Patient falling alarm method and system based on visual identification

Abstract

The invention discloses a patient falling alarm method and system based on visual identification, and relates to the field of identification alarm, and the method comprises the following steps: 1, arranging a video obtaining part, carrying out the transmission through employing a real-time streaming transmission protocol, obtaining the video data of the body side of a patient in a preset range, carrying out the noise reduction of the video data, and obtaining a video collection part; the contrast ratio is enhanced; step 2, dynamically identifying obstacles within a preset range of the body side of the patient in the video data, acquiring positions, shapes and material features of the obstacles, and forming an obstacle data set within the preset range; by capturing and analyzing actions of a patient in real time, dynamic behaviors and surrounding environment information of the patient are comprehensively obtained, and by dynamically monitoring the actions of the patient and surrounding environment obstacles, the risk can be predicted and an alarm can be given out in time before falling, so that the effectiveness of falling early warning is improved by means of active intervention; the occurrence of actual falling of the patient can be reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of recognition alarm, and in particular to a patient fall alarm method and system based on visual recognition. Background Art

[0002] As public health requirements for medical institutions and care environments continue to increase, fall prevention has become an important medical safety indicator. Many countries and regions are promoting policies related to fall risk assessment and management, aiming to reduce the incidence of patient falls and ensure the safety and health of patients. The injuries caused by falls not only affect life and health, but also bring huge economic burdens to individuals and families. At the same time, hospitalization expenses, rehabilitation treatment and long-term care expenses have greatly increased the pressure on social medical resources. The society's awareness of fall risks and their consequences has gradually increased. More and more patients and their families pay attention to safety and quality of care when seeking medical services. Many existing fall alarm systems rely on physical sensors, such as accelerometers and gyroscopes, to detect the patient's movements. However, this method often has the problem of false positives and false negatives, because it can only monitor specific movements and cannot fully understand the patient's behavior and environment. In addition, in complex environments, the types and locations of obstacles vary greatly, and traditional methods are difficult to effectively identify such changes and corresponding risks. Existing fall alarm systems find it difficult to analyze patient behavior and changes in the surrounding environment in real time, and often only monitor under specific conditions. They have difficulty responding in a timely manner in rapidly changing scenarios. When a fall risk is identified, many systems can only issue an alarm, lack personalized response strategies and real-time communication functions, and cannot effectively guide patients or caregivers on how to respond. Summary of the invention

[0003] 1. Technical issues to be resolved In view of the above-mentioned shortcomings of the prior art, the present invention provides a patient fall alarm method and system based on visual recognition, which can effectively solve the problems of the prior art.

[0004] (II) Technical solution To achieve the above objectives, the present invention is implemented through the following technical solutions: The present invention discloses a patient fall alarm method based on visual recognition, comprising the following steps: Step 1: Arrange the video acquisition component, use the real-time streaming protocol for transmission, obtain the video data of the patient's side within a preset range, reduce the noise of the video data, and enhance the contrast; Step 2: Dynamically identify obstacles within a preset range beside the patient in the video data, obtain the position, shape and material characteristics of the obstacles, and form an obstacle data set within the preset range; Step 3: Capture the patient's movements and behaviors in the video data, and extract the patient's sitting, lying, and walking postures and the corresponding movement amplitudes of the postures through the key point detection algorithm to form a movement behavior dataset; Step 4: Establish a risk coefficient model that includes several types of action behaviors. Based on the current patient's action behaviors and surrounding environmental obstacle data, combined with machine learning algorithms and pre-acquired historical fall case data, the patient's current behavior risk coefficient is calculated through the trained risk coefficient model; Step 5: Preset a dynamic standard threshold, obtain the patient's current behavior risk coefficient obtained in step 4, and compare it with the current dynamic standard threshold. Once the patient's current behavior risk coefficient exceeds the standard threshold, it is immediately determined that this behavior has the risk of falling; Step 6: When a behavior that poses a risk of falling is detected, an alarm mechanism is immediately triggered, and an instant notification is sent to the patient and accompanying personnel; Step 7: Identify the specific type of the patient's current action behavior and match temporary coping measures in the preset coping strategy library; Step 8: Through the integrated voice broadcast function, the matched temporary response measures are sent to the patient and accompanying personnel in real time.

