Millimeter-wave radar human body existence sensing method capable of eliminating interference in blocks

Abstract

The invention discloses a millimeter wave radar human body existence sensing method capable of eliminating interference in blocks, and belongs to the technical field of human body existence sensing, and the method comprises the following steps: S1, a millimeter wave radar emits an electromagnetic wave signal indoors, and receives an echo signal when encountering a static object and a vibrating object; s2, Doppler frequency shift phenomenon information of a plurality of vibration objects in the echo signals is extracted, and feature vectors are formed; and S3, clustering the feature vectors, and dividing the spatial region into a plurality of blocks by using K-means and DBSCAN clustering algorithms. According to the method, feature vectors are formed by learning space [SNR, distance, angle, speed, micro-Doppler frequency] and other multi-dimensional information, a single threshold value is replaced, and environment features can be captured more comprehensively; and meanwhile, blocks are divided through clustering, false triggering risk levels are marked, a high-risk area is comprehensively judged in combination with a tracking perception algorithm, interference signals such as wall surface vibration, curtain vibration and air conditioner airflow are effectively filtered, and the misjudgment rate is remarkably reduced.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of human presence perception, and specifically relates to a method for human presence perception of millimeter wave radar with interference elimination in sub-blocks. BACKGROUND

[0002] In the existing technical field of human presence perception, the traditional detection method often relies on a single threshold to determine whether a human target exists, which leads to poor adaptability to environmental changes. For example, in an indoor environment, static objects such as walls, floors, and metal doors and windows, as well as objects that move with the wind such as curtains, plants, and clothes, and interference signals caused by devices such as air conditioners and fans, all affect the accuracy of human detection. In addition, for the recognition of moving targets, it is difficult to distinguish between humans and other moving objects (such as pets and mechanical movements) by relying solely on simple speed or distance information, which can easily lead to false positives.

[0003] The millimeter wave radar system in the prior art has a relatively simple antenna design, which makes it difficult to accurately capture the spatial position characteristics of the human body, especially in a complex indoor environment, where its positioning accuracy and tracking stability are challenged. At the same time, there is a lack of effective filtering mechanism for areas prone to false triggering, resulting in a high false alarm rate, which affects user experience, and therefore it is necessary to develop a new type of millimeter wave adaptive human presence perception method with 1T2R antenna design. SUMMARY

[0004] The present application aims to provide a method for human presence perception of millimeter wave radar with interference elimination in sub-blocks to solve the problems raised in the background.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical solution: a method for human presence perception of millimeter wave radar with interference elimination in sub-blocks, comprising the following steps:

[0006] S1: The millimeter wave radar transmits electromagnetic wave signals indoors and receives echo signals from static objects and moving objects;

[0007] S2: Extracting Doppler shift phenomenon information of a plurality of moving objects in the echo signal to form a feature vector;

[0008] S3: Clustering the feature vector, dividing the space area into a plurality of blocks using K-means and DBSCAN clustering algorithms, and labeling the risk level of false triggering of the corresponding block;

[0009] S4: Automatically increasing the detection sensitivity for high-noise blocks and reducing the false alarm rate for low-noise blocks;

[0010] S5: Using 1T2R antenna design to accurately track the position of the human body in combination with the ability of lateral angle measurement and top-mounted angle measurement;

[0011] S6: Recognize and distinguish human body and other moving objects by using micro-Doppler detection technology, and make a comprehensive judgment to improve the accuracy of human body detection;

[0012] As a further preferred embodiment of the present technical solution: in the ambient noise learning, the static objects include walls, floors, metal doors and windows, etc. typical indoor interference sources, and the vibrating objects include curtains, plants, clothes, etc. objects that move with the wind, as well as air conditioners, fans, exhaust outlets, air purifiers, etc.

