Non-face-face person dynamic tracking method and system for construction site

Abstract

The invention discloses a construction site-oriented non-face person dynamic tracking method and system, and the method comprises the steps: collecting the information data of entering persons at a construction site, comparing the information data with a construction face list, and screening suspected non-face persons; establishing a heterogeneous graph based on the information data of the suspected non-face-to-face persons and information data in a face-to-face person database, and performing local similar node aggregation based on the heterogeneous graph to identify target non-face-to-face persons; obtaining a movement track of the target non-face person, and calculating a deviation degree based on the movement track of the non-face person to obtain a track deviation degree index of the target non-face person; constructing a behavior causal graph model based on the trajectory deviation index, and judging the behavior state of the target non-face person based on the behavior causal graph model to generate a behavior risk judgment result; and generating a risk level label based on the behavior risk discrimination result.

Description

Technical Field

[0001] This invention relates to the field of behavior recognition technology, and more specifically, to a method and system for dynamic tracking of non-ticketed personnel at construction sites. Background Technology

[0002] With the continuous improvement of informatization in the construction industry, safety management and personnel access control at construction sites have become key aspects of smart construction site construction. Existing construction sites typically use access control card systems, video surveillance systems, and personnel positioning devices to manage construction workers. By comparing worker identification with the roster on the construction ticket, the aim is to control personnel access and verify identity. However, traditional real-name management methods mainly rely on manual patrols, card comparisons, or post-event data verification, which suffers from low identification efficiency, poor response time, and the inability to promptly detect unauthorized personnel entering the construction site.

[0003] In complex construction site environments, unregistered personnel (such as unregistered workers, visitors, or individuals using others' information) pose a high potential risk. Once they enter the work area, they may cause safety hazards, unclear responsibilities, or disruption to construction progress. Currently, the following problems exist: the existing system's single data source is easily affected by ambient lighting, obstructions, or equipment malfunctions, leading to inaccurate identification of unregistered personnel (i.e., those not registered on the construction roster), hindering continuous tracking and dynamic control; existing path analysis algorithms are mostly based on geometric distance or fixed area divisions, lacking constraints on the construction site's topology, and cannot accurately reflect the actual movement deviation characteristics of unregistered personnel; traditional anomaly detection mechanisms often use static threshold settings, failing to dynamically adjust risk assessment criteria based on contextual information such as construction plans and area permissions, easily leading to false alarms or missed alarms.

[0004] Therefore, a method for dynamic tracking of non-ticketed personnel at construction sites is needed. Summary of the Invention

[0005] This invention proposes a dynamic tracking method and system for non-ticketed personnel at construction sites to solve the problem of how to identify non-ticketed personnel at construction sites.

[0006] To address the aforementioned problems, according to one aspect of the present invention, a method for dynamic tracking of non-ticketed personnel at construction sites is provided, the method comprising:

[0007] Collect information on personnel entering the construction site and compare it with the list of personnel on the construction ticket to screen out suspected non-ticket personnel;

[0008] A heterogeneous graph is established based on the information data of the suspected non-ticket holders and the information data in the ticket holder database. Local similar nodes are aggregated based on the heterogeneous graph to identify target non-ticket holders with abnormal identity characteristics.

[0009] The movement trajectory of the target non-ticket person is obtained, and the deviation is calculated based on the movement trajectory of the non-ticket person to obtain the trajectory deviation index of the target non-ticket person.

[0010] A behavioral causal graph model is constructed based on the trajectory deviation index, and the behavioral causal graph model is used to determine the behavioral state of the target non-ticket personnel, generating behavioral risk judgment results.

[0011] Risk level labels are generated based on the behavioral risk assessment results.

[0012] Preferably, the nodes of the heterogeneous graph include personnel identity nodes, entry and exit record nodes, and scene location nodes, and the edges of the heterogeneous graph are used to represent the interactive relationship between personnel and their behavioral trajectories.

[0013] Preferably, the aggregation of locally similar nodes based on the heterogeneous graph to identify target non-ticket holders with abnormal identity characteristics includes:

[0014] The structure of the heterogeneous graph is optimized based on graph regularization constraint mechanism and sparse relation constraint mechanism;

[0015] A node aggregation mechanism based on local subgraphs is used to cluster and align the ticket holder nodes in the optimized heterogeneous graph to generate reference group features;

[0016] The node features of the suspected non-ticket holders are compared with the features of the reference group to identify target non-ticket holders whose identity features are abnormal.

[0017] Preferably, the deviation is calculated based on the movement trajectory of the non-ticket holder to obtain the trajectory deviation index of the target non-ticket holder, including:

[0018] Based on the movement trajectory, the spatial location information of the target non-ticket personnel at different times is mapped to the construction site topology model; wherein, the construction site topology model consists of multiple nodes and edges, the nodes correspond to key location units of the construction area, including: entrance and exit nodes, work area nodes and hazard source nodes, and the edges represent the path connection relationship that personnel can walk on;

[0019] In the construction site topology model, the path potential function is modeled for the trajectory of non-ticketed personnel and the authorized path in the construction area based on the graph potential field path constraint algorithm;

[0020] Based on the path potential function, combined with the edge weight constraints between nodes and the spatial topological distance, the trajectory of non-ticket personnel is decomposed and compared segment by segment to generate the corresponding deviation index sequence.

