Elevator alarm control methods, equipment, media and program products

Abstract

This application belongs to the field of big data processing, specifically relating to an elevator alarm control method, device, medium, and program product. It acquires initial alarm data within the elevator in real time. This initial alarm data is generated after a target is initially identified using a target detection algorithm. The initial alarm data is compared with a first false alarm feature library to determine whether it is a false alarm. The first false alarm feature library stores historical false alarm data generated by the corresponding elevator. If the initial alarm data is not a false alarm, it is determined to be the target alarm data, and the elevator is controlled to perform the corresponding alarm operation. This method improves the accuracy of elevator alarms, reduces invalid alarms, ensures that alarm operations are only performed on genuine anomalies, and improves the safety and reliability of elevator operation.

Description

Technical Field

[0001] This application relates to the field of big data processing, and in particular to an elevator alarm control method, device, medium and program product. Background Technology

[0002] As an indispensable vertical transportation tool in high-rise buildings and public places, the safe operation of elevators directly affects the lives and property of passengers. Current technology generally uses smart cameras to collect images or videos inside the elevator car in real time, and uses traditional deep learning models for object detection to identify key objects such as people and electric vehicles.

[0003] In existing technologies, elevator control methods typically rely on a fixed logical combination of "people inside the elevator + elevator doors closed + elevator stationary for more than a set time"; while the recognition of electric vehicles entering the elevator mainly relies on the model's matching and judgment of visual features such as the target's shape and outline.

[0004] However, this entrapment detection mechanism lacks contextual understanding of special work scenarios (such as cleaning and maintenance). Relying solely on static logic can easily misjudge normal work behaviors as entrapment events, leading to frequent false alarms and affecting the accuracy of elevator entrapment alarms. Summary of the Invention

[0005] This application provides elevator alarm control methods, devices, media, and program products to solve the problem of low accuracy in elevator entrapment alarms.

[0006] In a first aspect, embodiments of this application provide an elevator alarm control method, applied to an elevator monitoring system, the method comprising:

[0007] The system acquires initial alarm data inside the elevator in real time. The initial alarm data is the alarm data generated after the target is initially identified by the target detection algorithm.

[0008] The initial alarm data is compared with the first false alarm feature library to determine whether the initial alarm data is a false alarm data. The first false alarm feature library is used to store the false alarm data generated in the history of the corresponding elevator.

[0009] If the initial alarm data is not a false alarm, then the initial alarm data is determined to be the target alarm data, and the elevator is controlled to perform the corresponding alarm operation.

[0010] As an optional implementation, comparing the initial alarm data with a first false alarm feature database to determine whether the initial alarm data is a false alarm includes:

[0011] Extract target features from the initial alarm data, where the target features refer to the target visual attributes extracted by an image processing algorithm;

[0012] Calculate the similarity value between the target feature and the features stored in the first false positive feature library;

[0013] If the similarity value is greater than a preset threshold, the initial alarm data is determined to be false alarm data;

[0014] If the similarity value is less than or equal to the preset threshold, the initial alarm data is determined to be non-false alarm data, and the non-false alarm data is sent to the platform's duplicate false alarm review intelligent agent.

[0015] As an optional implementation, the control of the elevator to perform an alarm operation includes:

[0016] Identify the alarm category of the target alarm data;

[0017] If the alarm category is a direct alarm category, then the elevator alarm operation is executed;

[0018] If the alarm category is a delayed alarm category, then when the elevator door is closed and the elevator has been stationary for a period of time that reaches the first entrapment alarm duration, the number of targets inside the elevator is obtained.

[0019] When the target number exceeds a preset threshold, an elevator alarm is triggered.

[0020] When the target number is less than or equal to the preset number threshold, and the elevator remains stationary for a period of time that reaches the second entrapment alarm duration, an elevator alarm operation is performed, wherein the second entrapment alarm duration is longer than the first entrapment alarm duration.

[0021] As an optional implementation, it also includes:

[0022] If the initial alarm data is a false alarm, the elevator alarm operation will not be performed.

[0023] Secondly, embodiments of this application provide an elevator alarm control method, applied to a platform's repeated false alarm review intelligent agent, the method comprising:

[0024] Obtain non-false alarm data sent by the elevator monitoring system, wherein the non-false alarm data carries the elevator identifier of the corresponding elevator;

[0025] Based on the feature extraction network, features are extracted from the non-false alarm data to obtain non-false alarm feature data;

[0026] Based on the elevator alarm recognition model, target recognition is performed on the non-false alarm feature data, new false alarm feature data is selected, and the new false alarm feature data is stored in the second false alarm feature library.

[0027] Based on the elevator identifier corresponding to the new false alarm feature data, the false alarm feature data corresponding to the elevator identifier in the second false alarm feature library is sent to the elevator monitoring system to update the first false alarm feature library.

[0028] As an optional implementation, the method for constructing the feature extraction network includes:

[0029] Extract historical false alarm datasets from the second false alarm feature library;

[0030] The initial neural network model is trained using the historical false alarm dataset. The feature extraction network is constructed by minimizing the intra-class distance of false alarm features and maximizing the inter-class distance between false alarm features and other features.

[0031] As an optional implementation, the method for constructing the elevator alarm recognition model includes:

[0032] Collect historical real alarm data samples and historical false alarm data samples uploaded by all elevator monitoring systems;

[0033] The historical real alarm data samples are labeled with positive class labels, and the historical false alarm data samples are labeled with negative class labels;

[0034] The classification model is trained under supervision using the labeled dataset to obtain the elevator alarm recognition model.

[0035] Thirdly, embodiments of this application provide an elevator alarm control device applied to an elevator monitoring system, the method comprising:

[0036] The acquisition module is used to acquire the initial alarm data in the elevator in real time. The initial alarm data is the alarm data generated after the target is initially identified by the target detection algorithm.

[0037] The determination module is used to compare the initial alarm data with the first false alarm feature library to determine whether the initial alarm data is false alarm data. The first false alarm feature library is used to store the false alarm data generated in the history of the corresponding elevator.

[0038] The control module is used to determine that the initial alarm data is the target alarm data if the initial alarm data is not a false alarm, and to control the elevator to perform the corresponding alarm operation.

[0039] As an optional implementation, the elevator alarm control device further includes: an extraction module, a calculation module, and a transmission module;

[0040] The extraction module is used to extract target features from the initial alarm data, where the target features refer to the target visual attributes extracted by the image processing algorithm.

[0041] The calculation module is used to calculate the similarity value between the target feature and the features stored in the first false alarm feature library;

[0042] The determining module is further configured to determine the initial alarm data as false alarm data if the similarity value is greater than a preset threshold.

[0043] The sending module is used to determine that the initial alarm data is non-false alarm data if the similarity value is less than or equal to the preset threshold, and to send the non-false alarm data to the platform's duplicate false alarm review intelligent agent.

[0044] As an optional implementation, the elevator alarm control device further includes: an identification module and an execution module;

[0045] The identification module is used to identify the alarm category of the target alarm data;

[0046] The execution module is used to perform an elevator alarm operation if the alarm category is a direct alarm category.

