A millimeter wave radar-based indoor human target tracking method and system

Abstract

The application discloses a millimeter wave radar-based indoor human target tracking method and system, relates to the technical field of radar positioning, and aims at the problem of low tracking efficiency and accuracy of the human target tracking method in the prior art.The application can make the algorithm cluster according to the proportion of the distance between data instead of a fixed value, and can adaptively adjust the clustering according to the distance between the target and the radar through the regulation of the coefficient.The best clustering mode can be selected through the contour coefficient, so that the accuracy and efficiency of indoor human target detection are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of radar positioning, in particular to an indoor human target tracking method and system based on millimeter wave radar. BACKGROUND

[0002] Radar group target tracking refers to monitoring and tracking the movement of multiple targets in a radar system. Generally, a radar system can detect a large number of indoor human targets, and if these targets are not classified and tracked, it is easy to cause information confusion and target tracking errors, thereby reducing the efficiency and accuracy of the detection system. Clustering algorithms are widely used in radar group target tracking, which can classify and analyze the targets detected by the radar, improve the accuracy and real-time performance of target tracking. Compared with other target tracking algorithms, clustering algorithms can better identify the behavior patterns of multiple targets, better separate data sets, and can guarantee tracking efficiency while minimizing the possibility of misjudgment. Clustering algorithms can effectively group a large number of targets detected by radars and further analyze the behavior patterns and movement rules of each group. This information helps us better understand the movement changes of the targets, thereby achieving more accurate and real-time target tracking. Another important application scenario of clustering algorithms is data preprocessing for target tracking. For example, using clustering technology, redundant information in radar data can be deleted, thereby making subsequent target tracking more accurate. In addition, different grouping strategies can be applied in the target tracking process to apply density-based clustering algorithms, so that targets with different patterns can be tracked and predicted. In an indoor environment, using existing technologies for human target tracking can result in low tracking efficiency and accuracy. SUMMARY

[0003] The purpose of the present application is to solve the problem of low tracking efficiency and accuracy in the prior art human target tracking method, and to propose an indoor human target tracking method and system based on millimeter wave radar.

[0004] The technical solution adopted by the present application to solve the above technical problems is:

[0005] An indoor human target tracking method based on millimeter wave radar, comprising the following steps:

[0006] Step 1: Use millimeter wave radar to obtain the echo time domain raw data of indoor human targets, and reconstruct the raw data to form a matrix with a size of ADCNum*NumChirp*(NTx*NRx), then perform two-dimensional Fourier transform on each channel in the matrix to obtain an RD spectrum matrix;

[0007] Wherein, NumChirp represents the number of chirps, ADCNum represents the number of sampling points contained in each chirp, NTx represents the number of transmitting antennas, and NRx represents the number of receiving antennas;

[0008] Step two: after static clutter filtering and constant false alarm detection on the RD spectrum matrix, the distance of the target point from the radar is obtained;

[0009] Step three: the angle information of the target point is measured through the angle measurement algorithm, and then the horizontal and vertical coordinates of the target point are obtained through the distance of the target point from the radar and the angle information, so as to construct the data set;

[0010] Step four: the reachable distance sequence is obtained by using the horizontal and vertical coordinates of the target point in the data set;

[0011] Step five: each reachable distance in the reachable distance sequence is clustered and divided to obtain all cluster divisions of the target point;

[0012] Step six: the contour coefficients of each cluster division in all cluster divisions of the target point are calculated, and the division with the maximum contour coefficient is selected as the clustering result Result, that is, the state of the target at this moment;

[0013] Step seven: steps two to six are repeated to realize tracking of the indoor human target.

[0014] Further, the specific steps of step four are:

[0015] Step four one: the core distance of all target points in the data set is obtained according to the horizontal and vertical coordinates of the target points, then the target point corresponding to the maximum core distance is selected as the initial observation point, and the remaining target points are all set as unobserved points, and then the initial reachable distance of all unobserved points is set as the core distance of the initial observation point;

[0016] Step four two: the distance value sequence A between the observation point and all unobserved points and the core distance value B of the observation point itself are calculated respectively, each value in the distance value sequence A is compared with B in sequence respectively, and the maximum value in each comparison is reserved in sequence to obtain the M sequence, then each value in the current reachable distance of the unobserved point is compared with the corresponding value in the M sequence, if the value in the current reachable distance of the unobserved point is greater than the corresponding value in the M sequence, the corresponding value in the M sequence is replaced with the corresponding value in the reachable distance of the unobserved point, otherwise, it is not replaced;

[0017] Step four three: find the point with the minimum reachable distance among the unobserved points, and take it as a new observed point, then mark the previous observed point as observed, and no longer compare, repeat step four two until all points are observed, finally record each target point as an observation point, and sort, and construct the reachable distance table with the reachable distance of each target point in the sorting.

