An elevator leveling monitoring system and method integrating visual recognition and barometric pressure learning

Abstract

This invention discloses an elevator leveling monitoring system and method integrating visual recognition and barometric pressure learning. It includes a visual perception module connected to an edge computing module via a network cable for real-time acquisition of elevator panel images; a barometric pressure measurement module connected to the edge computing module for acquiring the ambient atmospheric pressure at the elevator car's location; a reference sensing module installed on the elevator equipment's stabilization platform and connected to the edge computing module for measuring the horizontal angle of the elevator equipment's working plane; an edge computing module deployed on the top of the elevator car or in the machine room for running a barometric pressure self-learning algorithm, a multi-source fusion leveling determination engine, and fault detection logic; and a cloud management module deploying services in the cloud and communicating with the edge computing module via an MQTT message queue. The advantages of this invention are: achieving contactless floor calibration and leveling detection; maintaining highly reliable leveling determination and enabling self-correction; and significantly improving safety and reliability.

Description

Technical Field

[0001] This invention relates to the fields of Internet of Things, cloud computing, automation control and data processing, and in particular to an elevator leveling monitoring system and method that integrates visual recognition and barometric pressure learning. Background Technology

[0002] With the rapid development of high-rise buildings in cities and increasingly stringent requirements for elevator safety management, especially in scenarios such as real-time monitoring of elevator operation status, precise leveling control, and fault early warning, higher demands are being placed on the level of equipment intelligence. Traditional mechanical limit switches or manual inspection methods are inefficient, prone to errors, and difficult to implement remote monitoring and fault early warning. Therefore, utilizing Internet of Things (IoT) technology, combining sensors, edge computing, and cloud computing services to achieve accurate judgment and automated linkage of key elevator mechanical movements, is an important direction for current technological development.

[0003] Existing elevator leveling detection and fault monitoring technologies have the following technical problems: High installation adaptability and retrofitting costs: Traditional leveling detection relies on shaft magnetic switches, absolute encoders, or photoelectric switches, which require drilling and wiring within the shaft, making installation complex and poorly adaptable to older elevators. Especially for elevators without machine rooms and buildings where the civil engineering has already been completed, the later installation of sensors is extremely difficult, resulting in a large number of older elevators being unable to be connected to intelligent monitoring systems.

[0004] Low reliability of single sensors: Existing solutions mostly use a single sensor (such as only encoder or only barometric pressure sensor) to determine the floor. Once the sensor drifts, is damaged, or is affected by environmental interference (such as sudden changes in barometric pressure or encoder slippage), the system will generate an incorrect leveling signal, which may lead to safety accidents such as people being trapped, rushing to the top, or falling to the bottom. In addition, there is a lack of effective cross-verification mechanism.

[0005] The disconnect between visual analysis and physical sensing: Some new solutions introduce cameras for panel recognition, but the image recognition system and the elevator control system are independent of each other and data fusion is not achieved. The AI ​​recognition results are not compared with physical quantities such as air pressure and encoders in real time, and self-checking and error correction are not possible. The misrecognition rate is high in edge scenarios such as panel aging, sudden changes in light, and reflection obstruction.

[0006] The lack of a real-time fault confirmation mechanism for non-leveling stops: Although existing IoT solutions can report fault codes, they mostly rely on the self-test signals of the elevator control system and cannot independently verify whether the car is actually stopped in the unlocked area. For "hidden" non-leveling stops caused by brake slippage, wire rope elongation, etc., the system cannot detect and force-report them in a timely manner, posing a serious safety hazard.

[0007] Insufficient collaboration between edge computing and the cloud: Existing solutions often directly upload raw data to the cloud for processing, resulting in large network latency and poor real-time performance; or they only perform simple threshold judgments at the edge, lacking the ability to dynamically assess confidence through multi-source data fusion, and are unable to complete complex leveling and fault diagnosis decisions at the edge.

