A non-contact fall detection method and device

Abstract

The present invention discloses a contactless fall detection method and device, which relates to the field of intelligent health monitoring technology, including receiving user data detected by sensors and preprocessing the user data; performing time synchronization on all sensors, calculating the position change by integration according to the human body position and initial speed at the initial moment; calculating the position change amount and the angular change amount, fusing the velocity change amount, the position change amount and the angular change amount to obtain a comprehensive feature, performing dimensionality reduction processing on the extracted comprehensive feature to obtain a fall feature; using a machine learning algorithm to build a user behavior model, inputting the fall feature into the user behavior model, performing pattern recognition on the extracted fall feature to determine whether there is a fall behavior; and continuously updating the user behavior model parameters to ensure that the system can adapt to environmental changes and differences in personal behavior habits, maintain high accuracy and robustness in long-term operation, and provide reliable protection for the safety of the elderly and people with limited mobility.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent health monitoring, and in particular to a non-contact fall detection method and device. Background Art

[0002] In recent years, with the aging of the population and the rapid development of smart home technology, fall detection technology has received more and more attention. Traditional fall detection methods mainly rely on wearable sensors (such as accelerometers, gyroscopes, etc.). These devices identify fall events by monitoring changes in human motion status. However, wearable sensors have many limitations, such as requiring users to wear them at all times, being susceptible to environmental interference, having limited battery life, and unstable data transmission. In addition, wearable devices are often difficult to cover large areas in actual applications, and there are also challenges in real-time monitoring of multiple individuals. Therefore, non-contact fall detection technology has gradually become a research hotspot. It not only overcomes many shortcomings of wearable sensors, but also provides a more flexible and efficient solution.

[0003] Contactless fall detection technology uses a variety of sensors, such as millimeter-wave radar, infrared camera, ultrasonic sensor, etc., to realize the non-sensing monitoring of human movement in the environment. Among them, millimeter-wave radar has become an important choice in this field due to its high resolution, strong anti-interference ability, and all-weather operation. The human body position and speed information obtained by millimeter-wave radar can be used to construct a complete human motion trajectory, and then extract features related to falls. However, the existing non-contact fall detection system still faces some problems, such as inaccurate feature extraction, insufficient model training data, and high false alarm rate. Especially in complex environments, such as multi-person activity scenes or environments with more occlusions, the detection accuracy and robustness of existing systems still need to be improved. In addition, most existing systems lack an effective feedback mechanism and cannot be adaptively adjusted according to new user data, which makes it difficult to ensure the performance stability in long-term operation. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a non-contact fall detection method to solve the problems of inaccurate feature extraction, insufficient model training data, high false alarm rate, etc., especially in complex environments, such as multi-person activity scenes or environments with a lot of occlusions, the detection accuracy and robustness of the existing system still need to be improved.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a non-contact fall detection method, which comprises:

[0008] receiving user data detected by multiple sensors installed in the detection area and preprocessing the user data;

[0009] All sensors are synchronized using a global time synchronization mechanism, and the position change is calculated by integration based on the initial position and initial velocity of the human body;

[0010] According to the human body position at the initial moment, the speed change is calculated. According to the position change, the position change and the direction angle change are calculated. The speed change, position change and direction angle change are fused to obtain comprehensive features. The principal component analysis PCA is used to reduce the dimension of the extracted comprehensive features to obtain the fall features.

[0011] Use machine learning algorithms to build a user behavior model, input fall features into the user behavior model, perform pattern recognition on the extracted fall features, and determine whether there is a fall behavior;

[0012] Continue to collect new user data and user feedback, and constantly update user behavior model parameters.

[0013] As a preferred solution of the non-contact fall detection method of the present invention, wherein: the receiving of user data detected by multiple sensors installed in the detection area and pre-processing of the user data comprises the following specific steps:

[0014] The sensor refers to a millimeter wave radar sensor;

[0015] The user data refers to the position and speed of the human body obtained by the millimeter wave radar sensor;

[0016] Preprocessing of user data mainly refers to filtering, noise reduction and smoothing, and finally standardization.

