Millimeter wave radar positioning processing method, device and millimeter wave radar

Abstract

The present application provides a positioning processing method, device and millimeter wave radar for a millimeter wave radar, the method comprising: obtaining scene scanning data through the millimeter wave radar, detecting whether there is an out-of-bounds person according to the scene scanning data, and outputting the action positioning information of the out-of-bounds person; determining the behavior type information of the person according to the action positioning information, and setting the out-of-bounds person as a first person and a second person according to the behavior type information; performing intention detection on the first person and the second person respectively according to the action positioning information to obtain the intention type information; according to the intention type information, matching a preset intention response table to obtain and output the warning response measures.

Description

Technical Field

[0001] The present application relates to the field of radar technology, and in particular to a positioning processing method and device for a millimeter-wave radar, and a millimeter-wave radar. Background Art

[0002] In recent years, with the rapid development of society and the acceleration of urbanization, the security needs of various important places, border areas, military facilities and other areas have been increasing. Existing security systems generally have problems such as low detection accuracy, high false alarm rate, and inability to accurately identify the intentions of intruders. Especially when dealing with multi-person cross-border scenarios, it is difficult to effectively distinguish the behavioral characteristics and intention types of different intruders, resulting in the security system being unable to make accurate threat assessments and timely response measures. At the same time, when tracking and locating intruders, existing systems often lack intelligent analysis methods and cannot accurately predict and classify intruder behaviors. Millimeter-wave radar has shown broad application prospects in the field of security due to its advantages such as all-weather working ability, strong penetration and high distance resolution. However, the current security system based on millimeter-wave radar is still insufficient in data processing and intelligent analysis, especially the lack of in-depth analysis and intelligent judgment of the intruder's behavioral intentions, and cannot formulate corresponding defense strategies for different types of intrusion behaviors, which largely limits the application effect of millimeter-wave radar in the field of security. Summary of the invention

[0003] The present application provides a millimeter wave radar positioning processing method, device and millimeter wave radar, which are used to output different defense measures for different intrusion behaviors and reduce false alarms.

[0004] In a first aspect, an embodiment of the present application provides a positioning processing method of a millimeter wave radar, the method comprising:

[0005] Acquire scene scanning data through millimeter wave radar, detect whether there is a person crossing the boundary according to the scene scanning data, and output the action positioning information of the person crossing the boundary;

[0006] Determine the behavior type information of the actor according to the action location information, and set the cross-border actor as: a first actor and a second actor according to the behavior type information;

[0007] Performing intention detection on the first actor and the second actor respectively according to the action positioning information to obtain intention type information;

[0008] According to the intention type information, the preset intention response table is matched to obtain and output the warning response measures.

[0009] In a second aspect, an embodiment of the present application provides a positioning processing device for a millimeter wave radar, the device comprising:

[0010] A human body detection module is used to obtain scene scanning data through a millimeter wave radar, detect whether there is a person crossing the boundary according to the scene scanning data, and output the action positioning information of the person crossing the boundary;

[0011] A behavior characterization module, used to determine the behavior type information of the actor according to the action location information, and set the cross-border actor as: a first actor and a second actor according to the behavior type information;

[0012] an intention analysis module, configured to detect the intentions of the first actor and the second actor respectively according to the action location information, and obtain intention type information;

[0013] The measure determination module is used to match the preset intention response table according to the intention type information, obtain and output the warning response measures.

[0014] In a third aspect, an embodiment of the present application provides a millimeter wave radar, wherein the millimeter wave radar includes a memory and a processor;

[0015] The memory is used to store computer programs;

[0016] The processor is used to execute the computer program and implement the millimeter wave radar positioning processing method as described in any one of the embodiments of the present application when executing the computer program.

[0017] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the processor enables the processor to implement a positioning processing method for a millimeter-wave radar as described in any one of the embodiments of the present application.

[0018] The embodiment of the present application provides a positioning and processing method for a millimeter-wave radar, the method comprising: obtaining scene scanning data through a millimeter-wave radar, detecting whether there is an out-of-bounds person according to the scene scanning data, and outputting the action positioning information of the out-of-bounds person; determining the behavior type information of the person according to the action positioning information, and setting the out-of-bounds person as: the first person and the second person according to the behavior type information; performing intention detection on the first person and the second person respectively according to the action positioning information to obtain the intention type information; matching a preset intention response table according to the intention type information, obtaining and outputting warning response measures. Through the above method, the behavior type analysis of the out-of-bounds person identified by the millimeter-wave radar is performed and divided into the first person and the second person, realizing the intelligent recognition and classification processing of multi-person out-of-bounds scenes, performing intention detection on different types of persons respectively, obtaining the intention type information, matching based on the preset intention response table, realizing the intelligent output of warning response measures, enabling the system to adopt corresponding defense strategies according to the behavioral intentions of different intruders, and improving the intelligence level and response efficiency of the security system. It organically combines behavior detection, intent analysis and response measures to build a complete intelligent security solution, improves the system's detection accuracy and processing efficiency, provides a more accurate decision-making basis for subsequent defense measures, and effectively reduces the false alarm rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0020] Figure 1 A schematic flow chart of a millimeter wave radar positioning processing method provided in an embodiment of the present application;

[0021] Figure 2 A schematic block diagram of a millimeter wave radar positioning processing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0024] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, the singular forms "a", "an" and "the" are intended to include plural forms unless the context clearly indicates otherwise.

