (LOS) and non-line-of-sight scenarios. However, when the environment changes from line-of-sight to non-line-of-sight, the ranging data will change suddenly, resulting in significantly higher ranging values.

[0004] Another method uses a deep learning method based on channel impulse response (CIR). This method analyzes the time domain and frequency domain characteristics of the channel impulse response data to reveal the delay and amplitude changes of different paths experienced by the signal due to multipath propagation, and uses various neural network models to identify non-line-of-sight. However, the deep learning method based on channel impulse response (CIR) requires a large amount of channel characteristic parameters to be collected and processed, resulting in a long processing time and is difficult to apply to real-time positioning systems. Summary of the invention

[0005] In view of the problems existing in the prior art, the purpose of the present invention is to provide a non-line-of-sight recognition method based on ultra-wideband ranging information clustering. The non-line-of-sight recognition method of the present invention clusters continuous ranging distances based on the K-Means algorithm, realizes non-line-of-sight recognition, and improves the non-line-of-sight recognition accuracy and recognition efficiency in ultra-wideband signals.

[0006] The present invention is achieved through the following technical solutions.

[0007] A non-line-of-sight recognition method based on ultra-wideband ranging information clustering comprises the following steps:

[0008] S1, use one ultra-wideband signal transceiver as a tag and another ultra-wideband signal transceiver as a base station, so that a scene is formed between the tag and the base station. The scene is a cardboard box blocking non-line-of-sight scene or an iron box blocking non-line-of-sight scene. The cardboard box blocking non-line-of-sight scene means that there is a cardboard box blocking between the tag and the base station, and the iron box blocking non-line-of-sight scene means that there is an iron box blocking between the tag and the base station.

[0009] S2, in the non-line-of-sight scenario with carton occlusion, under certain circumstances, the tag collects the continuous ranging distance between it and the base station every 100ms and forms a set of multiple continuous ranging distances collected as the carton occlusion ranging dataset D 2 In the non-line-of-sight scenario with iron box occlusion, under certain circumstances, the tag collects the continuous ranging distance between it and the base station every 100ms and forms a set of multiple continuous ranging distances collected as the iron box occlusion ranging data set D 3 ; In each scenario, the total duration of the tag collecting continuous ranging distance is T1;

[0010] S3, select a value and use it as m; the carton occlusion distance measurement dataset D 2 Substitute the input data set and m into the "method for obtaining distance sets" to obtain the distance set as the distance set d 2 ; Block the distance measurement data set D with the iron box 3 Substitute the input data set and m into the "method for obtaining distance sets" to obtain the distance set as the distance set d 3 ;

[0011] At the distance set d 2 and the distance set d 3 Select the distance d with the smallest value as the non-line-of-sight judgment value d min ;

[0012] The method for obtaining the distance set is as follows: P groups of to-be-detected data X are selected from the input data set, each group of to-be-detected data X is m consecutive distances measured in the input data set, X = {x i , ..., x m+i-1}, take the data to be detected X as the input of the K-Means algorithm, use the K-Means algorithm to cluster each group of data to be detected X, output the coordinates of two cluster centroids in a one-dimensional coordinate system after clustering each group of data to be detected X, subtract the coordinates of the two cluster centroids and calculate the absolute value to obtain the distance d between the cluster centroids, where P = n-m+1, n represents the number of continuous distances in the input data set; let i take all integers in the range of 1 to n-m+1, and form a distance set by clustering the P groups of data to be detected X in the order of i from small to large; where the classification cluster K in the K-Means algorithm is set to 2;

[0013] S4, obtain the distance measurement data set D 4 :

[0014] Make sure there is no obstruction between the tag and the base station to form an unobstructed line-of-sight scene. In the unobstructed line-of-sight scene, under certain circumstances, make the tag collect the continuous ranging distance between it and the base station every 100ms and form a set of multiple continuous ranging distances collected as the unobstructed ranging data set D 1 , unobstructed distance measurement dataset D 1 The number of continuous ranging distances is m-1;

[0015] The unobstructed distance measurement dataset D 1 And the distance measurement data set D to be measured 4 ′ to obtain the ranging data set D 4 , so that the distance data set D 4 The first m-1 continuous distance measurements are the unobstructed distance measurement data set D 1 Continuous ranging distance in ;

[0016] Among them, the ranging data set D 4 ′ is the continuous distance measurement between the tag and the base station collected under specific circumstances in the T2 time period; the tag and the base station are randomly unobstructed, obstructed by a carton, and obstructed by an iron box under specific circumstances. The tag collects the continuous distance measurement between it and the base station every 100ms and forms a set of multiple continuous distance measurement collected as the distance measurement data set D 4 '; Ranging data set D 4 The continuous distances in 'are out of order;

[0017] S5, for the distance measurement data set D 4 The continuous ranging distance is judged, including S5-1, S5-2, S5-3, S5-4, S5-5 and S5-6:

[0018] S5-1, let j=1;

[0019] S5-2, select the ranging data set D in turn4 The jth to m+j-1th continuous distance measurement in the data set Y to be detected, Y = {x j , ..., x m+j-1};

