Indoor people counting, positioning method based on millimeter wave radar point cloud data

Abstract

The application discloses an indoor personnel counting and positioning method based on millimeter wave radar point cloud data, and a processing procedure of point cloud data output by a millimeter wave radar comprises the following steps: preprocessing the point cloud data; superimposing the processed point cloud and a target at a previous moment on a two-dimensional plane, carrying out density clustering, associating the detection points with the target, reclassifying the detection points in a cluster containing multiple targets according to the distance between the detection points and the target in the cluster; updating the position of the target at the current moment; managing the target state, and counting the number and position of the personnel. The method clusters after superimposing the point cloud and the target to determine the relationship between the detection points and the target, so that the detection points associated with the target can be of any shape, improving the accuracy of data association; in addition, the reclassification of the cluster containing multiple targets reduces the data association error rate of the target under the condition that the distance is relatively short.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of personnel detection, and particularly to an indoor personnel counting and positioning method based on millimeter wave radar. BACKGROUND

[0002] Generally, the operation state of lighting, heating, ventilation, air conditioning and other devices in a room is mainly controlled by human, which is affected by the subjectivity of personnel judgment and their living habits. The human control method will cause great waste of resources and is not conducive to energy saving and emission reduction. The solution to the above problems is to intelligently control these devices according to the personnel information in the room, and to reduce unnecessary energy consumption and improve energy utilization efficiency through demand driving. This control scheme requires the control system to be able to perceive the state of the indoor personnel, such as the number and position of the personnel. The current relatively mature scheme is a camera, but it will lead to privacy leakage and is difficult to popularize and promote in indoor office areas. The newly emerging millimeter wave radar technology does not involve privacy issues and has comparable detection accuracy to the camera scheme, and has good application prospects.

[0003] In the current indoor personnel information detection method based on millimeter wave radar, the classic multi-target tracking algorithm is mainly used. In the prior patent CN2022102778411 "Indoor personnel information detection method based on millimeter wave radar point cloud data" applied by the present applicant, it is mentioned in the background art of the patent "Radar detection device motion and stationary target continuous tracking correlation method", patent application number: CN201811151782.3, when the detection point cloud disappears, the position, speed and acceleration information before the disappearance of the associated point cloud is predicted according to the inertia theorem to compensate for the missing point trace, so that the trajectory is continuous. The disadvantage of this method is that when the point cloud disappears due to the target being stationary, the predicted position will gradually deviate from the actual position of the target, resulting in a wrong trajectory. When the stationary target moves again, a new trajectory will be generated, causing false alarm. In the patent "Personnel detection tracking and counting algorithm based on millimeter wave radar", patent application number: CN202010243717.4, and the patent "Personnel counting and tracking method based on millimeter wave radar", patent application number: CN202110108936.6, the disappearance of the detection point cloud is determined to be caused by the target being stationary, and the last detected position before the disappearance of the point cloud is retained as the position of the target when it is stationary. If the disappearance of the detection point cloud is not caused by the target being stationary, such as the target moving to the radar detection blind area or the direction of the target motion being perpendicular to the straight line direction of the target and the radar, the method may still misjudge. When the detection point cloud appears again, it may not be associated with the target determined to be stationary, and a new target will be generated, causing false alarm. The disadvantage of using the multi-target tracking algorithm is that the indoor personnel has multiple motion states and belongs to non-maneuverable targets. The commonly used maneuverable target motion model is difficult to meet the objective situation. In addition, when the personnel is stationary, the point cloud will disappear, that is, the radar observation of the target is missing. If the target position is predicted for a long time according to the model, the predicted position will gradually deviate from the actual position of the target, resulting in a wrong trajectory. In addition, when the personnel are close to each other, the accuracy of the data association based on the wave door will be reduced, affecting the effect of target tracking. SUMMARY

[0004] The purpose of the present application is to provide an indoor personnel counting and positioning method based on millimeter wave radar point cloud data. By processing each frame of point cloud data output by the millimeter wave radar, the number and position information of the personnel are obtained, solving the deficiencies of the prior art.

