Landing call intent recognition method and device, and elevator landing call control system

Abstract

The present invention relates to a landing call intent recognition method, comprising: performing backtracking on a surveillance video of passenger button pressing and extracting key frames to obtain boarding key frames; sequentially performing button press detection and direction intent recognition on a plurality of boarding key frames to obtain a button press frame and direction intent information, and on the basis of the direction intent information, establishing an association between the button press frame and the plurality of boarding key frames to obtain a same-batch monitoring list; and on the basis of the same-batch monitoring list, performing target tracking on the plurality of boarding key frames, and on the basis of a target tracking result, determining whether a current passenger has a change in boarding intent: if not, continuing the target tracking, and if yes, considering that the passenger intends to stop calling an elevator. The landing call intent recognition method of the present invention has the following advantages: significantly reducing misjudgments in complex elevator waiting area scenarios; effectively reducing computational resources required for target tracking; enhancing resource utilization efficiency; and improving the stability and real-time performance of boarding intent recognition.

Description

A method, device, and elevator call control system for identifying elevator call intentions. Technical Field

[0001] This invention relates to the fields of intelligent transportation and elevator dispatching, and in particular to a method, device, elevator dispatching control system, electronic equipment, and computer storage medium for identifying the intention to call an elevator. Background Technology

[0002] With the continuous advancement of social and economic development and urbanization, elevators have become an indispensable vertical transportation tool in modern buildings. The basic working principle of a typical elevator is as follows: The elevator external call control system receives a passenger's call button (up or down) pressed on the elevator call box, as shown in Figure 1. Based on the corresponding call button, the control system directs the elevator to that floor, allowing the passenger to board, as shown in Figure 2. Then, the passenger selects their desired floor on the internal floor panel, and the elevator moves smoothly to that floor at a constant speed or with uniform acceleration. If, during its movement, a passenger presses the external call button in the same direction on a floor along its path, the elevator will decelerate and arrive at that floor to pick up passengers moving in the same direction.

[0003] However, in practical applications, changes in passenger intent often occur, leading to elevators running empty. Specific situations include: passengers pressing the wrong elevator call button for various reasons, passengers on lower floors waiting too long and choosing to take the stairs, or passengers changing their travel plans and stopping the elevator altogether. As a result, the elevator arrives at the corresponding floor after receiving a call, but no passengers board. This situation not only severely impacts the elevator experience for other passengers but also extends the waiting time for the next round, thus reducing the overall operating efficiency of the elevator.

[0004] Based on this, existing technologies typically employ passenger identification and passenger intent recognition to optimize elevator scheduling in order to avoid elevators running empty. Passenger identification is suitable for residential and office buildings where privacy is paramount, but its effectiveness is limited in buildings with extensive public areas. Passenger intent recognition is primarily performed through the following steps: real-time acquisition and preprocessing of surveillance video outside each elevator car; pose estimation of the preprocessed video to obtain the skeletal nodes and number of human skeletons for each passenger; analysis of key point coordinates in the skeletal node map using a support vector machine behavior analysis model to determine whether the current passenger intends to board the elevator; and the elevator external call control system calls the elevator based on the passenger's intent, such as whether to move the elevator to the current passenger's waiting floor for entry.

[0005] However, in buildings with multiple public areas, such as shopping malls and hospitals, where people are densely populated and move frequently, existing technologies that rely solely on support vector machines to determine elevator intentions are prone to misjudgments. Furthermore, the acquisition, preprocessing, attitude estimation, and behavior analysis of surveillance videos require significant computational resources. As the computational load increases, the system response speed may slow down, affecting the real-time performance and accuracy of elevator scheduling. Summary of the Invention

[0006] Based on this, the purpose of the present invention is to provide a method for identifying the intention to call an elevator.

[0007] A method for identifying the intention to use an elevator by external call, comprising the following steps:

[0008] S1: Review the monitoring video of passengers pressing buttons and extract key frames to obtain elevator key frames;

[0009] S2: Sequentially perform button detection and directional intent recognition on several key frames of elevator ride to obtain button action frames and directional intent information, and establish a correlation between button action frames and several key frames of elevator ride based on directional intent information to obtain a monitoring list of the same batch.

[0010] S3: Based on the monitoring list of the same batch, target tracking is performed on several key frames of elevator rides, and the result of target tracking is used to determine whether the current passenger's intention to ride the elevator has changed. If not, target tracking continues; if so, it is assumed that the passenger intends to stop calling the elevator.

[0011] The elevator booking intention recognition method described in this invention, compared to existing technologies, adopts a bidirectional single-round detection strategy—specifically, key press action detection—by combining practical application scenarios. This strategy focuses solely on recognizing and tracking the elevator booking intentions of the first batch of passengers, representing different batches of passengers waiting in the same direction. This significantly reduces misjudgments in complex elevator waiting areas. Furthermore, by tracking targets based on several elevator booking keyframes, the system effectively reduces the computational resources required for target tracking, achieving more efficient resource utilization and improving the stability and real-time performance of elevator booking intention recognition. This provides a more intelligent solution for elevator scheduling in complex scenarios.

[0012] Further, step S1 includes the following sub-steps:

[0013] S11: Review the monitoring video of passengers pressing buttons to obtain relevant video segments;

[0014] For the current time t, the relevant video segment is specifically represented as t. related ∈[ts,e],s <t

[0015] In the formula, t related[ts,e] represents the relevant video segment; [ts,e] represents the video segment within s seconds backward from the current time t, where s is the preset interaction time and e represents the time of the elevator intention analysis result, i.e., the current time of the monitoring video.

[0016] Get relevant video segment t related The corresponding video frame sequence Frame n The keyframes were extracted using the inter-frame difference method, resulting in several elevator keyframes. The specific calculation expression is as follows:

[0017] For consecutive frames i and Frame i+1 The average difference intensity D between i Represented as:

[0018] In the formula, Frame i (x,y) represents the pixel value of the i-th frame at pixel coordinates (x,y), W and H represent the width and height of each frame, respectively; i belongs to the interval [1,n], and n is the total number of the video frame sequence, which is calculated based on the monitored refresh rate fps, specifically: fps×(e-t+s); then, a differential threshold is used to evaluate all average differential intensity sets D={D1,D2,…,D n-1 After filtering, several elevator keyframes are obtained, and their differential threshold is: D Th =max(D)-ΔD var

[0019] In the formula, max(·) represents the maximum average difference intensity in the current set of all average difference intensities D; ΔD var A custom differential intensity variable; several elevator keyframes (KeyFrame) are represented as: KeyFrame = {(Frame...} i ,t j )|i∈[1,n],t j ∈[ts,e]}

[0020] In the formula, Frame i This represents the i-th frame in the current video frame sequence; t j Represented as a video frame i The timestamp it belongs to.

