Elevator trapped person alarm method and system based on signal detection and visual identification

Abstract

The invention discloses an elevator trapped person alarm method and system based on signal detection and visual identification, and the method comprises the steps: collecting an elevator operation electric signal and a car visual perception signal, taking a preset reference signal time identifier as a reference, dynamically correcting a visual signal time sequence deviation through self-adaptive calibration, and guaranteeing the time sequences of the two types of signals to be consistent; and constructing a multi-modal data collaborative analysis basis. And based on the alignment signal, parameters such as elevator door motor current and speed are extracted to construct a normal operation behavior baseline, the activity of people in the lift car is identified through human body posture estimation, and the elevator operation state is monitored synchronously. A signal conflict processing model is established, historical data verification is combined, real-time parameters and base lines are compared, and a person trapping event is judged by integrating three conditions. And after judgment, a local sound-light alarm is triggered, information is pushed to a property management and maintenance unit through the Internet of Things, and if no response is obtained, a city emergency management department is linked to start rescue. And reliable support is provided for elevator operation emergency rescue safety.

Description

Technical Field

[0001] This invention belongs to the field of elevator emergency alarm technology, and more specifically, relates to an elevator entrapment alarm method and system based on signal detection and visual recognition. Background Technology

[0002] With the widespread use of high-rise buildings in cities, elevators have become an indispensable vertical transportation tool for people's daily travel. Their operational safety and reliability are directly related to the safety of users' lives and property. Among them, elevator entrapment incidents are a relatively common safety hazard during elevator operation. Once they occur, they not only cause psychological panic and inconvenience to the trapped people, but if rescue is not timely or handled improperly, they may also lead to secondary safety accidents and cause serious social impact.

[0003] Current elevator entrapment alarms mainly rely on single signal detection or simple status judgment mechanisms, which presents many problems that urgently need to be solved. Some traditional systems only rely on the elevator's operating status electrical signals to determine whether an alarm is triggered, such as relying on single indicators like leveling signals or door opening / closing signals. This lacks effective verification of whether people are inside the car and is prone to false alarms due to signal interference and equipment malfunctions. This not only increases the ineffective workload of property management and maintenance units but may also reduce the efficiency of relevant personnel in responding to genuine alarms.

[0004] Some systems have attempted to incorporate visual recognition technology, but these have failed to effectively address the timing discrepancy between electrical and visual signals. The time-series deviation between these two types of signals leads to the failure of collaborative data analysis, making it impossible to accurately correlate elevator operating status with the situation of people inside the car, thus increasing the risk of missed alarms. Furthermore, existing technologies lack a robust signal conflict handling mechanism. When signals from different sources contradict each other, it becomes difficult to accurately determine the truth of the event, further impacting the reliability of alarms.

[0005] In addition, the emergency response mechanism of traditional alarm systems is not sound enough, the alarm information transmission path is single, and it can often only be reported to the property management or maintenance unit. Furthermore, it lacks the ability to respond redundantly to equipment failures. If the core detection module fails, the alarm function is easily paralyzed, making it impossible to promptly link with the emergency management department to initiate the rescue process, thus prolonging the waiting time for trapped personnel.

[0006] Therefore, there is an urgent need for an elevator entrapment alarm technology that can integrate the advantages of multi-source signals, resolve signal coordination contradictions, improve the accuracy of entrapment incident judgment, and possess comprehensive emergency linkage and fault redundancy capabilities. This technology can effectively reduce false alarm and missed alarm rates, shorten rescue response time, comprehensively improve the safety level of elevator operation, and meet the growing social demand for elevator safety. Summary of the Invention

[0007] This invention aims to address the problems of poor signal coordination, insufficient accuracy in judgment, and inadequate emergency response in traditional elevator entrapment alarms. By integrating electrical signal detection and visual recognition technologies, optimizing signal timing calibration, constructing behavioral baselines, establishing conflict resolution mechanisms, and implementing multi-level emergency response processes, it achieves accurate identification of entrapment incidents, rapid alarm activation, and efficient rescue, thereby enhancing elevator operational safety, reducing false alarms and missed alarms, and providing users with more reliable safety protection.

[0008] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides an elevator entrapment alarm method based on signal detection and visual recognition, comprising: S1. Collect electrical signals related to elevator operation and visual perception signals inside the car. Using the time marker of the preset reference signal as a reference, dynamically correct the timing deviation of the visual perception signal through an adaptive calibration method to ensure that the timing consistency between the electrical signal and the visual perception signal meets the preset requirements, thus building a foundation for multimodal data collaborative analysis. S2. Based on the aligned multimodal signals, on the one hand, signals including elevator door machine current, moving speed, acceleration, pause time and door opening and closing status are extracted to construct the behavioral baseline for normal elevator door operation; on the other hand, the human posture estimation method of visual image signals is used to identify whether there is human activity in the car; at the same time, elevator operation signals are collected synchronously to complete the real-time monitoring of elevator operation status. S3. Establish a multimodal signal conflict handling priority model, combine historical data reliability, prioritize the use of stable signal sources and trigger dual-modal cross-validation; compare the real-time monitored elevator door operation parameters with the normal behavior baseline, and make a comprehensive judgment by combining the door opening and closing status, the identification results of personnel activities in the car and the elevator operation signals. When the three conditions of "door closed for a long time", "personnel activities in the car" and "no floor change signal and elevator call response signal" are met, it is determined to be a entrapment event. S4. Once a entrapment incident is detected, the local audible and visual alarm inside the elevator car is immediately triggered; at the same time, the relevant information of the entrapment incident is pushed to the property monitoring platform and the elevator maintenance unit in real time through the elevator IoT module; if no response is received within a preset time after the push, the emergency rescue process is automatically initiated by contacting the city's elevator emergency management department.

[0009] Furthermore, the electrical signals in S1 include gating electrical signals and operating status electrical signals, which are respectively: Door control electrical signals are used to characterize the control logic and execution status of elevator doors, including but not limited to door opening and closing command signals, door arrival detection signals, door machine operating current signals, relay contact on / off signals, light curtain trigger signals, and safety touch panel feedback signals output by the door machine control system. Operating status electrical signals are used to characterize the overall operating conditions and position information of the elevator, including but not limited to elevator leveling detection signals, floor position coding signals, elevator call command response signals, car load signals, elevator running speed signals, acceleration feedback signals, and braking system status signals.