[0025] It should be further understood that the terms “and / or” used in the specification and appended claims refer to any and all possible combinations of one or more of the associated listed items, and include these combinations.

[0026] See also Figure 1 , Figure 1 The embodiment of the present application also provides a schematic flow chart of a millimeter wave radar positioning processing method. Figure 1 As shown, the specific steps of the millimeter wave radar positioning processing method include: S101-S104.

[0027] S101. Obtain scene scanning data through a millimeter wave radar, detect whether there is a person crossing the boundary based on the scene scanning data, and output the action positioning information of the person crossing the boundary.

[0028] Exemplarily, the millimeter wave radar transmits millimeter wave signals and receives echo signals reflected by the target to achieve continuous scanning of the monitoring area. The raw data received by the millimeter wave radar contains three-dimensional spatial information such as distance, azimuth, and pitch angle, as well as dynamic feature information such as Doppler velocity. The millimeter wave radar uploads the collected scene scanning data to the background server, which identifies the actor based on the scene scanning data to detect whether there is an intrusion of the actor in the monitored area.

[0029] The backend server pre-processes the received raw data, including noise suppression, clutter elimination, and signal enhancement. The backend server uses a preset actor detection algorithm to detect targets on the processed data and extract potential cross-border actor targets. For the detected targets, the backend server calculates their spatial coordinates, motion speed, acceleration and other kinematic parameters in real time to form complete action positioning information. The process also includes tracking and associating the target to ensure that a stable tracking effect can be maintained during the target's movement, so as to achieve continuous monitoring of cross-border actors.

[0030] Among them, the server can be an independent server or a server cluster, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN) and big data and artificial intelligence platforms.

[0031] S102, determining the behavior type information of the actor according to the action positioning information, and setting the cross-border actor as: the first actor and the second actor according to the behavior type information.

[0032] Exemplarily, based on the acquired action positioning information, the behavioral characteristics of the person who crosses the boundary are deeply analyzed. By extracting the motion trajectory characteristics of the target, including parameters such as motion speed changes, acceleration characteristics, and motion direction changes, a behavioral feature vector is established. Then, using the pre-trained behavior recognition model, these characteristics are classified and analyzed to identify different types of cross-border behaviors. According to the degree of danger and urgency of the behavior, the cross-border person is classified into the first person and the second person. Among them, the first person usually exhibits more threatening behavioral characteristics, such as running fast, carrying suspicious items, or showing obvious aggressive actions; the second person exhibits relatively mild behavioral characteristics, such as slow movement, wandering, and static observation. This classification can help the system take different levels of response measures in subsequent processing.

[0033] S103: Perform intention detection on the first actor and the second actor respectively according to the action positioning information to obtain intention type information.

[0034] Exemplarily, after completing the classification of the actors, in-depth intention detection and analysis are performed on different types of actors. For the first actor, focus on the directionality and purpose of his movement trajectory, analyze whether he is targeting a specific protection target, and combine factors such as his speed change and residence time to judge his possible sabotage intention. For the second actor, more attention is paid to his range of activities and behavior patterns to determine whether there are premeditated behaviors such as reconnaissance and mapping. A time-based intention reasoning algorithm is used to combine the actor's historical trajectory data, current behavior characteristics, and environmental context information to construct an intention prediction model. Through the analysis of this model, the actor's intention type information can be obtained, including multiple possibilities such as sabotage intention, reconnaissance intention, and misguided intention.

[0035] S104: According to the intention type information, match the preset intention response table to obtain and output the warning response measures.

[0036] Exemplarily, based on the obtained intention type information, by querying the preset intention response table, the preset intention response table contains standard response measures for different intention types. In the intention response table, based on the multi-level response mechanism, it is divided into multiple levels according to the degree of danger of the intention, and each level corresponds to different warning measures. For example, for the first actor with obvious destructive intentions, the highest level of alarm is triggered, the emergency plan is activated, and the security personnel are notified to deal with the scene; for the second actor who is judged to have entered by mistake, mild persuasion measures may be taken. In addition, the response strategy can be dynamically adjusted according to the real-time situation to ensure the appropriateness and effectiveness of the response measures. All warning response measures will be recorded and fed back in real time for subsequent optimization and improvement.

[0037] The embodiment of the present application provides a positioning and processing method for a millimeter-wave radar, the method comprising: obtaining scene scanning data through a millimeter-wave radar, detecting whether there is an out-of-bounds person according to the scene scanning data, and outputting the action positioning information of the out-of-bounds person; determining the behavior type information of the person according to the action positioning information, and setting the out-of-bounds person as: the first person and the second person according to the behavior type information; performing intention detection on the first person and the second person respectively according to the action positioning information to obtain the intention type information; matching a preset intention response table according to the intention type information, obtaining and outputting warning response measures. Through the above method, the behavior type analysis of the out-of-bounds person identified by the millimeter-wave radar is performed and divided into the first person and the second person, realizing the intelligent recognition and classification processing of multi-person out-of-bounds scenes, performing intention detection on different types of persons respectively, obtaining the intention type information, matching based on the preset intention response table, realizing the intelligent output of warning response measures, enabling the system to adopt corresponding defense strategies according to the behavioral intentions of different intruders, and improving the intelligence level and response efficiency of the security system. It organically combines behavior detection, intent analysis and response measures to build a complete intelligent security solution, improves the system's detection accuracy and processing efficiency, provides a more accurate decision-making basis for subsequent defense measures, and effectively reduces the false alarm rate.