[0005] In order to achieve the above-mentioned purpose, the indoor personnel counting and positioning method based on millimeter wave radar point cloud data of the present application, the processing process of the point cloud data output by the millimeter wave radar, includes the following steps:

[0006] S1. Preprocess the 3D point cloud data, including converting the point cloud positions from the spherical coordinate system to the rectangular coordinate system, and filtering the detection points in the specified area;

[0007] S2. Overlay the current point cloud with the target from the previous time step, and then perform density clustering to associate the detection points with the targets. At the same time, for clusters containing multiple targets, classify them again according to the distance between the detection points and the targets, and assign the detection points to the targets with the closest distance.

[0008] S3. Use the center position of the detection point associated with the target as the radar's observation of the target's position, and update the target's current position.

[0009] S4. Manage the status of targets based on the association of detection points, and count the number of people and location information.

[0010] In a preferred embodiment, the specific steps of the present invention are as follows:

[0011] S1. Preprocess the 3D point cloud data, including the following sub-steps:

[0012] S101, point cloud data information includes distance r Azimuth φ and pitch angle θ ;

[0013] S102. Transform the position representation of the point cloud from spherical coordinates to Cartesian coordinates, ignoring the height information of the point cloud. For each detection point... express, ,in and They represent the first i The coordinates of each detection point in the x and y directions.

[0014]

[0015] S103. Let the boundary of the specified detection area in the x-direction be... and The boundary in the y-direction is and The number of screening points is n The filtered point cloud set is then:

[0016]

[0017] S104, End;

[0018] S2. Associate the detection points with the target, including the following sub-steps:

[0019] S201, with Indicates the first iOne goal, among which and They represent k The coordinates of the target in the x and y directions at any given time, assuming k -1 time has a total of m If there are n objectives, then the set of these objectives is denoted as . R ,

[0020]

[0021] Will k -1 time m The target and the selected at the current moment n The set of multiple detection points superimposed together is denoted as set. G ,

[0022]

[0023] S202, Use density clustering algorithm to process the set G Cluster the objects in the data and use... Record number i The sequence number of the detection point in each cluster. Record number i The target index in each cluster;

[0024] S203, if The number of elements in the middle is 0, that is, the first... i If no cluster contains the target, then a new target is generated, let it be the i-th cluster. j One goal, to The corresponding detection points are associated with the new target, using Record;

[0025] S204, if The number of elements in the middle is 1, that is, the first... i The cluster has only one target, let's assume it is the i-th cluster. j One goal, to The corresponding detection point is associated with the target, using Record;

[0026] S205, if If the number of elements in the middle is greater than 1, that is, if it contains multiple targets, calculate... The detection points in The distance to all targets in the range is used to associate the detection point with the target closest to it. The sequence number of the detection point associated with each target is denoted by . Record;

[0027] S206, End;

[0028] S3. Update the target's current position, including the following sub-steps:

[0029] S301. For a target that existed in the previous time step, if the number of detection points associated with the target is 0, then the target position remains unchanged.

[0030]

[0031] If the number of detection points associated with the target is greater than 0, then the center position of the detection points associated with the target is used as the radar's observation of the target's position. The weighted average of the target's position at the previous moment and the observed position at the current moment is calculated as the target's current position.

[0032]

[0033] Where α is the weighting coefficient and 0 < α < 1, for The number of elements in the middle, that is, the number associated with the first element. i The number of detection points for each target;

[0034] S302. For a newly generated target at the current moment, calculate the center position of the detection points associated with the target, and use this position as the current position of the new target.

[0035]

[0036] S303, End;

[0037] S4. Update the target status and collect personnel information, including the following sub-steps:

[0038] S401. The target state is divided into detection, activation and release. In step 2, the cluster that does not contain a target is considered to be a new target and is initialized to the detection state.