[0010] Furthermore, the process of dynamically correcting the temporal deviation of the visual perception signal using an adaptive calibration method in S1 is as follows: Using the time marker of the preset reference signal as a reference, the first [number]th ... The timestamp of the group electrical signal is , For the first The time point at which the group of electrical signals is captured by the acquisition module corresponds to the first... The frame acquisition time of the group of visual perception signals is , For the first The time points at which the visual perception signals are captured by the visual acquisition device are used to calculate the original timing deviation between each group of electrical signals and the corresponding visual perception signal. ; Based on the division of elevator door opening and closing action phases, the duration of the current elevator door action phase is set as follows: , Given the time span from start to finish of the current action phase, extract all original timing deviations within that action phase. By calculating the cumulative deviation of the original timing deviation within this action phase and ,in Given the number of signal groups acquired during the current action phase, the frame delay characteristic value of the visual perception signal during this action phase is derived. : , Based on the timing fluctuation characteristics corresponding to the elevator start and stop states, the timing fluctuation threshold during elevator start and stop state switching is set as follows: , The critical value is determined based on the timing deviation fluctuation range during historical start-up and shutdown processes. If the original timing deviation is determined to be an abnormal deviation caused by start-stop fluctuations, the frame delay pattern characterization value is recalculated after removing this group of deviations. ; Based on the final determination The time axis of the visual perception signal is offset and corrected. The acquisition time of the corrected visual perception signal frame is... This causes the timing deviation between the corrected electrical signal and the visually perceived signal to be... It meets preset requirements and completes dynamic calibration of timing deviations in visual perception signals.

[0011] Furthermore, the specific process of establishing the behavioral baseline for normal elevator door operation in S2 is as follows: Based on the time-calibrated multimodal signals, dynamic characteristic parameters of elevator door operation are extracted, including door operator current, moving speed, acceleration, and pause time. The complete opening and closing cycle of the elevator door is set as the basic data acquisition unit for continuous data acquisition. The complete switching cycle data under normal operating conditions form the original feature dataset, in which The preset number of valid data collection cycles; Based on the phase division of elevator door opening and closing operations, each complete opening and closing cycle is broken down into four phases: door opening start, door opening steady state, door closing start, and door closing steady state. The [number]th phase is defined as... In the first complete switching cycle The duration of each action phase is , The value range is 1 to , The value range is 1 to 4, corresponding to the four action stages; extract each Internal gate current sequence Movement speed sequence , acceleration sequence and pause time points ,in Identify the time-series sampling points within each stage; Statistical analysis was performed on the characteristic parameter sequences of each action phase to calculate the first... The cumulative sum of the gantry current sequence during all complete switching cycles within each action phase. The total number of all time-series sampling points during this phase ,in For the first The first cycle The total number of sampling points in the first stage is obtained. Statistical representation of gantry current at each stage Similarly, the calculation yields the first... Statistical representation of movement speed at each stage Statistical representation of acceleration ; Calculate the first The range of pause times within each action phase This yields the characteristic value representing the fluctuation range of the pause time in this stage; Integrate the four action phases corresponding to each , , The fluctuation range of the pause time is characterized by a set of dynamic features covering all stages of the complete opening and closing cycle of the elevator door. This set is the behavioral baseline for the normal operation of the elevator door.

[0012] Furthermore, the elevator operation signals in S2 include floor change signals and call response signals, respectively: Floor change signals are used to characterize the dynamic changes in the actual floor position of the elevator, including but not limited to elevator leveling detection signals, floor code recognition signals, and car position positioning signals. They can provide real-time feedback on the elevator's floor movement status in the shaft and the current floor information. Elevator call response signals are used to characterize the elevator's reception and execution feedback information of external elevator call commands, including but not limited to elevator call command trigger signals, command registration confirmation signals, elevator running direction response signals, and command completion execution feedback signals. They can reflect whether the elevator responds to elevator call requests and the progress of command execution.

[0013] Furthermore, the multimodal signal conflict processing priority model in S3 is specifically as follows: The model first classifies the signal stability levels. Among the electrical signals, the door opening and closing command signals and door arrival detection signals of the door control type, and the leveling detection signal and floor position coding signal of the operation status type are judged as Level 1 stable signals because their acquisition links are short and there are few interference factors. Among the visual perception signals, the personnel presence status recognition results and area contour feature signals after time-series calibration are judged as Level 2 signals. Simultaneously, confidence indexes for the two types of signals within the historical period are extracted, and the effective output ratio of the electrical signal within the continuous historical acquisition period is calculated. ,in This refers to the number of effective outputs of a first-level stable signal in an electrical signal. The total number of electrical signal acquisitions; calculate the historical recognition accuracy of the visual perception signal. ,in This represents the number of times the visual perception signal is consistent with the first-order stable signal through cross-validation. This represents the total number of times the visual perception signal was recognized. When electrical signals conflict with visual perception signals, the model prioritizes the output of the first-level stable signal; if the first-level stable signal has no clear output or Then, taking the visual perception signal result as a reference, the dual-modal cross-validation process is triggered synchronously, and the signal response patterns in similar scenarios in the historical associated data group are called to perform secondary verification of conflicting signals.

[0014] Furthermore, the process for determining the entrapment event in S3 is as follows: First, the real-time monitored elevator door operating parameters are compared with the behavior baseline to extract the deviation between the real-time values ​​of door machine current, moving speed, and acceleration and the statistical characterization values ​​of the corresponding action stages. At the same time, the duration of the door opening and closing state is identified. When the real-time door opening and closing state is closed and the duration exceeds the preset threshold, it is marked as a door abnormal state signal. Secondly, based on the time-calibrated visual perception signal, the human posture estimation method is used to obtain the recognition results of personnel activity in the car. If the outline features and dynamic activity trajectory of the personnel are identified, it is marked as a personnel presence signal; if no personnel-related features are identified, it is marked as a no-person signal. Subsequently, the floor change signal and call response signal in the elevator operation signal are checked synchronously. If no change in the floor position code signal or a jump in the leveling detection signal is detected during the door abnormal status signal marking period, and there is no effective output of the call command trigger signal, command registration confirmation signal and running direction response signal, it is marked as an elevator running stop signal. Finally, the three types of signals are cross-validated using a multimodal signal conflict handling priority model: when the door abnormality signal, the presence of personnel signal, and the elevator stop signal are all satisfied, and the priority model determines that the three types of signals have no conflict or that the conflict verification results are consistent, a entrapment event can be identified; if any signal is missing or there is an unresolved conflict, the entrapment event is not identified for the time being, and the signals are continuously collected for dynamic monitoring until the judgment conditions are met or the signal conflict is resolved.