[0018] Further, the specific steps of step five are:

[0019] Step five one: traverse the reachable distance table, and the target point with a reachable distance change exceeding a specified coefficient ξ is a mutation point, and each two mutation points represent a division, if the mutation point is one, then the data set has only one division, namely the data set itself, and the division includes a start point and an end point;

[0020] Step five two: sort the divisions obtained through the mutation points according to the start points from small to large, and by default all divisions are not conflicting, and all divisions except the division corresponding to the first start point are unjudged divisions, and then the division corresponding to the first start point is taken as a reference division;

[0021] Step five three: start with the division corresponding to the smallest start point in the unjudged division sorting, and respectively judge whether the division and the reference division coincide from small to large, if they coincide, then the division coinciding with the reference division in the sorting is a conflict division, if they do not coincide, then the division is combined with the reference division to obtain a new reference division,

[0022] Step five four: repeat step five three until the last division in the sorting is judged, and the finally obtained reference division is taken as a clustering division result;

[0023] Step five five: delete the last conflict-causing division from the finally obtained reference division to obtain a new reference division, and take all divisions after the last conflict-causing division as unjudged divisions, and then execute step five three until all divisions are judged, and all clustering division results obtained are final clustering division results.

[0024] Further, the specific steps of step six are:

[0025] Step six: calculate the silhouette coefficients of each clustering division in the target point, select the division with the maximum silhouette coefficient as the clustering result Result, and determine the state of the target at the moment according to the clustering result Result and the radial velocity of the target point relative to the radar, i.e. the state of the target at the moment.

[0026] The radial velocity of the target point relative to the radar is obtained by filtering static clutter and constant false alarm detection on the RD spectrum matrix in step two.

[0027] Further, the profile coefficient is represented as:

[0028]

[0029] Wherein, a i represents the average distance of the ith point to other points in the same cluster, b i represents the minimum average distance of the point to other points in different clusters.

[0030] Further, the minimum number of points in the cluster division is 5.

[0031] A millimeter wave radar-based indoor human target tracking system, comprising: an indoor human target echo time domain raw data acquisition module, a data set construction module, and an indoor human target tracking module;

[0032] The indoor human target echo time domain raw data acquisition module is used to acquire the echo time domain raw data of the indoor human target by using the millimeter wave radar, and reconstruct the raw data to form a matrix, the size of the matrix being ADCNum*NumChirp*(NTx*NRx), then perform two-dimensional Fourier transform on each channel in the matrix respectively to obtain an RD spectrum matrix; wherein, NumChirp represents the number of chirps, ADCNum represents the number of sampling points contained in each chirp, NTx represents the number of transmitting antennas, and NRx represents the number of receiving antennas;

[0033] The data set construction module is used to obtain the distance of the target point from the radar after static clutter filtering and constant false alarm detection of the RD spectrum matrix, and measure the angle information of the target point by an angle measurement algorithm, then obtain the horizontal and vertical coordinates of the target point through the distance of the target point from the radar and the angle information, so as to construct a data set;

[0034] The indoor human target tracking module is used to obtain a reachable distance sequence by using the horizontal and vertical coordinates of the target point in the data set, then cluster divide each reachable distance in the reachable distance sequence to obtain all cluster divisions of the target point, then calculate the profile coefficient of each cluster division in all cluster divisions of the target point, and select the division with the maximum profile coefficient as the cluster result Result, i.e. the state of the target at this moment;

[0035] The indoor human target tracking is realized through the indoor human target echo time domain raw data acquisition module, the data set construction module, and the indoor human target tracking module.