[0047] The acquisition module is used to acquire the number of targets inside the elevator when the elevator door is closed and the elevator has been stationary for a period of time that reaches the first entrapment alarm duration, if the alarm type is a delayed alarm type.

[0048] The execution module is also used to execute an elevator alarm operation when the target number is greater than a preset number threshold.

[0049] The execution module is further configured to perform an elevator alarm operation when the target number is less than or equal to the preset number threshold and the elevator remains stationary for a period of time that reaches the second entrapment alarm duration, wherein the second entrapment alarm duration is longer than the first entrapment alarm duration.

[0050] As an optional implementation, the execution module is further configured to not perform elevator alarm operation if the initial alarm data is false alarm data.

[0051] Fourthly, this application provides an elevator alarm control device applied to a platform's repeated false alarm review intelligent agent, the method comprising:

[0052] The acquisition module is used to acquire non-false alarm data sent by the elevator monitoring system, wherein the non-false alarm data carries the elevator identifier of the corresponding elevator;

[0053] The extraction module is used to extract features from the non-false alarm data based on the feature extraction network to obtain non-false alarm feature data;

[0054] The storage module is used to perform target recognition on the non-false alarm feature data based on the elevator alarm recognition model, filter out new false alarm feature data, and store the new false alarm feature data in the second false alarm feature library.

[0055] The update module is used to send the false alarm feature data corresponding to the elevator identifier in the second false alarm feature library to the elevator monitoring system based on the elevator identifier corresponding to the new false alarm feature data, thereby updating the first false alarm feature library.

[0056] As an optional implementation, the elevator alarm control device further includes: a construction module;

[0057] The extraction module is also used to extract historical false alarm datasets from the second false alarm feature library;

[0058] The construction module is used to train the initial neural network model using the historical false alarm dataset, and to construct the feature extraction network by minimizing the intra-class distance of false alarm features and maximizing the inter-class distance between false alarm features and other features.

[0059] As an optional implementation, the elevator alarm control device further includes: a collection module, a labeling module, and a training module;

[0060] The collection module is used to collect historical real alarm data samples and historical false alarm data samples uploaded by all elevator monitoring systems;

[0061] The labeling module is used to label the historical real alarm data samples with positive class labels and the historical false alarm data samples with negative class labels;

[0062] The training module is used to supervise the training of the classification model using the labeled dataset to obtain the elevator alarm recognition model.

[0063] Fifthly, embodiments of this application provide an elevator alarm control device, including: a receiver, a transmitter, a memory, and a processor;

[0064] Receiver, used to receive instructions and data;

[0065] A transmitter is used to send commands and data;

[0066] The memory stores computer-executed instructions;

[0067] The processor executes computer execution instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0068] In a sixth aspect, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0069] In a seventh aspect, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0070] The elevator alarm control method provided in this application collects initial alarm data generated by a target detection algorithm in real time and compares it precisely with a first false alarm feature library storing historical false alarm data of the elevator. This effectively filters false alarm information caused by similar historical anomalies before alarm execution, reducing invalid alarms and false triggers from the source and avoiding resource waste and alarm fatigue caused by frequent false alarms. Only non-false alarm data that is confirmed as a real anomaly generates target alarm data and drives the elevator to execute the corresponding alarm operation, significantly improving the accuracy and effectiveness of elevator intelligent alarms. This ensures that abnormal events can be identified and responded to in a timely and accurate manner, thereby strengthening the safety protection capability and system reliability during elevator operation and providing more reliable technical protection for elevator passenger safety and stable operation. Attached Figure Description

[0071] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0072] Figure 1 Flowchart of the elevator alarm control method provided in this application Figure 1 ;

[0073] Figure 2 Flowchart of the elevator alarm control method provided in this application Figure 2 ;

[0074] Figure 3 Flowchart of the elevator alarm control method provided in this application Figure 3 ;

[0075] Figure 4 Schematic diagram of the elevator alarm control device provided in this application Figure 1 ;

[0076] Figure 5 Schematic diagram of the elevator alarm control device provided in this application Figure 2 ;

[0077] Figure 6 A schematic diagram of the elevator alarm control device provided in this application.

[0078] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0079] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions 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.

[0080] As a core vertical transportation tool in high-rise buildings and public places, the safe operation of elevators is directly related to the safety of people's lives and property.

[0081] In existing technologies, images or videos inside elevators are captured in real time using smart cameras, and traditional deep learning models are used for target detection to identify key targets such as people and vehicles. The determination of a person entrapment incident is usually based on a logical combination of "people inside the elevator + elevator doors closed + elevator stationary for more than a preset time", while the determination of an electric vehicle entering the elevator relies on the model's matching of features such as the target's shape and outline.

[0082] However, existing technology only uses a single logic of "person presence + elevator door closed + duration of stillness" to determine entrapment events, without considering special work scenarios such as cleaning staff, which leads to frequent false alarms triggered by routine work behaviors.

[0083] To address the aforementioned issues, the elevator alarm control method provided in this application identifies and generates initial alarm data within the elevator in real time using a target detection algorithm. This data is then compared with a first false alarm feature library storing historical false alarm data. False alarm data is filtered out, and only initial alarm data confirmed as genuine anomalies is identified as target alarm data, triggering the corresponding alarm operation. This improves the accuracy of elevator alarms, reduces the frequency of invalid alarms, ensures that alarm operations are only performed on genuine anomalies, and ultimately enhances the safety and reliability of elevator operation.

[0084] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with 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. The embodiments of this application will now be described with reference to the accompanying drawings.

[0085] Figure 1Flowchart of the elevator alarm control method provided in this application Figure 1 It is used in elevator monitoring systems. For example... Figure 1 As shown, the method includes:

[0086] S101: Real-time acquisition of initial alarm data inside the elevator. The initial alarm data is the data to be alarmed generated after the target is initially identified by the target detection algorithm.

[0087] Initial alarm data refers to the data that is automatically generated by monitoring equipment, sensors and other terminals inside the elevator, which use target detection algorithms to monitor the scene inside the elevator in real time. This data is generated before false alarm verification.

[0088] Specifically, firstly, devices with target detection capabilities (such as high-definition cameras, infrared sensors, and vibration sensors) are deployed in key locations such as inside the elevator car, on the car top, and at the bottom of the shaft. All devices establish real-time communication connections with the elevator monitoring system. Secondly, the devices have built-in preset target detection algorithms (such as the YOLO algorithm and CNN convolutional neural network algorithm) adapted to elevator scenarios, capable of identifying targets such as people lingering, abnormal door jamming, foreign objects obstructing the door, and abnormal shaking of equipment components. Finally, when the algorithm identifies the above targets, it collects relevant data (including target type, identification time, elevator location, equipment operating parameters, etc.), integrates this data to generate initial alarm data, and transmits it to the elevator alarm processing terminal in real time, ensuring that data acquisition is delayed and complete.