[0015] Furthermore, S4 also includes: if any signal acquisition module fails during this process, automatically switching to another signal detection mode and upgrading the alarm level, the specific process being as follows: The system monitors the operational effectiveness of the electrical signal acquisition module and the visual perception signal acquisition module in real time. By verifying the continuity of the acquired data, the connectivity of the transmission link, and the effectiveness of the signal output, the system determines the failure status of the modules. Once any signal acquisition module is determined to be faulty, the cross-modal emergency detection switching mechanism is immediately activated, automatically switching to a single-modal detection mode dominated by another effective signal acquisition module. This single-modal detection mode maintains coordination with the aforementioned timing calibration logic, multimodal signal feature association rules, and core conditions for determining entrapment: the dominant signal after switching must use the previously established timing association basis to capture the door closing status and duration, elevator stop status, and relevant information about people inside the car. At the same time, the alarm level will be upgraded. In addition to the regular local audible and visual alarms and information push to the property monitoring platform and elevator maintenance unit, the early warning effect will be further enhanced, including shortening the information push interval, clearly marking the type of failed module and the type of signal that the current detection relies on. If the detection mode is switched to electrical signal-dominated mode, the correlation between the presence of people and electrical signal characteristics in historical data can be used to assist in the determination. If the detection mode is switched to a visual perception signal-dominated mode, the visual feature recognition of the door status is enhanced to ensure the accuracy of the judgment. After switching, the status of the failed module is continuously monitored. If the module recovers to normal, it automatically returns to the multimodal collaborative detection mode and the alarm level drops synchronously. If the module continues to fail, the single-modal dominant detection mode and the upgraded alarm level are maintained until the on-site handling is completed.

[0016] As a second aspect of the present invention, an elevator entrapment alarm system based on signal detection and visual recognition is also provided, comprising: The multimodal signal timing calibration unit is used to collect electrical signals related to elevator operation and visual perception signals inside the car. Using the time marker of a preset reference signal as a reference, it dynamically corrects the timing deviation of the visual perception signal through an adaptive calibration method, so that the timing consistency between the electrical signal and the visual perception signal meets the preset requirements, thus building the foundation for multimodal data collaborative analysis. The baseline construction and status monitoring unit is used to extract signals including elevator door machine current, moving speed, acceleration, pause time and door opening and closing status based on the aligned multimodal signals to construct the behavioral baseline for normal elevator door operation; on the other hand, it identifies whether there is human activity in the car through human posture estimation method of visual image signals; at the same time, it synchronously collects elevator operation signals to complete the real-time monitoring of elevator operation status. The conflict handling and entrapment determination unit is used to establish a multimodal signal conflict handling priority model. Combining the reliability of historical data, it prioritizes the use of stable signal sources and triggers dual-modal cross-validation. It compares the real-time monitored elevator door operating parameters with the normal behavior baseline, and makes a comprehensive judgment based on the door opening and closing status, the identification results of personnel activities in the car, and the elevator operating signals. When the three conditions of "door closed for a long time", "personnel activities in the car", and "no floor change signal and elevator call response signal" are met, it is determined to be an entrapment event. The entrapment alarm and emergency linkage unit is used to immediately trigger the local audible and visual alarm in the elevator car once an entrapment incident is detected; at the same time, through the elevator IoT module, the relevant information of the entrapment incident is pushed to the property monitoring platform and the elevator maintenance unit in real time; if no response is received within a preset time after the push, it will automatically link with the city elevator emergency management department to start the emergency rescue process.

[0017] As a third aspect of the invention, a computer-readable storage medium is also provided, on which a computer program is stored, which is executed by a processor as described in any one of the claims, a method for alarming elevator entrapment based on signal detection and visual recognition.

[0018] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: 1. The elevator entrapment alarm method based on signal detection and visual recognition of the present invention collects electrical signals related to elevator operation and visual perception signals inside the car. Using a preset reference signal's time signature as a reference, it dynamically corrects the timing deviation of the visual perception signals through an adaptive calibration method, ensuring that the timing consistency of the two types of signals meets preset requirements. This technical feature effectively solves the time misalignment problem that easily occurs during the acquisition of multi-source signals, allowing the elevator operating conditions reflected by the electrical signals to accurately correspond to the scene inside the car captured by the visual signals. This removes timing obstacles for subsequent multi-dimensional data collaborative analysis, ensuring that various monitoring data are comparable and correlated in the time dimension, providing an accurate and synchronous data foundation for the entire alarm method, and improving the reliability of subsequent analysis and judgment.

[0019] 2. The elevator entrapment alarm method based on signal detection and visual recognition of the present invention extracts key parameters such as elevator door operator current and moving speed from aligned multimodal signals to construct a behavioral baseline for normal elevator door operation. It identifies the activity of people inside the car through human posture estimation and simultaneously collects elevator operation signals for real-time monitoring. This technical feature provides a clear basis for judging whether the elevator door operation status is abnormal by establishing a normal operation benchmark. Simultaneously, it directly acquires the presence of people inside the car and core elevator operation information. These three types of data form a complementary and verifying monitoring system, comprehensively covering the key dimensions required for entrapment event judgment. This avoids the limitations of single signal monitoring, provides sufficient data analysis support for subsequent event judgment, and reduces misjudgments or omissions caused by incomplete information.

[0020] 3. The elevator entrapment alarm method based on signal detection and visual recognition of this invention establishes a multimodal signal conflict processing priority model, performs dual-modal cross-validation by combining historical data reliability, compares real-time monitoring parameters with behavioral baselines, and comprehensively determines an entrapment event based on door opening / closing status, personnel activity results, and elevator operation signals. Subsequently, it triggers a local audible and visual alarm, pushes information to relevant units, and coordinates with emergency departments. This technical feature ensures the accuracy of signal analysis through a scientific conflict processing mechanism, ensures the precision of entrapment event determination through multi-condition comprehensive judgment, and constructs a progressive emergency response process. This effectively avoids false alarms caused by a single signal anomaly and can quickly initiate rescue links after confirming an entrapment event, shortening rescue response time, providing timely safety guarantees for trapped personnel, and improving the safety protection level of elevator operation. Attached Figure Description