[0021] This invention uses the instantaneous time of the indicator light illuminating after a passenger presses a button to backtrack, ensuring that the passenger's button-pressing action is successfully captured within the time frame of the surveillance video. It also uses average differential intensity to highlight key action frames, providing representative data for subsequent target recognition and tracking. This ensures the system can accurately identify passenger intent even in complex scenarios, reducing the possibility of misjudgments. Furthermore, by processing only key frames within the relevant time frame, the system effectively reduces data processing volume and optimizes the utilization of computing resources, significantly improving the system's real-time performance and response speed. This allows the system to accurately identify passenger intent in complex elevator waiting scenarios and quickly respond to elevator scheduling, preventing empty elevator runs and improving the overall system efficiency and reliability.

[0022] Further, step S2 includes the following sub-steps:

[0023] S21: Using an image classification model, the data at timestamp t is classified. j Key press detection is performed on several keyframes within [ts,t] to obtain key press action frames;

[0024] S22: Determine whether the current button action frame is a single frame or a double frame: If it is a single frame, obtain the directional intention information based on the direction of the indicator light triggered when the passenger presses the button; if it is a double frame, determine the timestamp order of the double frames based on the order of the directions of the indicator lights triggered when the passenger presses the button, and obtain the directional intention information.

[0025] S23: Based on the directional intent information, identify the unique feature information of the passenger who performed the key press action in the key press action frame, and use it as the Key Passenger KeyID, which is represented as KeyID = {Feature KeyID Based on the unique characteristic information of key passengers, the human skeleton spatiotemporal relationship model M is adopted. r Correlation detection is performed on several keyframes of elevator rides to obtain the set of passengers traveling together, PID, specifically represented as: PID = {M} r (Feature KeyID ,(Frame i ,t j ))|i∈[1,n],t j ∈[ts,t]}={PID1,PID2,…,PID i}

[0026] In the formula, M r (Feature KeyID ,(Frame i ,t j )) represents the unique characteristics of key passengers based on the input; (Frame) i ,t j) represents the elevator keyframes within the time interval [ts,t].

[0027] By associating key passengers and their accompanying passengers, a monitoring list for the same batch is obtained, which is represented as: Group up ={KeyID,PID1,PID2,…,PID i}

[0028] And / or, Group down ={KeyID,PID1,PID2,…,PID j}

[0029] In the formula, Group up Group is used to represent the same batch of monitoring lists in the upward direction. down Used to represent the same batch of monitored passengers in the downward direction; i and j represent the total number of passengers traveling in the same direction in the upward and downward directions, respectively; where, PID i This includes the posture and motion characteristics of the i-th passenger.

[0030] This invention identifies the passenger who presses the first button as a key passenger and associates them with their companions. This allows the system to focus on key passengers and their companions, effectively determining the passenger's intention to board the elevator before its arrival. Simultaneously, it reduces the need for comprehensive monitoring of all waiting passengers, centralizing the processing of the boarding intentions of the first batch of passengers traveling in the same direction. This further reduces unnecessary resource consumption and increases the system's real-time performance and response speed. Furthermore, the association strategy of this invention improves the system's robustness and stability in complex scenarios, ensuring that even when multiple batches of passengers are waiting simultaneously, the system can accurately identify and respond to passengers' actual needs, and precisely track and manage key passengers and their companions.

[0031] Further, step S3 includes the following sub-steps:

[0032] S31: Based on the monitoring list of the same batch (Group) up and / or Group down The target tracking model M is adopted. Tracking Target tracking is performed on several passenger keyframes to obtain several real-time action (Act) and trajectory (Track) lists. For the key passenger (KeyID) in the same batch of monitoring lists, the specific expression for target tracking is as follows: M Tracking (KeyID,(Frame i ,t j )),t j ∈[et,e] ={Act KeyID ={act1,...,act iTrack KeyID ={position1,...,position i}}

[0033] In the formula, act i The position represents the action type corresponding to the key passenger in frame i. i This indicates the location of the key passenger in the i-th frame; in the time interval [et, e], e is updated in real time according to the current time of the surveillance video.

[0034] S32: Based on the real-time action and trajectory list, determine whether the current passenger's intention to use the elevator has changed: if yes, it is assumed that the passenger intends to stop calling the elevator; if no, proceed to step S31.

[0035] Among them, for the same batch of monitoring list Group up and / or Group down The expression for determining the elevator intention of the j-th passenger is as follows:

[0036] In the formula, act i Let be the action type of the i-th frame, and act i The action list of the j-th passenger in the same batch of monitoring list; distance(·) is used to calculate the distance between two input data, and left is the distance threshold for determining the position of leaving the elevator; while This is used to determine if, among all subsequent frames k, there exists at least one passenger in frame k whose distance from the elevator is less than or equal to the distance threshold; and This is used to determine the case where, in all subsequent frames k, the distance between the passenger and the elevator is greater than the stated distance threshold; E represents the elevator's coordinate position; position k Let position be the passenger trajectory position in the k-th frame, and position k The trajectory list of the j-th passenger belonging to the same batch of monitoring list; where k∈[i,n], and n is the total number of video frame sequences.

[0037] This invention meticulously analyzes the actions and trajectories of all passengers in the same batch of monitoring lists, and sets clear conditions for judgment. That is, only when all passengers in the same batch of monitoring lists are in a moving state and have moved away from the elevator to a certain distance is it determined that the passengers collectively intend to leave the elevator waiting area, and this is used as the condition for stopping the elevator call. This ensures that the intention of a single passenger will not affect the elevator experience of other passengers in the same batch, and also significantly reduces the phenomenon of elevator running empty due to misjudgment, thus optimizing the overall scheduling efficiency of the elevator system.

[0038] A device for identifying the intention to take an elevator by external call includes a monitoring keyframe acquisition unit, a target direction intention association unit for the same batch, and an elevator intention identification unit;

[0039] The monitoring key frame acquisition unit is used to backtrack the monitoring video of the passenger button and extract key frames to obtain elevator key frames.

[0040] The same batch target directional intent association unit is used to sequentially perform key detection and directional intent recognition on several elevator key frames to obtain key action frames and directional intent information, and establish an association between key action frames and several elevator key frames based on directional intent information to obtain a same batch monitoring list.

[0041] The elevator intention recognition unit is used to perform target tracking on several elevator key frames according to the same batch monitoring list, and determine whether the current passenger's elevator intention has changed based on the target tracking result: if not, then continue target tracking; if so, then consider that the passenger intends to stop calling the elevator.