[0020] S5-3, the data set Y to be tested is used as the input of the K-Means algorithm, the classification cluster K is set to 2, and the K-Means algorithm is used to cluster the data Y to be tested, and the coordinates of the two cluster centroids and x in the one-dimensional coordinate system are obtained. m+j-1 The size of the classification cluster to which it belongs, where in a clustering, the cluster with the larger cluster centroid coordinates is regarded as the large classification cluster, and the cluster with the smaller cluster centroid coordinates is regarded as the small classification cluster; subtract the coordinates of the two cluster centroids in S5-3 and calculate the absolute value to obtain the distance d between the two cluster centroids j ;

[0021] S5-4, according to d min , distance d j and x m+j-1 The size of the classification cluster to which it belongs, for x m+j-1 Perform category judgment and get x m+j-1 The category is line-of-sight data or non-line-of-sight data;

[0022] S5-5, if the value of j is less than s-m+1, then add 1 to the value of j and substitute it into S5-2, and execute steps S5-2 to S5-5; if the value of j is equal to s-m+1, then execute S5-6, where s is the distance measurement data set D 4 The number of consecutive ranging distances in ;

[0023] S5-6, according to the ranging data set D 4 The categories of the mth to sth continuous ranging distances in the data set are composed of the continuous ranging distances belonging to the non-line-of-sight data set, and the continuous ranging distances belonging to the line-of-sight data set are composed of the line-of-sight data set;

[0024] x m+j-1 The method for classifying is, when j is greater than 1, execute according to ① or ②:

[0025] ①When {x j-1 , ..., x m+j-2}When there is no NLOS data:

[0026] If d j >d min And x m+j-1 Belongs to a large classification cluster, then x m+j-1 is NLOS data; otherwise x m+j-1 is the LOS data;

[0027] ②When {x j-1, ..., x m+j-2}When NLOS data exists in:

[0028] If d j ≥d min And x m+j-1 Belongs to a large classification cluster, then x m+j-1 is NLOS data; if d j ≥d min And x m+j-1 Belongs to a small classification cluster, then x m+j-1 is the LOS data;

[0029] If d j <d min And x m+j-2 is LOS data, then x m+j-1 is the LOS data; if d j <d min And x m+j-2 is NLOS data, then x m+j-1 It is NLOS data;

[0030] When j=1, execute: If d j >d min And x m+j-1 Belongs to a large classification cluster, then x m+j-1 is NLOS data; otherwise x m+j-1 This is the LOS data.

[0031] The non-line-of-sight identification method is used to determine whether the ultra-wideband signal between two ultra-wideband signal transceivers is propagated in a non-line-of-sight or line-of-sight environment in a specific situation, and the specific situation is that the positions of the two ultra-wideband signal transceivers remain unchanged or change relatively.

[0032] In S1, the model of the ultra-wideband signal transceiver is HR-RTLS1.

[0033] In S2, the tag collects continuous distance measurement through wireless serial communication; the carton occlusion distance measurement data set D 2 and the iron box occlusion ranging dataset D 3 The total amount of data is 150; T1 = 15 seconds.

[0034] In S3, the method for selecting the value of m includes:

[0035] When the positions of the tag and the base station are relatively fixed, the value of m is a random integer in the range of 3 to 15. When the positions of the tag and the base station are relatively changing, the value of m is selected according to the following steps:

[0036] Step 1: Keep the positions of the tag and the base station unchanged in the carton-occluded non-line-of-sight scene, and make the tag collect the continuous ranging distance between it and the base station every 100ms, and form a set of multiple continuous ranging distances collected as the carton-occluded ranging data set D 6 ; Keep the positions of the tag and the base station unchanged in the non-line-of-sight scene blocked by the iron box, and make the tag collect the continuous ranging distance between it and the base station every 100ms, and form a set of multiple continuous ranging distances collected as the iron box blocking ranging data set D 7 ;

[0037] Step 2: Randomly select an integer value in the range of 3 to 15 as m′, and convert the carton occlusion distance measurement dataset D 6 Substitute the input data set and m′ as m into the “method for obtaining distance sets” to obtain the distance set as the distance set d 6 ; Block the distance measurement data set D with the iron box 7 Substitute the input data set and m′ as m into the “method for obtaining distance sets” to obtain the distance set as the distance set d 7 ; In the distance set d 6 and the distance set d 7 Select the distance d with the smallest value as the non-line-of-sight judgment value d min ′;

[0038] Step 3: Change the relative positions of the two UWB signal transceivers in the unobstructed line-of-sight scene, and make the tag collect the continuous ranging distance between it and the base station every 100ms, and form a set of the collected multiple continuous ranging distances as the unobstructed ranging data set D 5 , randomly select multiple integer values ​​in the range of 3 to 10 and use them as m′′ respectively, and convert the unobstructed ranging data set D 5 Substitute the input data set and m″ as m into the “method for obtaining distance set” to obtain the distance set d 5 , select the distance set d 5 The maximum distance is less than d min ′’s distance set d 5 The corresponding value of m″ is taken as m 5 ;

[0039] Step 4: Determine m 5 Number of: If m 5 If the number of is one, then m 5 The value of is taken as m;

[0040] If m 5 If there are multiple , calculate each m 5 The difference between the two, select the m corresponding to the maximum difference 5 The value of is taken as m, where each m 5The calculation method of the difference includes: the carton occludes the distance measurement data set D 2 As input dataset, m 5 Substitute the value as m into the "method for obtaining distance sets" to obtain the distance set as the distance set d 2 ′; Block the distance measurement data set D with the iron box 3 As input dataset, m 5 Substitute the value as m into the "method for obtaining distance sets" to obtain the distance set as the distance set d 3 ′; in the distance set d 2 ′ and the distance set d 3 ′, select the smallest distance d as the non-line-of-sight judgment value d min ″, the non-line-of-sight judgment value d min ″ and the distance set d 5 Subtract the maximum value from the average value to get the difference.