[0021] Figure 1 This is a flowchart of an elevator entrapment alarm method based on signal detection and visual recognition according to an embodiment of the present invention; Figure 2This is a schematic diagram of car data acquisition according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the monitoring platform according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the system units in an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0023] Example 1 Please refer to Figure 1 This embodiment 1 provides an elevator entrapment alarm method based on signal detection and visual recognition, including: S1. Collect electrical signals related to elevator operation and visual perception signals inside the car. Using the time marker of the preset reference signal as a reference, dynamically correct the timing deviation of the visual perception signal through an adaptive calibration method to ensure that the timing consistency between the electrical signal and the visual perception signal meets the preset requirements, thus building a foundation for multimodal data collaborative analysis. S2. Based on the aligned multimodal signals, on the one hand, signals including elevator door machine current, moving speed, acceleration, pause time and door opening and closing status are extracted to construct the behavioral baseline for normal elevator door operation; on the other hand, the human posture estimation method of visual image signals is used to identify whether there is human activity in the car; at the same time, elevator operation signals are collected synchronously to complete the real-time monitoring of elevator operation status. S3. Establish a multimodal signal conflict handling priority model, combine historical data reliability, prioritize the use of stable signal sources and trigger dual-modal cross-validation; compare the real-time monitored elevator door operation parameters with the normal behavior baseline, and make a comprehensive judgment by combining the door opening and closing status, the identification results of personnel activities in the car and the elevator operation signals. When the three conditions of "door closed for a long time", "personnel activities in the car" and "no floor change signal and elevator call response signal" are met, it is determined to be a entrapment event. S4. Once a entrapment incident is detected, the local audible and visual alarm inside the elevator car is immediately triggered; at the same time, the relevant information of the entrapment incident is pushed to the property monitoring platform and the elevator maintenance unit in real time through the elevator IoT module; if no response is received within a preset time after the push, the emergency rescue process is automatically initiated by contacting the city's elevator emergency management department.

[0024] Please refer to Figure 2 as well as Figure 3This embodiment 1 further elaborates on the above steps.

[0025] (1) Multimodal signal timing calibration In the multimodal data acquisition and analysis of elevator entrapment alarms, the timing consistency between electrical signals and visual perception signals is a key prerequisite for ensuring accurate subsequent judgments. However, differences in response speed and transmission links between different signal acquisition devices can easily lead to timing deviations between the two types of signals. Therefore, appropriate calibration methods are needed to achieve dynamic correction.

[0026] The collected electrical signals are mainly divided into two categories: door control and operation status. The door control electrical signals focus on the control logic and execution status of the elevator door, including the door opening and closing command signals output by the door machine control system, the door position detection signal used to confirm whether the door is fully closed or open, the door machine operating current signal reflecting the door machine operating load, the relay contact on / off signal reflecting the on / off status of the control circuit, the light curtain trigger signal to monitor whether there are obstacles in the door closing path, and the safety touch plate feedback signal to assist in anti-pinch protection. Operating status electrical signals are used to present the overall operating conditions and location information of the elevator, including leveling detection signals to confirm whether the elevator has accurately stopped at a floor, floor position coding signals to identify the current floor of the elevator, call command response signals to provide feedback on the elevator's response to external call requests, car load signals to display the weight carried in the car, operating speed signals to characterize the speed of the elevator's ascent and descent, acceleration feedback signals to reflect the elevator's acceleration or deceleration status, and braking system status signals to reflect the working status of the braking system.

[0027] The two types of electrical signals comprehensively cover the elevator door control logic and overall machine operation at the hardware level, providing an objective and stable basis for determining entrapment status. However, electrical signals cannot directly reflect the situation of people inside the car and need to be analyzed in conjunction with visual perception signals. The consistency of their timing is the core prerequisite for ensuring the accuracy of the collaborative analysis. Therefore, the following adaptive calibration method is needed to dynamically correct the timing deviation of the visual perception signals: Using the time marker of the preset reference signal as a reference, the first [number]th ... The timestamp of the group electrical signal is , For the first The time point at which the group of electrical signals is captured by the acquisition module corresponds to the first... The frame acquisition time of the group of visual perception signals is , For the first The time points at which the visual perception signals are captured by the visual acquisition device are used to calculate the original timing deviation between each group of electrical signals and the corresponding visual perception signal. ; Based on the division of elevator door opening and closing action phases, the duration of the current elevator door action phase is set as follows: , Given the time span from start to finish of the current action phase, extract all original timing deviations within that action phase. By calculating the cumulative deviation of the original timing deviation within this action phase and ,in Given the number of signal groups acquired during the current action phase, the frame delay characteristic value of the visual perception signal during this action phase is derived. : , Based on the timing fluctuation characteristics corresponding to the elevator start and stop states, the timing fluctuation threshold during elevator start and stop state switching is set as follows: , The critical value is determined based on the timing deviation fluctuation range during historical start-up and shutdown processes. If the original timing deviation is determined to be an abnormal deviation caused by start-stop fluctuations, the frame delay pattern characterization value is recalculated after removing this group of deviations. ; Based on the final determination The time axis of the visual perception signal is offset and corrected. The acquisition time of the corrected visual perception signal frame is... This causes the timing deviation between the corrected electrical signal and the visually perceived signal to be... It meets preset requirements and completes dynamic calibration of timing deviations in visual perception signals.

[0028] (2) Baseline construction and status monitoring Abnormalities in the operation of elevator doors are an important basis for judging entrapment incidents. Only by clearly defining the standard characteristics of normal elevator door operation can we accurately identify abnormal situations in subsequent operation. Therefore, it is necessary to build a behavioral baseline covering the complete operation cycle of elevator doors, and at the same time collect relevant elevator operation signals to provide comprehensive support for entrapment determination.

[0029] The construction of the behavioral baseline relies on time-calibrated multimodal signals. Through phased feature extraction and statistical analysis, the dynamic patterns of normal elevator door operation are characterized. The specific steps are as follows: Based on the time-calibrated multimodal signals, dynamic characteristic parameters of elevator door operation are extracted, including door operator current, moving speed, acceleration, and pause time. The complete opening and closing cycle of the elevator door is set as the basic data acquisition unit for continuous data acquisition. The complete switching cycle data under normal operating conditions form the original feature dataset, in which The preset number of valid data collection cycles; Based on the phase division of elevator door opening and closing operations, each complete opening and closing cycle is broken down into four phases: door opening start, door opening steady state, door closing start, and door closing steady state. The [number]th phase is defined as... In the first complete switching cycle The duration of each action phase is , The value range is 1 to , The value range is 1 to 4, corresponding to the four action stages; extract each Internal gate current sequence Movement speed sequence , acceleration sequence and pause time points ,in Identify the time-series sampling points within each stage; Statistical analysis was performed on the characteristic parameter sequences of each action phase to calculate the first... The cumulative sum of the gantry current sequence during all complete switching cycles within each action phase. The total number of all time-series sampling points during this phase ,in For the first The first cycle The total number of sampling points in the first stage is obtained. Statistical representation of gantry current at each stage Similarly, the calculation yields the first... Statistical representation of movement speed at each stage Statistical representation of acceleration ; Calculate the first The range of pause times within each action phase This yields the characteristic value representing the fluctuation range of the pause time in this stage; Integrate the four action phases corresponding to each , , The fluctuation range of the pause time is characterized by a set of dynamic features covering all stages of the complete opening and closing cycle of the elevator door. This set is the behavioral baseline for the normal operation of the elevator door.