[0008] Poor adaptability to extreme environments: During the operation of ultra-high-speed elevators, the air pressure changes nonlinearly, causing traditional air pressure ranging algorithms to fail; Under harsh environments such as high temperature, high humidity, and strong electromagnetic interference, the failure rate of a single sensor increases significantly, and existing technologies lack effective environmental adaptation and sensor redundancy switching mechanisms. Summary of the Invention

[0009] The present invention aims to overcome the above-mentioned deficiencies in the prior art and provides an elevator leveling monitoring system and method that integrates visual recognition and air pressure learning to achieve accurate leveling determination.

[0010] To achieve the above objectives, the present invention adopts the following technical solution: An elevator leveling monitoring system integrating visual recognition and barometric pressure learning includes a visual perception module, a barometric pressure measurement module, a reference sensing module, an edge computing module, and a cloud management module. The visual perception module is connected to the edge computing module via a network cable to collect and recognize elevator panel images in real time. The barometric pressure measurement module is connected to the edge computing module to collect the ambient atmospheric pressure at the location of the elevator car. The reference sensing module is installed on the elevator equipment stabilization platform and connected to the edge computing module to measure the horizontal angle of the elevator equipment's working plane. The edge computing module is deployed on the top of the elevator car or in the machine room to receive data from the visual perception module, the barometric pressure measurement module, and the reference sensing module to run a barometric pressure self-learning algorithm, a visual-barometric pressure fusion leveling determination engine, and fault detection logic. The cloud management module deploys services in the cloud and communicates with the edge computing module via an MQTT message queue.

[0011] This invention achieves contactless floor calibration and leveling detection by combining visual recognition with a barometric pressure sensor. It eliminates the need for drilling holes in the elevator shaft for wiring or installing mechanical sensors, making it particularly suitable for machine-room-less elevators and the intelligent retrofitting of older elevators in existing buildings, significantly reducing installation costs and construction difficulty. Through a visual-barometric pressure dual-modal fusion algorithm, the recognized panel direction arrows, floor numbers, barometric pressure differences, and running distance are compared in real time to establish a consistency verification mechanism. This ensures highly reliable leveling determination even under challenging scenarios such as complex lighting, panel aging, and sudden weather changes, enabling self-checking and error correction.

[0012] Preferably, the visual perception module includes at least one network AI camera, which is connected to the edge computing module via a network cable, for real-time acquisition of elevator panel images and identification of floor numbers, directional arrows, and panel status.

[0013] Preferably, the air pressure measurement module includes at least one air pressure sensor based on MEMS technology, which integrates a temperature compensation algorithm to collect the ambient atmospheric pressure at the location of the elevator car and output digital air pressure measurement values ​​in hPa or Pa.

[0014] Preferably, the reference sensing module includes at least one tilt sensor, which is rigidly mounted on the elevator equipment stabilization platform to measure the horizontal angle of the elevator equipment's working plane and provide an attitude reference for air pressure altitude conversion.

[0015] Preferably, the edge computing module includes at least one edge computing unit deployed on the top of the elevator car or in the machine room, for running a pressure self-learning algorithm, a visual pressure fusion leveling engine, and fault detection logic.

[0016] Preferably, the cloud management module deploys services in the cloud, including an MQTT Broker message queue, a device management unit, a monitoring and alarm unit, and a data analysis unit.

[0017] This invention also provides an elevator leveling monitoring method that integrates visual recognition and air pressure learning, specifically including the following steps: (1) The floor numbers displayed on the elevator panel are identified by the visual perception module, and the air pressure calibration of the floor is automatically triggered. The floor-air pressure mapping table, i.e. FP table, is established by combining the air pressure measurement module and the reference sensing module, so as to realize the self-learning construction of the floor-air pressure mapping table without human intervention. (2) The elevator panel image is analyzed by AI video analysis through the network AI camera in the visual perception module to obtain the visual leveling status VLS of the elevator. The floor-pressure mapping table is queried through the air pressure sensor in the air pressure measurement module to obtain the air pressure leveling status BLS. The running distance is obtained for consistency verification. The fused leveling confirmation is only output when the visual leveling status and the air pressure leveling status are consistent and the confidence index meets the standard. (3) After detecting the elevator stop command, the fault confidence index (FCI) is calculated by the edge computing module by combining VLS, BLS, elevator built-in encoder zero speed signal and door lock status. When FCI≥0.95 and the car position exceeds the unlocking area, the non-leveling elevator stop fault is confirmed, and the forced reporting mechanism is triggered. The triple mechanism of MQTT QoS 2, Retained message and Last Will is used to ensure that the platform reaches the cloud management module, and the fault freeze state is entered to prohibit the door from opening.