[0041] In step 1, the carton occlusion distance measurement dataset D 6 and iron box occlusion ranging dataset D 7 The total amount of data is 150; the distance between the tag and the base station in the non-line-of-sight scenario blocked by the cardboard box is 7 meters.

[0042] In step 3, the unobstructed distance measurement data set D 5 The total amount of data is 150; the relative change of position is specifically that the tag moves in a random direction at a speed not exceeding 2m / s within a range of 5m to 20m from the base station.

[0043] In S3, the one-dimensional coordinate system takes the tag as the origin and the direction from the origin to the base station as the positive direction.

[0044] In S4, T2=30 seconds.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] 1. The non-line-of-sight recognition method of the present invention adopts UWB technology with high temporal resolution and anti-interference ability, collects centimeter-level ranging with high accuracy in complex environments, and provides an accurate data basis for NLOS data recognition.

[0047] 2. The non-line-of-sight recognition method of the present invention uses the K-Means algorithm to cluster UWB ranging information to achieve real-time NLOS data (non-line-of-sight data) recognition, which is particularly important for application scenarios that require rapid response (such as unmanned driving, smart homes, etc.).

[0048] 3. The non-line-of-sight recognition method of the present invention effectively filters out outliers and noise in the ranging information through a cluster analysis method, thereby improving the stability and robustness of NLOS data (non-line-of-sight data) recognition. The non-line-of-sight recognition method of the present invention can also maintain a high recognition rate in an environment with complex signals or interference. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 The distance set d is the distance set d under the fixed label of the present invention. 2 , distance set d 2 The value of the middle distance d;

[0050] Figure 2 This is a layout diagram of tags and base stations in Example 3;

[0051] Figure 3 is the distance set d for different m″ values ​​in Example 3 5 ;

[0052] Figure 4 is m in Example 3 5 =4 when the distance set d 2 ′, distance set d 3 The value range of ′. DETAILED DESCRIPTION

[0053] The technical solution of the present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0054] Example 1

[0055] A non-line-of-sight identification method is used to determine whether an ultra-wideband signal between two ultra-wideband signal transceivers is propagated in a non-line-of-sight or line-of-sight environment in a specific case. The specific case is that the positions of the two ultra-wideband signal transceivers remain unchanged or change relatively. In this embodiment, the model of the ultra-wideband signal transceiver is HR-RTLS1, and the non-line-of-sight identification method includes S1 to S5. The specific steps are as follows:

[0056] S1, use an ultra-wideband signal transceiver as a tag (the tag collects continuous ranging distance through wireless serial port communication), and another ultra-wideband signal transceiver as a base station, so that a scene is formed between the tag and the base station. The scene is a carton-blocked non-line-of-sight scene or an iron box-blocked non-line-of-sight scene. The carton-blocked non-line-of-sight scene means that there is a carton block between the tag and the base station, and the iron box-blocked non-line-of-sight scene means that there is an iron box block between the tag and the base station;

[0057] S2, in the non-line-of-sight scenario with carton occlusion, under certain circumstances, the tag collects the continuous ranging distance between it and the base station every 100ms and forms a set of multiple continuous ranging distances collected as the carton occlusion ranging dataset D 2(The total amount of data is 150); In the non-line-of-sight scenario with iron box occlusion, under certain circumstances, the tag collects the continuous ranging distance between it and the base station every 100ms and the collected multiple continuous ranging distances are combined into a set as the iron box occlusion ranging data set D 3 (The total amount of data is 150); wherein, in each scene, the total duration of the tag collecting continuous ranging distance is T1 = 15 seconds.

[0058] S3, select a value and use it as m; the carton occlusion distance measurement dataset D 2 Substitute the input data set and m into the "method for obtaining distance sets" to obtain the distance set as the distance set d 2 ; Block the distance measurement data set D with the iron box 3 Substitute the input data set and m into the "method for obtaining distance sets" to obtain the distance set as the distance set d 3 ;

[0059] At the distance set d 2 and the distance set d 3 Select the distance d with the smallest value as the non-line-of-sight judgment value d min ;

[0060] The method for obtaining the distance set is as follows: P groups of to-be-detected data X are selected from the input data set, each group of to-be-detected data X is m consecutive distances measured in the input data set, X = {x i , ..., x m+i-1}, take the data to be detected X as the input of the K-Means algorithm, and use the K-Means algorithm (Ye Shuaichen, Bu Zhe. Analysis of abnormal traffic identification ability based on K-means clustering algorithm [J]. Information Technology and Informatization, 2024, (06): 127-130.) to cluster each group of data to be detected X. After clustering, each group of data to be detected X outputs the coordinates of the two cluster centroids in a one-dimensional coordinate system (the one-dimensional coordinate system takes the label as the origin and the direction from the origin to the base station as the positive direction). Subtract the coordinates of the two cluster centroids and calculate the absolute value to obtain the distance d between the cluster centroids, where P = n-m+1, n represents the number of continuous ranging distances in the input data set; let i take all integers in the range of 1 to n-m+1, and the distances d obtained after clustering the P groups of data to be detected X are organized into a distance set in the order of i values ​​from small to large; where the classification cluster K in the K-Means algorithm is set to 2.

