技术领域Technical field
本发明涉及电子信息技术领域，特别是指一种基于毫米波雷达的人员特征提取及异常行为判别方法。The invention relates to the field of electronic information technology, and in particular, to a method for extracting human features and identifying abnormal behaviors based on millimeter wave radar.
背景技术Background technique
随着社会的快速发展，特定场景的人员特征提取检测与异常行为的判别具有越来越多的应用需求，如银行安防、保卫工作等。暴力抢夺、长时间人员逗留、取款人员生命体征异常是一种常见的事故，这就亟需应用一种高效、准确的人员异常行为检测技术，以确保无人值守时特定区域的安全与正常运行。With the rapid development of society, there are more and more application requirements for personnel feature extraction and detection in specific scenarios and the identification of abnormal behaviors, such as bank security and security work. Violent robbery, long-term personnel stay, and abnormal vital signs of cashiers are common accidents. This requires the application of an efficient and accurate abnormal personnel behavior detection technology to ensure the safety and normal operation of specific areas when no one is on duty. .
传统的声光传感技术是当前特定区域人员行为检测及判别采用的较为常用的技术。通过声音传感检测区域内异常声音，是环境异常感知的有效手段。但声音传感技术很容易受到周围环境的影响。同时，当前通过全方位布置多个摄像头能够实现面孔识别、环境感知及异常行为检测，使得视频监控安保技术在人员异常行为检测与识别应用最为广泛。视频技术分为可见光视频和红外视频，其中，可见光视频在一定光照度场景成像分辨率高，红外视频环境适应能力强，但成像分辨率低识别能力弱。同时视频监控探头需要暴露在外，其隐秘性较差，容易被不法人员破坏或屏蔽。Traditional sound and light sensing technology is currently a more commonly used technology for detecting and identifying people's behavior in specific areas. Detecting abnormal sounds in an area through sound sensing is an effective means of sensing environmental anomalies. But sound sensing technology is easily affected by the surrounding environment. At the same time, face recognition, environment perception and abnormal behavior detection can now be achieved by arranging multiple cameras in all directions, making video surveillance security technology the most widely used in the detection and identification of abnormal personnel behavior. Video technology is divided into visible light video and infrared video. Among them, visible light video has high imaging resolution in a certain illumination scene, and infrared video has strong adaptability to the environment, but low imaging resolution and weak recognition ability. At the same time, the video surveillance probe needs to be exposed to the outside, and its privacy is poor, and it is easy to be destroyed or blocked by illegal personnel.
发明内容Contents of the invention
为解决上述技术问题，本发明提供一种基于毫米波雷达的人员特征提取及异常行为判别方法，具有环境适应能力强，分辨率高、检测与判别快速并准确的优点。In order to solve the above technical problems, the present invention provides a method for human feature extraction and abnormal behavior identification based on millimeter wave radar, which has the advantages of strong environmental adaptability, high resolution, fast and accurate detection and identification.
本发明提供技术方案如下：The technical solutions provided by the present invention are as follows:
一种基于毫米波雷达的人员特征提取及异常行为判别方法，所述方法包括：A method for extracting personnel features and identifying abnormal behaviors based on millimeter wave radar. The method includes:
S1：利用毫米波雷达MIMO天线技术，实现待测区内静态目标和动态目标的点云数据采集；S1: Use millimeter wave radar MIMO antenna technology to collect point cloud data of static targets and dynamic targets in the area to be measured;
S2：根据采集的点云数据，分析所述静态目标和动态目标的数据特征，实现待测区内人员数量统计、定位和跟踪；S2: Based on the collected point cloud data, analyze the data characteristics of the static targets and dynamic targets to achieve statistics, positioning and tracking of the number of people in the area to be measured;
S3：根据所述静态目标和动态目标各维度速度、加速度分析，参照人员日常行为特征参数，实现人体异常行为判定。S3: Based on the speed and acceleration analysis of each dimension of the static target and dynamic target, and with reference to the daily behavioral characteristic parameters of personnel, the abnormal human behavior determination is realized.
进一步的，所述S1包括：Further, the S1 includes:
S11：安装毫米波雷达，并设置所述毫米波雷达的工作边界参数；S11: Install a millimeter wave radar and set the working boundary parameters of the millimeter wave radar;
S12：对所述毫米波雷达的工作参数进行配置，利用毫米波雷达MIMO天线技术进行静态目标和动态目标的点云数据采集。S12: Configure the working parameters of the millimeter wave radar, and use millimeter wave radar MIMO antenna technology to collect point cloud data of static targets and dynamic targets.