[0038] In order to more clearly introduce the technical solution of the present application, the technical solution of the present application will be introduced through specific embodiments below. It should be noted that the specific embodiments are used to expand the technical solution of the present application, but are not intended to limit the present application.

[0039] In some embodiments, the first actor and the second actor are respectively detected for their intentions based on the action location information to obtain the intention type information, including:

[0040] If it is detected according to the action positioning information that the first person has entered the preset first warning interval, a breathing detection is performed on the first person to obtain the first breathing detection information. Specifically, when the millimeter-wave radar performs breathing detection, it is usually required that the target is within a relatively moderate distance range, that is, the first warning interval, such as within a range of 10-20 meters. This is because breathing detection requires higher signal quality and resolution, and too far a distance may affect the detection accuracy. A distance range for a warning interval is set in advance. Generally speaking, this distance range can be set according to the scene requirements. For example, it can be set to a range of 20-40 meters from the protected target as the first warning interval. This distance is set according to the performance of the millimeter-wave radar. It should be far enough to give enough warning time, but not too far to reduce the detection accuracy.

[0041] Send a warning signal to the first person, and perform a breathing test on the first person to obtain second breathing detection information. Specifically, the issuance of warning signals (such as sound and light alarms) needs to consider the effective action distance. The sound warning generally needs to ensure that it is clearly audible within a range of 20-30 meters, and the optical warning needs to be clearly visible within a range of 30-40 meters. At the same time, continue to monitor the person's breathing changes. If the breathing rate is detected to further increase to 35 times per minute after the warning, this change may indicate that the person has a nervous reaction to the warning, or is still approaching the restricted area.

[0042] The first intention information of the first actor is determined based on the action positioning information, the first breathing detection information, and the second breathing detection information. Specifically, by analyzing the person's position change (such as still approaching the restricted area after the warning), the initial breathing state (nervous), and the breathing change after the warning (more nervous), the system can determine that the person may have a clear intention to invade and mark his intention type as a "high-threat intruder".

[0043] If the frequency of the second person's appearance detected according to the action positioning information is greater than the preset frequency threshold within the preset time period, the second intention information is generated. Specifically, assuming that the preset time period is 24 hours, if a person with abnormal patrol behavior is found multiple times within 24 hours, other technical means, such as high-definition cameras, can be used to collect the ID features of the person with abnormal patrol behavior. When a person with the same ID feature is found to have appeared around the restricted area 5 times (the preset frequency threshold may be 3 times / day), this abnormal patrol behavior may indicate that the person is conducting scouting or surveillance activities, and the system marks his intention as a "suspicious scouting person".

[0044] Intent type information is generated based on the first intent information and the second intent information. Specifically, the first actor's "high threat intruder" judgment and the second actor's "suspicious scouting personnel" judgment are combined to generate a comprehensive intent type information, such as "cooperative intrusion behavior detected, threat level: high", indicating that there may be an organized intrusion attempt.

[0045] In the above technical solution, a comprehensive behavioral intention assessment system is constructed by integrating technical features such as action positioning and partition detection, double verification of breathing detection, multi-dimensional data correlation analysis, and time series frequency monitoring. The system can not only accurately perform spatial positioning and physiological state monitoring in real time, but also improve the reliability of judgment through cross-validation of multi-source data. By introducing the time dimension and frequency analysis, the system realizes the intelligent identification of abnormal behavior patterns. The organic combination of these technical features ultimately achieves the improvement of monitoring accuracy, the reduction of false alarm rate, and the timely implementation of preventive intervention, providing more reliable technical support for the analysis of behavioral intentions.

[0046] In some embodiments, performing intention detection on the first actor and the second actor respectively to obtain the intention type information further includes:

[0047] If the approach speed of the first person detected by the action positioning information is greater than the preset speed threshold, a warning message is sent to the first person. Specifically, suppose that at the boundary of a certain supervision area, the boundary distance is 20 meters and the preset speed threshold is 7 meters per second. Through the millimeter wave radar, a person (the first person) is monitored in real time at a speed of 8 meters per second to the restricted area. After detecting this abnormal approach speed, the following warning measures are immediately triggered: the tweeter in the area is activated to play the warning voice "The front is a supervision area, please stop moving immediately", and the warning light system is activated at the same time, projecting a red warning beam to the person, and displaying the "Do not enter" warning sign on the LED display screen around the area.