[0089] For example, in the elevator of Unit 2, Building 3 of a certain residential complex, high-definition cameras and infrared sensors are deployed inside the car, with a built-in YOLO target detection algorithm. The preset alarm target is "person trapped (person remains in the elevator for more than 10 minutes after it stops)". At 3:15 pm one day, the elevator stopped between the 5th and 6th floors due to a malfunction. There was one passenger in the car. The camera and infrared sensor captured the passenger's figure in real time. The algorithm initially identified the target as "person remaining in the elevator" and collected data such as the passenger's stay time (initially 10 seconds), the elevator's current position (between the 5th and 6th floors), and the elevator's operating status (stopped running, doors closed).

[0090] S102: Compare the initial alarm data with the first false alarm feature library to determine whether the initial alarm data is false alarm data. The first false alarm feature library is used to store the false alarm data generated in the history of the corresponding elevator.

[0091] The first false alarm feature database is a database used to store all false alarm data that has been generated during the elevator's historical operation. The false alarm data stored in it needs to be manually verified and recorded by the system, and includes key information such as false alarm scenarios, false alarm data characteristics, and false alarm triggering reasons.

[0092] Specifically, firstly, the system reviews past false alarm cases of the elevator, extracting and organizing the initial alarm data for each false alarm (including target type, trigger time, elevator status, environmental parameters, etc.), and labeling the reasons for the false alarms (e.g., camera obstructed by dust causing algorithm misjudgment, infrared sensor misidentification due to strong light interference, misjudgment due to brief personnel stay during temporary elevator maintenance, etc.). A first false alarm feature library is established and updated regularly (adding false alarm cases and deleting invalid false alarm data). Secondly, upon receiving the transmitted initial alarm data, the system automatically extracts the core features of the data (e.g., target type, elevator location, trigger conditions, etc.) and compares them one by one with the features of all false alarm data in the first false alarm feature library, setting reasonable comparison thresholds (e.g., a feature matching degree of 85% or higher is considered a suspected false alarm). Finally, based on the comparison results, the system determines whether the initial alarm data is a false alarm: if the feature matching degree reaches the threshold, it is determined to be false alarm data; if it does not reach the threshold, it is determined to be non-false alarm data.

[0093] For example, the first false alarm feature database for the elevator in Unit 2, Building 3 of the residential complex stores historical false alarm data such as "the camera was obstructed by dust, misidentifying debris in the elevator as a person, triggering an initial alarm for a trapped person." The characteristics of this false alarm data are "the elevator was running normally, no one was present, the camera image was blurry, and the target outline in the initial alarm data was unclear." When the system receives the generated initial alarm data, it extracts its core features: the elevator has stopped running, the target outline is clear (clearly a person), the camera image is normal, and the timer for the person remaining in the elevator is increasing. These features are compared one by one with the false alarm data in the first false alarm feature database. The feature matching degree is found to be only 10%, which does not reach the preset 85% comparison threshold. Therefore, the initial alarm data is determined to be non-false alarm data, and the process proceeds to the next step. If the initial alarm data is generated because the camera was obstructed by dust, mistakenly identifying a backpack in the elevator car as a person, and its features match the "camera obstruction misjudgment" feature in the false alarm feature database with a matching degree of 90%, then it is determined to be false alarm data.

[0094] S103: If the initial alarm data is not a false alarm, then determine the initial alarm data as the target alarm data and control the elevator to perform the corresponding alarm operation.

[0095] Among them, target alarm data refers to alarm data that, after false alarm verification, confirms that there are no false alarms and truly reflects abnormal scenarios inside the elevator (such as people being trapped, equipment failure, etc.). It is the core basis for triggering elevator alarm operations and has authenticity, effectiveness, and urgency.

[0096] Specifically, firstly, the elevator monitoring system receives a judgment result. If it confirms that the initial alarm data is not a false alarm, it marks the initial alarm data as the target alarm data and extracts information about the target alarm data (such as alarm type, elevator location, abnormal scene details, etc.). Secondly, it triggers a preset alarm procedure.

[0097] For example, after the elevator monitoring system determines that the initial alarm data is not a false alarm, it marks it as the target alarm data and extracts key information: the alarm type is "person trapped," the elevator location is between the 5th and 6th floors of Unit 2, Building 3, the number of trapped persons is 1, and the alarm time is 3:15 PM. Subsequently, the elevator monitoring system executes the corresponding alarm operations: it controls the emergency broadcast in the elevator car to automatically play a reassuring voice saying, "The elevator has temporarily malfunctioned. Please do not panic. We have notified staff to come to the rescue. Please do not forcibly pry open the doors." At the same time, it automatically dials the emergency phone number of the community property management (and broadcasts the alarm information simultaneously after the call is connected) and the phone number of the elevator maintenance company; it sends the alarm details to the mobile phones of the property manager and maintenance personnel via SMS, and stops the alarm operation after the maintenance personnel arrive at the scene, rescue the trapped passenger, and the abnormal scenario is resolved.

[0098] The elevator alarm control method provided in this application acquires initial alarm data inside the elevator in real time. The initial alarm data is the data to be alarmed generated after the target is initially identified by the target detection algorithm. The initial alarm data is compared with a first false alarm feature library to determine whether the initial alarm data is false alarm data. The first false alarm feature library is used to store false alarm data generated in the history of the corresponding elevator. If the initial alarm data is not false alarm data, it is determined that the initial alarm data is the target alarm data, and the elevator is controlled to perform the corresponding alarm operation. This method improves the accuracy of elevator alarms, reduces invalid alarms, ensures that alarm operations are only performed on real anomalies, and improves the safety and reliability of elevator operation.

[0099] Figure 2 Flowchart of the elevator alarm control method provided in this application Figure 2 ,like Figure 2 As shown, in this embodiment... Figure 1 Based on the embodiments, the elevator alarm control method is described in detail, which includes:

[0100] S201: Real-time acquisition of initial alarm data inside the elevator. The initial alarm data is the data to be alarmed generated after the target is initially identified by the target detection algorithm.

[0101] Step S201 is similar to step S101, and will not be described again here.

[0102] S202: Extract target features from the initial alarm data. Target features refer to the visual attributes of the target extracted through image processing algorithms.

[0103] Among them, target features refer to the visual attributes of the target obtained after analyzing and extracting image-type information from the initial alarm data through image processing algorithms. These include quantifiable and comparable visual information such as the target's outline, size, color, texture, and posture.

[0104] Specifically, after receiving the initial alarm data, the elevator monitoring system automatically filters out the image data (such as the image inside the elevator car captured by a camera); it calls the preset image processing algorithm (such as edge detection algorithm and feature point extraction algorithm) to preprocess the image data (denoise removal and enhancement), and then extracts the core visual features of the target, including the target outline shape, size, color distribution, texture details, etc.; the extracted target features are quantized and converted into feature vectors that can be used for calculation.

[0105] For example, after receiving the initial alarm data of the elevator, the elevator monitoring system extracts the images of the car interior captured by the camera, performs noise reduction on the images using image processing algorithms to remove minor interference in the image; then it extracts the visual features of the "person" target in the image, including the person's outline (human outline), size (height approximately 1.7 meters), color (black shirt, blue pants), and posture (standing position), and converts these features into feature vectors.

[0106] S203: Calculate the similarity value between the target feature and the features stored in the first false alarm feature library.