[0021] Extract the cumulative deviation and the maximum single-segment deviation from the deviation index sequence as two feature parameters to quantify the degree of trajectory abnormality of the target non-ticket personnel, and output the trajectory deviation index.

[0022] Preferably, the graph potential field path constraint algorithm includes:

[0023]

[0024] Among them, E risk (p) represents the comprehensive risk potential value of non-ticket holders under trajectory p; p t q represents the spatial location node of a person not represented by a ticket at time t; t This represents the standard location node of the authorized path in the construction area at time t; d(p t ,q t ) indicates the actual location p of the person not represented on the ticket. t With standard node q t Spatial topological distance between them; w(p) t ,q t ) represents the connection node p in the topology graph. t With q t The path edge weights; T represents the total temporal length of the trajectory; This represents the maximum deviation of a single segment in the trajectory; π(a t |s t ) indicates that in state s t The person who took the action of not receiving the ticket a t The policy probability; α represents the weight coefficient of cumulative deviation; β represents the power coefficient of deviation; γ represents the weight coefficient of the maximum single-segment deviation; λ represents the reinforcement learning risk correction factor.

[0025] Preferably, the behavioral causal graph model includes:

[0026]

[0027] Among them, C anom (u) represents the causal anomaly score for non-ticket recipient u; M represents the number of behavioral outcome variables involved in the behavioral causal graph model; Y i This represents the i-th actually observed behavioral outcome; This represents the i-th expected value obtained from the causal model inference; denoted as the variance of the i-th behavioral outcome in the historical normal ticket holder group; N represents the number of dimensions for comparing causal effects; This represents the observed causal effect of non-ticket holders in the j-th dimension; ρ represents the reference causal effect value for normal ticket holders in the same dimension; L represents the number of detected permission conflict events; k Let ξ represent the risk factor of the k-th permission conflict event; ξ represent the weight factor of the residual term; ν represent the weight factor of the causal effect deviation term; and χ represent the weight factor of the permission conflict term.

[0028] Preferably, the behavioral risk assessment result is generated by determining the behavioral state of the target non-ticket holder based on the behavioral causal graph model, including:

[0029] In the behavioral causal graph model, a causal inference-driven anomaly detection mechanism is introduced. Structural equation modeling and comparative intervention methods are used to compare the causal effects of the behavioral sequence of the target non-ticket holder with the baseline behavioral pattern of the normal ticket holder, determine the potential causal triggers of the abnormal behavior, and extract behavioral state feature parameters. Among them, the behavioral state feature parameters include: the duration of the anomaly, the sensitivity of the deviation area, and the frequency of permission conflicts.

[0030] Based on the aforementioned behavioral state characteristic parameters, a behavioral risk score is calculated for the target non-ticket personnel in a preset construction environment; wherein, the risk score is based on three dimensions: trajectory deviation, plan deviation, and permission violation.

[0031] Based on the correspondence between the behavioral risk score and the preset multi-level risk threshold range, the behavioral risk judgment result of the target non-ticket personnel is generated.

[0032] According to another aspect of the present invention, a non-ticketed personnel dynamic tracking system for construction sites is provided, the system comprising:

[0033] The screening unit is used to collect information data of personnel entering the construction site and compare it with the roster on the construction ticket to screen out suspected non-ticket personnel.

[0034] The aggregation and identification unit is used to establish a heterogeneous graph based on the information data of the suspected non-ticket personnel and the information data in the ticket personnel database, and to aggregate local similar nodes based on the heterogeneous graph in order to identify target non-ticket personnel with abnormal identity characteristics.

[0035] The deviation calculation unit is used to obtain the movement trajectory of the target non-ticket person, calculate the deviation based on the movement trajectory of the non-ticket person, and obtain the trajectory deviation index of the target non-ticket person.

[0036] The risk assessment unit is used to construct a behavioral causal graph model based on the trajectory deviation index, and to assess the behavioral state of the target non-ticket personnel based on the behavioral causal graph model, thereby generating a behavioral risk assessment result.

[0037] An alarm unit is used to generate a risk level label based on the behavioral risk judgment result.

[0038] Preferably, the nodes of the heterogeneous graph include personnel identity nodes, entry and exit record nodes, and scene location nodes, and the edges of the heterogeneous graph are used to represent the interactive relationship between personnel and their behavioral trajectories.

[0039] Preferably, the aggregation and identification unit, based on the heterogeneous graph, aggregates locally similar nodes to identify target non-ticket personnel with abnormal identity characteristics, including:

[0040] The structure of the heterogeneous graph is optimized based on graph regularization constraint mechanism and sparse relation constraint mechanism;

[0041] A node aggregation mechanism based on local subgraphs is used to cluster and align the ticket holder nodes in the optimized heterogeneous graph to generate reference group features;

[0042] The node features of the suspected non-ticket holders are compared with the features of the reference group to identify target non-ticket holders whose identity features are abnormal.

[0043] Preferably, the deviation calculation unit calculates the deviation based on the movement trajectory of the non-ticket holder to obtain the trajectory deviation index of the target non-ticket holder, including:

[0044] Based on the movement trajectory, the spatial location information of the target non-ticket personnel at different times is mapped to the construction site topology model; wherein, the construction site topology model consists of multiple nodes and edges, the nodes correspond to key location units of the construction area, including: entrance and exit nodes, work area nodes and hazard source nodes, and the edges represent the path connection relationship that personnel can walk on;

[0045] In the construction site topology model, the path potential function is modeled for the trajectory of non-ticketed personnel and the authorized path in the construction area based on the graph potential field path constraint algorithm;

[0046] Based on the path potential function, combined with the edge weight constraints between nodes and the spatial topological distance, the trajectory of non-ticket personnel is decomposed and compared segment by segment to generate the corresponding deviation index sequence.

