Dynamic monitoring and trajectory tracking method and system for hospital personnel

Abstract

The invention discloses a hospital personnel dynamic monitoring and trajectory tracking method and system, relates to the technical field of hospital operation and maintenance monitoring, and solves the problems that an existing hospital illegal intrusion identification method depends on equipment or manpower and is insufficient in real-time performance and accuracy. Detecting and identifying the identity type of the person from the monitoring video in real time by combining the face feature vector and the clothing color texture feature; distributing a unique track ID for each detected person, storing the spatial position and the timestamp of the person, forming activity track data, and storing the activity track data in a database; dividing a monitoring area into a plurality of sub-areas according to a hospital map, and distributing the sub-areas with permissions as permissible areas according to the identity type of each person; when the movement track data of the person enters the right area, an abnormal alarm is triggered; and illegal invasion of unauthorized personnel is prevented by combining trajectory tracking and authorization management.

Description

Technical Field

[0001] The present invention relates to the technical field of hospital operation and maintenance monitoring, and more specifically, to a method and system for dynamic monitoring and trajectory tracking of hospital personnel. Background Art

[0002] Smart hospital is an emerging concept in the field of modern medicine. It realizes integrated hospital operation and maintenance monitoring and management through advanced Internet of Things technology and emerging technologies combined with artificial intelligence, aiming to improve the safety, operation efficiency and service quality of hospitals. As a public place with high-density flow of people, the key points of operation and maintenance monitoring and management of hospitals include: illegal intrusion identification. Illegal intrusion identification refers to tracking the activity trajectories of visitors and patients to prevent unauthorized persons from entering highly sensitive areas (such as pharmacies, operating rooms, and ICUs).

[0003] Most of the existing hospitals restrict access to areas through access cards or code scanning, but this relies on device authorization and is difficult to cover the monitoring of dynamic crowd behavior; there are also methods that use manual observation or simple alarm rules to protect areas, but the cost of manual intervention is high, and the real-time and accuracy are insufficient; there are also methods based on simple target detection and tracking, but it only supports a single person or a small number of targets, and it is difficult to cope with the complex scenarios of high-density crowds in hospitals; it can be seen that the current hospital illegal intrusion identification methods rely on equipment or manual work, and lack real-time and accuracy.

[0004] Therefore, the present application provides a method and system for dynamic monitoring and trajectory tracking of hospital personnel to solve the above problems. Summary of the invention

[0005] The purpose of this application is to provide a method and system for dynamic monitoring and trajectory tracking of hospital personnel, so as to solve the problem that the current hospital illegal intrusion identification method relies on equipment or manual work and lacks real-time and accuracy. This application combines trajectory tracking with authorization management to achieve dynamic monitoring and trajectory tracking of doctors, nurses, patients and visitors, prevent unauthorized intrusion, and improve the overall safety management level of the hospital.

[0006] The present application first provides a method for dynamic monitoring and trajectory tracking of hospital personnel, including: S1, obtaining hospital surveillance video, combining facial feature vectors and clothing color and texture features to detect and identify the identity type of personnel in real time from the surveillance video, and the identity types include: doctors, nurses, patients and visitors; S2, assigning a unique trajectory ID to each detected person, predicting the position of the person through Kalman filtering, matching the detection box and the historical trajectory through the Hungarian algorithm, matching based on IOU and appearance feature similarity, saving the spatial position and timestamp of the person, forming activity trajectory data and saving it to a database; S3, dividing the monitoring area into multiple sub-areas according to the hospital map, and assigning sub-areas with permissions according to the identity type of each person as permission areas; S4, when the activity trajectory data of the person enters outside the permission area, an abnormal alarm is triggered.

[0007] In a possible implementation, S1, obtain hospital surveillance video, combine facial feature vectors and clothing color texture features to detect and identify the identity type of personnel in real time from the surveillance video, and the identity types include: doctors, nurses, patients and visitors; including: collecting facial images and full-body images of current doctors, nurses, patients and visitors in the hospital, extracting facial feature vectors from facial images through ResNet-50 and marking corresponding identity categories, storing them in a facial feature vector library, extracting clothing color texture features from full-body images through color histogram statistics and LBP algorithm and marking corresponding identity categories, and storing them in different clothing feature libraries according to identity categories; when obtaining surveillance video, detecting personnel areas in the surveillance video screen through the YOLOv5 algorithm, extracting facial feature vectors in the personnel area through ResNet-50, querying consistent facial feature vectors in the existing facial feature vector library, and determining the identity type of the personnel; when no consistent facial feature vectors can be found in the existing facial feature vector library, extracting clothing color texture features in the personnel area through color histogram statistics and LBP algorithm, and comparing them with clothing color texture features in each clothing feature library to determine the identity type of the personnel.

