Intelligent bus kiosk monitoring method and system with abnormal behavior intelligent monitoring function

Abstract

The invention relates to the technical field of intelligent monitoring, in particular to an intelligent bus kiosk monitoring method and system with an abnormal behavior intelligent monitoring function. The method comprises the following steps: acquiring multi-source monitoring data of a target bus kiosk, and based on the multi-source monitoring data, performing character and article linkage behavior analysis on the multi-source monitoring data to generate an entity interaction feature set; extracting a timestamp sequence and a space coordinate sequence based on the entity interaction feature set, and generating a space-time parameter set based on the timestamp sequence and the space coordinate sequence; based on the entity interaction feature set, according to the space-time parameter set, analyzing whether the current entity interaction feature has an abnormal behavior in the corresponding bus kiosk space-time scene, and if yes, extracting abnormal behavior information; and according to the abnormal behavior information, determining and outputting an abnormal early warning signal. According to the method, the abnormal characteristics of the behaviors under the bus kiosk are analyzed in a multi-dimensional mode, safety early warning is achieved, and the public traffic order is guaranteed.

Description

Technical Field

[0001] This application relates to the field of intelligent monitoring technology, and in particular to a smart bus shelter monitoring method and system with intelligent monitoring of abnormal behavior. Background Technology

[0002] Bus shelter monitoring systems are a technology that uses data on the status of people and facilities in waiting areas to provide early warnings for safety. Its core function is to temporarily store and analyze behavioral information to ensure public transportation order. This technology has become a standard feature of modern urban public transportation management.

[0003] However, when existing technologies are applied to monitor abnormal behavior in complex and dynamic scenarios at bus shelters, they focus on the abnormal characteristics of the behavior itself in a single dimension, failing to consider the influencing factors in the environment in which the behavior occurs. This results in low accuracy and weak real-time performance in judging abnormal behavior, leading to untimely and inaccurate early warnings of abnormal behavior. Summary of the Invention

[0004] This application provides a smart bus shelter monitoring method and system with intelligent monitoring of abnormal behavior to solve the above-mentioned technical problems.

[0005] Firstly, this application provides a smart bus shelter monitoring method with intelligent monitoring of abnormal behavior, the method comprising: Acquire multi-source monitoring data of the target bus shelter, and based on the multi-source monitoring data, analyze the interaction behavior of people and objects to generate a set of entity interaction features; Based on the entity interaction feature set, timestamp sequences and spatial coordinate sequences are extracted, and a spatiotemporal parameter set is generated based on the timestamp sequences and spatial coordinate sequences. Based on the entity interaction feature set, and according to the spatiotemporal parameter set, analyze whether there is abnormal behavior of the current entity interaction feature in the corresponding bus shelter spatiotemporal scenario. If so, extract the abnormal behavior information. Based on the abnormal behavior information, an abnormal warning signal is determined and output.

[0006] This solution analyzes the interaction behavior of people and objects based on multi-source monitoring data to generate a set of entity interaction features, providing high-precision structured input for anomaly analysis. By extracting spatiotemporal parameter sets, it accurately depicts the spatiotemporal coupling relationship of behavioral events. It uses a spatiotemporal scene rule engine to dynamically analyze abnormal behavior information, breaking through the limitations of static rules and significantly reducing the false alarm rate. Finally, through risk level-driven anomaly early warning signals, it achieves graded response and visualized handling, comprehensively improving the accuracy, scene adaptability, and emergency response efficiency of bus shelter safety monitoring.

[0007] Optionally, the multi-source monitoring data includes real-time video stream data, thermal imaging data, standing position distribution data, real-time bus operation data, and bus facility status sensor data; The real-time video stream data, the thermal imaging data, the standing position distribution data, the real-time bus operation data, and the bus facility status sensor data are all aligned across dimensions in time and space through time synchronization protocols and spatial synchronization protocols. The time synchronization protocol is used to ensure that all data points collected from all data sources are timestamped based on a unified high-precision clock source, eliminating inherent clock errors between different devices and ensuring strict consistency of behavioral events on the timeline. The spatial synchronization protocol is used to establish a unified mapping relationship between the data collected from various data sources in the physical spatial coordinate system of the bus shelter, and to convert the target position, attitude, heat map and other information from different sources to the same reference coordinate system.

[0008] This solution utilizes multi-source monitoring data to comprehensively reflect the complex dynamics of bus shelter operation, characterizing passenger behavior, service status, facility health, and environmental comfort, thereby improving the comprehensiveness of panoramic situational awareness. Simultaneously, by using time and spatial synchronization protocols to achieve cross-dimensional spatiotemporal alignment of data, it greatly enhances the accuracy and reliability of subsequent complex event correlation analysis, causal relationship inference, and precise decision support.

[0009] Optionally, the step of analyzing the multi-source monitoring data to perform interaction behavior analysis on the multi-source monitoring data and generating a set of entity interaction features includes: Based on the multi-source monitoring data, a multi-target tracking algorithm is used to screen people and objects in the multi-source monitoring data and extract crowd monitoring features and object monitoring features. Based on the real-time video stream data, the standing position, and the thermal imaging data, the crowd monitoring features are analyzed to identify the human-to-human aggregation patterns, limb movement features, and eye contact features, thereby generating a human-to-human interaction feature set. Based on the real-time video stream data and the real-time bus operation data, analyze the human-object contact behavior characteristics, object displacement characteristics and interactive action characteristics in the object monitoring characteristics, and generate a human-object interaction feature set; Integrate the human-to-human interaction feature set with the human-to-object interaction feature set to generate an entity interaction feature set.

[0010] This solution utilizes multi-source monitoring data and multi-target tracking algorithms to accurately screen and continuously track the characteristics of people and objects. It then deeply analyzes human-to-human interaction patterns (gathering patterns, body movements, eye contact) and human-object interaction states (contact behavior, object displacement, interactive actions), ultimately fusing these to generate a panoramic set of entity interaction features. Based on spatiotemporal correlation, this set comprehensively represents the complex dynamic relationships between all entities within the bus shelter, significantly improving the depth and accuracy of perception and understanding of passenger behavior patterns, service interaction processes, facility usage status, and potential abnormal events. This lays a core behavioral cognitive foundation for intelligent public transportation service management, safety assurance, and efficiency optimization.

[0011] Optionally, the step of integrating the human-to-human interaction feature set and the human-to-object interaction feature set to generate an entity interaction feature set includes: Based on the multi-source monitoring data, individual behavior, group emergence, and environmental coupling mechanisms are analyzed. Through a group behavior dynamics model, contact behavior characteristics, object displacement characteristics, and interactive action characteristics are determined to form a human-to-human interaction feature set. Based on the video stream data and the public transportation facility status sensor data, the changes in the status of the items are analyzed. Combining physical attributes, spatiotemporal trajectories, and human behavior triggering conditions, the item status migration analysis algorithm is used to determine contact behavior characteristics, item displacement characteristics, and interactive action characteristics, thus forming a human-item interaction feature set.

[0012] This solution utilizes multi-source monitoring data and multi-target tracking algorithms to accurately screen and continuously track the characteristics of people and objects. It then deeply analyzes human-to-human interaction patterns (gathering patterns, body movements, eye contact) and human-object interaction states (contact behavior, object displacement, interactive actions), ultimately fusing these to generate a panoramic set of entity interaction features. Based on spatiotemporal correlation, this set comprehensively represents the complex dynamic relationships between all entities within the bus shelter, significantly improving the depth and accuracy of perception and understanding of passenger behavior patterns, service interaction processes, facility usage status, and potential abnormal events. This lays a core behavioral cognitive foundation for intelligent public transportation service management, safety assurance, and efficiency optimization.

[0013] Optionally, the step of extracting a timestamp sequence and a spatial coordinate sequence based on the entity interaction feature set, and generating a spatiotemporal parameter set based on the timestamp sequence and spatial coordinate sequence, includes: Based on the entity interaction feature set, the timestamp sequence of each feature event is extracted, and a histogram of event occurrence frequency distribution is generated through a time window sliding mechanism. Based on the real-time video stream data, the thermal imaging data, and the standing position distribution data, the spatial coordinate sequence of each entity is determined, a three-dimensional density distribution cloud map is constructed, and the coordinates of spatial clustering hotspots are determined by analyzing the three-dimensional density distribution cloud map. Spatiotemporal coding is performed based on the frequency distribution histogram and the spatial hotspot coordinates to determine the time dimension parameters, spatial dimension parameters, and coupling coefficients. The spatiotemporal parameter set is generated based on the time dimension parameters, spatial dimension parameters, and coupling coefficients.

[0014] This scheme extracts event timestamp sequences and spatial coordinate sequences based on entity interaction feature sets. It then utilizes a time window sliding mechanism to generate event frequency histograms and 3D density reconstruction technology to generate spatial hotspot distributions. Furthermore, it fuses temporal patterns and spatial models through spatiotemporal coding, outputting quantified temporal and spatial parameters and coupling coefficients, ultimately integrating them into a high-information-density spatiotemporal parameter set. This parameter set distills the complex entity interactions within bus shelters into computable and predictable spatiotemporal equations, achieving a leap from phenomenon description to pattern modeling. This provides core decision operators for precise passenger flow scheduling, facility optimization, and risk control.

[0015] Optionally, spatiotemporal coding is performed based on the frequency distribution histogram and the coordinates of the spatially clustered hotspots to determine the time dimension parameters, spatial dimension parameters, and coupling coefficients, including: Based on the timestamp sequence of each feature event, the time dimension parameter is obtained by dynamically segmenting the timestamp sequence to determine the periodic feature vector with time context association. Based on the three-dimensional density distribution cloud map, the bus shelter area is divided into several adaptive grids. By analyzing the spatial clustering hotspot coordinates, the distance between each sub-area and the bus facilities is obtained, thus obtaining the spatial dimension parameters. Based on the time dimension parameters and spatial dimension parameters, the coupling relationship between different spatial coordinates under the same time window and the same spatial coordinate under different time windows is analyzed, and the coupling coefficient is generated.

[0016] This solution endows the system with adaptive capabilities for perceptual behavior temporal patterns through dynamic segmentation and periodic feature vector extraction of time-dimensional parameters; spatial parameters achieve intelligent deconstruction of scene spatial structure through breathing adaptive grids and spatial affinity calculations; and coupling coefficients reveal the propagation and evolution of behavioral patterns in the spatiotemporal field through horizontal spatial collaborative analysis and vertical spatiotemporal genetic analysis. The spatiotemporal parameter set composed of these three (time, space, and time + space coupling) transforms the bus shelter from a "static container" into a "living space with memory, breathing, and resonant intelligence."

[0017] Optionally, based on the entity interaction feature set and according to the spatiotemporal parameter set, the analysis is performed to determine whether there are abnormal behaviors in the corresponding bus shelter spatiotemporal scenario for the current entity interaction features. If so, abnormal behavior information is extracted, including: Based on historical spatiotemporal data of normal operation, construct a spatiotemporal scenario benchmark model; Based on the time dimension parameters, spatial dimension parameters and coupling coefficients of the spatiotemporal parameter set, a multi-dimensional comparison is performed with the spatiotemporal scene benchmark model to determine the time anomaly overlap, spatial anomaly overlap and coupling anomaly overlap, and the real-time spatiotemporal anomaly overlap is calculated. Based on the entity interaction features and the real-time spatiotemporal anomaly overlap, abnormal behavior information is extracted.

[0018] This solution constructs a dynamically evolving spatiotemporal scenario benchmark model based on historical normal operation data, defines the boundaries of legal behavior features, and compares real-time entity interaction features with the model in multiple dimensions (time, space, coupling) to determine the degree of anomaly overlap, and calculates the comprehensive real-time spatiotemporal anomaly overlap. When the overlap is below a threshold, abnormal behavior information is deeply extracted and structurally described, embedding anomaly detection deeply into the spatiotemporal context. This significantly improves alarm accuracy, reduces the probability of false alarms and missed alarms, and enhances information operability, providing intelligent and context-aware decision support for public transportation safety early warning, emergency dispatch, and service optimization.