[0047] Extract the cumulative deviation and the maximum single-segment deviation from the deviation index sequence as two feature parameters to quantify the degree of trajectory abnormality of the target non-ticket personnel, and output the trajectory deviation index.

[0048] Preferably, the graph potential field path constraint algorithm includes:

[0049]

[0050] Among them, E risk (p) represents the comprehensive risk potential value of non-ticket holders under trajectory p; p t q represents the spatial location node of a person not represented by a ticket at time t; t This represents the standard location node of the authorized path in the construction area at time t; d(p t ,q t ) indicates the actual location p of the person not represented on the ticket. t With standard node q t Spatial topological distance between them; w(p) t ,q t ) represents the connection node p in the topology graph. t With q t The path edge weights; T represents the total temporal length of the trajectory; This represents the maximum deviation of a single segment in the trajectory; π(a t |s t ) indicates that in state s t The person who took the action of not receiving the ticket a t The policy probability; α represents the weight coefficient of cumulative deviation; β represents the power coefficient of deviation; γ represents the weight coefficient of the maximum single-segment deviation; λ represents the reinforcement learning risk correction factor.

[0051] Preferably, the behavioral causal graph model includes:

[0052]

[0053] Among them, C anom (u) represents the causal anomaly score for non-ticket recipient u; M represents the number of behavioral outcome variables involved in the behavioral causal graph model; Y i This represents the i-th actually observed behavioral outcome; This represents the i-th expected value obtained from the causal model inference; denoted as the variance of the i-th behavioral outcome in the historical normal ticket holder group; N represents the number of dimensions for comparing causal effects; This represents the observed causal effect of non-ticket holders in the j-th dimension; ρ represents the reference causal effect value for normal ticket holders in the same dimension; L represents the number of detected permission conflict events; kξ represents the risk factor of the k-th permission conflict event; ξ represents the weight factor of the residual term; v represents the weight factor of the causal effect deviation term; and χ represents the weight factor of the permission conflict term.

[0054] Preferably, the risk judgment unit, based on the behavioral causal graph model, judges the behavioral state of the target non-ticket holder and generates a behavioral risk judgment result, including:

[0055] In the behavioral causal graph model, a causal inference-driven anomaly detection mechanism is introduced. Structural equation modeling and comparative intervention methods are used to compare the causal effects of the behavioral sequence of the target non-ticket holder with the baseline behavioral pattern of the normal ticket holder, determine the potential causal triggers of the abnormal behavior, and extract behavioral state feature parameters. Among them, the behavioral state feature parameters include: the duration of the anomaly, the sensitivity of the deviation area, and the frequency of permission conflicts.

[0056] Based on the aforementioned behavioral state characteristic parameters, a behavioral risk score is calculated for the target non-ticket personnel in a preset construction environment; wherein, the risk score is based on three dimensions: trajectory deviation, plan deviation, and permission violation.

[0057] Based on the correspondence between the behavioral risk score and the preset multi-level risk threshold range, the behavioral risk judgment result of the target non-ticket personnel is generated.

[0058] According to another aspect of the present invention, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the steps of a method for dynamic tracking of non-ticketed personnel at a construction site.

[0059] According to another aspect of the present invention, the present invention provides an electronic device, comprising:

[0060] The aforementioned computer-readable storage medium; and

[0061] One or more processors for executing a program in the computer-readable storage medium.

[0062] This invention provides a method and system for dynamic tracking of non-ticketed personnel at construction sites, comprising: collecting information data of personnel entering the construction site and comparing it with the construction ticket roster to screen suspected non-ticketed personnel; establishing a heterogeneous graph based on the information data of the suspected non-ticketed personnel and the information data in the ticketed personnel database, and aggregating locally similar nodes based on the heterogeneous graph to identify target non-ticketed personnel with abnormal identity characteristics; acquiring the movement trajectory of the target non-ticketed personnel, calculating the deviation degree based on the movement trajectory of the non-ticketed personnel, and obtaining the trajectory deviation index of the target non-ticketed personnel; constructing a behavioral causal graph model based on the trajectory deviation index, and judging the behavioral state of the target non-ticketed personnel based on the behavioral causal graph model to generate a behavioral risk judgment result; and generating a risk level label based on the behavioral risk judgment result. This invention can realize a fully intelligent closed loop from personnel identification and trajectory analysis to behavioral alarm. The method of this invention can not only significantly improve the automation and intelligence level of construction site safety management, but also effectively reduce the incidence of safety hazards, and has high engineering application value and promotion significance. Attached Figure Description

[0063] Exemplary embodiments of the present invention can be more fully understood by referring to the following figures:

[0064] Figure 1 A flowchart of a non-ticket-based personnel dynamic tracking method 100 for construction sites according to an embodiment of the present invention;

[0065] Figure 2 This is a flowchart illustrating the determination of trajectory deviation index according to an embodiment of the present invention;

[0066] Figure 3 A flowchart illustrating the behavioral risk assessment process according to an embodiment of the present invention;

[0067] Figure 4 This is a structural schematic diagram of a non-ticketed personnel dynamic tracking system 400 for construction sites according to an embodiment of the present invention. Detailed Implementation

[0068] Exemplary embodiments of the invention will now be described with reference to the accompanying drawings. However, the invention may be embodied in many different forms and is not limited to the embodiments described herein. These embodiments are provided to fully and completely disclose the invention and to fully convey its scope to those skilled in the art. The terminology used in the exemplary embodiments illustrated in the drawings is not intended to limit the invention. In the drawings, the same units / elements are referred to by the same reference numerals.