进一步的，所述S11包括：Further, the S11 includes:
S111：安装毫米波雷达，所述毫米波雷达的检测安装高度为h，所述毫米波雷达的俯仰方向按照水平下倾斜α度；S111: Install a millimeter-wave radar. The detection installation height of the millimeter-wave radar is h, and the pitch direction of the millimeter-wave radar is tilted α degrees downward from the horizontal;
S112：定义空间坐标系，将平行于所述毫米波雷达的探测方向的水平方向定义为X轴，将垂直于所述毫米波雷达的探测方向的水平方向定义为Y轴，将垂直方向定义为Z轴；S112: Define a spatial coordinate system, define the horizontal direction parallel to the detection direction of the millimeter wave radar as the X axis, define the horizontal direction perpendicular to the detection direction of the millimeter wave radar as the Y axis, and define the vertical direction as Z axis;
S113：设置所述毫米波雷达的工作边界参数；S113: Set the working boundary parameters of the millimeter wave radar;
其中，所述工作边界参数包括静态边界条件和动态边界条件，所述静态边界条件为X轴的x0至x1范围、Y轴的y0至y1范围和Z轴的z0至z1范围，所述动态边界条件为X轴的x2至x3范围、Y轴的y2至y3范围和Z轴的z2至z3范围，x0、x1、y0、y1、z0、z1、x2、x3、y2、y3、z2和z3根据待测区内测量异常行为的需要确定。Wherein, the working boundary parameters include static boundary conditions and dynamic boundary conditions. The static boundary conditions are the range of x0 to x1 of the X axis, the range of y0 to y1 of the Y axis, and the range of z0 to z1 of the Z axis. range, the dynamic boundary conditions are the range ofx2 tox3 on the X axis, the range ofy2 toy3 on the Y axis, and the range ofz2 toz3 on the Z axis,x0 ,x1 ,y0 ,y1 , z0 , z1 , x2 , x3 , y2 , y3 , z2 and z3 are determined according to the need to measure abnormal behavior in the area to be measured.
进一步的，所述S12包括：Further, the S12 includes:
S121：配置雷达工作调谐Chirp参数和配置雷达帧Frame工作参数；S121: Configure radar operation tuning Chirp parameters and configure radar frame Frame operation parameters;
S122：通过毫米波雷达MIMO天线技术，虚拟出方位向、俯仰向各有N个接收天线，并进行点云数据采集。S122: Using millimeter wave radar MIMO antenna technology, N receiving antennas are virtualized in the azimuth and pitch directions, and point cloud data is collected.
进一步的，所述S2包括：Further, the S2 includes:
根据人员目标不同，进行距离维FFT、水平方向Capon BF、距离/水平方向CFAR、垂直方向BF和垂直角度估计、速度估计以及聚类跟踪。According to different human targets, distance dimension FFT, horizontal Capon BF, distance/horizontal CFAR, vertical BF and vertical angle estimation, speed estimation and cluster tracking are performed.
进一步的，所述距离维FFT包括：Further, the distance dimension FFT includes:
对每个接收天线的ADC数据进行一维加窗和一维FFT；Perform one-dimensional windowing and one-dimensional FFT on the ADC data of each receiving antenna;
所述水平方向Capon BF包括：The horizontal Capon BF includes:
进行静态载波去除，并在水平方向的N个接收天线上计算空间相关矩阵，通过Capon BF生成距离/水平方向热图；Perform static carrier removal, calculate the spatial correlation matrix on N receiving antennas in the horizontal direction, and generate a range/horizontal heat map through Capon BF;
所述距离/水平方向CFAR包括：The distance/horizontal CFAR includes:
通过二维CFAR，在所述距离/水平方向热图上检测出目标点，得到包含各个目标点的距离和方位角的第一组点云；Through two-dimensional CFAR, the target points are detected on the distance/horizontal heat map, and a first set of point clouds containing the distance and azimuth angle of each target point are obtained;
所述垂直方向BF和垂直角度估计包括：The vertical direction BF and vertical angle estimation include:
在所述第一组点云的每个目标点上计算所有接收天线上的空间相关矩阵，并通过垂直方向的BF生成目标点在垂直方向的热图，在所述热图上通过峰值搜索获得目标点的垂直高度，得到包含各个目标点的距离、方位角和俯仰角的第二组点云；Calculate the spatial correlation matrix on all receiving antennas on each target point of the first set of point clouds, and generate a heat map of the target point in the vertical direction through BF in the vertical direction, which is obtained by peak search on the heat map The vertical height of the target point is used to obtain a second set of point clouds containing the distance, azimuth angle and pitch angle of each target point;
所述速度估计包括：Said speed estimates include:
对所述第二组点云的每个目标点计算空间BF向量，利用所述BF向量对目标点距离bin进行空间滤波并计算Doppler谱，在所述Doppler谱上通过峰值搜索获得目标点的速度，得到包含各个目标点的距离、方位角、俯仰角和速度的第三组点云；Calculate a spatial BF vector for each target point of the second group of point clouds, use the BF vector to perform spatial filtering on the target point distance bin and calculate a Doppler spectrum, and obtain the speed of the target point through peak search on the Doppler spectrum. , obtain the third set of point clouds containing the distance, azimuth angle, pitch angle and speed of each target point;
所述聚类跟踪包括：The cluster tracking includes:
对所述第三组点云，通过聚类跟踪算法进行目标聚类，并对目标进行跟踪和预测，得到跟踪目标列表，所述跟踪目标列表中每个目标均包含空间位置，速度和加速度信息。For the third group of point clouds, the clustering tracking algorithm is used to perform target clustering, and the targets are tracked and predicted to obtain a tracking target list. Each target in the tracking target list contains spatial position, speed and acceleration information. .