[0042] Furthermore, the monitoring key frame acquisition unit includes a monitoring backtracking module and a monitoring key frame extraction module;

[0043] The monitoring backtracking module is used to backtrack the monitoring video of passengers pressing buttons and obtain relevant video segments;

[0044] For the current time t, the relevant video segment is specifically represented as t. related ∈[ts,e],s <t

[0045] In the formula, t related [ts,e] represents the relevant video segment; [ts,e] represents the video segment within s seconds backward from the current time t, where s is the preset interaction time and e represents the time of the elevator intention analysis result, i.e., the current time of the monitoring video.

[0046] The monitoring keyframe extraction module is used to obtain relevant video segments t related The corresponding video frame sequence Frame n The keyframes were extracted using the inter-frame difference method, resulting in several elevator keyframes. The specific calculation expression is as follows:

[0047] For consecutive frames i and Frame i+1 The average difference intensity D between i Represented as:

[0048] In the formula, Frame i(x,y) represents the pixel value of the i-th frame at pixel coordinates (x,y), W and H represent the width and height of each frame, respectively; i belongs to the interval [1,n], and n is the total number of the video frame sequence, calculated based on the monitored refresh rate fps, specifically: fps×(e-t+s); then, a differential threshold is used to evaluate all average differential intensity sets D={D1,D2,...,D n-1 After filtering, several elevator keyframes are obtained, and their differential threshold is: D Th =max(D)-ΔD var

[0049] In the formula, max(·) represents the maximum average difference intensity in the current set of all average difference intensities D; ΔD var A custom differential intensity variable; several elevator keyframes (KeyFrame) are represented as: KeyFrame = {(Frame...} i ,t j )|i∈[1,n],t j ∈[ts,e]}

[0050] In the formula, Frame i This represents the i-th frame in the current video frame sequence; t j Represented as a video frame i The timestamp it belongs to.

[0051] Furthermore, the target directional intent association unit in the same batch includes a key detection module, a directional intent annotation module, and a directional intent association module;

[0052] The key detection module is used to employ an image classification model to detect keys at timestamp t. j Key press detection is performed on several keyframes within [ts,t] to obtain key press action frames;

[0053] The directional intent annotation module is used to determine whether the current key press action frame is a single frame or a double frame: if it is a single frame, the directional intent information is obtained according to the direction of the indicator light triggered when the passenger presses the key; if it is a double frame, the directional intent information is obtained by determining the timestamp order of the double frames according to the order of the directions of the indicator lights triggered when the passenger presses the key.

[0054] The directional intent association module is used to identify the unique feature information of the passenger performing the key press action in the key press action frame based on the directional intent information, and use it as the Key Passenger KeyID, which is represented as KeyID = {Feature KeyID Based on the unique characteristic information of key passengers, the human skeleton spatiotemporal relationship model M is adopted. rCorrelation detection is performed on several keyframes of elevator rides to obtain the set of passengers traveling together, PID, specifically represented as: PID = {M} r (Feature KeyID ,(Frame i ,t j ))|i∈[1,n],t j ∈[ts,t]}={PID1,PID2,...,PID i}

[0055] In the formula, M r (Feature KeyID ,(Frame i ,t j )) represents the unique characteristics of key passengers based on the input; (Frame) i ,t j ) represents the elevator keyframes within the time interval [ts,t].

[0056] By associating key passengers and their accompanying passengers, a monitoring list for the same batch is obtained, which is represented as: Group up ={KeyID,PID1,PID2,…,PID i}

[0057] And / or, Group down ={KeyID,PID1,PID2,...,PID j}

[0058] In the formula, Group up Group is used to represent the same batch of monitoring lists in the upward direction. down Used to represent the same batch of monitored passengers in the downward direction; i and j represent the total number of passengers traveling in the same direction in the upward and downward directions, respectively; where, PID i This includes the posture and motion characteristics of the i-th passenger;

[0059] The elevator intention recognition unit includes a batch target tracking module and a batch target elevator intention judgment module;

[0060] The same batch target tracking module is used to track targets based on the same batch monitoring list (Group). up and / or Group down The target tracking model M is adopted. Tracking Target tracking is performed on several passenger keyframes to obtain several real-time action (Act) and trajectory (Track) lists. For the key passenger (KeyID) in the same batch of monitoring lists, the specific expression for target tracking is as follows: M Tracking (KeyID,(Framei ,t j )),t j ∈[et,e] ={Act KeyID ={act1,…,act i Track KeyID ={position1,…,position i}}

[0061] In the formula, act i The position represents the action type corresponding to the key passenger in frame i. i This indicates the location of the key passenger in the i-th frame; in the time interval [et, e], e is updated in real time according to the current time of the surveillance video.

[0062] The same batch of target elevator intention judgment module is used to determine whether the current passenger has changed their elevator intention based on the real-time action and trajectory list: if yes, it is considered that the passenger intends to stop calling the elevator; if no, step S31 is executed.

[0063] Among them, for the same batch of monitoring list Group up and / or Group down The expression for determining the elevator intention of the j-th passenger is as follows:

[0064] In the formula, act i Let be the action type of the i-th frame, and act i The action list of the j-th passenger in the same batch of monitoring list; distance(·) is used to calculate the distance between two input data, and left is the distance threshold for determining the position of leaving the elevator; while This is used to determine if, among all subsequent frames k, there exists at least one passenger in frame k whose distance from the elevator is less than or equal to the distance threshold; and This is used to determine the case where, in all subsequent frames k, the distance between the passenger and the elevator is greater than the stated distance threshold; E represents the elevator's coordinate position; position k Let position be the passenger trajectory position in the k-th frame, and position k The trajectory list of the j-th passenger belonging to the same batch of monitoring list; where k∈[i,n], and n is the total number of video frame sequences.

[0065] An elevator external call control system includes a plurality of camera devices, a plurality of elevator external call buttons, and an external call intention recognition device that is electrically and / or communicatively connected to the plurality of camera devices and the plurality of elevator external call buttons.

[0066] The aforementioned camera devices are installed in the elevator waiting area to monitor the elevator call buttons on the corresponding floors and send the monitoring video to the elevator call intention recognition device.

[0067] The aforementioned elevator external call buttons are located in the elevator waiting area, allowing passengers to press the corresponding button in the appropriate direction to call an elevator to the corresponding waiting floor.

[0068] The external elevator call intention recognition device is used to identify whether the current passenger's elevator call intention has changed by analyzing the surveillance video: if not, the recognition continues; if so, the external elevator call button in the corresponding waiting area is canceled, and the elevator call is cancelled.

[0069] The device for identifying the intention to call for elevator service is the same as the device for identifying the intention to call for elevator service described above.