[0107] The similarity value refers to the degree of matching between the target features extracted from the initial alarm data and the historical false alarm data features stored in the first false alarm feature library, and is represented by a value between 0 and 1 (the closer the value is to 1, the higher the matching degree).

[0108] The first false alarm feature database is a database that stores the historical false alarm data features of the elevator (such as the visual features of misjudged debris and interference).

[0109] Specifically, the system calls the target feature vector stored in the temporary database and retrieves all historical false alarm feature vectors in the first false alarm feature library. Using a preset similarity calculation algorithm (such as Euclidean distance algorithm or cosine similarity algorithm), the system calculates the similarity between the target feature vector and each false alarm feature vector in the first false alarm feature library one by one, obtaining multiple similarity values. The highest similarity value is selected as the final similarity value between the target feature and the false alarm feature.

[0110] The elevator's first false alarm feature database stores false alarm feature vectors caused by the elevator monitoring system's inaccurate target detection and weak ability to distinguish between cleaning staff and ordinary people (such as features that misidentify cleaning staff as trapped ordinary people, including features such as the cleaning staff's outline, clothing, and posture), as well as false alarm feature vectors caused by the elevator monitoring system's weak ability to distinguish between bicycles and electric vehicles (such as features that misidentify bicycles as electric vehicles, including features such as the bicycle's outline, size, and number of wheels).

[0111] For example, the system calls to extract the target feature vector of "personnel". The false alarm feature vector related to the misjudgment of cleaning personnel in the database is calculated using the cosine similarity algorithm, and the similarity value is 0.1. At the same time, it is calculated one by one with other false alarm feature vectors such as bicycles being misjudged as electric vehicles in the database, and the similarity value is 0.2.

[0112] S204: If the similarity value is greater than the preset threshold, the initial alarm data is determined to be false alarm data.

[0113] The preset threshold is a critical value (usually set to 0.85-0.9) that is pre-set by the system to determine whether the target feature matches the false alarm feature.

[0114] False alarm data refers to data generated in the initial alarm data that does not actually exist due to algorithm misjudgment or other reasons, and therefore does not require triggering an alarm operation.

[0115] Specifically, the system obtains the similarity value and compares it with a preset threshold. If the similarity value is greater than the preset threshold, it means that the target features of the current initial alarm data are highly matched with the historical false alarm features in the first false alarm feature library, and the initial alarm data can be determined to be false alarm data.

[0116] For example, the extracted target features of the cleaning staff have a similarity of 0.92 with the features of "misjudgment of cleaning staff" in the first false alarm feature library, while the system's preset threshold is 0.85; since 0.92 > 0.85, the system determines that the initial alarm data is false alarm data.

[0117] S205: If the similarity value is less than or equal to the preset threshold, the initial alarm data is determined to be non-false alarm data, and the non-false alarm data is sent to the platform's duplicate false alarm review intelligent agent.

[0118] Among them, non-false alarm data refers to the initial alarm data that, after comparison of similarity values, confirms that the target features and false alarm features have a low degree of matching and truly reflect the abnormal scene in the elevator, thus possessing authenticity and validity.

[0119] The platform's duplicate false alarm review agent is used to conduct secondary review of non-false alarm data, investigate duplicate alarms, and confirm the authenticity of alarms.

[0120] Specifically, the elevator monitoring system compares the obtained similarity value with a preset threshold. If the similarity value is less than or equal to the preset threshold, the initial alarm data is determined to be non-false alarm data. The system marks the data as non-false alarm and integrates the core information of the data (target features, elevator status, alarm time, etc.). According to the preset communication protocol, the non-false alarm data is sent to the platform's repeat false alarm review intelligent agent in real time.

[0121] For example, if the similarity between the target feature (personnel) and the false alarm feature calculated by the system is 0.1 and the preset threshold is 0.85, since 0.1≤0.85, the system determines that the initial alarm data is not a false alarm data; then it integrates the core information of the data (personnel features, elevator stopped between the 5th and 6th floors, alarm time 3:15 pm) and sends it to the repeated false alarm review intelligent agent of the community elevator management platform according to the communication protocol.

[0122] S206: If the initial alarm data is not a false alarm, then the initial alarm data is determined to be the target alarm data.

[0123] Specifically, if the initial alarm data is confirmed to be non-false alarm data and there are no repeated alarms, then the non-false alarm data is marked as target alarm data; at the same time, the key information of the target alarm data (alarm type, elevator location, number of trapped people, abnormal duration, etc.) is extracted.

[0124] For example, if the initial alarm data is confirmed to be non-false alarm data and there is no duplicate alarm, the system marks the data as target alarm data; extract the key information: the alarm scenario is "person trapped", the elevator location is between the 5th and 6th floors of Unit 2, Building 3, the number of trapped persons is 1, and the alarm time is 3:15 pm.

[0125] S207: Identify the alarm category of the target alarm data.

[0126] The alarm category is a classification of alarm data based on the type of abnormal scenario reflected by the target alarm data, divided into direct alarm category and delayed alarm category.

[0127] The direct alarm category refers to scenarios where an alarm operation is performed (such as a violently shaking elevator or a person being trapped due to a sudden illness).

[0128] The delayed alarm category refers to scenarios where an alarm operation can be performed after a certain period of time (such as normal personnel staying in the elevator or temporary elevator malfunction).

[0129] Specifically, the elevator monitoring system retrieves target alarm data from the target alarm database, extracts key information such as details of abnormal scenarios and elevator operating status; it calls the preset alarm category recognition algorithm, combined with pre-set category determination rules (such as "severe elevator shaking, sudden illness of personnel" is determined as a direct alarm category, and "personnel stranded, elevator temporarily stopped" is determined as a delayed alarm category), to classify and identify the target alarm data; after the identification is completed, the specific category of the target alarm data is marked.

[0130] For example, the elevator monitoring system extracts key information from the target alarm data: the elevator stops between the 5th and 6th floors due to a malfunction, there is one passenger in the car, there is no violent shaking, no sudden illness or other emergency, which meets the judgment rule of "personnel stranded, elevator temporarily stopped"; through the alarm category recognition algorithm, the alarm category of the target alarm data is determined to be a delayed alarm category.

[0131] S208: If the alarm category is a direct alarm category, then perform the elevator alarm operation.

[0132] Specifically, when the elevator monitoring system receives the identified alarm category, if it is determined to be a direct alarm category, it triggers the preset direct alarm procedure; it controls the emergency broadcast in the elevator car to play an emergency reassurance voice, and at the same time automatically dials the property emergency number and the elevator maintenance unit's number, and simultaneously sends the alarm details (elevator location, anomaly type, alarm time) to the mobile phone of the relevant person in charge; it controls the elevator to stop running (to prevent the fault from escalating), turns on the emergency lighting in the car, and continuously collects real-time data in the elevator until the anomaly is resolved.

[0133] For example, if the elevator shakes violently and passengers in the car suddenly feel dizzy and unwell, the target alarm data is identified as a direct alarm category; the elevator monitoring system executes the alarm operation: the emergency broadcast plays "The elevator has experienced an emergency abnormality, please remain calm, rescue personnel have set off", the elevator location (between the 5th and 6th floors of Unit 2, Building 3) and the abnormal situation (elevator shaking, passengers feeling unwell) are sent to the person in charge's mobile phone, the emergency lighting is turned on, and the situation inside the car is continuously monitored until the rescue personnel arrive.