[0005] Furthermore, the video acquisition components in step 1 are: an indoor fixed infrared camera and a personnel-wearable motion infrared camera.

[0006] Furthermore, the method for setting the dynamic standard threshold in step 4 is: formulate an initial threshold template corresponding to each type of behavior, and extract the real-time feature data of the current patient's action behavior, set the segmentation threshold based on the feature distribution of the real-time feature data, and dynamically adjust the initial threshold template of the corresponding type of behavior.

[0007] Furthermore, the expression of the risk coefficient model established in step 4 is: ; Where: represents the risk factor of the patient's current behavior, represents the frequency of occurrence of the patient's i-th state, n represents the number of all states of the patient, Represents the risk weight of each patient status. Represents the number of action types, represents the weighting factor of the corresponding action, represents the dangerousness measure of various actions, Represents the number of types of obstacles, Representative The weight of the impact of each obstacle on the patient, Representative The danger value of an obstacle, Represents environmental factors, and combines various environmental characteristics to map out corresponding risk factors. Represents the frequency of falls in historical fall cases among patients with similar behavioral characteristics to the current patient. Represents the total number of historical cases, Represents the number of available historical case types, represents the number of patients at risk, represents the number of patients with similar behavior to the current patient, Represents the patient status factor weight, represents the risk weight of different patient status, and is used to measure the impact of different status on the risk of falling. It stands for risk weight, which indicates the risk weight of the patient's movement range and is used to assess the risk of falling when the patient performs a specific movement. Represents the obstacle factor weight, indicating the degree of influence of obstacles in the environment on the patient's fall risk. Represents the weight of environmental factors, which is used to indicate the impact of environmental conditions on the risk of patients falling. Represents the weight of the historical data factor, which is used to measure the impact of historical data on the current patient fall risk assessment.

[0008] Furthermore, the working logic of the expression of the risk coefficient model is: Calculate the frequency and risk weight of each action; Identify and extract the patient's motion range characteristics, and calculate the risk measurement and weighting factors of the motion range according to the defined risk measurement method; Identify obstacles around the patient and assign a criticality value and impact weight to each obstacle; Record environmental conditions and calculate environmental factors through functions; Analyze historical case data to calculate the frequency, total number of cases, and total number of patients with similar fall events to the current patient behavior; Optimize the weight parameters of each factor, and finally substitute all required factors into the calculation to output the risk coefficient.

[0009] Furthermore, the strategy types in the response strategy library in step 7 include: Responses to rapid turning behaviors: Turn obstacle prompts, turn deceleration prompts, and use of assistive devices prompts; Responses to unstable standing behavior: caregivers provide walking reminders, balance maintenance reminders, posture adjustment reminders, and rest advice reminders; Responses to anxious walking behavior: walking slowdown reminders, reminders for accompanying walking personnel, and obstacle identification reminders; Responses to squatting or bending over: tool-assisted standing prompts, caregiver-assisted standing prompts, and posture status prompts.

[0010] Patient fall alarm system based on visual recognition, including: The management module, as the core control terminal of each module and unit, edits and sends control instructions; A video capture unit, used for acquiring image data around the patient by using a real-time streaming protocol, and performing denoising and contrast enhancement processing on the image; The obstacle detection module is used to analyze video data, identify and locate obstacles around the patient, extract the location information, shape characteristics and material properties of the obstacles, and generate an obstacle attribute data set; The motion detection module is used to capture the patient's movements from the video data, extract the patient's sitting, lying and walking postures and their movement amplitudes through key point detection technology, and form a motion behavior data set; The risk assessment module is used to build a risk coefficient model. Based on the patient's current behavior and information about obstacles in the surrounding environment, the patient's behavior risk coefficient is calculated through a machine learning algorithm combined with historical fall data. The threshold comparison module is used to set a dynamic standard threshold, and compare the patient behavior risk factor calculated by the motion detection module with the threshold. When the risk factor exceeds the threshold, it can be determined that there is a risk of falling; An alarm trigger module is used to immediately activate the alarm mechanism when a fall risk behavior is detected, and send an instant warning notification to the patient and caregiver; The strategy matching module is used to identify the patient's current action type, match it in the preset response strategy library, and select the associated temporary response measures; The voice announcement module is used to convey matching temporary response measures to patients and caregivers in real time through integrated voice functions.