[0017] As a preferred solution of the non-contact fall detection method of the present invention, the expression for calculating the position change by integration is:

[0018] ;

[0019] in, Indicates at time location, Indicates the time interval, Indicates that at any time point in the integration process The speed at represents the integral variable, Represents the integral variable An infinitesimal increment of represents the initial time, Indicates the initial time The human body position, Indicates that at the initial moment speed.

[0020] As a preferred solution of the non-contact fall detection method of the present invention, the expression of the comprehensive feature is:

[0021] ;

[0022] in, Indicates at time The comprehensive characteristics of The weight coefficient representing the speed change, The weight coefficient representing the position change, Indicates the maximum speed change, Indicates the maximum position change, Indicates the maximum direction angle change, Indicates at time The speed change, Indicates at time The position change, Indicates at time The change in direction angle.

[0023] As a preferred solution of the non-contact fall detection method of the present invention, the user behavior model is constructed using a machine learning algorithm, and the specific steps are:

[0024] Choose to use the combined algorithm of convolutional neural network CNN and long short-term memory network LSTM to build user behavior model;

[0025] Set the architecture of the user behavior model based on the convolutional neural network CNN and the long short-term memory network LSTM;

[0026] The architecture of the user behavior model refers to the input layer, convolution layer, pooling layer, LSTM layer and fully connected layer;

[0027] Collect and annotate a large dataset of behavioral patterns of normal activities and fall events;

[0028] The labeled behavior pattern dataset is divided into training set, test set and validation set in proportion, and input into the user behavior model for training, testing and validation of the user behavior model respectively.

[0029] As a preferred solution of the non-contact fall detection method of the present invention, wherein: the fall feature is input into the user behavior model, the extracted fall feature is subjected to pattern recognition, and whether there is a fall behavior is determined, the specific steps are:

[0030] The comprehensive features identified by the CNN+LSTM combination algorithm Input the user behavior model and calculate the probability of falling behavior. The expression is:

[0031] ;

[0032] in, Indicates at time The probability of falling behavior, represents the Sigmoid activation function, Represents the weight vector The transpose of Represents the bias term.

[0033] As a preferred solution of the non-contact fall detection method of the present invention, the method of continuously collecting new user data and user feedback and continuously updating the user behavior model parameters comprises the following specific steps:

[0034] Use statistical methods to screen newly collected user data and user feedback;

[0035] Merge the filtered behavior pattern dataset with the existing historical user data, and then calculate new comprehensive features And the corresponding probability of falling behavior, then adjust the user behavior model parameters to optimize performance, the expression is:

[0036] ;

[0037] in, Indicates at time The change in the parameters of the user behavior model, represents the learning rate, represents the gradient of the user behavior model parameters, represents the set of user behavior model parameters, Indicates at time The actual label, represents the loss function; at regular intervals, a batch of new user data and user feedback is accumulated to form a training data set;

[0038] Use the accumulated training data set to input the user behavior model, perform comprehensive retraining, and update the user behavior model parameters. The update expression is:

[0039] ;

[0040] in, Represents the new model parameters obtained after retraining, represents the number of samples in the accumulated new data set, The index variable representing the sample, Indicates The key features of the samples are Indicates The actual labels of samples, represents the loss function, represents the regularization coefficient, represents the regularization term, Indicates found The smallest value in .

[0041] In a second aspect, the present invention provides a non-contact fall detection device, including an acquisition module, a motion trajectory module, a feature extraction module, a user behavior module and an update module:

[0042] The acquisition module is used to receive the user's position and speed information detected by multiple millimeter wave radar sensors installed in the detection area, and pre-process the information;

[0043] The motion trajectory module calculates the position change by integration according to the human body position and speed at the initial moment to form a complete human body motion trajectory;

[0044] The feature extraction module calculates the speed change, position change and direction angle change according to the human body motion trajectory, and fuses the speed change, position change and direction angle change into a comprehensive feature;

[0045] The user behavior module selects a combination algorithm of a convolutional neural network (CNN) and a long short-term memory network (LSTM) to construct a user behavior model, inputs comprehensive features into the user behavior model, and calculates the probability of falling behavior;

[0046] The update module uses statistical methods to screen newly collected user data and user feedback, and merges them with historical data, adjusts user behavior model parameters, optimizes performance, regularly accumulates new user data, conducts comprehensive retraining, and updates user behavior model parameters.

[0047] In a third aspect, the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the non-contact fall detection method as described in the first aspect of the present invention is implemented.