[0061] S4, obtain the ranging data set D 4 :

[0062] Make sure there is no obstruction between the tag and the base station to form an unobstructed line-of-sight scene. In the unobstructed line-of-sight scene, under certain circumstances, make the tag collect the continuous ranging distance between it and the base station every 100ms and form a set of multiple continuous ranging distances collected as the unobstructed ranging data set D 1 , unobstructed distance measurement dataset D 1 The number of continuous ranging distances is m-1;

[0063] The unobstructed distance measurement dataset D 1 And the distance measurement data set D to be measured 4 ′ to obtain the ranging data set D 4 , so that the distance data set D 4 The first m-1 continuous distance measurements are the unobstructed distance measurement data set D 1 Continuous ranging distance in ;

[0064] Among them, the ranging data set D 4 ′ is the continuous ranging distance between the tag and the base station collected under specific circumstances in the T2 time period; in this embodiment, T2 = 30 seconds, so that the tag and the base station are randomly unobstructed, blocked by a carton, and blocked by an iron box under specific circumstances, and the tag collects the continuous ranging distance between it and the base station every 100ms and forms a set of multiple continuous ranging distances collected as the ranging data set D 4 '; Ranging data set D 4 The continuous ranging distances in ′ are out of order.

[0065] S5, for the distance measurement data set D 4 The continuous ranging distance is judged, including S5-1, S5-2, S5-3, S5-4, S5-5 and S5-6:

[0066] S5-1, let j=1;

[0067] S5-2, select the ranging data set D in turn 4 The jth to m+j-1th continuous distance measurement in the data set Y to be detected, Y = {x j , ..., x m+j-1};

[0068] S5-3, the data set Y to be tested is used as the input of the K-Means algorithm, the classification cluster K is set to 2, and the K-Means algorithm is used to cluster the data Y to be tested, and the coordinates of the two cluster centroids and x in the one-dimensional coordinate system are obtained. m+j-1 The size of the classification cluster to which it belongs, where in a clustering, the cluster with the larger cluster centroid coordinates is regarded as the large classification cluster, and the cluster with the smaller cluster centroid coordinates is regarded as the small classification cluster; subtract the coordinates of the two cluster centroids in S5-3 and calculate the absolute value to obtain the distance d between the two cluster centroidsj ;

[0069] S5-4, according to d min , distance d j and x m+j-1 The size of the classification cluster to which it belongs, for x m+j-1 Perform category judgment and get x m+j-1 The category is line-of-sight (LOS) data or non-line-of-sight (NLOS) data;

[0070] S5-5, if the value of j is less than s-m+1, then add 1 to the value of j and substitute it into S5-2, and execute steps S5-2 to S5-5; if the value of j is equal to s-m+1, then execute S5-6, where s is the distance measurement data set D 4 The number of consecutive ranging distances in ;

[0071] S5-6, based on the ranging data set D 4 The categories of the mth to sth continuous ranging distances in the data set are composed of the continuous ranging distances belonging to the non-line-of-sight data and the continuous ranging distances belonging to the line-of-sight data are composed of the line-of-sight data set.

[0072] x m+j-1 The method for classifying is, when j is greater than 1, execute according to ① or ②:

[0073] ①When {x j-1 , ..., x m+j-2}When there is no NLOS data:

[0074] If d j >d min And x m+j-1 Belongs to a large classification cluster, then x m+j-1 is NLOS data; otherwise x m+j-1 is the LOS data;

[0075] ②When {x j-1 , ..., x m+j-2}When NLOS data exists in:

[0076] If d j ≥d min And x m+j-1 Belongs to a large classification cluster, then x m+j-1 is NLOS data; if d j ≥d min And x m+j-1 Belongs to a small classification cluster, then x m+j-1 is the LOS data;

[0077] If d j <d min And xm+j-2 is LOS data, then x m+j-1 is the LOS data; if d j <d min And x m+j- is NLOS data, then x m+j-1 It is NLOS data;

[0078] When j=1, execute: If d j >d min And x m+j-1 Belongs to a large classification cluster, then x m+j-1 is NLOS data; otherwise x m+j-1 This is the LOS data.

[0079] Example 2

[0080] A non-line-of-sight identification method, based on Example 1, when the specific situation is that the positions of the two ultra-wideband signal transceivers remain unchanged (the positions of the two ultra-wideband signal transceivers are relatively fixed and the distance is 7 meters), the m value in S3 is randomly selected from the range of 3 to 15. In this embodiment, m=10, s=309.