[0039] S402, Use Record the number of times the target in the detection state is detected by the radar within frame T1. Not an empty set, Add 1, if It is an empty set. Subtract 1 when Greater than the set threshold When the target transitions from the detection state to the active state, When the value is 0, the target transitions from the probe state to the release state, meaning the target is discarded.

[0040] S403, Use Record the number of times the active target is detected within frame T2. Not an empty set, Add 1, if It is an empty set. Subtract 1 when When the value is 0, the target changes from the active state to the released state, that is, the target is discarded;

[0041] S404. Count the number of targets in the active state and their locations, and output the detection results;

[0042] S405, End.

[0043] In the preferred embodiment, during step S2, the height information of the detection points is ignored and they are superimposed on the target on a two-dimensional plane; and a density clustering algorithm is used to associate the detection points with the target.

[0044] In addition, when counting people in S4, only targets in the active state are counted.

[0045] The beneficial effects of this invention are that, through the aforementioned indoor personnel counting and positioning method based on millimeter-wave radar point cloud data, the number and location of personnel indoors can be measured more accurately. Considering the characteristics of slow movement and continuous spatial changes of personnel indoors, this invention proposes clustering point clouds and targets on a plane, thereby associating detection points with targets. This solves the problem of gate shape limitations in gate-based data association methods, improving the accuracy of data association. During data association, for clusters containing multiple targets, they are further classified according to the distance between the point cloud and the targets, reducing the incidence of association errors such as target overlap and target loss when multiple targets are close together. The status of targets is managed based on the data association status, and an activation confirmation process is set for personnel targets to filter out interference caused by radar clutter. Confirmed targets are maintained to prevent stationary targets from being missed, thus improving the accuracy of personnel detection. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating the steps of an embodiment of the method of the present invention.

[0047] Figure 2 This is a geometrical diagram of the detection point cloud in an embodiment of the method of the present invention.

[0048] Figure 3 This is a schematic diagram of clustering after the detection points and the target are superimposed in an embodiment of the method of the present invention.

[0049] Figure 4 This is a schematic diagram illustrating the reclassification of a cluster containing multiple targets in an embodiment of the method of the present invention.

[0050] Figure 5 This is a target state transition diagram in an embodiment of the method of the present invention. Detailed Implementation

[0051] The technical solution of the present invention will now be described in detail with reference to the embodiments and accompanying drawings.

[0052] The indoor people counting and positioning method based on millimeter-wave radar point cloud data described in this invention includes the following steps:

[0053] Step 1: Preprocess the 3D point cloud data, including the following sub-steps:

[0054] (1) Point cloud data information includes distance ( r ), azimuth ( φ ) and pitch angle ( θ );

[0055] (2) Transform the position representation of the point cloud from spherical coordinates to Cartesian coordinates, ignoring the height information of the point cloud, and assign each detection point a positional coordinate system. express, ,in and They represent the first i The coordinates of each detection point in the x and y directions.

[0056]

[0057] (3) Let the boundary of the specified detection area in the x-direction be... and The boundary in the y-direction is and The number of screening points is n The filtered point cloud set is then:

[0058]

[0059] (4) End;

[0060] Step 2: Associate the detection points with the target, including the following sub-steps:

[0061] (1) with Indicates the first i One goal, among which and They represent k The coordinates of the target in the x and y directions at any given time, assuming k -1 time has a total of m If there are n objectives, then the set of these objectives is denoted as . R ,

[0062]

[0063] Will k -1 time m The target and the selected at the current moment n The set of multiple detection points superimposed together is denoted as set. G ,

[0064]

[0065] (2) Use density clustering algorithm to cluster the set G Cluster the objects in the data and use... Record number i The sequence number of the detection point in each cluster. Record number i The target index in each cluster;

[0066] (3) If The number of elements in the middle is 0, that is, the first... i If no cluster contains the target, then a new target is generated, let it be the i-th cluster. j One goal, to The corresponding detection points are associated with the new target, using Record;