[0070] To better understand and implement this invention, the following detailed description is provided in conjunction with the accompanying drawings. Attached Figure Description

[0071] Figure 1 is a schematic diagram of the appearance of the elevator external call button;

[0072] Figure 2 is a schematic diagram illustrating an example of the elevator usage process;

[0073] Figure 3 is a schematic diagram of the structure of the external call elevator intention recognition device of the present invention;

[0074] Figure 4 is a schematic diagram of the method for identifying the intention to take an elevator by external call according to the present invention;

[0075] Figure 5 is a schematic diagram of the target identification and tracking strategy of the present invention. Detailed Implementation

[0076] To address the shortcomings of existing technologies in determining elevator travel intentions, particularly in terms of real-time performance and accuracy, this invention acquires key frames from surveillance video and identifies passenger button presses within these frames. This process generates button press frames and corresponding passenger directional intention information. A correlation is established between the button press frames and key frames based on this directional intention information, creating a batch monitoring list. This list is then used to track corresponding passengers, resulting in a real-time action and trajectory list. The system continuously assesses whether a passenger's travel intention has changed based on the real-time action and trajectory list. If so, a cancellation command is sent to the elevator dispatch control system. Therefore, by constructing a batch monitoring list, this invention enables the system to more accurately track and analyze the behavior and trajectories of passenger groups, thereby reducing misjudgments caused by individual behaviors. Furthermore, the correlation analysis based on key frames not only improves the accuracy of determining travel intentions but also reduces unnecessary computational burden caused by misjudgments, significantly enhancing the system's real-time performance.

[0077] Based on the above design, this invention proposes a method for identifying the intention to call an elevator, and based on this method, proposes a device for identifying the intention to call an elevator.

[0078] Please refer to Figures 3 and 4. Figure 3 is a structural schematic diagram of the external call elevator intention recognition device according to the present invention, and Figure 4 is a flowchart of the external call elevator intention recognition method according to the present invention.

[0079] The device for identifying the intention to take the elevator includes a monitoring keyframe acquisition unit 1, a target direction intention association unit 2, and an elevator intention identification unit 3.

[0080] The monitoring key frame acquisition unit 1 is used to perform step S1: backtrack the monitoring video of the passenger pressing the buttons and extract key frames to obtain elevator key frames.

[0081] Specifically, the monitoring key frame acquisition unit 1 includes a monitoring backtracking module 11 and a monitoring key frame extraction module 12.

[0082] The monitoring backtracking module 11 is used to perform step S11: backtrack the monitoring video of the passenger's button presses to obtain relevant video segments.

[0083] Specifically, for the current time t, the relevant video segment is represented as ti. related ∈[ts,e],s <t

[0084] In the formula, t related [ts,e] represents the relevant video segment; [ts,e] represents the video segment within s seconds backward from the current time t, where s is the preset interaction time, which defaults to 1-2 seconds, and e represents the time of the elevator intention analysis result, that is, the current time of the monitoring video (real-time), until the elevator intention analysis result is obtained.

[0085] Depending on the different elevator systems or actual scenarios, such as shopping malls and residences, the preset interaction time s may vary. Therefore, this invention does not specifically limit the time, but only needs to ensure that the relevant video segment includes the time period from when the passenger presses the button to when the indicator light illuminates.

[0086] This invention captures the time segment when a passenger presses the elevator call button, thereby ensuring that all behaviors and contexts related to the passenger's button press are captured, providing key data for subsequent intent analysis.

[0087] The monitoring keyframe extraction module 12 is used to perform step S12: extract keyframes from relevant video segments to obtain several elevator keyframes.

[0088] Specifically, obtain the relevant video segment t related The corresponding video frame sequence Framen The keyframes were extracted using the inter-frame difference method, resulting in several elevator keyframes. The specific calculation expression is as follows:

[0089] For consecutive frames i and Frame i+1 The average difference intensity D between i Represented as:

[0090] In the formula, Frame i (x,y) represents the pixel value of the i-th frame at pixel coordinates (x,y), and W and H represent the width and height of each frame, respectively; i∈[1,n] and n is the total number of the video frame sequence, calculated based on the monitored refresh rate fps×(e-t+s), and n increases in real time according to the time e of the elevator intention analysis result. The refresh rate fps varies depending on the device performance, and is not specifically limited here.

[0091] Next, a difference threshold is applied to all average difference intensity sets D = {D1, D2, ..., D...} n-1 After filtering, several elevator keyframes are obtained, and their differential threshold D is determined. Th For: D Th =max(D)-ΔD var

[0092] In the formula, max(·) represents the maximum average difference intensity in the current set of all average difference intensities D; ΔD var This is a custom differential intensity variable. Different differential intensity variables can be defined according to specific needs to adjust the acquisition range of elevator keyframes. This invention does not specifically limit the value of its differential intensity variable.

[0093] Its keyframes for elevator access can be represented as: KeyFrame = {(Frame i ,t j )|i∈[1,n],t j ∈[ts,e]}

[0094] In the formula, Frame i This represents the i-th frame in the current video frame sequence; t j Represented as a video frame i The timestamp it belongs to.

[0095] This invention extracts keyframes from a large number of surveillance video frames to identify those that best reflect passenger behavior. This highlights the elevator-riding intentions of the most frequent passengers and relevant contextual information, while effectively removing redundant information from the video frames, thereby improving the system's accuracy in determining elevator-riding intentions. Simultaneously, it reduces the amount of image data the system needs to process, lowering computational resource requirements and improving the overall processing efficiency of the system.

[0096] Furthermore, this invention uses differential intensity variables to adjust the screening range of elevator keyframes, ensuring that the system can select the most representative keyframes according to the actual situation, rather than relying solely on a fixed differential intensity threshold, in order to adapt to various complex situations, optimize the keyframe extraction effect, and thus improve the overall system performance and reliability.

[0097] The same batch target directional intent association unit 2 is used to perform step S2: sequentially perform key detection and directional intent recognition on several elevator key frames to obtain key action frames and directional intent information, and establish an association between key action frames and several elevator key frames based on directional intent information to obtain a same batch monitoring list.

[0098] Specifically, the same batch of target directional intent association unit 2 includes a key detection module 21, a directional intent annotation module 22, and a directional intent association module 23.

[0099] The button detection module 21 is used to perform step S21: perform button detection on several elevator key frames to obtain button action frames.

[0100] Specifically, an image classification model is adopted. a For timestamp t j Key press detection is performed on several keyframes within [ts,t] to obtain key press action frames.