[0081] like Figure 1 As shown (the horizontal axis is the group number of each group of data to be detected X, and the vertical axis is the value of the distance d in the distance set), when m=10, the distance set d 3 for Figure 1 Large NLOS in, distance set d 3 The medium distance d is in the range of 0.22m to 0.45m, and the distance set d 2 for Figure 1 Small NLOS in, distance set d 2 The medium distance d is in the range of 0.17m to 0.25m, so the non-line-of-sight judgment value d min =0.17 meters.

[0082] The CNN-CAM, RF (multiple feature parameters) algorithms and CNN-Stacked-LSTM algorithms are used to perform non-line-of-sight recognition on the CIR data (channel impulse response) collected by the tag in Example 1 while collecting the continuous ranging distance, wherein the CIR data includes: the first channel path, communication power and signal strength;

[0083] The non-line-of-sight recognition method and KNN algorithm of the present invention in Example 1 are used to respectively analyze the test set samples (distance measurement data set D 4 ) to perform non-line-of-sight recognition. The non-line-of-sight recognition method of the present invention performs non-line-of-sight recognition on the continuous ranging distances in the ranging data set D between m=10 and d min=0.17 meters for non-line-of-sight recognition. The recognition results of the above different algorithms are compared, and the accuracy, line-of-sight recall rate, non-line-of-sight recall rate and F1 score are used as performance evaluation indicators (Su Jia, Zhang Jingjing, Yi Qingwu, et al. Research on NLOS / LOS recognition method based on CNN-CAM [J / OL]. Radio Engineering, 1-11 [2024-07-28]. http: / / kns.cnki.net / kcms / detail / 13.1097.TN.20230921.1857.014.html.) The recognition results are analyzed as shown in Table 1.

[0084] Table 1 Performance indicators of different algorithms (the positions of the two UWB signal transceivers remain unchanged)

[0085]

[0086] According to Table 1, when the positions of the two ultra-wideband signal transceivers remain unchanged (the tags are fixed), the accuracy of the non-line-of-sight recognition method of the present invention reaches 95.48%, the line-of-sight recall rate reaches 96.87%, the non-line-of-sight recall rate reaches 94.68%, and the F1 score reaches 95.88%. The recognition effect of the non-line-of-sight recognition method of the present invention is the best.

[0087] Example 3

[0088] A non-line-of-sight identification method, when the specific situation is that the positions of two ultra-wideband signal transceivers change relatively, on the basis of embodiment 1, one of the two ultra-wideband signal transceivers is fixed, and the other ultra-wideband signal transceiver moves randomly, so that the moving ultra-wideband signal transceiver is used as a tag, and the other ultra-wideband signal transceiver is used as a base station, specifically: the tag is moved at a random speed and in a random direction within a range of 5m to 20m from the base station (the moving speed of the tag does not exceed 2m / s), and the relative position of the tag and the base station in this embodiment is as follows Figure 2 As shown in the figure, the tag movement trajectory moves counterclockwise along the rectangular outer frame, the shortest distance (L_min) between the base station and the tag movement trajectory is 5 meters, and the longest distance (L_max) between the base station and the tag movement trajectory is 20 meters.

[0089] The method for selecting the m value includes:

[0090] Step 1: Keep the positions of the tag and the base station unchanged in the carton-occluded non-line-of-sight scene (the distance is 7 meters), and let the tag collect the continuous ranging distance between it and the base station every 100ms, and form a set of multiple continuous ranging distances collected as the carton-occluded ranging data set D 6(The total amount of data is 150); keep the positions of the tag and the base station unchanged in the non-line-of-sight scene blocked by the iron box, and make the tag collect the continuous ranging distance between it and the base station every 100ms, and form a set of multiple continuous ranging distances collected as the iron box blocking ranging data set D 7 (The total amount of data is 150);

[0091] Step 2: Randomly select an integer value in the range of 3 to 15 as m′, and convert the carton occlusion distance measurement dataset D 6 Substitute the input data set and m′ as m into the “method for obtaining distance sets” to obtain the distance set as the distance set d 6 ; Block the distance measurement data set D with the iron box 7 Substitute the input data set and m′ as m into the “method for obtaining distance sets” to obtain the distance set as the distance set d 7 ; In the distance set d 6 and the distance set d 7 Select the distance d with the smallest value as the non-line-of-sight judgment value d min ′;

[0092] Step 3: Change the relative positions of the two UWB signal transceivers in the unobstructed line-of-sight scene, and make the tag collect the continuous ranging distance between it and the base station every 100ms, and form a set of the collected multiple continuous ranging distances as the unobstructed ranging data set D 5 (The total amount of data is 150), randomly select multiple integer values ​​in the range of 3 to 10 and use them as m″, and convert the unobstructed ranging data set D 5 Substitute the input data set and m″ as m into the “method for obtaining distance set” to obtain the distance set d 5 , select the distance set d 5 The maximum distance is less than d min ′’s distance set d 5 The corresponding value of m"" is taken as m 5 ;

[0093] Step 4: Determine m 5 Number of: If m 5 If the number of is one, then m 5 The value of is taken as m;

[0094] If m 5 If there are multiple , calculate each m 5 The difference between the two, select the m corresponding to the maximum difference 5 The value of is taken as m, where each m 5 The calculation method of the difference includes: the carton occludes the distance measurement data set D 2 As input dataset, m 5Substitute the value as m into the "method for obtaining distance sets" to obtain the distance set as the distance set d 2 ′; Block the distance measurement data set D with the iron box 3 As input dataset, m 5 Substitute the value as m into the "method for obtaining distance sets" to obtain the distance set as the distance set d 3 ′; in the distance set d 2 ′ and the distance set d 3 ′, select the smallest distance d as the non-line-of-sight judgment value d min ″, the non-line-of-sight judgment value d min ″ and (the m 5 Corresponding) distance set d 5 Subtract the maximum value from the average value to get the difference.