[0134] S209: If the alarm type is a delayed alarm type, then when the elevator door is closed and the elevator has been stationary for a period of time that reaches the first entrapment alarm duration, the number of targets inside the elevator is obtained.

[0135] Specifically, after confirming that the alarm category is a delayed alarm category, the elevator monitoring system monitors the elevator door status (closed / open) and elevator operation status (stationary / running) in real time. When the elevator door is detected to be completely closed and the elevator stationary time accumulates to the first entrapment alarm duration, a target quantity acquisition command is triggered. The system controls the cameras, infrared sensors, and other equipment inside the elevator to conduct a comprehensive inspection of the car and count the number of targets inside the elevator (with a focus on counting the number of people) through a target detection algorithm.

[0136] For example, if the target alarm data is a delayed alarm, the elevator monitoring system presets the first entrapment alarm duration to 10 minutes. The elevator monitoring system monitors the elevator status in real time, confirms that the elevator doors are closed, and starts timing from when the elevator stops running at 3:15 pm. When the stationary time reaches 10 minutes (3:25 pm), the elevator monitoring system triggers the target quantity acquisition command. It controls the camera and infrared sensor in the car to detect inside the car and uses an algorithm to count the target quantity inside the elevator as 1 person.

[0137] S210: When the target quantity exceeds the preset quantity threshold, execute the elevator alarm operation.

[0138] Specifically, the preset number threshold is a critical value set in advance by the elevator monitoring system to determine whether an alarm operation needs to be executed in a delayed alarm scenario (usually set to 1 person, which can be adjusted according to the elevator usage scenario); when the target number exceeds this threshold, it indicates that there are many trapped people and the risk is high, so an alarm operation needs to be executed.

[0139] For example, if the elevator monitoring system has a preset threshold of 1 person, and assuming that the target number of people in the elevator at 3:25 pm is 2 (2 passengers are trapped), since 2 people > 1 person, the elevator monitoring system will perform the elevator alarm operation: control the emergency broadcast to play a reassuring voice, automatically dial the property emergency phone and maintenance phone, send information such as the elevator location and the number of people trapped (2 people) to the person in charge's mobile phone, turn on the emergency lights, continuously monitor the situation inside the car, and wait for the rescue personnel to arrive.

[0140] S211: When the target number is less than or equal to the preset number threshold, and the elevator remains stationary for a period of time that reaches the second entrapment alarm duration, an elevator alarm operation is executed. The second entrapment alarm duration is longer than the first entrapment alarm duration.

[0141] The second entrapment alarm duration is a pre-set alarm threshold duration for the elevator monitoring system in the delayed alarm category when the "target number is less than or equal to the preset number threshold". This duration is longer than the first entrapment alarm duration (the second entrapment alarm duration is usually set to 5-10 minutes).

[0142] Specifically, the system compares the target number with a preset threshold. If the target number is less than or equal to the preset threshold, it continues to monitor the elevator's stationary time in real time, accumulating the time from when the elevator stops running. When the accumulated stationary time reaches the second entrapment alarm time, it indicates that the abnormal situation has not been alleviated. Even if the number of trapped people is small, an alarm operation must be performed. The system triggers the alarm process and continues to collect real-time data inside the elevator until the rescue is completed.

[0143] For example, the elevator monitoring system has a preset threshold of 1 person and a second entrapment alarm duration of 10 minutes. If the target number of people in the elevator is 1 at 3:25 pm (1 passenger trapped), the system continues to accumulate the elevator's stationary time. When the stationary time reaches 10 minutes (3:35 pm), the system determines that the abnormality has not been relieved and executes the elevator alarm operation: playing a reassuring voice, calling the property management and maintenance numbers, sending alarm details, turning on emergency lighting, and continuously monitoring the status of passengers in the car until maintenance personnel arrive to rescue them.

[0144] S212: If the initial alarm data is a false alarm, the elevator alarm operation will not be performed.

[0145] Specifically, after the elevator monitoring system confirms that the initial alarm data is a false alarm, it terminates all processes related to the alarm operation and does not trigger any alarm action; at the same time, it records the detailed information of the false alarm data (false alarm time, false alarm reason, target characteristics, similarity value, etc.), and the false alarm reason is highlighted as insufficient accuracy of camera target detection, weak ability to distinguish between cleaning staff and ordinary people, or weak ability to distinguish between bicycles and electric vehicles, and the false alarm characteristics are updated to the first false alarm characteristic database.

[0146] For example, if the initial alarm data for the elevator is generated because the camera's target detection is inaccurate and its ability to distinguish between bicycles and electric vehicles is weak, mistakenly identifying a bicycle entering the elevator as an electric vehicle, and this is determined to be false alarm data, the elevator monitoring system will immediately terminate the alarm-related process, without playing reassuring voice messages or making any emergency calls. It will only record the false alarm information (false alarm time 3:15 PM, reason for false alarm: inaccurate camera target detection, weak ability to distinguish between bicycles and electric vehicles, similarity 0.92), update the target features of the bicycle to the first false alarm feature database, and send a prompt to the property management platform to remind staff to optimize the camera's target detection algorithm to improve the recognition accuracy of bicycles and electric vehicles and avoid further false alarms.

[0147] The elevator alarm control method provided in this application first acquires initial alarm data generated by a target detection algorithm in real time. Then, it extracts target visual attribute features from this data using an image processing algorithm. The similarity score between this feature and features stored in a first false alarm feature library is calculated. Based on the comparison between the score and a preset threshold, false alarm data and non-false alarm data are distinguished. Non-false alarm data is pushed to a platform's repeated false alarm review agent for further confirmation. Once confirmed as target alarm data, its alarm category is identified. For direct alarm categories, elevator alarm operation is executed immediately. For delayed alarm categories, the elevator door is first closed or stationary. When the stationary duration reaches the first entrapment alarm duration, the number of targets inside the elevator is detected. If the number is greater than a preset threshold, an alarm is triggered directly. If the number is less than or equal to the preset threshold, the alarm operation is executed only after the stationary duration reaches a longer second entrapment alarm duration. False alarm data does not trigger an alarm. The overall method effectively improves the accuracy of elevator alarms, reduces invalid alarms, and only alarms for genuine anomalies, thereby improving the safety and reliability of elevator operation.

[0148] Figure 3 Flowchart of the elevator alarm control method provided in this application Figure 3 ,like Figure 3 As shown, the elevator alarm control method is described in detail for the intelligent agent applied to the platform for reviewing repeated false alarms. The method includes:

[0149] S301: Obtain non-false alarm data sent by the elevator monitoring system. The non-false alarm data carries the elevator identifier of the corresponding elevator.

[0150] Elevator identification refers to the unique identification information used to distinguish each elevator, such as elevator ID, building unit number, etc.