[0036] Further, the specific steps for obtaining the reachable distance sequence by using the horizontal and vertical coordinates of the target point in the data set in the indoor human target tracking module are as follows:

[0037] Step 1: Obtain the core distance of all target points in the data set according to the horizontal and vertical coordinates of the target points, then select the target point corresponding to the maximum core distance as the initial observation point, and set the remaining target points as unobserved points, then set the initial reachable distance of all unobserved points as the core distance of the initial observation point;

[0038] Step 2: respectively obtain the distance value sequence A between the observation point and all unobserved points and the core distance value B of the observation point itself, compare each value in the distance value sequence A with B in sequence, and keep the maximum value in each comparison in sequence to obtain the M sequence, then compare each value in the current reachable distance of the unobserved point with the value at the corresponding position in the M sequence, if the value in the current reachable distance of the unobserved point is greater than the value at the corresponding position in the M sequence, replace the value at the corresponding position in the M sequence with the value in the current reachable distance of the unobserved point, otherwise do not replace;

[0039] Step 3: find the point with the minimum reachable distance among the unobserved points, and take it as a new observation point, then mark the previous observation point as observed, then no longer compare, repeat step 2 until all points are observed, finally record each target point as an observation point and sort them, and construct a reachable distance table with the reachable distance corresponding to each target point in the sorting.

[0040] Further, the indoor human target tracking module clusters each reachable distance in the reachable distance sequence to obtain all cluster divisions of the target points, and the specific steps are as follows:

[0041] Step A: traverse the reachable distance table, wherein the target points with a reachable distance change exceeding a specified coefficient ξ are mutation points, and each two mutation points represent a division, if the mutation point is one, then the data set has only one division, i.e. the data set itself, the division includes a start point and an end point;

[0042] Step B: sort the divisions obtained through the mutation points according to the start points in the divisions from small to large, by default all divisions do not conflict, and all divisions except the division corresponding to the first start point are unjudged divisions, then the division corresponding to the first start point is taken as a reference division;

[0043] Step C: start with the division corresponding to the smallest start point in the unjudged division sorting, from small to large, respectively judge whether the division and the reference division coincide, if they coincide, then the division coinciding with the reference division in the sorting is a conflict division, if they do not coincide, then the division is combined with the reference division to obtain a new reference division,

[0044] Step D: repeat step C until the last division in the sorting is judged, and the finally obtained reference division is taken as a cluster division result;

[0045] Step E: delete the last conflict-causing partition from the last obtained reference partition, obtain a new reference partition, and take all partitions after the last conflict-causing partition as unjudged partitions, then execute step C until all partitions have been judged, and the obtained all clustering partition results are final clustering partition results.

[0046] Further, the contour coefficient is represented as:

[0047]

[0048] Wherein, a i represents the average distance of the ith point to other points in the same cluster, b i represents the minimum average distance of the point to other points in different clusters.

[0049] The beneficial effects of the present application are:

[0050] The present application can make the algorithm cluster according to the proportion of the distance between data instead of a fixed value by defining the coefficient, and the clustering can be adaptively adjusted according to the distance of the target radar; through the contour coefficient, the best clustering method can be selected, and the accuracy and efficiency of indoor human target detection are improved.

[0051] In the case of indoor tracking, the specific number of targets is not known in advance, so the K-means clustering algorithm needs to cluster different target numbers respectively, and then determine the best target number according to additional judgment parameters. In order to reduce the dependence of the K-means algorithm on the initial value, random values can be taken multiple times for clustering, which will make the algorithm more complicated and redundant; hierarchical clustering algorithm will result in too many clustering results and long algorithm running time; in the method based on density clustering, DBSCAN can overcome the disadvantage that the K-means clustering center has a greater impact on the clustering result to a certain extent, but the clustering effect also depends on the given clustering radius. In the indoor environment, the millimeter wave echo points of different distances are different in dispersion degree, and the fixed clustering radius will make it difficult to meet the requirements of indoor tracking. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 The flowchart of the present application;

[0053] Figure 2 The reachable distance diagram;

[0054] Figure 3 The original data diagram;

[0055] Figure 4 The clustering result diagram calculated by the method of the present application;

[0056] Figure 5 A schematic diagram of the clustering result of the existing DBSCAN algorithm is shown in FIG. 1.

[0057] Figure 6 A schematic diagram of step four is shown in FIG. 4. DETAILED DESCRIPTION

[0058] It should be particularly noted that the various embodiments disclosed in the present application can be combined with each other without conflict.