[0018] This invention employs a multi-source redundancy design combining AI video analysis (panel orientation, floor numbers) and barometric pressure sensors. A dynamic confidence assessment mechanism enables cross-validation, ensuring accurate leveling status determination even if a single sensor fails or is affected by environmental interference (such as sudden pressure changes or panel glare), significantly improving safety and reliability. Through a multi-source fusion fault confirmation mechanism, it integrates visual leveling status, barometric pressure leveling status, encoder zero-speed signal, and door lock circuit status to calculate a fault confidence index. Mandatory reporting is triggered only when the confidence level meets the threshold, satisfying the stringent reliability requirements of elevator IoT safety monitoring.

[0019] Preferably, in step (1), the following steps are taken: when the elevator is in the initial state, the network AI camera identifies the current elevator car on a certain floor on the elevator panel, the air pressure sensor confirms that the elevator is stationary and outputs the air pressure measurement value, and notifies the elevator equipment to calibrate the FP table; during the subsequent continuous operation of the elevator, the air pressure sensor confirms that the elevator is stationary each time and calculates the running distance based on the air pressure measurement value. Combined with the reference sensor module, it is determined whether the elevator is at the level position. Combined with the network AI camera to identify the floor and panel direction on the elevator panel, the calibrated FP table is continuously self-learned and updated synchronously, so as to realize the self-learning construction of the floor table without human intervention.

[0020] As a preferred option, in step (2), the following steps are taken: the visual leveling status (VLS) is obtained through AI video analysis. The visual leveling status is determined by whether there is an arrow on the elevator panel identified by the network AI camera. If there is an arrow, the elevator is running; if there is no arrow, the elevator is level or stationary. The air pressure leveling status (BLS) is obtained by querying the FP table through the air pressure sensor. That is, after the air pressure sensor calculates the running distance, it compares the air pressure at the leveling position in the floor-air pressure mapping table to determine whether the elevator is level. The consistency verification between the floor identified by the video and the stationary state of the elevator detected by the air pressure sensor is performed. Only when the dual-source states of VLS and BLS are consistent and the confidence index exceeds the set threshold is the final fused leveling confirmation output output to avoid misjudgment by a single sensor.

[0021] Preferably, in step (3), the cloud management module communicates with the edge computing module through an MQTT message queue to remotely configure FP table parameters, leveling tolerance threshold, and sensor calibration instructions. The leveling tolerance threshold refers to whether the air pressure sensor is within the threshold after measuring the operating distance and comparing it with the FP table. The sensor calibration instruction refers to the calibration and re-self-learning calibration triggered after the network AI camera recognizes the discrepancy between the air pressure sensor's calculation and the actual measurement. The module also receives the equipment operating status, leveling judgment results, and fault alarms in real time, providing a visual monitoring interface and data traceability function.

[0022] The beneficial effects of this invention are: it enables contactless floor calibration and leveling detection, eliminating the need for drilling holes and wiring or installing mechanical sensors in the shaft, making it particularly suitable for machine-room-less elevators and the intelligent renovation of old elevators in existing buildings, significantly reducing installation costs and construction difficulty; it can maintain highly reliable leveling determination even under edge scenarios such as complex lighting, panel aging, and sudden weather changes, achieving self-checking and error correction; even if a single sensor fails or is affected by environmental interference (such as sudden air pressure changes or panel reflections), it can still accurately determine the leveling status, significantly improving safety and reliability; and it meets the stringent reliability requirements of elevator IoT safety monitoring. Detailed Implementation

[0023] The present invention will be further described below with reference to specific embodiments.

