Fall identification method based on millimeter wave radar point cloud data and height threshold detection

Abstract

The invention relates to the field of identification and detection, in particular to a fall identification method based on millimeter wave radar point cloud data and a height threshold, which comprises the following steps of: 1, acquiring radar point cloud data; 2, extracting accurate point cloud data from target point information; 3, calculating three-dimensional point cloud data through an algorithm; 4, height information extraction and dynamic height threshold value fitting; 5, tumble identification is carried out; according to the method, the high-precision perception capability of the millimeter wave radar and the dynamic fitting model of the height threshold are combined, the accuracy of fall detection can be effectively improved, and the method has high environmental adaptability and privacy protection characteristics and is suitable for families, nursing homes, hospitals and other scenes.

Description

Technical Field

[0001] The present invention relates to the field of recognition and detection, and in particular to a fall recognition method based on millimeter wave radar point cloud data and a height threshold. Background Art

[0002] As the problem of population aging intensifies, falling has become one of the major risks among the elderly. Existing fall detection technologies usually rely on video surveillance, accelerometers or other sensors, which have problems such as insufficient recognition accuracy, privacy issues and poor environmental adaptability. Millimeter-wave radar has the advantages of strong penetration, good privacy protection, and no influence from light. Therefore, using millimeter-wave radar point cloud data for fall recognition is an important technical means. Existing radar point cloud fall detection algorithms mostly use fixed threshold methods, which cannot be flexibly adjusted according to individual characteristics and environment, resulting in a high false alarm rate. Therefore, there is an urgent need for a fall recognition method based on millimeter-wave radar point cloud data and dynamic height threshold detection, which can improve detection accuracy and reduce false alarms. Summary of the invention

[0003] To solve one of the above problems, the present invention adopts the following technical solution, and the method steps are as follows: a fall recognition method based on millimeter wave radar point cloud data and height threshold detection, characterized in that it includes: the first step, through radar point cloud data collection; the second step, target point information extraction accurate point cloud data; the third step, through the algorithm calculation of three-dimensional point cloud data; the fourth step, height information extraction and dynamic height threshold fitting; the fifth step, fall recognition; the sixth step, alarm triggering to complete the recognition.

[0004] Furthermore, the radar point cloud data collection includes the steps of millimeter wave signal transmission and reception processing.

[0005] Furthermore, the target point information extraction of accurate point cloud data includes processing the data of the distance dimension and the Doppler dimension twice through an algorithm, including static target removal and filtering;

[0006] Use algorithms to extract target point information from the processed point cloud data and eliminate interference points.

[0007] Furthermore, the calculation of three-dimensional point cloud data by an algorithm includes calculating the orientation information of the target point, including radiality and azimuth, by an algorithm, and calculating the precise spatial position of the target point according to the phase difference of the radar received signal to generate three-dimensional point cloud data.

[0008] Furthermore, in the height information extraction and dynamic height threshold fitting, the height information extraction includes extracting the height information of the target human body from the three-dimensional point cloud data collected by the millimeter wave radar; the dynamic height threshold fitting includes data collection, linear regression model, least squares method, dynamic threshold generation, and adaptive adjustment.

[0009] Furthermore, fall recognition includes determining the data observation window and historical observation window size required for fall event judgment, real-time height monitoring, dynamic height threshold comparison, fall status confirmation, and activity status distinction.

[0010] Furthermore, the alarm is triggered when the system confirms that the target has a fall event, automatically triggering an alarm and sending an alarm message to the user or emergency contact via a mobile device or other communication means.

[0011] Furthermore, the height information is the Z-axis height information of the target human body, and also includes point cloud screening and center of mass calculation.

[0012] Furthermore, the fall state confirmation consists of the characteristics of the fall state, height change discrimination and threshold setting.

[0013] Furthermore, the activity status distinction includes combining the height information of the previous and next time points. The system determines the current activity status through the following logic: current H previous The rate of change of height is R th If H current Significantly lower than H th And R <R th , the system determines it as a fall.