[0095] In this embodiment, m″=4, 6, 8, 10, P=100 groups, and the obtained distance set d 5 like Figure 3 As shown by Figure 3 It can be seen that when m″=4, the distance set d 5 The value of is in the range of 0.06m to 0.10m; when m″=6, the distance set d 5 The value of is in the range of 0.095m to 0.14m; when m″=8, the distance set d 5 The value of is in the range of 0.13m to 0.18m; when m″=10, the distance set d 5 The value is in the range of 0.16m to 0.22m.

[0096] In this embodiment, the non-line-of-sight determination value d min ′=0.17 meters. In summary, m″=4 and m″=6 both satisfy the distance set d 5 The maximum distance is less than d min ′=0.17 meters.

[0097] like Figure 4 As shown, in m 5 =4, the distance set d 3 'for Figure 4 Large NLOS in, distance set d 3 ′ is in the range of 0.290m to 0.505m, the distance set d 2 'for Figure 4 Small NLOS in, distance set d 2 ′ is in the range of 0.165m to 0.310m, so in m 5 =4 when non-line-of-sight judgment value d min ″=0.165 meters. Figure 4 The LOS in is m 5=4 when the distance set d 5 .

[0098] In m 5 =6, the distance set d 3 ′ is within the range of 0.238~0.471m, the distance set d 2 ′ is in the range of 0.175 to 0.283 meters. Therefore, in m 5 =6 when non-line-of-sight judgment value d min ″=0.175m.

[0099] In m 5 =4 when non-line-of-sight judgment value d min ″=0.165 meters, distance set d 5 In the range of 0.06m to 0.10m, the maximum distance is 0.10m, d min ″ and the distance set d 5 The difference of the upper limit of the value is 0.065;

[0100] In m 5 =6 when non-line-of-sight judgment value d min ″=0.175 meters, distance set d 5 In the range of 0.95m to 0.14m, the maximum distance is 0.14m, d min ″ and the distance set d 5 The difference of the upper limit of the value is 0.035;

[0101] Therefore, in this embodiment, m=4 is selected, d min =0.165 m.

[0102] For comparison, let m be 4 and 6 respectively, and measure the distance data set D under label movement. 4 The recognition and classification are performed, and the recognition results are compared, as shown in Table 2.

[0103] Table 2 Comparison of model performance with different numbers of clustering data

[0104]

[0105] According to Table 2, when m is 6, although there is also a high non-line-of-sight recall rate and F1 score, there is a gap in accuracy and line-of-sight recall rate, which shows that when m is 6, the non-line-of-sight recognition method of the present invention is better at recognizing non-line-of-sight data than recognizing line-of-sight data. Therefore, when m is 4, the classification performance of the non-line-of-sight recognition method of the present invention is the best.

[0106] The KNN algorithm (ZHAO Z, HUANG W, LIANG Y, et al. ANLOS detection method based on machine learning in UWB indoor location system [C] / / 2020 International Conference on Virtual Reality and Intelligent Systems (ICVRIS), Zhangjiajie, China: IEEE, 2021: 214-217.), CNN-CAM, RF (multiple feature parameters) algorithm and CNN-Stacked-LSTM algorithm (Su Jia, Zhang Jingjing, Yi Qingwu, et al. Research on NLOS / LOS recognition method based on CNN-CAM [J / OL]. Radio Engineering, 1-11 [2024-07-28]. http: / / kns.cnki.net / kcms / detail / 13.1097.TN.20230921.1857.014.html.) and the non-line-of-sight recognition method of the present invention are selected in m=4 and d min = 0.165 meters for the test set samples (the distance measurement data set D under label movement) 4 ) are used for identification and the identification results are compared, as shown in Table 3.

[0107] Table 3 Performance indicators of different algorithms (label movement scenario)

[0108]

[0109] When the tag moves, the accuracy of the non-line-of-sight recognition method of the present invention reaches 94.73%, the line-of-sight recall rate reaches 94.52%, the non-line-of-sight recall rate reaches 95.50%, and the F1 score reaches 94.59%. According to Table 3, the classification effect of the method proposed by the present invention is the best.

[0110] According to all the above embodiments, the differences in the fixation and movement of tags are fully considered, and comparative tests are carried out using the non-line-of-sight recognition method of the present invention. From the experimental results, it can be seen that the method proposed by the present invention has the best effect.