[0024] An elevator leveling monitoring system integrating visual recognition and barometric pressure learning includes a visual perception module, a barometric pressure measurement module, a reference sensing module, an edge computing module, and a cloud management module. The visual perception module is connected to the edge computing module via a network cable to collect and recognize elevator panel images in real time. The barometric pressure measurement module is connected to the edge computing module to collect the ambient atmospheric pressure at the elevator car's location. The reference sensing module is installed on the elevator equipment's stabilization platform and connected to the edge computing module to measure the horizontal angle of the elevator equipment's working plane. The edge computing module is deployed on the top of the elevator car or in the machine room to receive data from the visual perception module, the barometric pressure measurement module, and the reference sensing module to run a barometric pressure self-learning algorithm, a visual-barometric pressure fusion leveling judgment engine, and fault detection logic. The cloud management module deploys services in the cloud and communicates with the edge computing module via an MQTT message queue. Details are as follows: The visual perception module includes at least one network AI camera, which is connected to the edge computing module via a network cable to collect elevator panel images in real time and identify floor numbers, directional arrows (up / down / stationary), and panel status.

[0025] The air pressure measurement module includes at least one air pressure sensor based on MEMS technology, which integrates a temperature compensation algorithm to collect the ambient atmospheric pressure at the location of the elevator car and output digital air pressure measurement values ​​in hPa or Pa.

[0026] The reference sensing module includes at least one tilt sensor, which is rigidly mounted on the elevator equipment stabilization platform to measure the horizontal angle of the elevator equipment's working plane and provide an attitude reference for barometric altitude conversion.

[0027] The edge computing module includes at least one edge computing unit (DTU / industrial gateway), deployed on the top of the elevator car or in the machine room, for running air pressure self-learning algorithms, visual air pressure fusion leveling determination engines, and fault detection logic.

[0028] The cloud management module deploys services in the cloud, including an MQTT Broker message queue, a device management unit, a monitoring and alarm unit, and a data analysis unit.

[0029] This invention also provides an elevator leveling monitoring method that integrates visual recognition and air pressure learning, specifically including the following steps: (1) The floor numbers displayed on the elevator panel are identified by the visual perception module, and the air pressure calibration of the floor is automatically triggered. The floor-air pressure mapping table, i.e. FP table, is established by combining the air pressure measurement module and the reference sensing module, so as to realize the self-learning construction of the floor-air pressure mapping table without human intervention.

[0030] Specifically, when the elevator is in its initial state, the network AI camera identifies the current floor of the elevator car on the elevator panel, the air pressure sensor confirms that the elevator is stationary and outputs the air pressure measurement value, notifying the elevator equipment to calibrate the FP table; during the subsequent continuous operation of the elevator, the air pressure sensor confirms that the elevator is stationary each time and calculates the running distance based on the air pressure measurement value. Combined with the reference sensor module, it determines whether the elevator is level with the floor. Combined with the network AI camera's recognition of the floor and panel direction on the elevator panel, the calibrated FP table is continuously self-learned and updated synchronously, realizing the self-learning construction of the floor table without human intervention.

[0031] It's important to note that elevator floor data acquired through AI video analysis using network AI cameras can only be used for floor data correction (e.g., checking if it matches the floor calculated based on air pressure), and cannot be used to determine leveling. Leveling must still be determined based on air pressure. When AI video analysis identifies the floor and the air pressure sensor detects the elevator is stationary, the FP (Flat Function) table needs to be continuously learned and updated because air pressure can drift with changes in temperature and time.

[0032] (2) The elevator panel image is analyzed by the network AI camera in the visual perception module (the network AI camera is equipped to identify the floor on the elevator panel and whether the elevator is currently in operation). The visual leveling status VLS of the elevator is obtained (based on the presence or absence of directional arrows). The floor-pressure mapping table is queried by the air pressure sensor in the air pressure measurement module to obtain the air pressure leveling status BLS. The running distance is obtained for consistency verification. The fused leveling confirmation is only output when the visual leveling status and the air pressure leveling status are consistent and the confidence index meets the standard.