[0067] (4) If The number of elements in the middle is 1, that is, the first... i The cluster has only one target, let's assume it is the i-th cluster. j One goal, to The corresponding detection point is associated with the target, using Record;

[0068] (5) If If the number of elements in the middle is greater than 1, that is, if it contains multiple targets, calculate... The detection points in The distance to all targets in the range is used to associate the detection point with the target closest to it. The sequence number of the detection point associated with each target is denoted by . Record;

[0069] (6) End;

[0070] Step 3: Update the target's current position, including the following sub-steps:

[0071] (1) For a target that already existed in the previous time step, if the number of detection points associated with the target is 0, then the target position remains unchanged, i.e.

[0072]

[0073] If the number of detection points associated with the target is greater than 0, then the center position of the detection points associated with the target is used as the radar's observation of the target's position. The weighted average of the target's position at the previous moment and the observed position at the current moment is calculated as the target's current position.

[0074]

[0075] Where α is the weighting coefficient and 0 < α < 1, for The number of elements in the middle, that is, the number associated with the first element. i The number of detection points for each target;

[0076] (2) For a newly generated target at the current moment, calculate the center position of the detection points associated with the target, and use this position as the current position of the new target.

[0077]

[0078] (3) End;

[0079] Step 4: Update the target status and compile personnel information, including the following sub-steps:

[0080] (1) The target state is divided into detection, activation and release. In step 2, the cluster that does not contain a target is considered to be a new target and is initialized to the detection state.

[0081] (2) Use Record the number of times the target in the detection state is detected by the radar within frame T1. Not an empty set, Add 1, if It is an empty set. Subtract 1 when Greater than the set threshold When the target transitions from the detection state to the active state, When the value is 0, the target transitions from the probe state to the release state, meaning the target is discarded.

[0082] (3) Use Record the number of times the active target is detected within frame T2. Not an empty set, Add 1, if It is an empty set. Subtract 1 when When the value is 0, the target changes from the active state to the released state, that is, the target is discarded;

[0083] (4) Count the number of targets in the active state and their locations, and output the detection results;

[0084] (5) End.

[0085] The flowchart of the indoor people counting and positioning method based on millimeter-wave radar point cloud data in this embodiment is shown below. Figure 1After receiving point cloud data, the system first performs coordinate transformation and point cloud filtering; then it overlays the filtered point cloud with the target from the previous time step on a two-dimensional plane, and associates the detection points with the target through density clustering. If a cluster contains multiple targets, the detection points in the cluster need to be reclassified; finally, it updates the target position, manages the target state, counts the number and position of active targets, and outputs the detection results. Figure 2 The figure shown is a geometric diagram of point cloud data. Considering that people move in the plane indoors, the target height information is ignored in the embodiment, and target tracking is performed in a two-dimensional plane. Figure 3 The diagram shows a clustering of point clouds and targets after they are superimposed on a plane, with the same cluster enclosed in an ellipse. Figure 4 This diagram illustrates cluster reclassification involving multiple targets, where detection points are assigned to the nearest target. Targets have three states: detected, activated, and released. See the state transition diagram below. Figure 5 A new target is initially in the detection state. When the number of detections reaches a threshold, the target is transferred to the active state. If the number of detections of the target within a specified frame is 0, it is transferred to the release state, that is, the target is discarded.

Claims

1. A method for indoor personnel counting and positioning based on millimeter-wave radar point cloud data, characterized in that, The processing of point cloud data output by millimeter-wave radar includes the following steps: S1. Preprocess the 3D point cloud data, including converting the point cloud positions from the spherical coordinate system to the rectangular coordinate system, and filtering the detection points in the specified area; S2. Overlay the current point cloud with the target from the previous time step, and then perform density clustering to associate the detection points with the targets. At the same time, for clusters containing multiple targets, classify them again according to the distance between the detection points and the targets, and assign the detection points to the targets with the closest distance. S3. Use the center position of the detection point associated with the target as the radar's observation of the target's position, and update the target's current position. S4. Manage the status of targets based on the association of detection points, and count the number of people and location information.