[0008] In a possible implementation, S1, obtaining hospital surveillance video, combining facial feature vectors and clothing color texture features to detect and identify the identity type of personnel in real time from the surveillance video, the identity types including: doctors, nurses, patients and visitors; also including: when a consistent facial feature vector is queried in an existing facial feature vector library, the identity type of the personnel is obtained; the clothing color texture features in the personnel area are extracted through color histogram statistics and LBP algorithm, and compared with the clothing color texture features of each clothing feature library to verify the identity type of the personnel.

[0009] In a possible implementation, S4, when the activity trajectory data of a person enters outside the authorized area, an abnormal alarm is triggered; including: grading the abnormal behavior according to the stay time and number of stays of the activity trajectory data outside the authorized area, and executing different alarm strategies according to the grade.

[0010] In a possible implementation, abnormal behaviors are graded according to the residence time and number of stays of the activity trajectory data outside the permission area, and different alarm strategies are executed according to the grades; including: when the residence time of the activity trajectory data outside the permission area is less than a time threshold and the number of stays is less than a frequency threshold, it is a low-risk behavior; when the residence time of the activity trajectory data outside the permission area is greater than or equal to the time threshold and the number of stays is less than the frequency threshold, it is a medium-risk behavior; when the residence time of the activity trajectory data outside the permission area is greater than or equal to the time threshold and the number of stays is greater than or equal to the frequency threshold, it is a high-risk behavior.

[0011] The present application also provides a hospital personnel dynamic monitoring and trajectory tracking system, including: a target detection and identity recognition module, which is used to obtain hospital monitoring videos, and detect and identify the identity types of personnel in real time from the monitoring videos in combination with facial feature vectors and clothing color and texture features. The identity types include: doctors, nurses, patients and visitors; a multi-target trajectory tracking module, which is used to assign a unique trajectory ID to each detected person, predict the person's position through Kalman filtering, match the detection box and historical trajectory through the Hungarian algorithm, match based on IOU and appearance feature similarity, save the spatial position and timestamp of the person, form activity trajectory data and save it to a database; an area authorization management module, which is used to divide the monitoring area into multiple sub-areas according to the hospital map, and assign sub-areas with permissions according to the identity type of each person as permission areas; an abnormal alarm and data storage module, which is used to trigger an abnormal alarm when the activity trajectory data of a person enters outside the permission area.

[0012] In a possible implementation, the target detection and identity recognition module includes: a data acquisition module, which is used to collect facial images and full-body images of current doctors, nurses, patients and visitors in the hospital, extract facial feature vectors from facial images through ResNet-50 and mark corresponding identity categories, store them in a facial feature vector library, extract clothing color and texture features of full-body images through color histogram statistics and LBP algorithm and mark corresponding identity categories, and store them in different clothing feature libraries according to identity categories; a face recognition module, which is used to detect personnel areas in the surveillance video screen through the YOLOv5 algorithm when acquiring surveillance video, extract facial feature vectors in the personnel area through ResNet-50, query consistent facial feature vectors in an existing facial feature vector library, and determine the identity type of the personnel; a clothing color and texture feature recognition module, which is used to extract clothing color and texture features in the personnel area through color histogram statistics and LBP algorithm when no consistent facial feature vectors are found in the existing facial feature vector library, and compare them with clothing color and texture features in each clothing feature library to determine the identity type of the personnel.

[0013] In a possible implementation, the target detection and identity recognition module also includes: an identification verification module, which is used to obtain the identity type of the person when a consistent facial feature vector is queried in an existing facial feature vector library; extract the clothing color texture features in the person area through color histogram statistics and LBP algorithm, and compare them with the clothing color texture features of each clothing feature library to verify the identity type of the person.

[0014] In a possible implementation, the abnormal alarm and data storage module is specifically used to: classify abnormal behaviors according to the duration and number of stays outside the permission area of ​​the activity trajectory data, and execute different alarm strategies according to the classification.