[0101] The image classification model a The system includes a feature extraction module, a feature fusion module, and a key classification module. The feature extraction module extracts features from several elevator keyframes to obtain feature maps of different scales. The feature fusion module fuses these feature maps to obtain fused feature maps of different scales. The key classification module detects key presses on the fused feature maps of different scales to obtain key action frames.

[0102] Among them, the image classification model aYOLOv10 (You Only Look Once Version 10) can be selected, and a training dataset is constructed by manually labeling the buttons and human actions of people waiting in the elevator car video. Then, this training dataset is used to train the image classification model. a Train and validate the model to make it a If the accuracy of recognizing buttons and elevators meets the standard, then the trained image classification model is obtained. a .

[0103] Since different image classification models can be selected based on different hardware conditions, computing resources, or accuracy requirements, this invention does not specifically limit the structure and selection of its models.

[0104] It is particularly important to emphasize that the button press action frames obtained by this invention are actually the video frames of the indicator light illuminating after the first passenger presses the button. If the indicator light remains on, repeated button presses by other passengers in the same batch will not be captured. Therefore, this invention uses the illumination of the indicator light after a button press as the starting point for detecting a batch of waiting passengers.

[0105] The directional intent annotation module 22 is used to perform step S22: annotate the key action frame through time correspondence to obtain directional intent information.

[0106] Specifically, determine whether the current button action frame is a single frame or a double frame: if it is a single frame, obtain the directional intention information based on the direction of the indicator light triggered when the passenger presses the button;

[0107] If it is a two-frame sequence, the directional intention information is obtained by determining the timestamp sequence of the two frames based on the order of the directions of the indicator lights triggered when the passenger presses the button.

[0108] Since there is a temporal sequence in which the indicator lights illuminate after a passenger presses a button, and the passenger's destination is only either up or down, and they will not choose both directions simultaneously, the directional intention of the key passenger can be accurately determined simply by following the order in which the indicator lights illuminate corresponding to the button press action frame. This allows the directional intention marking to be completed efficiently and accurately, ensuring the intelligent response of the elevator dispatching system.

[0109] The directional intent association module 23 is used to perform step S23: based on the directional intent information, establish an association between the button action frame and several elevator key frames to obtain a monitoring list of the same batch.

[0110] Specifically, based on directional intent information, the unique feature information of the passenger performing the key press action in the key press action frame is identified and used as the Key Passenger KeyID, which is represented as KeyID = {Feature KeyID}

[0111] Based on the unique characteristics of key passengers, the human skeleton spatiotemporal relationship model M is adopted. r Correlation detection is performed on several keyframes of elevator rides to obtain the set of passengers traveling together, PID, specifically represented as: PID = {M} r (Feature KeyID ,(Frame i ,t j ))|i∈[1,n],t j ∈[ts,t]}={PID1,PID2,…,PID i}

[0112] In the formula, M r (Feature KeyID ,(Frame i ,t j )) represents the unique characteristics of key passengers based on the input; (Frame) i ,t j ) represents the elevator keyframe within the time interval [ts,t].

[0113] By associating key passengers and their accompanying passengers, a monitoring list for the same batch is obtained, which is represented as: Group up ={KeyID,PID1,PID2,…,PID i}

[0114] And / or, Group down ={KeyID,PID1,PID2,…,PID j}

[0115] In the formula, Group up Group is used to represent the same batch of monitoring lists in the upward direction. down This is used to represent the list of monitored passengers in the same batch in the downward direction. i and j represent the total number of passengers traveling in the same direction in the upward and downward directions, respectively.

[0116] The same batch monitoring list can determine whether a passenger group belongs to the uphill, downhill, or both, based on the directional intent information, depending on the key action frame corresponding to the given directional intent information.

[0117] Among them, the spatiotemporal relationship model M of the human skeleton r It includes a posture estimation module, a spatiotemporal feature dimension extraction module, a spatiotemporal feature fusion module, and a PID generation module. The posture estimation module is used to extract the posture of elevator keyframes within the time interval [ts,t] to obtain the posture features of all passengers.

[0118] The spatiotemporal feature extraction module is used to extract spatial and temporal features from the posture features of all passengers and obtain several spatiotemporal feature information.

[0119] The spatiotemporal feature fusion module is used to fuse the spatiotemporal feature information of each passenger to generate the action features of each passenger.

[0120] The PID generation module is used to match and identify passengers traveling with the key passenger based on the unique characteristics of the key passenger and the action characteristics of all passengers, obtain the characteristics of passengers traveling with the key passenger, assign corresponding PIDs to the traveling passengers, and output the set of traveling passengers PID. Wherein, PID = {PID1, PID2, ..., PID...} i}, and PID i This includes the posture and motion features of the i-th fellow passenger, which provides sufficient information for subsequent target tracking.

[0121] Furthermore, the elevator keyframes within the time interval [ts,t] input to the human skeleton spatiotemporal relationship model can be dynamically expanded. Specifically, based on the current elevator keyframe (Frame... i ,t j ), expanding to a sequence of neighboring video frames as input, the expression for which is as follows: {Frame i-c ,…,Frame i ,…,Frame i+c |c≤fps}

[0122] By providing more temporal information to the human skeleton spatiotemporal model through adjacent video frame sequences, the model's accuracy in recognizing fellow passengers can be improved.

[0123] Furthermore, the spatiotemporal relationship model M of the human skeleton r An optional Spatial Temporal Graph Convolutional Network (ST-GCN) is used, trained on a self-constructed training dataset of fellow passengers, to identify and assign PIDs to fellow passengers. However, users can choose different spatiotemporal relationship models to identify and assign PIDs to fellow passengers based on different needs and different keyframe extraction methods. Therefore, this invention does not specifically limit the choice of the human skeleton spatiotemporal relationship model.

[0124] In real-world elevator scenarios, once the first group of passengers presses the up or down button (indicating the indicator light), subsequent passengers do not need to press the button again; they simply wait until the indicator light goes out. Therefore, for the target of interest, determining the intent of the first group of passengers is sufficient to indicate whether there are passengers waiting in the current round, as shown in Figure 5. This reduces the computational burden of target recognition and tracking while ensuring the system's effectiveness in recognizing changes in passenger intent.

[0125] The elevator intention recognition unit 3 is used to perform step S3: according to the same batch monitoring list, target tracking is performed on several elevator key frames, and the current passenger's elevator intention is determined based on the target tracking result: if not, target tracking continues; if so, it is considered that the passenger intends to stop calling the elevator.

[0126] Specifically, the elevator intention recognition unit 3 includes a batch target tracking module 31 and a batch target elevator intention judgment module 32.

[0127] The same batch target tracking module 31 is used to perform step S31: according to the same batch monitoring list, target tracking is performed on several elevator key frames to obtain several real-time action and trajectory lists.