[0011] Furthermore, the management module is interactively connected to a storage module via a wireless network, and the storage module is used for centrally managing historical data and real-time input of captured data, and for classification and distributed storage.

[0012] Furthermore, during the operation phase, the threshold comparison module receives the behavioral risk coefficient submitted by the risk assessment module and compares it with the dynamic standard threshold. If it does not exceed the current dynamic standard threshold, it jumps to the obstacle detection module and the motion detection module for further operation. If it exceeds the current dynamic standard threshold, it jumps to the alarm trigger module and the strategy matching module for further operation.

[0013] Furthermore, the management module is interactively connected to the video capture unit through a wireless network, the video capture unit is interactively connected to the obstacle detection module and the motion detection module through a wireless network, the danger assessment module is interactively connected to the obstacle detection module, the motion detection module and the threshold comparison module through a wireless network, the alarm trigger module is interactively connected to the threshold comparison module and the strategy matching module through a wireless network, and the strategy matching module is interactively connected to the voice broadcast module through a wireless network.

[0014] (III) Beneficial effects Compared with the known prior art, the technical solution provided by the present invention has the following beneficial effects: By using visual recognition technology, the patient's movements can be captured and analyzed in real time, the patient's movements and behaviors can be classified and evaluated, and the patient's dynamic behavior and surrounding environmental information can be fully obtained. Using key point detection algorithms, the patient's sitting, lying and walking postures can be accurately identified. By dynamically identifying the location and characteristics of surrounding obstacles and combining environmental factors for comprehensive evaluation, the system can work effectively in different environments, be more adaptable, and be able to respond immediately to potential fall risks. Furthermore, by dynamically monitoring the patient's movements and surrounding environmental obstacles, the risk can be predicted before a fall occurs and an alarm can be issued in time, thereby improving the effectiveness of fall warnings through active intervention and reducing the occurrence of actual falls in patients.

[0015] By comprehensively considering the patient's behavior, environmental obstacles and historical data, and dynamically adjusting the risk assessment model through machine learning algorithms, a more accurate risk assessment can be achieved. Compared with single-dimensional technology, multi-dimensional analysis can more effectively identify potential fall risks.

[0016] Through real-time monitoring and analysis of historical data, the standard threshold of action behavior is dynamically adjusted to improve the adaptability to specific patient conditions and environmental changes, avoiding the misjudgment or missed judgment problems caused by static thresholds. This flexibility enables the system to adapt to the individual differences and changes of different patients, thereby improving the accuracy of overall action recognition and reducing the possibility of false alarms and missed alarms.

[0017] By combining alarms and response strategy matching, automatic alarms and immediate responses are achieved, and a voice broadcast function is integrated. Once a risk is detected, the alarm can be automatically triggered and a response plan can be provided, reducing the workload of medical staff, improving the timeliness of rescue, and enhancing user experience and emergency response capabilities. The collaborative work of the two effectively reduces the consequences of falls. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0019] Figure 1 It is a flow chart of the patient fall alarm method based on visual recognition in the present invention; Figure 2 It is a schematic diagram of the framework of the patient fall alarm system based on visual recognition in the present invention.

[0020] The numbers in the figure represent: 1. Management module; 2. Video capture unit; 3. Obstacle detection module; 4. Motion detection module; 5. Danger assessment module; 6. Threshold comparison module; 7. Alarm trigger module; 8. Strategy matching module; 9. Voice broadcast module; 10. Storage module. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0022] The present invention will be further described below in conjunction with the embodiments.