[0069] Unless otherwise stated, the terms used herein (including technical terms) have their common meaning as understood by one of ordinary skill in the art. Furthermore, it is understood that terms defined in commonly used dictionaries should be understood to have a meaning consistent with the context of their relevant field, and not to be interpreted as having an idealized or overly formal meaning.

[0070] Figure 1 This is a flowchart of a non-ticket-based personnel dynamic tracking method 100 for construction sites according to an embodiment of the present invention. Figure 1 As shown, the non-ticketed personnel dynamic tracking method for construction sites provided by this invention can realize a fully intelligent closed loop from personnel identification and trajectory analysis to behavior alarm. This method not only significantly improves the automation and intelligence level of construction site safety management but also effectively reduces the incidence of safety hazards, possessing high engineering application value and promotional significance. The non-ticketed personnel dynamic tracking method 100 provided by this invention begins at step 101. In step 101, information data of personnel entering the construction site is collected and compared with the construction ticket roster to screen suspected non-ticketed personnel.

[0071] Preferably, the nodes of the heterogeneous graph include personnel identity nodes, entry and exit record nodes, and scene location nodes, and the edges of the heterogeneous graph are used to represent the interactive relationship between personnel and their behavioral trajectories.

[0072] In this invention, video surveillance, access control card swipe records, and positioning sensors are used to collect information data on personnel entering the site, which is then compared with the roster on the construction permit to screen out suspected non-permitted personnel. The information data includes personnel image information, identification codes, access control entry and exit times, spatial location information, and behavioral image characteristics.

[0073] Specifically, multi-source sensing devices, including high-definition video surveillance cameras, access control card terminals, Bluetooth or UWB positioning tag base stations, and infrared thermal sensors, are deployed at the entrances and exits, main passages, work areas, and safety warning zones of the construction site. Data from all devices is aggregated and preprocessed through a unified edge computing node.

[0074] High-definition video surveillance equipment with night vision and dustproof capabilities is installed at fixed locations in entrances / exits, floor corridors, and core construction areas. Cameras capture video streams at a rate of 2-5 frames per second, and edge computing units perform face detection and key point localization, automatically generating facial feature codes and behavioral feature vectors (such as walking posture, carrying actions, and clothing identification features) for each person entering the site. To improve recognition accuracy in strong light and dusty environments, the system employs infrared-assisted light compensation and automatic exposure algorithms. Only encrypted feature vectors and timestamp information are retained in the image data to ensure privacy and security.

[0075] Each entrance and exit is equipped with a card reader and a facial recognition terminal. The access control system records personnel's "employee ID number," "entry / exit time," and "access point number" in real time. When the system detects someone entering without a valid card swipe or when the card swipe information does not match the facial features, an "abnormal entry event" is generated at the edge node, and the record is temporarily labeled as suspicious. All access control data is uploaded to the backend database via secure communication protocols (such as TLS encryption) to ensure the integrity and immutability of the data chain.

[0076] UWB positioning base stations are deployed in key areas of the construction site (tower crane operating area, material storage area, aerial work platform, etc.), and each certified worker wears a positioning tag with a unique ID. The system collects location coordinates and movement speed information in real time and uploads it at a frequency of 1Hz. For personnel who are not wearing tags but are captured by cameras on site, the system compares the visual detection results with the positioning gap area and marks them as "unauthorized positioning signal personnel," further improving the identification coverage of suspected non-certificate personnel.

[0077] Edge nodes synchronously fuse video features, access control card swipe records, and location data according to a unified timestamp, generating a multi-source information dataset. Each record includes the following fields: personnel image features, identification code (or missing status), access control entry / exit time, spatial location information, and dynamic behavior features. The backend system calls the construction ticket roster database, which contains fields such as "name, ID number, job type, construction unit, entry / exit registration information, and authorized work area." By matching the "identity code" and "facial feature code" fields, the following three types of abnormal samples are filtered out: personnel entering the site without access control records but detected by the image; personnel with card swipe identification codes but not registered in the roster; and personnel in the roster but located outside the authorized area. These samples are uniformly marked as a candidate set of suspected non-ticket personnel, along with information on the collection time, location of occurrence, and confidence level.

[0078] The system performs automatic initial screening based on the type and frequency of anomalies: if the same person is detected entering the site three times consecutively and identity verification fails, the system marks them as "highly suspected"; if an anomaly is detected only once, they are temporarily added to the observation list. All suspected personnel information is stored in the local database and provides input data for the next step of heterogeneous graph structure modeling.

[0079] In step 102, a heterogeneous graph is established based on the information data of the suspected non-ticket holders and the information data in the ticket holder database. Local similar nodes are aggregated based on the heterogeneous graph to identify target non-ticket holders with abnormal identity characteristics.