[0048] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the non-contact fall detection method as described in the first aspect of the present invention is implemented.

[0049] The beneficial effects of the present invention are as follows: the present invention realizes large-area non-contact monitoring by installing multiple millimeter-wave radar sensors in the detection area, ensures the accuracy and reliability of data by filtering, noise reduction and smoothing, constructs a complete human motion trajectory by using a global time synchronization mechanism and an integral calculation method, improves the sensitivity and accuracy of fall detection, extracts features such as speed change, position change and angular change from the motion trajectory, reduces the dimension by principal component analysis (PCA), enhances the compactness and discrimination of feature representation, and constructs a user behavior model by using a combination algorithm of a convolutional neural network (CNN) and a long short-term memory network (LSTM), gives full play to the ability of CNN to capture local spatial features and LSTM to process time series data, improves the intelligence level and classification performance of pattern recognition, simplifies the decision-making process and improves the response speed of the system by inputting comprehensive features into the user behavior model and calculating the probability of falling. In addition, the present invention introduces a continuous data update and model optimization mechanism to ensure that the system can adapt to environmental changes and differences in personal behavior habits, maintain high accuracy and robustness in long-term operation, and provide reliable protection for the safety of the elderly and people with limited mobility. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0051] Figure 1 This is a flow chart of the non-contact fall detection method in Example 1.

[0052] Figure 2 Schematic diagram of the non-contact fall detection device in Example 1. DETAILED DESCRIPTION

[0053] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0054] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0055] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0056] Example 1, reference Figure 1 , which is the first embodiment of the present invention, provides a non-contact fall detection method, comprising the following steps:

[0057] S1: receiving user data detected by multiple sensors installed in the detection area and pre-processing the user data;

[0058] The sensor refers to a millimeter wave radar sensor;

[0059] User data refers to the position and speed of the human body obtained by the millimeter wave radar sensor;

[0060] Preprocessing of user data mainly refers to filtering, noise reduction and smoothing, and finally standardization;

[0061] Specifically, the user data is filtered using a low-pass filter, noise is reduced using wavelet transform, the user data is smoothed using a Savitzky-Golay filter, and the user data collected by all sensors is standardized using Z-score standardization;

[0062] The millimeter wave radar sensor was chosen not only because of its superior hardware performance, but more importantly, it can monitor the position and speed of the human body in real time without contact, which is especially suitable for long-term health monitoring of the elderly or people with limited mobility;

[0063] Traditional wearable sensors have problems such as inconvenience in wearing and limited battery life, while millimeter-wave radar sensors overcome these limitations, providing a non-contact monitoring solution and significantly improving the scope of user experience.

[0064] S2: Use the global time synchronization mechanism to synchronize the time of all sensors, and calculate the position change by integration based on the initial position and initial velocity of the human body;

[0065] Use the global time synchronization mechanism to synchronize the time of all millimeter wave radar sensors;

[0066] Set at the initial time Human body position , at the initial moment The speed is ;

[0067] according to and , calculate the position change by integration, the expression is:

[0068] ;

[0069] in, Indicates at time location, Indicates the time interval, Indicates that at any time point in the integration process The speed at represents the integral variable, Represents the integral variable An infinitesimal increment of represents the initial time, Indicates the initial time The human body position, Indicates that at the initial moment speed;

[0070] This step converts the velocity data into position data to form a complete human motion trajectory.

[0071] S3: Calculate the speed change according to the human body position at the initial moment, calculate the position change and the direction angle change according to the position change, fuse the speed change, position change and direction angle change to obtain comprehensive features, use principal component analysis PCA to reduce the dimension of the extracted comprehensive features to obtain the fall features;

[0072] According to the initial speed and in time speed, calculation speed , the speed change expression is:

[0073] ;

[0074] in, Indicates at time speed;

[0075] Varies by location , calculate the position change , the position change expression is:

[0076] ;

[0077] Varies by location , calculate the change in direction angle , the expression of the direction angle change is:

[0078] ;

[0079] in, Indicates at time The change in direction angle, represents the inverse tangent function, Indicates at time The position changes Axis component, Indicates at time The position changes Axis component, Indicates that at the initial moment The direction angle of

[0080] Speed ,Location and the change in direction angle Fusion, get comprehensive features, the expression is:

[0081] ;

[0082] in, Indicates at time The comprehensive characteristics of The weight coefficient representing the speed change, The weight coefficient representing the position change, Indicates the maximum speed change, Indicates the maximum position change, Indicates the maximum direction angle change, Indicates at time The speed change, Indicates at time The position change, Indicates at time The change in direction angle;

[0083] The value range of is [0, 1]. When it is equal to 1, it means there is a fall behavior. When it is equal to 0, it indicates normal activity;

[0084] Principal component analysis (PCA) was used to reduce the dimensionality of the extracted comprehensive features and retain the key dimensions that represent the characteristics of falls;

[0085] By fusing the velocity change, position change and angular change into a comprehensive feature and performing normalization, the present invention can effectively integrate multiple features and improve the compactness and discrimination of feature representation;

[0086] A single feature is difficult to fully reflect the complexity of falling behavior, but multi-feature fusion can make up for this deficiency. By introducing weight coefficients, the present invention can adjust the importance of each feature according to actual conditions to ensure adaptability in different application scenarios. In addition, normalization processing makes different features comparable, avoids deviation problems caused by dimensional differences, and improves the effect of user behavior model training.

[0087] S4: Use machine learning algorithms to build a user behavior model, input fall features into the user behavior model, perform pattern recognition on the extracted fall features, and determine whether there is a fall behavior;

[0088] We chose to use a combination of convolutional neural networks (CNN) and long short-term memory (LSTM) to build a user behavior model. CNN is good at capturing local spatial features, while LSTM is good at processing long-term dependencies in time series data. This combination can more effectively identify fall features.

[0089] The combination algorithm of CNN and LSTM is selected to give full play to the advantages of the two models. CNN is good at capturing local spatial features, while LSTM is good at processing long-term dependencies in time series data. This combination can more comprehensively describe the characteristics of human motion and improve the accuracy of fall detection.

[0090] Existing technologies often rely on only a single model, which makes it difficult to capture complex spatial and temporal features at the same time. By introducing the CNN+LSTM combination, the present invention can capture changes in human motion status on a wider range of spatiotemporal scales, thereby significantly improving the sensitivity and accuracy of fall detection. For example, in complex multi-person activity scenes, a single model may cause false positives or false negatives due to the lack of time-dependent analysis, while the CNN+LSTM combination can better handle such situations.

[0091] Set the architecture of the user behavior model based on the convolutional neural network CNN and the long short-term memory network LSTM;

[0092] The architecture of the user behavior model refers to the input layer, convolution layer, pooling layer, LSTM layer and fully connected layer;

[0093] Collect and annotate a large dataset of behavioral patterns of normal activities and fall events;

[0094] The labeled behavior pattern data set is divided into training set, test set and validation set in proportion, and input into the user behavior model to train, test and validate the user behavior model respectively;

[0095] The comprehensive features identified by the CNN+LSTM combination algorithm Input the user behavior model and calculate the probability of falling behavior. The expression is:

[0096] ;

[0097] in, Indicates at time The probability of falling behavior, represents the Sigmoid activation function, Represents the weight vector The transpose of represents the bias term;

[0098] The Sigmoid activation function is used to map the result of the linear combination to the probability interval of [0, 1], making the output result more intuitive and easier to explain. This method not only simplifies the decision-making process, but also improves the response speed.

[0099] S5: Continuously collect new user data and user feedback, and continuously update user behavior model parameters;

[0100] Use statistical methods to screen newly collected user data and user feedback;

[0101] Merge the filtered behavior pattern dataset with the existing historical user data, and then calculate new comprehensive features And the corresponding probability of falling behavior, then adjust the user behavior model parameters to optimize performance, the expression is:

[0102] ;

[0103] in, Indicates at time The change in the parameters of the user behavior model, represents the learning rate, represents the gradient of the user behavior model parameters, represents the set of user behavior model parameters, Indicates at time The actual label, indicating that at time Whether a fall event occurred, obtained from the labeled data set, is used for model training and verification in the supervised learning process. represents the loss function; at regular intervals, a batch of new user data and user feedback is accumulated to form a training data set;

[0104] Use the accumulated training data set to input the user behavior model, perform comprehensive retraining, and update the user behavior model parameters. The update expression is:

[0105] ;

[0106] in, Represents the new model parameters obtained after retraining, represents the number of samples in the accumulated new data set, The index variable representing the sample, Indicates The key features of the samples are Indicates The actual labels of samples, represents the loss function, represents the regularization coefficient, represents the regularization term, Indicates found The smallest value among

[0107] Find the optimal model parameters by minimizing the overall loss function and adding regularization terms , this formula not only takes into account the loss of all samples, but also prevents overfitting through regularization, ensuring that the model has good generalization ability;

[0108] A single update method may cause large fluctuations in model performance and make it difficult to maintain stability. By introducing a comprehensive update formula, the present invention can gradually optimize model parameters in multiple iterations to ensure that it is always in the optimal state. For example, in long-term monitoring, the model can adapt to behavioral changes in different time periods through continuous updates, providing a more reliable fall warning service.

[0109] See also Figure 2 This embodiment also provides a non-contact fall detection device, including: a collection module, a motion trajectory module, a feature extraction module, a user behavior module and an update module:

[0110] An acquisition module, used to receive the user's position and speed information detected by multiple millimeter wave radar sensors installed in the detection area, and pre-process the information;

[0111] The motion trajectory module calculates the position change by integration based on the human body position and speed at the initial moment to form a complete human body motion trajectory;

[0112] The feature extraction module calculates the speed change, position change and direction angle change according to the human body motion trajectory, and fuses the speed change, position change and direction angle change into a comprehensive feature;

[0113] User behavior module: A combination of convolutional neural network (CNN) and long short-term memory (LSTM) is used to construct a user behavior model. The comprehensive features are input into the user behavior model to calculate the probability of falling behavior.

[0114] Update module, use statistical methods to screen newly collected user data and user feedback, merge them with historical data, adjust user behavior model parameters, optimize performance, regularly accumulate new user data, conduct comprehensive retraining, and update user behavior model parameters.

[0115] This embodiment also provides a computer device, which is applicable to the case of a non-contact fall detection method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the non-contact fall detection method proposed in the above embodiment.

[0116] The computer device may be a terminal, and 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, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0117] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the non-contact fall detection method proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0118] In summary, the present invention realizes large-area contactless monitoring by installing multiple millimeter-wave radar sensors in the detection area, and ensures the accuracy and reliability of data through filtering, noise reduction and smoothing. It uses the global time synchronization mechanism and integral calculation method to construct a complete human motion trajectory, improves the sensitivity and accuracy of fall detection, extracts features such as speed change, position change and angular change from the motion trajectory, and reduces the dimension through principal component analysis PCA, thereby enhancing the compactness and discrimination of feature representation. It uses the convolutional neural network CNN and the long short-term memory network LSTM combined algorithm to construct a user behavior model, giving full play to the ability of CNN to capture local spatial features and LSTM to process time series data, improving the intelligence level and classification performance of pattern recognition, and simplifying the decision-making process and improving the response speed of the system by inputting comprehensive features into the user behavior model and calculating the probability of falling. In addition, the present invention introduces a continuous data update and model optimization mechanism to ensure that the system can adapt to environmental changes and differences in personal behavior habits, maintain high accuracy and robustness in long-term operation, and provide reliable protection for the safety of the elderly and people with limited mobility.

[0119] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A non-contact fall detection method, characterized in that: include: receiving user data detected by multiple sensors installed in the detection area and preprocessing the user data; All sensors are synchronized using a global time synchronization mechanism, and the position change is calculated by integration based on the initial position and initial velocity of the human body; According to the human body position at the initial moment, the speed change is calculated. According to the position change, the position change and the direction angle change are calculated. The speed change, position change and direction angle change are fused to obtain comprehensive features. The principal component analysis PCA is used to reduce the dimension of the extracted comprehensive features to obtain the fall features. Use machine learning algorithms to build a user behavior model, input fall features into the user behavior model, perform pattern recognition on the extracted fall features, and determine whether there is a fall behavior; Continuously collect new user data and user feedback, and continuously update user behavior model parameters; The expression of the comprehensive feature is: ； in, Indicates at time The comprehensive characteristics of The weight coefficient representing the speed change, The weight coefficient representing the position change, Indicates the maximum speed change, Indicates the maximum position change, Indicates the maximum direction angle change, Indicates at time The speed change, Indicates at time The position change, Indicates at time The change in direction angle.