[0151] Specifically, a stable connection is established with all elevator monitoring systems via the MQTT communication protocol; the real-time monitoring port automatically parses the message content when it detects a data message sent by the elevator monitoring system, separating the non-false alarm data body (including alarm image, target features, alarm time) and elevator identification; after parsing, the data integrity is verified, and if the verification passes, the non-false alarm data is bound to the elevator identification.

[0152] For example, the elevator monitoring system (smart camera) of Unit 2, Building 3 in a certain residential community initially identified one person in the car and generated initial alarm data due to insufficient target detection accuracy. After comparing it with the false alarm feature database for cleaning staff, the similarity was 0.1 (≤0.85), and it was determined to be non-false alarm data. The monitoring terminal bound this non-false alarm data (including the image of the person in the car, the target feature vector, and the alarm time 15:15) with the elevator identifier "Unit 2, Building 3-01" and sent it to the community elevator management platform via the MQTT protocol. After detecting the message, the binding information was parsed and the integrity verification was completed.

[0153] S302: Extract features from non-false alarm data using a feature extraction network to obtain non-false alarm feature data.

[0154] Among them, the feature extraction network refers to a neural network model customized for elevator alarm scenarios. Its core function is to extract highly discriminative deep visual features from image information that is not a false alarm, thereby compensating for the insufficient target detection accuracy of the elevator monitoring camera.

[0155] Non-false alarm feature data refers to the high-dimensional feature vectors generated by quantization after being extracted by this network, which can more accurately represent the core attributes of the alarm target.

[0156] Specifically, non-false alarm data is retrieved, and alarm target images are selected and preprocessed (normalized and size-uniformed). The preprocessed images are then input into a pre-trained feature extraction network, which is optimized for elevator false alarm scenarios and can accurately capture subtle visual differences in targets (such as cleaning staff's tools, and the structural features of bicycles and electric vehicles). The network outputs high-dimensional feature vectors through convolution, pooling, and other layer operations, which are then labeled as non-false alarm feature data.

[0157] For example, the platform system retrieves non-false alarm data for elevator "Building 3, Unit 2-01" from a temporary buffer, filters out the alarm image of one person inside the car, preprocesses it into a uniform-sized feature map, and then inputs it into a feature extraction network. This network compensates for the camera's weak ability to recognize people, extracting not only the person's outline and clothing features, but also capturing deeper features such as not carrying cleaning tools and standing in a flustered posture, ultimately generating 128-dimensional non-false alarm feature data. The system then binds this feature data to the elevator identifier "Building 3, Unit 2-01".

[0158] Optional methods for constructing feature extraction networks include:

[0159] Extract historical false alarm datasets from the second false alarm feature library;

[0160] The initial neural network model was trained using a historical false alarm dataset. A feature extraction network was constructed by minimizing the intra-class distance of false alarm features and maximizing the inter-class distance between false alarm features and other features.

[0161] The second false alarm feature library refers to the database in which the platform's repeated false alarm review agent centrally stores all historical new false alarm feature data of elevators.

[0162] Specifically, firstly, historical false alarm datasets are extracted in batches from the platform's second false alarm feature library and classified and labeled according to false alarm type (misjudgment by cleaning staff, confusion between bicycles and electric vehicles, etc.); then, ResNet18 is selected as the initial neural network model and trained based on the historical false alarm dataset. During training, the core objective of the loss function is to "minimize intra-class distance and maximize inter-class distance". By iteratively optimizing the model parameters, the model learns the consistency of features of the same false alarm category and the differences of features of different categories; after training, the classification layer is stripped, and only the backbone network is retained, thus obtaining the feature extraction network suitable for elevator alarm scenarios.

[0163] For example, when building the feature extraction network, the platform extracted a historical false alarm dataset from the second false alarm feature library, covering nearly six months. This dataset contains two core data categories: "cleaners misclassified as trapped individuals" and "bicycles misclassified as electric vehicles," totaling 1000 samples, which were then labeled by category. ResNet18 was selected as the initial model and trained on this dataset. A loss function was used to gradually reduce the intra-class distance of features for the "cleaners" category (e.g., the feature similarity between different cleaners approached 0.9), while simultaneously increasing the inter-class distance between the "cleaners" and "bicycle misclassification" categories (similarity decreased to below 0.1). After 50 rounds of iterative training, the classification layer was removed, resulting in a customized feature extraction network.

[0164] S303: Based on the elevator alarm recognition model, target recognition is performed on non-false alarm feature data, new false alarm feature data is selected, and the new false alarm feature data is stored in the second false alarm feature library.

[0165] Among them, the elevator alarm recognition model refers to a supervised learning model built on a binary classification task, which is used to accurately identify non-false alarm feature data and distinguish between real alarm features and "hidden false alarm" features caused by insufficient camera recognition accuracy.

[0166] New false alarm feature data refers to feature vectors that are identified as false alarms by the model and are not stored in the second false alarm feature library. It is the core data for realizing the dynamic updating of the feature library.

[0167] Specifically, the platform's repeated false alarm review agent retrieves non-false alarm feature data bound to elevator identifiers from a dedicated feature data database and inputs it into the pre-trained elevator alarm recognition model. The model performs binary classification on the feature data according to a preset classification threshold and outputs the recognition result of "real alarm" or "false alarm". If it is determined to be a false alarm, the feature data is marked as new false alarm feature data, and the same feature is checked against the second false alarm feature library. If it does not exist, it is bound to the elevator identifier and false alarm type and written into the second false alarm feature library to complete the data storage.

[0168] For example, the platform's repeated false alarm review agent inputs the non-false alarm feature data of elevator "Building 3, Unit 2-01" into the elevator alarm recognition model. Assuming the person is actually a newly hired cleaning staff member, and because they are not wearing cleaning tools, the monitoring end initially does not identify it as a false alarm, but the model captures deeper features corresponding to subtle characteristics such as their cleaning gloves and cleaning wipes in their pockets. It determines this feature data as a "cleaning staff misjudgment" type false alarm, and after checking the second false alarm feature library, confirms that this feature is new data. The system marks it as new false alarm feature data, binds it to the elevator identifier "Building 3, Unit 2-01" and the false alarm type "cleaning staff misjudgment," and writes it into the second false alarm feature library.

[0169] Optional methods for constructing elevator alarm recognition models include:

[0170] Collect historical real alarm data samples and historical false alarm data samples uploaded by all elevator monitoring systems;

[0171] Historical real alarm data samples are labeled with positive class labels, and historical false alarm data samples are labeled with negative class labels;

[0172] The classification model was trained under supervision using the labeled dataset to obtain the elevator alarm recognition model.

[0173] Among them, the historical real alarm data sample refers to the elevator real abnormal scene data collected by the platform and verified by manual review (such as real people being trapped, real electric vehicles entering the elevator).

[0174] Historical false alarm data samples refer to various historical false alarm data collected by the platform due to insufficient camera recognition accuracy (such as cleaning staff misjudging or bicycles being misjudged as electric vehicles).

[0175] Positive and negative labels are classification identifiers used to label two types of samples. Positive labels correspond to real alarms, while negative labels correspond to false alarms, providing a basis for supervised training.