[0080] Preferably, the aggregation of locally similar nodes based on the heterogeneous graph to identify target non-ticket holders with abnormal identity characteristics includes:

[0081] The structure of the heterogeneous graph is optimized based on graph regularization constraint mechanism and sparse relation constraint mechanism;

[0082] A node aggregation mechanism based on local subgraphs is used to cluster and align the ticket holder nodes in the optimized heterogeneous graph to generate reference group features;

[0083] The node features of the suspected non-ticket holders are compared with the features of the reference group to identify target non-ticket holders whose identity features are abnormal.

[0084] In this invention, a graph structure relationship is established between the suspected non-ticket holders and the ticket holder database. A graph regularized sparse representation algorithm is used to aggregate locally similar nodes to identify target non-ticket holders with abnormal identity characteristics.

[0085] Specifically, based on the suspected non-ticket personnel, a heterogeneous graph structure is constructed with the construction ticket personnel database. The nodes of the heterogeneous graph structure include personnel identity nodes, entry and exit record nodes, and scene location nodes. The edges of the heterogeneous graph are used to represent the interactive relationship between personnel and their behavioral trajectories.

[0086] Specifically, the system reads two types of data sources from the database:

[0087] Data set of suspected non-ticket holders (including facial feature codes, access control record status, location trajectory segments, and time periods of occurrence);

[0088] Construction ticket personnel database (including personnel identity information, job type, work team, authorized area and historical entry and exit records).

[0089] Based on this data, a heterogeneous graph structure containing multiple types of nodes and edges is constructed.

[0090] Node types include the following:

[0091] Personnel identification node: Corresponds to each ticket or suspected person. The node attributes include identification code, facial feature vector and unit information;

[0092] Entry / exit record node: corresponds to each entry / exit record in the access control system, with attributes including time, location, and access status (valid / invalid);

[0093] Scene location node: corresponds to the specific spatial area of ​​the construction site (such as the steel bar processing area, tower crane base area, edge protection area, etc.).

[0094] Edge relationship types include the following:

[0095] Personnel-Access Control Relationship: Represents the behavioral mapping between personnel and entry / exit records;

[0096] Person-Scene Relationship: Represents the degree of matching between the personnel activity area and the authorized area;

[0097] Access control - scene relationship: reflects the geographical connection between the entrance and the corresponding construction zone.

[0098] In this invention, graph regularization constraints are applied to the heterogeneous graph structure. A feature similarity matrix is ​​used to limit the smoothness of the local structure, and a sparse representation operator is introduced to suppress redundant associations. Specifically, to prevent redundant node connections caused by data acquisition noise, false alarms, and other factors, the system introduces a graph regularization constraint mechanism into the heterogeneous graph structure. Specifically, the system constructs a similarity matrix based on the feature similarity between nodes (such as cosine similarity of facial features, job similarity, activity time overlap rate, etc.) to guide the smoothing of the local structure in the graph, making the connections between feature-similar nodes tighter, while gradually weakening the connections between irrelevant nodes. Subsequently, a sparse relation constraint mechanism is introduced, limiting each node to retain only a few of the most representative connections (such as the top 5 neighboring nodes with the highest similarity), thereby removing data noise and false associations. This process ensures that the heterogeneous graph, while maintaining the integrity of its local structure, can highlight the relational structure that truly reflects behavioral consistency and identity feature associations.

[0099] In this invention, within the heterogeneous graph structure, a node aggregation mechanism based on local subgraphs is employed to cluster and align ticket personnel nodes with highly similar characteristics, generating reference group features. Specifically, the system extracts local subgraphs centered on the "ticket personnel node" within the graph structure. Each subgraph includes the ticket personnel node and its directly associated entry / exit record nodes and scene location nodes. By traversing and aggregating these local subgraphs, several "ticket personnel reference groups" can be formed. These groups represent legitimate individual groups with stable and consistent behavioral patterns and identity characteristics at the construction site.

[0100] The node aggregation mechanism operates according to the following principles during execution:

[0101] If multiple ticket holder nodes exhibit high consistency in attributes such as time, location, and access channels, they are merged into a single reference group node.

[0102] For groups that frequently appear in the same area, have similar behavioral trajectories, and overlap in working hours, the system further encodes their behavioral characteristics in a unified manner to form a comprehensive characteristic representation of the group;

[0103] All reference group nodes eventually form a "standard identity feature cluster," which is used as a reference model in subsequent comparison stages.

[0104] In this invention, the node features of suspected non-ticket holders are compared with the features of the reference group to identify non-ticket holders with abnormal identity features, and the corresponding anomaly identification results are output. Specifically, the system compares the node features (including face code, entry / exit record status, location information, behavioral patterns, etc.) of each suspected non-ticket holder with the generated reference group features one by one. If the facial features of a suspected person are more than a preset threshold away from the average feature distance of any reference group, or if their entry / exit behavior deviates significantly from the time distribution of the same type of work group, the system determines that the person has an abnormal identity feature; if the person's trajectory appears multiple times in unauthorized areas, and their access control record is missing or they are using someone else's identity code, the anomaly level is further increased. The identification results include: anomaly level (high, medium, low), corresponding reference group number, and main anomaly cause (identity feature mismatch / abnormal entry / exit pattern / spatial behavior deviation, etc.). All identification results will be used as input data for the subsequent trajectory deviation analysis process.

[0105] In step 103, the movement trajectory of the target non-ticket person is obtained, and the deviation is calculated based on the movement trajectory of the non-ticket person to obtain the trajectory deviation index of the target non-ticket person.