[0048] After the warning message is issued, if it is detected that the speed threshold of the first actor in the first warning interval is still greater than the preset speed threshold, the third intention information is generated. Specifically, in actual applications, the speed of the actor can quickly pass through the first warning area. If the warning measures are gradually upgraded, it may cause the actor to cause actual harm to the area to be protected. Therefore, when a fast-approaching actor with a strong intention to destroy is detected, the third intention information is directly generated to initiate the highest standard of protection measures.

[0049] Generate the intention type information based on the first intention information, the second intention information, and the third intention information. Specifically, in a specific scenario, there may be multiple people invading and destroying at the same time, and multiple people are working together. In this case, it is necessary to comprehensively score the possible hazards in this scenario. By assigning different weights to different intention information: the first intention information has a weight of 0.3, the second intention information has a weight of 0.3, and the third intention information has a weight of 0.4. After weighted calculation, the comprehensive score is 92 points (out of 100 points). According to the score query intention classification table, the intention type information is finally generated: "The current target is a malicious intrusion attempt, the threat level: extremely high, it is recommended to immediately activate the highest level defense plan."

[0050] In some embodiments, determining the first intention information of the first actor according to the action positioning information, the first breathing detection information and the second breathing detection information includes:

[0051] The first respiratory detection information and the second respiratory detection information are smoothed and filtered, and then the first-order derivative is calculated respectively. The time point when the first-order derivative value exceeds the preset derivative threshold is marked as the respiratory mutation point to obtain the respiratory mutation time series. Specifically, assuming that the acquired original respiratory data has noise fluctuations, it is first smoothed by Gaussian filtering. For example, the original data sequence is [15, 16, 15, 17, 25, 30, 28, 27, 16, 15] times / minute, and after filtering, [15, 15, 16, 20, 26, 28, 27, 22, 16, 15] is obtained. The first-order derivative is calculated to obtain [0, 1, 4, 6, 2, -1, -5, -6, -1]. If the derivative threshold is set to ±4, the mutation points can be marked at t=4s and t=8s.

[0052] According to the respiratory mutation time series, the time interval and amplitude difference between adjacent respiratory mutation points are counted. When the time interval is less than the preset time threshold and the amplitude difference is greater than the preset amplitude threshold, the time interval between adjacent respiratory mutation points is marked as an abnormal mutation interval. Specifically, suppose two adjacent mutation points are detected at t=10s and t=12s, respectively, with a time interval of 2s; the respiratory amplitudes of the mutation points are 15 and 25 times / minute, respectively, and the amplitude difference is 10 times / minute. If the preset time threshold is 3s and the preset amplitude threshold is 8 times / minute, then this interval will be marked as an abnormal mutation interval, because 2s<3s and 10>8.

[0053] In each abnormal mutation interval, the action location information of the first actor is extracted, and the speed change rate and direction change of the actor are calculated. If the speed change rate exceeds the third preset threshold or the direction change exceeds the fourth preset threshold, the abnormal mutation interval is marked as a focus interval. Specifically,

[0054] In a certain abnormal mutation interval (10s-12s), the extracted behavior movement data shows: the speed increases from 2m / s to 4m / s, the speed change rate is 100%; the direction deviates from due north to northeast, and the direction changes by 45 degrees. If the third preset threshold (speed change rate) is set to 80% and the fourth preset threshold (direction change) is set to 30 degrees, then this interval satisfies both conditions (100%>80% and 45 degrees>30 degrees) and should be marked as a concerned interval.

[0055] The peak and valley values ​​of the respiratory waveform in the focus interval are detected, and the respiratory rate of change is calculated. When the respiratory rate of change exceeds the preset change threshold, the focus interval is marked as a warning interval. Specifically, the respiratory waveform in the focus interval is analyzed and it is found that the respiratory rate rises rapidly from 15 times / minute to 25 times / minute, with a change rate of 66.7%. If the preset change threshold is 50%, the focus interval should be marked as a warning interval because 66.7%>50%.

[0056] The average respiratory rate, average movement speed and average direction change rate in the warning interval are calculated respectively, and these three feature quantities are classified according to the preset parameter range to obtain a three-dimensional feature vector. Specifically, in a certain warning interval, the following are calculated: the average respiratory rate is 25 times / minute, corresponding to level 3 (assuming the classification standard: <15 is level 1, 15-20 is level 2, and >20 is level 3), the average movement speed is 4m / s, corresponding to level 2 (assuming the classification standard: <3m / s is level 1, 3-5m / s is level 2, and >5m / s is level 3), and the average direction change rate is 45 degrees / s, corresponding to level 3 (assuming the classification standard: <30 degrees / s is level 1, 30-40 degrees / s is level 2, and >40 degrees / s is level 3). Finally, the three-dimensional feature vector [3, 2, 3] is obtained.

[0057] A preset intention feature mapping table is searched according to the classification result of the three-dimensional feature vector, and when the matching degree of the mapping result exceeds a preset matching threshold, the first intention information is output.

[0058] Specifically, the preset intention feature mapping table includes: a high-threat intention feature group, a medium-threat intention feature group, and a low-threat intention feature group.

[0059] High-threat intention feature group: [3, 2, 3] corresponds to “rapid approach intention”, with a matching degree of 0.9; [3, 3, 3] corresponds to “attack collision intention”, with a matching degree of 0.95; [3, 2, 2] corresponds to “tracking tail random pattern”, with a matching degree of 0.85.