进一步的，所述聚类跟踪之前还包括：Further, the cluster tracking also includes:
对所述第三组点云，利用平均相消算法实现静态目标的滤除，得到极坐标下多普勒时频谱图像，并对动态目标测量数据进行雷达极坐标到所述空间坐标系的转换。For the third group of point clouds, the average cancellation algorithm is used to filter out static targets, obtain Doppler time spectrum images in polar coordinates, and convert the dynamic target measurement data from radar polar coordinates to the spatial coordinate system. .
进一步的，进行目标聚类的方法包括：Further, methods for target clustering include:
结合待测区的特定环境，采用基于无监督学习的K均值聚类算法，通过不同粒度的聚类处理，实现不同特征值间耦合关系的学习，实现对目标特征的有效融合与识别；Combined with the specific environment of the area to be tested, the K-means clustering algorithm based on unsupervised learning is used to learn the coupling relationship between different feature values through clustering processing at different granularities, and achieve effective fusion and identification of target features;
对目标进行跟踪和预测的方法包括：Methods for tracking and predicting targets include:
利用特征跟踪算法，在目标匹配中，采用最近邻目标匹配跟踪算法，考虑待跟踪目标与下一帧每个目标的质心的欧氏距离，将欧式距离最小的两个目标认为是同一个目标。Using the feature tracking algorithm, in the target matching, the nearest neighbor target matching tracking algorithm is used, considering the Euclidean distance between the target to be tracked and the centroid of each target in the next frame, and the two targets with the smallest Euclidean distance are considered to be the same target.
进一步的，所述S3包括：Further, the S3 includes:
S31：对所述跟踪目标列表的X轴和Y轴的信息进行微分，得到人员运动速度和肢体摆动最大速度，实现水平方向的人体行为判定；S31: Differentiate the information on the X-axis and Y-axis of the tracking target list to obtain the movement speed of the person and the maximum speed of limb swing, so as to realize the judgment of human behavior in the horizontal direction;
S32：结合人体日常行为特点，分析所述跟踪目标列表的Z轴的信息。实现垂直方向的人体行为判定；S32: Analyze the Z-axis information of the tracking target list based on the daily behavioral characteristics of the human body. Realize vertical human behavior judgment;
S33：参照人员日常行为特征参数，对判定得到的水平方向的人体行为和垂直方向的人体行为进行人体异常行为判定。S33: Determine abnormal human behavior based on the determined human body behavior in the horizontal direction and the human body behavior in the vertical direction with reference to the daily behavioral characteristic parameters of the person.
本发明具有以下有益效果：The invention has the following beneficial effects:
本发明是一种非接触式、高分辨率的人员异常行为快速检测方法，可用于特定区域人员特征提取及异常行为的识别。通过将毫米波雷达及MIMO技术结合使用，实现快速波束赋形和扫描，解决了现有人员异常行为检测与识别系统中存在的分辨率低和环境适应能力差的问题，实现具有高分辨的毫米波雷达点云图像，提高对区域人员异常行为的特征检测和识别能力，在实现快速精准检测的同时具有低功耗、小型化等优点。本发明适用于多种环境多人员数量的特征提取及异常行为检测判别，可根据用户需求进行自适应设计，具有良好的可拓展性。The invention is a non-contact, high-resolution rapid detection method of abnormal human behavior, which can be used to extract human features and identify abnormal behaviors in specific areas. By combining millimeter wave radar and MIMO technology, rapid beamforming and scanning are achieved, which solves the problems of low resolution and poor environmental adaptability in existing abnormal human behavior detection and identification systems, and achieves high-resolution millimeter Wave radar point cloud images improve the feature detection and identification capabilities of abnormal behaviors of people in the area. It not only achieves fast and accurate detection, but also has the advantages of low power consumption and miniaturization. The invention is suitable for feature extraction and abnormal behavior detection and discrimination in multiple environments and multiple numbers of people. It can be designed adaptively according to user needs and has good scalability.