[0013] As a further preferred embodiment of the present technical solution: the dimension of the feature vector includes but is not limited to SNR, distance, angle, speed, micro-Doppler frequency;

[0014] As a further preferred embodiment of the present technical solution: the feature vector of the ambient noise can be expressed as: v = [R, Az, El, V, fμ, SNR,...], where R is the distance, Az is the horizontal angle, El is the pitch angle, V is the speed, fμ is the micro-Doppler frequency, and SNR is the signal-to-noise ratio;

[0015] As a further preferred embodiment of the present technical solution: the steps of noise clustering include:

[0016] S1: Collect the feature vectors and sort them in the order of R, Az, and El, with the correlation determination function constraint strength decreasing in turn;

[0017] S2: Extract each A vector into a matrix, and cut and integrate the set into B matrices;

[0018] S3: Use SNR, V, and fμ as the correlation coefficient respectively, merge the adjacent strong correlation items, and complete the block division;

[0019] As a further preferred embodiment of the present technical solution: the labeling of the false trigger risk level is based on the feature vector distribution of each block in the clustering result, and the false trigger risk identification block can be divided into easy false trigger identification block and low false trigger identification block;

[0020] And the easy false trigger identification block can be expressed as: the block with |V|>0 or fμ energy greater than the threshold τ;

[0021] And for the easy false trigger identification block, a matching filter can be matched in the local cloud filtering model library, and if the matching fails, the content on the vector matrix of the block is transmitted to the cloud server, and the AI server generates a matching filter model and sends it to the 1t2r antenna and the millimeter wave radar, and updates the local database;

[0022] Mis-triggering recognition zone body existence determination: for the detected target, if not in the mis-triggering recognition zone block, directly determine the body existence; if in the mis-triggering zone block, call the matched filter to remove external influencing factors after interference, and judge whether it is a human body combined with the historical trajectory;

[0023] As a further preferred of the technical solution: it further includes a user interface configuration, and the user can manually label the mis-triggering recognition zone block through the interface, and the labeling information is uploaded to the cloud server for filter model learning and information matching.

[0024] Compared with the prior art, the present application has the following advantages:

[0025] 1. In the present application, the feature vector is formed by learning the multi-dimensional information such as SNR, distance, angle, speed and micro-Doppler frequency, instead of a single threshold value, which can more comprehensively capture environmental characteristics; at the same time, the block is divided by clustering and the mis-triggering risk level is labeled, and the high-risk area is comprehensively judged by combining the tracking perception algorithm, which effectively filters out interference signals such as wall vibration, curtain vibration and air flow, and significantly reduces the misjudgment rate.

[0026] 2. In the present application, by using 1T2R antenna design, combined with lateral angle measurement and top-mounted angle measurement technology, multi-dimensional angle information such as horizontal angle (Az) and pitch angle (El) can be obtained at the same time, compared with traditional single antenna design, the spatial position characteristics of human body can be more comprehensively captured, and the positioning accuracy is greatly improved, at the same time, through continuous collection and analysis of feature vectors such as distance (R) and speed (V), combined with multi-dimensional information fusion, the human body movement trajectory can be continuously and stably tracked, even in complex indoor environment, the tracking stability can be maintained at a high level, at the same time, this method is also applicable to 1T1R, 2T2R, 4T4R and other multi-transmit and multi-receive antenna systems.

[0027] 3. In the present application, through the multi-path processing mechanism and the filter matching strategy of the mis-triggering zone block (local filter model library matching + AI server deep learning to generate special filter), the interference signals in specific areas can be filtered, the resistance of the system to complex environmental interference can be significantly enhanced, and the human body existence state can still be stably detected in the presence of vibration, multi-path reflection and other scenes. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 The flowchart of the method for partitioning and deinterfering the millimeter wave radar human existence perception according to the present application;

[0029] Figure 2 The main flowchart of the human existence determination of the method for partitioning and deinterfering the millimeter wave radar human existence perception according to the present application;

[0030] Figure 3The application discloses an environment background noise learning and easy false triggering block identification map of a method for human body existence sensing of a millimeter wave radar with block interference elimination.

[0031] Figure 4 The application discloses an easy false triggering area matching filter generation map of a method for human body existence sensing of a millimeter wave radar with block interference elimination. DETAILED DESCRIPTION

[0032] The technical solutions in the embodiments of the application will be apparently and completely described in combination with the drawings in the embodiments of the application. Apparently, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by a person of ordinary skill in the art without creative work belong to the protection scope of the application.