[0060] Moderate threat intention feature group: [3, 1, 3] corresponds to "wandering observation intention", with a matching degree of 0.7; [2, 3, 2] corresponds to "circumventing approach intention", with a matching degree of 0.75; [3, 1, 2] corresponds to "surveying and recording intention", with a matching degree of 0.72

[0061] Low threat intention feature group: [2, 2, 2] corresponds to “normal passing intention”, with a matching degree of 0.6; [1, 1, 2] corresponds to “wandering intention”, with a matching degree of 0.55; [2, 1, 1] corresponds to “temporary stay intention”, with a matching degree of 0.5.

[0062] In a specific embodiment, when the feature vector [3, 2, 3] is detected, the following determination is made:

[0063] The closest feature combination is searched in the intention feature mapping table, and the calculated match with "rapid approach intention" is the highest, which is 0.9. Since 0.9 is greater than the preset matching threshold of 0.8, the system confirms that the intention judgment is credible. At the same time, by recording the suboptimal matching result "attack collision intention" (matching degree 0.85) as an alternative, "rapid approach intention" is output as the first intention information, and its confidence is set to 90%. The continuous tracking mechanism is started. If the subsequent feature vector changes to [3, 3, 3], the intention judgment will automatically upgrade to "attack collision intention".

[0064] Through the above-mentioned multi-level intention determination mechanism, the misjudgment rate can be effectively reduced. At the same time, through continuous tracking and dynamic adjustment, the accuracy and timeliness of intention determination can be ensured.

[0065] In some embodiments, searching a preset intention feature mapping table according to the classification result of the three-dimensional feature vector, and outputting first intention information when the matching degree of the mapping result exceeds a preset matching threshold, includes:

[0066] The three-dimensional eigenvectors are processed by principal component analysis to obtain the feature contribution matrix. Specifically, the covariance matrix of the three-dimensional eigenvectors (respiratory frequency, movement speed, and direction change rate) is calculated, and the covariance matrix is ​​subjected to eigenvalue decomposition to obtain eigenvalues ​​and eigenvectors. The eigenvalues ​​represent the variance contribution of each dimension, and the eigenvectors constitute the feature contribution matrix. For example, suppose the analysis results show that the contribution of respiratory frequency is 0.5, the contribution of speed is 0.3, and the contribution of direction change is 0.2.

[0067] According to the feature contribution matrix, the three-dimensional feature vector is orthogonally transformed to obtain the reduced-dimensional feature vector. Specifically, the original three-dimensional feature vector is linearly transformed using the feature contribution matrix, and the data is projected into the principal component space. By retaining the first two principal components with the largest contribution (for example, the cumulative contribution rate reaches 85%), the dimension reduction from three dimensions to two dimensions is achieved, and the reduced-dimensional feature vector is obtained. This can reduce the computational complexity while retaining the main information.

[0068] The feature space is constructed based on the reduced-dimensional feature vector, and the feature space is gridded to obtain feature grid units. The sample points in the feature grid units are densely clustered to obtain a set of cluster center points. Specifically, the reduced-dimensional two-dimensional feature vector is mapped to a two-dimensional plane to form a feature space. The feature space is gridded, such as a 10×10 uniform grid. In each grid unit, the DBSCAN density clustering algorithm is used to cluster the sample points, and an appropriate neighborhood radius and minimum sample number threshold are set to obtain a number of cluster center points. These cluster center points represent the typical feature distribution of different intent types.

[0069] The Mahalanobis distance is calculated based on the set of cluster centers, and the Mahalanobis distance is weighted and summed to obtain the intent similarity score. Specifically, the Mahalanobis distance from the current sample point to each cluster center is calculated, taking into account the covariance structure of the feature distribution. The obtained distance values ​​are weighted and summed. The weights can be determined based on the number of samples or the compactness of each cluster to obtain a comprehensive intent similarity score, which reflects the degree of match between the current sample and various intent types.

[0070] The intent types in the intent feature map are probabilistically sorted according to the intent similarity scores to obtain an intent matching sequence. Specifically, the various intent types predefined in the intent feature map (such as "rapid approach", "wandering and observing", "normal passing", etc.) are probabilistically sorted according to the intent similarity scores. For example, a probability sequence such as ["rapid approach: 0.8", "wandering and observing: 0.6", "normal passing: 0.3"] may be obtained.

[0071] The intent matching sequence is subjected to sequence correlation analysis to obtain the matching coefficient. Specifically, the intent matching sequence is subjected to time series correlation analysis to examine the correlation between the intent determination results at multiple consecutive time points. The sliding window method is used to calculate the autocorrelation coefficient of the intent sequence within the window to obtain the final matching coefficient. This can reduce misjudgments caused by instantaneous fluctuations.

[0072] The first intention information is output according to the comparison result between the matching coefficient and the preset matching threshold. Specifically, the obtained matching coefficient is compared with the preset matching threshold (such as 0.75). If the matching coefficient exceeds the threshold, the corresponding intention type is output as the first intention information; if it is lower than the threshold, it may be necessary to continue to observe or adopt other judgment strategies. This can ensure the reliability of the output results.