附图说明Description of the drawings
图1为本发明的基于毫米波雷达的人员特征提取及异常行为判别方法的流程图；Figure 1 is a flow chart of the method of human feature extraction and abnormal behavior identification based on millimeter wave radar according to the present invention;
图2为毫米波雷达安装以及空间坐标系建立示意图；Figure 2 is a schematic diagram of millimeter wave radar installation and spatial coordinate system establishment;
图3为分析静态目标和动态目标的数据特征的实现过程示意图。Figure 3 is a schematic diagram of the implementation process of analyzing the data characteristics of static targets and dynamic targets.
具体实施方式Detailed ways
为使本发明要解决的技术问题、技术方案和优点更加清楚，下面将结合附图及具体实施例进行详细描述。In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.
本发明实施例提供一种基于毫米波雷达的人员特征提取及异常行为判别方法，如图1-3所示，该方法包括：Embodiments of the present invention provide a method for extracting human features and identifying abnormal behaviors based on millimeter wave radar, as shown in Figures 1-3. The method includes:
S1：利用毫米波雷达MIMO天线技术，实现待测区内静态目标和动态目标的点云数据采集。S1: Use millimeter wave radar MIMO antenna technology to collect point cloud data of static targets and dynamic targets in the area to be measured.
本发明的毫米波雷达前端是基于内含C674xDSP信号处理器的芯片设计，工作频段为毫米波波段，发射功率12.5dBm，易于实现雷达前端的小型化设计。同时，带宽设置为4G；满足最大距离分辨测量能力。The millimeter wave radar front end of the present invention is based on the chip design containing the C674xDSP signal processor. The working frequency band is the millimeter wave band and the transmission power is 12.5dBm. It is easy to realize the miniaturization design of the radar front end. At the same time, the bandwidth is set to 4G to meet the maximum distance resolution measurement capability.
在一个示例中，本步骤的实现方式包括：In an example, this step is implemented by:
S11：安装毫米波雷达，并设置毫米波雷达的工作边界参数。S11: Install the millimeter wave radar and set the working boundary parameters of the millimeter wave radar.
具体的，包括：Specifically, include:
S111：安装毫米波雷达，毫米波雷达为墙壁安装，毫米波雷达的检测安装高度为h(即距地面h)，毫米波雷达的雷达辐射方向俯仰方向按照水平下倾斜α度，如图2所示。S111: Install the millimeter wave radar. The millimeter wave radar is installed on the wall. The detection installation height of the millimeter wave radar is h (that is, h from the ground). The pitch direction of the millimeter wave radar is tilted downward by α degree from the horizontal, as shown in Figure 2. Show.
S112：定义空间坐标系，将平行于毫米波雷达的探测方向的水平方向定义为X轴，将垂直于毫米波雷达的探测方向的水平方向定义为Y轴，将垂直方向定义为Z轴，如图2所示。S112: Define the spatial coordinate system, define the horizontal direction parallel to the detection direction of the millimeter wave radar as the X axis, define the horizontal direction perpendicular to the detection direction of the millimeter wave radar as the Y axis, and define the vertical direction as the Z axis, such as As shown in Figure 2.
S113：根据特定空间内测量异常行为需要，设置毫米波雷达的工作边界参数。S113: Set the working boundary parameters of the millimeter wave radar based on the need to measure abnormal behavior in a specific space.
其中，工作边界参数包括静态边界条件和动态边界条件，静态边界条件为X轴的x0至x1范围、Y轴的y0至y1范围和Z轴的z0至z1范围，去除地面、天花板、前墙、后墙及左右墙壁的影响。动态边界条件为X轴的x2至x3范围、Y轴的y2至y3范围和Z轴的z2至z3范围，该范围外的动目标点云数据较少，为提高测量准确，动态边界条件外部的目标不做异常行为分析。x0、x1、y0、y1、z0、z1、x2、x3、y2、y3、z2和z3根据待测区内测量异常行为的需要确定。Among them, the working boundary parameters include static boundary conditions and dynamic boundary conditions. The static boundary conditions are the x0 to x1 range of the X axis, the y0 to y1 range of the Y axis, and the z0 to z1 range of the Z axis. The ground , ceiling, front wall, back wall and left and right walls.The dynamicboundary conditions are the rangeof x2 tox 3on the , targets outside the dynamic boundary conditions are not analyzed for abnormal behavior. x0 , x1 , y0 , y1 , z0 , z1 , x2 , x3 , y2 , y3 , z2 and z3 are determined according to the need to measure abnormal behavior in the area to be measured.