[0111] The above accuracy, line-of-sight recall rate, non-line-of-sight recall rate and F1 score reflect the non-line-of-sight recognition method of the present invention for the distance measurement data set D 4 The ability to identify and classify. The calculation formulas for accuracy, line-of-sight recall, non-line-of-sight recall, and F1 score are as follows:

[0112] Accuracy refers to the percentage of the number of continuous ranging data correctly classified by the non-line-of-sight recognition method of the present invention in the ranging data set D4 The ratio of the total number of continuous ranging data in .

[0113]

[0114] Recall LOS ) refers to the proportion of LOS data correctly identified by the non-line-of-sight identification method of the present invention.

[0115]

[0116] Non-line-of-sight recall NLOS ) refers to the proportion of NLOS data correctly identified by the non-line-of-sight identification method of the present invention.

[0117]

[0118] F1-score is the harmonic mean of precision and recall. F1-score is a comprehensive evaluation indicator.

[0119]

[0120] Among them, TP (True Positive) is the number of LOS data judged as LOS data, TN (True Negative) is the number of NLOS data judged as NLOS data, FP (False Positive) is the number of NLOS data judged as LOS data, and FN (False Negative) is the number of LOS data judged as NLOS data.

[0121] The present invention is described above by way of example. It should be noted that, without departing from the core of the present invention, any simple deformation, modification or other equivalent replacement that can be made by those skilled in the art without inventive effort falls within the protection scope of the present invention.

Claims

1. A non-line-of-sight recognition method based on ultra-wideband ranging information clustering, characterized in that: The following steps are involved: S1, use one ultra-wideband signal transceiver as a tag and another ultra-wideband signal transceiver as a base station, so that a scene is formed between the tag and the base station. The scene is a cardboard box blocking non-line-of-sight scene or an iron box blocking non-line-of-sight scene. The cardboard box blocking non-line-of-sight scene means that there is a cardboard box blocking between the tag and the base station, and the iron box blocking non-line-of-sight scene means that there is an iron box blocking between the tag and the base station. S2, in a non-line-of-sight scenario with cardboard boxes blocking, under certain circumstances, the tag collects the continuous ranging distance between it and the base station at intervals, and the multiple continuous ranging distances collected are combined into a set as the cardboard box blocking ranging data set D2; in a non-line-of-sight scenario with iron boxes blocking, under certain circumstances, the tag collects the continuous ranging distance between it and the base station at intervals, and the multiple continuous ranging distances collected are combined into a set as the iron box blocking ranging data set D3; S3, select a value and use it as m; Substitute the cardboard box occlusion distance measurement dataset D2 as the input dataset and m into the "method for obtaining distance sets" to obtain the distance set as distance set d2; substitute the iron box occlusion distance measurement dataset D3 as the input dataset and m into the "method for obtaining distance sets" to obtain the distance set as distance set d3; Select the distance d with the smallest value from the distance set d2 and the distance set d3 as the non-line-of-sight judgment value d min ; The method for obtaining the distance set is as follows: P groups of to-be-detected data x are selected from the input data set, each group of to-be-detected data x is m consecutive distances measured in the input data set, x = {x i , ..., x m+i-1 }, take the data to be detected x as the input of the K-Means algorithm, use the K-Means algorithm to cluster each group of data to be detected x, output the coordinates of two cluster centroids in a one-dimensional coordinate system after clustering each group of data to be detected X, subtract the coordinates of the two cluster centroids and calculate the absolute value to obtain the distance d between the cluster centroids, where P = n-m+1, n represents the number of continuous ranging distances in the input data set; let i take all integers in the range of 1 to n-m+1, and form a distance set by clustering the P groups of data to be detected X according to the value of i from small to large; S4, obtain the ranging data set D4: Make sure that there is no obstruction between the tag and the base station to form an unobstructed line-of-sight scene. In the unobstructed line-of-sight scene, under certain circumstances, make the tag collect continuous ranging distances between it and the base station at intervals and form a set of multiple collected continuous ranging distances as an unobstructed ranging data set D1. The number of continuous ranging distances in the unobstructed ranging data set D1 is m-1; The unobstructed ranging data set D1 and the ranging data set D4′ to be measured are spliced ​​to obtain the ranging data set D4, so that the first m-1 continuous ranging distances in the ranging data set D4 are the continuous ranging distances in the unobstructed ranging data set D1; The ranging data set D4′ is the continuous ranging distance between the tag and the base station collected under specific circumstances during the T2 period; S5, judging the continuous ranging distance of the ranging data set D4, including S5-1, S5-2, S5-3, S5-4, S5-5 and S5-6: S5-1, let j=1; S5-2, sequentially select the jth to m+j-1th continuous ranging distances in the ranging data set D4 to form a data set to be detected Y, Y = {x j , ..., x m+j-1 }; S5-3, take the data set Y to be tested as the input of K-Means algorithm, use K-Means algorithm to cluster the data Y to be tested, and obtain the coordinates of the centroids of two clusters and x in the one-dimensional coordinate system. m+j-1 The size of the classification cluster to which it belongs, where in a clustering, the cluster with the larger cluster centroid coordinates is regarded as the large classification cluster, and the cluster with the smaller cluster centroid coordinates is regarded as the small classification cluster; subtract the coordinates of the two cluster centroids in S5-3 and calculate the absolute value to obtain the distance dj between the two cluster centroids; S5-4, according to d min , distance d j and x m+j-1 The size of the classification cluster to which it belongs, for x m+j-1 Perform category judgment and get x m+j-1 The category is line-of-sight data or non-line-of-sight data; S5-5, if the value of j is less than s-m+1, then the value of j is increased by 1 and substituted into S5-2, and steps S5-2 to S5-5 are executed; if the value of j is equal to s-m+1, then S5-6 is executed, where s is the number of continuous ranging distances in the ranging data set D4; S5-6, according to the categories of the m-th to s-th continuous ranging distances in the ranging data set D4, the continuous ranging distances belonging to non-line-of-sight data are grouped into a non-line-of-sight data set, and the continuous ranging distances belonging to line-of-sight data are grouped into a line-of-sight data set; x m+j-1 The method for classifying is, when j is greater than 1, execute according to ① or ②: ①When {x j-1 , ..., x m+j-2 }When there is no NLOS data: If d j >d min And x m+j-1 Belongs to a large classification cluster, then x m+j-1 is NLOS data; otherwise x m+j-1 is the LOS data; ②When {x j-1 , ..., x m+j-2 }When NLOS data exists in: If d j ≥d min And x m+j-1 Belongs to a large classification cluster, then x m+j-1 is NLOS data; if d j ≥d min And x m+j-1 Belongs to a small classification cluster, then x m+j-1 is the LOS data; If d j <d min And x m+j-2 is LOS data, then x m+j-1 is the LOS data; if d j <d min And x m+j-2 is NLOS data, then x m+j-1 It is NLOS data; When j=1, execute: If d j >d min And x m+j-1 Belongs to a large classification cluster, then x m+j-1 is NLOS data; otherwise x m+j-1 This is the LOS data.