[0033] Specifically, the process involves: obtaining the visual leveling status (VLS) through AI video analysis, where the network AI camera identifies whether there are arrows on the elevator panel; if there are arrows, the elevator is running; otherwise, it is level or stationary. The air pressure leveling status (BLS) is obtained by querying the floor-air pressure mapping table using a barometric pressure sensor; this involves the barometric pressure sensor calculating the travel distance and comparing it to the air pressure at the leveling location in the floor-air pressure mapping table to determine if the elevator is level. The consistency between the floor identified by the video and the stationary elevator status detected by the air pressure sensor is verified. Only when the VLS and BLS statuses are consistent and the confidence index exceeds a set threshold is a fused leveling confirmation output generated, avoiding misjudgments from a single sensor. This involves using a network AI camera to identify directional arrows on the elevator panel (arrows indicate operation, no arrows indicate suspected leveling), while simultaneously using a barometric pressure sensor to measure air pressure changes and compare them with a floor-barometric pressure mapping table (FP table) to determine the height position. Only when the visual judgment (VLS) and barometric pressure judgment (BLS) are consistent (e.g., both determine leveling), and the confidence index (combining visual recognition accuracy, air pressure stability, and time persistence) exceeds a set threshold, will a fused leveling confirmation be output, thus avoiding misjudgment by a single sensor.

[0034] (3) Upon detecting a stop command, the Fault Confidence Index (FCI) is calculated using the edge computing module by integrating VLS, BLS, the zero-speed signal of the elevator's built-in encoder, and the door lock status. When the FCI ≥ 0.95 and the car position exceeds the unlocking area, a non-leveling stop fault is confirmed, triggering a mandatory reporting mechanism. A triple mechanism of MQTT QoS 2, Retained messages, and Last Will ensures that the platform reaches the cloud management module, and the system enters a fault freeze state, prohibiting door opening. Among these, the QoS 2 mechanism refers to ensuring accurate alarm delivery once the message is lost or duplicated during transmission; the Retained mechanism refers to ensuring that the alarm is traceable and not missed when the platform is offline; and the Last Will mechanism refers to ensuring that the abnormal state is detected when the equipment fails to report in time.

[0035] The cloud management module communicates with the edge computing module via an MQTT message queue to remotely configure FP meter parameters, leveling tolerance thresholds, and sensor calibration commands. The leveling tolerance threshold refers to whether the distance measured by the barometer sensor is within the threshold range when compared with the FP meter. The threshold is approximately 50 centimeters above or below the leveling position. The sensor calibration command is triggered when the network AI camera's recognition and the barometer sensor's calculation are inconsistent, thus initiating calibration and re-learning. The module also receives real-time equipment operating status, leveling determination results, and fault alarms, providing a visual monitoring interface and data traceability function.

[0036] Through the above technical solution, the present invention solves the following technical problems: (a) Solves the problem of difficult installation in the renovation of old elevators: This invention uses camera visual recognition combined with air pressure sensor to achieve contactless floor calibration and leveling detection. There is no need to drill holes and wires in the shaft or install mechanical sensors. It is especially suitable for the intelligent renovation of elevators without machine room and old elevators in existing buildings, which greatly reduces installation costs and construction difficulty.

[0037] (b) Solves the problems of low reliability and susceptibility to interference of single sensors: The present invention adopts a multi-source redundancy design of AI video analysis (panel direction, floor number) and barometric pressure sensor. Based on the calibrated FP meter, the barometric pressure sensor calculates the running distance and compares it with the FP meter to see if it is in the level position. The AI ​​camera identifies the floor in the panel and the panel direction to determine whether it is in the normal position. That is, cross-validation is achieved through dynamic confidence evaluation mechanism. Even if a single sensor fails or is affected by environmental interference (such as sudden air pressure change, panel reflection), the system can still accurately determine the level status, which significantly improves safety and reliability.

[0038] (c) It solves the problems of the separation between visual analysis and physical sensing and the high misidentification rate in edge scenes: The present invention uses a visual-barometric dual-modal fusion algorithm to compare the panel direction arrows and floor numbers identified by AI with the barometric height difference and running distance in real time, and establishes a consistency verification mechanism. It can still maintain high reliability in leveling judgment under edge scenes such as complex lighting, panel aging, and sudden weather changes, and achieve self-checking and error correction.