[0033] EMBODIMENT

[0034] Please refer to Figures 1-4 The application provides a technical solution: a method for human body existence sensing of a millimeter wave radar with block interference elimination, comprising the following steps:

[0035] S1: The millimeter wave radar transmits an electromagnetic wave signal indoors, and receives echo signals of encountering static objects and vibrating objects;

[0036] S2: Doppler shift phenomenon information of a plurality of vibrating objects in the echo signals is extracted to form a feature vector;

[0037] S3: The feature vector is clustered, K-means and DBSCAN clustering algorithms are adopted to divide a space region into a plurality of blocks, and a risk level of false triggering of the corresponding block is labeled;

[0038] S4: The detection sensitivity is automatically improved for a high-noise block, and the false alarm rate is reduced for a low-noise block;

[0039] S5: A 1T2R antenna design is adopted to combine the ability of lateral angle measurement and top-mounted angle measurement to accurately track the position of the human body;

[0040] S6: Micro-Doppler detection technology is used to identify and distinguish the human body from other moving objects, and a comprehensive judgment is made to improve the accuracy of human body detection;

[0041] As Figure 2 shown in the embodiment, the specific process steps are as follows:

[0042] S11: The radar transmits an FMCW signal and performs multi-channel AD sampling;

[0043] S22: Environment background noise detection learning is performed, and a feature vector is extracted;

[0044] S33: For the detected target, if it is not in the easy false trigger block, directly call the matching filter in the block to interfere and judge; if it is in the easy false trigger block, after interference, combine whether the historical trajectory conforms to the human body motion trajectory to determine whether there is a person or not;

[0045] As shown in Figure 3 In this embodiment, the ambient noise learning and easy false trigger block identification steps are as follows:

[0046] S12: The transmitting antenna transmits orthogonal FMCW signals in turn according to time slices, and the receiving antenna synchronously receives echoes, 1 second transmits m frames of data, each frame contains n continuous chirps, AD samples the data of each channel and saves it as an original data matrix, processes the data, and sequentially obtains distance R, speed V, horizontal angle Az, pitch angle El, micro-Doppler frequency fμ and signal-to-noise ratio SNR, to form a feature vector v = [R, Az, El, V, fμ, SNR,...];

[0047] S13: Ambient noise clustering: traverse the feature vector set, sort by R, Az and El, extract every n vectors into a matrix, and merge adjacent strongly correlated items with SNR, V and fμ as correlation coefficients, to complete block division;

[0048] S14: Easy false trigger block identification: when the block has |V|>0 or fμ energy greater than threshold value τ, mark it as an easy false trigger block, and perform multi-path learning;

[0049] S15: Ambient noise filter matching and filter model database learning update, combined with user manual annotation, complete ambient noise learning;

[0050] In this embodiment, specifically: the annotation of false trigger risk level is based on the feature vector distribution of each block in the clustering result, and the false trigger risk identification block can be divided into an easy false trigger identification block and a low false trigger identification block;

[0051] And the easy false trigger identification block can be represented as: a block with |V|>0 or fμ energy greater than threshold value τ;

[0052] And for the easy false trigger identification block, a matching filter can be matched in the local cloud filter model library, if the matching fails, the content on the vector matrix of the block is transmitted to the cloud server, the AI server generates a matching filter model and sends it to the 1t2r antenna and the millimeter wave radar, and updates the local database;

[0053] Easy false trigger identification block human existence determination: for the detected target, if it is not in the easy false trigger identification block, directly determine the human existence; if it is in the easy false trigger block, call the matching filter to remove external influencing factors after interference, and combine the historical trajectory to determine whether it is a human body;

[0054] Meanwhile, the process of generating the matched filter model by the cloud server comprises:

[0055] S21: after receiving the vector matrix, the AI server starts neural network deep learning to generate a matched filter model by taking the matrix as input;

[0056] S22: the generated filter model is recorded in a filter model library and is distributed to the radar, and the radar adopts a 1T2R radio frequency system, giving consideration to lateral angle measurement and top-mounted angle measurement capability.