[0059] Embodiment one: reference Figure 1 To specifically describe the present embodiment, the indoor human target tracking method based on a millimeter wave radar according to the present embodiment comprises the following steps:

[0060] Step one: obtaining the echo time-domain original data of the indoor human target by using the millimeter wave radar, re-constructing the original data to form a matrix, the size of the matrix being ADCNum*NumChirp*(NTx*NRx), and then performing two-dimensional Fourier transform on each channel in the matrix to obtain an RD spectrum matrix;

[0061] Wherein, NumChirp represents the number of chirps, ADCNum represents the number of sampling points contained in each chirp, NTx represents the number of transmitting antennas, and NRx represents the number of receiving antennas.

[0062] Step two: after performing static clutter filtering and constant false alarm detection on the RD spectrum matrix, the distance of the target point from the radar is obtained.

[0063] Step three: measuring the angle information of the target point by using an angle measurement algorithm, and then combining the distance of the target point from the radar and the angle information to obtain the horizontal and vertical coordinates of the target point, so as to construct a data set.

[0064] Step four: obtaining a reachable distance sequence by using the horizontal and vertical coordinates of the target points in the data set.

[0065] Step five: clustering and dividing each reachable distance in the reachable distance sequence to obtain all clustering divisions of the target points.

[0066] Step six: calculating the contour coefficient S of each clustering division in all clustering divisions of the target points, and selecting the division with the largest contour coefficient as the clustering result Result, i.e., the state of the target at this moment.

[0067] Step seven: repeating steps two to six to realize tracking of the indoor human target.

[0068] Clustering algorithm is a common machine learning technique that can divide a dataset into several categories, each containing a group of highly similar samples. In radar group target tracking, clustering algorithms are mainly used to distinguish the motion rules and behavior patterns of different categories of targets. The most commonly used clustering algorithms include K-means clustering algorithm, hierarchical clustering algorithm, and density clustering algorithm. Each of these algorithms has its own advantages and application scope.

[0069] K-means clustering algorithm is a center-based clustering algorithm. In this algorithm, the parameter k is first input, representing the division of the dataset into k different categories. Then, k samples are randomly selected from the dataset as the centers of the k categories. Next, all samples in the dataset are assigned to the k different categories according to the distance between the samples in the dataset and the k samples. The sum of the distances between the samples assigned to the same category is the smallest, and the assignment result satisfies the minimum distance principle. Finally, the center of each category is recalculated based on the assignment result. This process is repeated until the center no longer changes or other stopping conditions are met.

[0070] Hierarchical clustering algorithm is a top-down or bottom-up clustering method. In this algorithm, for a dataset, it is first taken as a unique cluster, then each cluster is successively split into several clusters, and this process is repeated until the predetermined conditions are met. The distance between samples in the hierarchical clustering process can have different definitions, such as single connectivity, full connectivity, and average connectivity. Among them, single connectivity refers to the shortest distance between two clusters, full connectivity refers to the longest distance between two clusters, and average connectivity refers to the average distance between all samples in two clusters.

[0071] Density clustering algorithm is a density-based clustering algorithm. In this algorithm, a cluster is defined as a dense region of sample points, and the distance between clusters is defined as the distance between sample points in the clusters. By calculating the density of sample points, the dataset can be divided into several clusters. Among them, the density of a sample point represents the number of samples contained in a circle with that point as the center and a certain radius as the radius. In the density clustering algorithm, two parameters are used when clustering and discriminating clusters, namely radius and density threshold. Density clustering algorithm can overcome the strong dependence on initial values in K-means algorithm, so it is widely used in target tracking field.

[0072] In indoor human tracking scenarios, since the exact number of targets is unknown beforehand, K-means clustering requires separate clustering for different target numbers, followed by determining the optimal target count based on additional parameters. To reduce the K-means algorithm's dependence on initial values, multiple random values ​​can be used for clustering, making the algorithm cumbersome and redundant. Hierarchical clustering algorithms result in numerous clustering results and long runtimes. Among density-based clustering methods, DBSCAN can overcome the significant influence of K-means cluster centers on the clustering results to some extent, but its clustering effectiveness still depends on a pre-defined cluster radius. In indoor environments, the dispersion of millimeter-wave echo points at different distances varies, and a fixed cluster radius may result in clustering results that fail to meet indoor tracking requirements. This application addresses these shortcomings of clustering algorithms.