2. The non-contact fall detection method according to claim 1, characterized in that: The receiving of user data detected by multiple sensors installed in the detection area and preprocessing the user data comprises the following specific steps: The sensor refers to a millimeter wave radar sensor; The user data refers to the position and speed of the human body obtained by the millimeter wave radar sensor; Preprocessing of user data mainly refers to filtering, noise reduction and smoothing, and finally standardization.

3. The non-contact fall detection method according to claim 2, characterized in that: The expression for calculating the position change by integration is: ； in, Indicates at time location, Indicates the time interval, Indicates any time point during the integration process The speed at represents the integral variable, Represents the integral variable An infinitesimal increment of represents the initial time, Indicates the initial time The position of the human body, Indicates that at the initial moment speed.

4. The non-contact fall detection method according to claim 3, characterized in that: The specific steps of using machine learning algorithms to build a user behavior model are as follows: Choose to use the combined algorithm of convolutional neural network CNN and long short-term memory network LSTM to build user behavior model; Set the architecture of the user behavior model based on the convolutional neural network CNN and the long short-term memory network LSTM; The architecture of the user behavior model refers to the input layer, convolution layer, pooling layer, LSTM layer and fully connected layer; Collect and annotate a large dataset of behavioral patterns of normal activities and fall events; The labeled behavior pattern dataset is divided into training set, test set and validation set in proportion, and input into the user behavior model for training, testing and validation of the user behavior model respectively.

5. The non-contact fall detection method according to claim 4, characterized in that: The step of inputting the fall feature into the user behavior model, performing pattern recognition on the extracted fall feature, and determining whether there is a fall behavior is as follows: The comprehensive features identified by the CNN+LSTM combination algorithm Input the user behavior model and calculate the probability of falling behavior. The expression is: ； in, Indicates at time The probability of falling behavior, represents the Sigmoid activation function, Represents the weight vector The transpose of Represents the bias term.

6. The non-contact fall detection method according to claim 5, characterized in that: The specific steps of continuously collecting new user data and user feedback and continuously updating the user behavior model parameters are as follows: Use statistical methods to screen newly collected user data and user feedback; Merge the filtered behavior pattern dataset with the existing historical user data, and then calculate new comprehensive features And the corresponding probability of falling behavior, then adjust the user behavior model parameters to optimize performance, the expression is: ； in, Indicates at time The change in the parameters of the user behavior model, represents the learning rate, represents the gradient of the user behavior model parameters, represents the set of user behavior model parameters, Indicates at time The actual label, represents the loss function; at regular intervals, a batch of new user data and user feedback is accumulated to form a training data set; Use the accumulated training data set to input the user behavior model, perform comprehensive retraining, and update the user behavior model parameters. The update expression is: ； in, Represents the new model parameters obtained after retraining, represents the number of samples in the accumulated new data set, The index variable representing the sample, Indicates The key features of the samples are Indicates The actual labels of samples, represents the loss function, represents the regularization coefficient, represents the regularization term, Indicates found The smallest value in .

7. A non-contact fall detection device, based on the non-contact fall detection method according to any one of claims 1 to 6, characterized in that: Including acquisition module, motion trajectory module, feature extraction module, user behavior module and update module: The acquisition module is used to receive the user's position and speed information detected by multiple millimeter wave radar sensors installed in the detection area, and pre-process the information; The motion trajectory module calculates the position change by integration according to the human body position and speed at the initial moment to form a complete human body motion trajectory; The feature extraction module calculates the speed change, position change and direction angle change according to the human body motion trajectory, and fuses the speed change, position change and direction angle change into a comprehensive feature; The user behavior module selects a combination algorithm of a convolutional neural network (CNN) and a long short-term memory network (LSTM) to construct a user behavior model, inputs comprehensive features into the user behavior model, and calculates the probability of falling behavior; The update module uses statistical methods to screen newly collected user data and user feedback, and merges them with historical data, adjusts user behavior model parameters, optimizes performance, regularly accumulates new user data, conducts comprehensive retraining, and updates user behavior model parameters.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the non-contact fall detection method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the non-contact fall detection method according to any one of claims 1 to 6 are implemented.

Images