S12：对毫米波雷达的工作参数进行配置，利用毫米波雷达MIMO天线技术进行静态目标和动态目标的点云数据采集。S12: Configure the working parameters of the millimeter wave radar, and use the millimeter wave radar MIMO antenna technology to collect point cloud data of static targets and dynamic targets.
具体的，包括：Specifically, include:
S121：配置雷达工作调谐Chirp参数和配置雷达帧(Frame)工作参数。S121: Configure radar operation tuning Chirp parameters and configure radar frame (Frame) operation parameters.
雷达带宽设计为4GHz，Chirp周期50us；距离测量分辨率为4cm，Frame参数设置满足低速运动目标速度提取，速度测量分辨率0.17m/s；采集人体静态散点分布和动态特征参数，可以较好的实现动目标异常行为检测。The radar bandwidth is designed to be 4GHz, the Chirp period is 50us; the distance measurement resolution is 4cm, the Frame parameter setting meets the speed extraction of low-speed moving targets, and the speed measurement resolution is 0.17m/s; the static scatter distribution and dynamic characteristic parameters of the human body are collected, which can be better To achieve abnormal behavior detection of moving targets.
S122：通过毫米波雷达MIMO天线技术，虚拟出方位向、俯仰向各有N个接收天线，并进行点云数据采集。S122: Using millimeter wave radar MIMO antenna technology, N receiving antennas are virtualized in the azimuth and pitch directions, and point cloud data is collected.
毫米波雷达体制设计采用M发N收天线体制，通过毫米波雷达多输入多输出(Multiple-Input Multiple-Output，MIMO)天线技术，可虚拟出方位向、俯仰向各有N个接收天线；方位向和俯仰方向角度分辨率为25度，丰富点云数据，提高了两个维度的空间探测能力。The millimeter wave radar system design adopts an M transmitting and N receiving antenna system. Through the millimeter wave radar multiple-input multiple-output (MIMO) antenna technology, N receiving antennas can be virtualized in the azimuth and pitch directions; azimuth The angle resolution in the direction and pitch direction is 25 degrees, which enriches the point cloud data and improves the spatial detection capability in two dimensions.
S2：根据采集的点云数据，分析静态目标和动态目标的数据特征，实现待测区内人员数量统计、定位和跟踪。S2: Based on the collected point cloud data, analyze the data characteristics of static targets and dynamic targets to achieve statistics, positioning and tracking of the number of people in the area to be measured.
在一个示例中，本步骤包括：根据人员目标不同，进行距离维FFT(快速傅里叶变换)、水平方向Capon BF(最小方差无畸变响应波束形成)、距离/水平方向CFAR(恒虚警率检测，Constant False-Alarm Rate)、垂直方向BF(Beam Forming，波束形成)和垂直角度估计、速度估计以及聚类跟踪，如图3所示。In one example, this step includes: performing range-dimensional FFT (Fast Fourier Transform), horizontal direction Capon BF (minimum variance distortion-free response beamforming), range/horizontal direction CFAR (constant false alarm rate) according to different human targets Detection, Constant False-Alarm Rate), vertical BF (Beam Forming, beam forming) and vertical angle estimation, speed estimation and cluster tracking, as shown in Figure 3.
具体的，距离维FFT包括：Specifically, the distance dimension FFT includes:
对每个接收天线的原始ADC(Analog Digital Converter,模拟数字转换器)数据进行一维加窗(1D Windowing)和一维FFT(1D FFT)。Perform one-dimensional windowing (1D Windowing) and one-dimensional FFT (1D FFT) on the original ADC (Analog Digital Converter, analog-to-digital converter) data of each receiving antenna.
水平方向Capon BF包括：Horizontal Capon BF includes:
首先进行静态载波去除，并在水平方向的N个接收天线上计算空间相关矩阵，最后通过Capon BF生成距离/水平方向热图(Heat map)。First, static carrier removal is performed, and the spatial correlation matrix is calculated on N receiving antennas in the horizontal direction. Finally, a distance/horizontal heat map (Heat map) is generated through Capon BF.