[0014] The beneficial effects of this application include at least the following:

[0015] 1. The purpose of the present invention is to provide a fall recognition method based on millimeter-wave radar point cloud data and height threshold detection. By collecting human body point cloud data through millimeter-wave radar and combining the linear model to fit the height threshold, fall events can be effectively identified and false alarms can be reduced.

[0016] 2. This application method is efficient, accurate and fast, can effectively improve the accuracy of fall detection, and has strong environmental adaptability and privacy protection characteristics, and is suitable for scenarios such as homes, nursing homes, and hospitals. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of the system flow of the present application method;

[0018] Figure 2 This is a schematic diagram of the fall recognition process for this application. DETAILED DESCRIPTION

[0019] The present application aims to describe a method and process for systematic data processing and algorithm identification. Figure 1 The system flow diagram of the present application method is as follows: Figure 1 As shown in the figure, the process includes six steps, such as the first step, collecting radar point cloud data; the second step, extracting accurate point cloud data from target point information; the third step, calculating three-dimensional point cloud data through algorithms; the fourth step, extracting height information and fitting dynamic height thresholds; the fifth step, fall recognition; the sixth step, triggering an alarm to complete recognition. The specific steps are as follows:

[0020] The first step is radar point cloud data collection

[0021] The target in the monitoring area is scanned by the millimeter wave radar device to collect the point cloud data of the target object. The specific steps include:

[0022] Millimeter wave signal transmission and reception: The millimeter wave radar transmits high-frequency signals and receives reflected signals to obtain point cloud data of the monitoring area. In order to achieve accurate filtering of radio frequency signals and significantly reduce out-of-band interference and noise, the received signal is mixed with the transmitted signal to generate an intermediate frequency signal;

[0023] Then 2D-FFT processing is performed: a two-dimensional fast Fourier transform (2D-FFT) is performed on the intermediate frequency signal to extract the distance and speed information of the target and generate a preliminary spectrum diagram.

[0024] The second step is to extract accurate point cloud data from target point information

[0025] The data of the distance dimension and Doppler dimension are processed twice by the CFAR (Constant False Alarm Rate) algorithm, including static target removal and filtering. First, the average value of noise or clutter when a "non-moving target" exists is calculated as the static background noise, and the "moving target" judgment threshold is obtained. Finally, static targets below the threshold are removed; then adaptive background noise is used to cancel clutter, and a first-order adaptive smoothing filter is used to improve the accuracy of the target point.

[0026] The DBSCAN algorithm is used to extract target point information from the point cloud data processed by CFAR, eliminate interference points, and obtain accurate point cloud data of the target object. The DBSCAN algorithm finds data-dense intervals and classifies them into the same cluster. These clusters represent possible target points of human body parts or other objects. Through clustering, the target can be effectively separated from complex point cloud data. After clustering, each cluster obtained represents a target object. By analyzing the center position, volume, and shape of each cluster, different parts of the human body, such as trunk, limbs, etc., can be further extracted, thereby providing key information for subsequent fall recognition.

[0027] The third step is to calculate the 3D point cloud data using the DOA (angle of arrival) algorithm.

[0028] The DOA algorithm is used to calculate the target point's position information, including radial and azimuth angles. Based on the phase difference of the radar received signal, the precise spatial position of the target point is calculated to generate three-dimensional point cloud data (X, Y, Z coordinates).

[0029] Step 4: Height information extraction and dynamic height threshold fitting

[0030] Highly Information Extraction:

[0031] Extract the Z-axis height information of the target human body from the 3D point cloud data collected by the millimeter-wave radar. The specific steps include:

[0032] Point cloud screening: Screen the 3D point cloud data, identify the point clouds related to the target human body, and exclude points related to the background or static objects.

[0033] Centroid calculation: By calculating the centroid of the filtered point cloud data, the height information (H) of the entire point set is obtained. The Z coordinate of the centroid can be calculated using the following formula:

[0034]

[0035] Where N is the number of points after screening, Z i is the Z coordinate of the ith point.