[0176] Specifically, firstly, all historical data uploaded by elevator monitoring terminals are collected. After manual review, historical real alarm data samples and historical false alarm data samples are selected to ensure a balance between the two types of samples. Then, historical real alarm data samples are uniformly labeled with positive class labels (such as "1"), and historical false alarm data samples are uniformly labeled with negative class labels (such as "0") to construct a labeled classification dataset, which is then divided into training and test sets in a 7:3 ratio. Finally, logistic regression or lightweight CNN is selected as the basic classification model, and supervised training is performed using the training set. The model accuracy is verified using the test set. When the accuracy reaches 95% or higher, the elevator alarm recognition model is completed.

[0177] For example, when the platform's intelligent agent for reviewing repeated false alarms was building an elevator alarm recognition model, it collected nearly a year's worth of uploaded data from 50 elevators in the jurisdiction. After manual review, 500 sets of historical real alarm data samples (such as people actually trapped and electric vehicles actually entering the elevator) and 500 sets of historical false alarm data samples (all due to insufficient camera recognition accuracy, including 300 sets of misjudgments by cleaning staff and 200 sets of misjudgments of bicycles as electric vehicles) were selected. Real alarm samples were labeled with a positive class label "1", and false alarm samples were labeled with a negative class label "0". A classification dataset was constructed and divided into training and test sets. A lightweight CNN was selected as the base model and used for supervised training on the training set. After 30 iterations, the accuracy on the test set reached 96%, meeting the usage requirements, thus completing the construction of the elevator alarm recognition model.

[0178] S304: Based on the elevator identifier corresponding to the new false alarm feature data, send the false alarm feature data corresponding to the elevator identifier in the second false alarm feature library to the elevator monitoring system to update the first false alarm feature library.

[0179] The first false alarm feature database refers to the false alarm feature database specifically for that elevator, which is deployed locally on the elevator monitoring terminal (smart camera).

[0180] Specifically, the platform's duplicate false alarm review agent filters out all false alarm feature data (including newly added and historically stored data) corresponding to the elevator identifier bound to the new false alarm feature data from the second false alarm feature library; through the MQTT message communication interface, it encapsulates these feature data into a delivery message in a preset format and sends it to the monitoring terminal of the corresponding elevator; after receiving the message, the monitoring terminal automatically parses out the feature data, overwrites or appends it to the local first false alarm feature library, completes the dynamic update of the library, and ensures that subsequent local preliminary comparison can accurately identify similar false alarms.

[0181] For example, the platform's intelligent agent for reviewing repeated false alarms filters all false alarm feature data corresponding to the elevator identifier "Building 3, Unit 2-01" from the second false alarm feature library. This includes the features of the newly hired cleaning staff and historically stored features of bicycles mistakenly identified as electric bikes. The system encapsulates this feature data into a message and sends it to the elevator's monitoring terminal via the MQTT interface. Upon receiving the message, the monitoring terminal parses the feature data and appends it to its local first false alarm feature library, completing the library update. Subsequently, when the monitoring terminal detects the newly hired cleaning staff again, it can accurately determine it as a false alarm through local comparison, effectively compensating for the camera's weak ability to identify cleaning staff.

[0182] The elevator alarm control method provided in this application embodiment acquires non-false alarm data sent by the elevator monitoring system, which carries the corresponding elevator identifier. It then uses a feature extraction network to extract features from this data to obtain non-false alarm feature data. Based on an elevator alarm recognition model, it performs target recognition on the non-false alarm feature data, filters out new false alarm feature data, and stores it in a second false alarm feature library. Subsequently, based on the elevator identifier corresponding to the new false alarm feature data, it sends the false alarm feature data matching the elevator identifier in the second false alarm feature library back to the elevator monitoring system, thereby updating the first false alarm feature library. This effectively improves the accuracy of elevator alarms, reduces invalid alarms, and only executes alarms for genuine abnormal situations, enhancing the safety and reliability of elevator operation.

[0183] Figure 4 Schematic diagram of the elevator alarm control device provided in this application Figure 1 ,like Figure 4 As shown, the elevator alarm control device 400 provided in this embodiment, applied to an elevator monitoring system, includes:

[0184] The acquisition module 401 is used to acquire the initial alarm data in the elevator in real time. The initial alarm data is the alarm data generated after the target is initially identified by the target detection algorithm.

[0185] The determination module 402 is used to compare the initial alarm data with the first false alarm feature library to determine whether the initial alarm data is false alarm data. The first false alarm feature library is used to store the false alarm data generated in the history of the corresponding elevator.

[0186] The control module 403 is used to determine the initial alarm data as the target alarm data if the initial alarm data is not a false alarm, and to control the elevator to perform the corresponding alarm operation.

[0187] As an optional implementation, the elevator alarm control device further includes: an extraction module 404, a calculation module 405, and a sending module 406;

[0188] Extraction module 404 is used to extract target features from the initial alarm data. Target features refer to the target visual attributes extracted by the image processing algorithm.

[0189] The calculation module 405 is used to calculate the similarity value between the target feature and the features stored in the first false alarm feature library;

[0190] The determination module 402 is also used to determine that the initial alarm data is false alarm data if the similarity value is greater than a preset threshold.

[0191] The sending module 406 is used to determine that the initial alarm data is non-false alarm data if the similarity value is less than or equal to a preset threshold, and to send the non-false alarm data to the platform's repeated false alarm review intelligent agent.

[0192] As an optional implementation, the elevator alarm control device further includes: an identification module 407 and an execution module 408;

[0193] The identification module 407 is used to identify the alarm category of the target alarm data;

[0194] Execution module 408 is used to execute elevator alarm operation if the alarm type is a direct alarm type;

[0195] The acquisition module 401 is used to acquire the number of targets inside the elevator when the elevator door is closed and the elevator has been stationary for a period of time that reaches the first entrapment alarm duration, if the alarm type is a delayed alarm type.

[0196] The execution module 408 is also used to execute an elevator alarm operation when the target number exceeds a preset number threshold;

[0197] The execution module 408 is also used to perform an elevator alarm operation when the target number is less than or equal to a preset number threshold and the elevator remains stationary for a period of time that reaches the second entrapment alarm duration. The second entrapment alarm duration is longer than the first entrapment alarm duration.

[0198] As an optional implementation, the execution module 408 is also configured to not perform elevator alarm operation if the initial alarm data is false alarm data.

[0199] Figure 5 Schematic diagram of the elevator alarm control device provided in this application Figure 2 ,like Figure 5 As shown, the elevator alarm control device 500 provided in this embodiment, which is applied to the intelligent agent for reviewing repeated false alarms on the platform, includes:

[0200] The acquisition module 501 is used to acquire non-false alarm data sent by the elevator monitoring system. The non-false alarm data carries the elevator identifier of the corresponding elevator.

[0201] Extraction module 502 is used to extract features from non-false alarm data based on feature extraction network to obtain non-false alarm feature data;

[0202] Storage module 503 is used to identify targets from non-false alarm feature data based on the elevator alarm recognition model, filter out new false alarm feature data, and store the new false alarm feature data in the second false alarm feature library.