[0106] Preferably, the deviation is calculated based on the movement trajectory of the non-ticket holder to obtain the trajectory deviation index of the target non-ticket holder, including:

[0107] Based on the movement trajectory, the spatial location information of the target non-ticket personnel at different times is mapped to the construction site topology model; wherein, the construction site topology model consists of multiple nodes and edges, the nodes correspond to key location units of the construction area, including: entrance and exit nodes, work area nodes and hazard source nodes, and the edges represent the path connection relationship that personnel can walk on;

[0108] In the construction site topology model, the path potential function is modeled for the trajectory of non-ticketed personnel and the authorized path in the construction area based on the graph potential field path constraint algorithm;

[0109] Based on the path potential function, combined with the edge weight constraints between nodes and the spatial topological distance, the trajectory of non-ticket personnel is decomposed and compared segment by segment to generate the corresponding deviation index sequence.

[0110] Extract the cumulative deviation and the maximum single-segment deviation from the deviation index sequence as two feature parameters to quantify the degree of trajectory abnormality of the target non-ticket personnel, and output the trajectory deviation index.

[0111] Preferably, the graph potential field path constraint algorithm includes:

[0112]

[0113] Among them, e risk (p) represents the comprehensive risk potential value of non-ticket holders under trajectory p; p t q represents the spatial location node of a person not represented by a ticket at time t; t This represents the standard location node of the authorized path in the construction area at time t; d(p t ,q t ) indicates the actual location p of the person not represented on the ticket. t With standard node q t Spatial topological distance between them; w(p) t ,q t ) represents the connection node p in the topology graph. t With q t The path edge weights; T represents the total temporal length of the trajectory; This represents the maximum deviation of a single segment in the trajectory; π(a t |s t ) indicates that in state s t The person who took the action of not receiving the ticket a t The policy probability; α represents the weight coefficient of cumulative deviation; β represents the power coefficient of deviation; γ represents the weight coefficient of the maximum single-segment deviation; λ represents the reinforcement learning risk correction factor.

[0114] In this invention, the movement trajectory of the target non-ticket holder is mapped to a construction site topology model. A graph potential field path constraint algorithm is used to calculate the deviation between their trajectory and the authorized path in the construction area, and the trajectory deviation index of the target non-ticket holder is output. The specific process is as follows: Figure 2 As shown, it includes:

[0115] Step 1: Map the spatial location information of the non-ticket personnel at different times to the construction site topology model. The construction site topology model consists of multiple nodes and edges. The nodes correspond to key location units in the construction area, including entrance and exit nodes, work area nodes, and hazard source nodes. The edges represent the path connections that personnel can walk on.

[0116] Specifically, the spatial location information of the non-ticket holder at different times is extracted. Each record includes a timestamp, planar coordinates (x, y), and height layer information (applicable to multi-story building scenarios).

[0117] The map data of the construction site was pre-processed into a topological model. This model consists of a series of nodes and edges:

[0118] Nodes: Represent key spatial units, such as construction entrances and exits, work areas, storage areas, edge protection areas, high-risk work areas, and tower crane operating radius boundary points, etc.

[0119] Edges: Represent the path relationships that people can travel on, including passageways, stairs, platforms, and construction walkways. Each edge includes attributes such as the direction of travel, path length, slope, and accessibility.

[0120] The system sequentially maps the trajectory points of non-ticket holders to the corresponding nodes or edges in the topology graph. When the location data falls within the boundary area of ​​multiple nodes, the system determines the node to which it belongs based on the shortest spatial distance and temporal continuity rules to ensure that the trajectory mapping is continuous and physically reasonable.

[0121] Step 2: In the construction site topology model, the path potential function is modeled based on the graph potential field path constraint algorithm to model the trajectory of non-ticketed personnel and the authorized path in the construction area. The path potential function is used to measure the relative deviation between the trajectory of non-ticketed personnel and the standard authorized path.

[0122] Specifically, the digital topology graph model of the construction site has established a basic spatial framework in step 101. In this stage, semantic information and path potential field distribution are further superimposed on this basis to construct a graph potential field path constraint model with behavioral constraints and spatial safety weights.

[0123] The system first transforms the construction site map into a semantic topology layer, as follows:

[0124] Node Layer: Includes entrance / exit nodes, work area nodes, hazard source nodes, management area nodes, and material storage nodes. Each node has additional attributes including job type, risk level, access priority, and safety control level.

[0125] Edge Layer: Composed of path elements such as passageways, stairs, ventilation shafts, and temporary scaffolding. Edge attributes include distance, slope, passage width, passage direction (one-way / two-way), and risk weight (mapped from the site safety level).

[0126] Hierarchical relationship: For multi-story construction structures, the topology model is extended in the form of "floor nodes + vertical connection edges", and a "vertical impedance" parameter is added to the vertical channel edge to reflect the difficulty of movement between different floors.

[0127] Each node and edge is assigned a different "potential field weight" based on its security attributes and authorization permissions. For example, the potential value is lowest for general passage areas (such as main passages); work area nodes are assigned different values ​​depending on the type of work they belong to, such as welding areas and hoisting areas with medium potential values; and hazardous source nodes (such as high-voltage distribution cabinets and tower crane slewing areas) have the highest potential value to represent the path risk to enter that area. The system generates a potential field distribution map in the model, which visually resembles a heat map, with color depth reflecting the level of risk in each area.