[0015] In a possible implementation, the abnormal alarm and data storage module is further specifically used for: when the stay time of the activity trajectory data outside the permission area is less than the time threshold and the number of stays is less than the frequency threshold, it is a low-risk behavior; when the stay time of the activity trajectory data outside the permission area is greater than or equal to the time threshold and the number of stays is less than the frequency threshold, it is a medium-risk behavior; when the stay time of the activity trajectory data outside the permission area is greater than or equal to the time threshold and the number of stays is greater than or equal to the frequency threshold, it is a high-risk behavior.

[0016] Compared with the prior art, the present application has the following beneficial effects: First, based on the special application scenario of the hospital, an identity type detection method combining facial feature vectors and clothing color texture features is proposed to improve the accuracy of identity type detection of hospital personnel; second, a method combining personnel activity trajectory data and authority areas is proposed to realize real-time detection of illegal intrusions; third, a graded alarm is performed by combining movement trajectory data and authority areas to effectively improve security. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0018] Figure 1 A flowchart of a method for dynamic monitoring and trajectory tracking of hospital personnel provided in an embodiment of the present application;

[0019] Figure 2 This is a structural diagram of the hospital personnel dynamic monitoring and trajectory tracking system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0020] Hereinafter, the term "include" or "may include" used in various embodiments of the present application indicates the presence of the function, operation or element applied for, and does not limit the increase of one or more functions, operations or elements. In addition, as used in various embodiments of the present application, the terms "include", "have" and their cognates are intended only to indicate specific features, numbers, steps, operations, elements, components or a combination of the foregoing items, and should not be understood as first excluding the presence of one or more other features, numbers, steps, operations, elements, components or a combination of the foregoing items or the possibility of increasing one or more features, numbers, steps, operations, elements, components or a combination of the foregoing items.

[0021] The terms used in the various embodiments of the application are only used to describe the purpose of specific embodiments and are not intended to limit the various embodiments of the application. As used herein, the singular form is intended to also include the plural form, unless the context clearly indicates otherwise. Unless otherwise limited, all terms used here (including technical terms and scientific terms) have the same meaning as the meanings commonly understood by ordinary technicians in the field of the various embodiments of the application. The terms (such as the terms defined in the dictionary generally used) will be interpreted as having the same meaning as the contextual meaning in the relevant technical field and will not be interpreted as having an idealized meaning or an overly formal meaning, unless clearly defined in the various embodiments of the application.

[0022] In order to make the objectives, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with examples and drawings. The illustrative implementation scheme of the present application and its description are only used to explain the present application and are not intended to limit the present application.

[0023] See also Figure 1 As shown, Figure 1 A flowchart of a method for dynamic monitoring and trajectory tracking of hospital personnel provided in an embodiment of the present application. The method includes: S1, obtaining hospital surveillance video, combining facial feature vectors and clothing color and texture features to detect and identify the identity type of personnel in real time from the surveillance video, and the identity types include: doctors, nurses, patients and visitors; S2, assigning a unique trajectory ID to each detected person, predicting the position of the person through Kalman filtering, matching the detection frame and the historical trajectory through the Hungarian algorithm, matching based on IOU and appearance feature similarity, saving the spatial position and timestamp of the person, forming activity trajectory data and saving it to the database; S3, dividing the monitoring area into multiple sub-areas according to the hospital map, and assigning sub-areas with permissions according to the identity type of each person as permission areas; S4, when the activity trajectory data of the person enters outside the permission area, triggering an abnormal alarm.

[0024] In a possible implementation, S1, obtain hospital surveillance video, combine facial feature vectors and clothing color texture features to detect and identify the identity type of personnel in real time from the surveillance video, and the identity types include: doctors, nurses, patients and visitors; including: collecting facial images and full-body images of current doctors, nurses, patients and visitors in the hospital, extracting facial feature vectors from facial images through ResNet-50 and marking corresponding identity categories, storing them in a facial feature vector library, extracting clothing color texture features from full-body images through color histogram statistics and LBP algorithm and marking corresponding identity categories, and storing them in different clothing feature libraries according to identity categories; when obtaining surveillance video, detecting personnel areas in the surveillance video screen through the YOLOv5 algorithm, extracting facial feature vectors in the personnel area through ResNet-50, querying consistent facial feature vectors in the existing facial feature vector library, and determining the identity type of the personnel; when no consistent facial feature vectors can be found in the existing facial feature vector library, extracting clothing color texture features in the personnel area through color histogram statistics and LBP algorithm, and comparing them with clothing color texture features in each clothing feature library to determine the identity type of the personnel.