[0073] The technical solution of this application specifically includes three stages: core distance and reachability distance calculation, clustering partitioning, and silhouette coefficient calculation.

[0074] Phase 1: Calculation of Core Distance and Reachable Distance

[0075] Suppose we have a dataset D that needs to be clustered, with the following parameters and conditions: each cluster contains at least MinPts points, and the distance is Euclidean distance. The neighborhood of a point q is represented by N. ε (q), where ε represents the neighborhood radius.

[0076] The core distance of a data point q is defined as the minimum neighborhood radius ε0 that makes its neighborhood contain MinPts points. Data points with smaller core distances are usually denser and relatively concentrated. For each data point q belonging to dataset D, we only need to calculate the distances of the other points, sort the distances in ascending order, and take the smallest MinPts distance as the core distance for that point.

[0077] Reachability distance is the distance between a data point p and another data point q, and it also needs to consider the core distance of q and the core distance of p. If there is a connected path from p to q, and the maximum core distance of all data points on the path is not higher than the core distance of q, then the reachability distance between p and q is the core distance of q.

[0078] The reachable distance needs to be calculated according to the following steps:

[0079] (1) Find the point with the maximum core distance among all points in dataset D as the observation point, which means that this point is the most likely noise point; and set the initial reachable distance of all points to the maximum value;

[0080] (2) Set the reachable distance of the points included in the core region of the observation point as the core distance of the observation point, and set the reachable distance of the points not in the core region as the distance between the point and the observation point; this distance is denoted as m;

[0081] (3) Find the point in the core region whose reachable distance is greater than m, and set it as m;

[0082] (4) Find the point with the smallest current reachable distance among the remaining points as a new observation point, and repeat steps (2) to (4) until the remaining points are zero in step 4, and end. It should be noted that during the processing, the points serving as observation points each time need to be recorded and sorted, and a reachable distance table is constructed with the reachable distance corresponding to each point.

[0083] Phase two, cluster division

[0084] In phase one, the reachable distance and its corresponding order are obtained. Here, first of all, it is clear that from the calculation method of the reachable distance, the reachable distance corresponding to each point represents the close relationship between the point and its previous point, so the smaller the reachable distance, the closer the point to the previous point, and vice versa. According to this characteristic, it can be known that there will be a region of abrupt change in reachable distance between two clusters, and the region between the two abrupt change regions is a cluster.

[0085] In the cluster division phase, the coefficient ξ needs to be specified, which determines under what circumstances a region can be defined as "abrupt change". After giving the value of ξ according to the actual situation, all the abrupt points are found by traversing the reachable distance table, and each two are combined to obtain a cluster. The clusters obtained by this method are usually overlapping, so further processing is needed for cluster division. The specific method is as follows:

[0086] (1) Set the specified coefficient ξ, and the points whose reachable distance changes more than the specified coefficient ξ are the abrupt points, and each two abrupt points represent a division. If the abrupt point is one, then the data set has only one division, i.e. the data set itself.

[0087] The division obtained by "abrupt change" includes the starting point and the ending point, and is sorted in ascending order according to the starting point. By default, all divisions do not conflict, and all divisions except the division corresponding to the first starting point are considered as unjudged divisions. The division corresponding to the first starting point is taken as the reference division,

[0088] (2) Starting from the smallest starting point in the un-judged partition, judge whether it coincides with the reference partition from small to large, if it coincides, the partition in the order coinciding with the reference partition is taken as the conflict partition, if it does not coincide, it is merged with the reference partition to form a new reference partition, and the above steps are repeated until the last partition in the order is judged. The last reference partition obtained is taken as a clustering partition result.

[0089] The last conflict-causing partition is deleted from the last reference partition to obtain a new reference partition, and all partitions conflicting with the last conflict-causing partition are taken as un-judged partitions, and the above steps are repeated until all partitions have been judged, and all clustering partition results obtained are final clustering partition results.

[0090] Stage three, contour coefficient calculation

[0091] In stage two, different clustering methods of a data set will be obtained, and the most suitable clustering method is selected according to the contour coefficients of different clustering.