[0128] Import the "work authorization path sets" for each trade from the construction management platform, including trade type, work time, passable node range, and path direction. The system marks these paths on the topology model and highlights them in green or blue as "legal path sets." Simultaneously, for high-risk areas, closed areas, and areas around hazardous equipment (such as concrete pumps and lifting platforms), the system sets "restricted areas" and shields them with a high potential field value, ensuring that any trajectory entering these areas triggers a deviation signal.

[0129] Considering dynamic changes at the construction site (such as temporary road closures, hoisting operations, etc.), the system has a dynamic update mechanism: when safety management personnel mark new closed areas or adjust access on the site management terminal, the topology map automatically updates the potential field weights; the edge computing nodes recalculate the path potential distribution and synchronize it to the server to achieve real-time scene adaptation.

[0130] Step 3: Using a deviation calculation method based on path potential function, combined with the edge weight constraints between nodes and spatial topological distance, the trajectory of non-ticket personnel is decomposed and compared segment by segment to generate the corresponding deviation index sequence.

[0131] Specifically, after obtaining complete trajectory data of non-ticket holders, the system needs to spatially align and analyze the differences between their real-time travel paths and authorized paths. This process consists of four stages: trajectory preprocessing, segmented modeling, path comparison, and deviation calculation.

[0132] Location data for non-ticket holders may originate from different devices (such as UWB tags, video recognition results, and access control card swipe location inference), resulting in temporal errors and spatial drift. The system first performs the following preprocessing:

[0133] Time synchronization: The timestamps of data from different sources are aligned according to the station's clock, with the error controlled within 1 second;

[0134] Spatial correction: Use multi-source positioning weighting mechanism (such as camera coordinates and UWB base station signal fusion) to correct positioning drift and ensure that the trajectory points fall within the topology map coordinate system;

[0135] Trajectory smoothing: A sliding window mechanism is used to eliminate short-term positioning jitter, making the trajectory segments physically continuous and reasonable.

[0136] The system segments the trajectory according to time or spatial length. For example, a sub-trajectory segment can be formed every 30 seconds or every 20 meters. Each trajectory segment records the sequence of topological nodes and path edge information traversed by the person within that time window. For cases involving cross-floor or overlapping paths, the system uses "critical node changes" as the dividing point; that is, when a person enters a new functional area or traverses a new floor, the trajectory segment is automatically split. This maintains the semantic integrity of the segments while facilitating subsequent path comparison.

[0137] For each segment of a non-ticketed person's trajectory, the system performs a comparison process with the standard authorized path, including:

[0138] Spatial path comparison: Calculate the correspondence between each node in the trajectory segment and the nearest node in the authorized path. If three or more consecutive nodes fail to match a corresponding node in the authorized path, it is determined to be a spatial out-of-bounds error.

[0139] Path relationship analysis: Compare the travel direction, frequency of travel, and connection sequence of the trajectory edge and the authorized edge. Once "reverse travel", "crossing a non-connected area" or "entering a forbidden edge" is found, it is recorded as an abnormal path segment.

[0140] Potential field energy difference assessment: The system reads the average potential field value of the area covered by the trajectory segment and performs a difference analysis with the average potential field value of the same location segment of the standard authorized path. If the difference exceeds a set threshold (e.g., more than twice the average value of normal work), it indicates that the person may have entered a high-risk or unauthorized area.

[0141] The system generates structured deviation feature data for each trajectory segment, including: the start and end node numbers; the corresponding authorized path number; the average potential field difference; the proportion of matching nodes; and whether it contains dangerous nodes or forbidden edges.

[0142] Meanwhile, the backend system will highlight the deviation from the path on the construction plan with a red curve and automatically generate a deviation statistics chart, showing the trend curve of the trajectory deviation rate over time, for safety management personnel to view intuitively.

[0143] Step 4: Extract two feature parameters, cumulative deviation and maximum single-segment deviation, from the deviation index sequence, and use them as the quantitative basis for the degree of trajectory abnormality of non-ticket personnel, and output the trajectory deviation index.

[0144] The formula for the graph potential path constraint algorithm is as follows:

[0145]

[0146] Among them, E risk (p) represents the comprehensive risk potential value of non-ticket holders under trajectory p; p t q represents the spatial location node of a person not represented by a ticket at time t; t This represents the standard location node of the authorized path in the construction area at time t; d(p t ,q t ) indicates the actual location p of the person not represented on the ticket. t With standard node q t Spatial topological distance between them; w(p) t ,q t ) represents the connection node p in the topology graph. t With qt The path edge weights; T

[0147] The weighting coefficient for deviation; β represents the power coefficient of deviation; γ represents the weighting coefficient for the maximum single-segment deviation; λ represents the reinforcement learning risk correction factor.

[0148] In step 104, a behavioral causal graph model is constructed based on the trajectory deviation index, and the behavioral causal graph model is used to determine the behavioral state of the target non-ticket personnel, generating a behavioral risk judgment result.

[0149] Preferably, the behavioral causal graph model includes:

[0150]

[0151] Among them, C anom (u) represents the causal anomaly score for non-ticket recipient u; M represents the number of behavioral outcome variables involved in the behavioral causal graph model; Y i This represents the i-th actually observed behavioral outcome; This represents the i-th expected value obtained from the causal model inference; denoted as the variance of the i-th behavioral outcome in the historical normal ticket holder group; N represents the number of dimensions for comparing causal effects; This represents the observed causal effect of non-ticket holders in the j-th dimension; ρ represents the reference causal effect value for normal ticket holders in the same dimension; L represents the number of detected permission conflict events; k Let ξ represent the risk factor of the k-th permission conflict event; ξ represent the weight factor of the residual term; ν represent the weight factor of the causal effect deviation term; and χ represent the weight factor of the permission conflict term.