[0203] The update module 504 is used to send the false alarm feature data corresponding to the elevator identifier in the second false alarm feature library to the elevator monitoring system based on the elevator identifier corresponding to the new false alarm feature data, and update the first false alarm feature library.

[0204] As an optional implementation, the elevator alarm control device further includes: a construction module 505;

[0205] The extraction module 502 is also used to extract historical false alarm datasets from the second false alarm feature library;

[0206] Module 505 is used to train the initial neural network model using a historical false alarm dataset. It constructs a feature extraction network by minimizing the intra-class distance of false alarm features and maximizing the inter-class distance between false alarm features and other features.

[0207] As an optional implementation, the elevator alarm control device further includes: a collection module 506, a labeling module 507, and a training module 508;

[0208] Collection module 506 is used to collect historical real alarm data samples and historical false alarm data samples uploaded by all elevator monitoring systems;

[0209] The annotation module 507 is used to annotate historical real alarm data samples with positive class labels and historical false alarm data samples with negative class labels.

[0210] Training module 508 is used to supervise the training of the classification model using the labeled dataset to obtain the elevator alarm recognition model.

[0211] Figure 6 This is a structural schematic diagram of the elevator alarm control device provided in this application. Figure 6 As shown, this application provides an elevator alarm control device 600, which includes: a receiver 601, a transmitter 602, a processor 603, and a memory 604.

[0212] Receiver 601 is used to receive instructions and data;

[0213] Transmitter 602 is used to send commands and data;

[0214] Memory 604 is used to store instructions executed by the computer;

[0215] The processor 603 is used to execute computer execution instructions stored in the memory 604 to implement the various steps of the elevator alarm control method in the above embodiments. For details, please refer to the relevant descriptions in the foregoing embodiments of the elevator alarm control method.

[0216] Alternatively, the memory 604 can be either standalone or integrated with the processor 603.

[0217] When the memory 604 is set up independently, the electronic device also includes a bus for connecting the memory 604 and the processor 603.

[0218] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the elevator alarm control method performed by the elevator alarm control device described above.

[0219] It will be understood by those skilled in the art that all or some of the steps, systems, or apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0220] The technical solutions of this application have been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it is readily understood by those skilled in the art that the scope of protection of this application is obviously not limited to these specific embodiments. The above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit them. Although this application 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 therein. 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 this application.

Claims

1. An elevator alarm control method, characterized in that, The method, applied to an elevator monitoring system, includes: The system acquires initial alarm data inside the elevator in real time. The initial alarm data is the alarm data generated after the target is initially identified by the target detection algorithm. The initial alarm data is compared with the first false alarm feature library to determine whether the initial alarm data is a false alarm. The first false alarm feature library is used to store the false alarm data generated in the history of the corresponding elevator. If the initial alarm data is not a false alarm, then the initial alarm data is determined to be the target alarm data, and the elevator is controlled to perform the corresponding alarm operation.

2. The method according to claim 1, characterized in that, The step of comparing the initial alarm data with the first false alarm feature database to determine whether the initial alarm data is a false alarm includes: Extract target features from the initial alarm data, where the target features refer to the target visual attributes extracted by an image processing algorithm; Calculate the similarity value between the target feature and the features stored in the first false positive feature library; If the similarity value is greater than a preset threshold, the initial alarm data is determined to be false alarm data; If the similarity value is less than or equal to the preset threshold, the initial alarm data is determined to be non-false alarm data, and the non-false alarm data is sent to the platform's duplicate false alarm review intelligent agent.

3. The method according to claim 1, characterized in that, The control of the elevator to perform an alarm operation includes: Identify the alarm category of the target alarm data; If the alarm category is a direct alarm category, then the elevator alarm operation is executed; If the alarm category is a delayed alarm category, then when the elevator door is closed and the elevator has been stationary for a period of time that reaches the first entrapment alarm duration, the number of targets inside the elevator is obtained. When the target number exceeds a preset threshold, an elevator alarm is triggered. When the target number is less than or equal to the preset number threshold, and the elevator remains stationary for a period of time that reaches the second entrapment alarm duration, an elevator alarm operation is performed, wherein the second entrapment alarm duration is longer than the first entrapment alarm duration.

4. The method according to claim 1, characterized in that, Also includes: If the initial alarm data is a false alarm, the elevator alarm operation will not be performed.

5. An elevator alarm control method, characterized in that, The method, applied to a platform's intelligent agent for reviewing duplicate false alarms, includes: Obtain non-false alarm data sent by the elevator monitoring system, wherein the non-false alarm data carries the elevator identifier of the corresponding elevator; Based on the feature extraction network, features are extracted from the non-false alarm data to obtain non-false alarm feature data; Based on the elevator alarm recognition model, target recognition is performed on the non-false alarm feature data, new false alarm feature data is selected, and the new false alarm feature data is stored in the second false alarm feature library. Based on the elevator identifier corresponding to the new false alarm feature data, the false alarm feature data corresponding to the elevator identifier in the second false alarm feature library is sent to the elevator monitoring system to update the first false alarm feature library.

6. The method according to claim 5, characterized in that, The method for constructing the feature extraction network includes: Extract historical false alarm datasets from the second false alarm feature library; The initial neural network model is trained using the historical false alarm dataset. The feature extraction network is constructed by minimizing the intra-class distance of false alarm features and maximizing the inter-class distance between false alarm features and other features.

7. The method according to claim 5, characterized in that, The method for constructing the elevator alarm recognition model includes: Collect historical real alarm data samples and historical false alarm data samples uploaded by all elevator monitoring systems; The historical real alarm data samples are labeled with positive class labels, and the historical false alarm data samples are labeled with negative class labels; The classification model is trained under supervision using the labeled dataset to obtain the elevator alarm recognition model.

8. An elevator alarm control device, characterized in that, Applications in elevator monitoring systems include: The acquisition module is used to acquire the initial alarm data in the elevator in real time. The initial alarm data is the alarm data generated after the target is initially identified by the target detection algorithm. The determination module is used to compare the initial alarm data with the first false alarm feature library to determine whether the initial alarm data is false alarm data. The first false alarm feature library is used to store the false alarm data generated in the history of the corresponding elevator. The control module is used to determine that the initial alarm data is the target alarm data if the initial alarm data is not a false alarm, and to control the elevator to perform the corresponding alarm operation.

9. An elevator alarm control device, characterized in that, Applications to the intelligent agent for reviewing duplicate false alarms on the platform include: The acquisition module is used to acquire non-false alarm data sent by the elevator monitoring system, wherein the non-false alarm data carries the elevator identifier of the corresponding elevator; The extraction module is used to extract features from the non-false alarm data based on the feature extraction network to obtain non-false alarm feature data; The storage module is used to perform target recognition on the non-false alarm feature data based on the elevator alarm recognition model, filter out new false alarm feature data, and store the new false alarm feature data in the second false alarm feature library. The update module is used to send the false alarm feature data corresponding to the elevator identifier in the second false alarm feature library to the elevator monitoring system based on the elevator identifier corresponding to the new false alarm feature data, thereby updating the first false alarm feature library.

10. An elevator alarm control device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-7.

Images