距离/水平方向CFAR包括：Distance/horizontal CFAR includes:
通过二维CFAR，在距离/水平方向热图上检测出目标点，得到包含各个目标点的距离和方位角的第一组点云。Through two-dimensional CFAR, the target points are detected on the distance/horizontal heat map, and the first set of point clouds containing the distance and azimuth angle of each target point are obtained.
垂直方向BF和垂直角度估计包括：Vertical BF and vertical angle estimates include:
在第一组点云的每个目标点上计算所有接收天线上的空间相关矩阵，并通过垂直方向的BF生成目标点在垂直方向的热图，在热图上通过峰值搜索获得目标点的垂直高度，得到包含各个目标点的距离、方位角和俯仰角的第二组点云。Calculate the spatial correlation matrix on all receiving antennas on each target point of the first set of point clouds, and generate a heat map of the target point in the vertical direction through BF in the vertical direction. Obtain the vertical direction of the target point through peak search on the heat map. height, and obtain a second set of point clouds containing the distance, azimuth angle and pitch angle of each target point.
速度估计包括：Speed estimates include:
对第二组点云的每个目标点计算空间BF向量，利用BF向量对目标点距离bin(单元)进行空间滤波并计算Doppler谱(多普勒光谱学)，在Doppler谱上通过峰值搜索获得目标点的速度，得到包含各个目标点的距离、方位角(即水平角)、俯仰角(即垂直角)和速度的第三组点云。Calculate the spatial BF vector for each target point of the second group of point clouds, use the BF vector to perform spatial filtering on the target point distance bin (unit) and calculate the Doppler spectrum (Doppler spectroscopy), which is obtained by peak search on the Doppler spectrum The speed of the target point is used to obtain a third set of point clouds containing the distance, azimuth angle (i.e. horizontal angle), pitch angle (i.e. vertical angle) and speed of each target point.
聚类跟踪包括：Cluster tracking includes:
对第三组点云，通过聚类跟踪算法进行目标聚类，并对目标进行跟踪和预测，得到跟踪目标列表，跟踪目标列表中每个目标均包含空间位置(即坐标)，速度和加速度信息。For the third group of point clouds, target clustering is performed through the cluster tracking algorithm, and the targets are tracked and predicted to obtain a tracking target list. Each target in the tracking target list contains spatial position (i.e. coordinates), speed and acceleration information. .
其中，进行目标聚类的方法包括：Among them, methods for target clustering include:
结合待测区的特定环境，采用基于无监督学习的K均值聚类算法(K-meansClustering Algorithm)，通过不同粒度的聚类处理，灵活实现不同特征值间耦合关系的学习，实现对目标特征的有效融合与识别。Combined with the specific environment of the area to be tested, the K-means Clustering Algorithm based on unsupervised learning is used to flexibly realize the learning of the coupling relationship between different feature values through clustering processing at different granularities, and realize the target features. Effective fusion and identification.
对目标进行跟踪和预测的方法包括：Methods for tracking and predicting targets include:
利用特征跟踪算法，在目标匹配中，采用最近邻目标匹配跟踪算法，考虑待跟踪目标与下一帧每个目标的质心的欧氏距离，将欧式距离最小的两个目标认为是同一个目标。Using the feature tracking algorithm, in the target matching, the nearest neighbor target matching tracking algorithm is used, considering the Euclidean distance between the target to be tracked and the centroid of each target in the next frame, and the two targets with the smallest Euclidean distance are considered to be the same target.
进一步的，在聚类跟踪之前还包括：Further, before cluster tracking, it also includes:
对第三组点云，利用平均相消算法实现静态目标的滤除，得到极坐标下多普勒时频谱图像，并对动态目标测量数据进行雷达极坐标到空间坐标系的转换。For the third group of point clouds, the average cancellation algorithm is used to filter out static targets, obtain Doppler time spectrum images in polar coordinates, and convert the dynamic target measurement data from radar polar coordinates to spatial coordinate systems.
利用平均相消算法实现静态目标的滤除时，采用基于向量叠加平均作为背景。第三组点云是雷达极坐标系下的点云图像，其各个点的位置和运动通过距离、方位角、俯仰角和速度参数表示，需要将其转换到建立的空间坐标系上，即实现X、Y、Z、速度的转换。When using the average cancellation algorithm to filter out static targets, the vector-based superposition average is used as the background. The third group of point clouds are point cloud images in the radar polar coordinate system. The position and movement of each point are represented by distance, azimuth angle, pitch angle and velocity parameters. They need to be converted to the established spatial coordinate system, that is, to achieve X, Y, Z, speed conversion.
