Dynamic target detection method, device, computer equipment and storage medium

Abstract

The present application relates to a method, apparatus, computer device, and storage medium for dynamic target detection. The method is applied to a millimeter-wave radar and includes: obtaining signal reception data of at least one receiving antenna of the millimeter-wave radar within a preset time range; performing background noise and dynamic target analysis based on the signal reception data; when there is a dynamic target in the signal reception data, generating a point cloud energy map with static components removed based on the signal reception data of at least one of the receiving antennas; when the dynamic target is within a preset defense area range of the point cloud energy map, inputting the point cloud energy map into a pre-trained target classification model to obtain a detection result of the dynamic target. Using this method can reduce the amount of calculation data and time consumption.

Description

Technical Field

[0001] The present application relates to the technical field of radar detection, and particularly to a method, device, computer device and storage medium for dynamic target detection. Background Art

[0002] As a radar operating in the millimeter wave band, a millimeter wave radar can calculate information such as the respiration and heart rate of a human body by collecting and analyzing phase information, achieving non-contact measurement of human body information. However, respiration and heart rate signals are tiny signals and are thus extremely vulnerable to environmental interference, such as reflection interference caused by strong reflection targets like iron rods and iron sheets, and may also be affected by other living bodies other than humans, such as pets. Currently, in traditional technologies, interference is excluded by establishing a grid map. However, this method requires collecting a large amount of scan data, and the computational amount for establishing the grid map is huge, resulting in an increase in detection time.

[0003] Therefore, there are still problems of large computational data volume and long time consumption when using a radar for non-contact measurement of human body information currently. Summary of the Invention

[0004] Based on this, in view of the above technical problems, it is necessary to provide a method, device, computer device and storage medium for dynamic target detection that can reduce the computational data volume and time consumption.

[0005] In a first aspect, the present application provides a method for dynamic target detection, which is applied to a millimeter wave radar. The method for dynamic target detection includes:

[0006] Obtain signal reception data of at least one receiving antenna of the millimeter wave radar within a preset time range;

[0007] Perform background noise and dynamic target analysis based on the signal reception data; when there is a dynamic target in the signal reception data, generate a point cloud energy map with static components removed based on the signal reception data of at least one of the receiving antennas;

[0008] When the dynamic target is within a preset defense area range of the point cloud energy map, input the point cloud energy map into a pre-trained target classification model to obtain a detection result of the dynamic target.

[0009] In one embodiment, the obtaining signal reception data of at least one receiving antenna of the millimeter wave radar within a preset time range includes:

[0010] Obtain range gate energy data obtained by the receiving antenna in each frame;

[0011] Based on the range gate energy data of each frame within the preset time range, construct two-dimensional data with the number of frames and the number of range gates as dimensions;

[0012] Use the two-dimensional data as the signal reception data of the receiving antenna within a preset time range.

[0013] In one embodiment, the analyzing background noise and dynamic targets based on the signal reception data includes:

[0014] Remove the static components of the signal reception data based on the average energy of each range gate in the signal reception data to obtain dynamic energy data;

[0015] Based on the dynamic energy data, obtain the maximum and minimum energy data; the maximum and minimum energy data includes: the maximum energy range gate, the maximum energy value, the minimum energy range gate, and the minimum energy value;

[0016] Based on the maximum and minimum energy data, determine the slope of the maximum and minimum change of the dynamic energy data;

[0017] Based on the slope of the maximum and minimum change and the maximum and minimum energy data, determine whether there are dynamic targets in the signal reception data.

[0018] In one embodiment, the determining whether there are dynamic targets in the signal reception data based on the slope of the maximum and minimum change and the maximum and minimum energy data includes:

[0019] If the slope of the maximum and minimum change and the maximum and minimum energy data meet the first background noise condition or the second background noise condition, then there is only background noise in the signal reception data; the first background noise condition includes: the slope of the maximum and minimum change is less than a preset slope threshold, or the maximum energy value is less than a first energy threshold; the second background noise condition includes: the sum of the maximum energy range gate and the minimum energy range gate is less than a preset range gate threshold, and the maximum energy value is less than a second energy threshold;

[0020] If the number of times that the slope of the maximum and minimum change and the maximum and minimum energy data do not meet the first background noise condition and the second background noise condition simultaneously reaches a preset number within a preset time range, then there are dynamic targets in the signal reception data.

[0021] In one embodiment, the generating a point cloud energy map with static components removed based on the signal reception data of at least one receiving antenna includes:

[0022] Calculate the energy difference between the energy value of each frame and the average energy of all frames under the same range gate of the same receiving antenna respectively to obtain dynamic three-dimensional data with static components removed; the dynamic three-dimensional data is three-dimensional data with the number of receiving antennas, the number of frames, and the number of range gates as dimensions;

[0023] Perform beamforming on the dynamic three-dimensional data based on a preset steering vector to obtain a range-angle energy distribution map;

[0024] Perform point cloud conversion based on the range-angle energy distribution map to obtain the point cloud energy map.

[0025] In one embodiment, the millimeter-wave radar includes at least two receiving antennas. Before performing beamforming on the dynamic three-dimensional data based on a preset steering vector to obtain a range-angle energy distribution map, it includes:

[0026] Determine the weight value of each receiving antenna at each receiving angle based on the number of receiving antennas, the spacing between adjacent receiving antennas, and the receiving angle range;

[0027] Obtain a preset steering vector with the number of receiving antennas and the receiving angle range as dimensions based on the weight values of multiple receiving antennas at multiple receiving angles.

[0028] In one embodiment, the dynamic target detection method further includes:

[0029] Obtain the number of target data points within a preset defense area in the point cloud energy map; the energy value of the target data point is greater than the energy values of multiple surrounding data points, and the signal-to-noise ratio of the target data point is higher than a preset signal-to-noise ratio threshold; the signal-to-noise ratio is determined based on the energy value of the target data point and a preset noise coefficient;

[0030] If the number of target data points within the preset defense area is at least one, then the dynamic target is located within the preset defense area of the point cloud energy map.

[0031] In one embodiment, before inputting the point cloud energy map into a pre-trained target classification model to obtain the detection result of the dynamic target, it further includes:

[0032] Construct a training data sample set; the training data sample set includes various types of point cloud data samples;

[0033] Perform transfer learning on a pre-trained deep learning model based on the training data sample set, an Adam optimizer, and a cross-entropy loss function to obtain the pre-trained target classification model.

[0034] In one embodiment, inputting the point cloud energy map into a pre-trained target classification model to obtain the detection result of the dynamic target includes:

[0035] Perform vital sign signal detection based on the point cloud energy map;

[0036] When the breathing signs and heart rate signs of the dynamic target meet the preset sign conditions, classify the dynamic target based on the target classification model to obtain the classification result of the dynamic target; the classification result includes human or animal.

[0037] Based on the classification result, determine the detection result of the dynamic target.

[0038] In a second aspect, the present application provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the method described above is implemented.

[0039] In a third aspect, the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method described above is implemented.

[0040] In a fourth aspect, the present application provides a computer program product. When the computer program product is executed by a processor, the method described above is implemented.

[0041] The above dynamic target detection method, device, computer device and storage medium obtain the signal reception data of at least one receiving antenna of a millimeter-wave radar within a preset time range; perform background noise and dynamic target analysis based on the signal reception data. When there is a dynamic target in the signal reception data, generate a point cloud energy map with static components removed based on the signal reception data of at least one of the receiving antennas; when the dynamic target is within the preset defense area range of the point cloud energy map, input the point cloud energy map into a pre-trained target classification model to obtain the detection result of the dynamic target. Among them, by distinguishing background noise and dynamic targets, and through the setting of the preset defense area range, irrelevant information can be effectively excluded, and dynamic targets within the preset defense area range can be analyzed centrally, thereby reducing the impact of environmental interference on the detection result. By using the target classification model to obtain the detection result, the impact of other animals such as pets on sign detection can also be effectively excluded, achieving the effect of improving the detection accuracy without constructing a grid map, reducing the amount of calculation data, and reducing the time consumption. Description of the Drawings

[0042] Figure 1 It is an application environment diagram of the dynamic target detection method in an embodiment;

[0043] Figure 2 It is a flowchart of the dynamic target detection method in an embodiment;

[0044] Figure 3 It is a flowchart of the dynamic target detection method in another embodiment;

[0045] Figure 4 Schematic diagram of updating a two-dimensional data storage matrix in an embodiment;

[0046] Figure 5 Schematic diagram of the process of data processing and background noise judgment in an embodiment;

[0047] Figure 6 Schematic diagram of the process of point cloud output and defense area setting in an embodiment;

[0048] Figure 7 Schematic diagram of the process of external interference judgment for the defense area in an embodiment;

[0049] Figure 8 Schematic diagram of the process of deep learning sample judgment in an embodiment;

[0050] Figure 9 Internal structure diagram of a computer device in an embodiment. Detailed implementation manners

[0051] In order to make the objectives, technical solutions, and advantages of the present application clearer and more understandable, the following further details the present application in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0052] The dynamic target detection method provided by the embodiments of the present application can be applied to, for example, Figure 1 the application environment shown. Among them, the terminal 102 communicates with the millimeter-wave radar 104 through the network. The millimeter-wave radar 104 emits signals in the millimeter-wave frequency band to the detection area and generates signal reception data based on the received signals. The terminal 102 obtains the signal reception data of at least one receiving antenna of the millimeter-wave radar within a preset time range; performs background noise and dynamic target analysis based on the signal reception data. When there is a dynamic target in the signal reception data, a point cloud energy map with static components removed is generated based on the signal reception data of at least one of the receiving antennas; when the dynamic target is within the preset defense area range of the point cloud energy map, the point cloud energy map is input into a pre-trained target classification model to obtain the detection result of the dynamic target. Among them, the terminal 102 can be, but is not limited to, various personal computers, laptop computers, smart phones, tablet computers, Internet of Things devices, and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart vehicle-mounted devices, etc. The portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc.

[0053] In one embodiment, as Figure 2 shown, a dynamic target detection method is provided, which is applied to a millimeter-wave radar. For this method applied to Figure 1The terminal 102 in the example is used as an example to illustrate, and the following steps are included:

[0054] Step S100: acquiring signal reception data of at least one receiving antenna of the millimeter wave radar within a preset time range.

[0055] Among them, millimeter-wave radar is a radar system that works in the millimeter wave frequency band, with high resolution, strong penetration, and the ability to detect small movements. The receiving antenna is used to capture the signal reflected by the target. The preset time range can be a pre-set time window, during which data from the received signal is continuously collected. The time range can be selected according to actual needs.

[0056] Step S200, performing background noise and dynamic target analysis based on the signal reception data; when there are dynamic targets in the signal reception data, generating a point cloud energy spectrum with static components removed based on the signal reception data of at least one of the receiving antennas.

[0057] The background noise may include signals reflected by static targets in the environment, and may also include random signals generated by non-target sources in the environment, such as thermal noise, interference from electronic equipment, etc. A dynamic target may be a target that moves or changes relative to the static background, for example, the rise and fall of a person's chest caused by breathing and heartbeat.

[0058] In this embodiment, background noise and dynamic target analysis may be used to determine whether there is a dynamic target in the signal reception data. When there is a dynamic target in the signal reception data, a point cloud energy spectrum is generated according to the signal reception data of at least one of the receiving antennas.

[0059] The point cloud energy spectrum can be a graphical representation of the target location and its energy distribution extracted from the signal reception data. Through the analysis of background noise and real dynamic targets, the important parts of the signal reception data can be effectively determined, and the main resources can be focused on the actual active objects, thereby improving the accuracy of recognition.

[0060] The static components may be removed by performing differentiation between consecutive frames, screening energy values, etc. The static components may be removed from the signal receiving data first, or after obtaining the point cloud energy spectrum, or a combination of the two may be performed, which is not limited in this embodiment.

[0061] Step S300, when the dynamic target is located within the preset defense zone of the point cloud energy spectrum, the point cloud energy spectrum is input into a pre-trained target classification model to obtain a detection result of the dynamic target.

[0062] Among them, determining whether a dynamic target is within a preset defense area range of the point cloud energy map can be a data point with a high energy value in the point cloud energy map, and it is determined whether this data point falls within the area delimited by the preset defense area range.

[0063] The preset defense area range can be a pre-delimited spatial area. The preset defense area range can be used to delimit the boundaries of the surrounding environment and the target detection area, or can be used to distinguish different types of interference, such as distinguishing the interference of living body activities and environmental noise interference, etc.

[0064] The pre-trained target classification model can be a model pre-trained using a machine learning model or a deep learning algorithm, so as to be able to identify different types of dynamic targets, such as people and other animals, etc.

[0065] The detection result can include the classification result of the dynamic target, and can also include physical sign information such as the respiration and heart rate of the dynamic target.

[0066] A dynamic target detection method provided in this embodiment includes: obtaining signal reception data of at least one receiving antenna of a millimeter-wave radar within a preset time range; performing background noise and dynamic target analysis based on the signal reception data. When there is a dynamic target in the signal reception data, generating a point cloud energy map with static components removed based on the signal reception data of at least one of the receiving antennas; when the dynamic target is within the preset defense area range of the point cloud energy map, inputting the point cloud energy map into the pre-trained target classification model to obtain the detection result of the dynamic target. Among them, by distinguishing background noise and dynamic targets, and through the setting of the preset defense area range, irrelevant information can be effectively excluded, and the dynamic targets within the preset defense area range can be centrally analyzed, thereby reducing the influence of environmental interference on the detection result. By using the target classification model to obtain the detection result, the influence caused by other animals such as pets on physical sign detection can also be effectively excluded, realizing the improvement of the detection accuracy without constructing a grid map, achieving the effects of reducing the amount of calculation data and reducing the time consumption.

[0067] In one embodiment, the obtaining signal reception data of at least one receiving antenna of the millimeter-wave radar within a preset time range includes:

[0068] Obtaining the range gate energy data obtained by the receiving antenna in each frame;

[0069] Based on the range gate energy data of each frame within the preset time range, constructing a two-dimensional data with the number of frames and the number of range gates as dimensions;

[0070] Taking the two-dimensional data as the signal reception data of the receiving antenna within the preset time range.

[0071] Among them, one frame can be a complete scan or sampling period completed by the radar within a specific time interval. The range gate is determined by the time delay between the transmitted signal and the received signal and can be used to determine the distance of the target. Each range gate corresponds to a specific distance interval, and each range gate can respectively correspond to the energy value of the measured reflected signal.

[0072] Within a preset time range, the radar transmits a transmitted signal in the millimeter wave band and receives the echo signals returned under different range gates, so as to obtain the range gate energy data obtained by the receiving antenna in each frame. The range gate energy data of each frame can reflect whether there is a target under each specific range gate and the corresponding intensity information.

[0073] A two-dimensional data with the number of frames and the number of range gates as dimensions can be constructed. It can be constructed with the number of frames as rows and the number of range gates as columns for the two-dimensional data, or it can be constructed with the number of frames as columns and the number of range gates as rows for the two-dimensional data. By constructing the two-dimensional data, the range gate energy information of all frames of the receiving antenna within the preset time range can be covered, thus completing a complete and continuous spatio-temporal data record.

[0074] Further, when entering the next moment, this moment is recorded as the second moment. It can be to overwrite the range gate energy data of the frame corresponding to the second moment on the range gate energy data of the earliest frame in the two-dimensional data, so as to realize data update on the basis of the existing two-dimensional data and avoid the extra time consumption caused by re-collecting the range gate energy data within the preset time range.

[0075] A dynamic target detection method provided in this embodiment, by constructing a two-dimensional data with the number of frames and the number of range gates as dimensions, enables the two-dimensional data to cover the range gate energy information of all frames within the preset time range, forming a complete and continuous spatio-temporal data collection, thereby improving the environmental assessment speed, improving the analysis speed of background noise and dynamic targets, and achieving the effect of reducing time consumption.

[0076] In one of the embodiments, the analysis of background noise and dynamic targets based on the signal reception data includes:

[0077] Removing the static components of the signal reception data based on the average energy of each range gate in the signal reception data to obtain dynamic energy data;

[0078] Obtaining the maximum and minimum energy data based on the dynamic energy data;

[0079] Determining the maximum and minimum change slopes of the dynamic energy data based on the maximum and minimum energy data;

[0080] Based on the slope of the extreme value change and the extreme value energy data, determine whether there is a dynamic target in the signal reception data.

[0081] Among them, the average energy value can be the average of the energy values corresponding to each range gate. The static component represents the fixed or slowly changing part of the environment, and the static part usually appears as a relatively stable reflected signal in the time dimension in the signal reception data.

[0082] Based on the average energy value of each range gate in the signal reception data, the static component of the signal reception data can be removed, which can be subtracted from the signal reception data according to the average energy value. Further, further processing such as absolute value conversion and normalization can be performed on the subtracted result to obtain the dynamic energy data after removing the static component.

[0083] The extreme value energy data can be the extreme data in the dynamic energy data, which is used to reflect the extreme situation in the dynamic energy data. The extreme value energy data includes: the maximum energy range gate, the maximum energy value, the minimum energy range gate, and the minimum energy value. Among them, the maximum energy value is the largest energy value in the dynamic energy data, and the maximum energy range gate is the range gate position corresponding to the maximum energy value. Correspondingly, the minimum energy value is the smallest energy value in the dynamic energy data, and the minimum energy range gate is the range gate position corresponding to the minimum energy value. The extreme value energy data can be obtained by traversing the energy values corresponding to each range gate in the dynamic energy data.

[0084] The slope of the extreme value change can be the rate of change of the energy in the dynamic energy data. Exemplarily, it can be calculated by the ratio of the difference between the maximum energy value and the minimum energy value to the distance between the maximum energy range gate and the minimum energy range gate.

[0085] It can be understood that when there is a dynamic target in the signal reception data, it means that the energy value will change over time in the signal reception data, and then there will also be a slope of the extreme value change higher than the threshold in the corresponding dynamic energy data. Based on the slope of the extreme value change and the extreme value energy data, to determine whether there is a dynamic target in the signal reception data, it can be determined whether there is a dynamic target according to whether the slope of the extreme value change meets the preset slope threshold and in combination with the conditions satisfied by the extreme value energy data. Exemplarily, it can be determined whether there is a dynamic target in the signal reception data by judging whether both the slope of the extreme value change is greater than the preset slope threshold and the maximum energy value is greater than the preset energy threshold.

[0086] A dynamic target detection method provided by this embodiment can concentrate on identifying the dynamic part in the signal reception data by removing the static components from the signal reception data. By combining the slope of the maximum and minimum value changes and the maximum and minimum value energy data for calculating dynamic targets, it can ensure the effective identification of tiny dynamic targets in the environment, thereby achieving the effect of improving the accuracy and sensitivity of target detection.

[0087] In one embodiment, determining whether there is a dynamic target in the signal reception data based on the slope of the maximum and minimum value changes and the maximum and minimum value energy data includes:

[0088] If the slope of the maximum and minimum value changes and the maximum and minimum value energy data satisfy the first background noise condition or the second background noise condition, then there is only background noise in the signal reception data;

[0089] If the number of times that the slope of the maximum and minimum value changes and the maximum and minimum value energy data do not satisfy the first background noise condition and the second background noise condition simultaneously reaches a preset number within a preset time range, then there is a dynamic target in the signal reception data.

[0090] In this embodiment, the first background noise condition and the second background noise condition are used to determine whether the current slope of the maximum and minimum value changes and the maximum and minimum value energy data are only background noise. That is, in this embodiment, it is to inversely infer whether there is a dynamic target by judging whether the slope of the maximum and minimum value changes and the maximum and minimum value energy data satisfy the background noise situation.

[0091] The first background noise condition includes: the slope of the maximum and minimum value changes is less than a preset slope threshold, or the maximum energy value is less than a first energy threshold. When the first background noise condition is met, it means that the energy change in the monitored area is very gentle. Even if there is energy fluctuation, the intensity of the energy fluctuation is low. Therefore, there is no situation sufficient to be recognized as a dynamic target.

[0092] The second background noise condition includes: the sum of the maximum energy distance gate and the minimum energy distance gate is less than a preset distance gate threshold, and the maximum energy value is less than a second energy threshold. When the second background noise condition is met, it means that the current energy change may be caused by the characteristics of the local environment rather than a real dynamic target.

[0093] Furthermore, if the number of times that the slope of the maximum and minimum value changes and the maximum and minimum value energy data do not satisfy the first background noise condition and the second background noise condition simultaneously reaches a preset number within a preset time range, then it is considered that there is a dynamic target in the signal reception data. Among them, the preset time range for determining whether there is a dynamic target can be determined according to the recent occurrence situation of dynamic targets or the energy change situation. This preset time range can be the same as or different from the preset time range for signal reception data acquisition.

[0094] In this embodiment, not only the one-time energy fluctuation needs to be concerned, but also for the accurate judgment of the dynamic target, it is necessary to determine according to the change slope of the maximum value and the situation where the maximum value energy data does not meet the background noise condition within a preset time range. Therefore, higher reliability of the detection result can be achieved, and false alarms caused by instantaneous interference or short-term anomalies can be avoided.

[0095] A dynamic target detection method provided in this embodiment can ensure the accurate and effective identification of tiny dynamic targets in the environment, reduce the false alarm rate, and achieve the effect of improving the accuracy and sensitivity of target detection by judging whether the change slope of the maximum value and the maximum value energy data meet the background noise condition to determine whether there is a dynamic target.

[0096] In one of the embodiments, generating a point cloud energy map with static components removed based on the signal reception data of at least one of the receiving antennas includes:

[0097] Calculate the energy difference between the energy value of each frame and the average energy of all frames under the same distance gate of the same receiving antenna respectively to obtain dynamic three-dimensional data with static components removed;

[0098] Perform beamforming on the dynamic three-dimensional data based on a preset steering vector to obtain a distance-angle energy distribution map;

[0099] Perform point cloud conversion based on the distance-angle energy distribution map to obtain the point cloud energy map.

[0100] Among them, the dynamic three-dimensional data is three-dimensional data with the number of receiving antennas, the number of frames, and the number of distance gates as dimensions. The dynamic three-dimensional data with static components removed is that under the same distance gate of the same receiving antenna, the energy value of each frame is calculated by subtracting the average energy of all frames from the initial energy value.

[0101] The preset steering vector is a parameter for forming a beam and defines the direction of signal arrival. Performing beamforming on the dynamic three-dimensional data based on the preset steering vector to obtain a distance-angle energy distribution map can be achieved by weighted combination of the signals of multiple antennas through the preset steering vector, so as to enhance the signal strength in a specific direction and suppress interference from other directions.

[0102] Performing point cloud conversion based on the distance-angle energy distribution map to obtain the point cloud energy map can be to convert the data in the distance-angle energy distribution map into a series of discrete points, and each point has a specific position and an energy value target.

[0103] A dynamic target detection method provided by this embodiment can generate a distance-angle capability distribution map by removing static components from three-dimensional data, and thus obtain a point cloud energy map, which can ensure the extraction of effective target information from signal reception data and greatly reduce the influence of background noise, achieving the effect of improving the accuracy and sensitivity of target detection.

[0104] In one embodiment, before beamforming the dynamic three-dimensional data based on the preset steering vector to obtain a distance-angle energy distribution map, it includes:

[0105] Determine the weight value of each receiving antenna at each receiving angle based on the number of receiving antennas, the spacing between adjacent receiving antennas, and the receiving angle range;

[0106] Obtain a preset steering vector with the number of receiving antennas and the receiving angle range as dimensions based on the weight values of multiple receiving antennas at multiple receiving angles.

[0107] In this embodiment, the millimeter-wave radar includes at least two receiving antennas. The number of receiving antennas is the number of antennas used to receive signals in the millimeter-wave radar. More receiving antennas can provide a more refined and accurate angle resolution.

[0108] The spacing between adjacent receiving antennas can be the physical distance between two adjacent receiving antennas. It can be understood that the spacing between adjacent receiving antennas may affect the spatial resolution of the system.

[0109] The receiving angle range can be the angle interval in which the radar or receiving antenna can effectively detect a target.

[0110] Determining the weight value of each receiving antenna at each receiving angle based on the number of receiving antennas, the spacing between adjacent receiving antennas, and the receiving angle range can be determined according to the array antenna theory.

[0111] In a specific embodiment, the preset steering vector can be a two-dimensional matrix with the receiving antenna serial number and the receiving angle range as dimensions. The weight value of the Nth antenna at the angle θ can be , where e is the exponential, j is the imaginary unit, λ is the wavelength, and d is the spacing between adjacent receiving antennas.

[0112] Obtaining a preset steering vector with the number of receiving antennas and the receiving angle range as dimensions based on the weight values of multiple receiving antennas at multiple receiving angles can be to fill the two-dimensional matrix according to the receiving antenna and receiving angle corresponding to each weight value to construct the preset steering vector.

[0113] A dynamic target detection method provided in this embodiment constructs a preset steering vector according to the array antenna theory, ensuring the effectiveness and accuracy in the beamforming process, and can obtain a more accurate distance-angle energy distribution map, making the detection of dynamic targets more reliable and accurate, achieving the effect of improving the accuracy and sensitivity of target detection.

[0114] In one embodiment, the dynamic target detection method further includes:

[0115] Obtain the number of target data points within a preset defense area in the point cloud energy map;

[0116] If the number of target data points within the preset defense area is at least one, the dynamic target is located within the preset defense area of the point cloud energy map.

[0117] Among them, the target data points meet the following conditions: the energy value of the target data point is greater than the energy values of multiple surrounding data points, and the signal-to-noise ratio of the target data point is higher than a preset signal-to-noise ratio threshold. The signal-to-noise ratio is determined based on the energy value of the target data point and a preset noise coefficient. The preset signal-to-noise ratio threshold can be determined based on prior knowledge or the energy value of static components.

[0118] The energy value of the target data point being greater than the energy values of multiple surrounding data points means that there is a strong reflection source at the position corresponding to this point, and there is a high probability that there is an object or target. The signal-to-noise ratio exceeding the preset signal-to-noise ratio threshold indicates that the signal strength of this data point is stronger relative to the background noise, thus ensuring the reliability of target detection.

[0119] After obtaining the number of target data points, count the positions of each target data point in the point cloud energy map. If there are target data points within the defined preset defense area, it can be considered that the dynamic target is located within the preset defense area of the point cloud energy map.

[0120] Furthermore, according to the different numbers of target data points, the decision time limit of the point cloud can also be adjusted.

[0121] A dynamic target detection method provided in this embodiment determines whether a dynamic target is located within a preset defense area based on the number of target data points within the preset defense area in the point cloud energy map, and can effectively reduce false alarms and improve detection accuracy by screening through the signal-to-noise ratio, achieving the effect of improving the accuracy and sensitivity of target detection.

[0122] In one embodiment, before inputting the point cloud energy map into a pre-trained target classification model to obtain the detection result of the dynamic target, it further includes:

[0123] Construct a training data sample set;

[0124] Perform transfer learning on the pre-trained deep learning model based on the training data sample set, the Adam optimizer, and the cross-entropy loss function to obtain the pre-trained target classification model.

[0125] Among them, the training data sample set includes various types of point cloud data samples. These samples are used to train the machine learning model, and each training sample can be composed of input features and corresponding labels. Exemplarily, the input features can be pre-processed point cloud energy maps, and the labels can be target categories, such as people, pets, etc.

[0126] Performing transfer learning on the pre-trained deep learning model based on the training data sample set, the Adam optimizer, and the cross-entropy loss function can be to adopt the transfer learning method to fine-tune the pre-trained deep learning model through the training data sample set, so that the deep learning model can quickly learn the corresponding classification application scenarios.

[0127] During the training process, the Adam optimizer and the cross-entropy loss function can be used for model compilation. Among them, the Adam optimizer can handle sparse gradients and non-stationary targets, thus accelerating convergence to find the local optimal solution. The cross-entropy loss function can be used to measure the difference between the prediction result and the true label, so that the output probability distribution is closer to the true label distribution.

[0128] A dynamic target detection method provided in this embodiment can significantly improve the performance of the model, make it more suitable for the non-contact measurement task of human body information in the actual environment, and can also reduce the time and resource costs required for training by performing transfer learning on the pre-trained deep learning model based on the training data sample set, the Adam optimizer, and the cross-entropy loss function, so as to achieve the effect of improving the accuracy and sensitivity of target detection.

[0129] In one of the embodiments, the inputting the point cloud energy map into the pre-trained target classification model to obtain the detection result of the dynamic target includes:

[0130] Perform vital sign detection based on the point cloud energy map;

[0131] When the breathing vital signs and heart rate vital signs of the dynamic target meet the preset vital sign conditions, classify the dynamic target based on the target classification model to obtain the classification result of the dynamic target; the classification result includes people or animals;

[0132] Determine the detection result of the dynamic target based on the classification result.

[0133] Among them, the vital sign signals can be the respiratory and heart rate signs in the received signals. Detecting the vital sign signals can be to analyze the respiratory and heart rate signs to determine whether the vital sign signals meet the preset vital sign conditions. Exemplarily, the normal respiratory rate of an adult is usually between 12 and 20 breaths per minute, and the resting heart rate is usually between 60 and 100 beats per minute. Accordingly, the corresponding preset vital sign conditions can be set based on the above respiratory rate and resting heart rate.

[0134] When the vital sign signals meet the preset vital sign conditions, the dynamic target is a living body. Then, the dynamic target can be classified based on the target classification model to obtain a classification result.

[0135] Based on the classification result, determining the detection result of the dynamic target can be to feedback one or more of the classification result, location, signal strength, and vital sign signals of the dynamic target as the detection result, or to perform further analysis based on the above content and record it in the detection result.

[0136] The dynamic target detection method provided in this embodiment screens for living bodies by combining vital sign signal detection and preset vital sign conditions, and then classifies the dynamic target based on the target classification model of deep learning. It can reduce the number of times the target classification model is used, reduce the calculation amount, and maintain the accuracy rate. It can obtain important physiological parameters of the target without contacting the target, achieving the effects of improving the accuracy and sensitivity of target detection.

[0137] To more clearly elaborate the technical solution of the present application, the present application also provides a detailed embodiment.

[0138] In one embodiment, a dynamic target detection method is provided, including:

[0139] Step S1, distinguishing the background noise and moving or micro-moving targets by distance gate energy, as Figure 3 shown, including:

[0140] Step S11, data acquisition. Obtain the information of the target to be measured through one receiving antenna of the radar and save the data in an array named SA. The size of the data obtained in a single frame, that is, the number of distance gates, is NR. Therefore, the dimension of the array is 1D and the size is NR. Then, the target information obtained at the t-th moment can be expressed as:

[0141] SA = [RX1(t) RX2(t) RX3(t) RX n (t)]

[0142] Among them, n represents the position sequence of the target, RX n (t) represents the energy of the n-th distance gate of one receiving antenna, and the larger n is, the farther the corresponding distance is. t represents the number of frames.

[0143] Step S12, data accumulation. The number of accumulated frames is NF, the allocated number of rows is the number of accumulated frames NF, the number of columns is the number of range gates NR, and a two-dimensional matrix storage unit of size NF×NR is constructed and denoted as MA:

[0144] ;

[0145] Step S13, data update. As Figure 4 shown, shift the data of all columns in each row to the same position in the corresponding column of the previous row, that is, satisfy the formula:

[0146] MA[i - 1][j] = MA[i][j];

[0147] where i = 2, 3, …, NF - 1, j = 1, 2, …, NR, and the above operation needs to be performed NF - 1 times in total.

[0148] Step S14, data update. Repeat step S11, and assign the result SA of step S11 to the last row of the array MA in step S12, MA[NF][j] = SA[j], where j = 1, 2, …, NR is the range.

[0149] Step S2, data processing. As Figure 5 shown, perform stationary target removal.

[0150] Step S21, perform non-coherent accumulation on the array obtained in step S14 with the number of frames NF, and perform mean removal, modulus calculation, and summation on the NF-frame data corresponding to the chest cavity at the corresponding range. That is, take the two-dimensional array MA column by column, and add up the corresponding all rows (rows represent the number of frames) of each column (columns represent different range gates) respectively (i.e., add multiple frames) to obtain a one-dimensional array of size NR, denoted as ColSA.

[0151] Step S22, divide each value of the result ColSA in step S21 by the number of frames NF to obtain a one-dimensional average value array ColSA_Ave of size NR, denoted as:

[0152] ;

[0153] where FrameNum is the number of frames NF.

[0154] Step S23, subtract the data of each column in the two-dimensional accumulation array MA with the number of rows NF and the number of columns equal to the total number of range gates NR generated in step S14 from the corresponding column position of the one-dimensional average value array ColSA_Ave obtained in step S22, that is, satisfy the following equation:

[0155] MA[i][j] = MA[i][j] - ColSA_Ave[j];

[0156] where \(i = 2, 3, \cdots, N_F - 1\) and \(j = 1, 2, \cdots, N_R\).

[0157] Step S24: Take the absolute value of the result of step S23 to obtain a two-dimensional array \(MA_{ABS}\) after taking the absolute value, that is:

[0158] \(MA_{ABS}=|MA|\);

[0159] Step S25: Perform non-coherent accumulation on the array \(MA_{ABS}\) obtained in step S24 with the number of frames being \(FrameNum\), that is, add up all the rows (each row represents each frame) corresponding to each column (each column represents each different range bin) in the two-dimensional array \(MA_{ABS}\) to obtain a one-dimensional array of size \(N_R\), denoted as \(RBA\).

[0160] Step 3: Background noise judgment, for background noise and target discrimination.

[0161] Step 31: Use the mean sum result described in step S25 to obtain a one-dimensional array \(RBA\) of size \(N_R\), and obtain the maximum and minimum values of the accumulation and their corresponding position coordinates. Calculate the maximum value \(Max\_MA_{ABS}\) and the maximum value position \(MaxId\_MA_{ABS}\), the minimum value \(Min\_MA_{ABS}\) and the minimum value position \(MinId\_MA_{ABS}\) of \(MA_{ABS}\) respectively.

[0162] Step S32: Set an array \(Max\_MA_{ABS\_Buffer}\) of size 10 to store the results of the maximum value of \(Max\_MA_{ABS}\) for 10 consecutive seconds. Put the maximum value result of a single processing into \(Max\_MA_{ABS\_Buffer}\) and update it in real time.

[0163] Step S33: Calculate the difference between the maximum value \(Max\_MA_{ABS\_Buffer}[9]\) at the 10th second and the maximum value \(Max\_MA_{ABS\_Buffer}[0]\) at the 1st second, denoted as \(\Delta Max\_MA_{ABS}\), that is:

[0164] \(\Delta Max\_MA_{ABS}= Max\_MA_{ABS\_Buffer}[9]-Max\_MA_{ABS\_Buffer}[0]\)

[0165] Step S34: Respectively judge the relationship between the head and tail difference \(\Delta Max\_MA_{ABS}\) and the threshold and the relationship between the maximum value \(Max\_MA_{ABS}\) of the current one-dimensional array \(RBA\) and the threshold. If one of them is greater than the threshold, reset the point cloud decision delay to 10, otherwise execute S35.

[0166] Step S35: Calculate the slope \(Slope\).

[0167] ;

[0168] Among them, the slope is the slope of the maximum value change, which is equal to the difference between the maximum and minimum energies, divided by the difference between the maximum and minimum positions, and the absolute value of the ratio.

[0169] Step S36, if either of the following two conditions is met, it means that the count value (global variable) is 0 and the slope flag bit (global variable) is 0, and it can be determined that there is only background noise in the current environment, and there are no moving or micro-moving targets:

[0170] Condition 1, the slope is less than the preset slope threshold of 0.008, or the accumulated maximum value is less than the first energy threshold of 0.3.

[0171] Condition 2, the maximum and minimum positions are less than the preset distance gate threshold of 15, and the accumulated maximum value is less than the second energy threshold of 0.5.

[0172] Step S37, if the conditions in Step S36 are not met, then perform in-vivo recognition within the defense area.

[0173] Step S371, if the count value < the preset number of times 3, then the count value is incremented by 1, and the point cloud decision delay is reset to 10.

[0174] Step S372, if the detected count value ≥ the preset number of times 3, the count value remains 3, and the slope flag is reset to 1, then it is considered that there are people or other moving and micro-moving targets in the environment.

[0175] Step 4, point cloud output and defense area setting. As Figure 6 shown.

[0176] Step S41, construct a 4×80×32 three-dimensional array zero_dopple_buf, where 4×80×32 respectively represent the upper limit values of [the number of receiving antennas][the number of distance indices][the number of frames] of the three-dimensional array zero_dopple_buf. The array is updated frame by frame. One frame of data contains data from four receiving antennas. The data of one receiving antenna contains the energy values corresponding to different distance gates arranged in sequence from near to far. The total number of accumulated frames is 32. Therefore, the size of the array is 4×80×32.

[0177] Step S42, perform non-coherent accumulation on the multi-frame data of the same antenna and the same distance gate in the three-dimensional array zero_dopple_buf, calculate the mean values of the real part and the imaginary part respectively, and calculate the difference by subtracting the mean value from the original value.

[0178] Step S43: Allocate a three-dimensional array zerodopple_remove_buf of 4×80×32, and put the difference between the original value zero_dopple_buf and the average value in step S42 into the corresponding position of zerodopple_remove_buf.

[0179] Step S44: Repeat step S42 and step S43, calculate the differences of multiple frames of data under different distances of the same antenna, and put the results into the corresponding positions of zerodopple_remove_buf.

[0180] Step S45: Repeat step S42 to step S44, calculate the differences of multiple frames of data under different distances of different antennas, and put the results into the corresponding positions of zerodopple_remove_buf.

[0181] Step S46: Since there are 4 receiving antennas, construct a two-dimensional preset steering vector of size 4×AngleTotal according to the following formula, and denote this preset steering vector as AngleArray.

[0182] ;

[0183] where, e is the exponent, j is the imaginary unit, λ is the wavelength, N represents the number of receiving antennas. In this embodiment, N = 4, d is the spacing between adjacent two receiving antennas (here it is half wavelength λ / 2), θ is the angle of the receiving antenna, θ∈[θ1,θ AngleTotal , in this embodiment, θ∈[θ1,θ AngleTotal are respectively θ1 = -60°, θ AngleTotal = 60°.

[0184] Step S47: Integrate the beam. After transposing the steering vector obtained in step S47 to get the transposed matrix AngleArray T , multiply the energy values corresponding to different receiving antennas at the same distance and the same angle by the corresponding steering vectors respectively, and then add them up to obtain a distance-angle energy map constructed by the product of the two, with the distance as the abscissa (the total number is RanTotal = 80) and the angle as the ordinate (the total number is AngleTotal = 24, the variation range is ±60°, and the step size is 5°), denoted as PA.

[0185] PA = AngleArray T ×RanArray;

[0186] Then the array size of the two-dimensional energy spectrum is AngleTotal×RanTotal. In this embodiment, it is 24×80.

[0187] Step S48: Calculate the mean values of the upper, lower, left, and right boundary points of the point cloud data map to obtain the noise coefficient Noise.

[0188] Step S49: Set the signal-to-noise ratio threshold as needed. The formula for calculating the signal-to-noise ratio is SNR = 10 × log10(Signal / Noise), and thus the signal-to-noise ratio of the point cloud energy map can be calculated.

[0189] Step S5: Configure the preset defense area range through the terminal on the Web visual interface, and determine whether the coordinates of the data points are within the defense area. If the energy value of the current data point is greater than the energy values of the four surrounding data points, and the signal-to-noise ratio of this point is higher than the signal-to-noise ratio threshold, then calculate the horizontal and vertical coordinates (i.e., the abscissa and ordinate) of the point cloud.

[0190] Among them, the selection conditions for the data points include:

[0191] (1) The energy value of the data point is greater than 0 and greater than the energy values of the four surrounding upper, lower, left, and right data points.

[0192] (2) Calculate the abscissa Range × sin(Angle) and ordinate Range × cos(Angle) of the point cloud, and determine whether this point is within the defense area.

[0193] If the number of selected data points ≥ 1, the point cloud flag position is set to 1; otherwise, the point cloud flag position is set to 0; if the number of selected data points ≥ 5, the point cloud decision delay count value is reset to 10.

[0194] The above thresholds are the thresholds obtained based on the offline data and online tests of a specific millimeter-wave radar, and considering the delay effect. The thresholds can also be adjusted according to the actual situations such as the point cloud attributes of the target data points (moving point cloud or micro-moving point cloud), the angle and distance step sizes after integrating the beams, and the differences in the quantity and quality of the point cloud. This embodiment does not make any limitations here.

[0195] Step S6: Judgment of external interference in the defense area. As Figure 7 shown.

[0196] Step S612: Judge the slope flag bit. If the slope flag bit that satisfies Step S372 is equal to 1, it is considered that there are people or other dynamic targets in the environment. At this time, set the presence detection flag position to 1, and reset the no-person decision delay to 3. Otherwise, execute Step S62.

[0197] Step S613: Determine whether the current frame number > the interval frame number × the cumulative frame number, and whether the point cloud decision delay is greater than 0. In this embodiment, the interval frame number × the cumulative frame number is 5 × 32, which is used to determine whether the array is full. If the condition is not met, then execute Step S614. When the point cloud decision delay is greater than 0, trigger the point cloud decision. Since there may be no point cloud after a person completely stops moving, the filtering of interference by the point cloud is only triggered during the time of human movement with a delay. Exemplarily, the total trigger delay is 10 s, and the count value is decremented by 1 for each operation.

[0198] Step S6131: If the condition of Step S613 is satisfied, then repeat Step S4, decrement the point cloud decision delay count by 1, and perform point cloud generation and processing.

[0199] Step S6132: Determine whether the position of the maximum distance gate energy ≥ 20 and ≤ 30, and the maximum energy > 3. If satisfied, set the point cloud decision delay to 0, and execute Step S614.

[0200] Step S614: Determine whether the point cloud flag bit is 0.

[0201] Step S6141: If the point cloud flag bit is 0, and the position of the maximum distance gate energy < 20 and > 30, output that there is no person inside the detection zone, and the information of movement or slight movement in Step S372 is interference outside the detection zone.

[0202] Step S6142: If the position of the maximum distance gate energy > 35 and the energy > 1.5, reset the point cloud decision delay to 10, and output that there is a moving or slightly moving target inside the detection zone.

[0203] Step S62: If the condition of Step S36 is satisfied, then determine the current output state. If the current state is no person, clear the register.

[0204] Step S621: Determine whether the no-person decision delay count value is greater than 0. If > 0, decrement the no-person decision delay count value by 1, and output that there is a person.

[0205] Step S622: Output no person.

[0206] Step S7: Identify living bodies within the defense zone. Use deep learning samples to determine whether the detected target is a human or an animal. As Figure 8 shown.

[0207] Step S71: Input training samples. If there is a detected output with a target, then use the data after heart rate filtering as a sample for training.

[0208] Step S72, data augmentation and transformation. Map the energy of the data to the corresponding image in different colors for training. In a specific embodiment, the ImageDataGenerator class in Keras can be used to perform image transformation.

[0209] Step S73, transfer learning. Exemplarily, the VGG16 model can be used as a pre-trained model.

[0210] Step S74, sample synthesis. Exemplarily, generative adversarial networks (GANs) can be used for sample synthesis.

[0211] Step S75, generate a training model. The Adam optimizer and the cross-entropy loss function can be used to compile the model and train the samples.

[0212] Step S8, result judgment. Detect the vital sign signals of the object to be detected.

[0213] Step S81, static strong reflection and non-human living body recognition. Calculate whether the energy of respiration and heart rate exceeds the threshold. If it exceeds the threshold, it is a living body; if not, it is a strong reflection such as metal.

[0214] Step S82, compare whether the sample meets the training model. If it does not meet the requirements, judge that the living body in the defense area is a pet rather than a human. If the conditions are met, judge that there is a person and output the result.

[0215] Step S83, if there is no information such as the human getting out of bed later, or the energy of respiration and heart rate suddenly decreases for a long time, then do not make a judgment on whether there is a detection.

[0216] Step S84, if a person enters again after getting out of bed, then re-perform Steps S1 to S73 to distinguish whether the data points are the environment or interference.

[0217] A dynamic target detection method provided by this embodiment judges the target and the background noise through range gate energy analysis, thereby avoiding the complex grid map construction process in the traditional technology; filters out the interference outside the defense area through point cloud technology, ensuring that only the important areas within the defense area are monitored; realizes the intelligent recognition of the target types within the defense area through deep learning algorithms, improving the accuracy and reliability of presence detection. According to the dynamic target detection method provided by this embodiment, the interference of the environment can be effectively excluded, achieving the effects of reducing the amount of calculated data and the time consumption.

[0218] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are sequentially shown according to the indications of the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear indication in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of the steps or stages in other steps or other steps.

[0219] Based on the same inventive concept, an embodiment of the present application also provides a dynamic target detection device for implementing the above-mentioned dynamic target detection method. The solution provided by this device to solve the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the dynamic target detection device provided below can refer to the limitations on the dynamic target detection method in the above text, and will not be repeated here.

[0220] In one embodiment, a dynamic target detection device is provided, which is applied to a millimeter-wave radar and includes: a data acquisition module, a map generation module, and a classification detection module, where:

[0221] The data acquisition module is used to acquire signal reception data of at least one receiving antenna of the millimeter-wave radar within a preset time range;

[0222] The map generation module is used to perform background noise and dynamic target analysis based on the signal reception data; when there is a dynamic target in the signal reception data, a point cloud energy map with static components removed is generated based on the signal reception data of at least one of the receiving antennas;

[0223] The classification detection module is used to input the point cloud energy map into a pre-trained target classification model when the dynamic target is within a preset defense area range of the point cloud energy map, and obtain the detection result of the dynamic target.

[0224] In one of the embodiments, the data acquisition module is further used to:

[0225] Acquire the range gate energy data obtained by the receiving antenna in each frame;

[0226] Based on the range gate energy data of each frame within the preset time range, construct a two-dimensional data with the number of frames and the number of range gates as dimensions;

[0227] Use the two-dimensional data as the signal reception data of the receiving antenna within a preset time range.

[0228] In one embodiment, the map generation module is further configured to:

[0229] Remove the static component of the signal reception data based on the average energy value of each range gate in the signal reception data to obtain dynamic energy data;

[0230] Based on the dynamic energy data, obtain the maximum and minimum energy data; the maximum and minimum energy data includes: the maximum energy range gate, the maximum energy value, the minimum energy range gate, and the minimum energy value;

[0231] Based on the maximum and minimum energy data, determine the slope of the maximum and minimum changes of the dynamic energy data;

[0232] Based on the slope of the maximum and minimum changes and the maximum and minimum energy data, determine whether there is a dynamic target in the signal reception data.

[0233] In one embodiment, the map generation module is further configured to:

[0234] If the slope of the maximum and minimum changes and the maximum and minimum energy data satisfy the first background noise condition or the second background noise condition, then there is only background noise in the signal reception data; the first background noise condition includes: the slope of the maximum and minimum changes is less than a preset slope threshold, or the maximum energy value is less than the first energy threshold; the second background noise condition includes: the sum of the maximum energy range gate and the minimum energy range gate is less than a preset range gate threshold, and the maximum energy value is less than the second energy threshold;

[0235] If the number of times that the slope of the maximum and minimum changes and the maximum and minimum energy data do not satisfy the first background noise condition and the second background noise condition simultaneously reaches a preset number within a preset time range, then there is a dynamic target in the signal reception data.

[0236] In one embodiment, the map generation module is further configured to:

[0237] Calculate the energy difference between the energy value of each frame and the average energy value of all frames under the same range gate of the same receiving antenna respectively, to obtain dynamic three-dimensional data after removing the static component; the dynamic three-dimensional data is three-dimensional data with the number of receiving antennas, the number of frames, and the number of range gates as dimensions;

[0238] Based on a preset steering vector, perform beamforming on the dynamic three-dimensional data to obtain a range-angle energy distribution map;

[0239] Based on the range-angle energy distribution map, perform point cloud conversion to obtain the point cloud energy map.

[0240] In one embodiment, the millimeter-wave radar includes at least two receiving antennas, and the map generation module is further configured to:

[0241] Determine the weight value of each receiving antenna at each receiving angle based on the number of receiving antennas, the spacing between adjacent receiving antennas, and the receiving angle range;

[0242] Obtain a preset steering vector with the number of receiving antennas and the receiving angle range as dimensions based on the weight values of multiple receiving antennas at multiple receiving angles.

[0243] In one embodiment, the classification and detection module is further configured to:

[0244] Obtain the number of target data points within a preset defense area in the point cloud energy map; the energy value of the target data point is greater than the energy values of multiple surrounding data points, and the signal-to-noise ratio of the target data point is higher than a preset signal-to-noise ratio threshold; the signal-to-noise ratio is determined based on the energy value of the target data point and a preset noise coefficient;

[0245] If the number of target data points within the preset defense area is at least one, the dynamic target is located within the preset defense area of the point cloud energy map.

[0246] In one embodiment, the dynamic target detection device further includes a model training module, which is configured to:

[0247] Construct a training data sample set; the training data sample set includes various types of point cloud data samples;

[0248] Perform transfer learning on a pre-trained deep learning model based on the training data sample set, an Adam optimizer, and a cross-entropy loss function to obtain the pre-trained target classification model.

[0249] In one embodiment, the classification and detection module is further configured to:

[0250] Detect vital sign signals based on the point cloud energy map;

[0251] When the breathing vital signs and heart rate vital signs of the dynamic target meet the preset vital sign conditions, classify the dynamic target based on the target classification model to obtain the classification result of the dynamic target; the classification result includes human or animal;

[0252] Determine the detection result of the dynamic target based on the classification result.

[0253] Each module in the above dynamic target detection device can be implemented in whole or in part by software, hardware, or a combination thereof. Each of the above modules can be embedded in the processor of the computer device in hardware form or be independent of it, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to each of the above modules.