[0025] Furthermore, S1, obtaining hospital surveillance video, combining facial feature vectors and clothing color texture features to detect and identify the identity type of personnel in real time from the surveillance video, the identity types including: doctors, nurses, patients and visitors; also including: when a consistent facial feature vector is queried in an existing facial feature vector library, the identity type of the personnel is obtained; the clothing color texture features in the personnel area are extracted through color histogram statistics and LBP algorithm, and compared with the clothing color texture features of each clothing feature library to verify the identity type of the personnel.

[0026] Step S1 is to detect and identify the identity type of a person in real time from the hospital surveillance video. The technical content involved in step S1 is described in detail below.

[0027] Regarding target detection: Use the YOLOv5 algorithm to detect people in the surveillance image;

[0028]

[0029] Among them, P(Object) is the probability of whether the object exists; IOU is the intersection-over-union ratio of the predicted bounding box and the true bounding box.

[0030] Regarding face feature vector recognition: ResNet-50 is used to extract face feature vectors. ResNet-50 is a deep residual network that is widely used in the field of image recognition. It solves the gradient vanishing problem in deep network training through residual learning. Given the formula:

[0031] f(x)=σ(W·x+b)

[0032] Among them, f(x) is the feature vector, which represents the abstract features of the face image after network processing, x is the input face image data; W and b are the network weights and biases, which are learned and adjusted during the training process; σ is the activation function, usually a nonlinear function such as ReLU (Rectified Linear Unit), which is used to introduce nonlinear characteristics to help the network learn more complex patterns.

[0033] Regarding clothing color and texture feature recognition: In actual application scenarios, such as identifying doctors, nurses, patients, and visitors in a hospital environment, relying solely on face recognition may have limitations, especially in occluded or long-distance scenarios. In this case, clothing recognition can serve as a supplement to solve the recognition problem in occluded and long-distance scenarios.

[0034] Color features: Color features can be extracted by performing color histogram statistics on the image. For example, doctors usually wear white or light-colored work clothes, nurses may wear nurse uniforms of specific colors, and the colors of patients and visitors' clothing are more diverse. Perform histogram statistics on the RGB channels of the image to obtain color distribution features as auxiliary information for identity recognition.

[0035] Texture features: Texture feature extraction methods such as LBP (Local Binary Pattern) can be used. LBP generates binary codes by comparing the grayscale values ​​of the central pixel and the neighboring pixels, reflecting the local texture features of the image. Medical staff's work clothes usually have specific textures (such as doctors' white coats are relatively flat, nurses' coats may have stripes, etc.), while the textures of patients and visitors' clothing are more diverse.

[0036] During the training phase: collect facial images and full-body images (including clothing) of doctors, nurses, patients, and visitors. Use ResNet-50 to train facial images, extract facial feature vectors, and mark the corresponding identity categories. Extract color and texture features from full-body images, and also mark the corresponding identity categories.

[0037] In the recognition stage: obtain hospital surveillance video images, first extract facial feature vectors through ResNet-50, compare with the existing feature vector library, and make a preliminary identity judgment. If the facial image is blocked or unclear, use the color and texture features of the clothing to assist in judgment. For example, if an image is captured in a hospital corridor, the face is partially blurred, but the clothing is white and has the typical texture features of a doctor's uniform, it can be inferred that the person is a doctor. Further, machine learning algorithms (such as support vector machines, deep learning classifiers, etc.) can be used to fuse and classify facial feature vectors and clothing features to improve the accuracy and robustness of recognition.

[0038] It can be understood that step S1 dynamically identifies the target identity through multimodal data (face and clothing), which can more accurately identify people of different identities in complex scenarios such as hospitals and ensure the management order and safety of the hospital.

[0039] Step S2 is to track all personnel in the monitoring area in real time, generate trajectories and update the position status. The technical content involved in step S2 is described in detail below.

[0040] About trajectory initialization: Assign a unique trajectory ID to each newly detected target (person) and initialize its trajectory.

[0041] About trajectory update: Use Kalman filtering to predict the target position, and combine the Hungarian algorithm to match the detection box and historical trajectory:

[0042]

[0043] Among them, A and B are the state transfer matrix and control matrix; is the predicted position;

[0044] Use IOU and appearance feature similarity for matching:

[0045] C=λ·(1-IOU)+(1-λ)·(1-CosSim)

[0046] Among them, CosSim is cosine similarity.

[0047] About trajectory storage: Save the spatial position and timestamp of the target to form activity trajectory data.

[0048] T={(x 1 ,y 1 ,t 1 ),(x 2 ,y 2 ,t 2 ),...}.