S3：根据静态目标和动态目标各维度速度、加速度分析，参照人员日常行为特征参数，实现人体异常行为判定。S3: Based on the speed and acceleration analysis of each dimension of static targets and dynamic targets, and with reference to the daily behavioral characteristic parameters of personnel, the abnormal human behavior determination is realized.
本步骤通过分析特征目标特征点的数据特征，速度变化、加速度变化，分析目标快速蹲下、跌倒、挥动手臂等行为。具体的：This step analyzes the data characteristics of the characteristic target feature points, speed changes, and acceleration changes to analyze the target's behaviors such as squatting, falling, and waving arms quickly. specific:
S31：雷达墙体侧壁安装时，X、Y轴为水平坐标平面，对跟踪目标列表的X轴和Y轴的信息进行微分，得到人员运动速度和肢体摆动最大速度等，实现水平方向的人体行为判定.S31: When the radar is installed on the side wall of the wall, the X and Y axes are horizontal coordinate planes. Differentiate the X-axis and Y-axis information of the tracking target list to obtain the person's movement speed and the maximum speed of limb swing, etc., to realize the horizontal direction of the human body. Behavioral judgment.
S32：结合人体日常行为特点，分析跟踪目标列表的Z轴的信息。实现垂直方向的人体行为判定。S32: Combined with the daily behavioral characteristics of the human body, analyze the Z-axis information of the tracking target list. Realize human behavior judgment in vertical direction.
Z轴为地面高度维，结合人们日常行为特点，分析高度维数据特征，即分析距离、多维度速度、加速度等特征，通过分析数据特征如速度变化、加速度变化等，判别人员异常行为，可实现判断跌倒、蹲下等垂直方向的行为检测。The Z-axis is the height dimension of the ground. Combined with the characteristics of people's daily behaviors, the height-dimensional data characteristics are analyzed, that is, the characteristics of distance, multi-dimensional speed, acceleration, etc. are analyzed. By analyzing the data characteristics such as speed changes, acceleration changes, etc., it is possible to identify abnormal human behaviors. Detect vertical behaviors such as falling and squatting.
S33：参照人员日常行为特征参数，对判定得到的水平方向的人体行为和垂直方向的人体行为进行人体异常行为判定。S33: Determine abnormal human behavior based on the determined human body behavior in the horizontal direction and the human body behavior in the vertical direction with reference to the daily behavioral characteristic parameters of the person.
本发明根据人员目标各维度速度、加速度分析，参照人员日常行为特征参数，实现人体异常行为判定，使得区域内检测工作能够更加准确、灵活及高效。实现如银行ATM、自助营业厅内暴力抢劫、人员长时间滞留等特定环境区域的异常行为检测与判别。检测结果上传上位机，在上位机输出显示人员数量统计、定位、跟踪及行为特征检测结果和人体行为判别结果。The invention realizes the determination of abnormal human behavior based on the analysis of the speed and acceleration of each dimension of the human target and with reference to the daily behavioral characteristic parameters of the human body, making the detection work in the area more accurate, flexible and efficient. Realize the detection and identification of abnormal behaviors in specific environmental areas such as bank ATMs, self-service business halls, violent robberies, and people staying for long periods of time. The detection results are uploaded to the host computer, and the host computer outputs and displays personnel statistics, positioning, tracking, behavioral characteristics detection results and human behavior discrimination results.
本发明是一种非接触式、高分辨率的人员异常行为快速检测方法，可用于特定区域人员特征提取及异常行为的识别。通过将毫米波雷达及MIMO技术结合使用，实现快速波束赋形和扫描，解决了现有人员异常行为检测与识别系统中存在的分辨率低和环境适应能力差的问题，实现具有高分辨的毫米波雷达点云图像，提高对区域人员异常行为的特征检测和识别能力，在实现快速精准检测的同时具有低功耗、小型化等优点。本发明适用于多种环境多人员数量的特征提取及异常行为检测判别，可根据用户需求进行自适应设计，具有良好的可拓展性。The invention is a non-contact, high-resolution rapid detection method of abnormal human behavior, which can be used to extract human features and identify abnormal behaviors in specific areas. By combining millimeter wave radar and MIMO technology, rapid beamforming and scanning are achieved, which solves the problems of low resolution and poor environmental adaptability in existing abnormal human behavior detection and identification systems, and achieves high-resolution millimeter Wave radar point cloud images improve the feature detection and identification capabilities of abnormal behaviors of people in the area. It not only achieves fast and accurate detection, but also has the advantages of low power consumption and miniaturization. The invention is suitable for feature extraction and abnormal behavior detection and discrimination in multiple environments and multiple numbers of people. It can be designed adaptively according to user needs and has good scalability.