[0030] Furthermore, while establishing the behavioral baseline, elevator operation signals are simultaneously collected. These signals mainly include floor change signals and call response signals. Floor change signals characterize the dynamic changes in the elevator's actual floor position, including elevator leveling detection signals, floor code identification signals, and car position positioning signals. These signals provide real-time feedback on whether the elevator has moved floors within the shaft and its current floor information. Call response signals characterize the elevator's reception and execution of external call commands, including call command trigger signals, command registration confirmation signals, elevator direction response signals, and command completion feedback signals. These signals clearly reflect whether the elevator has received a call request, whether the command has been registered, and the progress of command execution. Together, these two types of signals enable real-time monitoring of the elevator's operating status.

[0031] Meanwhile, based on the time-calibrated visual perception signal, a human posture estimation algorithm is used to analyze the image frames inside the car frame by frame. By extracting the coordinate information of key human skeletal nodes in the image, including feature points such as head, shoulders, elbows, hips, and knees, a human posture feature vector is constructed. The matching degree of the feature vector with the preset human posture template is compared to filter out interference features caused by the shaking of objects inside the car and changes in light, and to distinguish between human activity and non-human interference factors. When a stable human posture feature vector is detected in multiple consecutive frames of images, and the displacement changes of the feature points conform to the laws of human activity, it is determined that there is human activity inside the car. If no key human feature points are detected in the image frame, or only scattered and irregular feature points appear, it is determined that there is no human activity inside the car, forming a clear identification result of the presence of personnel, providing a core basis for the determination of entrapment events.

[0032] (3) Conflict handling and entrapment determination In the process of collaboratively judging entrapment events using multimodal signals, signals from different sources may conflict due to factors such as the acquisition environment and equipment status. Furthermore, the determination of entrapment events requires the integration of key information from multiple dimensions. Therefore, it is necessary to establish a suitable signal conflict handling mechanism and a clear judgment logic to ensure the accuracy of the results.

[0033] The signal conflict handling priority model is based on the temporal consistency of electrical signals and visual perception signals. It combines the inherent stability characteristics of the signals with the confidence level of historical data to construct a hierarchical conflict coordination mechanism. The model first classifies signals into stability levels. Among electrical signals, door opening / closing command signals and door arrival detection signals (related to gate control), as well as leveling detection signals and floor location encoding signals (related to operational status), are classified as Level 1 stable signals due to their short acquisition links, fewer external interference factors, and higher data reliability. Among visual perception signals, personnel presence status recognition results after temporal calibration and area contour feature signals are classified as Level 2 signals.

[0034] Simultaneously, confidence indexes for the two types of signals within the historical period are extracted, and the effective output ratio of the electrical signal within the continuous historical acquisition period is calculated. ,in This refers to the number of effective outputs of a first-level stable signal in an electrical signal. The total number of electrical signal acquisitions; calculate the historical recognition accuracy of the visual perception signal. ,in This represents the number of times the visual perception signal is consistent with the first-order stable signal through cross-validation. This represents the total number of times the visual perception signal was recognized. When electrical signals conflict with visual perception signals, the model prioritizes the output of the first-level stable signal; if the first-level stable signal has no clear output or Then, taking the visual perception signal result as a reference, the dual-modal cross-validation process is triggered synchronously, and the signal response patterns in similar scenarios in the historical associated data group are called to perform secondary verification of conflicting signals.

[0035] Based on the aforementioned multimodal signal conflict processing priority model, conflicting signals can be coordinated. From conflicting electrical and visual perception signals, signals that are more consistent with reality and have higher reliability can be selected as the basis for judgment, avoiding misjudgments due to signal inconsistencies. Based on this, a tiered judgment process for entrapment events is carried out by combining multi-dimensional monitoring signals. The specific process is as follows: First, anomaly identification of elevator door operation status is conducted. Real-time collected operating parameters such as door operator current, moving speed, and acceleration are compared item by item with the statistical representation values ​​of the corresponding action stages in a pre-established baseline of normal elevator door operation behavior to determine the degree of real-time deviation of each parameter. Simultaneously, the opening and closing status of the elevator door is continuously tracked, recording the duration the door remains closed. When this duration exceeds a preset safety threshold, an abnormal elevator door operation is determined, and an abnormal door status signal is simultaneously marked, providing the first layer of evidence for subsequent entrapment determination.

[0036] Secondly, the presence of people inside the elevator car is confirmed. Based on time-calibrated visual perception signals, a human posture estimation algorithm is used to analyze real-time image frames inside the car, extracting the contour features and dynamic movement trajectories of the human body in the images. If feature information consistent with human activity patterns is detected in multiple consecutive image frames, it is marked as a signal of the presence of people; if no human-related features are identified, it is marked as a signal of no people. This clarifies whether there are people trapped inside the car, constituting a key prerequisite for determining whether someone is trapped.

[0037] Subsequently, the overall elevator operation status is verified. Floor change signals and call response signals from the elevator operation signals are retrieved simultaneously, and a specific check is performed on the time period marked by the door abnormality signal. If, during this period, the floor position code signal does not change, the leveling detection signal does not jump, and the call command trigger signal, command registration confirmation signal, and running direction response signal all have no effective output, it indicates that the elevator has neither moved floors nor responded to any call requests and is in a stationary state. The elevator is then marked as stationary.

[0038] Finally, a comprehensive cross-validation of multi-dimensional signals is carried out. Combined with the multi-modal signal conflict processing priority model, the consistency of the above-mentioned door abnormal status signal, personnel presence signal, and elevator operation stop signal is checked. When all three types of signals are satisfied at the same time, and the priority model determines that there is no signal conflict, or the conflicting signals reach a consensus after secondary verification, it can be determined as a entrapment event; if any signal is missing, or there is an unsolvable conflict between the signals, the entrapment judgment will not be made for the time being, and various signals will be continuously collected for dynamic monitoring until all judgment conditions are met, or the signal conflict problem is effectively resolved. (4) Entrapment alarm and emergency linkage In the emergency response process for people trapped in accidents, in addition to the standard alarm and rescue coordination mechanism, it is also necessary to consider the potential failure of the signal acquisition module. The normal operation of the signal acquisition module is fundamental to the accuracy of trapped person detection; if any module fails, it may lead to missing or abnormal monitoring data, affecting subsequent response decisions. Therefore, it is necessary to establish an emergency switching and alarm enhancement mechanism in case of module failure to ensure the continuity of trapped person monitoring and emergency response. The specific process is as follows: Once the system detects a entrapment incident, it immediately activates a dual emergency response: First, it triggers a local audible and visual alarm inside the elevator car, alerting the trapped individuals with sound and light to alleviate their anxiety and inform them that they have been detected. Second, using the elevator's IoT module, it pushes specific information about the entrapment incident (including elevator location, entrapment time, door status, etc.) to the property management monitoring platform and elevator maintenance company in real time, ensuring that relevant responsible parties are informed of the situation immediately and prepare for rescue. If no response is received from the property management or maintenance company within a preset reasonable time after the information is pushed, the system will automatically escalate the rescue linkage level, establish a connection with the city's elevator emergency management department, and initiate official emergency rescue procedures to ensure the timely progress of rescue work.