[0073] In some embodiments, scene scanning data is acquired by a millimeter wave radar, and whether there is a person crossing the boundary is detected according to the scene scanning data, and the action location information of the person crossing the boundary is output, including:

[0074] The reflected signals in the horizontal and vertical directions collected by the millimeter-wave radar are processed by Fourier transform to obtain the time-frequency domain feature matrix.

[0075] The radar scanning area is divided into multiple resolution grids according to the time-frequency domain feature matrix to obtain a set of scene grid units. Specifically, the spatial range boundary of the scene is determined, multiple resolution levels (such as 1m, 0.5m, 0.25m) are set, each resolution level is gridded, a hierarchical relationship between grids of different resolutions is established, a unique identifier is assigned to each grid unit, and the adjacency relationship information of the grid units is stored.

[0076] The signal strength within the scene grid unit set is threshold segmented to obtain the target candidate area. Specifically, the average signal strength of the scene background is calculated, a dynamic threshold is set (such as 1.5 times the background strength), the signal strength of each grid unit is compared, adjacent grids that exceed the threshold are merged, an ID is assigned to each candidate area, and candidate areas that are too small are filtered out to obtain the target candidate area.

[0077] The target candidate region is subjected to multi-scale feature extraction to obtain a spatial feature vector group, a three-dimensional projection map is constructed based on the spatial feature vector group, and the three-dimensional projection map is subjected to depth information encoding to obtain a human joint point coordinate sequence. Specifically, the target candidate region is subjected to wavelet decomposition to obtain a multi-scale sub-band coefficient matrix, the multi-scale sub-band coefficient matrix is ​​subjected to local binary pattern calculation to obtain a texture feature descriptor, and a multi-scale feature map is constructed based on the texture feature descriptor; the target candidate region is subjected to non-maximum suppression based on the multi-scale feature map to obtain a key point candidate set, shape context feature extraction is performed on the key point candidate set to obtain a target contour description vector, and a spatial feature vector group is constructed based on the target contour description vector; the spatial feature vector group is subjected to principal curvature analysis to obtain a curvature feature matrix, a local coordinate system is constructed based on the curvature feature matrix, and a spatial transformation is performed on the local coordinate system to obtain a three-dimensional projection map; a voxel grid is constructed based on the three-dimensional projection map, and an octree decomposition is performed on the voxel grid to obtain a hierarchical spatial structure, and a deep learning encoding is performed on the hierarchical spatial structure to obtain a human joint point coordinate sequence.

[0078] According to the coordinate sequence of human joints, a spatiotemporal feature tensor is constructed, and the attention mechanism is used to calculate the spatiotemporal feature tensor to obtain the posture feature mapping matrix. Specifically, the spatial relationship between the joints is extracted, the motion trajectory of the joints is calculated, the feature sequence of the time dimension is constructed, the spatial and temporal features are integrated, the feature tensor is standardized, and the attention mechanism is used to highlight the important features, that is, the posture feature mapping matrix.

[0079] The posture feature mapping matrix is ​​subjected to time series correlation analysis to obtain the behavior trajectory sequence. Specifically, the posture similarity of adjacent moments is calculated, continuous action segments are extracted, key posture change points are identified, a posture state transition diagram is constructed, the complete behavior trajectory sequence is extracted, and the trajectory is smoothed to obtain the behavior trajectory sequence.

[0080] The behavior trajectory sequence is processed by boundary detection to obtain the boundary crossing judgment result, and the behavior trajectory sequence is segmented according to the boundary crossing judgment result to obtain the boundary crossing behavior fragment. Specifically, the preset boundary range information is obtained, whether the trajectory point crosses the boundary is determined, the start and end time of the boundary crossing is marked, the complete boundary crossing behavior fragment is extracted, the boundary crossing duration is calculated, and the severity of the boundary crossing behavior is evaluated.

[0081] The spatial coordinate transformation process is performed on the cross-border behavior fragment to obtain the action positioning information, and the cross-border person is uniquely marked according to the action positioning information, and the action positioning information of the cross-border person is output. Specifically, the relative coordinates are converted into absolute coordinates, the target's movement speed and direction are calculated, the target's feature descriptor is extracted, the target ID association mapping table is established, a unique identifier is generated, and the action positioning information is output.

[0082] In the above process, the accuracy of scene perception is improved through Fourier transform and multi-resolution grid division, and accurate human posture recognition is achieved by multi-scale feature extraction and deep information encoding. The behavior understanding ability is enhanced through spatiotemporal feature tensors and attention mechanisms, and the accurate positioning and tracking of cross-border behaviors are ensured through boundary detection and spatial coordinate transformation. This multi-level technical architecture not only improves detection accuracy and reliability, but also realizes real-time monitoring and early warning of cross-border behaviors, providing effective technical support for security applications.

[0083] In some embodiments, according to the intention type information, a preset intention response table is matched to obtain and output alert response measures, including: querying a preset intention response table according to the intention type information to determine the target alert level, wherein the intention response table contains a mapping relationship between the intention type and the corresponding alert level. According to the target alert level, a corresponding response measure is selected, wherein the response measure includes: issuing a warning sound, activating a warning light, sending an alarm signal, and recording one or more of the behavior video data. According to the target alert level, the corresponding response measure duration and response interval are set. The response measures are executed according to the set duration and response interval, and the execution results are recorded in the alert log.

[0084] See also Figure 2 , Figure 2 The embodiment of the present application also provides a schematic block diagram of a millimeter wave radar positioning processing device, wherein the millimeter wave radar positioning processing device 200 is used to execute the aforementioned millimeter wave radar positioning processing method. The millimeter wave radar positioning processing device 200 can be configured in a server.

[0085] Among them, the server can be an independent server or a server cluster, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN) and big data and artificial intelligence platforms.

[0086] like Figure 2 As shown, the millimeter wave radar positioning processing device 200 includes: a human body detection module 201, a behavior qualitative module 202, an intention analysis module 203 and a measure determination module 204.

[0087] The human body detection module 201 is used to obtain scene scanning data through the millimeter wave radar, detect whether there is a person who crosses the boundary according to the scene scanning data, and output the action positioning information of the person who crosses the boundary.

[0088] The behavior characterization module 202 is used to determine the behavior type information of the actor according to the action location information, and set the cross-border actor as: a first actor or a second actor according to the behavior type information.

[0089] The intention analysis module 203 is used to perform intention detection on the first actor and the second actor respectively according to the action positioning information to obtain intention type information.

[0090] The measure determination module 204 is used to match the preset intention response table according to the intention type information, obtain and output the warning response measures.

[0091] An embodiment of the present application provides a millimeter wave radar, which includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and implement a positioning processing method of the millimeter wave radar as any one of the embodiments of the present application when executing the computer program.

[0092] An embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the processor implements a positioning processing method for a millimeter-wave radar as described in any one of the embodiments of the present application.

[0093] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A millimeter wave radar positioning processing method, characterized in that: The method comprises: The reflected signals in the horizontal and vertical directions collected by the millimeter-wave radar are processed by Fourier transform to obtain a time-frequency domain feature matrix; the radar scanning area is divided into multiple resolution grids according to the time-frequency domain feature matrix to obtain a scene grid unit set; the signal intensity in the scene grid unit set is subjected to threshold segmentation to obtain a target candidate area; the target candidate area is subjected to multi-scale feature extraction to obtain a spatial feature vector group, a three-dimensional projection image is constructed according to the spatial feature vector group, and the three-dimensional projection image is subjected to depth information encoding to obtain a human joint point coordinate sequence; according to the human joint The point coordinate sequence is used to construct a spatiotemporal feature tensor, and the attention mechanism is used to calculate the spatiotemporal feature tensor to obtain a posture feature mapping matrix; the posture feature mapping matrix is ​​subjected to time series correlation analysis to obtain a behavior trajectory sequence; the behavior trajectory sequence is subjected to boundary detection to obtain an out-of-bounds determination result, and the behavior trajectory sequence is segmented according to the out-of-bounds determination result to obtain an out-of-bounds behavior segment; the out-of-bounds behavior segment is subjected to spatial coordinate transformation to obtain action positioning information, and the out-of-bounds person is uniquely marked according to the action positioning information, and the action positioning information of the out-of-bounds person is output; Determine the behavior type information of the actor according to the action location information, and set the cross-border actor as: a first actor and a second actor according to the behavior type information; Performing intention detection on the first actor and the second actor respectively according to the action positioning information to obtain intention type information; According to the intention type information, the preset intention response table is matched to obtain and output the warning response measures.

2. The millimeter wave radar positioning processing method according to claim 1, characterized in that: The performing intention detection on the first actor and the second actor respectively according to the action positioning information to obtain intention type information includes: If it is detected according to the action positioning information that the first person has entered a preset first warning interval, a breathing detection is performed on the first person to obtain first breathing detection information; Sending a warning signal to the first actor, and performing a breathing test on the first actor to obtain second breathing test information; determining first intention information of the first actor according to the action positioning information, the first breathing detection information and the second breathing detection information; If within a preset time period, the appearance frequency of the second actor detected according to the action location information is greater than a preset frequency threshold, second intention information is generated; The intent type information is generated according to the first intent information and the second intent information.

3. The millimeter wave radar positioning processing method as claimed in claim 2, characterized in that: The performing intention detection on the first actor and the second actor respectively to obtain the intention type information further includes: If it is detected according to the action positioning information that the approach speed of the first actor is greater than a preset speed threshold, a warning message is sent to the first actor; After issuing the warning information, if it is detected that the rate threshold of the first actor in the first warning interval is still greater than the preset rate threshold, generating third intention information; The intent type information is generated according to the first intent information, the second intent information and the third intent information.