[0023] Embodiment 1: The patient fall alarm method based on visual recognition in this embodiment, such as Figure 1 As shown, the following steps are included: Step 1: Arrange video acquisition components, use real-time streaming protocol for transmission, obtain video data of the patient's side within a preset range, reduce noise on the video data, and enhance contrast; the video acquisition components are: indoor fixed infrared cameras and personnel-worn motion infrared cameras; Step 2: Dynamically identify obstacles within a preset range beside the patient in the video data, obtain the position, shape and material characteristics of the obstacles, and form an obstacle data set within the preset range; Step 3: Capture the patient's movements and behaviors in the video data, and extract the patient's sitting, lying, and walking postures and the corresponding movement amplitudes of the postures through the key point detection algorithm to form a movement behavior dataset; Step 4: Establish a risk coefficient model including several types of action behaviors. Based on the current patient's action behaviors and surrounding environmental obstacle data, combined with machine learning algorithms and pre-acquired historical fall case data, the patient's current behavior risk coefficient is calculated through the trained risk coefficient model; the method for setting the dynamic standard threshold is: formulate an initial threshold template corresponding to each type of behavior, and extract the real-time feature data of the current patient's action behavior, set the segmentation threshold based on the feature distribution of the real-time feature data, and dynamically adjust the initial threshold template of the corresponding type of behavior; Step 5: Preset a dynamic standard threshold, obtain the patient's current behavior risk coefficient obtained in step 4, and compare it with the current dynamic standard threshold. Once the patient's current behavior risk coefficient exceeds the standard threshold, it is immediately determined that this behavior has the risk of falling; Step 6: When a behavior that poses a risk of falling is detected, an alarm mechanism is immediately triggered, and an instant notification is sent to the patient and accompanying personnel; Step 7: Identify the specific type of the patient's current action behavior, and match temporary response measures in the preset response strategy library; the strategy types in the response strategy library in step 7 include: response to rapid turning behavior: turning obstacle prompts, turning deceleration prompts, and use of assistive devices prompts; response to unstable standing behavior: accompanying walking prompts, maintaining balance prompts, adjusting posture prompts, and rest suggestion prompts; response to anxious walking behavior: walking deceleration prompts, accompanying walking prompts, and obstacle identification prompts; response to squatting or bending over: tool-assisted standing prompts, accompanying standing prompts, and posture status prompts; Step 8: Through the integrated voice broadcast function, the matched temporary response measures are sent to the patient and accompanying personnel in real time.

[0024] Compared with existing technologies, the use of video monitoring and key point detection algorithms can obtain the patient's movement status and surrounding environment information in real time. Compared with traditional fall monitoring systems, this method can more accurately analyze the patient's specific behavior, reduce the false alarm rate, and improve the accuracy of the alarm. It can dynamically identify the obstacle information around the patient, including location, shape and material, and adapt to environmental changes in various situations. Compared with detection methods based on gravity or acceleration, visual recognition methods can more comprehensively understand the relationship between patients and the environment; Through the risk factor model, combined with the patient's historical fall data and real-time behavior analysis, personalized risk assessment is achieved. The risk assessment can be automatically adjusted according to each patient's movement characteristics and historical data, improving the accuracy of early warning. The dynamic standard threshold is adopted instead of a fixed alarm setting. The threshold can be adjusted in real-time monitoring to adapt to the patient's behavioral changes in different states and avoid false alarms or missed alarms. After identifying the risk of falling, not only can an alarm be issued, but personalized response measures can also be provided based on the specific risk behavior to ensure that the patient is protected at all times. Information can be conveyed directly to patients and accompanying personnel, which improves the effect and eliminates the need for them to actively check or respond to the alarm device, thereby improving response efficiency.

[0025] Embodiment 2: In other aspects, this embodiment also provides a patient fall alarm system based on visual recognition, such as Figure 2 As shown, including: Management module 1, as the core control terminal of each module and unit, edits and sends control instructions; The video capture unit 2 is used to obtain image data around the patient by using a real-time streaming protocol, and to perform denoising and contrast enhancement processing on the image; the management module 1 is interactively connected to a storage module 10 via a wireless network, and the storage module 10 is used to centrally manage historical data and input the captured data in real time, and to classify and distribute the data; Obstacle detection module 3, used to analyze video data, identify and locate obstacles around the patient, extract the location information, shape characteristics and material properties of the obstacles, and generate an obstacle attribute data set; Motion detection module 4 is used to capture the patient's movements from the video data, extract the patient's sitting, lying and walking postures and their movement amplitudes through key point detection technology, and form a motion behavior data set; The risk assessment module 5 is used to construct a risk coefficient model, which calculates the patient's behavior risk coefficient based on the patient's current behavior and information about obstacles in the surrounding environment through a machine learning algorithm combined with historical fall data; The threshold comparison module 6 is used to set a dynamic standard threshold, and compare the patient behavior risk coefficient calculated by the motion detection module 4 with the threshold. When it is found that the risk coefficient exceeds the threshold, it can be determined that there is a risk of falling. In the operation stage, the threshold comparison module 6 receives the behavior risk coefficient submitted by the risk assessment module 5 and compares it with the dynamic standard threshold. If it does not exceed the current dynamic standard threshold, it jumps to the obstacle detection module 3 and the motion detection module 4 for further operation. If it exceeds the current dynamic standard threshold, it jumps to the alarm trigger module 7 and the strategy matching module 8 for further operation. The alarm trigger module 7 is used to immediately activate the alarm mechanism when a behavior that poses a risk of falling is detected, and send an instant warning notification to the patient and the caregiver; The strategy matching module 8 is used to identify the patient's current action type, match it in the preset coping strategy library, and select the associated temporary coping measures; The voice broadcast module 9 is used to convey matching temporary response measures to patients and nursing staff in real time through the integrated voice function.

