A smart hospital information management method and system

Abstract

The application provides a kind of intelligent hospital information management method and system, it is related to hospital information management technical field.The method comprises: obtaining monitoring video frame and finding target object being interactive;The number of interactions of each target object is counted, and the to-be-determined audit object is filtered based on this;Verification target object is filtered and its verification position information is obtained to verify the to-be-determined audit object, obtain verification result, and obtain audit prompt information.According to the application, the target object with interactive behavior can be found in the monitoring video of the outpatient area, and the number of interactions can filter possible number traffickers and medical agents, etc., so that number traffickers and medical agents can be found in advance, the timeliness of supervision is improved, the human consumption is reduced, and the possibility of medical agents and number traffickers infringing on the rights and interests of patients is also reduced.Furthermore, the behavior and moving route of the target object can be verified, and the to-be-determined audit object is verified, so as to improve the accuracy of identifying number traffickers and medical agents, and reduce the probability of misidentification.

Description

Technical Field

[0001] This invention relates to the field of hospital information management technology, and in particular to a smart hospital information management method and system. Background Technology

[0002] In related technologies, hospitals can set up management systems to collect and manage doctor and patient information and perform various processing, such as monitoring patients' medication use and matching patients with suitable doctors. However, many hospitals face shortages of medical resources, which may lead to illegal and irregular activities by medical touts, scalpers, and other individuals. It is difficult to supervise and prevent medical touts and scalpers, which usually require manual identification and removal. However, manual identification is not only labor-intensive but also prone to problems such as untimely supervision, which may result in patients' rights being infringed upon by medical touts and scalpers. Summary of the Invention

[0003] This invention provides a smart hospital information management method and system that can solve the technical problem that related technologies are unable to timely monitor and prevent medical touts and scalpers.

[0004] According to a first aspect of the present invention, a smart hospital information management method is provided, comprising:

[0005] Acquire surveillance video of the outpatient area within the current monitoring period, and parse the surveillance video to obtain multiple surveillance video frames;

[0006] The image processing model is used to process each monitoring video frame to determine whether there is a target object in each monitoring video frame that is performing interactive behavior.

[0007] If it exists, record the identification information of the target object that is currently interacting, and count the target objects corresponding to multiple monitoring video frames to determine the number of people interacting with each target object in the current monitoring period.

[0008] The target objects are filtered based on the number of interactions corresponding to each target object to obtain the objects to be reviewed;

[0009] Among all target objects, filter out the verification target objects that have interacted with the pending review objects during the current monitoring period;

[0010] The image processing model is used to determine the verification location information of the target object in multiple monitoring video frames within the monitoring period;

[0011] Based on the verification location information, the target to be reviewed is verified to obtain the verification result;

[0012] Based on the verification results, an audit notification message will be received.

[0013] According to the present invention, each monitoring video frame is processed using an image processing model to determine whether a target object is performing interactive behavior in each monitoring video frame, including:

[0014] The image processing model is used to obtain the location information of each object in each monitoring video frame. The location information of the target object includes the location information of the smallest rectangle that selects the object.

[0015] Obtain the centroid coordinates of the region where each object is located;

[0016] Based on the centroid coordinates of the region, at the... From the first to the i-th monitoring video frame, determine whether there exists a group of undetermined targets whose centroid distance is always less than a first preset distance threshold. This undetermined target group includes two undetermined target objects, and the centroid distance between the centroid coordinates of the regions where the two undetermined target objects are located is less than a first preset distance threshold. The distance is always less than the first preset distance threshold from the i-th monitoring video frame to the i-th monitoring video frame, where i is a positive integer. This is the first preset quantity;

[0017] If it exists, the image processing model is used to process the regions where the two target objects in the target group are located to obtain the facial orientation information of the two target objects, and the facial region is determined in the region where the target objects are located.

[0018] If in the first If, in the i-th to i-th monitoring video frames, the proportion of monitoring video frames in which the facial orientation information of the two target objects points to the facial area of ​​the other exceeds the first preset proportion, then in the i-th monitoring video frame, it is determined that there are target objects in the process of interacting, and the target objects in the process of interacting are the two target objects.

[0019] According to the present invention, target objects are filtered based on the number of interactions corresponding to each target object to obtain objects to be reviewed, including:

[0020] If the number of interactions corresponding to the j-th target object is greater than or equal to the second preset number, then the j-th target object is determined as an object to be reviewed.

[0021] According to the present invention, the verification of the target audit object is performed based on the verification location information to obtain the verification result, including:

[0022] Based on the verification location information of the target object in the monitoring video frame, determine the geographical location information of the target object within the outpatient area;

[0023] By using an image processing model, the region where the verification target object is located is processed to obtain the facial orientation information of the verification target object;

[0024] Based on the facial orientation information, determine the facial orientation angle of the verification target object within the outpatient area;

[0025] Based on the geographical location information, facial orientation angle, and the sequence number of the monitoring video frame, determine the state feature information of the k-th verification target object when the i-th monitoring video frame is captured;

[0026] The structure of the kth subgraph is determined based on the state feature information of the kth verification target object when capturing each monitoring video frame;

[0027] The first graph neural network model is used to process each subgraph structure to obtain the subgraph feature information of each subgraph structure;

[0028] The graph structure is obtained based on the feature information of the subgraphs corresponding to multiple verification target objects;

[0029] Based on the second graph neural network model, the graph structure is processed to obtain the feature information of each verification target object;

[0030] Based on the feature information of each verification target object, relevant target objects are selected from multiple verification target objects;

[0031] The verification results are obtained based on the target object and related target objects.

[0032] According to the present invention, the structure of the kth subgraph is determined based on the state feature information of the kth verification target object when capturing each monitoring video frame, including:

[0033] The state feature information of the kth verification target object when the i-th monitoring video frame is captured is used as the first input feature information of the i-th node in the k-th subgraph structure;

[0034] According to the formula

[0035]

[0036]

[0037] Obtain the adjacency matrix of the k-th subgraph structure, where, This represents the data in the i-th row and i+1-th column of the adjacency matrix of the k-th subgraph structure. This represents the data in the (i+1)th row and i-th column of the adjacency matrix of the k-th subgraph structure. This represents the data in the i-th row and i+s-th column of the adjacency matrix of the k-th subgraph structure. Let s be the data in the (i+s)th row and i-th column of the adjacency matrix of the k-th subgraph structure, where s≠1. Let be the distance between the geographical location information of the k-th verification target object when the i-th monitoring video frame is captured and the geographical location information when the (i+1)-th monitoring video frame is captured. This is the average distance between the geographical locations of each verification target when capturing adjacent monitoring video frames;

[0038] The degree matrix of the k-th subgraph structure is determined based on the adjacency matrix of the k-th subgraph structure.

[0039] The structure of the k-th subgraph is determined based on the adjacency matrix, degree matrix, and first input feature information of each node.

[0040] According to the present invention, each subgraph structure is processed according to a first graph neural network model to obtain subgraph feature information of each subgraph structure, including:

[0041] The first graph neural network model is used to process the k-th subgraph structure to obtain the node feature information of each node in the k-th subgraph structure;

[0042] The node feature information of each node is processed by the attention mechanism of the first graph neural network model to obtain the subgraph feature information of the k-th subgraph structure.

[0043] According to the present invention, a graph structure is obtained based on the subgraph feature information corresponding to multiple verification target objects, including:

[0044] The subgraph feature information of the kth subgraph structure is used as the second input feature information of the kth node in the graph structure;

[0045] According to the formula

[0046]

[0047] Obtain the adjacency matrix of the graph structure, where, The data in the k-th row and s-th column of the adjacency matrix of the graph structure. The data in the s-th row and k-th column of the adjacency matrix of the graph structure. Let be the facial orientation angle of the k-th verification target when the i-th monitoring video frame is captured. Let be the facial orientation angle of the s-th verification target object when the ith monitoring video frame is captured. To verify the geographical location information of the s-th target object when the ith surveillance video frame was captured, Let n be the geographical location information of the k-th verification target object when the i-th monitoring video frame is captured, and n be the number of monitoring video frames in the current monitoring period. To capture multiple surveillance video frames, the average distance between the k-th verification target object and the s-th verification target object, where i ≤ n, and i and n are both positive integers;

[0048] Based on the adjacency matrix of the graph structure, obtain the degree matrix of the graph structure;

[0049] The graph structure is obtained by using the second input feature information, adjacency matrix, and degree matrix of each node in the graph structure.

[0050] According to the present invention, obtaining verification results based on the verification target object and related target objects includes:

[0051] Related target objects that are mutually related are identified as related target object groups;

[0052] Determine the number of relevant target object groups;

[0053] The verification results are obtained based on the number of target objects, the number of related target object groups, and the number of related target objects.

[0054] According to a second aspect of the present invention, a smart hospital information management system is provided, comprising:

[0055] The acquisition module is used to acquire surveillance videos within the outpatient area during the current monitoring period, and to parse the surveillance videos to obtain multiple surveillance video frames.

[0056] The target object module is used to process each monitoring video frame through an image processing model to determine whether there is a target object in each monitoring video frame that is performing interactive behavior.

[0057] The interaction number module is used to record the identification information of the target object that is currently interacting, if it exists, and to count the target objects corresponding to multiple monitoring video frames to determine the number of people interacting with each target object in the current monitoring period.

[0058] The "Pending Review Objects" module is used to filter target objects based on the number of interactions corresponding to each target object, and obtain objects pending review.

[0059] The Target Object Verification Module is used to filter out target objects from all target objects that have interacted with the object to be reviewed during the current monitoring period.

[0060] The location information verification module is used to determine the verification location information of the target object in multiple monitoring video frames within the monitoring period through an image processing model.

[0061] The verification module is used to verify the target to be reviewed based on the verification location information and obtain the verification result;

[0062] The audit notification module is used to obtain audit notification information based on the verification results.

[0063] By adopting the above technical solution, the present invention can achieve the following technical effects:

[0064] According to this invention, target objects exhibiting interactive behavior can be identified through surveillance video in the outpatient area. The number of interactions can be used to screen for potential scalpers and medical touts, allowing for the early detection of scalpers and medical touts, improving the timeliness of supervision, reducing manpower consumption, and decreasing the likelihood of scalpers and medical touts infringing on patients' rights. Furthermore, the behavior and movement routes of multiple verification target objects interacting with the target can be used to verify the target, thereby improving the accuracy of identifying scalpers and medical touts and reducing the probability of false identification. When screening target objects exhibiting interactive behavior, the nearest target objects can be selected from multiple objects in the surveillance video frame, and their facial orientation information can be used for verification, improving the accuracy of finding target objects exhibiting ongoing interactive behavior and providing an accurate data foundation for screening scalpers and medical touts. When determining the subgraph structure, the trajectory information of each verification target object can be described using the subgraph structure. Adding the sequence number of the monitoring video frame to the state feature information of each node increases the temporal dimension features, improving the feature recognition of the trajectory information. Furthermore, the relationship between nodes can be determined based on the distance between the geographical locations of each video frame, thus accurately describing the features of the trajectory information and improving the accuracy of trajectory feature-based verification. When determining whether verification target objects are related, the angular correlation between the two verification target objects can be determined by the angle between their facial orientations and whether one verification target object is facing another. The distance correlation can be determined by their average distance. Therefore, the correlation between their trajectories can be determined based on the angular and distance correlations, thus determining the adjacency matrix. The feature information of each verification target object can be determined using a second-graph neural network model, improving the distinguishability between the trajectories of related and unrelated verification target objects, thereby improving the accuracy of determining whether two verification target objects are related. Attached Figure Description

[0065] Figure 1 An exemplary flowchart of a smart hospital information management method according to an embodiment of the present invention is shown.

[0066] Figure 2 An exemplary schematic diagram of verifying a target for review according to an embodiment of the present invention is shown.

[0067] Figure 3 A block diagram of a smart hospital information management system according to an embodiment of the present invention is shown as an example. Detailed Implementation

[0068] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0069] Figure 1 An exemplary flowchart of a smart hospital information management method according to an embodiment of the present invention is shown, the method comprising:

[0070] Step S1: Obtain surveillance video of the outpatient area within the current monitoring period, and parse the surveillance video to obtain multiple surveillance video frames;

[0071] Step S2: Process each monitoring video frame using an image processing model to determine whether there is a target object engaging in interactive behavior in each monitoring video frame.

[0072] Step S3: If it exists, record the identification information of the target object that is currently interacting, and count the target objects corresponding to multiple monitoring video frames to determine the number of people interacting with each target object in the current monitoring period.

[0073] Step S4: Filter the target objects according to the number of interactions corresponding to each target object to obtain the objects to be reviewed;

[0074] Step S5: Among all target objects, filter the verification target objects that have interacted with the object to be reviewed during the current monitoring period;

[0075] Step S6: Using an image processing model, determine the verification location information of the target object in multiple monitoring video frames within the monitoring period;

[0076] Step S7: Verify the target to be audited based on the verification location information and obtain the verification result;

[0077] Step S8: Obtain audit prompt information based on the verification results.

[0078] According to embodiments of the present invention, the smart hospital information management method can identify target objects exhibiting interactive behavior through surveillance videos of the outpatient area, and filter for potential scalpers and medical touts based on the number of interactions. This allows for the early detection of scalpers and medical touts, improving the timeliness of supervision, reducing manpower consumption, and decreasing the likelihood of scalpers and medical touts infringing on patients' rights. Furthermore, the method can verify the target object based on the behavior and movement routes of multiple verification target objects that have interacted with the target object, thereby improving the accuracy of identifying scalpers and medical touts and reducing the probability of false identification.

[0079] According to one embodiment of the present invention, in step S1, each monitoring cycle can be 15 minutes, 30 minutes, etc., and the present invention does not limit this. At least one camera can be installed in the outpatient area to acquire monitoring video within the current monitoring cycle. At the end of the monitoring cycle, the monitoring video captured within the monitoring cycle can be analyzed. All video frames in the monitoring video can be used as the monitoring video frame, or multiple video frames can be sampled, for example, sampling can be performed every 1 second to obtain the monitoring video frame.

[0080] According to an embodiment of the present invention, in step S2, the image processing model can be a convolutional neural network model, such as the YOLO model, which can process each monitoring video frame and determine whether there is a target object in each monitoring video frame that is interacting. For example, if two target objects are talking in a certain monitoring video frame, it can be determined that there is a target object in the monitoring video frame that is interacting.

[0081] According to an embodiment of the present invention, in step S2, each monitoring video frame is processed by an image processing model to determine whether there is a target object performing interactive behavior in each monitoring video frame. This includes: obtaining the location information of the region where each object is located in each monitoring video frame by using an image processing model, wherein the location information of the region where the target object is located includes the location information of the region where the smallest rectangle used to select the object is located; obtaining the centroid coordinates of the region where each object is located; and, based on the centroid coordinates, determining whether there is a target object performing interactive behavior in each monitoring video frame. From the first to the i-th monitoring video frame, determine whether there exists a group of undetermined targets whose centroid distance is always less than a first preset distance threshold. This undetermined target group includes two undetermined target objects, and the centroid distance between the centroid coordinates of the regions where the two undetermined target objects are located is less than a first preset distance threshold. The distance is always less than the first preset distance threshold from the i-th monitoring video frame to the i-th monitoring video frame, where i is a positive integer. The first preset quantity; if it exists, then through the image processing model, the regions where the two target objects in the target group are located are processed to obtain the facial orientation information of the two target objects, and the facial region is determined in the region where the target objects are located; if in the first If, in the i-th to i-th monitoring video frames, the proportion of monitoring video frames in which the facial orientation information of the two target objects points to the facial area of ​​the other exceeds the first preset proportion, then in the i-th monitoring video frame, it is determined that there are target objects in the process of interacting, and the target objects in the process of interacting are the two target objects.

[0082] According to one embodiment of the present invention, the image processing model can select the region where each object is located in a monitoring video frame and obtain the location information of the region where each object is located. For example, the region where each object is located can be selected using a minimum bounding box, and the location of the minimum bounding box can be recorded, for example, the coordinates of the four vertices of the minimum bounding box can be recorded.

[0083] According to one embodiment of the present invention, the centroid coordinates of the region where the object is located are the average of the coordinates of the four vertices. That is, the x-coordinate of the centroid coordinates is the average of the x-coordinates of the four vertices, and the y-coordinate of the centroid coordinates is the average of the y-coordinates of the four vertices.

[0084] According to one embodiment of the present invention, if two objects are engaged in interactive activities such as conversation, the distance between the two objects should be relatively close during the conversation. Therefore, objects that are relatively close in a series of consecutive video frames can be identified as potential target objects, and two potential target objects form a group of potential targets. In the example, objects that are relatively close in a series of consecutive video frames can be identified as potential target objects. In the first to the i-th monitoring video frame, two objects whose centroid distance between their region centroid coordinates is always less than a first preset distance threshold (e.g., a distance of 20 pixels) can be considered as two objects in the i-th monitoring video frame. If the distance between the i-th monitoring video frame and the i-th monitoring video frame is always close, then the two objects can be regarded as potential target objects, and the two potential target objects form a potential target group.

[0085] According to one embodiment of the present invention, the distance between two target objects cannot be used to directly determine whether they are interacting. For example, if two target objects are queuing at a registration window, their distance remains relatively close for a certain period of time, but they are not interacting. Therefore, it is necessary to further determine whether the faces of the two target objects are facing each other. This can be achieved by processing the region where the target objects are located using an image processing model to obtain the facial orientation information and the region where the faces are located. For example, a unit vector parallel to the direction of facial orientation can be obtained as facial orientation information.

[0086] According to one embodiment of the present invention, taking the centroid of the facial region of one of the target objects as the starting point, a straight line is drawn according to the facial orientation information. If this straight line intersects with the facial region of another target object, it can be determined that the facial orientation information of the first target object points to the facial region of the second target object. Similarly, it can be determined whether the facial orientation information of the second target object points to the facial region of the third target object. If the facial orientation information of both target objects points to the facial region of the third target object, the monitoring video frame can be statistically analyzed. Furthermore, the number of monitoring video frames in which the facial orientation information of both target objects points to the facial region of the third target object can be counted, and the relationship between this number and the first target object can be solved. The total number of surveillance video frames up to the i-th surveillance video frame (i.e., If the ratio of the two target objects to the number of potential targets exceeds a first preset ratio (e.g., 0.8), then it is determined that there is a high probability that the two potential target objects are interacting face-to-face, and it can be determined that there are target objects currently interacting, and these target objects are the two potential target objects mentioned above. Furthermore, this method can also be used to determine whether there are other target objects interacting.

[0087] In this way, it is possible to filter out nearby target objects from multiple objects in a surveillance video frame and verify them by using the facial orientation information of the target objects. This improves the accuracy of finding target objects that are interacting with each other and provides an accurate data foundation for screening scalpers and medical touts.

[0088] According to one embodiment of the present invention, in step S3, scalpers and medical touts typically need to talk to multiple target individuals, while normal target individuals usually only talk to a small number of their companions. Therefore, target individuals can be screened based on the number of interactions with each target individual, and those with higher suspicion can be selected for further review. Identification information of target individuals currently interacting can be obtained, and based on this identification information, the number of interactions with each target individual can be counted. For example, the number of people interacting with target individual A is 5.

[0089] According to an embodiment of the present invention, in step S4, target objects are filtered based on the number of interactions corresponding to each target object to obtain objects to be reviewed. This includes: if the number of interactions corresponding to the j-th target object is greater than or equal to a second preset number, then the j-th target object is determined as an object to be reviewed. The second preset number can be determined based on a first preset number. For example, if the first preset number is 119, then the monitoring video frames can analyze the number of interactions of the target objects in the past 120 seconds. If the number of interactions of a target object in the past 120 seconds is greater than or equal to the second preset number (e.g., 4 people), then the target object can be determined as an object to be reviewed. In the example, the second preset number can be equal to (first preset number + 1) / 30, that is, it is expected that a medical tout or scalper will talk to at least one person every 30 seconds. If the number of interactions of a target object is greater than or equal to the second preset number, then the target object is more likely to be a scalper or medical tout and can be set as an object to be reviewed. Normal target objects usually only talk to a small number of people (e.g., 1 or 2), that is, the number of interactions is less than the second preset number.

[0090] According to one embodiment of the present invention, in step S5, within the current monitoring period, there may be multiple target objects in all monitored video frames. Target objects that have interacted with the subject of the pending review can be used as verification target objects, and their behavior is analyzed to verify the subject of the pending review, thereby further determining whether the subject of the pending review is a medical tout or a scalper. For example, although there may be a large number of people interacting with the subject of the pending review, these people are all colleagues. For instance, if the subject of the pending review interacts with multiple colleagues separately, the subject of the pending review may also be a colleague, but due to the large number of interactions, it may be mistakenly judged as a scalper or a medical tout. Conversely, if there are a large number of people interacting with the subject of the pending review, and these people are unrelated, and the subject of the pending review interacts with multiple unrelated people in a short period of time, it can be considered that the subject of the pending review is randomly looking for sales targets, and the subject of the pending review is more likely to be a scalper or a medical tout. To further reduce the possibility of misjudgment, it is possible to verify whether the target objects have any relationship, thereby verifying whether the personnel to be reviewed are suspected of being scalpers or medical touts, and improving the accuracy of identifying scalpers or medical touts.

[0091] According to an embodiment of the present invention, in step S6, the verification position information of the verification target object in multiple monitoring video frames within the detection period can be determined by an image processing model, that is, the position information of the smallest rectangular box for selecting the target object.

[0092] According to one embodiment of the present invention, in step S7, the verification location information of each verification target object can be analyzed, and the movement trajectory of each verification target object within the current monitoring period can be determined. The similarity of the movement trajectories of each verification target object can also be analyzed. If the movement trajectories of two or more verification target objects are highly similar, then the two or more verification target objects are related, and the candidate for review may be misjudged. Conversely, if the movement trajectories of two or more verification target objects are not highly similar, then the two or more verification target objects are not related to each other, and it can be determined that the candidate for review has interacted with multiple unrelated individuals within a short period of time; that is, the candidate for review is highly likely to be a medical tout or a scalper.

[0093] Figure 2 An exemplary schematic diagram of verifying a target for review according to an embodiment of the present invention is shown.

[0094] According to an embodiment of the present invention, in step S7, the target audit object is verified according to the verification location information to obtain the verification result, including: determining the geographical location information of the target audit object in the outpatient area according to the verification location information of the target audit object in the monitoring video frame; processing the area where the target audit object is located through an image processing model to obtain the facial orientation information of the target audit object; determining the facial orientation angle of the target audit object in the outpatient area according to the facial orientation information; determining the state feature information of the kth target audit object when the i-th monitoring video frame is captured according to the geographical location information, the facial orientation angle, and the sequence number of the monitoring video frame; determining the kth subgraph structure according to the state feature information of the kth target audit object when each monitoring video frame is captured; processing each subgraph structure according to a first graph neural network model to obtain the subgraph feature information of each subgraph structure; obtaining a graph structure according to the subgraph feature information corresponding to multiple target audit objects; processing the graph structure according to a second graph neural network model to obtain the feature information of each target audit object; filtering relevant target objects among multiple target audit objects according to the feature information of each target audit object; and obtaining the verification result according to the target audit object and the relevant target objects.

[0095] According to one embodiment of the present invention, the method for obtaining the verification location information of the target object in the monitoring video frame is the same as the method for obtaining location information described above, and will not be repeated here. Furthermore, in some cases, the field of view of a single monitoring camera cannot fully cover the outpatient area. Therefore, multiple monitoring cameras can be used to film the outpatient area. However, due to different shooting angles, the verification location information of the target object in the monitoring video frames captured by each camera is different. Therefore, it is not possible to directly process the verification location information. Therefore, the verification location information of the target object in the monitoring video frame can be converted into geographical location information within the outpatient area. For example, the verification location information can be converted into geographical location information through the camera's calibration parameters. This allows the verification location information in monitoring video frames captured by different cameras to be converted into geographical location information in a unified coordinate system (i.e., an established geographic coordinate system within the outpatient area), facilitating subsequent processing. Similarly, the facial orientation information of the verification target object obtained by the image processing model can be processed to obtain the facial orientation angle of the verification target object within the outpatient area. That is, the angle in the image is converted into an angle in the geographic coordinate system. For example, the facial orientation information can be a unit vector in the monitoring video frame. A straight line is drawn with the centroid of the facial region as the starting point of the unit vector, and any point on the line is taken as a reference point. Furthermore, the coordinates of the centroid of the facial region in the monitoring video frame and the coordinates of the reference point in the monitoring video frame are both converted to the above-mentioned unified coordinate system through calibration parameters. Then, the orientation angle of the vector between the corresponding position of the centroid of the facial region in the unified coordinate system and the corresponding position of the reference point in the unified coordinate system (e.g., the angle with the X-axis in the unified coordinate system) can be used as the facial orientation angle of the verification target object within the outpatient area.

[0096] According to one embodiment of the present invention, geographic location information and facial orientation angle can be used to describe the state of the k-th verification target object when the i-th monitoring video frame is captured, that is, the position and orientation when the i-th monitoring video frame is captured. The sequence number of the monitoring video frame is used to add a time dimension and improve the feature recognition of the trajectory of the verification target object. For example, if the trajectory of one verification target object passes through positions B, C, and D, and the trajectory of another verification target object also passes through positions B, C, and D, without the sequence number of the monitoring video frame, the trajectories of the two unrelated verification target objects would be the same. However, if the sequence numbers of the surveillance video frames are used for auxiliary description, such that the aforementioned verification target object passes through position B when the first surveillance video frame is captured, position C when the tenth surveillance video frame is captured, and position D when the twentieth surveillance video frame is captured, and the aforementioned other verification target object passes through position B when the fiftyth surveillance video frame is captured, position C when the sixtyth surveillance video frame is captured, and position D when the seventyth surveillance video frame is captured, then the time dimension is added by the sequence numbers of the surveillance video frames, making the state feature information of the two unrelated verification target objects inconsistent, thereby improving the feature recognition of the trajectory of the verification target object. The geographical location information, facial orientation angle, and sequence number of the monitoring video frame when the k-th verification target object is captured in the i-th monitoring video frame can be concatenated to obtain a state vector. This state vector can be used as the state feature information of the k-th verification target object when the i-th monitoring video frame is captured. Alternatively, the state vector can be input into a fully connected layer to obtain a high-dimensional vector with higher dimensionality and richer feature description, and this high-dimensional vector can be used as the state feature information of the k-th verification target object when the i-th monitoring video frame is captured.

[0097] According to an embodiment of the present invention, the trajectory features of each verification target object can be described by a graph structure. Multiple state feature information of each verification target object can constitute a subgraph structure, and the subgraph structures corresponding to multiple verification target objects can constitute a graph structure, thereby determining the trajectory similarity of each target object based on the graph structure.

[0098] According to an embodiment of the present invention, the kth subgraph structure is determined based on the state feature information of the kth verification target object when capturing each monitoring video frame, including: using the state feature information of the kth verification target object when capturing the ith monitoring video frame as the first input feature information of the ith node in the kth subgraph structure; obtaining the adjacency matrix of the kth subgraph structure according to formulas (1) and (2).

[0099] (1)

[0100] (2)

[0101] in, This represents the data in the i-th row and i+1-th column of the adjacency matrix of the k-th subgraph structure. This represents the data in the (i+1)th row and i-th column of the adjacency matrix of the k-th subgraph structure. This represents the data in the i-th row and i+s-th column of the adjacency matrix of the k-th subgraph structure. Let s be the data in the (i+s)th row and i-th column of the adjacency matrix of the k-th subgraph structure, where s≠1. Let be the distance between the geographical location information of the k-th verification target object when the i-th monitoring video frame is captured and the geographical location information when the (i+1)-th monitoring video frame is captured. The average distance between the geographical locations of each verification target object when capturing adjacent monitoring video frames is calculated; the degree matrix of the k-th subgraph structure is determined based on the adjacency matrix of the k-th subgraph structure; the k-th subgraph structure is determined based on the adjacency matrix, degree matrix, and first input feature information of each node.

[0102] According to one embodiment of the present invention, in the subgraph structure, each node may have input feature information, and the state feature information of the k-th verification target object when the i-th monitoring video frame is captured can be used as the first input feature information of the i-th node in the k-th subgraph structure. That is, the i-th node can be used to describe the i-th state in the trajectory of the verification target object (i.e., the state when the i-th monitoring video frame is captured).

[0103] According to an embodiment of the present invention, since the object described by the subgraph structure is the trajectory of the verification target object, the nodes in the subgraph structure describe the nodes at a certain moment in the trajectory. In the trajectory, any node is only connected to the previous node and the next node. Therefore, in the adjacency matrix of the subgraph structure, only the data corresponding to the edge between adjacent nodes is not 0, that is, it indicates that adjacent nodes are connected, while there is no edge between non-adjacent nodes, and the corresponding data in their adjacency matrix is ​​0. For example, in formula (2), Let be the data in the i-th row and i+s-th column of the adjacency matrix, representing the relationship between the i-th node and the i+s-th node in the subgraph structure. When s≠1, the two nodes are not adjacent; therefore, there is no edge between them. Thus, there is no relationship between the i-th node and the i+s-th node in the subgraph structure. Similarly, .

[0104] According to an embodiment of the present invention, in formula (1), The data in the i-th row and i+1-th column of the adjacency matrix represents the relationship between the i-th node and the (i+1)-th node in the subgraph structure, i.e., the relationship between adjacent nodes. In the trajectory, the closer the geographical location information when the i-th surveillance video frame is captured is to the geographical location information when the (i+1)-th surveillance video frame is captured, the stronger the association between the two nodes. As The adjacency matrix is ​​a symmetric matrix, therefore, the value of the adjacency matrix describes the strength of the association between two nodes. Based on the above analysis, all data in the adjacency matrix can be obtained, thus yielding the adjacency matrix of the k-th subgraph structure.

[0105] According to one embodiment of the present invention, a degree matrix can be obtained based on the adjacency matrix. The degree matrix contains non-zero data only on the diagonal, with all other data being 0. Furthermore, the i-th data on the diagonal is equal to the sum of all data in the i-th row of the adjacency matrix.

[0106] According to one embodiment of the present invention, the above adjacency matrix, degree matrix and the first input feature information of each node can form the kth subgraph structure.

[0107] In this way, the trajectory information of each verification target object can be described through a subgraph structure, and the sequence number of the monitoring video frame is added to the state feature information of each node to increase the time dimension features and improve the feature recognition of the trajectory information. Furthermore, the relationship between nodes can be determined based on the distance between the geographical locations when each video frame was captured, thereby accurately describing the features of the trajectory information and improving the accuracy of verification based on trajectory features.

[0108] According to one embodiment of the present invention, processing each subgraph structure according to a first graph neural network model to obtain subgraph feature information of each subgraph structure includes: processing the k-th subgraph structure through the first graph neural network model to obtain node feature information of each node in the k-th subgraph structure; and processing the node feature information of each node through the attention mechanism of the first graph neural network model to obtain subgraph feature information of the k-th subgraph structure.

[0109] According to one embodiment of the present invention, the first graph neural network model can process each subgraph structure, thereby operating on the first input feature information of each node based on the degree matrix and adjacency matrix to obtain the updated feature information (i.e., node feature information) of each node. It can also perform a weighted average of the node feature information using an attention mechanism to obtain the subgraph feature information of each subgraph structure. The operation parameters and attention mechanism parameters of the first graph neural network model when operating on the degree matrix, adjacency matrix, and first input feature information can be obtained through training. During training, an end-to-end training method can be adopted, that is, the first graph neural network model and the second graph neural network model process information such as geographical location information, facial orientation angle, and the sequence number of the monitoring video frame to obtain the predicted value of the verification result, and compare it with the manually labeled verification result to obtain the loss function. Then, the first graph neural network model and the second graph neural network model are trained as a whole using the loss function. The present invention does not limit the specific training method.

[0110] According to an embodiment of the present invention, a graph structure is obtained based on the subgraph feature information corresponding to multiple verification target objects, including: using the subgraph feature information of the k-th subgraph structure as the second input feature information of the k-th node in the graph structure; and obtaining the adjacency matrix of the graph structure according to formula (3).

[0111] (3)

[0112] in, The data in the k-th row and s-th column of the adjacency matrix of the graph structure. The data in the s-th row and k-th column of the adjacency matrix of the graph structure. Let be the facial orientation angle of the k-th verification target when the i-th monitoring video frame is captured. Let be the facial orientation angle of the s-th verification target object when the ith monitoring video frame is captured. To verify the geographical location information of the s-th target object when the ith surveillance video frame was captured, Let n be the geographical location information of the k-th verification target object when the i-th monitoring video frame is captured, and n be the number of monitoring video frames in the current monitoring period. To capture multiple surveillance video frames, the average distance between the k-th and s-th verification target objects, i≤n, where i and n are both positive integers; the degree matrix of the graph structure is obtained based on the adjacency matrix of the graph structure; the graph structure is obtained based on the second input feature information of each node in the graph structure, the adjacency matrix, and the degree matrix.

[0113] According to one embodiment of the present invention, the trajectory information of each verification target object can correspond one-to-one with the feature information of the subgraph. Therefore, the subgraph feature information can be used as the second input feature information of the nodes in the graph structure. When obtaining the adjacency matrix of the graph structure, Let be the data in the s-th row and k-th column of the adjacency matrix of the graph structure, representing the correlation between the trajectory information of the k-th verification target object and the trajectory information of the s-th verification target object. The adjacency matrix is ​​a symmetric matrix; therefore, ,and and The meaning is similar and will not be elaborated here.

[0114] According to one embodiment of the present invention, in formula (3), Let represent the cosine of the difference in facial orientation angle between the k-th and s-th verification target objects (i.e., the angle between their facial orientations) when the i-th surveillance video frame is captured. The smaller the value, the larger the cosine value, indicating a higher consistency in the direction of their line of sight. In other words, it indicates that they are more likely to move in the same direction, and thus the higher the correlation between their trajectory information.

[0115] According to one embodiment of the present invention, in formula (3), This vector represents the geographic location information of the k-th verification target object when the i-th surveillance video frame is captured, pointing to the geographic location information of the s-th verification target object. The unit vector in the direction of the facial orientation angle of the kth verification target object. The higher the cosine value between the k-th and s-th target objects, the more likely the k-th verification object is to be associated with the s-th target object, and thus the stronger the correlation between them. Similarly, This vector represents the geographic location information of the s-th verification target object when the ith monitoring video frame is captured, pointing to the geographic location information of the k-th verification target object. The unit vector in the direction of the facial orientation angle of the kth verification target object. The higher the cosine value between the two, the more likely the s-th verification object is to be related to the k-th target object, and the stronger the correlation between them.

[0116] According to one embodiment of the present invention, This indicates that the maximum value of the correlation determined based on the facial orientation angle when capturing the i-th monitoring video frame is taken as the angular correlation between the s-th and k-th verification objects in the i-th monitoring video frame. Furthermore, the average angular correlation can be taken for capturing each monitoring video frame. , which represents the angular correlation between the s-th verification object and the k-th verification object.

[0117] According to one embodiment of the present invention, it is possible to The distance correlation between the s-th and k-th verification objects is considered; the closer the average distance, the higher the distance correlation. The angular correlation and distance correlation can be multiplied to obtain the trajectory correlation between the s-th and k-th verification objects. This correlation can then be used as the correlation between the s-th and k-th nodes in the graph structure, which can be used as the data in the k-th row and s-th column of the adjacency matrix, thus obtaining the adjacency matrix of the graph structure. Furthermore, the degree matrix of the graph structure can be obtained in the same way as the degree matrix of the aforementioned subgraph structure. The second input feature information of each node in the graph structure, the adjacency matrix, and the degree matrix together form the graph structure.

[0118] According to an embodiment of the present invention, a second graph neural network model can be used to process the graph structure to obtain feature information of each verification target object. The feature information of each verification target object can be used to describe the features of the action trajectory of the verification target object. Furthermore, when the second graph neural network model obtains the feature information of a certain verification target object, it can aggregate the second input feature information of the verification target object with the second input feature information of other verification target objects through an adjacency matrix. This allows the influence of the trajectories of other verification target objects on the trajectory of the verification target object to be considered, thereby improving the accuracy of the feature information of the verification target object in representing its trajectory. It can also further increase the accuracy of the representation of the trajectory of the verification target object with correlation and without correlation. The distinction between the trajectories of the verification target objects is improved. For example, if two verification target objects are not related, if one verification target object appears at a certain location, the trajectory of the other verification target object will usually avoid that location. However, if the two verification target objects are related, this phenomenon usually does not occur. This increases the distinction between the trajectories of related verification target objects and those of unrelated verification target objects. Therefore, by aggregating the second input feature information of each verification target object through the second graph neural network model, the feature information of a verification target object can reflect whether it avoids the trajectories of other verification target objects, thereby further improving the accuracy of judging whether two verification target objects are related.

[0119] In this way, the angular correlation between two verification target objects can be determined by the angle between their facial orientations and whether one object is facing the other. The distance correlation is determined by their average distance. Based on both angular and distance correlations, the correlation between their trajectories can be determined, thus establishing the adjacency matrix. Furthermore, a second-graph neural network model can be used to determine the feature information of each verification target object, improving the distinguishability between the trajectories of correlated and uncorrelated verification target objects, thereby enhancing the accuracy of determining whether two verification target objects are correlated.

[0120] According to one embodiment of the present invention, relevant target objects are screened among multiple verification target objects based on the feature information of each verification target object. For example, the cosine similarity of the feature information of each pair of verification target objects is determined. If the cosine similarity is higher than a preset similarity threshold (e.g., 0.6), it is determined that the two verification target objects are related, that is, the two verification target objects are related target objects.

[0121] According to one embodiment of the present invention, obtaining a verification result based on the verification target object and related target objects includes: determining mutually related target objects as related target object groups; determining the number of related target object groups; and obtaining the verification result based on the number of verification target objects, the number of related target object groups, and the number of related target objects. In the example, verification target object 1 and verification target object 2 are mutually related, and verification target object 2 and verification target object 3 are mutually related. Therefore, verification target object 1 and verification target object 3 can be considered mutually related, and the three belong to the same related target object group. For two or more related target objects within a related target object group, regardless of how many related target objects the pending review object has interacted with, it is considered that it has only interacted with one person. Therefore, the number of interactions of the pending review object will decrease by (number of related target objects - number of related target object groups). Thus, the number of interactions of the pending review object becomes [number of verification target objects - (number of related target objects - number of related target object groups)]. This number of interactions can be used as the verification result and compared with the second preset number. If the verification result is still greater than or equal to the second preset number, the pending review object can be determined to be a scalper or medical tout. Otherwise, the pending review object is determined not to be a scalper or medical tout.

[0122] According to one embodiment of the present invention, in step S8, if the candidate for review is determined to be a scalper or medical tout, a review prompt message is generated to prompt security personnel to review the identity information of the candidate. If it is confirmed that the candidate is indeed a scalper or medical tout, appropriate measures can be taken. This allows for the automatic detection and screening of potential scalpers or medical touts, improving screening efficiency, enhancing the timeliness of supervision, reducing manpower consumption, and decreasing the possibility of medical touts and scalpers infringing on patients' rights.

[0123] According to an embodiment of the present invention, the smart hospital information management method can identify target objects exhibiting interactive behavior through surveillance videos of the outpatient area, and filter for potential scalpers and medical touts based on the number of interactions. This allows for the early detection of scalpers and medical touts, improving the timeliness of supervision, reducing manpower consumption, and decreasing the likelihood of scalpers and medical touts infringing on patients' rights. Furthermore, the method can verify the target object based on the behavior and movement routes of multiple verification target objects interacting with the target object, thereby improving the accuracy of identifying scalpers and medical touts and reducing the probability of false identification. When filtering target objects exhibiting interactive behavior, the method can select the closest target objects from multiple objects in the surveillance video frame and verify them using the facial orientation information of the target objects, improving the accuracy of finding target objects exhibiting ongoing interactive behavior and providing an accurate data foundation for filtering scalpers and medical touts. When determining the subgraph structure, the trajectory information of each verification target object can be described using the subgraph structure. Adding the sequence number of the monitoring video frame to the state feature information of each node increases the temporal dimension features, improving the feature recognition of the trajectory information. Furthermore, the relationship between nodes can be determined based on the distance between the geographical locations of each video frame, thus accurately describing the features of the trajectory information and improving the accuracy of trajectory feature-based verification. When determining whether verification target objects are related, the angular correlation between the two verification target objects can be determined by the angle between their facial orientations and whether one verification target object is facing another. The distance correlation can be determined by their average distance. Therefore, the correlation between their trajectories can be determined based on the angular and distance correlations, thus determining the adjacency matrix. The feature information of each verification target object can be determined using a second-graph neural network model, improving the distinguishability between the trajectories of related and unrelated verification target objects, thereby improving the accuracy of determining whether two verification target objects are related.

[0124] Figure 3 An exemplary block diagram of a smart hospital information management system according to an embodiment of the present invention is shown, the system comprising:

[0125] The acquisition module is used to acquire surveillance videos within the outpatient area during the current monitoring period, and to parse the surveillance videos to obtain multiple surveillance video frames.

[0126] The target object module is used to process each monitoring video frame through an image processing model to determine whether there is a target object in each monitoring video frame that is performing interactive behavior.

[0127] The interaction number module is used to record the identification information of the target object that is currently interacting, if it exists, and to count the target objects corresponding to multiple monitoring video frames to determine the number of people interacting with each target object in the current monitoring period.

[0128] The "Pending Review Objects" module is used to filter target objects based on the number of interactions corresponding to each target object, and obtain objects pending review.

[0129] The Target Object Verification Module is used to filter out target objects from all target objects that have interacted with the object to be reviewed during the current monitoring period.

[0130] The location information verification module is used to determine the verification location information of the target object in multiple monitoring video frames within the monitoring period through an image processing model.

[0131] The verification module is used to verify the target to be reviewed based on the verification location information and obtain the verification result;

[0132] The audit notification module is used to obtain audit notification information based on the verification results.

[0133] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0134] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functions and structural principles of the present invention have been demonstrated and explained in the embodiments, and any variations or modifications may be made to the implementation of the present invention without departing from the stated principles.

[0135] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A smart hospital information management method, characterized in that, include: Acquire surveillance video of the outpatient area within the current monitoring period, and parse the surveillance video to obtain multiple surveillance video frames; The image processing model is used to process each monitoring video frame to determine whether there is a target object in each monitoring video frame that is performing interactive behavior. If it exists, record the identification information of the target object that is currently interacting, and count the target objects corresponding to multiple monitoring video frames to determine the number of people interacting with each target object in the current monitoring period. The target objects are filtered based on the number of interactions corresponding to each target object to obtain the objects to be reviewed; Among all target objects, filter out the verification target objects that have interacted with the pending review objects during the current monitoring period; The image processing model is used to determine the verification location information of the target object in multiple monitoring video frames within the monitoring period; Based on the verification location information, the target to be reviewed is verified to obtain the verification result; Based on the verification results, you will receive an audit notification. By using an image processing model, each monitoring video frame is processed to determine whether there is a target object engaging in interactive behavior in each monitoring video frame, including: The image processing model is used to obtain the location information of each object in each monitoring video frame. The location information of the target object includes the location information of the smallest rectangle that selects the object. Obtain the centroid coordinates of the region where each object is located; Based on the centroid coordinates of the region, at the... From the first to the i-th monitoring video frame, determine whether there exists a group of undetermined targets whose centroid distance is always less than a first preset distance threshold. This undetermined target group includes two undetermined target objects, and the centroid distance between the centroid coordinates of the regions where the two undetermined target objects are located is less than a first preset distance threshold. The distance is always less than the first preset distance threshold from the i-th monitoring video frame to the i-th monitoring video frame, where i is a positive integer. This is the first preset quantity; If it exists, the image processing model is used to process the regions where the two target objects in the target group are located to obtain the facial orientation information of the two target objects, and the facial region is determined in the region where the target objects are located. If in the first If the proportion of monitoring video frames from the first monitoring video frame to the i-th monitoring video frame where the facial orientation information of the two undetermined target objects points to the facial area of ​​the other exceeds the first preset proportion, then it is determined that there are target objects in the i-th monitoring video frame that are interacting, and the target objects in the interaction are two undetermined target objects. Based on the verification location information, the target object to be audited is verified to obtain the verification result, including: Based on the verification location information of the target object in the monitoring video frame, determine the geographical location information of the target object within the outpatient area; By using an image processing model, the region where the verification target object is located is processed to obtain the facial orientation information of the verification target object; Based on the facial orientation information, determine the facial orientation angle of the verification target object within the outpatient area; Based on the geographical location information, facial orientation angle, and the sequence number of the monitoring video frame, determine the state feature information of the k-th verification target object when the i-th monitoring video frame is captured; The structure of the kth subgraph is determined based on the state feature information of the kth verification target object when capturing each monitoring video frame; The first graph neural network model is used to process each subgraph structure to obtain the subgraph feature information of each subgraph structure; The graph structure is obtained based on the feature information of the subgraphs corresponding to multiple verification target objects; Based on the second graph neural network model, the graph structure is processed to obtain the feature information of each verification target object; Based on the feature information of each verification target object, relevant target objects are selected from multiple verification target objects; The verification results are obtained based on the target object and related target objects.

2. The smart hospital information management method according to claim 1, characterized in that, Target objects are filtered based on the number of interactions corresponding to each target object, resulting in a list of objects awaiting review, including: If the number of interactions corresponding to the j-th target object is greater than or equal to the second preset number, then the j-th target object is determined as an object to be reviewed.

3. The smart hospital information management method according to claim 1, characterized in that, Based on the state feature information of the k-th verification target object when capturing each monitoring video frame, the structure of the k-th subgraph is determined, including: The state feature information of the kth verification target object when the i-th monitoring video frame is captured is used as the first input feature information of the i-th node in the k-th subgraph structure; According to the formula Obtain the adjacency matrix of the k-th subgraph structure, where, This represents the data in the i-th row and i+1-th column of the adjacency matrix of the k-th subgraph structure. This represents the data in the (i+1)th row and i-th column of the adjacency matrix of the k-th subgraph structure. This represents the data in the i-th row and i+s-th column of the adjacency matrix of the k-th subgraph structure. Let s be the data in the (i+s)th row and i-th column of the adjacency matrix of the k-th subgraph structure, where s≠1. Let be the distance between the geographical location information of the k-th verification target object when the i-th monitoring video frame is captured and the geographical location information when the (i+1)-th monitoring video frame is captured. This is the average distance between the geographical locations of each verification target when capturing adjacent monitoring video frames; The degree matrix of the k-th subgraph structure is determined based on the adjacency matrix of the k-th subgraph structure. The structure of the k-th subgraph is determined based on the adjacency matrix, degree matrix, and first input feature information of each node.

4. The smart hospital information management method according to claim 3, characterized in that, The first graph neural network model is used to process each subgraph structure to obtain the subgraph feature information of each subgraph structure, including: The first graph neural network model is used to process the k-th subgraph structure to obtain the node feature information of each node in the k-th subgraph structure; The node feature information of each node is processed by the attention mechanism of the first graph neural network model to obtain the subgraph feature information of the k-th subgraph structure.

5. The smart hospital information management method according to claim 1, characterized in that, Based on the feature information of the subgraphs corresponding to multiple verification target objects, the graph structure is obtained, including: The subgraph feature information of the kth subgraph structure is used as the second input feature information of the kth node in the graph structure; According to the formula Obtain the adjacency matrix of the graph structure, where, The data in the k-th row and s-th column of the adjacency matrix of the graph structure. The data in the s-th row and k-th column of the adjacency matrix of the graph structure. Let be the facial orientation angle of the k-th verification target when the i-th monitoring video frame is captured. Let be the facial orientation angle of the s-th verification target object when the ith monitoring video frame is captured. To verify the geographical location information of the s-th target object when the ith surveillance video frame was captured, Let n be the geographical location information of the k-th verification target object when the i-th monitoring video frame is captured, and n be the number of monitoring video frames in the current monitoring period. To capture multiple surveillance video frames, the average distance between the k-th verification target object and the s-th verification target object, where i ≤ n, and i and n are both positive integers; Based on the adjacency matrix of the graph structure, obtain the degree matrix of the graph structure; The graph structure is obtained by using the second input feature information, adjacency matrix, and degree matrix of each node in the graph structure.

6. The smart hospital information management method according to claim 1, characterized in that, Based on the target object and related target objects, the verification results are obtained, including: Related target objects that are mutually related are identified as related target object groups; Determine the number of relevant target object groups; The verification results are obtained based on the number of target objects, the number of related target object groups, and the number of related target objects.

7. A smart hospital information management system, used to perform the method as described in any one of claims 1-6, characterized in that, include: The acquisition module is used to acquire surveillance videos within the outpatient area during the current monitoring period, and to parse the surveillance videos to obtain multiple surveillance video frames. The target object module is used to process each monitoring video frame through an image processing model to determine whether there is a target object in each monitoring video frame that is performing interactive behavior. The interaction number module is used to record the identification information of the target object that is currently interacting, if it exists, and to count the target objects corresponding to multiple monitoring video frames to determine the number of people interacting with each target object in the current monitoring period. The "Pending Review Objects" module is used to filter target objects based on the number of interactions corresponding to each target object, and obtain objects pending review. The Target Object Verification Module is used to filter out target objects from all target objects that have interacted with the object to be reviewed during the current monitoring period. The location information verification module is used to determine the verification location information of the target object in multiple monitoring video frames within the monitoring period through an image processing model. The verification module is used to verify the target to be reviewed based on the verification location information and obtain the verification result; The audit notification module is used to obtain audit notification information based on the verification results.

Images