4. The millimeter wave radar positioning processing method according to claim 2, characterized in that: The determining, according to the action positioning information, the first breathing detection information, and the second breathing detection information, the first intention information of the first actor includes: Performing smoothing filtering on the first breathing detection information and the second breathing detection information, and then performing first-order derivative calculations respectively, marking a time point at which the first-order derivative value exceeds a preset derivative threshold as a breathing mutation point, and obtaining a breathing mutation time series; According to the respiratory mutation time series, the time interval and amplitude difference between adjacent respiratory mutation points are counted, and when the time interval is less than a preset time threshold and the amplitude difference is greater than a preset amplitude threshold, the time interval between adjacent respiratory mutation points is marked as an abnormal mutation interval; In each abnormal mutation interval, extract the action location information of the first actor, calculate the speed change rate and direction change of the actor, and if the speed change rate exceeds a third preset threshold or the direction change exceeds a fourth preset threshold, mark the corresponding abnormal mutation interval as a focus interval; Perform peak and valley value detection on the respiratory waveform in the concerned interval, calculate the respiratory frequency change rate, and when the respiratory frequency change rate exceeds a preset change threshold, mark the concerned interval as a warning interval; The average respiratory rate, average movement speed and average direction change rate within the warning interval are calculated respectively, and the three characteristic quantities are classified according to a preset parameter range to obtain a three-dimensional characteristic vector; A preset intention feature mapping table is searched according to the classification result of the three-dimensional feature vector, and when the matching degree of the mapping result exceeds a preset matching threshold, the first intention information is output.

5. The millimeter wave radar positioning processing method as claimed in claim 4, characterized in that: The step of searching a preset intention feature mapping table according to the classification result of the three-dimensional feature vector, and outputting the first intention information when the matching degree of the mapping result exceeds a preset matching threshold, includes: Performing principal component analysis on the three-dimensional feature vector to obtain a feature contribution matrix; Performing an orthogonal transformation on the three-dimensional feature vector according to the feature contribution matrix to obtain a reduced-dimensional feature vector; Constructing a feature space according to the dimension-reduced feature vector, performing grid division processing on the feature space to obtain feature grid units, and performing density clustering processing on sample points in the feature grid units to obtain a set of cluster center points; Calculating the Mahalanobis distance according to the cluster center point set, performing weighted sum processing on the Mahalanobis distance to obtain an intent similarity score; Probabilistically sorting the intent types in the intent feature mapping table according to the intent similarity scores to obtain an intent matching sequence; Performing sequence correlation analysis on the intention matching sequence to obtain a matching coefficient; The first intention information is output according to a comparison result between the matching degree coefficient and the preset matching threshold.

6. The millimeter wave radar positioning processing method according to claim 1, characterized in that: The method of matching the preset intention response table according to the intention type information to obtain and output the warning response measures includes: According to the intention type information, a preset intention response table is searched to determine the target alert level, wherein the intention response table contains a mapping relationship between the intention type and the corresponding alert level; Selecting corresponding response measures according to the target alert level, wherein the response measures include: issuing a warning sound, activating a warning light, sending an alarm signal, and recording behavior video data or one or more thereof; Set corresponding response measures duration and response interval according to the target alert level; Execute the countermeasures according to the set duration and response interval, and record the execution results in the alert log.

7. A millimeter wave radar positioning processing device, characterized in that: The millimeter wave radar positioning processing device is used to execute the millimeter wave radar positioning processing method according to any one of claims 1 to 6, and the millimeter wave radar positioning processing device includes: The human body detection module is used to perform Fourier transform processing on the horizontal and vertical reflection signals collected by the millimeter wave radar to obtain a time-frequency domain feature matrix; perform multi-resolution grid division processing on the radar scanning area according to the time-frequency domain feature matrix to obtain a scene grid unit set; perform threshold segmentation processing on the signal intensity in the scene grid unit set to obtain a target candidate area; perform multi-scale feature extraction processing on the target candidate area to obtain a spatial feature vector group, construct a three-dimensional projection image according to the spatial feature vector group, perform depth information encoding processing on the three-dimensional projection image to obtain a human joint point coordinate sequence; according to the The human joint point coordinate sequence constructs a spatiotemporal feature tensor, and the attention mechanism is used to calculate the spatiotemporal feature tensor to obtain a posture feature mapping matrix; the posture feature mapping matrix is ​​subjected to time series correlation analysis to obtain a behavior trajectory sequence; the behavior trajectory sequence is subjected to boundary detection to obtain an out-of-bounds determination result, and the behavior trajectory sequence is segmented according to the out-of-bounds determination result to obtain an out-of-bounds behavior segment; the out-of-bounds behavior segment is subjected to spatial coordinate transformation to obtain action positioning information, and the out-of-bounds person is uniquely marked according to the action positioning information, and the action positioning information of the out-of-bounds person is output; A behavior characterization module, used to determine the behavior type information of the actor according to the action location information, and set the cross-border actor as: a first actor and a second actor according to the behavior type information; an intention analysis module, configured to detect the intentions of the first actor and the second actor respectively according to the action location information, and obtain intention type information; The measure determination module is used to match the preset intention response table according to the intention type information, obtain and output the warning response measures.

8. A millimeter wave radar, characterized in that: The millimeter wave radar includes a memory and a processor; The memory is used to store computer programs; The processor is used to execute the computer program and implement the millimeter wave radar positioning processing method as described in any one of claims 1 to 6 when executing the computer program.

Images