[0026] As an implementation method in this embodiment, Figure 2 As shown, the management module 1 is interactively connected with the video capture unit 2 through a wireless network, the video capture unit 2 is interactively connected with the obstacle detection module 3 and the motion detection module 4 through a wireless network, the danger assessment module 5 is interactively connected with the obstacle detection module 3, the motion detection module 4 and the threshold comparison module 6 through a wireless network, the alarm trigger module 7 is interactively connected with the threshold comparison module 6 and the strategy matching module 8 through a wireless network, and the strategy matching module 8 is interactively connected with the voice broadcast module 9 through a wireless network.

[0027] In the specific implementation of this embodiment, the system can instantly identify the patient's movement status and promptly discover potential fall risks through real-time image acquisition and processing. Through environmental modeling and obstacle recognition, the system can comprehensively consider the impact of the surrounding environment on the patient's behavior, thereby providing a more accurate risk assessment. Based on machine learning, the system can automatically adjust the risk factor model according to the specific behavior pattern and historical data of each patient, thereby realizing personalized safety monitoring. It takes into account multiple dimensions such as the patient's range of motion, environmental factors, and historical fall cases to provide a more comprehensive risk assessment. When high-risk behaviors are detected, it can quickly issue an alarm, provide instant notifications and suggestions to patients and caregivers, and enhance the effectiveness of intervention.

[0028] Embodiment 3: This embodiment is an improvement based on Embodiment 1. In this embodiment, the risk coefficient model is expressed as follows: ; Where: represents the risk factor of the patient's current behavior, represents the frequency of occurrence of the patient's i-th state, n represents the number of all states of the patient, Represents the risk weight of each patient status. Represents the number of action types, represents the weighting factor of the corresponding action, represents the dangerousness measure of various actions, Represents the number of types of obstacles, Representative The weight of the impact of each obstacle on the patient, Representative The danger value of an obstacle, Represents environmental factors, and combines various environmental characteristics to map out corresponding risk factors. Represents the frequency of falls in historical fall cases among patients with similar behavioral characteristics to the current patient. Represents the total number of historical cases, Represents the number of available historical case types, represents the number of patients at risk, represents the number of patients with similar behavior to the current patient, Represents the patient status factor weight, which represents the risk weight of different patient statuses, such as sitting, standing, and walking. It is used to measure the impact of different statuses on the risk of falling. The larger the value, the more important the status is to the risk of falling. It stands for risk weight, which indicates the risk weight of the patient's movement range. It is used to assess the risk of falling when the patient performs a specific movement. The greater the change in walking speed and turning angle, the higher the potential risk of falling. Therefore, this weight reflects the importance of this factor in the overall risk assessment. Represents the weight of the obstacle factor, which indicates the degree of influence of obstacles in the environment on the risk of falling of patients. Different types of obstacles contribute differently to the risk, so this weight can assess the relative importance of environmental factors in the risk of falling. Represents the weight of environmental factors, which is used to indicate the impact of environmental conditions on the risk of falling. Different environmental conditions can affect the patient's gait and reaction ability, so it plays a key role in the model. Represents the weight of historical data factors, which is used to measure the impact of historical data on the current patient fall risk assessment. By weighting historical data, the time dimension can be taken into consideration, making the assessment more comprehensive and accurate; The working logic of the expression of the hazard coefficient model is: Calculate the frequency and risk weight of each action; Identify and extract the patient's motion range characteristics, and calculate the risk measurement and weighting factors of the motion range according to the defined risk measurement method; Identify obstacles around the patient and assign a criticality value and impact weight to each obstacle; Record environmental conditions and calculate environmental factors through functions; Analyze historical case data to calculate the frequency, total number of cases, and total number of patients with similar fall events to the current patient behavior; Optimize the weight parameters of each factor, and finally substitute all required factors into the calculation to output the risk coefficient.