[0049] It is understandable that step S2 combines Kalman filtering and Hungarian algorithm to match and update the trajectories of multiple targets. This method is suitable for dynamic and complex hospital environments and has high real-time performance.

[0050] Step S3 is to divide the authority of the hospital area. The technical content involved in step S3 is described in detail below.

[0051] Regarding area division: the monitoring area is divided into multiple sub-areas R = {R1, R2, ..., Rn} according to the hospital map.

[0052] Regarding permission matching: Each target is assigned a permission sub-area according to the identity type (doctor, nurse, patient, visitor). For example, P target = {R1, R3, R5}. When the target enters outside the permission area, an abnormal alarm is triggered.

[0053] It can be understood that step S3 sets different authority areas according to the different identities of doctors, nurses, patients and visitors to ensure that the activities of different personnel in the hospital comply with regulations.

[0054] In a possible implementation, S4, when the activity trajectory data of a person enters outside the authorized area, an abnormal alarm is triggered; including: grading the abnormal behavior according to the stay time and number of stays of the activity trajectory data outside the authorized area, and executing different alarm strategies according to the grade.

[0055] Furthermore, according to the residence time and number of stays of the activity trajectory data outside the permission area, the abnormal behavior is graded, and different alarm strategies are executed according to the grade; including: when the residence time of the activity trajectory data outside the permission area is less than the time threshold and the number of stays is less than the frequency threshold, it is a low-risk behavior; when the residence time of the activity trajectory data outside the permission area is greater than or equal to the time threshold and the number of stays is less than the frequency threshold, it is a medium-risk behavior; when the residence time of the activity trajectory data outside the permission area is greater than or equal to the time threshold and the number of stays is greater than or equal to the frequency threshold, it is a high-risk behavior.

[0056] Step S4 is to generate a multi-level alarm for unauthorized behavior. The technical content involved in step S4 is described in detail below.

[0057] About abnormal alarm: When the target enters the restricted area, the abnormal alarm is triggered;

[0058]

[0059] About graded alarms: abnormal behaviors are graded according to the duration and number of times the activity trajectory data stays outside the authorized area. Low risk means that short-term crossing the boundary may be a mistaken entry, medium risk means continuous stay in the restricted area, and high risk means multiple attempts to enter the critical area. Specifically, you can set: low-risk behavior, the duration of activity trajectory data staying outside the authorized area is less than the time threshold and the number of stays is less than the frequency threshold; medium-risk behavior, the duration of activity trajectory data staying outside the authorized area is greater than or equal to the time threshold and the number of stays is less than the frequency threshold; high-risk behavior, the duration of activity trajectory data staying outside the authorized area is greater than or equal to the time threshold and the number of stays is greater than or equal to the frequency threshold.

[0060] Different alarm strategies: Different alarm strategies can be implemented according to risk classification, such as SMS notification of security personnel, linkage of sound and light alarms, and sending real-time images to the control center.

[0061] Regarding early warning data storage: the target trajectory data can also be saved to the database for inspection.

[0062] D={T 1 ,T 2 ,...,T n}

[0063] Ti is the target trajectory. The trajectory data of all monitored targets are saved in time series and can be used for post-analysis and responsibility tracing.

[0064] It is understandable that step S4 associates the trajectory of the personnel with the permission area. Applying this association of "personnel trajectory + permission area" to hospital scenarios can promptly alarm when the trajectory of the personnel is abnormal. Abnormal behaviors are divided into low risk (mistaken entry), medium risk (continuous stay), and high risk (multiple attempts to enter). Compared with the traditional simple cross-border alarm system, the classification is more detailed and the response is more reasonable. According to the risk level, measures such as SMS notifications, sound and light alarms, and real-time screen push can be triggered respectively, which improves the flexibility and security of the system. In addition, according to the different security levels of key areas of the hospital (such as ICU and pharmacy), the abnormal behavior classification can be adaptively adjusted to provide higher customized security monitoring capabilities for hospital scenarios.

[0065] The method for dynamic monitoring and trajectory tracking of hospital personnel provided in this application, first, based on the special application scenario of the hospital, proposes an identity type detection method that combines facial feature vectors and clothing color texture features to improve the accuracy of identity type detection of hospital personnel; second, a method that combines personnel activity trajectory data and permission areas is proposed to achieve real-time detection of illegal intrusions; third, a graded alarm is performed by combining movement trajectory data and permission areas to effectively improve security. In addition, the method for dynamic monitoring and trajectory tracking of hospital personnel provided in this application achieves millisecond-level response through lightweight target detection and trajectory update algorithms, and has high real-time performance; supports multi-target monitoring and regional management in complex scenarios, and has strong adaptability; can be seamlessly integrated with the hospital's access control, attendance, visitor management and other systems, and has high scalability; has automatic warning and self-learning capabilities, can be continuously optimized through incremental learning, significantly reduces the need for manual intervention, and has high intelligence.