[0092] The contour coefficient (silhouette value) of a point in a cluster is defined as

[0093]

[0094] Where a i represents the average distance from the ith point to other points in the same cluster;

[0095] b i represents the minimum average distance from the point to other points in different clusters.

[0096] The contour coefficients of all points in the cluster are calculated, and the average value is taken as the contour coefficient of the entire clustering method. In this way, the contour coefficients of different clustering partition methods can be obtained, and the clustering method with the largest contour coefficient is the most suitable clustering method for the data set.

[0097] Taking a frame of actual radar data as an example, it is known that there are three targets in the current frame, as shown in Figure 2 MinPts=5 is set,

[0098] Specific process

[0099] (1) The reachable distance table of all points is calculated, as shown in Figure 3 .

[0100] (2) According to the mutation value, small block clusters are obtained, which are [13, 19], [13, 100], [34, 98], [46, 98], [57, 62],

[0101] [81,100], [106,120], [106,208], [106,218], [123,131], [208,218], [232,253];

[0102] (3) The small block clusters are merged to obtain different classification results as follows:

[0103] {[13, 19], [34, 98], [106, 120], [123, 131], [208, 218], [232, 253]};

[0104] {[13, 19], [34, 98], [106, 208], [208, 218], [232, 253]};

[0105] {[13, 19], [34, 98], [106, 218], [232, 253]};

[0106] {[13, 19], [34, 98], [123, 131], [208, 218], [232, 253]};

[0107] {[13, 19], [46, 98], [106, 120], [123, 131], [208, 218], [232, 253]};

[0108] {[13, 19], [57, 62], [81, 100], [106, 120], [123, 131], [208, 218], [232, 253]};

[0109] {[13, 100], [106, 120], [123, 131], [208, 218], [232, 253]};

[0110] {[13, 100], [106, 208], [208, 218], [232, 253]};

[0111] {[13, 100], [106, 218], [232, 253]};

[0112] {[13, 100], [123, 131], [208, 218], [232, 253]};

[0113] (4) The contour coefficients of these small block clusters are solved respectively, and the cluster corresponding to the largest contour coefficient is taken out as

[0114] {[13, 100], [106, 218], [232, 253]};

[0115] (5) Final clustering is {[13, 100], [106, 218], [232, 253]};

[0116] Effect comparison: Compared with the DBSCAN clustering method, the application can consider multiple clustering methods to obtain the best performance.

[0117] Suppose the radar transmits FMCW frequency-modulated continuous wave, a total of many frames are transmitted during the entire measurement process, for each frame, its parameters are as follows: a total of NumChirp chirps are contained, each chirp has ADCNum sampling points, a total of NTx transmitting antennas and NRx receiving antennas are contained. In the clustering process, the minimum number of points in a cluster is set to MinPts, and the threshold coefficient is ξ.

[0118] It should be noted that the specific embodiments are only an explanation and description of the technical solutions of the present application, and cannot limit the scope of protection. Any partial change made according to the claims and description of the present application shall still fall within the protection scope of the present application.