[0039] (d) Solved the problem of non-leveling elevator stop faults being unable to be confirmed in real time: This invention uses a multi-source fusion fault confirmation mechanism to integrate visual leveling status, air pressure leveling status, encoder zero speed signal and door lock circuit status to calculate the fault confidence index. It only triggers mandatory reporting when the confidence level meets the standard, thus meeting the stringent reliability requirements of elevator IoT safety monitoring.

[0040] (e) This invention addresses the issues of insufficient collaboration between edge computing and the cloud, and poor real-time performance: It deploys a self-learning air pressure algorithm and a leveling engine at the edge to achieve localized real-time decision-making; through MQTT protocol and cloud collaboration, it employs QoS-level transmission, retained messages, and a Last Will mechanism to ensure the delivery of critical fault information, balancing real-time performance and reliability. Specifically, after the device is installed, the AI ​​camera identifies the current floor and notifies the device to calibrate the FP meter. After the FP meter calibration is complete, the air pressure sensor determines whether the elevator has reached the leveling position based on the elevator's travel distance.

[0041] The key terms used in the above technical solutions are explained below: Air pressure self-learning: Through vision-air pressure multi-source fusion, high-precision positioning of elevator floors is achieved under non-contact installation conditions, which is particularly suitable for the renovation of old elevators and ultra-high-speed elevators (where air pressure changes are non-linear).

[0042] Barometric pressure sensor: an absolute pressure measuring element based on MEMS (microelectromechanical systems) technology, integrating a temperature compensation algorithm, used to collect the ambient atmospheric pressure at the location of the elevator car, and output digital measurement values ​​in hPa or Pa.

[0043] Operating distance: Pressure ranging enables high-precision leveling determination without additional wellbore sensors.

[0044] AI video analysis: Through visual-barometric dual-modal fusion, it can still maintain highly reliable leveling determination even in edge scenarios such as complex lighting, panel aging, and sudden weather changes.

[0045] The elevator car stops outside the unlocking area: Through a multi-source fusion fault confirmation + mandatory reporting mechanism, the stringent reliability requirements of elevator IoT security monitoring are met.

Claims

1. An elevator leveling monitoring system integrating visual recognition and barometric pressure learning, characterized in that, The system includes a visual perception module, a barometric pressure measurement module, a reference sensing module, an edge computing module, and a cloud management module. The visual perception module is connected to the edge computing module via a network cable to collect and identify elevator panel images in real time. The barometric pressure measurement module is connected to the edge computing module to collect the ambient atmospheric pressure at the location of the elevator car. The reference sensing module is installed on the elevator equipment stabilization platform and connected to the edge computing module to measure the horizontal angle of the elevator equipment's working plane. The edge computing module is deployed on the top of the elevator car or in the machine room to receive data from the visual perception module, the barometric pressure measurement module, and the reference sensing module to run a barometric pressure self-learning algorithm, a visual-barometric pressure fusion leveling engine, and fault detection logic. The cloud management module deploys services in the cloud and communicates with the edge computing module via an MQTT message queue.

2. The elevator leveling monitoring system integrating visual recognition and barometric pressure learning according to claim 1, characterized in that, The visual perception module includes at least one network AI camera, which is connected to the edge computing module via a network cable to collect elevator panel images in real time and identify floor numbers, directional arrows, and panel status.

3. The elevator leveling monitoring system integrating visual recognition and barometric pressure learning according to claim 1, characterized in that, The air pressure measurement module includes at least one air pressure sensor based on MEMS technology, which integrates a temperature compensation algorithm to collect the ambient atmospheric pressure at the location of the elevator car and output digital air pressure measurement values ​​in hPa or Pa.

4. The elevator leveling monitoring system integrating visual recognition and barometric pressure learning according to claim 1, characterized in that, The reference sensing module includes at least one tilt sensor, which is rigidly mounted on the elevator equipment stabilization platform to measure the horizontal angle of the elevator equipment's working plane and provide an attitude reference for barometric altitude conversion.

5. The elevator leveling monitoring system integrating visual recognition and barometric pressure learning according to claim 1, characterized in that, The edge computing module includes at least one edge computing unit, deployed on the top of the elevator car or in the machine room, for running a self-learning air pressure algorithm, a visual air pressure fusion leveling engine, and fault detection logic.