[0036] Dynamic height threshold fitting:

[0037] In order to improve the accuracy of fall detection, the system needs to dynamically adjust the threshold based on real-time height information. The specific method is as follows:

[0038] Data collection: The system continuously monitors and records the target’s altitude data, forming an altitude time series data set D = {H1, H2, ..., H n}, where H i Represents the height information at the i-th time point.

[0039] Linear regression model: A linear regression model is used to fit the height data to determine the dynamic height threshold. The model assumptions are: H fit =a·t+b, where H fit is the height of the fit, t is the time, a is the slope, and b is the intercept.

[0040] Least Squares Method: The model parameters a and b are calculated by the least squares method to minimize the following sum of squared errors:

[0041] Dynamic threshold generation: After the model parameters are determined, the dynamic height threshold H is generated according to the height change of the target. th , can be set as a percentage of the fitting result, for example: H fit =H fit (1-δ) where δ is a constant less than 1 and is used to adjust the sensitivity of the threshold.

[0042] Adaptive adjustment: The system can adaptively adjust the parameter δ according to environmental changes and individual characteristics to ensure that the dynamic height threshold can effectively reflect the normal height range of the target.

[0043] Step 5: Fall Recognition

[0044] The fall recognition process diagram is as follows: Figure 2 As shown, in real-time monitoring, the system obtains the current height information H of the target current And the fitted dynamic height threshold H th By comparing these two values, it is determined whether the target is in a falling state. The specific steps are as follows:

[0045] Determine the data observation window and historical observation window size required for fall event judgment: Since the fall action event is an instantaneous action with a short duration, it is necessary to determine a suitable data "observation window" size T1 (for example, T1 = 50ms) to accumulate data within T time to accurately estimate the "real-time height". At the same time, it is necessary to set a "historical observation window" with a size of T2 (for example, T2 = 500ms) before the "observation window" moment to fit and calculate the "dynamic height threshold".

[0046] Real-time height monitoring: The system continuously obtains the height information of the target, calculates the centroid of the filtered point cloud data through 4 height information extraction methods, and obtains the current height value, which is recorded as H current .

[0047] Dynamic height threshold comparison: The real-time acquired height H current With dynamic height threshold H th Make a comparison.

[0048] If H current <H th , it is preliminarily determined to be a fall state; at this time, the system enters the further confirmation stage.

[0049] Fall status confirmation: Considering the height characteristics under different activity states, the system further confirms whether a fall has occurred according to the following rules:

[0050] Characteristics of a fall: When a person falls, there is usually a significant loss of height. Compared with other activities (such as walking, sitting or squatting), the height change during a fall is more dramatic.

[0051] Altitude change determination: Calculate the altitude change within several time points (e.g. T=1s, T=2s) and define the altitude change rate as:

[0052]

[0053] Among them, H previous It is the height data of the previous moment.

[0054] Threshold setting: Set the threshold R of the height change rate th For example, R th = -30%. If R <R th , further confirming that the target is in a falling state.

[0055] Activity status distinction: Combining the height information of the previous and next time points, the system determines the current activity status through the following logic:

[0056] If H current H previous The rate of change of height is R th The system determines it as walking, sitting or other states within the range.

[0057] If H current Significantly lower than H th And R <R th , the system determines it as a fall.

[0058] Step 6: Alarm Trigger

[0059] When the system confirms that the target has fallen, it automatically triggers an alarm and sends an alarm message to the user or emergency contact via mobile devices or other communication methods. The system can further trigger an alarm through voice prompts or smart devices.

[0060] Example

[0061] Example 1

[0062] In indoor environments, millimeter-wave radar equipment is installed to monitor the activity areas of the elderly or people with limited mobility. The millimeter-wave radar collects point cloud data of the target in real time, and generates preliminary point cloud data of the target through 2D-FFT, static target removal and filtering. The real point cloud data of the target is extracted through CFAR processing, and the orientation information of the target is calculated using the DOA algorithm to obtain the three-dimensional coordinates.

[0063] Based on the Z-axis height data, the linear model is fitted to generate a dynamic height threshold. When the target's real-time height is lower than the threshold, combined with posture change analysis, the system determines that the target has fallen and sends an alarm to the guardian through a mobile device.