[0057] Through learning of the bottom noise of static objects (wall surface, ground, etc.) and shaken objects (equipment, floating objects, etc.), the system can pre-master the characteristics of typical interference sources in the environment, and filter out the interference in a targeted manner during detection. Even in a scene with more dynamic interference (such as when the window is opened and the curtain is moving, or the fan is running), the system can still reliably identify the presence of a human body. Meanwhile, the feature vector covers multiple dimensions such as distance (R), horizontal angle (Az), pitch angle (El), speed (V), micro-Doppler frequency (fμ), signal-to-noise ratio (SNR), etc., providing comprehensive data support for block division, risk level labeling and filtering processing, so that the system can still maintain stable performance in a complex indoor environment.

[0058] Although embodiments of the present application have been shown and described, it is to be understood that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for human presence sensing using a millimeter-wave radar with block-based interference removal, characterized by: The following steps are involved: S1: The millimeter-wave radar transmits electromagnetic wave signals indoors and receives echo signals from static and vibrating objects. S2: Extract the Doppler frequency shift information of multiple vibrating objects in the echo signal to form a feature vector; S3: Cluster the feature vectors and use the K-means and DBSCAN clustering algorithms to divide the spatial area into several blocks, and mark the risk level of false triggering of the corresponding blocks; S4: Automatically improves detection sensitivity for high-noise blocks and reduces false alarm rate for low-noise blocks; S5: uses a 1T2R antenna design combined with side and top-mounted angle measurement capabilities to accurately track the human body position; S6: Use micro-Doppler detection technology to identify and distinguish the human body from other moving objects, and make comprehensive judgments to improve the accuracy of human body detection.

2. A method for human presence sensing using millimeter-wave radar with block-based interference removal according to claim 1, characterized in that: In environmental noise floor learning, static objects include typical indoor interference sources such as walls, floors, metal doors and windows, and vibrating objects include objects fluttering in the wind, such as curtains, flowers, clothing, as well as air conditioners, fans, exhaust vents, and air purifiers.

3. A method for human presence sensing using millimeter-wave radar with block-based interference removal according to claim 2, characterized in that: The dimensions of the feature vector include but are not limited to SNR, distance, angle, velocity, and micro-Doppler frequency.

4. A method for human presence sensing using millimeter-wave radar with block-based interference removal according to claim 3, characterized in that: The feature vector of the extracted ambient noise can be expressed as: v = [R, Az, El, V, fμ, SNR, ...], where R is the distance, Az is the horizontal angle, El is the pitch angle, V is the velocity, fμ is the micro-Doppler frequency, and SNR is the signal-to-noise ratio.

5. A method for human presence sensing by millimeter-wave radar with block-based interference removal according to claim 4, characterized in that: The steps of background noise clustering include: S1: Collect the eigenvectors and sort them in the order of R, Az, and El, with the constraint strength of the correlation determination function decreasing in sequence; S2: Extract each A vector into a matrix, and cut and integrate the set into B matrices; S3: Use SNR, V, and fμ as correlation coefficients, merge adjacent strongly correlated items, and complete block division.

6. A method for human presence sensing using millimeter-wave radar with block-based interference removal according to claim 5, characterized in that: The false trigger risk level is labeled based on the distribution of feature vectors of each block in the clustering results, and the false trigger risk identification blocks can be divided into easy false trigger identification blocks and low false trigger identification blocks; And the blocks prone to false triggering can be expressed as: blocks where |V|>0 or fμ energy is greater than the threshold value τ; Furthermore, for blocks prone to false triggering, a matching filter can be applied to the local cloud filter model library. If the match fails, the content of the vector matrix of the block is transmitted to the cloud server. The AI ​​server generates a matched filter model and sends it to the 1t2r antenna and millimeter-wave radar, and updates the local database. At the same time, the filter model is sent to antennas not limited to 1t2r antennas. Human presence determination in easily falsely triggered recognition blocks: If the detected target is not in the easily falsely triggered recognition block, it is directly determined that a human is present; If it is in a block that is prone to false triggering, the matched filter is called to remove interference and external influencing factors, and the historical trajectory is combined to determine whether it is a human body.

7. A method for human presence sensing by millimeter-wave radar with block-based interference removal according to claim 6, characterized in that: It also includes user interface configuration, through which users can manually mark easily mistriggered recognition blocks, and upload the marked information to the cloud server for filter model learning and information matching.

Images