[0029] In summary, the present invention can monitor the patient's activities in real time without manual intervention, ensuring that timely responses can be made to the patient's dynamics at any time. Once a potential fall risk is detected, the system can immediately issue an alarm to notify the nursing staff to intervene and reduce the possibility of a fall event. Through computer vision technology, the patient's posture and movement can be accurately analyzed to reduce the risk of false positives and false negatives. At the same time, environmental factors and obstacles are taken into account, making the fall risk assessment more comprehensive and reliable. By utilizing historical fall case data and machine learning algorithms, we can continuously optimize the risk factor model, improve the scientificity and accuracy of risk assessment, and provide personalized risk assessment and intervention recommendations based on each patient's unique behavior patterns and environment to meet the needs of different patients.

[0030] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A patient fall alarm method based on visual recognition, characterized in that: The following steps are involved: Step 1: Arrange the video acquisition component, use the real-time streaming protocol for transmission, obtain the video data of the patient's side within a preset range, reduce the noise of the video data, and enhance the contrast; Step 2: Dynamically identify obstacles within a preset range beside the patient in the video data, obtain the position, shape and material characteristics of the obstacles, and form an obstacle data set within the preset range; Step 3: Capture the patient's movements and behaviors in the video data, and extract the patient's sitting, lying, and walking postures and the corresponding movement amplitudes of the postures through the key point detection algorithm to form a movement behavior dataset; Step 4: Establish a risk coefficient model that includes several types of action behaviors. Based on the current patient's action behaviors and surrounding environmental obstacle data, combined with machine learning algorithms and pre-acquired historical fall case data, the patient's current behavior risk coefficient is calculated through the trained risk coefficient model; Step 5: Preset a dynamic standard threshold, obtain the patient's current behavior risk coefficient obtained in step 4, and compare it with the current dynamic standard threshold. Once the patient's current behavior risk coefficient exceeds the standard threshold, it is immediately determined that this behavior has the risk of falling; Step 6: When a behavior that poses a risk of falling is detected, an alarm mechanism is immediately triggered, and an instant notification is sent to the patient and accompanying personnel; Step 7: Identify the specific type of the patient's current action behavior and match temporary coping measures in the preset coping strategy library; Step 8: Through the integrated voice broadcast function, the matched temporary response measures are sent to the patient and accompanying personnel in real time.

2. The patient fall alarm method based on visual recognition according to claim 1, characterized in that: The video acquisition components in step 1 are: an indoor fixed infrared camera and a wearable motion infrared camera.

3. The patient fall alarm method based on visual recognition according to claim 1, characterized in that: The method for setting the dynamic standard threshold in step 4 is: formulate an initial threshold template corresponding to each type of behavior, extract the real-time feature data of the current patient's action behavior, set the segmentation threshold based on the feature distribution of the real-time feature data, and dynamically adjust the initial threshold template of the corresponding type of behavior.

4. The patient fall alarm method based on visual recognition according to claim 1, characterized in that: The expression of the risk coefficient model established in step 4 is: ； Where: represents the risk factor of the patient's current behavior, represents the frequency of occurrence of the patient's i-th state, n represents the number of all states of the patient, Represents the risk weight of each patient status. Represents the number of action types, represents the weighting factor of the corresponding action, represents the riskiness of various actions, Represents the number of types of obstacles, Representative The weight of the impact of each obstacle on the patient, Representative The danger value of an obstacle, Represents environmental factors, Represents the frequency of falls in historical fall cases among patients with similar behavioral characteristics to the current patient. Represents the total number of historical cases, Represents the number of available historical case types, represents the number of patients at risk, represents the number of patients with similar behavior to the current patient, represents the patient status factor weight, represents the risk weight, represents the obstacle factor weight, Represents the weight of environmental factors, Represents the historical data factor weights.