[0064] Example 2

[0065] In outdoor public areas, millimeter-wave radars are installed at specific monitoring points. The radar equipment detects the target's moving speed and azimuth changes through the Doppler effect, updates the height threshold in real time, and identifies fall events. In complex environments, the DOA algorithm accurately calculates the target point position, and dynamically adjusts the height threshold to adapt to different human activity states.

[0066] The significance of the implementation effect of the present application is at least reflected in the fact that the purpose of the present application is to provide a fall recognition method based on millimeter-wave radar point cloud data and height threshold detection. By collecting human point cloud data by millimeter-wave radar and combining the linear model to fit the height threshold, fall events can be effectively identified and false alarms can be reduced. The method of the present application is efficient, accurate and fast, can effectively improve the accuracy of fall detection, and has strong environmental adaptability and privacy protection characteristics, and is suitable for scenarios such as homes, nursing homes, and hospitals.

[0067] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A fall recognition method based on millimeter wave radar point cloud data and height threshold detection, characterized in that: include: The first step is to collect point cloud data through radar; the second step is to extract accurate point cloud data from target point information; the third step is to calculate three-dimensional point cloud data through algorithms; The fourth step is to extract height information and fit the dynamic height threshold; the fifth step is to identify falls; and the sixth step is to trigger an alarm to complete the identification.

2. A fall recognition method based on millimeter wave radar point cloud data and height threshold detection as claimed in claim 1, characterized in that: The radar point cloud data collection includes the steps of millimeter wave signal transmission and reception processing.

3. A fall recognition method based on millimeter wave radar point cloud data and height threshold detection as claimed in claim 1, characterized in that: The target point information extraction of accurate point cloud data includes two processing steps of the distance dimension and the Doppler dimension data through an algorithm, including static target removal and filtering; Use algorithms to extract target point information from the processed point cloud data and eliminate interference points.

4. A fall recognition method based on millimeter wave radar point cloud data and height threshold detection as claimed in claim 1, characterized in that: The three-dimensional point cloud data is calculated by the algorithm, including the position information of the target point, including the radial degree and the azimuth, and the precise spatial position of the target point is calculated according to the phase difference of the radar receiving signal to generate the three-dimensional point cloud data.

5. A fall recognition method based on millimeter wave radar point cloud data and height threshold detection as claimed in claim 1, characterized in that: In the height information extraction and dynamic height threshold fitting, the height information extraction includes extracting the height information of the target human body from the three-dimensional point cloud data collected by the millimeter wave radar; the dynamic height threshold fitting includes data collection, linear regression model, least squares method, dynamic threshold generation, and adaptive adjustment.

6. A fall recognition method based on millimeter wave radar point cloud data and height threshold detection as claimed in claim 1, characterized in that: Fall recognition includes determining the data observation window and historical observation window size required for fall event judgment, real-time height monitoring, dynamic height threshold comparison, fall status confirmation, and activity status distinction.

7. A fall recognition method based on millimeter wave radar point cloud data and height threshold detection as claimed in claim 1, characterized in that: The alarm trigger is when the system confirms that the target has a fall event, it automatically triggers an alarm and sends an alarm message to the user or emergency contact via a mobile device or other communication means.

8. A fall recognition method based on millimeter wave radar point cloud data and height threshold detection as claimed in claim 5, characterized in that: The height information is the Z-axis height information of the target human body, and also includes point cloud screening and center of mass calculation.

9. A fall recognition method based on millimeter wave radar point cloud data and height threshold detection as claimed in claim 6, characterized in that: The fall state confirmation consists of the characteristics of the fall state, height change discrimination and threshold setting.

10. A fall recognition method based on millimeter wave radar point cloud data and height threshold detection as claimed in claim 6, characterized in that: The activity status distinction includes combining the height information of the previous and next time points. The system determines the current activity status through the following logic. If H current H previous The rate of change of height is R th If H current Significantly lower than H th And R <R th , the system determines it as a fall.

Images