[0039] Throughout the entire emergency response process, the system continuously monitors the operational effectiveness of the electrical signal acquisition module and the visual perception signal acquisition module in real time. The monitoring primarily includes three core indicators: the continuity of acquired data (determining whether there are data interruptions or missing data); the connectivity of the transmission link (verifying whether the signal transmission channel from the acquisition end to the processing end is unobstructed); and the validity of the signal output (verifying whether the acquired signal conforms to a normal data format and has practical analytical value). Through comprehensive verification of these three indicators, the system can accurately determine the module failure status—when any module experiences no data output, continuous data anomalies, or a transmission link interruption, the module is deemed to have failed.

[0040] If any signal acquisition module fails, the system will immediately activate the cross-modal emergency detection switching mechanism, automatically switching to a single-modal detection mode dominated by another still-active signal acquisition module. To ensure the continuity of the detection logic and the consistency of the judgment criteria after the switch, this single-modal detection mode will strictly adhere to the previously established timing calibration logic, multi-modal signal feature association rules, and core conditions for entrapment judgment. The dominant signal after the switch must continue to accurately capture the key information required for entrapment judgment based on the original timing association, including the closing status and duration of the elevator doors, whether the elevator is in a stopped state, and whether there are people in the car, ensuring that entrapment monitoring is not interrupted.

[0041] At the same time, the system will simultaneously upgrade the alarm level. While retaining basic early warning measures such as local audible and visual alarms in the elevator car and information push to the property monitoring platform and elevator maintenance unit, the system further enhances the early warning effect: First, it shortens the information push interval, allowing relevant units to continuously obtain the latest monitoring information and avoid affecting rescue due to information lag; second, it clearly marks the specific type of the failed module (electrical signal acquisition module or visual perception signal acquisition module) and the type of detection signal that the current system relies on in the pushed information, providing relevant units with accurate reference for judging the situation and formulating rescue plans.

[0042] Depending on the type of dominant signal after the switch, the system will optimize the detection strategy accordingly: If switched to the electrical signal-dominated detection mode, since electrical signals cannot directly identify the status of personnel, the system will combine the correlation patterns between personnel presence and electrical signal characteristics summarized from historical data (such as typical changes in signals such as door operator current and car load when personnel are present) to assist in determining the presence status of personnel; if switched to the visual perception signal-dominated detection mode, the system will strengthen the visual feature recognition of the door status, and ensure the accuracy of judging the closing status and duration of the elevator door by accurately analyzing the position and shape changes of the door outline in the image frame, thus compensating for the impact of missing electrical signals.

[0043] After switching detection modes, the system will continuously monitor the operating status of the failed module. If the failed module is detected to have returned to normal operation, it will automatically switch back to the multimodal collaborative detection mode and simultaneously reduce the alarm level to the normal level. If the failed module remains in an abnormal state, it will maintain the single-modal dominant detection mode and the increased alarm level until rescue personnel arrive on site to complete the handling work and repair and restore the failed module. Then, it will be manually or automatically adjusted to the normal monitoring mode by the system.

[0044] The multimodal intelligent judgment and emergency response method for elevator entrapment proposed in this embodiment can be directly integrated into existing elevator IoT monitoring systems without large-scale hardware modifications, possessing strong engineering adaptability and low-cost implementation advantages. This method effectively reduces the false positive and false negative rates of entrapment incidents through timing calibration of electrical and visual signals, hierarchical priority conflict handling, and multi-dimensional feature cross-validation, significantly improving the safety monitoring level of elevator operation. It is particularly suitable for elevator safety management in densely populated places such as high-rise residential buildings, commercial complexes, and office buildings, providing property management and maintenance units with accurate early warning information and decision-making basis, and helping to achieve intelligent and refined upgrades in elevator safety management.

[0045] From an industry development perspective, the multimodal signal collaborative analysis framework constructed in this embodiment can be extended to monitoring and early warning scenarios for other types of elevator faults, forming an intelligent safety monitoring system covering the entire life cycle of elevators. With the continuous growth of the number of elevators in cities, the traditional manual inspection mode can no longer meet the needs of efficient safety management. The promotion and application of this method can not only effectively protect the lives of elevator passengers and reduce safety hazards and disputes caused by untimely handling of entrapment incidents, but also provide standardized monitoring data and response solutions for urban special equipment safety supervision platforms, promoting the elevator safety management industry towards data-driven and intelligent decision-making, and has broad market application prospects and social value.

[0046] Example 2 Please refer to Figure 4 This embodiment 2 provides an elevator entrapment alarm system based on signal detection and visual recognition, including: The multimodal signal timing calibration unit is used to collect electrical signals related to elevator operation and visual perception signals inside the car. Using the time marker of a preset reference signal as a reference, it dynamically corrects the timing deviation of the visual perception signal through an adaptive calibration method, so that the timing consistency between the electrical signal and the visual perception signal meets the preset requirements, thus building the foundation for multimodal data collaborative analysis. The baseline construction and status monitoring unit is used to extract signals including elevator door machine current, moving speed, acceleration, pause time and door opening and closing status based on the aligned multimodal signals to construct the behavioral baseline for normal elevator door operation; on the other hand, it identifies whether there is human activity in the car through human posture estimation method of visual image signals; at the same time, it synchronously collects elevator operation signals to complete the real-time monitoring of elevator operation status. The conflict handling and entrapment determination unit is used to establish a multimodal signal conflict handling priority model. Combining the reliability of historical data, it prioritizes the use of stable signal sources and triggers dual-modal cross-validation. It compares the real-time monitored elevator door operating parameters with the normal behavior baseline, and makes a comprehensive judgment based on the door opening and closing status, the identification results of personnel activities in the car, and the elevator operating signals. When the three conditions of "door closed for a long time", "personnel activities in the car", and "no floor change signal and elevator call response signal" are met, it is determined to be an entrapment event. The entrapment alarm and emergency linkage unit is used to immediately trigger the local audible and visual alarm in the elevator car once an entrapment incident is detected; at the same time, through the elevator IoT module, the relevant information of the entrapment incident is pushed to the property monitoring platform and the elevator maintenance unit in real time; if no response is received within a preset time after the push, it will automatically link with the city elevator emergency management department to start the emergency rescue process.