[0128] Specifically, according to the same batch monitoring list Group up and / or Group down The target tracking model M is adopted. Tracking Target tracking is performed on several passenger keyframes to obtain several real-time action (Act) and trajectory (Track) lists. For the key passenger (KeyID) in the same batch of monitoring lists, the specific expression for target tracking is as follows: M Tracking (KeyID,(Frame i ,t j )),t j ∈[et,e] ={Act KeyID ={act1,…,act i Track KeyID ={position1,…,position i}}

[0129] In the formula, act i The position represents the action type corresponding to the key passenger in frame i. i This represents the location of the key passenger in the i-th frame. In the time interval [et, e], e is a dynamically selected value, updated in real time according to the current time. It represents the target tracking interval from the associated passenger to the current time. For example, the target tracking interval t in the first period... j∈[10s,20s], in the next cycle when e = e + 10, the target tracking interval is updated to t. j ∈[20s,30s].

[0130] Wherein, the target tracking model M Tracking It includes a feature extraction module, a feature fusion module, a target association module, an action recognition module, and a trajectory recognition module.

[0131] The feature extraction module is used to extract features from several passenger keyframes to obtain several feature vectors.

[0132] The feature fusion module is used to fuse several feature vectors to obtain several passenger fused feature vectors.

[0133] The target association module is used to match the passenger fusion feature vector with the passenger information in the same batch monitoring list to obtain a matching fusion feature vector, so as to ensure continuous tracking of the target passenger.

[0134] The action recognition module is used to perform action recognition on the matched and fused feature vectors to obtain a list of actions corresponding to each passenger, specifically: Act j ={act1,…,act i}

[0135] In the formula, j is the j-th passenger in the same batch of monitoring list, act i The action type is the action type corresponding to the i-th frame. The action type can be selected as movement or waiting for a ladder. Different action types can be selected according to different scenarios and requirements. This invention does not specifically limit the types.

[0136] The trajectory recognition module is used to track the trajectory of the matched and fused feature vectors to obtain a list of trajectories for each passenger, specifically: Track j ={position1,…,position i}

[0137] In the formula, j is the j-th passenger in the same batch of monitoring list, and position i The trajectory coordinates corresponding to the i-th frame can be joint coordinates, detection box coordinates, etc., and are not specifically limited in this invention.

[0138] Furthermore, the target tracking model M Tracking An optional DeepSORT multi-object tracking algorithm can be used, combined with the aforementioned human skeleton spatiotemporal relationship model M. rAssociation is established to achieve multi-target tracking. However, users can choose different target tracking models to achieve multi-target tracking based on different needs and different types of data input. Therefore, this invention does not specifically limit the choice of target tracking model.

[0139] The same batch of target elevator intention judgment module 32 is used to execute step S32: based on the real-time action and trajectory list, determine whether the current passenger has changed their elevator intention; if so, send a cancel external call control command to the elevator external call control system; if not, call the same batch of target tracking module 31.

[0140] Specifically, for the same batch of monitoring lists (Group) up and / or Group down The expression for determining the elevator intention of the j-th passenger is as follows:

[0141] In the formula, act i Let be the action type of the i-th frame, and act i The action list of the j-th passenger in the same batch of monitoring list; distance(·) is used to calculate the distance between two input data, left is the distance threshold for judging the position of leaving the elevator, and different distance thresholds can be defined in actual situations. This invention does not specifically limit the choice of the distance threshold; and This is used to determine if, among all subsequent frames k, there exists at least one passenger in frame k whose distance from the elevator is less than or equal to the distance threshold; and This is used to determine if, in all subsequent frames k, the distance between the passenger and the elevator is greater than the stated distance threshold; E represents the elevator's coordinates, which are known data and can be defined according to the actual scenario; position k Let position be the passenger trajectory position in the k-th frame, and position k The trajectory list of the j-th passenger belonging to the same batch of monitoring list. Where k∈[i,n], and n is the total number of video frame sequences.

[0142] Therefore, if the monitoring list of the same batch is Group up and / or Group down If at least one passenger's intention to use the elevator is to wait for it, then it is assumed that the current passenger's intention to use the elevator has not changed, and the same batch target tracking module 31 is invoked. If the same batch monitoring list Group up and / or Group down If all passengers' intention to use the elevator is to leave, it is considered that the current passenger's intention to use the elevator has changed, and a cancellation command is sent to the elevator call control system to cancel the elevator call.

[0143] This invention analyzes real-time actions and trajectories to accurately distinguish whether a passenger is waiting for or leaving the elevator. Furthermore, to ensure that waiting passengers are not misjudged, this invention only determines a change of intent to ride the elevator when all passengers in the same batch of monitoring lists show signs of leaving. This balances the passenger experience for both waiting passengers and passengers already in the elevator, thereby improving passenger satisfaction.

[0144] Compared to existing technologies, this invention, combined with real-world elevator riding scenarios, focuses on target identification and tracking of the first batch of passengers, representing the riding intentions of other batches of passengers traveling in the same direction. This enables the system to effectively identify passengers' true intentions in complex scenarios, significantly reducing the false positive rate. Simultaneously, this invention performs behavioral analysis based on keyframes of elevator riding, ensuring the representativeness of the analysis results, thereby reducing the system's computational resource consumption and improving the real-time performance and accuracy of elevator dispatching responses.

[0145] Based on the same inventive concept, this application also provides an electronic device, which can be a server, desktop computing device, or mobile computing device (e.g., laptop computing device, handheld computing device, tablet computer, netbook, etc.) or other terminal device. The device includes one or more processors and a memory, wherein the processor is used to execute a program to implement the method for recognizing the intention to call for elevators according to embodiments of the present invention; the memory is used to store computer programs executable by the processor.

[0146] Based on the same inventive concept, this application also provides a computer-readable storage medium corresponding to the aforementioned embodiment of a method for identifying the intention to call an elevator. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the method for identifying the intention to call an elevator as described in any of the above embodiments.

[0147] This application may take the form of a computer program product implemented on one or more storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing program code. Computer storage media include permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information may be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to: phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0148] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and the present invention also intends to include these modifications and variations.

Claims

1. A method for recognizing the intention to call an elevator, characterized in that, Includes the following steps: S1: Review the monitoring video of passengers pressing buttons and extract key frames to obtain elevator key frames; S2: Sequentially perform button detection and directional intent recognition on several key frames of elevator ride to obtain button action frames and directional intent information, and establish a correlation between button action frames and several key frames of elevator ride based on directional intent information to obtain a monitoring list of the same batch. S3: Based on the monitoring list of the same batch, target tracking is performed on several key frames of elevator rides, and the result of target tracking is used to determine whether the current passenger's intention to ride the elevator has changed. If not, target tracking continues; if so, it is assumed that the passenger intends to stop calling the elevator.