[0066] See also Figure 2 As shown, Figure 2 The structure diagram of the hospital personnel dynamic monitoring and trajectory tracking system provided by the embodiment of the present application. The system includes: a target detection and identity recognition module, which is used to obtain hospital monitoring videos, and detect and identify the identity types of personnel in real time from the monitoring videos in combination with facial feature vectors and clothing color and texture features. The identity types include: doctors, nurses, patients and visitors; a multi-target trajectory tracking module, which is used to assign a unique trajectory ID to each detected person, predict the position of the person through Kalman filtering, match the detection box and the historical trajectory through the Hungarian algorithm, match based on IOU and appearance feature similarity, save the spatial position and timestamp of the person, form activity trajectory data and save it to the database; a regional authorization management module, which is used to divide the monitoring area into multiple sub-areas according to the hospital map, and assign sub-areas with permissions according to the identity type of each person as permission areas; an abnormal alarm and data storage module, which is used to trigger an abnormal alarm when the activity trajectory data of a person enters outside the permission area.

[0067] In a possible implementation, the target detection and identity recognition module includes: a data acquisition module, which is used to collect facial images and full-body images of current doctors, nurses, patients and visitors in the hospital, extract facial feature vectors from facial images through ResNet-50 and mark corresponding identity categories, store them in a facial feature vector library, extract clothing color and texture features of full-body images through color histogram statistics and LBP algorithm and mark corresponding identity categories, and store them in different clothing feature libraries according to identity categories; a face recognition module, which is used to detect personnel areas in the surveillance video screen through the YOLOv5 algorithm when acquiring surveillance video, extract facial feature vectors in the personnel area through ResNet-50, query consistent facial feature vectors in an existing facial feature vector library, and determine the identity type of the personnel; a clothing color and texture feature recognition module, which is used to extract clothing color and texture features in the personnel area through color histogram statistics and LBP algorithm when no consistent facial feature vectors are found in the existing facial feature vector library, and compare them with clothing color and texture features in each clothing feature library to determine the identity type of the personnel.

[0068] Furthermore, the target detection and identity recognition module also includes: an identification verification module, which is used to obtain the identity type of the person when a consistent facial feature vector is queried in the existing facial feature vector library; extract the clothing color texture features in the person area through color histogram statistics and LBP algorithm, and compare them with the clothing color texture features of each clothing feature library to verify the identity type of the person.

[0069] In a possible implementation, the abnormal alarm and data storage module is specifically used to classify abnormal behaviors according to the stay time and number of stays of the activity trajectory data outside the permission area, and execute different alarm strategies according to the classification.

[0070] Furthermore, the abnormal alarm and data storage module is specifically used for: when the stay time of the activity trajectory data outside the permission area is less than the time threshold and the number of stays is less than the frequency threshold, it is a low-risk behavior; when the stay time of the activity trajectory data outside the permission area is greater than or equal to the time threshold and the number of stays is less than the frequency threshold, it is a medium-risk behavior; when the stay time of the activity trajectory data outside the permission area is greater than or equal to the time threshold and the number of stays is greater than or equal to the frequency threshold, it is a high-risk behavior.

[0071] Specifically, the target detection and identity recognition module: advanced deep learning algorithms (such as YOLO, FasterR-CNN, etc.) can be used for target detection, accurately identifying people in the hospital from the video stream, combining facial feature vector recognition (such as ResNet-50 and other networks to extract facial features) and clothing color and texture feature recognition (based on color and texture features) to accurately identify the identity of the person and record the identity type (doctor, nurse, visitor, patient).

[0072] Multi-target trajectory tracking module: Combines Kalman filtering and Hungarian algorithm to track the target trajectory of detected personnel, records the movement routes of personnel, and dynamically updates the generated trajectory data.

[0073] Regional authorization management module: Different permission areas are set according to the identity type of the personnel, for example, visitors can only move in specific areas, patients can only enter their own ward areas, etc. The permission areas can be stored in the database and associated with the identity type of the personnel. For patients, facial and clothing feature information can be collected when they are admitted to the hospital, and associated with ward information, treatment area information, etc., to set their permission areas; for example, internal medicine patients can only move in internal medicine wards and related examination areas. For visitors, identity information can be obtained through the visitor registration system, and temporary passes with identity identifiers (which can be electronic passes, such as QR codes or Bluetooth beacons, etc.) are issued. The activity area is set according to the information of the patient to be visited by the visitor, such as only being able to move on the floor where the patient's ward is located. In addition, the permission area can be dynamically adjusted according to the patient's treatment progress and doctor's orders. For example, the range of activities of surgical patients before and after surgery may be different, and the system automatically updates the permissions according to the doctor's orders. For visitors, when the visiting time ends or the patient is transferred to a ward, the visitor's permissions are automatically updated to prevent them from walking around the hospital at will.

[0074] Abnormal alarm and data storage module: Real-time analysis of personnel trajectory and behavior, for example, analyzing the personnel's residence time, movement speed, movement route, etc. in a certain area; judging whether the personnel behavior is abnormal in combination with the permission area, for example, visitors entering the permission area, patients wandering in areas other than their own wards for a long time, etc. are all considered abnormal behavior; graded warnings are carried out according to the severity of the abnormal behavior. For low-risk behaviors (such as patients staying briefly in the area near the ward), a prompt sound can be issued. For medium-risk behaviors (such as visitors entering restricted floors), text messages or APP push notifications can be sent to security personnel or relevant medical staff. For high-risk behaviors (such as visitors entering the operating room area and staying there), the sound and light alarm device can be activated and all hospital security personnel and relevant management departments can be notified at the same time.

[0075] It should be noted that the various modules can communicate and work together through message queues or event-driven mechanisms. For example, when the multi-target trajectory tracking module detects a new person entering the monitoring area, it immediately notifies the multi-target trajectory tracking module to start recording the trajectory, and at the same time notifies the regional authorization management module to extract the permission area of ​​the identity type, and pass the permission area of ​​the personnel type to the abnormal alarm and data storage module for identity type permission judgment. When abnormal behavior is found, an alarm of the corresponding level is issued according to the situation.

[0076] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for dynamic monitoring and trajectory tracking of hospital personnel, characterized in that: include: S1. Obtain hospital surveillance video, and detect and identify the identity types of people in real time from the surveillance video by combining facial feature vectors and clothing color and texture features. The identity types include: doctors, nurses, patients, and visitors; S2. Assign a unique trajectory ID to each detected person, predict the person's location through Kalman filtering, match the detection box and historical trajectory through the Hungarian algorithm, match based on IOU and appearance feature similarity, save the person's spatial location and timestamp, form activity trajectory data and save it to the database; S3, dividing the monitoring area into multiple sub-areas according to the hospital map, and assigning sub-areas with authority according to the identity type of each person as the authority area; S4. When the activity trajectory data of a person enters outside the authorized area, an abnormal alarm is triggered.

2. A method for dynamic monitoring and trajectory tracking of hospital personnel according to claim 1, characterized in that: S1. Obtain hospital surveillance video, and detect and identify the identity types of people in real time from the surveillance video by combining facial feature vectors and clothing color and texture features. The identity types include: doctors, nurses, patients and visitors; including: Collect facial images and full-body images of current doctors, nurses, patients, and visitors in the hospital, extract facial feature vectors from the facial images through ResNet-50, mark the corresponding identity categories, and store them in the facial feature vector library; extract clothing color and texture features of full-body images through color histogram statistics and LBP algorithm, mark the corresponding identity categories, and store them in different clothing feature libraries according to the identity categories; When obtaining surveillance video, the YOLOv5 algorithm is used to detect the personnel area in the surveillance video screen, and the facial feature vector in the personnel area is extracted through ResNet-50. The consistent facial feature vector is searched in the existing facial feature vector library to determine the identity type of the person; When no consistent facial feature vector can be found in the existing facial feature vector library, the clothing color and texture features in the personnel area are extracted through color histogram statistics and LBP algorithm, and compared with the clothing color and texture features in each clothing feature library to determine the identity type of the person.

3. A method for dynamic monitoring and trajectory tracking of hospital personnel according to claim 2, characterized in that: S1. Obtain hospital surveillance video, and detect and identify the identity types of people in real time from the surveillance video by combining facial feature vectors and clothing color and texture features. The identity types include: doctors, nurses, patients and visitors; and also include: When a consistent facial feature vector is found in the existing facial feature vector library, the identity type of the person is obtained; The clothing color and texture features in the personnel area are extracted through color histogram statistics and LBP algorithm, and compared with the clothing color and texture features in each clothing feature library to verify the identity type of the person.

4. A method for dynamic monitoring and trajectory tracking of hospital personnel according to claim 1, characterized in that: S4. When the activity trajectory data of a person enters outside the authorized area, an abnormal alarm is triggered; including: grading the abnormal behavior according to the stay time and number of stays of the activity trajectory data outside the authorized area, and executing different alarm strategies according to the grading.

5. A method for dynamic monitoring and trajectory tracking of hospital personnel according to claim 4, characterized in that: According to the time and number of times the activity trajectory data stays outside the authorized area, the abnormal behavior is classified and different alarm strategies are implemented according to the classification; including: When the activity trajectory data stays outside the permission area for less than the time threshold and the number of stays is less than the frequency threshold, it is a low-risk behavior; When the activity trajectory data stays outside the permission area for a time greater than or equal to the time threshold and the number of stays is less than the frequency threshold, it is considered a medium-risk behavior; When the activity trajectory data stays outside the permission area for a time greater than or equal to the time threshold and the number of stays is greater than or equal to the frequency threshold, it is a high-risk behavior.

6. A hospital personnel dynamic monitoring and trajectory tracking system, characterized in that: include: The target detection and identity recognition module is used to obtain hospital surveillance videos, and detect and recognize the identity types of people in real time from the surveillance videos by combining facial feature vectors and clothing color and texture features. The identity types include: doctors, nurses, patients and visitors; The multi-target trajectory tracking module is used to assign a unique trajectory ID to each detected person, predict the person's position through Kalman filtering, match the detection box and historical trajectory through the Hungarian algorithm, match based on IOU and appearance feature similarity, save the spatial position and timestamp of the person, form activity trajectory data and save it to the database; The regional authorization management module is used to divide the monitoring area into multiple sub-areas according to the hospital map, and assign the sub-areas with authority according to the identity type of each person as the authority area; The abnormal alarm and data storage module is used to trigger an abnormal alarm when the activity trajectory data of a person enters outside the authorized area.

7. A hospital personnel dynamic monitoring and trajectory tracking system according to claim 6, characterized in that: The target detection and identity recognition module includes: The data collection module is used to collect the face images and full-body images of the current doctors, nurses, patients and visitors in the hospital, extract the face feature vectors of the face images through ResNet-50 and mark the corresponding identity categories, and store them in the face feature vector library; extract the clothing color and texture features of the full-body images through color histogram statistics and LBP algorithm and mark the corresponding identity categories, and store them in different clothing feature libraries according to the identity categories; The face recognition module is used to obtain surveillance video, detect the personnel area in the surveillance video screen through the YOLOv5 algorithm, extract the face feature vector in the personnel area through ResNet-50, query the consistent face feature vector in the existing face feature vector library, and determine the identity type of the person; The clothing color and texture feature recognition module is used to extract the clothing color and texture features in the personnel area through color histogram statistics and LBP algorithm when no consistent facial feature vector can be found in the existing facial feature vector library, and compare them with the clothing color and texture features in each clothing feature library to determine the identity type of the person.

8. A hospital personnel dynamic monitoring and trajectory tracking system according to claim 7, characterized in that: The target detection and identity recognition module further includes: The identification and verification module is used to obtain the identity type of a person when a consistent facial feature vector is found in the existing facial feature vector library; the clothing color and texture features in the person area are extracted through color histogram statistics and LBP algorithm, and compared with the clothing color and texture features in each clothing feature library to verify the identity type of the person.

9. A hospital personnel dynamic monitoring and trajectory tracking system according to claim 6, characterized in that: The abnormal alarm and data storage module is specifically used to classify abnormal behaviors according to the residence time and number of residences of the activity trajectory data outside the permission area, and execute different alarm strategies according to the classification.

10. A hospital personnel dynamic monitoring and trajectory tracking system according to claim 9, characterized in that: The abnormal alarm and data storage module is also specifically used for: When the activity trajectory data stays outside the permission area for less than the time threshold and the number of stays is less than the frequency threshold, it is a low-risk behavior; When the activity trajectory data stays outside the permission area for a time greater than or equal to the time threshold and the number of stays is less than the frequency threshold, it is considered a medium-risk behavior; When the activity trajectory data stays outside the permission area for a time greater than or equal to the time threshold and the number of stays is greater than or equal to the frequency threshold, it is a high-risk behavior.

Images