[0019] Optionally, the step of extracting abnormal behavior information based on the entity interaction features and the real-time spatiotemporal anomaly overlap includes: Extract the human-human interaction feature set and the human-object interaction feature set from the entity interaction feature set, and combine them with the time anomaly overlap degree to analyze the time coupling effect of human-human interaction and human-object interaction, so as to obtain the abnormal behavior information of entity interaction features in the time dimension. The entity interaction feature set extracts the human-human interaction feature set and the human-object interaction feature set. Combined with the spatial anomaly overlap, the spatial coupling effect of human-human interaction and human-object interaction is analyzed to obtain abnormal behavior information of entity interaction features in the spatial dimension. Based on the time dimension parameter and the space dimension parameter in the spatiotemporal parameter set, a spatiotemporal dynamic alignment dataset is constructed, and a spatiotemporal coupling analysis framework is generated based on the spatiotemporal dynamic alignment dataset. Based on the entity interaction feature set, and combined with the spatiotemporal coupling analysis framework, the coupling effect of the entity interaction feature set in time and space is analyzed. Combined with the coupling anomaly overlap degree, abnormal behavior information of entity interaction features under coupling relationship is obtained.

[0020] This approach analyzes temporal / spatial coupling effects from two perspectives: temporal linkage of behaviors (such as the event sequence coordination of human-to-human and human-object interactions) and regional spatial transmission (such as the behavior density mapping of hotspot areas). This approach accurately captures abnormal interaction features in a single dimension. Simultaneously, the spatiotemporal coupling analysis framework employs a triple mechanism of gridded behavior density fusion, coupling coefficient quantification, and behavior chain verification to deeply analyze complex cross-dimensional patterns such as "human-to-human collaboration driving human-object anomalies" or "human-object anomalies inducing human-to-human aggregation." This constructs an analytical system from single-dimensional anomalies (temporal / spatial) to coupled anomalies, significantly improving the ability to identify hidden collaborative behaviors.

[0021] Optionally, determining and outputting an abnormal warning signal based on the abnormal behavior information includes: Based on the real-time spatiotemporal anomaly overlap degree and the entity interaction features, when an abnormal correlation signal of the time dimension parameter or spatial dimension parameter of the spatiotemporal parameter set is detected in the entity interaction feature set, and the abnormal overlap degree of the corresponding dimension in the spatiotemporal parameter set exceeds the basic threshold, a first-level anomaly warning signal is triggered and output. Based on the real-time spatiotemporal anomaly overlap degree and the entity interaction features, when the anomaly overlap degree of the coupling coefficient of the time dimension parameter and the spatial dimension parameter exceeds the upgrade threshold, and the coupling anomaly overlap degree in the entity interaction features forms a logical association, a level two anomaly warning signal is triggered and output.

[0022] This solution utilizes real-time spatiotemporal anomaly overlap to perform preliminary quantitative screening of single-dimensional anomalies (Level 1 warning). Building upon this, a hierarchical anomaly identification framework is constructed by analyzing the coupling degree between temporal and spatial anomalies and the logical chains formed in entity interaction behavior (Level 2 warning). This not only improves the response speed to single significant anomaly events (Level 1) but also significantly enhances the insight into complex and hidden cross-spatiotemporal collaborative anomaly patterns (Level 2), thereby comprehensively improving the comprehensiveness and accuracy of the anomaly warning system.

[0023] Secondly, this application provides a smart bus shelter monitoring system with intelligent monitoring of abnormal behavior, the system comprising: The perception and analysis module is used to acquire multi-source monitoring data of the target bus shelter, and based on the multi-source monitoring data, analyze the interaction behavior of people and objects to generate a set of entity interaction features. The extraction modeling module is used to extract timestamp sequences and spatial coordinate sequences based on the entity interaction feature set, and to generate a spatiotemporal parameter set based on the timestamp sequences and spatial coordinate sequences; The detection and recognition module is used to analyze whether there is abnormal behavior of the current entity interaction features in the corresponding bus shelter spatiotemporal scenario based on the entity interaction feature set and the spatiotemporal parameter set; if so, it extracts abnormal behavior information. The decision response module is used to determine and output an abnormal warning signal based on the abnormal behavior information.

[0024] Optionally, in the perception and analysis module, the multi-source monitoring data includes real-time video stream data, thermal imaging data, standing position distribution data, real-time bus operation data, and bus facility status sensor data; The real-time video stream data, the thermal imaging data, the standing position distribution data, the real-time bus operation data, and the bus facility status sensor data are all aligned across dimensions in time and space through time synchronization protocols and spatial synchronization protocols. The time synchronization protocol is used to ensure that all data points collected from all data sources are timestamped based on a unified high-precision clock source, eliminating inherent clock errors between different devices and ensuring strict consistency of behavioral events on the timeline. The spatial synchronization protocol is used to establish a unified mapping relationship between the data collected from various data sources in the physical spatial coordinate system of the bus shelter, and to convert the target position, attitude, heat map and other information from different sources to the same reference coordinate system.

[0025] Optionally, when the perception and analysis module analyzes the interaction behavior of people and objects based on the multi-source monitoring data to generate a set of entity interaction features, it is specifically used for: Based on the multi-source monitoring data, a multi-target tracking algorithm is used to screen people and objects in the multi-source monitoring data and extract crowd monitoring features and object monitoring features. Based on the real-time video stream data, the standing position, and the thermal imaging data, the crowd monitoring features are analyzed to identify the human-to-human aggregation patterns, limb movement features, and eye contact features, thereby generating a human-to-human interaction feature set. Based on the real-time video stream data and the real-time bus operation data, analyze the human-object contact behavior characteristics, object displacement characteristics and interactive action characteristics in the object monitoring characteristics, and generate a human-object interaction feature set; Integrate the human-to-human interaction feature set with the human-to-object interaction feature set to generate an entity interaction feature set.

[0026] Optionally, when the perception and analysis module integrates the human-to-human interaction feature set and the human-to-object interaction feature set to generate the entity interaction feature set, it is specifically used for: Based on the multi-source monitoring data, individual behavior, group emergence, and environmental coupling mechanisms are analyzed. Through a group behavior dynamics model, contact behavior characteristics, object displacement characteristics, and interactive action characteristics are determined to form a human-to-human interaction feature set. Based on the video stream data and the public transportation facility status sensor data, the changes in the status of the items are analyzed. Combining physical attributes, spatiotemporal trajectories, and human behavior triggering conditions, the item status migration analysis algorithm is used to determine contact behavior characteristics, item displacement characteristics, and interactive action characteristics, thus forming a human-item interaction feature set.

[0027] Optionally, the extraction and modeling module integrates the human-to-human interaction feature set and the human-to-object interaction feature set to generate an entity interaction feature set, specifically for: Based on the multi-source monitoring data, individual behavior, group emergence, and environmental coupling mechanisms are analyzed. Through a group behavior dynamics model, contact behavior characteristics, object displacement characteristics, and interactive action characteristics are determined to form a human-to-human interaction feature set. Based on the video stream data and the public transportation facility status sensor data, the changes in the status of the items are analyzed. Combining physical attributes, spatiotemporal trajectories, and human behavior triggering conditions, the item status migration analysis algorithm is used to determine contact behavior characteristics, item displacement characteristics, and interactive action characteristics, thus forming a human-item interaction feature set.

[0028] Optionally, when the extraction and modeling module extracts the timestamp sequence and spatial coordinate sequence based on the entity interaction feature set, and generates a spatiotemporal parameter set based on the timestamp sequence and spatial coordinate sequence, it is specifically used for: Based on the entity interaction feature set, the timestamp sequence of each feature event is extracted, and a histogram of event occurrence frequency distribution is generated through a time window sliding mechanism. Based on the real-time video stream data, the thermal imaging data, and the standing position distribution data, the spatial coordinate sequence of each entity is determined, a three-dimensional density distribution cloud map is constructed, and the coordinates of spatial clustering hotspots are determined by analyzing the three-dimensional density distribution cloud map. Spatiotemporal coding is performed based on the frequency distribution histogram and the spatial hotspot coordinates to determine the time dimension parameters, spatial dimension parameters, and coupling coefficients. The spatiotemporal parameter set is generated based on the time dimension parameters, spatial dimension parameters, and coupling coefficients.

[0029] Optionally, the detection and identification module performs spatiotemporal coding based on the frequency distribution histogram and the spatial hotspot coordinates to determine the time dimension parameters, spatial dimension parameters, and coupling coefficients, specifically for: Based on the timestamp sequence of each feature event, the time dimension parameter is obtained by dynamically segmenting the timestamp sequence to determine the periodic feature vector with time context association. Based on the three-dimensional density distribution cloud map, the bus shelter area is divided into several adaptive grids. By analyzing the spatial clustering hotspot coordinates, the distance between each sub-area and the bus facilities is obtained, thus obtaining the spatial dimension parameters. Based on the time dimension parameters and spatial dimension parameters, the coupling relationship between different spatial coordinates under the same time window and the same spatial coordinate under different time windows is analyzed, and the coupling coefficient is generated.

[0030] Optionally, the detection and recognition module extracts abnormal behavior information based on the entity interaction features and the real-time spatiotemporal anomaly overlap, specifically in the following ways: Extract the human-human interaction feature set and the human-object interaction feature set from the entity interaction feature set, and combine them with the time anomaly overlap degree to analyze the time coupling effect of human-human interaction and human-object interaction, so as to obtain the abnormal behavior information of entity interaction features in the time dimension. The entity interaction feature set extracts the human-human interaction feature set and the human-object interaction feature set. Combined with the spatial anomaly overlap, the spatial coupling effect of human-human interaction and human-object interaction is analyzed to obtain abnormal behavior information of entity interaction features in the spatial dimension. Based on the time dimension parameter and the space dimension parameter in the spatiotemporal parameter set, a spatiotemporal dynamic alignment dataset is constructed, and a spatiotemporal coupling analysis framework is generated based on the spatiotemporal dynamic alignment dataset. Based on the entity interaction feature set, and combined with the spatiotemporal coupling analysis framework, the coupling effect of the entity interaction feature set in time and space is analyzed. Combined with the coupling anomaly overlap degree, abnormal behavior information of entity interaction features under coupling relationship is obtained.

[0031] Optionally, the detection and recognition module, based on the entity interaction feature set and the spatiotemporal parameter set, analyzes whether there is abnormal behavior in the corresponding bus shelter spatiotemporal scenario for the current entity interaction features. If so, it extracts abnormal behavior information, specifically for: Based on historical spatiotemporal data of normal operation, construct a spatiotemporal scenario benchmark model; Based on the time dimension parameters, spatial dimension parameters and coupling coefficients of the spatiotemporal parameter set, a multi-dimensional comparison is performed with the spatiotemporal scene benchmark model to determine the time anomaly overlap, spatial anomaly overlap and coupling anomaly overlap, and the real-time spatiotemporal anomaly overlap is calculated. Based on the entity interaction features and the real-time spatiotemporal anomaly overlap, abnormal behavior information is extracted.

[0032] Optionally, the decision response module determines and outputs an abnormal warning signal based on the abnormal behavior information, specifically for: Based on the real-time spatiotemporal anomaly overlap degree and the entity interaction features, when an abnormal correlation signal of the time dimension parameter or spatial dimension parameter of the spatiotemporal parameter set is detected in the entity interaction feature set, and the abnormal overlap degree of the corresponding dimension in the spatiotemporal parameter set exceeds the basic threshold, a first-level anomaly warning signal is triggered and output. Based on the real-time spatiotemporal anomaly overlap degree and the entity interaction features, when the anomaly overlap degree of the coupling coefficient of the time dimension parameter and the spatial dimension parameter exceeds the upgrade threshold, and the coupling anomaly overlap degree in the entity interaction features forms a logical association, a level two anomaly warning signal is triggered and output. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is a schematic diagram illustrating an application scenario provided in one embodiment of this application; Figure 2 A flowchart illustrating a smart bus shelter monitoring method with intelligent monitoring of abnormal behavior, provided as an embodiment of this application; Figure 3 This application provides a schematic diagram of the structure of a smart bus shelter monitoring system with intelligent monitoring of abnormal behavior, as an embodiment of the present application. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0036] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0037] The embodiments of this application will now be described in further detail with reference to the accompanying drawings.

[0038] When existing technologies are applied to monitor abnormal behavior in complex and dynamic scenarios at bus shelters, they focus on the abnormal characteristics of the behavior itself in a single dimension, failing to consider the influencing factors in the environment in which the behavior occurs. This results in low accuracy and weak real-time performance in judging abnormal behavior, leading to untimely and inaccurate early warnings of abnormal behavior.

[0039] Based on this, this application provides a smart bus shelter monitoring method and system with intelligent monitoring of abnormal behavior. It generates an entity interaction feature set by analyzing the interaction behavior of people and objects based on multi-source monitoring data, providing high-precision structured input for anomaly analysis; by extracting spatiotemporal parameter sets, it accurately depicts the spatiotemporal coupling relationship of behavioral events; it uses a spatiotemporal scene rule engine to dynamically analyze abnormal behavior information, breaking through the limitations of static rules and significantly reducing the false alarm rate; finally, through risk level-driven anomaly early warning signals, it achieves graded response and visualized handling, comprehensively improving the accuracy, scene adaptability, and emergency response efficiency of bus shelter safety monitoring.

[0040] Figure 1 This is a schematic diagram illustrating an application scenario provided by this application. During bus operations, the method provided in this application enables tiered response and visualized handling of abnormal behaviors within bus shelters, comprehensively improving the accuracy, scenario adaptability, and emergency response efficiency of bus shelter safety monitoring.

[0041] Specifically, the method of this application is applied to any server that communicates with heterogeneous sensors to acquire multi-source monitoring data provided by the sensors. Based on the multi-source monitoring data, the method analyzes the interaction behavior of people and objects to generate a set of entity interaction features, providing high-precision structured input for anomaly analysis. By extracting spatiotemporal parameter sets, the method accurately depicts the spatiotemporal coupling relationship of behavioral events. By using a spatiotemporal scene rule engine to dynamically analyze abnormal behavior information, the method overcomes the limitations of static rules and significantly reduces the false alarm rate. Finally, the method provides risk level-driven anomaly warning signals to safety maintenance personnel, enabling graded response and visualized handling, and comprehensively improving the accuracy, scene adaptability, and emergency response efficiency of bus shelter safety monitoring.

[0042] For specific implementation details, please refer to the following examples.

[0043] Figure 2 This is a flowchart illustrating a smart bus shelter monitoring method with intelligent abnormal behavior detection, provided as an embodiment of this application. The method of this embodiment can be applied to bus shelters in the above-mentioned scenarios. Figure 2 As shown, the method includes: S201. Obtain multi-source monitoring data of the target bus shelter. Based on the multi-source monitoring data, analyze the interaction behavior of people and objects to generate a set of entity interaction features.

[0044] Multi-source monitoring data can be a collection of data collected by heterogeneous sensors installed inside the target bus shelter, including real-time video stream data, thermal imaging data, standing position distribution data, real-time bus operation data, and bus facility status sensor data.

[0045] Interactive behaviors between people and objects can be related to the movement trajectories of people and changes in the state of objects inside a bus shelter.

[0046] The entity interaction feature set can be structured data generated after parsing, including entity interaction feature interaction labels, interaction strength coefficients, and confidence scores.

[0047] Specifically, current bus shelter monitoring systems only record time or location information. This static, single-dimensional analysis rule cannot adapt to the needs of dynamic scenarios, lacks cross-modal information collaboration capabilities, and cannot capture collaborative anomalies. This step constructs entity-level interaction features through multi-source data fusion and cross-modal behavior analysis. By synchronously receiving multi-source monitoring data and combining it with computer vision algorithms to identify the spatiotemporal correlation of entity interaction features, the generated structured data provides high-precision input for subsequent anomaly analysis.

[0048] S202. Based on the entity interaction feature set, extract the timestamp sequence and spatial coordinate sequence, and generate a spatiotemporal parameter set based on the timestamp sequence and spatial coordinate sequence.

[0049] The timestamp sequence and spatial coordinate sequence can be a data sequence consisting of the occurrence time of each entity interaction event and the position of the interaction event in the three-dimensional space of the bus shelter.

[0050] The spatiotemporal parameter set can be a matrix composed of timestamp sequences and spatial coordinate sequences, or a parameter set composed of the time sequence, spatial location, and regional semantic labels of behavioral events.

[0051] Specifically, existing bus shelter abnormal behavior monitoring technologies only focus on the characteristics of abnormal behaviors generated by people within the bus shelter itself, without analyzing their correlation with corresponding time and space, resulting in overly one-sided judgments of abnormal behavior. This step parses event timestamps from the entity interaction feature set, sorts them in order of occurrence to create a behavioral event tag sequence, and converts the two-dimensional pixel coordinates of the events into three-dimensional spatial coordinates within the bus shelter based on camera calibration parameters. This yields the location coordinates of the behavioral events in the three-dimensional space beneath the bus shelter, and integrates the timestamp sequence and spatial location coordinates into a structured dataset, outputting a spatiotemporal parameter set that demonstrates the coupling relationship between time and space. This provides reliable data in both the temporal and spatial dimensions for subsequent definition and judgment of abnormal behaviors under the bus shelter.

[0052] S203. Based on the entity interaction feature set, and according to the spatiotemporal parameter set, analyze whether there is any abnormal behavior of the current entity interaction feature in the corresponding bus shelter spatiotemporal scenario. If so, extract the abnormal behavior information.

[0053] Abnormal behavior information can be either explicit or implicit, which violate the normal behavior pattern in the bus shelter scenario.

[0054] Specifically, existing technologies focus on the abnormal characteristics of entity interaction behavior in a single dimension, causing them to overlook the anomalies that seemingly normal behaviors may cause under specific time or space conditions. For example, in the context of a bus shelter, the gathering of people waiting inside the shelter is easily misjudged as abnormal gathering behavior if its time and space are ignored. However, when this behavior occurs within a safe area or time period, it should not be defined as abnormal behavior under the bus shelter. Existing technologies, without considering the time and space of the behavior, are prone to such misjudgments. This step uses a spatiotemporal scene rule engine to achieve dynamic anomaly determination. It uses a spatiotemporal parameter set to locate the specific area where the event occurs, inputs the entity interaction feature set and the spatiotemporal parameter set into the rule engine, matches the area rules corresponding to the event's spatiotemporal coordinates, calculates the dynamic behavior anomaly overlap degree, and if the anomaly overlap degree exceeds the limit, extracts the anomaly type, risk level, and evidence data index.

[0055] S204. Based on the abnormal behavior information, determine and output the abnormal warning signal.

[0056] An abnormal warning signal can be a set of instructions including the warning level, target location, and recommended handling measures.

[0057] Specifically, traditional early warning systems only trigger audible alarms, lacking a tiered response mechanism, leading to over-response for low-risk events and insufficient information for handling high-risk events. This step improves response efficiency and accuracy by generating differentiated early warning signals based on risk levels, implementing multi-level early warning strategies, and providing visual output, thus tiering abnormal early warning signals to safety maintenance personnel.

[0058] This solution analyzes the interaction behavior of people and objects based on multi-source monitoring data to generate a set of entity interaction features, providing high-precision structured input for anomaly analysis. By extracting spatiotemporal parameter sets, it accurately depicts the spatiotemporal coupling relationship of behavioral events. It uses a spatiotemporal scene rule engine to dynamically analyze abnormal behavior information, breaking through the limitations of static rules and significantly reducing the false alarm rate. Finally, through risk level-driven anomaly early warning signals, it achieves graded response and visualized handling, comprehensively improving the accuracy, scene adaptability, and emergency response efficiency of bus shelter safety monitoring.

[0059] In some embodiments, real-time video stream data, thermal imaging data, standing position distribution data, real-time bus operation data, and bus facility status sensor data are all aligned across dimensions through time synchronization protocols and spatial synchronization protocols. The time synchronization protocol ensures that all data points collected from all data sources are timestamped based on a unified high-precision clock source, eliminating inherent clock errors between different devices and ensuring strict consistency of behavioral events on the timeline. The spatial synchronization protocol is used to establish a unified mapping relationship between the data collected from each data source in the physical spatial coordinate system of the bus shelter, transforming target positions, attitudes, heat maps, and other information from different sources to the same reference coordinate system.

[0060] Multi-source monitoring data includes real-time video stream data, thermal imaging data, standing position distribution data, real-time bus operation data, and bus facility status sensor data.

[0061] Real-time video stream data can capture continuous visual information of dynamic scenes inside the bus shelter, intuitively presenting passengers' macro-behavioral patterns (such as entry and exit flow, gathering status, queuing order), facial emotional cues (after compliant desensitization processing), and characteristics of carried items (such as suitcases, strollers).

[0062] Thermal imaging data can be spatial distribution information of human infrared radiation, which can be used for non-contact detection of the presence, precise location, posture characteristics (such as distinguishing between standing, leaning, and squatting), and changes in the body surface temperature field (to assist in the preliminary screening of abnormal body temperature).

[0063] The data on standing location distribution can include the population density, distribution hotspots, and dynamic changes in various areas within the bus shelter.

[0064] Real-time bus operation data can include information on vehicles arriving at the station (route, estimated arrival time), current vehicle location, and passenger occupancy rate.

[0065] Data from public transport facility status sensors can include the occupancy status of waiting seats, the operational status of electronic bus stop signs, energy consumption of lighting / air conditioning, and the status of safety facilities (such as emergency buttons).

[0066] Time synchronization protocols can be the underlying mechanism for ensuring high consistency of timestamps among multi-source heterogeneous data acquisition devices (such as based on NTP or PTP protocols), eliminating inherent clock drift between devices and aligning cross-dimensional events on the timeline.

[0067] Spatial synchronization protocols can map video coordinates, thermal image coordinates, and sensor positions to the physical coordinate system of the bus shelter.

[0068] Specifically, differences in sampling frequencies among different devices can lead to fragmented behavioral events and independent spatial reference systems for each device. This results in inconsistent descriptions of the physical location of the same target, and abnormal behaviors often manifest as multi-feature coupling in the spatiotemporal dimensions. If spatiotemporal alignment is not achieved, a logical chain for human-object-environment interaction cannot be established, and additional calculations are required to compensate for the deviation, significantly increasing processing latency. The time synchronization protocol used for time synchronization executes as follows: a GPS / PTP time synchronization server is deployed as the master clock source, outputting nanosecond-level time signals; each data acquisition device connects to the time synchronization network via wired / Wi-Fi 6 and periodically calibrates its local clock; during data acquisition, the device calls the time synchronization API to inject a unified timestamp (format: UTC + milliseconds + device ID); data is transmitted to the edge computing node, where the time alignment module verifies timestamp consistency and discards data packets exceeding the time difference threshold (>10ms). The spatial synchronization protocol for spatial synchronization is executed in the following steps: Calibration phase: A checkerboard calibration board is placed inside the bus shelter. Multiple cameras are used for joint shooting to calculate the internal and external parameters of the cameras. The LiDAR scans the structure inside the shelter to generate a 3D point cloud map and marks the coordinates of key facilities (such as bus stop signs and seats). Mapping table construction: A transformation matrix is ​​established from the device coordinate system to the bus shelter's reference coordinate system (such as camera image plane coordinates → 3D world coordinates). A temperature-space mapping function is generated between the thermal imager pixel coordinates and the physical location. Real-time conversion: When data is input, a preset transformation matrix / function is called to convert the original coordinates into (X,Y,Z) values ​​in the reference coordinate system. For standing position data, coordinate interpolation is used to fill the sparse sampling gaps of the sensors. The spatiotemporal alignment mechanism implemented in this embodiment can effectively avoid misassociations or missed associations caused by time deviations, accurately match the high-temperature points in thermal imaging with the human body positions in the video, solve the problem of "ambiguous target positioning", and the spatiotemporally aligned data can be directly input into the spatiotemporal parameter set generation module to ensure the spatial topological authenticity of the frequency distribution histogram and the three-dimensional density cloud map, reduce false alarms caused by environmental noise, support multimodal feature fusion, provide reliable input for coupled abnormal behavior recognition, and optimize the real-time performance of the system.

[0069] This solution utilizes multi-source monitoring data to comprehensively reflect the complex dynamics of bus shelter operation, characterizing passenger behavior, service status, facility health, and environmental comfort, thereby improving the comprehensiveness of panoramic situational awareness. Simultaneously, by using time and spatial synchronization protocols to achieve cross-dimensional spatiotemporal alignment of data, it greatly enhances the accuracy and reliability of subsequent complex event correlation analysis, causal relationship inference, and precise decision support.

[0070] In some embodiments, based on multi-source monitoring data, a multi-target tracking algorithm is used to screen people and objects in the multi-source monitoring data to extract crowd monitoring features and object monitoring features; based on real-time video stream data, standing position, and thermal imaging data, the crowd monitoring features are analyzed to identify human-to-human aggregation patterns, limb movement features, and eye-to-eye interaction features, generating a human-to-human interaction feature set; based on real-time video stream data and real-time public transportation operation data, the object monitoring features are analyzed to identify human-to-object contact behavior features, object displacement features, and interactive action features, generating a human-to-object interaction feature set; the human-to-human interaction feature set and the human-to-object interaction feature set are integrated to generate an entity interaction feature set.

[0071] Multi-source monitoring data is used to analyze the interactive behaviors of people and objects, generating a set of entity interaction features.

[0072] Multi-target tracking algorithms are algorithms that use computer vision technology to continuously locate, identify, and predict the trajectories of multiple moving targets (such as pedestrians and objects) in a video stream.

[0073] Crowd monitoring characteristics can be quantitative indicators that reflect the dynamic behavior of crowds, extracted from monitoring data, including gathering density, direction of movement, and range of motion of limbs.

[0074] The monitoring characteristics of items can be parameters of the interaction between public facilities (such as seats, trash cans, and electronic screens) and people in bus shelters, including contact frequency, displacement distance, and usage duration.

[0075] The characteristics of human-human aggregation patterns can be a set of characteristics of social behavior among individuals.

[0076] Human-to-human physical movement characteristics can be the changes in physical movements resulting from individual posture changes between people.

[0077] Human-to-human eye contact characteristics can be inferred from the direction of a passenger's head and the focus of their gaze (based on video).

[0078] Human-to-human interaction feature set can be formed by integrating features such as gathering patterns, body movements, and eye contact.

[0079] Human-object contact behavior characteristics can refer to the interaction between a person and a physical facility / object.

[0080] The displacement characteristics of objects between people and objects can be used to monitor unexpected movements of objects in space.

[0081] The interaction characteristics of people and objects can be an integration of the contact behavior characteristics of people and objects, the displacement characteristics of objects, and the interaction characteristics.

[0082] The entity interaction feature set can be a panoramic behavioral representation that integrates the human-human interaction feature set and the human-object interaction feature set, and a complex, interconnected dynamic interaction network among all entities (people, objects, facilities) in the bus shelter.

[0083] Specifically, in existing bus shelter monitoring scenarios, traditional solutions suffer from data fragmentation due to the independent analysis of video streams, thermal imaging, and sensor data, failing to connect the interactive logic of "people-objects-environment." Relying solely on video target detection ignores the tension implied by thermal distribution (such as abnormal body temperature clusters) or the behavioral analysis of abnormal facility states reflected by sensor data (such as continuous shaking of electronic screens). Furthermore, the highly dynamic nature of pedestrian flow at bus shelters (such as dense crowds during morning and evening rush hours) means that static threshold models are ill-suited to distinguish between normal crowding and aggressive pushing. This embodiment focuses on multi-target cross-modal tracking, constructing a high-precision entity interaction model through the extraction and integration of human-to-human and human-to-object dual-dimensional features, providing a data foundation for intelligent monitoring of abnormal behavior in bus shelter scenarios. By inputting multi-source monitoring data (real-time video stream, thermal imaging, standing position distribution, and public transportation facility sensor data) aligned with time and space, a multi-target tracking algorithm is used to detect and bind people and objects in each frame of data, generating continuous trajectories. Crowd features (such as skeletal motion vectors and thermal aggregation areas) and object features (such as sensor trigger sequences and displacement trajectories) are separated from the trajectory data. Specifically, crowd monitoring features are analyzed as follows: crowd distribution shape is calculated using a convex hull algorithm to analyze crowd aggregation patterns; abnormal limb amplitude values ​​(such as sudden arm swings) are calculated based on the velocity vectors of posture key points to analyze limb movements; the duration of sustained eye contact is statistically analyzed using the spatial intersection of head orientation vectors and targets; and a structured set of human-to-human interaction features is output. The analysis of object monitoring features includes: marking contact intentions (e.g., brief touch / continuous pressing) based on the overlap rate of hand-object bounding boxes and sensor pressure abrupt change points in the video to analyze contact behavior; calculating the probability of displacement caused by non-human factors (e.g., wind) by combining sensor displacement data and video optical flow methods to analyze object displacement; identifying interaction types in the video (e.g., normal use / kicking / graffiti) through a classification model to analyze interactive actions; and outputting a structured human-object interaction feature set. The integration of entity interaction feature sets involves aligning the human-human interaction feature set with the human-object interaction feature set according to a time window to generate a unified entity interaction feature set.

[0084] This solution utilizes multi-source monitoring data and multi-target tracking algorithms to accurately screen and continuously track the characteristics of people and objects. It then deeply analyzes human-to-human interaction patterns (gathering patterns, body movements, eye contact) and human-object interaction states (contact behavior, object displacement, interactive actions), ultimately fusing these to generate a panoramic set of entity interaction features. Based on spatiotemporal correlation, this set comprehensively represents the complex dynamic relationships between all entities within the bus shelter, significantly improving the depth and accuracy of perception and understanding of passenger behavior patterns, service interaction processes, facility usage status, and potential abnormal events. This lays a core behavioral cognitive foundation for intelligent public transportation service management, safety assurance, and efficiency optimization.

[0085] In some embodiments, based on multi-source monitoring data, individual behavior, group emergence, and environmental coupling mechanisms are analyzed. Through a group behavior dynamics model, contact behavior characteristics, object displacement characteristics, and interactive action characteristics are determined to form a human-to-human interaction feature set. Based on video stream data and public transportation facility status sensor data, changes in object status are analyzed. Combining physical attributes, spatiotemporal trajectories, and human behavior triggering conditions, an object status migration analysis algorithm is used to determine contact behavior characteristics, object displacement characteristics, and interactive action characteristics to form a human-object interaction feature set.

[0086] Integrate human-to-human interaction feature sets and human-to-object interaction feature sets to generate entity interaction feature sets.

[0087] Individual behavior can be the action primitives of a single individual in a specific time and space, such as choosing a walking path, shifting gaze, expressing gestures, or adjusting body posture.

[0088] Group emergence can be a macroscopic orderly pattern that spontaneously forms from nonlinear interactions between individuals. For example, individuals moving towards a bus stop can trigger a vortex of crowd gathering; a few people queuing can trigger a collective imitation of order; and local pushing behavior can spread into ripples of overall unrest.

[0089] Environmental coupling mechanisms can be the dynamic mutual shaping relationship between group behavior and the physical / digital environment. For example: updating information on electronic bus stop signs (environmental stimulus) → triggering crowd gathering for inquiries (group response) → causing channel blockage (change in environmental state) → triggering new path selection strategies (behavioral feedback).

[0090] Group behavior dynamics models can be computational frameworks based on physical and sociological principles, treating individuals as particles driven by "social forces" (such as following, avoiding, attracting, and repelling). They can predict macro-emergence (group patterns) by simulating micro-interactions (individual behavior) and quantify the modulating effects of the environment (such as facility layout and information dissemination).

[0091] The status change of an item can be a dynamic transformation of the key attributes of a facility or item. For example: a seat goes from idle to occupied to idle; a bus stop screen goes from normal display to lagging to black screen; a suitcase goes from being carried to being stationary to being moved without owner; an emergency button goes from never being triggered to being pressed to being in alarm state.

[0092] Physical properties can be inherent material constraints of an item, such as: the load-bearing deformation threshold of a seat (to determine if it is overloaded), the touch sensitivity of a bus stop screen (to distinguish between valid and invalid operations), the size / weight characteristics of a suitcase (affecting the mode of transportation), and the opening and closing status of the facility's outer shell (to detect illegal opening).

[0093] A spatiotemporal trajectory can be a continuous evolutionary path of an object's position and state.

[0094] Human behavior triggers can be actions that cause an item's state to transition, such as: a passenger continuously pressing a threshold → triggering an emergency button alarm (state transition); a passenger leaving their seat for an extended period without moving their luggage → triggering a lost item warning (state transition); a specific finger swipe sequence → switching the bus stop sign to the route search page (state transition).

[0095] Item state transition analysis algorithms can be finite state machines or temporal pattern mining techniques, which track the physical properties and spatiotemporal trajectories of items.

[0096] Specifically, in the bus shelter scenario, abnormal behavior may manifest as isolated individual vandalism or abnormal group coordination. Analyzing only a single feature set can lead to several issues: the person-to-person feature set may miss incidents of object destruction, and the person-to-object feature set may fail to identify group conflicts due to its biased analysis. This embodiment employs a dual-channel analysis: the generation process of the person-to-person interaction feature set involves first inputting multi-source monitoring data (video stream, thermal imaging, standing position), then using a group behavior dynamics model to calculate contact behavior characteristics (duration of minimum distance between individuals), limb movement characteristics (frequency and direction of waving / pushing, etc.), and eye-to-eye convergence characteristics (focus overlap when multiple people are looking at the same target). When "group centripetal acceleration > threshold and eye-to-eye convergence suddenly increases," the "aggregation conflict" feature is output to generate the person-to-person interaction feature set. Interaction feature set; the generation process of the human-object interaction feature set is as follows: first, input video stream and facility sensor data, then generate the human-object interaction feature set by running the item state migration (physical attribute constraint (marking movable items (luggage) and fixed facilities (bus stop signs)), spatiotemporal trajectory verification (if the luggage movement path does not overlap with any passenger trajectory, trigger the "unclaimed item" mark), and behavior trigger condition matching (when "the hand touches the fire hydrant for a certain number of seconds and there is no water-taking action", output the "suspicious operation" feature)) analysis algorithm; and align the human-human feature set and the human-object feature set according to the time window. If both feature sets have abnormal values ​​in the same time period, generate coupled abnormal events and store them in the entity interaction feature set.

[0097] This solution utilizes multi-source monitoring data and multi-target tracking algorithms to accurately screen and continuously track the characteristics of people and objects. It then deeply analyzes human-to-human interaction patterns (gathering patterns, body movements, eye contact) and human-object interaction states (contact behavior, object displacement, interactive actions), ultimately fusing these to generate a panoramic set of entity interaction features. Based on spatiotemporal correlation, this set comprehensively represents the complex dynamic relationships between all entities within the bus shelter, significantly improving the depth and accuracy of perception and understanding of passenger behavior patterns, service interaction processes, facility usage status, and potential abnormal events. This lays a core behavioral cognitive foundation for intelligent public transportation service management, safety assurance, and efficiency optimization.

[0098] In some embodiments, based on the entity interaction feature set, the timestamp sequence of each feature event is extracted, and a frequency distribution histogram of events is generated through a time window sliding mechanism; based on real-time video stream data, thermal imaging data, and standing position distribution data, the spatial coordinate sequence of each entity is determined, a three-dimensional density distribution cloud map is constructed, the three-dimensional density distribution cloud map is analyzed, and the coordinates of spatial clustering hotspots are determined; spatiotemporal encoding is performed based on the frequency distribution histogram and the coordinates of spatial clustering hotspots to determine the time dimension parameters, spatial dimension parameters, and coupling coefficients; based on the time dimension parameters, spatial dimension parameters, and coupling coefficients, a spatiotemporal parameter set is generated.

[0099] A timestamp sequence can be an array of precise moments in time from the set of entity interaction features.

[0100] The time window sliding mechanism can be a statistical box that moves dynamically along the time axis, like a "spatiotemporal probe" that continuously scans the event stream, capturing local fluctuations and global trends in event frequency during the sliding process, and avoiding feature loss caused by fixed segments.

[0101] An event frequency distribution histogram can be a statistical visualization result output by a time window mechanism. Its horizontal axis represents the time window (e.g., morning rush hour 7:00-9:00), and its vertical axis represents the cumulative frequency of similar events within the window (e.g., the number of "conflict encounters" and "bus stop query" times), forming an event tide map reflecting "when events are most frequent."

[0102] The spatial coordinate sequence of each entity can be a dataset of the three-dimensional positions (X, Y, height) of all tracked targets (people, objects) at continuous time points, constituting the "dynamic point cloud flow" of the physical space of the bus shelter.

[0103] A three-dimensional density distribution cloud map can be created by applying spatial kernel density estimation techniques to transform a discrete sequence of spatial coordinates into a continuous probability density surface, where highlighted areas represent "thermal peaks" in the probability of entity occurrence. This cloud map serves as a holographic sand table for understanding spatial usage patterns.

[0104] Spatial hotspot coordinates can be local density maxima (e.g., (X1, Y1, Z1), (X2, Y2, Z2)) identified in a 3D density cloud map, representing areas of high concentration of people or goods (e.g., in front of bus stops, seating areas, boarding gates). Each hotspot is accompanied by a concentration intensity value (density value) and a spatial radius.

[0105] Spatiotemporal coding can be a mathematical transformation process that jointly models time statistics (histograms) and spatial hotspot information. Its core is to establish a ternary mapping relationship of "time-space-event".

[0106] Time dimension parameters can be indicators for quantifying the temporal patterns of events.

[0107] Spatial dimension parameters can be indicators that describe the characteristics of hotspot spaces.

[0108] The coupling coefficient can be a key parameter for measuring the strength of the dynamic correlation between temporal patterns and spatial hotspots.

[0109] The spatiotemporal parameter set can be a structured output that integrates time dimension parameters, spatial dimension parameters, and coupling coefficients.

[0110] Specifically, existing bus shelter monitoring only analyzes time or space dimensions independently, failing to capture spatiotemporally coupled anomalies. This results in temporal isolation or spatial staticity, hindering the accurate identification of high-precision abnormal behavior and the synergistic effect of multi-source data. This embodiment addresses this by fusing timestamp sequences and spatial coordinate sequences through spatiotemporal coding to generate a spatiotemporal parameter set. A time window sliding mechanism generates an event frequency distribution histogram, which, combined with spatial hotspot coordinates, identifies whether high-frequency events occur in atypical areas, thus solving the problem of temporal isolation in simple event frequency statistics, which cannot distinguish between normal queuing and abnormal delays. Furthermore, by analyzing the density distribution cloud map to dynamically reflect spatial clustering trends, it addresses the problem of spatial staticity by preventing fixed-area monitoring from ignoring behavioral changes over time. Finally, it extracts timestamps from the entity interaction feature set (e.g., the moment of human-object contact), arranges them in ascending chronological order to generate a timestamp sequence, and then uses video... Flow and thermal imaging data are used to locate entity positions and map them to a unified coordinate system, forming a coordinate sequence and generating a spatial coordinate sequence for spatiotemporal feature extraction. A sliding time window is set to count the number of events within each window, and a bar chart is generated based on the statistical results, with the horizontal axis representing the time interval and the vertical axis representing the number of events to analyze temporal patterns. By aggregating all spatial coordinate points, a 3D heat map is generated through density calculation, with red areas indicating high concentration. The darkest areas in the heat map are automatically identified, and their center points are recorded as hotspot coordinates to model spatial patterns. Finally, time dimension parameters, spatial dimension parameters, and time-space coupling coefficients are generated to produce a spatiotemporal parameter set.

[0111] This scheme extracts event timestamp sequences and spatial coordinate sequences based on entity interaction feature sets. It then utilizes a time window sliding mechanism to generate event frequency histograms and 3D density reconstruction technology to generate spatial hotspot distributions. Furthermore, it fuses temporal patterns and spatial models through spatiotemporal coding, outputting quantified temporal and spatial parameters and coupling coefficients, ultimately integrating them into a high-information-density spatiotemporal parameter set. This parameter set distills the complex entity interactions within bus shelters into computable and predictable spatiotemporal equations, achieving a leap from phenomenon description to pattern modeling. This provides core decision operators for precise passenger flow scheduling, facility optimization, and risk control.

[0112] In some embodiments, based on the timestamp sequence of each feature event, the time dimension parameter is obtained by dynamically segmenting the timestamp sequence to determine the periodic feature vector with temporal context association; based on the three-dimensional density distribution cloud map, the bus shelter area is divided into several adaptive grids, and the distance between each sub-area and the bus facility is analyzed by spatially aggregating hotspot coordinates to obtain the spatial dimension parameter; based on the time dimension parameter and the spatial dimension parameter, the coupling relationship between different spatial coordinates under the same time window and the same spatial coordinate under different time windows is analyzed to generate the coupling coefficient.

[0113] Dynamic segmentation can be a time-series segmentation algorithm that adaptively adjusts the window length based on event density.

[0114] Periodic feature vectors with temporal context can be mathematical expressions that characterize the periodic patterns of events.

[0115] An adaptive grid can be a spatial unit that dynamically adjusts according to population density.

[0116] The distance between each sub-area and public transportation facilities can be a key indicator for quantifying spatial relationships.

[0117] The coupling relationship between different spatial coordinates within the same time window and the same spatial coordinates within different time windows can be used for spatial synergy effects and analysis of tracking the life cycle of spatial hotspots.

[0118] Specifically, existing monitoring and analysis methods often process time-series data and spatial location data independently, perhaps only counting the number of people per unit time or only marking hotspots of crowd gathering. This fragmented analysis loses the inherent intrinsic connections and coordinated changes in behavior across the spatiotemporal dimensions. However, abnormal behavior within bus shelters often manifests as a disconnect or misalignment of spatiotemporal patterns; for example, crowds gathering at unusual locations during off-peak hours or objects suddenly moving after a long period of stillness. Identifying these anomalies requires analyzing the coupling relationship between temporal and spatial patterns. This embodiment utilizes a method of acquiring temporal and spatial dimension parameters and generating coupling coefficients to optimize computational resources for bus shelters, enhancing their ability to represent spatiotemporal characteristics, analyzing spatiotemporal fusion, and providing refined and interpretable early warning decisions. The process involves obtaining time dimension parameters: extracting the timestamp sequence corresponding to all feature events from the entity interaction feature set, and analyzing the distribution density and interval pattern of the timestamps through the spatiotemporal covariance function. In periods where events occur frequently and change drastically (such as peak traffic), shorter and more refined time segments are automatically divided; in periods where events are sparse and changes are gradual (such as late at night), they are merged into longer time segments. The goal is to ensure that the events within each time segment have a strong temporal contextual relationship. The timestamp sequence is dynamically segmented, and within each dynamically segmented time segment, the periodicity, frequency distribution trend, and explosiveness of events are further analyzed. The peak position, mean, and variance of the frequency of events within the segment are calculated, or periodic components are extracted using a model. The quantitative features that characterize the time distribution pattern and periodicity of events within each time segment are then combined to form a periodic feature vector, which is output as the required time dimension parameter. The spatial dimension parameter is obtained by analyzing each sub-region of the segmented grid (or focusing on hotspot areas). The Euclidean distance or path distance from the coordinates of the spatial clustered hotspots (or their representative points) within the sub-region to various key public transportation facilities (such as seats, bus stop signs, information screens, trash cans, charging ports, billboards, emergency buttons, etc.) in the bus shelter is calculated. This distance information is analyzed, for example, by statistically analyzing the distance distribution from hotspots to the nearest facilities and the frequency of hotspots appearing within a specific distance range of a specific facility.Then, the density characteristics of each sub-region and its distance characteristics from key facilities are combined to form a set of quantitative indicators describing the spatial structure characteristics and hotspot-facility relationships of the entire bus shelter, i.e., spatial dimension parameters. The coupling coefficient is generated by ensuring that the time dimension parameters (periodic feature vectors) and spatial dimension parameters (a set containing density and distance characteristics) are aligned on the analysis window. By calculating the spatiotemporal covariance function, the analysis examines how behavioral patterns (characterized by their spatial dimension parameters) occurring in different spatial grids (sub-regions) within the same dynamic time segment (such as a minute during the morning rush hour) are interconnected, influence each other, or change collaboratively. The coupling relationship data of different spatial coordinates under the window is analyzed to see how behavioral patterns (characterized by their time dimension parameters) evolve and correlate within different dynamic time segments (such as several consecutive five-minute segments) on the same spatial grid (sub-region). For example, whether there are continuous weak abnormal heat signals in a corner area during several consecutive time periods at night (which may indicate that someone is lingering in the bus shelter), or whether the frequency of human-object interaction near a facility fluctuates abnormally in a specific time series (which may indicate that someone is abnormally operating the equipment in the bus shelter). The coupling relationship analysis data of the same spatial coordinate under different time windows is obtained; and the quantified value of the spatiotemporal coupling strength obtained from the analysis is used as the coupling coefficient.

[0119] This solution endows the system with adaptive capabilities for perceptual behavior temporal patterns through dynamic segmentation and periodic feature vector extraction of time-dimensional parameters; spatial parameters achieve intelligent deconstruction of scene spatial structure through breathing adaptive grids and spatial affinity calculations; and coupling coefficients reveal the propagation and evolution of behavioral patterns in the spatiotemporal field through horizontal spatial collaborative analysis and vertical spatiotemporal genetic analysis. The spatiotemporal parameter set composed of these three (time, space, and time + space coupling) transforms the bus shelter from a "static container" into a "living space with memory, breathing, and resonant intelligence."

[0120] In some embodiments, a spatiotemporal scenario benchmark model is constructed based on historical normal operation spatiotemporal data; a multi-dimensional comparison is performed between the time dimension parameters, spatial dimension parameters, and coupling coefficients of the spatiotemporal parameter set and the spatiotemporal scenario benchmark model to determine the time anomaly overlap, spatial anomaly overlap, and coupling anomaly overlap, and the real-time spatiotemporal anomaly overlap is calculated; based on entity interaction features and combined with the real-time spatiotemporal anomaly overlap, abnormal behavior information is extracted.

[0121] Based on the entity interaction feature set, and according to the spatiotemporal parameter set, analyze whether there is any abnormal behavior of the current entity interaction feature in the corresponding bus shelter spatiotemporal scenario. If so, extract the abnormal behavior information.

[0122] The spatiotemporal scenario benchmark model can be a dynamic behavior pattern library built based on historical normal operation data, which represents the legal behavior characteristics boundary of the bus shelter under specific spatiotemporal conditions (such as weekday morning rush hour, rainy or snowy weather).

[0123] Multidimensional comparison can be used to perform difference quantification analysis between the time, space and coupling dimension parameters of real-time data and the corresponding dimensions of the benchmark model.

[0124] Real-time spatiotemporal anomaly overlap can be a weighted evaluation value that combines temporal anomaly overlap, spatial anomaly overlap, and coupled anomaly overlap, reflecting the global anomaly probability of the current behavior deviating from the baseline model.

[0125] Abnormal behavior information can be a structured description that includes the type of abnormality (such as violent conflict or facility damage), spatiotemporal location (timestamp + coordinates), and confidence level.

[0126] Specifically, existing bus shelter anomaly detection technologies suffer from several limitations. Traditional threshold rules fail to adapt to dynamic scenarios, leading to the failure of static models. They also suffer from fragmented analysis by independently detecting temporal anomalies (such as prolonged loitering) or spatial anomalies (such as trespassing into restricted areas), ignoring the coupling effect between these two factors (e.g., late-night loitering combined with frequent abnormal operations of equipment within the bus shelter constitutes a real threat). Furthermore, they struggle with the degradation of video quality in rainy or foggy weather, resulting in the failure of single-video analysis models. Sensor false alarms (such as seat displacement caused by strong winds) lack multi-source verification mechanisms, resulting in weak scene generalization capabilities. These limitations make it difficult to adapt to complex and dynamic spatiotemporal scenarios, leading to high false alarm rates or missed detections of critical risks. This embodiment introduces a spatiotemporal scene benchmark model, combined with entity interaction feature sets, to achieve more accurate and contextualized perception of abnormal behavior. A multi-dimensional (time, space, time + space coupling) parallel comparison mechanism effectively addresses the shortcomings of single-dimensional analysis. A dynamic "normal behavior map"—the spatiotemporal scene benchmark model—is constructed as the core reference for anomaly detection. Based on massive historical normal operation spatiotemporal data, this model learns and characterizes the legal behavioral feature boundaries of various entity interaction features (human-to-human interaction, human-to-object interaction) in specific spatiotemporal scenarios (such as waiting areas during weekday morning rush hour on sunny days). It defines the dynamic range of reasonable behavior in this scenario and establishes time dimension benchmarks (learning typical interaction patterns in specific time periods such as morning rush hour), spatial dimension benchmarks (learning feature distributions in specific areas such as boarding points), and key coupling coefficient benchmarks (learning the influence rules of dynamic interaction between time and space on features, such as how to correct waiting area behavior patterns on rainy days). Finally, it forms a probabilistic normality map that adaptively evolves with the data. When performing contextualized anomaly detection, the system inputs the real-time acquired set of entity interaction features and the current spatiotemporal parameter set (time, location, environment), and performs a multi-dimensional comparison process: First, it extracts the baseline pattern and coupling correction factor under the corresponding time and space parameters; then, through the spatiotemporal covariance function, it analyzes the degree to which the real-time features deviate from the time and space baselines, respectively, to obtain the time anomaly overlap degree and the space anomaly overlap degree; simultaneously, through the spatiotemporal covariance function, it analyzes the degree to which the features deviate from the baseline after coupling correction, to obtain the coupling anomaly overlap degree. The system creatively integrates the overlap degrees of these three dimensions, and through weighted integration or logical combination, generates a comprehensive real-time spatiotemporal anomaly overlap scalar (0-1). The higher the value, the lower the overlap with the normal map and the greater the probability of an anomaly. When the overlap exceeds the contextual threshold, abnormal behavior is determined to exist, and abnormal behavior information is extracted: based on the entity interaction feature set, the entity that caused the abnormality and the specific interaction type are deeply located (such as abnormal gathering pattern in a certain area, abnormal contact between a passenger and a package, and violation of the status migration of a certain facility), the degree and pattern of the abnormality are quantified (such as sudden increase in density, unreasonable trajectory), the specific spatiotemporal context in which it occurred is associated, and a preliminary abnormality category inference is generated (such as crowding risk, lost item), and finally a structured report containing these elements is output, which significantly improves the accuracy of alarms and the operability of information.

[0127] This solution constructs a dynamically evolving spatiotemporal scenario benchmark model based on historical normal operation data, defines the boundaries of legal behavior features, and compares real-time entity interaction features with the model in multiple dimensions (time, space, coupling) to determine the degree of anomaly overlap, and calculates the comprehensive real-time spatiotemporal anomaly overlap. When the overlap is below a threshold, abnormal behavior information is deeply extracted and structurally described, embedding anomaly detection deeply into the spatiotemporal context. This significantly improves alarm accuracy, reduces the probability of false alarms and missed alarms, and enhances information operability, providing intelligent and context-aware decision support for public transportation safety early warning, emergency dispatch, and service optimization.

[0128] In some embodiments, human-to-human interaction feature sets and human-to-object interaction feature sets are extracted from the entity interaction feature set. Combined with temporal anomaly overlap, the temporal coupling effect of human-to-human and human-to-object interactions is analyzed to obtain abnormal behavior information of entity interaction features in the time dimension. Human-to-human interaction feature sets and human-to-object interaction feature sets are also extracted from the entity interaction feature set. Combined with spatial anomaly overlap, the spatial coupling effect of human-to-human and human-to-object interactions is analyzed to obtain abnormal behavior information of entity interaction features in the spatial dimension. A spatiotemporal dynamic alignment dataset is constructed based on the temporal and spatial dimension parameters in the spatiotemporal parameter set. A spatiotemporal coupling analysis framework is generated based on the spatiotemporal dynamic alignment dataset. Based on the entity interaction feature set and the spatiotemporal coupling analysis framework, the coupling effect of the entity interaction feature set in time and space is analyzed. Combined with coupling anomaly overlap, abnormal behavior information of entity interaction features under coupling relationships is obtained.

[0129] Temporal coupling effect analysis can be used to analyze the synergistic patterns of human-human / human-object interaction characteristics over time.

[0130] Spatial coupling effect analysis can be used to analyze the correlation of interactive features in spatial distribution.

[0131] Spatiotemporal dynamic alignment datasets can be structured databases that spatiotemporally synchronize and correlate time dimension parameters (event frequency), spatial dimension parameters (hotspot coordinates), and entity interaction features.

[0132] The spatiotemporal coupling analysis framework can be a rule engine for quantifying the three-dimensional correlation of "time-space-behavior" and diagnosing cross-dimensional abnormal logic chains.

[0133] Specifically, existing technologies suffer from dimensional silos, manifested in the following ways: the detection of temporal anomalies (such as late-night loitering) and spatial anomalies is independent; existing methods ignore the behavioral chain breakage problem of causal chains in human-to-human and human-to-object interactions; and latent risks are overlooked. This embodiment implements a complete process for refined anomaly detection. The complete process for refined anomaly detection is as follows: First, in the temporal dimension coupling analysis, the process first performs multi-source data collection and preprocessing, integrating human-to-human interaction data and human-to-object interaction data from sources such as video surveillance and Bluetooth probes. Then, feature separation and extraction are performed, extracting human-to-human interaction feature sets (such as density fluctuations and contact frequency) and human-to-object interaction feature sets (such as device occupancy rate). Next, single-dimensional anomaly detection is performed, using the HOI method to obtain entity interaction features, obtaining entity interaction feature frequencies through these interaction features, and identifying anomalies in entity interaction features based on percentile thresholds. The next crucial stage is coupling effect analysis. This involves timestamp overlap detection for two sets of abnormal time periods. When a decrease in "person-to-person contact frequency" (contact frequency less than a preset frequency) is detected during a specific time period (e.g., morning rush hour), a set of abnormal person-to-person feature points is output. Conversely, when "person-to-object contact duration increases (person-to-object contact frequency greater than a preset number of times)," a set of abnormal person-to-object feature points is output. Based on the rule base, these are identified as coupling events where "abnormal congestion leads to deterioration of traffic efficiency." A sliding window overlap detection method, rather than a fixed window, is used to avoid time boundary deviations. Finally, a collaborative warning is generated. If such abnormal coupling continues for more than the expected duration, a "peak-hour collaborative congestion risk" alarm is triggered, and a warning message containing the specific time period, feature change values, and traffic management suggestions is automatically output. The system separates the human-human interaction feature set (such as crowd density fluctuations) from the physical interaction feature set, and analyzes the mutual influence between the two in the same time period by combining the temporal anomaly overlap degree. For example, when a sudden drop in human-human contact frequency and a surge in card swiping time are detected at the same time during the morning peak, the system reveals the temporal coupling effect of "abnormal lingering leading to deterioration of traffic efficiency" and generates a warning of "coordinated congestion risk during peak hours". At the same time, spatial dimension coupling analysis is performed. Based on the same spatial location parameters, human-human interaction features (such as gathering patterns) and human-object interaction features (such as seat occupancy status) are fused. Through spatial anomaly overlap degree, spatial gridding and feature synchronization are performed to build a spatial coupling feature library. Real-time spatial grid scanning is matched. For example, when a circular crowd gathers in the waiting area and the central seat is vacant for a long time, the system determines the spatial coupling effect of "abnormal onlookers leading to resource idleness" and outputs the decision clue of "suspicious items attracting abnormal gathering".Further, through deep modeling of spatiotemporal coupling, a spatiotemporal dynamic alignment dataset is constructed based on a spatiotemporal parameter set. The entire platform data is indexed using timestamps and spatial grid coordinates, and a spatiotemporal coupling analysis framework is established. This framework quantifies the interactive influence of parameters using spatiotemporal association rules (e.g., "rainy day + boarding point → pushing probability > probability threshold"). It also utilizes cross-dimensional feature transmission algorithms (e.g., the lag effect coefficient of entrance congestion on waiting area density), a domain-knowledge-driven rule base, and data-driven transmission relationship learning and coefficient calibration to construct an interpretable computational model. Finally, at the coupling relationship layer, entity features are input into this framework, and combined with coupling anomaly overlap, composite anomalies are identified. For example, in the scenario of "large event dispersal + platform exit," when the coupling effect between the intensity of human-to-human pushing and the rate of vehicle passage under the bus shelter significantly deviates from the baseline, abnormal behavioral information of "risk of traffic paralysis caused by instantaneous overload" is extracted, and key congestion transmission paths are located, forming a closed-loop decision-making basis with a spatiotemporal causal chain.

[0134] This approach analyzes temporal / spatial coupling effects from two perspectives: temporal linkage of behaviors (such as the event sequence coordination of human-to-human and human-object interactions) and regional spatial transmission (such as the behavior density mapping of hotspot areas). This approach accurately captures abnormal interaction features in a single dimension. Simultaneously, the spatiotemporal coupling analysis framework employs a triple mechanism of gridded behavior density fusion, coupling coefficient quantification, and behavior chain verification to deeply analyze complex cross-dimensional patterns such as "human-to-human collaboration driving human-object anomalies" or "human-object anomalies inducing human-to-human aggregation." This constructs an analytical system from single-dimensional anomalies (temporal / spatial) to coupled anomalies, significantly improving the ability to identify hidden collaborative behaviors.

[0135] In some embodiments, based on real-time spatiotemporal anomaly overlap and entity interaction features, when an abnormal correlation signal of the time dimension parameter or spatial dimension parameter of the spatiotemporal parameter set is detected in the entity interaction feature set, and the abnormal overlap of the corresponding dimension in the spatiotemporal parameter set exceeds the basic threshold, a first-level anomaly warning signal is triggered and output; based on real-time spatiotemporal anomaly overlap and entity interaction features, when the abnormal overlap of the coupling coefficient of the time dimension parameter and the spatial dimension parameter exceeds the upgrade threshold, and the coupling abnormal overlap in the entity interaction features forms a logical correlation, a second-level anomaly warning signal is triggered and output.

[0136] A Level 1 anomaly warning signal can be a primary alarm signal indicating a significant anomaly in a single dimension (time or space).

[0137] The Level 2 anomaly warning signal can be an upgraded alarm signal for spatiotemporal coupling anomalies.

[0138] Anomaly correlation signals can be indicators of abnormal matches between time / space dimension parameters and entity behavioral characteristics.

[0139] The basic threshold can be the lowest critical value for a single-dimensional anomaly that triggers a level-one warning.

[0140] The upgrade threshold can be the lowest critical value of the spatiotemporal coupling anomaly that triggers a level-two warning.

[0141] Logical connections can be causal chains between spatiotemporal anomalies and behavioral characteristics.

[0142] Specifically, in spatiotemporal monitoring environments, anomalies in the temporal dimension and anomalies in the spatial dimension often have a deep-seated interactive relationship. For example, sudden spatial clustering (high overlap of spatial anomalies) may directly lead to abnormal compression or extension of the interaction time of entities in that area (time dimension anomaly associated signal). If only time and space parameters or time + space coupling coefficients are analyzed, it will be difficult to understand the essence of this cross-dimensional linkage. By deploying a multi-source sensor network, the system captures human-to-human and human-to-object interaction features in real time and transforms the raw data into behavioral indicators. The system automatically constructs dynamic spatiotemporal units—binding event cycles and physical risk domains to generate independent analysis containers. In the cross-dimensional anomaly correlation analysis phase: first, it compares historical baselines to mark time-dimensional collaborative offsets and scans spatial differences to identify local mutations; second, it traces back the behavioral logic chain and verifies the transmission integrity through device logs; based on this, it executes coupled risk decisions: when a single-dimensional anomaly continuously exceeds the limit, a level-one warning is triggered; if human-to-human / human-to-object behavioral collaborative deviations occur within the spatiotemporal unit and the behavioral chain is verified, a level-two warning is triggered; finally, it drives a tiered response: a level-one warning initiates voice guidance, and a level-two warning simultaneously activates real-time alarms and broadcasts diversion routes, forming a closed-loop risk prevention and control system. The core of this embodiment's tiered warning mechanism lies in establishing a dynamic assessment system for the severity of spatiotemporal anomalies. The system first identifies anomalous correlation signals in the time or spatial dimensions of spatiotemporal parameters by scanning human-to-human and human-to-object interaction features in real time. These signals manifest as sudden, coordinated deviations of multiple features within a specific spatiotemporal unit. For example, during the morning rush hour, both the frequency of person-to-person contact and the frequency of person-to-object interaction are detected to rise simultaneously, and their fluctuation curves show a strong statistical correlation. When the real-time spatiotemporal anomaly overlap corresponding to this type of abnormal correlation signal, such as the time-dimensional anomaly overlap, exceeds a preset basic threshold, the system immediately triggers a Level 1 anomaly warning signal. This signal not only locates the abnormal spatiotemporal unit, such as the morning rush hour, but also clearly reveals the abnormal coupling relationship between features; a strong correlation exists between the decrease in contact frequency and the increase in person-to-object interaction frequency. The output directly points to the root cause of the single-dimensional risk, specifically described as: Level 1 warning: A strong correlation is detected between the increase in person-to-person contact frequency and the increase in person-to-object interaction frequency during the morning rush hour, and the time anomaly overlap exceeds the basic threshold. This type of warning is suitable for handling risks that are localized and induced by single-dimensional anomalies, such as abnormal behavior of people under bus shelters during peak hours. When the risk evolves to the spatiotemporal interaction level, the system activates a Level 2 warning mechanism. Its triggering depends on two key conditions. First, the degree of abnormal overlap corresponding to the coupling coefficient between the time dimension parameters and the spatial dimension parameters must exceed the upgrade threshold. This coupling coefficient reflects the intensity of spatiotemporal interaction, such as the probability weight of pushing on a platform in combination with rainy weather. Second, there must be a verifiable logical connection between human-to-human interaction features and human-to-object interaction features, that is, a causal transmission chain must be formed across feature outliers, such as the mathematical dependency that a surge in human-to-human pushing intensity directly leads to an increase in the frequency of human-to-object interaction.In scenarios where temporal parameters, such as the dispersal of large events, and spatial parameters, such as platform exits, are superimposed, if the coupling coefficient between pushing intensity and the frequency of human interaction deviates from the baseline value, and the degree of abnormal coupling overlap equals the escalation threshold, and data verification confirms a causal chain where crowd congestion is the cause leading to traffic paralysis, the system generates a Level 2 anomaly warning signal. The signal is described as: Level 2 Warning: In scenarios involving the dispersal of large events and platform exits, the coupling coefficient between pushing intensity and the characteristics of physical interaction and the set of spatial parameters deviates from the baseline value, the degree of abnormal coupling overlap is equal to or greater than the escalation threshold, and the logical correlation is that crowd congestion is the cause leading to traffic paralysis. This signal reveals the systemic risks arising from spatiotemporal interactions, prompting emergency departments to simultaneously activate composite contingency plans such as crowd evacuation, equipment repair, and the opening of backup channels.

[0143] This solution utilizes real-time spatiotemporal anomaly overlap to perform preliminary quantitative screening of single-dimensional anomalies (Level 1 warning). Building upon this, a hierarchical anomaly identification framework is constructed by analyzing the coupling degree between temporal and spatial anomalies and the logical chains formed in entity interaction behavior (Level 2 warning). This not only improves the response speed to single significant anomaly events (Level 1) but also significantly enhances the insight into complex and hidden cross-spatiotemporal collaborative anomaly patterns (Level 2), thereby comprehensively improving the comprehensiveness and accuracy of the anomaly warning system.

[0144] Figure 3 A schematic diagram of a smart bus shelter monitoring system with intelligent monitoring of abnormal behavior is provided as an embodiment of this application, as shown below. Figure 3 As shown, the intelligent bus shelter monitoring system 300 with abnormal behavior intelligent monitoring in this embodiment includes: a perception and analysis module 301, an extraction and modeling module 302, a detection and identification module 303, and a decision response module 304.

[0145] The perception and analysis module 301 is used to acquire multi-source monitoring data of the target bus shelter, and based on the multi-source monitoring data, analyze the interaction behavior of people and objects to generate a set of entity interaction features. The modeling extraction module 302 is used to extract timestamp sequences and spatial coordinate sequences based on the entity interaction feature set, and to generate a spatiotemporal parameter set based on the timestamp sequences and spatial coordinate sequences; The detection and recognition module 303 is used to analyze whether there is abnormal behavior of the current entity interaction features in the corresponding bus shelter spatiotemporal scene based on the entity interaction feature set and the spatiotemporal parameter set; if so, it extracts abnormal behavior information. The decision response module 304 is used to determine and output an abnormal warning signal based on the abnormal behavior information.

[0146] Optionally, in the perception and analysis module 301, the multi-source monitoring data includes real-time video stream data, thermal imaging data, standing position distribution data, real-time bus operation data, and bus facility status sensor data; The real-time video stream data, the thermal imaging data, the standing position distribution data, the real-time bus operation data, and the bus facility status sensor data are all aligned across dimensions in time and space through time synchronization protocols and spatial synchronization protocols. The time synchronization protocol is used to ensure that all data points collected from all data sources are timestamped based on a unified high-precision clock source, eliminating inherent clock errors between different devices and ensuring strict consistency of behavioral events on the timeline. The spatial synchronization protocol is used to establish a unified mapping relationship between the data collected from various data sources in the physical spatial coordinate system of the bus shelter, and to convert the target position, attitude, heat map and other information from different sources to the same reference coordinate system.

[0147] Optionally, when the perception and analysis module 301 analyzes the interaction behavior of people and objects based on the multi-source monitoring data to generate a set of entity interaction features, it is specifically used for: Based on the multi-source monitoring data, a multi-target tracking algorithm is used to screen people and objects in the multi-source monitoring data and extract crowd monitoring features and object monitoring features. Based on the real-time video stream data, the standing position, and the thermal imaging data, the crowd monitoring features are analyzed to identify the human-to-human aggregation patterns, limb movement features, and eye contact features, thereby generating a human-to-human interaction feature set. Based on the real-time video stream data and the real-time bus operation data, analyze the human-object contact behavior characteristics, object displacement characteristics and interactive action characteristics in the object monitoring characteristics, and generate a human-object interaction feature set; Integrate the human-to-human interaction feature set with the human-to-object interaction feature set to generate an entity interaction feature set.

[0148] Optionally, when the perception and analysis module 301 integrates the human-to-human interaction feature set and the human-to-object interaction feature set to generate the entity interaction feature set, it is specifically used for: Based on the multi-source monitoring data, individual behavior, group emergence, and environmental coupling mechanisms are analyzed. Through a group behavior dynamics model, contact behavior characteristics, object displacement characteristics, and interactive action characteristics are determined to form a human-to-human interaction feature set. Based on the video stream data and the public transportation facility status sensor data, the changes in the status of the items are analyzed. Combining physical attributes, spatiotemporal trajectories, and human behavior triggering conditions, the item status migration analysis algorithm is used to determine contact behavior characteristics, item displacement characteristics, and interactive action characteristics, thus forming a human-item interaction feature set.

[0149] Optionally, the extraction and modeling module 302, in integrating the human-to-human interaction feature set and the human-to-object interaction feature set to generate an entity interaction feature set, is specifically used for: Based on the multi-source monitoring data, individual behavior, group emergence, and environmental coupling mechanisms are analyzed. Through a group behavior dynamics model, contact behavior characteristics, object displacement characteristics, and interactive action characteristics are determined to form a human-to-human interaction feature set. Based on the video stream data and the public transportation facility status sensor data, the changes in the status of the items are analyzed. Combining physical attributes, spatiotemporal trajectories, and human behavior triggering conditions, the item status migration analysis algorithm is used to determine contact behavior characteristics, item displacement characteristics, and interactive action characteristics, thus forming a human-item interaction feature set.

[0150] Optionally, the extraction and modeling module 302 extracts timestamp sequences and spatial coordinate sequences based on the entity interaction feature set, and generates a spatiotemporal parameter set based on the timestamp sequences and spatial coordinate sequences, specifically for: Based on the entity interaction feature set, the timestamp sequence of each feature event is extracted, and a histogram of event occurrence frequency distribution is generated through a time window sliding mechanism. Based on the real-time video stream data, the thermal imaging data, and the standing position distribution data, the spatial coordinate sequence of each entity is determined, a three-dimensional density distribution cloud map is constructed, and the coordinates of spatial clustering hotspots are determined by analyzing the three-dimensional density distribution cloud map. Spatiotemporal coding is performed based on the frequency distribution histogram and the spatial hotspot coordinates to determine the time dimension parameters, spatial dimension parameters, and coupling coefficients. The spatiotemporal parameter set is generated based on the time dimension parameters, spatial dimension parameters, and coupling coefficients.

[0151] Optionally, the detection and identification module 303 performs spatiotemporal coding based on the frequency distribution histogram and the spatial hotspot coordinates to determine the time dimension parameters, spatial dimension parameters, and coupling coefficients, specifically for: Based on the timestamp sequence of each feature event, the time dimension parameter is obtained by dynamically segmenting the timestamp sequence to determine the periodic feature vector with time context association. Based on the three-dimensional density distribution cloud map, the bus shelter area is divided into several adaptive grids. By analyzing the spatial clustering hotspot coordinates, the distance between each sub-area and the bus facilities is obtained, thus obtaining the spatial dimension parameters. Based on the time dimension parameters and spatial dimension parameters, the coupling relationship between different spatial coordinates under the same time window and the same spatial coordinate under different time windows is analyzed, and the coupling coefficient is generated.

[0152] Optionally, the detection and recognition module 303 extracts abnormal behavior information based on the entity interaction features and the real-time spatiotemporal anomaly overlap, specifically in the following ways: Extract the human-human interaction feature set and the human-object interaction feature set from the entity interaction feature set, and combine them with the time anomaly overlap degree to analyze the time coupling effect of human-human interaction and human-object interaction, so as to obtain the abnormal behavior information of entity interaction features in the time dimension. The entity interaction feature set extracts the human-human interaction feature set and the human-object interaction feature set. Combined with the spatial anomaly overlap, the spatial coupling effect of human-human interaction and human-object interaction is analyzed to obtain abnormal behavior information of entity interaction features in the spatial dimension. Based on the time dimension parameter and the space dimension parameter in the spatiotemporal parameter set, a spatiotemporal dynamic alignment dataset is constructed, and a spatiotemporal coupling analysis framework is generated based on the spatiotemporal dynamic alignment dataset. Based on the entity interaction feature set, and combined with the spatiotemporal coupling analysis framework, the coupling effect of the entity interaction feature set in time and space is analyzed. Combined with the coupling anomaly overlap degree, abnormal behavior information of entity interaction features under coupling relationship is obtained.

[0153] Optionally, the detection and recognition module 303, based on the entity interaction feature set and according to the spatiotemporal parameter set, analyzes whether there is abnormal behavior in the corresponding bus shelter spatiotemporal scenario for the current entity interaction features. If so, it extracts abnormal behavior information, specifically for: Based on historical spatiotemporal data of normal operation, construct a spatiotemporal scenario benchmark model; Based on the time dimension parameters, spatial dimension parameters and coupling coefficients of the spatiotemporal parameter set, a multi-dimensional comparison is performed with the spatiotemporal scene benchmark model to determine the time anomaly overlap, spatial anomaly overlap and coupling anomaly overlap, and the real-time spatiotemporal anomaly overlap is calculated. Based on the entity interaction features and the real-time spatiotemporal anomaly overlap, abnormal behavior information is extracted.

[0154] Optionally, the decision response module 304 determines and outputs an abnormal warning signal based on the abnormal behavior information, specifically for: Based on the real-time spatiotemporal anomaly overlap degree and the entity interaction features, when an abnormal correlation signal of the time dimension parameter or spatial dimension parameter of the spatiotemporal parameter set is detected in the entity interaction feature set, and the abnormal overlap degree of the corresponding dimension in the spatiotemporal parameter set exceeds the basic threshold, a first-level anomaly warning signal is triggered and output. Based on the real-time spatiotemporal anomaly overlap degree and the entity interaction features, when the anomaly overlap degree of the coupling coefficient of the time dimension parameter and the spatial dimension parameter exceeds the upgrade threshold, and the coupling anomaly overlap degree in the entity interaction features forms a logical association, a level two anomaly warning signal is triggered and output.

[0155] The system in this embodiment can be used to execute the methods of any of the above embodiments, and its implementation principle and technical effect are similar, so they will not be described again here.

Claims

1. A smart bus shelter monitoring method with intelligent monitoring of abnormal behavior, characterized in that, include: Acquire multi-source monitoring data of the target bus shelter, and based on the multi-source monitoring data, analyze the interaction behavior of people and objects to generate a set of entity interaction features; Based on the entity interaction feature set, timestamp sequences and spatial coordinate sequences are extracted, and a spatiotemporal parameter set is generated based on the timestamp sequences and spatial coordinate sequences. Based on the entity interaction feature set, and according to the spatiotemporal parameter set, analyze whether there is abnormal behavior of the current entity interaction feature in the corresponding bus shelter spatiotemporal scenario. If so, extract the abnormal behavior information. Based on the abnormal behavior information, an abnormal warning signal is determined and output.

2. The method according to claim 1, characterized in that, The multi-source monitoring data includes real-time video stream data, thermal imaging data, standing position distribution data, real-time bus operation data, and bus facility status sensor data. The real-time video stream data, the thermal imaging data, the standing position distribution data, the real-time bus operation data, and the bus facility status sensor data are all aligned across dimensions in time and space through time synchronization protocols and spatial synchronization protocols. The time synchronization protocol is used to ensure that all data points collected from all data sources are timestamped based on a unified high-precision clock source, eliminating inherent clock errors between different devices and ensuring strict consistency of behavioral events on the timeline. The spatial synchronization protocol is used to establish a unified mapping relationship between the data collected from various data sources in the physical spatial coordinate system of the bus shelter, and to convert the target position, attitude, heat map and other information from different sources to the same reference coordinate system.

3. The method according to claim 2, characterized in that, The step involves analyzing the multi-source monitoring data to understand the interaction behavior of people and objects, generating a set of entity interaction features, including: Based on the multi-source monitoring data, a multi-target tracking algorithm is used to screen people and objects in the multi-source monitoring data and extract crowd monitoring features and object monitoring features. Based on the real-time video stream data, the standing position, and the thermal imaging data, the crowd monitoring features are analyzed to identify the human-to-human aggregation patterns, limb movement features, and eye contact features, thereby generating a human-to-human interaction feature set. Based on the real-time video stream data and the real-time bus operation data, analyze the human-object contact behavior characteristics, object displacement characteristics and interactive action characteristics in the object monitoring characteristics, and generate a human-object interaction feature set; Integrate the human-to-human interaction feature set with the human-to-object interaction feature set to generate an entity interaction feature set.

4. The method according to claim 3, characterized in that, The process of integrating the human-to-human interaction feature set and the human-to-object interaction feature set to generate an entity interaction feature set includes: Based on the multi-source monitoring data, individual behavior, group emergence, and environmental coupling mechanisms are analyzed. Through a group behavior dynamics model, contact behavior characteristics, object displacement characteristics, and interactive action characteristics are determined to form a human-to-human interaction feature set. Based on the video stream data and the public transportation facility status sensor data, the changes in the status of the items are analyzed. Combining physical attributes, spatiotemporal trajectories, and human behavior triggering conditions, the item status migration analysis algorithm is used to determine contact behavior characteristics, item displacement characteristics, and interactive action characteristics, thus forming a human-item interaction feature set.

5. The method according to claim 1, characterized in that, The process involves extracting a timestamp sequence and a spatial coordinate sequence based on the entity interaction feature set, and generating a spatiotemporal parameter set based on the timestamp sequence and spatial coordinate sequence, including: Based on the entity interaction feature set, the timestamp sequence of each feature event is extracted, and a histogram of event occurrence frequency distribution is generated through a time window sliding mechanism. Based on the real-time video stream data, the thermal imaging data, and the standing position distribution data, the spatial coordinate sequence of each entity is determined, a three-dimensional density distribution cloud map is constructed, and the coordinates of spatial clustering hotspots are determined by analyzing the three-dimensional density distribution cloud map. Spatiotemporal coding is performed based on the frequency distribution histogram and the spatial hotspot coordinates to determine the time dimension parameters, spatial dimension parameters, and coupling coefficients. The spatiotemporal parameter set is generated based on the time dimension parameters, spatial dimension parameters, and coupling coefficients.

6. The method according to claim 5, characterized in that, Spatiotemporal coding is performed based on the frequency distribution histogram and the coordinates of the spatially clustered hotspots to determine the time dimension parameters, spatial dimension parameters, and coupling coefficients, including: Based on the timestamp sequence of each feature event, the time dimension parameter is obtained by dynamically segmenting the timestamp sequence to determine the periodic feature vector with time context association. Based on the three-dimensional density distribution cloud map, the bus shelter area is divided into several adaptive grids. By analyzing the spatial clustering hotspot coordinates, the distance between each sub-area and the bus facilities is obtained, thus obtaining the spatial dimension parameters. Based on the time dimension parameters and spatial dimension parameters, the coupling relationship between different spatial coordinates under the same time window and the same spatial coordinate under different time windows is analyzed, and the coupling coefficient is generated.

7. The method according to claim 1, characterized in that, Based on the entity interaction feature set and the spatiotemporal parameter set, the system analyzes whether there are abnormal behaviors in the corresponding bus shelter spatiotemporal scenario for the current entity interaction features. If so, it extracts abnormal behavior information, including: Based on historical spatiotemporal data of normal operation, construct a spatiotemporal scenario benchmark model; Based on the time dimension parameters, spatial dimension parameters and coupling coefficients of the spatiotemporal parameter set, a multi-dimensional comparison is performed with the spatiotemporal scene benchmark model to determine the time anomaly overlap, spatial anomaly overlap and coupling anomaly overlap, and the real-time spatiotemporal anomaly overlap is calculated. Based on the entity interaction features and the real-time spatiotemporal anomaly overlap, abnormal behavior information is extracted.

8. The method according to claim 7, characterized in that, The step of extracting abnormal behavior information based on the entity interaction features and the real-time spatiotemporal anomaly overlap includes: Extract the human-human interaction feature set and the human-object interaction feature set from the entity interaction feature set, and combine them with the time anomaly overlap degree to analyze the time coupling effect of human-human interaction and human-object interaction, so as to obtain the abnormal behavior information of entity interaction features in the time dimension. The entity interaction feature set extracts the human-human interaction feature set and the human-object interaction feature set. Combined with the spatial anomaly overlap, the spatial coupling effect of human-human interaction and human-object interaction is analyzed to obtain abnormal behavior information of entity interaction features in the spatial dimension. Based on the time dimension parameter and the space dimension parameter in the spatiotemporal parameter set, a spatiotemporal dynamic alignment dataset is constructed, and a spatiotemporal coupling analysis framework is generated based on the spatiotemporal dynamic alignment dataset. Based on the entity interaction feature set, and combined with the spatiotemporal coupling analysis framework, the coupling effect of the entity interaction feature set in time and space is analyzed. Combined with the coupling anomaly overlap degree, abnormal behavior information of entity interaction features under coupling relationship is obtained.

9. The method according to claim 7, characterized in that, The step of determining and outputting an abnormal warning signal based on the abnormal behavior information includes: Based on the real-time spatiotemporal anomaly overlap degree and the entity interaction features, when an abnormal correlation signal of the time dimension parameter or spatial dimension parameter of the spatiotemporal parameter set is detected in the entity interaction feature set, and the abnormal overlap degree of the corresponding dimension in the spatiotemporal parameter set exceeds the basic threshold, a first-level anomaly warning signal is triggered and output. Based on the real-time spatiotemporal anomaly overlap degree and the entity interaction features, when the anomaly overlap degree of the coupling coefficient of the time dimension parameter and the spatial dimension parameter exceeds the upgrade threshold, and the coupling anomaly overlap degree in the entity interaction features forms a logical association, a level two anomaly warning signal is triggered and output.

10. A smart bus shelter monitoring system with intelligent monitoring of abnormal behavior, characterized in that, include: The perception and analysis module is used to acquire multi-source monitoring data of the target bus shelter, and based on the multi-source monitoring data, analyze the interaction behavior of people and objects to generate a set of entity interaction features. The extraction modeling module is used to extract timestamp sequences and spatial coordinate sequences based on the entity interaction feature set, and to generate a spatiotemporal parameter set based on the timestamp sequences and spatial coordinate sequences; The detection and recognition module is used to analyze whether there is abnormal behavior of the current entity interaction features in the corresponding bus shelter spatiotemporal scenario based on the entity interaction feature set and the spatiotemporal parameter set; if so, it extracts abnormal behavior information. The decision response module is used to determine and output an abnormal warning signal based on the abnormal behavior information.

Images