2. The indoor personnel counting and positioning method based on millimeter-wave radar point cloud data according to claim 1, characterized in that, Includes the following steps: S1. Preprocess the 3D point cloud data, including the following sub-steps: S101, point cloud data information includes distance r Azimuth φ and pitch angle θ ; S102. Transform the position representation of the point cloud from spherical coordinates to Cartesian coordinates, ignoring the height information of the point cloud. For each detection point... express, ,in and They represent the first i The coordinates of each detection point in the x and y directions. S103. Let the boundary of the specified detection area in the x-direction be... and The boundary in the y-direction is and The number of screening points is n The filtered point cloud set is then: S104, End; S2. Associate the detection points with the target, including the following sub-steps: S201, with Indicates the first i One goal, among which and They represent k The coordinates of the target in the x and y directions at any given time, assuming k -1 time has a total of m If there are n objectives, then the set of these objectives is denoted as . R , Will k -1 time m The target and the selected at the current moment n The set of multiple detection points superimposed together is denoted as set. G , S202, Use density clustering algorithm to process the set G Cluster the objects in the data and use... Record number i The sequence number of the detection point in each cluster. Record number i The target index in each cluster; S203, if The number of elements in the middle is 0, that is, the first... i If no cluster contains the target, then a new target is generated, let it be the i-th cluster. j One goal, to The corresponding detection points are associated with the new target, using Record; S204, if The number of elements in the middle is 1, that is, the first... i The cluster has only one target, let's assume it is the i-th cluster. j One goal, to The corresponding detection point is associated with the target, using Record; S205, if If the number of elements in the middle is greater than 1, that is, if it contains multiple targets, calculate... The detection points in The distance to all targets in the range is used to associate the detection point with the target closest to it. The sequence number of the detection point associated with each target is denoted by . Record; S206, End; S3. Update the target's current position, including the following sub-steps: S301. For a target that existed in the previous time step, if the number of detection points associated with the target is 0, then the target position remains unchanged. If the number of detection points associated with the target is greater than 0, then the center position of the detection points associated with the target is used as the radar's observation of the target's position. The weighted average of the target's position at the previous moment and the observed position at the current moment is calculated as the target's current position. Where α is the weighting coefficient and 0 < α < 1, for The number of elements in the middle, that is, the number associated with the first element. i The number of detection points for each target; S302. For a newly generated target at the current moment, calculate the center position of the detection points associated with the target, and use this position as the current position of the new target. S303, End; S4. Update the target status and collect personnel information, including the following sub-steps: S401. The target state is divided into detection, activation and release. In step 2, the cluster that does not contain a target is considered to be a new target and is initialized to the detection state. S402, Use Record the number of times the target in the detection state is detected by the radar within frame T1. Not an empty set, Add 1, if It is an empty set. Subtract 1 when Greater than the set threshold When the target transitions from the detection state to the active state, When the value is 0, the target transitions from the probe state to the release state, meaning the target is discarded. S403, Use Record the number of times the active target is detected within frame T2. Not an empty set, Add 1, if It is an empty set. Subtract 1 when When the value is 0, the target changes from the active state to the released state, that is, the target is discarded; S404. Count the number of targets in the active state and their locations, and output the detection results; S405, End.

3. The indoor personnel counting and positioning method based on millimeter-wave radar point cloud data according to claim 1, characterized in that, In step S2, during the overlay process, the height information of the detection points is ignored, and they are overlaid on the target in a two-dimensional plane; a density clustering algorithm is then used to associate the detection points with the target.

4. The indoor personnel counting and positioning method based on millimeter-wave radar point cloud data according to claim 1, characterized in that, In S4, when counting people, only targets that are active are counted.

Images