[0047] Example 3 This embodiment 3 also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can implement any step of an elevator entrapment alarm method based on signal detection and visual recognition.

[0048] The computer-readable storage medium may include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0049] For a description of the computer-readable storage medium provided in this application, please refer to the above method embodiments; further details will not be repeated here.

[0050] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0051] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An elevator trapped person alarm method based on signal detection and visual recognition, characterized by, Comprise: S1. Collecting electrical signals related to elevator operation and visual perception signals in the car, taking the time mark of the preset reference signal as the reference, dynamically correcting the timing deviation of the visual perception signal through an adaptive calibration method, making the timing consistency of the electrical signal and the visual perception signal meet the preset requirements, and constructing a multi-modal data collaborative analysis basis; S2. Based on the aligned multi-modal signals, on the one hand, signals including elevator door motor current, moving speed, acceleration, pause time and door switch state are extracted to construct the behavior baseline of normal operation of the elevator door; on the other hand, whether there is personnel activity in the car is identified through human pose estimation method of visual image signal; at the same time, the elevator operation signal is synchronously collected to complete the real-time monitoring of the elevator operation state; S3. Establishing a multi-modal signal conflict processing priority model, combining the historical data credibility, preferentially collecting stable signal sources and triggering double-modal cross verification; The real-time monitored elevator door operation parameters are compared with the normal behavior baseline, and the door switch state, the personnel activity recognition result in the car and the elevator operation signal are comprehensively judged, when the three conditions of "long time door closed", "personnel activity in the car" and "no floor change signal and call response signal" are met, the person trapped event is determined; S4. Once the person trapped event is determined, the local sound and light alarm in the elevator car is triggered immediately; at the same time, through the elevator Internet of Things module, the person trapped event related information is pushed to the property monitoring platform and the elevator maintenance unit in real time; if no response is received within the preset time after pushing, the urban elevator emergency management department is automatically linked, and the emergency rescue process is started.

2. The method of claim 1, wherein the method further comprises: The electrical signals in S1 include door control type electrical signals and running state type electrical signals, which are respectively: The door control type electrical signal is used to represent the control logic and execution state of the elevator door, including but not limited to the switch door instruction signal output by the door machine control system, the door in place detection signal, the door machine running current signal, the relay contact on-off signal, the light curtain trigger signal, and the safety touch plate feedback signal; The running state type electrical signal is used to represent the overall running condition and position information of the elevator, including but not limited to the elevator leveling detection signal, the floor position coding signal, the call instruction response signal, the car load signal, the elevator running speed signal, the acceleration feedback signal, and the brake system state signal.

3. The method of claim 1, wherein the method further comprises: The process of dynamically correcting the timing deviation of the visual perception signal in S1 through the adaptive calibration method is: With the time identifier of the preset reference signal as a reference reference, the time stamp of the first group of electrical signals in a continuous acquisition period is set as , The time node when the first group of electrical signals is captured by the acquisition module is The frame acquisition time of the first group of visual perception signals is , The time node when the first group of visual perception signals is captured by the visual acquisition device is calculated. The original time sequence deviation of each group of electrical signals and the corresponding visual perception signal is calculated Based on the division of the elevator door switch action stage, the duration of the current action stage of the elevator door is set as , The time span from the start to the end of the current action stage is set as T , and all original time sequence deviations in the action stage are extracted , and the cumulative deviation of the original time sequence deviation in the action stage is calculated , wherein n is the number of signal groups collected in the action stage, and the frame delay rule representation value of the visual perception signal in the action stage is derived ， In combination with the time sequence fluctuation characteristics corresponding to the elevator start-stop state, the time sequence fluctuation threshold at the time of switching of the elevator start-stop state is set as , The critical value is determined based on the time sequence deviation fluctuation range in the historical start-stop process. When , it is determined that the group of original time sequence deviations is abnormal deviation caused by start-stop fluctuation, and the frame delay regularity representation value is recalculated after eliminating the group of deviations ; According to the final determination The time axis of the visual perception signal is offset and corrected, and the frame acquisition time of the corrected visual perception signal is The time sequence deviation between the corrected electric signal and the visual perception signal is The preset requirement is met, and the dynamic calibration of the time sequence deviation of the visual perception signal is completed.

4. The method of claim 1, wherein the method further comprises: The specific process of constructing the behavior baseline of normal operation of the elevator door in S2 is: Based on the multi-modal signals calibrated in time sequence, dynamic characteristic parameters of the elevator door operation are extracted, including door machine current, moving speed, acceleration and pause time, a complete opening and closing cycle of the elevator door is set as a basic data acquisition unit, complete opening and closing cycle data in a normal operation state are continuously acquired to form an original characteristic data set, wherein is a preset number of effective data acquisition cycles. Based on the phase division of elevator door opening and closing operations, each complete opening and closing cycle is broken down into four phases: door opening start, door opening steady state, door closing start, and door closing steady state. The [number]th phase is defined as... In the first complete switching cycle The duration of each action phase is , The value range is 1 to , The value range is 1 to 4, corresponding to the four action stages; extract each Internal gate current sequence Movement speed sequence , acceleration sequence and pause time points ,in Identify the time-series sampling points within each stage; The characteristic parameter sequence of each action stage is statistically analyzed, and the cumulative sum of the door machine current sequence in all complete switching cycles in the first action stage is calculated , the total number of all timing sampling points in the stage is calculated , wherein is the total number of sampling points of the first cycle in the first stage, and the statistical representation value of the door machine current in the first stage is obtained . Similarly, the statistical representation value of the moving speed in the first stage is calculated , and the statistical representation value of the acceleration is calculated ; calculating the range of the pause time points in the first action stage , obtaining the fluctuation range representation value of the pause time in the stage . Integrating the respective corresponding , , and the fluctuation range of the pause time representation value, a dynamic feature set covering each phase of the complete opening and closing cycle of the elevator door is formed, which is the behavior baseline of the normal operation of the elevator door.