Claims

1. A method of tracking an indoor human target based on millimeter wave radar, characterized by The method comprises the following steps: Step one: obtaining echo time-domain original data of an indoor human target by using a millimeter wave radar, reconstructing the original data to form a matrix, the size of the matrix being ADCNum*NumChirp*(NTx*NRx), and then performing two-dimensional Fourier transform on each channel in the matrix to obtain an RD spectrum matrix; wherein NumChirp represents the number of chirps, ADCNum represents the number of sampling points contained in each chirp, NTx represents the number of transmitting antennas, and NRx represents the number of receiving antennas; Step two: obtaining the distance of a target point from the radar after static clutter filtering and constant false alarm detection on the RD spectrum matrix; Step three: measuring the angle information of the target point by using an angle measurement algorithm, and then obtaining the horizontal and vertical coordinates of the target point by using the distance of the target point from the radar and the angle information, so as to construct a data set; Step four: obtaining a reachable distance sequence by using the horizontal and vertical coordinates of the target points in the data set; Step five: clustering and dividing each reachable distance in the reachable distance sequence to obtain all cluster divisions of the target point; Step six: calculating the contour coefficients of each cluster division in all cluster divisions of the target point, and selecting the division with the maximum contour coefficient as the cluster result Result, that is, the state of the target at the current moment; Step seven: repeating steps two to six to realize tracking of the indoor human target. 2.The method of claim 1, wherein The specific steps of step four are as follows: Step four one: obtaining the core distance of all target points in the data set according to the horizontal and vertical coordinates of the target points, then selecting the target point corresponding to the maximum core distance as an initial observation point, and setting the remaining target points as unobserved points, and then setting the initial reachable distance of all unobserved points as the core distance of the initial observation point; Step four two: respectively obtaining a distance value sequence A between the observation point and all unobserved points and a core distance value B of the observation point itself, comparing each value in the distance value sequence A with B in sequence, and retaining the maximum value in each comparison in sequence to obtain an M sequence, and then comparing each value in the current reachable distance of the unobserved point with the value at the corresponding position in the M sequence, if the value in the current reachable distance of the unobserved point is greater than the value at the corresponding position in the M sequence, replacing the value at the corresponding position in the M sequence with the value at the corresponding position in the reachable distance of the unobserved point, otherwise, not replacing; Step four three: finding the point with the minimum reachable distance among the unobserved points, and taking it as a new observation point, then marking the previous observation point as observed, and then no longer comparing, repeating step four two until all points are observed, finally recording the target points serving as observation points each time and sorting them, and constructing a reachable distance table by using the reachable distances corresponding to each target point in the sorting. 3.The method of claim 2, wherein The specific steps of step five are as follows: Step five: traverse the reachable distance table, wherein the reachable distance changes more than a specified coefficient The target point, i.e. the mutation point, and every two mutation points represent a division. If the mutation point is one, the data set has only one division, i.e. the data set itself. The division includes the start point and the end point. Step five two: sorting the divisions obtained through the mutation points according to the starting points in the divisions from small to large, defaulting that all divisions are not in conflict, and taking all divisions except the division corresponding to the first starting point as unjudged divisions, and then taking the division corresponding to the first starting point as a reference division; Step five three: starting from the partition corresponding to the smallest starting point in the un-judged partition in the order, judging whether the partition and the reference partition coincide from small to large, if coinciding, taking the partition coinciding with the reference partition in the order as the conflict partition, if not coinciding, merging the partition and the reference partition to obtain a new reference partition, Step five four: repeating step five three until the last partition in the order is judged, taking the last obtained reference partition as a clustering partition result; Step five five: deleting the last conflict-causing partition from the last obtained reference partition to obtain a new reference partition, taking all the partitions after the last conflict-causing partition as un-judged partitions, and then executing step five three until all the partitions are judged, and all the clustering partition results obtained are final clustering partition results. 4.The method of claim 1, wherein The specific steps of the step six are: Step six: calculating the contour coefficients of each clustering partition in all the clustering partitions of the target point, selecting the partition with the largest contour coefficient as a clustering result Result, and determining the state of the target at the moment according to the clustering result Result and the radial velocity of the target point relative to the radar, i.e., the state of the target at the moment; The radial velocity of the target point relative to the radar is obtained by static clutter filtering and constant false alarm rate detection on the RD spectrum matrix in step two.

5. The method according to claim 3 or 4, characterized in that The contour coefficient is expressed as: wherein, represents the distance of the point to the other points in the same cluster, represents the average distance of the point to the other points in the same cluster, represents the minimum average distance of the point to the other points in different clusters.