6. The elevator leveling monitoring system integrating visual recognition and barometric pressure learning according to claim 1, characterized in that, The cloud management module deploys services in the cloud, including an MQTT Broker message queue, a device management unit, a monitoring and alarm unit, and a data analysis unit.

7. An elevator leveling monitoring method integrating visual recognition and barometric pressure learning according to claim 1, 2, 3, 4, 5, or 6, characterized in that, Specifically, the steps include the following: (1) The floor numbers displayed on the elevator panel are identified by the visual perception module, and the air pressure calibration of the floor is automatically triggered. The floor-air pressure mapping table, i.e. FP table, is established by combining the air pressure measurement module and the reference sensing module, so as to realize the self-learning construction of the floor-air pressure mapping table without human intervention. (2) The elevator panel image is analyzed by AI video analysis through the network AI camera in the visual perception module to obtain the visual leveling status VLS of the elevator. The floor-pressure mapping table is queried through the air pressure sensor in the air pressure measurement module to obtain the air pressure leveling status BLS. The running distance is obtained for consistency verification. The fused leveling confirmation is only output when the visual leveling status and the air pressure leveling status are consistent and the confidence index meets the standard. (3) After detecting the elevator stop command, the fault confidence index (FCI) is calculated by the edge computing module by combining VLS, BLS, elevator built-in encoder zero speed signal and door lock status. When FCI≥0.95 and the car position exceeds the unlocking area, the non-leveling elevator stop fault is confirmed, and the forced reporting mechanism is triggered. The triple mechanism of MQTT QoS 2, Retained message and Last Will is used to ensure that the platform reaches the cloud management module, and the fault freeze state is entered to prohibit the door from opening.

8. The elevator leveling monitoring method integrating visual recognition and barometric pressure learning according to claim 7, characterized in that, in In step (1), specifically: when the elevator is in the initial state, the network AI camera identifies the current elevator car on a certain floor on the elevator panel, the air pressure sensor confirms that the elevator is stationary and outputs the air pressure measurement value, and notifies the elevator equipment to calibrate the FP table; during the subsequent continuous operation of the elevator, the air pressure sensor confirms that the elevator is stationary each time and calculates the running distance based on the air pressure measurement value. Combined with the reference sensor module, it determines whether the elevator is at the level position. Combined with the network AI camera to identify the floor and panel direction on the elevator panel, the calibrated FP table is continuously self-learned and updated synchronously, so as to realize the self-learning construction of the floor table without human intervention.

9. The elevator leveling monitoring method integrating visual recognition and barometric pressure learning according to claim 7, characterized in that, In step (2), specifically: the visual leveling status (VLS) is obtained through AI video analysis. The visual leveling status is determined by whether there is an arrow on the elevator panel identified by the network AI camera. If there is an arrow, the elevator is running; if there is no arrow, the elevator is level or stationary. The air pressure leveling status (BLS) is obtained by querying the FP table through the air pressure sensor. That is, after the air pressure sensor calculates the running distance, it compares the air pressure at the leveling position in the floor-air pressure mapping table to determine whether the elevator is level. The consistency verification between the floor identified by the video and the stationary state of the elevator detected by the air pressure sensor is performed. Only when the dual-source states of VLS and BLS are consistent and the confidence index exceeds the set threshold is the final fused leveling confirmation output output to avoid misjudgment by a single sensor.

10. The elevator leveling monitoring method integrating visual recognition and barometric pressure learning according to claim 7, characterized in that, in In step (3), the cloud management module communicates with the edge computing module through the MQTT message queue to remotely configure FP table parameters, leveling tolerance threshold and sensor calibration instructions. The leveling tolerance threshold refers to whether the air pressure sensor is within the threshold after measuring the running distance and comparing it with the FP table. The sensor calibration instructions refer to the calibration and re-self-learning calibration triggered after the network AI camera recognizes the discrepancy between the air pressure sensor's calculation and the network AI camera's recognition. It also receives real-time equipment operating status, leveling results, and fault alarms, providing a visual monitoring interface and data traceability function.