2. The method for identifying the intention to call an elevator according to claim 1, characterized in that, Step S1 includes the following sub-steps: S11: Review the monitoring video of passengers pressing buttons to obtain relevant video segments; For the current time t, the relevant video segment is specifically represented as follows: t related ∈[t-s,e],s<t In the formula, t related [ts,e] represents the relevant video segment; [ts,e] represents the video segment within s seconds backward from the current time t, where s is the preset interaction time and e represents the time of the elevator intention analysis result, i.e., the current time of the monitoring video. S12: Obtain relevant video segment t related The corresponding video frame sequence Frame n The keyframes were extracted using the inter-frame difference method, resulting in several elevator keyframes. The specific calculation expression is as follows: For consecutive frames i and Frame i+1 The average difference intensity D between i Represented as: In the formula, Frame i (x,y) represents the pixel value of the i-th frame at pixel coordinates (x,y), W and H represent the width and height of each frame, respectively; i belongs to the interval [1,n], and n is the total number of the video frame sequence, which is calculated based on the monitored refresh rate fps, specifically: fps×(e-t+s); then, a differential threshold is used to evaluate all average differential intensity sets D={D1,D2,…,D n-1 After filtering, several elevator keyframes are obtained, and their differential threshold D is determined. Th for: D Th =max(D)-ΔD var In the formula, max(·) represents the maximum average difference intensity in the current set of all average difference intensities D; ΔD var A custom differential intensity variable; several elevator keyframes (KeyFrames) are represented as follows: KeyFrame={(Frame i ,t j )|i∈[1,n],t j ∈[t-s,e]} In the formula, Frame i This represents the i-th frame in the current video frame sequence; t j Represented as a video frame i The timestamp it belongs to.

3. The method for recognizing the intention to use an elevator by calling another passenger, as described in claim 2, is characterized in that... Step S2 includes the following sub-steps: S21: Using an image classification model, the data at timestamp t is classified. j Key press detection is performed on several keyframes within [ts,t] to obtain key press action frames; S22: Determine whether the current button action frame is a single frame or a double frame: If it is a single frame, obtain the directional intention information based on the direction of the indicator light triggered when the passenger presses the button; If it is a two-frame sequence, the directional intention information is obtained by determining the timestamp sequence of the two frames based on the order of the directions of the indicator lights triggered when the passenger presses the button. S23: Based on the directional intent information, identify the unique feature information of the passenger who performed the key press action in the key press action frame, and use it as the Key Passenger KeyID, which is represented as KeyID = {Feature KeyID Based on the unique characteristic information of key passengers, the human skeleton spatiotemporal relationship model M is adopted. r Correlation detection is performed on several keyframes of elevator rides to obtain the PID set of passengers traveling together, specifically represented as follows: PID={M r (Feature KeyID ,(Frame i ,t j ))|i∈[1,n],t j ∈[t-s,t]}={PID1,PID2,…,PID i } In the formula, M r (Feature KeyID ,(Frame i ,t j This represents the unique characteristics of the key passengers based on the input. (Frame i ,t j ) represents the elevator keyframes within the time interval [ts,t]. By associating key passengers with their accompanying passengers, a monitoring list for the same batch is obtained, which is represented as follows: Group up ={KeyID,PID1,PID2,…,PID i } And / or, Group down ={KeyID,PID1,PID2,…,PID j } In the formula, Group up Group is used to represent the same batch of monitoring lists in the upward direction. down Used to represent the same batch of monitored passengers in the downward direction; i and j represent the total number of passengers traveling in the same direction in the upward and downward directions, respectively; where, PID i This includes the posture and motion characteristics of the i-th passenger.

4. The method for identifying the intention to call an elevator according to claim 3, characterized in that, Step S3 includes the following sub-steps: S31: Based on the monitoring list of the same batch (Group) up and / or Group down The target tracking model M is adopted. Tracking Target tracking is performed on several passenger keyframes to obtain several real-time action (Act) and trajectory (Track) lists. The specific expression for target tracking for key passenger KeyIDs in the same batch of monitoring lists is as follows: M Tracking (KeyID,(Frame i ,t j )),t j ∈[e-t,e] ={Act KeyID ={act1,…,act i },Track KeyID ={position1,…,position i }} In the formula, act i The position represents the action type corresponding to the key passenger in frame i. i This indicates the location of the key passenger in the i-th frame; in the time interval [et, e], e is updated in real time according to the current time of the surveillance video. S32: Based on the real-time action and trajectory list, determine whether the current passenger's intention to use the elevator has changed: if yes, it is assumed that the passenger intends to stop calling the elevator; if no, proceed to step S31. Among them, for the same batch of monitoring list Group up and / or Group down The expression for determining the elevator intention of the j-th passenger is as follows: In the formula, act i Let be the action type of the i-th frame, and act i The action list of the j-th passenger in the same batch of monitoring list; distance(·) is used to calculate the distance between two input data, and left is the distance threshold for determining the position of leaving the elevator; while This is used to determine if, among all subsequent frames k, there exists at least one passenger in frame k whose distance from the elevator is less than or equal to the distance threshold; and This is used to determine the case where, in all subsequent frames k, the distance between the passenger and the elevator is greater than the stated distance threshold; E represents the elevator's coordinate position; position k Let position be the passenger trajectory position in the k-th frame, and position k The trajectory list of the j-th passenger belonging to the same batch of monitoring list; where k∈[i,n], and n is the total number of video frame sequences.

5. A device for recognizing the intention to call an elevator, characterized in that, This includes a monitoring keyframe acquisition unit, a batch target direction intent association unit, and an elevator ride intent recognition unit; The monitoring key frame acquisition unit is used to backtrack the monitoring video of the passenger button and extract key frames to obtain elevator key frames. The same batch target directional intent association unit is used to sequentially perform key detection and directional intent recognition on several elevator key frames to obtain key action frames and directional intent information, and establish an association between key action frames and several elevator key frames based on directional intent information to obtain a same batch monitoring list. The elevator intention recognition unit is used to perform target tracking on several elevator key frames according to the same batch monitoring list, and determine whether the current passenger's elevator intention has changed based on the target tracking result: if not, then continue target tracking; if so, then consider that the passenger intends to stop calling the elevator.