[0254] In one embodiment, a computer device is provided. The computer device can be a terminal, and its internal structure diagram can be as Figure 9 shown. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner. The wireless manner can be implemented through WIFI, a mobile cellular network, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a dynamic target detection method. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the housing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0255] Those skilled in the art can understand that Figure 9 the structure shown in

[0256] is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0257] Obtain the signal reception data of at least one receiving antenna of the millimeter-wave radar within a preset time range;

[0258] Perform background noise and dynamic target analysis based on the signal reception data; when there is a dynamic target in the signal reception data, generate a point cloud energy map with static components removed based on the signal reception data of at least one of the receiving antennas;

[0259] When the dynamic target is within the preset defense area of the point cloud energy map, input the point cloud energy map into a pre-trained target classification model to obtain the detection result of the dynamic target.

[0260] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the dynamic target detection method of any of the above embodiments is implemented:

[0261] Obtain the signal reception data of at least one receiving antenna of the millimeter-wave radar within a preset time range;

[0262] Perform background noise and dynamic target analysis based on the signal reception data; when there is a dynamic target in the signal reception data, generate a point cloud energy map with static components removed based on the signal reception data of at least one of the receiving antennas;

[0263] When the dynamic target is within the preset defense area of the point cloud energy map, input the point cloud energy map into a pre-trained target classification model to obtain the detection result of the dynamic target.

[0264] In one embodiment, a computer program product is provided. When the computer program product is executed by a processor, the dynamic target detection method of any of the above embodiments is implemented:

[0265] Obtain the signal reception data of at least one receiving antenna of the millimeter-wave radar within a preset time range;

[0266] Perform background noise and dynamic target analysis based on the signal reception data; when there is a dynamic target in the signal reception data, generate a point cloud energy map with static components removed based on the signal reception data of at least one of the receiving antennas;

[0267] When the dynamic target is within the preset defense area of the point cloud energy map, input the point cloud energy map into a pre-trained target classification model to obtain the detection result of the dynamic target.

[0268] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data that have been authorized by the user or fully authorized by all parties.

[0269] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.

[0270] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0271] The above-described embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A dynamic target detection method, characterized in that, Applied to a millimeter-wave radar, the dynamic target detection method includes: Obtaining signal reception data of at least one receiving antenna of the millimeter-wave radar within a preset time range; Performing background noise and dynamic target analysis based on the signal reception data; when there is a dynamic target in the signal reception data, generating a point cloud energy map with static components removed based on the signal reception data of at least one of the receiving antennas; the performing background noise and dynamic target analysis based on the signal reception data includes: removing the static components of the signal reception data based on the average energy value of each range gate in the signal reception data to obtain dynamic energy data; obtaining the maximum and minimum energy data based on the dynamic energy data; the maximum and minimum energy data includes: the maximum energy range gate, the maximum energy value, the minimum energy range gate, and the minimum energy value; determining the maximum and minimum change slopes of the dynamic energy data based on the maximum and minimum energy data; and determining whether there is a dynamic target in the signal reception data based on the maximum and minimum change slopes and the maximum and minimum energy data; When the dynamic target is within the preset defense area range of the point cloud energy map, inputting the point cloud energy map into a pre-trained target classification model to obtain the detection result of the dynamic target.

2. The dynamic target detection method according to claim 1, wherein The obtaining signal reception data of at least one receiving antenna of the millimeter-wave radar within a preset time range includes: Obtaining the range gate energy data obtained by the receiving antenna in each frame; Constructing two-dimensional data with the number of frames and the number of range gates as dimensions based on the range gate energy data of each frame within the preset time range; Taking the two-dimensional data as the signal reception data of the receiving antenna within the preset time range.

3. The dynamic target detection method according to claim 1, characterized in that The determining whether there is a dynamic target in the signal reception data based on the maximum and minimum change slopes and the maximum and minimum energy data includes: If the maximum and minimum change slopes and the maximum and minimum energy data meet the first background noise condition or the second background noise condition, then there is only background noise in the signal reception data; the first background noise condition includes: the maximum and minimum change slope is less than a preset slope threshold, or the maximum energy value is less than a first energy threshold; the second background noise condition includes: the sum of the maximum energy range gate and the minimum energy range gate is less than a preset range gate threshold, and the maximum energy value is less than a second energy threshold; If the number of times that the maximum and minimum change slopes and the maximum and minimum energy data do not meet the first background noise condition and the second background noise condition simultaneously within a preset time range reaches a preset number of times, then there is a dynamic target in the signal reception data.

4. The dynamic target detection method according to claim 1, wherein The generating a point cloud energy map with static components removed based on the signal reception data of at least one of the receiving antennas includes: Calculating the energy difference between the energy value of each frame and the average energy value of all frames at the same range gate of the same receiving antenna respectively to obtain dynamic three-dimensional data with static components removed; the dynamic three-dimensional data is three-dimensional data with the number of receiving antennas, the number of frames, and the number of range gates as dimensions; Performing beamforming on the dynamic three-dimensional data based on a preset steering vector to obtain a range-angle energy distribution map; Based on the distance-angle energy distribution map, perform point cloud conversion to obtain the point cloud energy map.

5. The dynamic target detection method according to claim 4, wherein The millimeter-wave radar includes at least two receiving antennas. Before obtaining the distance-angle energy distribution map by performing beamforming on the dynamic three-dimensional data based on a preset steering vector, it includes: Based on the number of receiving antennas, the spacing between adjacent receiving antennas, and the receiving angle range, determine the weight value of each receiving antenna at each receiving angle. Based on the weight values of multiple receiving antennas at multiple receiving angles, obtain a preset steering vector with the number of receiving antennas and the receiving angle range as dimensions.

6. The dynamic target detection method according to claim 1, wherein The dynamic target detection method further includes: Obtain the number of target data points within a preset defense area in the point cloud energy map; the energy value of the target data point is greater than the energy values of multiple surrounding data points, and the signal-to-noise ratio of the target data point is higher than a preset signal-to-noise ratio threshold; the signal-to-noise ratio is determined based on the energy value of the target data point and a preset noise coefficient. If the number of target data points within the preset defense area is at least one, the dynamic target is located within the preset defense area of the point cloud energy map.

7. The dynamic target detection method according to claim 1, wherein Before inputting the point cloud energy map into a pre-trained target classification model to obtain the detection result of the dynamic target, it further includes: Construct a training data sample set; the training data sample set includes various types of point cloud data samples. Based on the training data sample set, the Adam optimizer, and the cross-entropy loss function, perform transfer learning on a pre-trained deep learning model to obtain the pre-trained target classification model.

8. The dynamic target detection method according to claim 1, wherein Inputting the point cloud energy map into the pre-trained target classification model to obtain the detection result of the dynamic target includes: Perform vital sign signal detection based on the point cloud energy map. When the breathing vital signs and heart rate vital signs of the dynamic target meet the preset vital sign conditions, classify the dynamic target based on the target classification model to obtain the classification result of the dynamic target; the classification result includes human or animal. Based on the classification result, determine the detection result of the dynamic target.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method according to any one of claims 1 to 8.

11. A computer program product, characterized in that, When the computer program product is executed by the processor, it implements the method according to any one of claims 1 to 8.

Images