5. The patient fall alarm method based on visual recognition according to claim 4 is characterized in that: The working logic of the expression of the risk coefficient model is: Calculate the frequency and risk weight of each action; Identify and extract the patient's motion range characteristics, and calculate the risk measurement and weighting factors of the motion range according to the defined risk measurement method; Identify obstacles around the patient and assign a criticality value and impact weight to each obstacle; Record environmental conditions and calculate environmental factors through functions; Analyze historical case data to calculate the frequency, total number of cases, and total number of patients with similar fall events to the current patient behavior; Optimize the weight parameters of each factor, and finally substitute all required factors into the calculation to output the risk coefficient.

6. The patient fall alarm method based on visual recognition according to claim 1, characterized in that: The strategy types in the response strategy library in step 7 include: Responses to rapid turning behaviors: Turn obstacle prompts, turn deceleration prompts, and use of assistive devices prompts; Responses to unstable standing behavior: caregivers provide walking reminders, balance maintenance reminders, posture adjustment reminders, and rest advice reminders; Responses to anxious walking behavior: walking slowdown reminders, reminders for accompanying walking personnel, and obstacle identification reminders; Responses to squatting or bending over: tool-assisted standing prompts, caregiver-assisted standing prompts, and posture status prompts.

7. A patient fall alarm system based on visual recognition, the system being an implementation system of the patient fall alarm method based on visual recognition according to any one of claims 1 to 6, characterized in that: include: The management module (1) is the core control terminal of each module and unit, and is responsible for editing and sending control instructions; A video capture unit (2) is used to obtain image data around the patient by using a real-time streaming protocol, and to perform denoising and contrast enhancement processing on the image; The obstacle detection module (3) is used to analyze the video data, identify and locate obstacles around the patient, extract the location information, shape characteristics and material properties of the obstacles, and generate an obstacle property data set; A motion detection module (4) is used to capture the patient's movements from the video data, extract the patient's sitting, lying and walking postures and their movement amplitudes through key point detection technology, and form a motion behavior data set; The risk assessment module (5) is used to construct a risk coefficient model, based on the patient's current behavior and information about obstacles in the surrounding environment, and to calculate the patient's behavioral risk coefficient by combining historical fall data with a machine learning algorithm; A threshold comparison module (6) is used to set a dynamic standard threshold, compare the patient behavior risk factor calculated by the motion detection module (4) with the threshold, and when it is found that the risk factor exceeds the threshold, it can be determined that there is a risk of falling; An alarm trigger module (7) is used to immediately activate the alarm mechanism when a behavior that poses a risk of falling is detected, and to send an instant warning notification to the patient and the caregiver; A strategy matching module (8) is used to identify the patient's current action type, match it in a preset coping strategy library, and select associated temporary coping measures; The voice announcement module (9) is used to convey matching temporary response measures to patients and nursing staff in real time through an integrated voice function.

8. The patient fall alarm system based on visual recognition according to claim 7, characterized in that: The management module (1) is interactively connected to a storage module (10) via a wireless network, and the storage module (10) is used for centrally managing historical data and real-time input of captured data, and for classification and distributed storage.

9. The patient fall alarm system based on visual recognition according to claim 7, characterized in that: During the operation phase, the threshold comparison module (6) receives the behavior risk coefficient submitted by the risk assessment module (5) and compares it with the dynamic standard threshold. If it does not exceed the current dynamic standard threshold, it jumps to the obstacle detection module (3) and the motion detection module (4) for further operation. If it exceeds the current dynamic standard threshold, it jumps to the alarm trigger module (7) and the strategy matching module (8) for further operation.

10. The patient fall alarm system based on visual recognition according to claim 7, characterized in that: The management module (1) is interactively connected to the video capture unit (2) via a wireless network; the video capture unit (2) is interactively connected to the obstacle detection module (3) and the motion detection module (4) via a wireless network; the danger assessment module (5) is interactively connected to the obstacle detection module (3), the motion detection module (4) and the threshold comparison module (6) via a wireless network; the alarm trigger module (7) is interactively connected to the threshold comparison module (6) and the strategy matching module (8) via a wireless network; and the strategy matching module (8) is interactively connected to the voice broadcast module (9) via a wireless network.

Images