6. The method of claim 5, wherein The minimum number of points in the clustering partition is 5.

7. A millimeter wave radar-based indoor human target tracking system, characterized by It comprises: An indoor human target echo time domain original data acquisition module, a data set construction module, and an indoor human target tracking module; The indoor human target echo time domain original data acquisition module is used to acquire the echo time domain original data of the indoor human target by using a millimeter wave radar, reconstruct the original data to form a matrix, the size of the matrix is ADCNum*NumChirp*(NTx*NRx), then perform two-dimensional Fourier transform on each channel in the matrix to obtain an RD spectrum matrix; wherein NumChirp represents the number of chirps, ADCNum represents the number of sampling points contained in each chirp, NTx represents the number of transmitting antennas, and NRx represents the number of receiving antennas; The data set construction module is used to obtain the distance of the target point from the radar after static clutter filtering and constant false alarm rate detection on the RD spectrum matrix, measure the angle information of the target point by an angle measurement algorithm, and then obtain the horizontal and vertical coordinates of the target point by the distance of the target point from the radar and the angle information, so as to construct a data set; The indoor human target tracking module is used to obtain an accessible distance sequence by using the horizontal and vertical coordinates of the target point in the data set, then cluster each accessible distance in the accessible distance sequence to obtain all the clustering partitions of the target point, then calculate the contour coefficients of each clustering partition in all the clustering partitions of the target point, and select the partition with the largest contour coefficient as a clustering result Result, i.e., the state of the target at the moment; The indoor human target tracking module is used to obtain an accessible distance sequence by using the horizontal and vertical coordinates of the target point in the data set, then cluster each accessible distance in the accessible distance sequence to obtain all the clustering partitions of the target point, then calculate the contour coefficients of each clustering partition in all the clustering partitions of the target point, and select the partition with the largest contour coefficient as a clustering result Result, i.e., the state of the target at the moment; The indoor human target tracking module is used to obtain an accessible distance sequence by using the horizontal and vertical coordinates of the target point in the data set, then cluster each accessible distance in the accessible distance sequence to obtain all the clustering partitions of the target point, then calculate the contour coefficients of each clustering partition in all the clustering partitions of the target point, and select the partition with the largest contour coefficient as a clustering result Result, i.e., the state of the target at the moment; 8. The indoor human target tracking system based on millimeter wave radar according to claim 7, characterized in that The specific steps for obtaining the reachable distance sequence in the indoor human target tracking module by using the horizontal and vertical coordinates of the target points in the data set are as follows: Step 1: Obtain the core distance of all target points in the data set according to the horizontal and vertical coordinates of the target points, then select the target point corresponding to the maximum core distance as the initial observation point, and set the remaining target points as unobserved points, and then set the initial reachable distance of all unobserved points as the core distance of the initial observation point; Step 2: respectively obtain the distance value sequence A between the observation point and all unobserved points and the core distance value B of the observation point itself, compare each value in the distance value sequence A with B in sequence, and reserve the maximum value in each comparison in sequence to obtain the M sequence, then compare each value in the current reachable distance of the unobserved point with the corresponding value in the M sequence, if the value in the current reachable distance of the unobserved point is greater than the corresponding value in the M sequence, replace the corresponding value in the M sequence with the corresponding value in the reachable distance of the unobserved point, otherwise do not replace; Step 3: find the point with the minimum reachable distance among the unobserved points, and take it as a new observation point, then mark the previous observation point as observed, and then no longer compare, repeat step 2 until all points are observed, finally record the target points as observation points each time and sort them, and construct the reachable distance table with the reachable distance corresponding to each target point in the sorting.

9. The indoor human target tracking system based on millimeter wave radar according to claim 8, characterized in that The specific steps for clustering and dividing each reachable distance in the reachable distance sequence to obtain all cluster divisions of the target points in the indoor human target tracking module are as follows: Step A: Traverse the reachable distance table, where the reachable distance changes more than a specified coefficient The target point, i.e. the mutation point, and every two mutation points represent a division. If the mutation point is one, the data set has only one division, i.e. the data set itself. The division includes the start point and the end point. Step B: sort the divisions obtained through the mutation points according to the starting points in the divisions from small to large, default all divisions are not in conflict, and all divisions except the division corresponding to the first starting point are taken as unjudged divisions, then the division corresponding to the first starting point is taken as the reference division; Step C: start with the division corresponding to the smallest starting point in the unjudged division sorting, from small to large, respectively judge whether the division and the reference division coincide, if they coincide, the divisions coinciding with the reference division in the sorting are taken as conflict divisions, if they do not coincide, the division is combined with the reference division to obtain a new reference division, Step D: repeat step C until the last division in the sorting is judged, and the last obtained reference division is taken as a cluster division result; Step E: delete the last conflict-causing division from the last obtained reference division to obtain a new reference division, and take all divisions after the last conflict-causing division as unjudged divisions, then execute step C until all divisions are judged, and the obtained all cluster division results are the final cluster division results.

10. The indoor human target tracking system based on millimeter wave radar according to claim 9, characterized in that The contour coefficient is represented as: wherein, represents the distance from the point to other points in the same cluster, represents the average distance from the point to other points in the same cluster, represents the minimum average distance from the point to other points in different clusters.

Images