应说明的是：以上所述实施例，仅为本发明的具体实施方式，用以说明本发明的技术方案，而非对其限制，本发明的保护范围并不局限于此，尽管参照前述实施例对本发明进行了详细的说明。本领域的普通技术人员应当理解：任何熟悉本技术领域的技术人员在本发明揭露的技术范围内，其依然可以对前述实施例所记载的技术方案进行修改或可轻易想到变化，或者对其中部分技术特征进行等同替换；而这些修改、变化或者替换，并不使相应技术方案的本质脱离本发明实施例技术方案的精神和范围。都应涵盖在本发明的保护范围之内。It should be noted that the above-mentioned embodiments are only specific implementations of the present invention and are used to illustrate the technical solutions of the present invention rather than to limit them. The protection scope of the present invention is not limited thereto. Although reference is made to the foregoing implementations Examples illustrate the invention in detail. Those of ordinary skill in the art should understand that any person familiar with the art can still modify or easily think of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed by the present invention, or change part of them. The technical features are equivalently substituted; and these modifications, changes or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. All are covered by the protection scope of the present invention.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311724729.9ACN117690185A (en) | 2023-12-14 | 2023-12-14 | Personnel feature extraction and abnormal behavior discrimination method based on millimeter wave radar |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311724729.9ACN117690185A (en) | 2023-12-14 | 2023-12-14 | Personnel feature extraction and abnormal behavior discrimination method based on millimeter wave radar |
| Publication Number | Publication Date |
|---|---|
| CN117690185Atrue CN117690185A (en) | 2024-03-12 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311724729.9APendingCN117690185A (en) | 2023-12-14 | 2023-12-14 | Personnel feature extraction and abnormal behavior discrimination method based on millimeter wave radar |
| Country | Link |
|---|---|
| CN (1) | CN117690185A (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118609203A (en)* | 2024-04-22 | 2024-09-06 | 安徽大学 | A group behavior detection system and method based on laser radar point cloud data |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118609203A (en)* | 2024-04-22 | 2024-09-06 | 安徽大学 | A group behavior detection system and method based on laser radar point cloud data |
| Publication | Publication Date | Title |
|---|---|---|
| CN110118966B (en) | Personnel detection and counting system based on millimeter wave radar | |
| CN113093170B (en) | Millimeter wave radar indoor personnel detection method based on KNN algorithm | |
| CN112782664A (en) | Toilet fall detection method based on millimeter wave radar | |
| CN112859033A (en) | Target detection method, device and related equipment | |
| Li et al. | Robust human targets tracking for MIMO through-wall radar via multi-algorithm fusion | |
| CN112946630A (en) | Personnel counting and tracking method based on millimeter wave radar | |
| CN113835074B (en) | Dynamic people flow monitoring method based on millimeter wave radar | |
| Kalyanaraman et al. | Forma track: tracking people based on body shape | |
| CN116125458A (en) | Personnel positioning method based on millimeter wave radar | |
| WO2022130350A1 (en) | Radar detection and tracking | |
| CN106443625B (en) | Based on the one-dimensional target identification method merged as information of high-resolution | |
| CN105913074A (en) | Combined SAR image moving target clustering method based on amplitude and radial speed | |
| CN117690185A (en) | Personnel feature extraction and abnormal behavior discrimination method based on millimeter wave radar | |
| Zhang et al. | Vector phase symmetry for stable hyperbola detection in ground-penetrating radar images | |
| Zhang et al. | Waffle: A waterproof mmwave-based human sensing system inside bathrooms with running water | |
| Liu et al. | An intelligent signal processing method for motional vital signs detection system based on deep learning | |
| Zhou et al. | Human body recognition based on the sparse point cloud data from MIMO millimeter-wave radar for smart home | |
| Aytutuldu et al. | Indoor human counting based on range-Doppler spectrum in the presence of material penetration and occlusion | |
| Raeis et al. | InARMS: Individual activity recognition of multiple subjects with FMCW radar | |
| Kumar et al. | UAV detection multi-sensor data fusion | |
| Sun et al. | Omnidirectional Human Behavior Recognition Method Based on Frequency-Modulated Continuous-Wave Radar | |
| CN117214854A (en) | Debris flow detection method based on millimeter wave radar | |
| TWI772208B (en) | Method for counting number of people based on mmwave radar | |
| EP4184074A1 (en) | Sneezing sensing method based on wireless signal, and related apparatus | |
| Choi et al. | Enhanced Radar False Alarm Mitigation in Low-RCS Target Detection Using Time-Varying Trajectories on Range-Doppler Diagrams with DCNN |
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |