Chair resource state estimation method and system based on time-series heterogeneous data fusion

Abstract

The application discloses a chair resource state estimation method and system based on time sequence heterogeneous data fusion, and relates to the technical field of intelligent medical resource management. The method comprises the following steps: collecting camera visual detection data, doctor calling data and electronic medical record operation data of a chair; mapping the three kinds of heterogeneous data to a unified time axis to generate a three-dimensional observation vector at each time point; based on a recursive Bayesian filter, taking the three-dimensional observation vector as an observation input, using the time correlation of the chair occupation state between adjacent time points to alternately iterate state prediction and observation update, and outputting a continuous estimation value of the chair occupation state; and comparing the continuous estimation value with a preset decision threshold to determine the final occupation state of the chair. The application does not require hardware modification, realizes low-cost and high-robust chair state continuous estimation through multi-source heterogeneous data fusion, and is suitable for intelligent management of chair resources in the oral diagnosis and treatment scene.

Description

Technical Field

[0001] This application relates to the field of intelligent management technology of medical resources, specifically to a method and system for estimating the status of chair resources based on time-series heterogeneous data fusion. Background Technology

[0002] Dental chairs are a core hardware resource in dental hospitals and clinics. Their occupancy status, usage time, and idle time directly affect patient waiting times, doctor scheduling efficiency, and departmental resource utilization. Therefore, real-time monitoring of chair status is a crucial prerequisite for achieving refined operation and intelligent scheduling in medical institutions.

[0003] Currently, there are several technical solutions in the industry for monitoring the status of dental chairs: one solution is a manual inspection and visual judgment solution, in which nurses or triage staff periodically check the chair usage status on-site and record it; another solution is a visual image detection solution based on a single camera, which uses a camera deployed above the chair to collect images and uses target detection algorithms to identify human targets to determine whether the chair is occupied; another solution is a recording solution based on the hospital's business information system, which indirectly determines the chair status by relying on the call logs of the queuing system or the operation logs of the electronic medical record system; another solution is a hardware modification solution that adds gravity sensors or infrared sensors to the chair, which determines whether someone is seated by detecting changes in pressure values ​​or human body heat sources; and a third solution is a recording solution triggered by the patient swiping their card, in which the patient actively swipes their card to mark the start and end times of their visit.

[0004] However, all of the aforementioned existing technologies have significant drawbacks. Manual patrol solutions are labor-intensive, cannot achieve continuous 24 / 7 monitoring, and suffer from subjectivity and delayed information updates. Single-camera solutions are susceptible to interference from complex scenarios such as patients briefly leaving their seats (e.g., getting up to rinse their mouth, take X-rays, pay fees), medical staff temporarily sitting down, body obstruction, and changes in lighting, making it impossible to distinguish between "brief departure" and "departure after treatment," leading to instantaneous misjudgments and status jumps. In single-business system solutions, the queuing system only records the call time and cannot confirm whether the patient is actually seated; the operation time recorded by the electronic medical record system (especially post-operative supplementary entry and archiving operations) deviates significantly from the actual time the patient occupies the chair, resulting in delayed and fragmented data. Hardware modification solutions require disassembling and modifying all treatment chairs, leading to high costs and construction difficulties per chair. Furthermore, gravity sensors are easily interfered with by heavy objects such as instruments and consumables, and infrared sensors have a fixed detection range, unable to accurately correspond to individual chair positions, resulting in a high overall false alarm rate and high maintenance costs. The card-swiping triggering scheme is cumbersome, relies on patient intervention, and suffers from issues such as missed swipes and premature exits, resulting in extremely low automation. Overall, existing technologies have the following core technical problems: 1. Single data source perception is incomplete; cameras, queuing systems, and medical record systems each have different error blind spots, unable to cope with various complex interferences in the medical setting; 2. Data from different sources cannot be used collaboratively; the time granularity of the three data sources is inconsistent, and the data semantics are not interoperable, lacking standardized fusion methods within the industry; 3. Existing technologies determine status instantaneously and independently, failing to utilize the continuity and inertia of chair occupancy status over time, leading to frequent status output jumps and poor stability; 4. Hardware modification solutions are costly, while pure software solutions lack sufficient accuracy, failing to balance low cost and high precision. Summary of the Invention

[0005] This application aims to address the technical problems in existing technologies, such as one-sided perception from a single data source, inability to collaboratively utilize heterogeneous data, failure to utilize temporal state patterns, and high hardware modification costs. It provides a method and system for estimating the state of chair resources based on the fusion of temporal heterogeneous data.

[0006] To achieve the above objectives, this application adopts the following technical solution: In a first aspect, this application provides a method for estimating the state of chair resources based on temporal heterogeneous data fusion, comprising the following steps: Three types of heterogeneous data are collected from the chair positions, including camera visual detection data, doctor's call data, and electronic medical record operation data. The camera visual detection data, the doctor's queuing data, and the electronic medical record operation data are respectively mapped to a unified time axis with a preset time granularity to generate a three-dimensional observation vector for each time point. The three-dimensional observation vector includes the camera status dimension, the queuing status dimension, and the medical record status dimension. Based on the recursive Bayesian filter, the three-dimensional observation vector is used as the observation input. The state prediction and observation update are alternately iterated by utilizing the temporal correlation between adjacent time points of the chair occupancy state, and the continuous estimated value of the chair occupancy state is output. The continuous estimated value is compared with a preset decision threshold to obtain a comparison result. The final occupancy status of the chair is determined based on the comparison result.

[0007] Furthermore, the camera visual detection data includes chair number, detection status value, detection start time and detection end time, and the detection status value includes occupied status, unoccupied status and abnormal status; In the process of mapping the camera visual detection data to a unified time axis, the time stamp is in minutes. If the current minute timestamp is between the detection start time and the detection end time corresponding to the detection status value of "person in" or "abnormal", then the value of the camera status dimension corresponding to the current minute is set to the first value; otherwise, it is set to the second value.

[0008] Furthermore, the doctor's call data includes the doctor's identifier, the call time, the patient's identifier, and the target chair number; In the process of mapping the doctor's call data to a unified timeline, a time window is set with the call time as the starting point and minutes as the timestamp unit. If the current minute timestamp is within the time window, the value of the call status dimension corresponding to the current minute is set to the third value; otherwise, it is set to the fourth value.

[0009] Furthermore, the electronic medical record operation data includes doctor identifier, patient identifier, operating chair number, medical record writing start time, and medical record writing end time; In the process of mapping the electronic medical record operation data to a unified time axis, the time stamp unit is minutes. If the current minute timestamp is between the start time and end time of the medical record writing, the value of the medical record status dimension corresponding to the current minute is set to the fifth value; otherwise, it is set to the sixth value.

[0010] Furthermore, the specific process of using the three-dimensional observation vector as the observation input, based on the recursive Bayesian filter, and alternately iterating between state prediction and observation update by utilizing the temporal correlation between chair occupancy states at adjacent time points to output a continuous estimate of the chair occupancy state is as follows: Based on the recursive Bayesian filter, the three-dimensional observation vector is used as the observation input, the continuous estimated value output at the previous time step is used as the state prediction value at the current time step, and process noise is superimposed on the state prediction value to complete the state prediction. The scalar observation value is obtained by multiplying each state dimension in the three-dimensional observation vector by its corresponding confidence weight and then summing the results. The state prediction value and the scalar observation value are then weighted and fused using Kalman gain to obtain the continuous estimate value at the current time, thus completing the observation update. Repeat the state prediction and observation update to output a continuous estimate of the chair occupancy state at each time step.

[0011] Furthermore, in the mapping of the electronic medical record operation data, only the start time and end time of medical record writing are used as the effective mapping intervals, and the root extraction time and archiving time in the electronic medical record system are not used as the mapping basis.

[0012] Furthermore, the Kalman gain is adaptively adjusted based on the relative magnitudes of the prediction error variance and the observation noise variance: if the prediction error variance is larger relative to the observation noise variance, the Kalman gain is larger, and the continuous estimate at the current moment is more inclined towards the scalar observation value; if the prediction error variance is smaller relative to the observation noise variance, the Kalman gain is smaller, and the continuous estimate at the current moment is more inclined towards the state prediction value.

[0013] Furthermore, the duration of the time window is a configurable integer minute parameter, which is set according to the patient flow rhythm of the dental clinic or the type of departmental treatment.

[0014] Secondly, this application provides a chair resource status estimation system based on temporal heterogeneous data fusion, comprising: The data acquisition module is used to collect three types of heterogeneous data from the chair position, including camera visual detection data, doctor's call data, and electronic medical record operation data. The data mapping module is used to map the camera visual detection data, the doctor's queuing data and the electronic medical record operation data to a unified time axis with a preset time granularity, and generate a three-dimensional observation vector for each time point. The three-dimensional observation vector includes the camera status dimension, the queuing status dimension and the medical record status dimension. The state fusion module has a built-in recursive Bayesian filter, which uses the three-dimensional observation vector as the observation input and utilizes the temporal correlation between adjacent time points of the chair occupancy state to perform alternating iterations of state prediction and observation update, and outputs a continuous estimate of the chair occupancy state. The status determination module is used to compare the continuous estimated value with a preset decision threshold, obtain a comparison result, and determine the final occupancy status of the chair based on the comparison result.

[0015] In a second aspect, this application provides a computer device, including: a processor and a computer-readable storage medium; A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements a method for estimating the state of chair resources based on time-series heterogeneous data fusion.

[0016] Compared with the prior art, this application has the following beneficial effects: This application provides a chair occupancy resource status estimation method based on temporal heterogeneous data fusion. It maps three types of heterogeneous data—camera visual detection data, doctor's queuing data, and electronic medical record operation data—to a unified time axis to construct a three-dimensional observation vector. Then, based on a recursive Bayesian filter, it iteratively predicts and updates the chair occupancy status using the temporal correlation between adjacent time points. Finally, it outputs the chair occupancy status by determining a threshold. This method solves the technical problems of limited perception from a single data source, inability to collaboratively utilize heterogeneous data, failure to utilize temporal state patterns, and high hardware modification costs. Specifically, the three data sources each have different error patterns: cameras provide direct data but are susceptible to interruptions due to brief periods of absence; queuing data provides forward-looking signals but cannot confirm actual occupancy; and medical records reflect synchronous examination information but are lagging in archiving operations. After time alignment through a unified time axis mapping, these data are sequentially fused within a recursive Bayesian filtering framework. Instantaneous deviations from any data source can be complementaryly corrected by the other data sources. Meanwhile, the state prediction step of the recursive Bayesian filter utilizes the temporal inertia of chair occupancy, using the continuous estimate of the previous moment's output as the basis for the current moment's state prediction. This ensures that momentary "unattended" observations such as when a patient briefly gets up to rinse their mouth, take an X-ray, or pay will not cause a state jump, effectively distinguishing between "brief departure" and "departure at the end of the consultation." State prediction and observation updates iterate alternately at each time point, with the output of the previous moment serving as the input for the next moment, allowing for adaptive operation without any manually labeled data. The entire process relies entirely on the existing data from the hospital's cameras, queuing system, and electronic medical record system. There is no need to disassemble or modify the treatment chair, add any new sensor equipment, or perform manual operations such as card swiping or registration by patients or medical staff. This significantly reduces deployment and maintenance costs while achieving highly robust state estimation.

[0017] Specifically, during the observation update process, each state dimension in the three-dimensional observation vector is multiplied by its corresponding confidence weight and then summed to obtain a scalar observation value. The Kalman gain is adaptively adjusted based on the relative magnitude of the prediction error variance and the observation noise variance. When the prediction error variance is larger relative to the observation noise variance, the gain is larger, and the continuous estimate at the current moment is more inclined to the observed value; conversely, the gain is smaller, and the continuous estimate at the current moment is more inclined to the predicted value. This mechanism enables the filter to respond quickly to observation changes when real changes occur in the state (such as a patient taking their seat or leaving their seat), while maintaining state smoothness when the observation is subject to momentary interference (such as a patient briefly getting up or medical staff temporarily passing by), achieving a dynamic balance between "rapid response" and "smooth noise reduction". The decision threshold can be flexibly adjusted according to the different medical institutions' business sensitivity to missed or false alarms: the threshold can be lowered for waiting queue management scenarios to reduce the missed alarm rate, and the threshold can be raised for resource statistics scenarios to reduce the false alarm rate, giving this method good cross-scenario generalization ability. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the overall process of the chair resource status estimation method based on temporal heterogeneous data fusion provided in the embodiments of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] See Figure 1 This application provides a method for estimating the state of chair resources based on temporal heterogeneous data fusion, including the following steps: S1, collect three types of heterogeneous data from the chair position, including camera visual detection data, doctor's call data, and electronic medical record operation data; In a more specific embodiment provided in this application, the camera visual detection data is collected in real time by a camera deployed above the chair, and the presence or absence of a person in the chair is output through an image recognition algorithm. Each piece of camera visual detection data includes a chair number, a detection status value, a detection start time, and a detection end time. The detection status value includes three types: 0 indicates no one is present, 1 indicates someone is present, and 2 indicates an abnormality. An abnormal status can be considered as someone being present. Doctor call data is generated by the hospital's queuing and calling system, recording the precise moment when a doctor calls a patient. Each doctor call record includes fields such as doctor identifier (doctor ID), call time, patient identifier (patient ID), and target chair number; Electronic medical record operation data is generated by the electronic medical record system and records the start and end times of doctors writing medical records. Each electronic medical record operation data includes fields such as doctor identifier (doctor ID), patient identifier (patient ID), operating chair number, medical record writing start time, medical record writing end time, and diagnosis information.

[0022] S2, map the camera visual detection data, the doctor's queuing data and the electronic medical record operation data to a unified time axis with a preset time granularity, and generate a three-dimensional observation vector for each time point. The three-dimensional observation vector includes the camera status dimension, the queuing status dimension and the medical record status dimension. In a more specific embodiment provided in this application, a discrete time series is constructed using a unified time axis with a sampling granularity of 1 minute. For each chair in the chair set, an observation sequence is constructed. ,in This is a three-dimensional observation vector, corresponding to observations in the camera status dimension, the queuing status dimension, and the medical record status dimension, respectively. Each of the three data sources is independently mapped to this unified timeline, with the specific mapping method as follows: Mapping of camera visual inspection data: Using minutes as the timestamp unit, if the current minute timestamp is between the detection start time and detection end time corresponding to the detection status value of "occupied" (status=1) or "abnormal" (status=2), then the value of the camera status dimension corresponding to the current minute is set to the first value (e.g., 1); otherwise, it is set to the second value (e.g., 0).

[0023] In other words, for each discrete time point (minute), it is determined whether that time point falls within the continuous time period of "someone" or "abnormal" output by the camera. If so, it is recorded as 1; otherwise, it is recorded as 0. This step achieves time alignment with business data by discretizing the continuous video detection results into minute-level state values.

[0024] It should be noted that abnormal states are treated as if someone is present (i.e., mapped to the first value), which is a preferred implementation of this application. In practical applications, cameras may experience abnormal situations such as signal loss, algorithm timeouts, or confidence levels below the threshold. These abnormalities are all treated as if someone is present, avoiding missed status detections due to temporary camera malfunctions.

[0025] Mapping of doctor call number data: Using minutes as the timestamp unit, a time window is set starting from the call time. If the current minute timestamp is within the time window, the value of the call status dimension corresponding to the current minute is set to the third value (e.g., 1); otherwise, it is set to the fourth value (e.g., 0).

[0026] That is, define the mapping rule: if the first... The minutes fall within a preset time window following a certain call event (i.e.) ),but ,otherwise .in, The configurable call-out time window parameter can be flexibly adjusted according to the patient flow rhythm and departmental treatment types of different outpatient clinics. The setting of this parameter closely matches real clinical scenarios.

[0027] In real-world dental settings, after a dentist initiates a call for a patient, there is a buffer period for the patient to walk to the examination room, put on protective gear, and prepare for their turn. By setting an adaptive time window, the patient's arrival time can be precisely adapted to different scenarios. For example, orthodontic patients have a relatively fixed follow-up appointment process and a shorter arrival time; implant surgery preparation time is longer, resulting in a longer arrival time. Configuring different window durations based on the characteristics of different departments allows the call data to more accurately reflect the patient's actual arrival time.

[0028] Mapping of electronic medical record operation data: The electronic medical record operation data includes the start time and end time of medical record writing. Using minutes as the timestamp unit, if the current minute timestamp falls between the start and end times of medical record writing, the value of the medical record status dimension corresponding to the current minute is set to the fifth value (e.g., 1); otherwise, it is set to the sixth value (e.g., 0).

[0029] In a preferred embodiment of this application, the mapping of electronic medical record operation data uses only the start and end times of medical record writing as the effective mapping interval. The prescription and archiving times in the electronic medical record system are not used as mapping criteria. This is because, according to actual clinical observation, the time doctors spend writing medical records is usually roughly synchronized with the patient's consultation process, while prescription writing is relatively short and archiving is often completed after the patient leaves the chair, failing to reflect the actual occupancy of the chair. Using only the start and end times of medical record writing as the mapping criteria effectively avoids interference from delayed operations such as "supplementary entry" and "archiving" in the medical record system on state estimation.

[0030] S3, based on the recursive Bayesian filter, using the three-dimensional observation vector as the observation input, and utilizing the temporal correlation between adjacent time points of the chair occupancy status to perform alternating iterations of state prediction and observation update, outputting a continuous estimate of the chair occupancy status; In this step, continuous variables are used. Indicates the chair position at time The occupancy level is determined (0 indicates completely empty, 1 indicates completely occupied), and the hidden state is estimated using a recursive Bayesian filter. The state estimation consists of two steps: "prediction" and "update," which are executed iteratively.

[0031] Specifically, the process of using a recursive Bayesian filter, taking the three-dimensional observation vector as the observation input, and iteratively performing state prediction and observation updates based on the temporal correlation between adjacent time points to output a continuous estimate of the chair occupancy state is as follows: Based on the assumption of temporal continuity of chair occupancy status, the continuous estimate output at the previous moment is used as the state prediction value at the current moment, and process noise is superimposed on the state prediction value to complete the state prediction. The variance of the process noise is a fixed small value.

[0032] In a preferred embodiment of this application, the process noise follows a Gaussian distribution with a fixed small variance. This noise is used to model the natural fluctuations in chair occupancy between adjacent time points. For example, actions such as a patient briefly getting up to rinse their mouth, take an X-ray, or retrieve a report may cause the camera to momentarily detect "no one," but the patient's consultation process is not yet over, and the actual occupancy of the chair has not changed; it should still be considered "occupied." By superimposing the process noise, the filter can adaptively track minute changes in the state while maintaining overall smoothness.

[0033] The scalar observation value is obtained by multiplying each state dimension in the three-dimensional observation vector by its corresponding confidence weight and then summing the results. The state prediction value and the scalar observation value are then weighted and fused using Kalman gain to obtain the continuous estimate value at the current time, thus completing the observation update.

[0034] In the observation update, the three-dimensional observation vector is first... scalar observations are obtained through linear weighted projection.

[0035] in For the camera status dimension, The value for the call status dimension. This refers to the value of the medical record status dimension; These represent the credibility weights for the three data sources. Camera visual detection data directly reflects whether someone is in the chair, and its credibility weight is... It can be set to 0.95; electronic medical record operation data directly reflects the time interval of doctors performing diagnosis and treatment operations at the chair position, and the credibility weight is... It can be set to 0.90; the call number data reflects the signal that the patient is about to be seated, which is an indirect observation, and the reliability weight is... It can be set to 0.80. The credibility weight of each data source in the fusion process is a configurable parameter that can be set according to different application scenarios.

[0036] Through Kalman gain The state predictions and scalar observations are weighted and fused to obtain the fused posterior state estimate (i.e., the continuous estimate at the current time):

[0037] Furthermore, the Kalman gain is adaptively adjusted based on the relative magnitudes of the prediction error variance and the observation noise variance: if the prediction error variance... Relative to the observation noise variance The larger the Kalman gain (i.e., the higher the prediction uncertainty), the higher the Kalman gain. The larger the value, the more the continuous estimate at the current moment tends to resemble the scalar observation. The filter responds quickly to observed changes; if the prediction error variance Relative to the observation noise variance The smaller the value (i.e., the more certain the prediction), the higher the Kalman gain. The smaller the value, the more the continuous estimate at the current moment tends to be the predicted state value. The filter removes transient interference from the observation.

[0038] In this step, when the prediction uncertainty is high but the observation is relatively reliable, the filter trusts the observation more and quickly tracks state changes; when the prediction is relatively certain but the observation noise is large, the filter trusts the prediction more and filters out transient interference in the observation. This mechanism allows the filter to achieve a dynamic balance between fast response and smooth noise reduction, effectively solving the problem of misjudgment caused by interference from a single data source.

[0039] The state prediction and observation update are repeated, meaning that at each time step, the prediction and update operations are performed sequentially, and the continuous estimates output at the previous time step are used. As input for the state prediction at the next time step, a recursive loop is formed, outputting a continuous estimate of the chair occupancy state at each time step. This continuous estimate ranges from 0 to 1, reflecting the probability or degree to which the chair is occupied.

[0040] S4. Compare the continuous estimated value with the preset decision threshold to obtain the comparison result. Determine the final occupancy status of the chair based on the comparison result.

[0041] Specifically, the continuous estimate output in step S3 With preset decision threshold Comparison: If If the condition is met, the chair is determined to be "occupied"; otherwise, it is determined to be "unoccupied". The default decision threshold is 0.5.

[0042] Specifically, the preset decision threshold is a configurable parameter, which can be adjusted upwards or downwards based on the business sensitivity requirements for underreporting or false reporting. In practical applications, if the outpatient department is more concerned with preventing patients from waiting too long in the waiting area (preferring to overreport patients rather than underreport), the threshold can be lowered, for example, to 0.4, so that more ambiguous states are judged as "occupied," prioritizing a reduction in the underreporting rate. If the outpatient department is more concerned with the accuracy of resource utilization statistics (preferring to underreport rather than false report), the threshold can be raised, for example, to 0.6, so that more ambiguous states are judged as "unoccupied," prioritizing a reduction in the false reporting rate. Through this adjustable mechanism, this method can flexibly adapt to the operational strategies and management needs of different medical institutions.

[0043] Steps S1 to S4 are repeated at each time point to form a continuous estimation sequence of chair occupancy status, thereby realizing real-time, online monitoring of chair status.

[0044] In another embodiment provided in this application, a chair resource status estimation system based on temporal heterogeneous data fusion is provided, comprising: The data acquisition module is used to collect three types of heterogeneous data from the chair position, namely, camera visual detection data, doctor's call data, and electronic medical record operation data. The data mapping module is used to map the camera visual detection data, the doctor's queuing data and the electronic medical record operation data to a unified time axis with a preset time granularity, and generate a three-dimensional observation vector for each time point. The three-dimensional observation vector includes the camera status dimension, the queuing status dimension and the medical record status dimension. The state fusion module has a built-in recursive Bayesian filter, which uses the three-dimensional observation vector as the observation input and utilizes the temporal correlation between adjacent time points of the chair occupancy state to perform alternating iterations of state prediction and observation update, and outputs a continuous estimate of the chair occupancy state. The status determination module is used to compare the continuous estimated value with a preset decision threshold, obtain a comparison result, and determine the final occupancy status of the chair based on the comparison result.

[0045] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0046] The various embodiments in this specification are described in a progressive manner. For directly identical or similar parts of the embodiments, refer to each other. Each embodiment focuses on its differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0047] The embodiments described above are merely preferred embodiments of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various improvements and substitutions without departing from the technical principles of this application, and these improvements and substitutions should also be considered within the scope of protection of this application. Therefore, the scope of protection of this patent application should be determined by the scope of the claims.

Claims

1. A method for estimating the state of chair resources based on temporal heterogeneous data fusion, characterized in that, Includes the following steps: Three types of heterogeneous data are collected from the chair positions, including camera visual detection data, doctor's call data, and electronic medical record operation data. The camera visual detection data, the doctor's queuing data, and the electronic medical record operation data are respectively mapped to a unified time axis with a preset time granularity to generate a three-dimensional observation vector for each time point. The three-dimensional observation vector includes the camera status dimension, the queuing status dimension, and the medical record status dimension. Based on the recursive Bayesian filter, the three-dimensional observation vector is used as the observation input. The state prediction and observation update are alternately iterated by utilizing the temporal correlation between adjacent time points of the chair occupancy status, and the continuous estimate of the chair occupancy status is output. The continuous estimated value is compared with a preset decision threshold to obtain a comparison result. The final occupancy status of the chair is determined based on the comparison result.

2. The chair resource status estimation method based on temporal heterogeneous data fusion according to claim 1, characterized in that, The camera visual detection data includes chair number, detection status value, detection start time and detection end time, and the detection status value includes occupied status, unoccupied status and abnormal status. In the process of mapping the camera visual detection data to a unified time axis, the time stamp is in minutes. If the current minute timestamp is between the detection start time and the detection end time corresponding to the detection status value of "person in" or "abnormal", then the value of the camera status dimension corresponding to the current minute is set to the first value; otherwise, it is set to the second value.

3. The chair resource status estimation method based on temporal heterogeneous data fusion according to claim 1, characterized in that, The doctor's call data includes the doctor's identifier, the call time, the patient's identifier, and the target chair number; In the process of mapping the doctor's call data to a unified timeline, a time window is set with the call time as the starting point and minutes as the timestamp unit. If the current minute timestamp is within the time window, the value of the call status dimension corresponding to the current minute is set to the third value; otherwise, it is set to the fourth value.

4. The chair resource status estimation method based on temporal heterogeneous data fusion according to claim 1, characterized in that, The electronic medical record operation data includes doctor identifier, patient identifier, operating chair number, medical record writing start time, and medical record writing end time; In the process of mapping the electronic medical record operation data to a unified time axis, the time stamp unit is minutes. If the current minute timestamp is between the start time and end time of the medical record writing, the value of the medical record status dimension corresponding to the current minute is set to the fifth value; otherwise, it is set to the sixth value.

5. The chair resource status estimation method based on temporal heterogeneous data fusion according to claim 1, characterized in that, The specific process of using a recursive Bayesian filter, taking the three-dimensional observation vector as the observation input, and iteratively performing state prediction and observation updates based on the temporal correlation between adjacent time points to output a continuous estimate of the chair occupancy state is as follows: Based on the recursive Bayesian filter, the three-dimensional observation vector is used as the observation input, the continuous estimated value output at the previous time step is used as the state prediction value at the current time step, and process noise is superimposed on the state prediction value to complete the state prediction. The scalar observation value is obtained by multiplying each state dimension in the three-dimensional observation vector by its corresponding confidence weight and then summing the results. The state prediction value and the scalar observation value are then weighted and fused using Kalman gain to obtain the continuous estimate value at the current time, thus completing the observation update. Repeat the state prediction and observation update to output a continuous estimate of the chair occupancy state at each time step.

6. The chair resource status estimation method based on temporal heterogeneous data fusion according to claim 1, characterized in that, In the mapping of electronic medical record operation data, only the start time and end time of medical record writing are used as the effective mapping intervals. The root extraction time and archiving time in the electronic medical record system are not used as the mapping basis.

7. The chair resource status estimation method based on temporal heterogeneous data fusion according to claim 5, characterized in that, The Kalman gain is adaptively adjusted based on the relative magnitudes of the prediction error variance and the observation noise variance: if the prediction error variance is larger relative to the observation noise variance, the Kalman gain is larger, and the continuous estimate at the current time is more inclined to the scalar observation value. If the variance of the prediction error is smaller relative to the variance of the observation noise, then the Kalman gain is smaller, and the continuous estimate at the current moment is more likely to be the state prediction.

8. The chair resource status estimation method based on temporal heterogeneous data fusion according to claim 3, characterized in that, The duration of the time window is a configurable integer minute parameter, which is set according to the patient flow rhythm of the dental clinic or the type of treatment in the department.

9. A chair resource status estimation system based on temporal heterogeneous data fusion, characterized in that, include: The data acquisition module is used to collect three types of heterogeneous data from the chair position, namely, camera visual detection data, doctor's call data, and electronic medical record operation data. The data mapping module is used to map the camera visual detection data, the doctor's queuing data and the electronic medical record operation data to a unified time axis with a preset time granularity, and generate a three-dimensional observation vector for each time point. The three-dimensional observation vector includes the camera status dimension, the queuing status dimension and the medical record status dimension. The state fusion module has a built-in recursive Bayesian filter, which uses the three-dimensional observation vector as the observation input and utilizes the temporal correlation between adjacent time points of the chair occupancy state to perform alternating iterations of state prediction and observation update, and outputs a continuous estimate of the chair occupancy state. The status determination module is used to compare the continuous estimated value with a preset decision threshold, obtain a comparison result, and determine the final occupancy status of the chair based on the comparison result.

10. A computer device, characterized in that, include: Processor and computer-readable storage media; A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements a chair resource state estimation method based on temporal heterogeneous data fusion as described in any one of claims 1 to 8.

Images