6. The device for identifying the intention to call an elevator according to claim 5, characterized in that, The monitoring key frame acquisition unit includes a monitoring backtracking module and a monitoring key frame extraction module; The monitoring backtracking module is used to backtrack the monitoring video of passengers pressing buttons and obtain relevant video segments; For the current time t, the relevant video segment is specifically represented as follows: t related ∈[t-s,e],s<t In the formula, t related [ts,e] represents the relevant video segment; [ts,e] represents the video segment within s seconds backward from the current time t, where s is the preset interaction time and e represents the time of the elevator intention analysis result, i.e., the current time of the monitoring video. The monitoring keyframe extraction module is used to obtain relevant video segments t related The corresponding video frame sequence Frame n The keyframes were extracted using the inter-frame difference method, resulting in several elevator keyframes. The specific calculation expression is as follows: For consecutive frames i and Frame i+1 The average difference intensity D between i Represented as: In the formula, Frame i (x,y) represents the pixel value of the i-th frame at pixel coordinates (x,y), W and H represent the width and height of each frame, respectively; i belongs to the interval [1,n], and n is the total number of the video frame sequence, which is calculated based on the monitored refresh rate fps, specifically: fps×(e-t+s); then, a differential threshold is used to evaluate all average differential intensity sets D={D1,D2,…,D n-1 After filtering, several elevator keyframes are obtained, and their differential threshold D is determined. Th for: D Th =max(D)-ΔD var In the formula, max(·) represents the maximum average difference intensity in the current set of all average difference intensities D; ΔD var A custom differential intensity variable; several elevator keyframes (KeyFrames) are represented as follows: KeyFrame={(Frame i ,t j )|i∈[1,n],t j ∈[t-s,e]} In the formula, Frame i This represents the i-th frame in the current video frame sequence; t j Represented as a video frame i The timestamp it belongs to.

7. The device for identifying the intention to call an elevator according to claim 6, characterized in that, The target directional intent association unit in the same batch includes a key detection module, a directional intent annotation module, and a directional intent association module; The key detection module is used to employ an image classification model to detect keys at timestamp t. j Key press detection is performed on several keyframes within [ts,t] to obtain key press action frames; The directional intent annotation module is used to determine whether the current button action frame is a single frame or a double frame: if it is a single frame, the directional intent information is obtained according to the direction corresponding to the indicator light triggered when the passenger presses the button. If it is a two-frame sequence, the directional intention information is obtained by determining the timestamp sequence of the two frames based on the order of the directions of the indicator lights triggered when the passenger presses the button. The directional intent association module is used to identify the unique feature information of the passenger performing the key press action in the key press action frame based on the directional intent information, and use it as the Key Passenger KeyID, which is represented as KeyID = {Feature KeyID Based on the unique characteristic information of key passengers, the human skeleton spatiotemporal relationship model M is adopted. r Correlation detection is performed on several keyframes of elevator rides to obtain the PID set of passengers traveling together, specifically represented as follows: PID={M r (Feature KeyID ,(Frame i ,t j ))|i∈[1,n],t j ∈[t-s,t]}={PID1,PID2,…,PID i } In the formula, M r (Feature KeyID ,(Frame i ,t j This represents the unique characteristics of the key passengers based on the input. (Frame i ,t j ) represents the elevator keyframes within the time interval [ts,t]. By associating key passengers with their accompanying passengers, a monitoring list for the same batch is obtained, which is represented as follows: Group up ={KeyID,PID1,PID2,…,PID i } And / or, Group down ={KeyID,PID1,PID2,…,PID j } In the formula, Group up Group is used to represent the same batch of monitoring lists in the upward direction. down Used to represent the same batch of monitored passengers in the downward direction; i and j represent the total number of passengers traveling in the same direction in the upward and downward directions, respectively; where, PID i This includes the posture and motion characteristics of the i-th passenger; The elevator intention recognition unit includes a batch target tracking module and a batch target elevator intention judgment module; The same batch target tracking module is used to track targets based on the same batch monitoring list (Group). up and / or Group down The target tracking model M is adopted. Tracking Target tracking is performed on several passenger keyframes to obtain several real-time action (Act) and trajectory (Track) lists. The specific expression for target tracking for key passenger KeyIDs in the same batch of monitoring lists is as follows: M Tracking (KeyID,(Frame i ,t j )),t j ∈[e-t,e] ={Act KeyID ={act1,…,act i },Track KeyID ={position1,…,position i }} In the formula, act i The position represents the action type corresponding to the key passenger in frame i. i This indicates the location of the key passenger in the i-th frame; in the time interval [et, e], e is updated in real time according to the current time of the surveillance video. The same batch of target elevator intention judgment module is used to determine whether the current passenger has changed their elevator intention based on the real-time action and trajectory list: if yes, it is considered that the passenger intends to stop calling the elevator; if no, step S31 is executed. Among them, for the same batch of monitoring list Group up and / or Group down The expression for determining the elevator intention of the j-th passenger is as follows: In the formula, act i Let be the action type of the i-th frame, and act i The action list of the j-th passenger in the same batch of monitoring list; distance(·) is used to calculate the distance between two input data, and left is the distance threshold for determining the position of leaving the elevator; while This is used to determine if, among all subsequent frames k, there exists at least one passenger in frame k whose distance from the elevator is less than or equal to the distance threshold; and This is used to determine the case where, in all subsequent frames k, the distance between the passenger and the elevator is greater than the stated distance threshold; E represents the elevator's coordinate position; position k Let position be the passenger trajectory position in the k-th frame, and position k The trajectory list of the j-th passenger belonging to the same batch of monitoring list; where k∈[i,n], and n is the total number of video frame sequences.

8. An elevator external call control system, characterized in that, It includes several camera devices, several elevator call buttons, and an identification device for identifying the intention to call an elevator that is electrically and / or communicatively connected to the several camera devices and several elevator call buttons; The aforementioned camera devices are installed in the elevator waiting area to monitor the elevator call buttons on the corresponding floors and send the monitoring video to the elevator call intention recognition device. The aforementioned elevator external call buttons are located in the elevator waiting area, allowing passengers to press the corresponding button in the appropriate direction to call an elevator to the corresponding waiting floor. The external elevator call intention recognition device is used to identify whether the current passenger's elevator call intention has changed by analyzing the surveillance video: if not, the recognition continues; if so, the external elevator call button in the corresponding waiting area is canceled, and the elevator call is cancelled. The device for identifying the intention to call for elevators is the same as the device for identifying the intention to call for elevators as described in any one of claims 5-7.

9. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, when the processor executes the computer program, it implements a method for identifying an intention to call an elevator as described in any one of claims 1 to 4.

10. A computer-readable storage medium storing computer-executable instructions, characterized in that, The computer-executable instructions are used for the method of identifying the intention to call an elevator as described in any one of claims 1 to 4.

Images