2. The non-line-of-sight recognition method according to claim 1, characterized in that: The non-line-of-sight identification method is used to determine whether the ultra-wideband signal between two ultra-wideband signal transceivers is propagated in a non-line-of-sight or line-of-sight environment under a specific situation; the specific situation is that the positions of the two ultra-wideband signal transceivers remain unchanged or change relatively.

3. The non-line-of-sight recognition method according to claim 1, characterized in that: In S3, the method for selecting the value of m includes: When the positions of the tag and the base station are relatively fixed, the value of m is a random integer in the range of 3 to 15; When the positions of the tag and the base station change relative to each other, select the value of m by following the steps below: Step 1: Keep the positions of the tag and the base station unchanged in the non-line-of-sight scene blocked by a carton, and make the tag collect the continuous ranging distance between it and the base station at intervals, and use the multiple continuous ranging distances collected as a set of carton occlusion ranging data set D6; keep the positions of the tag and the base station unchanged in the non-line-of-sight scene blocked by an iron box, and make the tag collect the continuous ranging distance between it and the base station at intervals, and use the multiple continuous ranging distances collected as a set of iron box occlusion ranging data set D7; Step 2, randomly select an integer value in the range of 3 to 15 as m′, substitute the carton occlusion distance measurement data set D6 as the input data set and m′ as m into the "method for obtaining a distance set", and obtain a distance set as distance set d6; substitute the iron box occlusion distance measurement data set D7 as the input data set and m′ as m into the "method for obtaining a distance set", and obtain a distance set as distance set d7; select the distance d with the smallest value in distance set d6 and distance set d7 as the non-line-of-sight judgment value d min ′; Step 3: Make the positions of the two ultra-wideband signal transceivers change relatively in the unobstructed line-of-sight scene, make the tag collect the continuous ranging distance between it and the base station at intervals, and form a set of multiple continuous ranging distances collected as the unobstructed ranging data set D5, randomly select multiple integer values ​​in the range of 3 to 10 and use them as m″ respectively, substitute the unobstructed ranging data set D5 as the input data set and m″ as m into the "method for obtaining distance set", obtain the distance set d5, select the distance set d5 with the maximum value less than d min The value of m″ corresponding to the distance set d5 of ′ is taken as m5; Step 4, determine the number of m5: if the number of m5 is one, take the value of m5 as m; If there are multiple m5s, the difference of each m5 is calculated, and the value of m5 corresponding to the maximum difference is selected as m, wherein the method for calculating the difference of each m5 includes: substituting the carton occlusion distance measurement data set D2 as the input data set and the m5 value as m into the "method for obtaining a distance set", and obtaining a distance set as a distance set d2′; substituting the iron box occlusion distance measurement data set D3 as the input data set and the m5 value as m into the "method for obtaining a distance set", and obtaining a distance set as a distance set d3′; selecting the smallest distance d in the distance set d2′ and the distance set d3′ as the non-line-of-sight judgment value d min ″, the non-line-of-sight judgment value d min Subtract ″ from the maximum value in the distance set d5 to obtain the difference.

4. The non-line-of-sight recognition method according to claim 1, characterized in that: In S3, the one-dimensional coordinate system takes the tag as the origin and the direction from the origin to the base station as the positive direction.

5. The method for selecting the value of m according to claim 3, characterized in that: The relative position change is specifically: the tag moves in a random direction at a speed not exceeding 2 m / s within a range of 5 m to 20 m from the base station.

Images