5. The method of claim 1, wherein the method further comprises: The elevator operation signal in S2 includes floor change signal and call response signal, which are respectively: The floor change signal is used to represent the dynamic change information of the actual floor position of the elevator, including but not limited to the elevator leveling detection signal, the floor coding recognition signal, and the car position positioning signal, which can feedback the floor moving state and the current stop floor information of the elevator in the shaft in real time; The call response signal is used to represent the feedback information of the receiving and executing of the external call instruction by the elevator, including but not limited to the call instruction trigger signal, the instruction registration confirmation signal, the elevator running direction response signal and the instruction completion execution feedback signal, and can reflect whether the elevator responds to the call demand and the instruction execution progress.

6. The method of claim 1, wherein the method further comprises: The S3 multimodal signal conflict processing priority model is specifically: The model first divides the signal stability level. The switch door instruction signal of the gate type electrical signal, the door in place detection signal and the running state type electrical signal, the flat layer detection signal and the floor position coding signal, are determined as first-level stable signals because of their short acquisition link and few interference factors. The personnel existence state recognition result and the area contour feature signal in the visual perception signal are determined as second-level signals after time sequence calibration. Meanwhile, confidence representation indexes of the two types of signals in the historical period are extracted, and the effective output proportion of the electrical signal in the continuous historical collection period is calculated wherein is the effective output times of the primary stable signal in the electrical signal, is the total collection times of the electrical signal; the historical recognition accuracy of the visual perception signal is calculated wherein is the number of times of cross-validation consistency between the visual perception signal and the primary stable signal, is the total recognition times of the visual perception signal; When the electrical signal and the visual perception signal conflict, the model gives priority to the output result of the first stable signal; if the first stable signal has no clear output or the visual perception signal result is taken as the reference, and a cross-verification process of the dual-mode is triggered synchronously, and the signal response law in the same scene in the historical associated data set is called to verify the conflict signal twice.

7. The method of claim 1, wherein the method further comprises: The S3 person-trapped event determination process is: Firstly, the real-time monitored elevator door operation parameters are compared with the behavior baseline to extract the deviation degree of the real-time values of the door machine current, moving speed and acceleration from the corresponding action stage statistical representation values, and to identify the duration of the door opening and closing state. When the real-time door opening and closing state is closed and the duration exceeds the preset threshold, it is marked as a door abnormal state signal. Secondly, based on the time sequence calibrated visual perception signal, the personnel activity recognition result in the car is obtained through the human body posture estimation method. If the personnel contour feature and dynamic activity track are recognized, it is marked as a personnel existence signal. If no personnel related feature is recognized, it is marked as a no personnel signal. Then, the floor change signal in the elevator operation signal and the call response signal are synchronously checked. If no change of the floor position coding signal and no jump of the flat layer detection signal are detected during the door abnormal state signal marking period, and no call instruction trigger signal, instruction registration confirmation signal and running direction response signal are effectively output, it is marked as an elevator operation stagnation signal. Finally, the above three types of signals are cross-verified by the multimodal signal conflict processing priority model. When the door abnormal state signal, the personnel existence signal and the elevator operation stagnation signal are simultaneously satisfied, and the three types of signals are consistent after the conflict verification, it is determined as a person-trapped event. If any signal is missing or there is an unresolved conflict, the person-trapped event is not determined, and the signals are continuously collected for dynamic monitoring until the determination condition is met or the signal conflict is resolved.

8. The method of claim 1, wherein the method further comprises: The S4 further includes: during this process, if any signal acquisition module fails, automatically switch to another signal detection mode and raise the alarm level. The specific process is: The running effectiveness of the electrical signal acquisition module and the visual perception signal acquisition module is monitored in real time. The module failure state is determined through the continuity of the collected data, the connectivity of the transmission link and the effectiveness verification of the signal output. Once it is determined that any signal acquisition module is invalid, the cross-modal emergency detection switching mechanism is started immediately, and the single-modal detection mode dominated by another valid signal acquisition module is switched to automatically, and the single-modal detection mode keeps coordination with the aforementioned timing calibration logic, multi-modal signal feature association rule and person-trapped judgment core condition: the dominant signal after switching needs to follow the previously established timing association basis to capture the door closing state and duration, elevator operation stagnation state, and information related to the people in the car; At the same time, the alarm level is simultaneously improved, and on the basis of conventional local audible and visual alarms and information pushing to the property monitoring platform and elevator maintenance unit, the early warning effect is further strengthened, including shortening the information pushing interval, clearly marking the type of invalid module and the type of signal currently relied on for detection; If the signal dominated detection mode is switched to, the association rule between the presence of people in the historical data and the signal features is combined to assist in the judgment; If the visual perception signal dominated detection mode is switched to, the visual feature recognition of the door state is strengthened to ensure the judgment accuracy; After switching, the status of the invalid module is continuously monitored, if the module returns to normal, the multi-modal collaborative detection mode is automatically returned, and the alarm level is simultaneously lowered; if the module remains invalid, the single-modal dominant detection mode and the improved alarm level are maintained until the on-site disposal is completed.

9. An elevator person-trapped alarm system based on signal detection and visual recognition, characterized by, It includes: A multi-modal signal timing calibration unit for acquiring elevator operation related electrical signals and car visual perception signals, taking the time identifier of a preset reference signal as a reference, dynamically correcting the timing deviation of the visual perception signal through an adaptive calibration method, making the timing consistency of the electrical signal and the visual perception signal meet the preset requirements, and building a multi-modal data collaborative analysis foundation; A baseline construction and state monitoring unit for extracting signals including elevator door motor current, moving speed, acceleration, dwell time and door switch state based on the aligned multi-modal signals, and constructing the behavior baseline of the normal operation of the elevator door; on the other hand, whether there is personnel activity in the car is identified through a human pose estimation method of visual image signals; at the same time, the elevator operation signals are synchronously acquired to complete real-time monitoring of the elevator operation state; A conflict processing and person-trapped judgment unit for establishing a multi-modal signal conflict processing priority model, combining the credibility of historical data, and preferentially using stable signal sources and triggering double-modal cross verification; The real-time monitored elevator door operation parameters are compared with the normal behavior baseline, and at the same time, the door switch state, the personnel activity recognition result in the car and the elevator operation signal are comprehensively judged, when the three conditions of "long time door closing", "personnel activity in the car" and "no floor change signal and call response signal" are met, the person-trapped event is determined; A person-trapped alarm and emergency linkage unit for triggering local audible and visual alarms in the elevator car as soon as a person-trapped event is determined; at the same time, through the elevator Internet of Things module, the person-trapped event related information is pushed to the property monitoring platform and the elevator maintenance unit in real time; if no response is received within the preset time after pushing, the urban elevator emergency management department is automatically linked, and the emergency rescue process is started.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by a processor to implement the elevator trapping alarm method based on signal detection and visual recognition according to any one of claims 1-8.

Images