[0152] Preferably, the behavioral risk assessment result is generated by determining the behavioral state of the target non-ticket holder based on the behavioral causal graph model, including:

[0153] In the behavioral causal graph model, a causal inference-driven anomaly detection mechanism is introduced. Structural equation modeling and comparative intervention methods are used to compare the causal effects of the behavioral sequence of the target non-ticket holder with the baseline behavioral pattern of the normal ticket holder, determine the potential causal triggers of the abnormal behavior, and extract behavioral state feature parameters. Among them, the behavioral state feature parameters include: the duration of the anomaly, the sensitivity of the deviation area, and the frequency of permission conflicts.

[0154] Based on the aforementioned behavioral state characteristic parameters, a behavioral risk score is calculated for the target non-ticket personnel in a preset construction environment; wherein, the risk score is based on three dimensions: trajectory deviation, plan deviation, and permission violation.

[0155] Based on the correspondence between the behavioral risk score and the preset multi-level risk threshold range, the behavioral risk judgment result of the target non-ticket personnel is generated.

[0156] In this invention, a causal inference-driven anomaly detection model is introduced into the trajectory deviation index. This model combines construction plan information and regional access rules to determine the behavioral status of non-ticket holders and generate behavioral risk assessment results. Specifically, the process is as follows: Figure 3 As shown, it includes:

[0157] Step 1: Construct a behavioral causal graph model based on the trajectory deviation index. The nodes of the behavioral causal graph model include construction plan nodes, regional permission nodes, and personnel behavior nodes. The edges of the heterogeneous graph are used to represent the causal dependencies between different behavioral events.

[0158] Specifically, based on the trajectory deviation index data, and combined with the construction site's Management Information System (MIS) and safety control platform, a behavioral cause-effect graph model is established. This model has a core structure of "construction activity events—personnel behavior—area constraints," and consists of the following three types of nodes:

[0159] 1. Construction plan milestones: These represent the planned work content, work areas, and participating trades at the construction site at different time periods.

[0160] Node attributes include: job type (such as concrete pouring, electric welding, tower crane hoisting), job time period, construction team, and on-site instruction number;

[0161] Data source: Construction plan table in the construction project management system.

[0162] 2. Area permission node: Represents the spatial safety constraints and access rules of each work area.

[0163] Node attributes include: area risk level, permitted occupations, permitted time periods, and maximum number of entries and exits;

[0164] Data source: Security management platform and access control database.

[0165] 3. Personnel behavior nodes: Characterize the trajectory behavior characteristics of non-ticket personnel within a specific time period.

[0166] Node attributes include: trajectory deviation, dwell time, access area type, entry and exit paths, and behavior category (such as following, loitering, and crossing boundaries).

[0167] When constructing the causal graph, the system defines the connections between nodes as directed edges to represent the causal dependencies of "changes in construction plans → changes in area permissions → changes in personnel behavior patterns". For example, when an area is temporarily closed or the work plan is changed, this causal chain will directly affect the passage behavior and deviation patterns of non-ticketed personnel.

[0168] Step 2: In the behavioral causal graph model, an anomaly detection mechanism driven by causal inference is introduced. Structural equation modeling and comparative intervention methods are used to compare the causal effects of the behavioral sequences of non-ticket holders with the baseline behavioral patterns of normal ticket holders, determine the potential causal triggers of abnormal behavior, and extract behavioral state feature parameters, including the duration of abnormality, sensitivity to deviation areas, and frequency of permission conflicts.

[0169] Specifically, based on the behavioral causal graph model, the system introduces a causal inference-driven anomaly detection mechanism to identify the trigger sources of abnormal behavior by non-ticket holders. This mechanism mainly includes the following steps:

[0170] 1. Structural Modeling Phase: The system uses structural equation modeling to constrain and map the variable relationships in the behavioral cause-effect graph. For example, the degree of dependence between personnel deviation and construction plan nodes, and the influence strength between changes in regional permissions and behavior frequency, are all defined as directed dependency paths in the structural model.

[0171] 2. Comparative intervention analysis phase: The system selects the behavioral sequences of normal ticket swiping personnel in the same working period and similar locations as the baseline sample, and conducts comparative intervention analysis with the behavioral sequences of non-ticket swiping personnel.

[0172] If individuals not represented by tickets exhibit significant deviations from the planned route without any changes to the schedule, it can be inferred that they engaged in "illegal entry" or "loitering and observing."

[0173] If you enter a high-risk area during a temporary adjustment of regional authority, it may be considered "delayed response to plan change" or "mistaken entry".

[0174] If there is a frequent occurrence of entering and exiting the same dangerous area, it is considered to have a "high-risk tendency to violate regulations".

[0175] 3. Causal Effect Extraction and Feature Generation: The system compares behavioral differences before and after intervention, extracting feature parameters representing abnormal behavior, including:

[0176] Duration of abnormality: The total time that non-ticket holders stayed in the abnormal area;

[0177] Area deviation sensitivity: The correlation between trajectory deviation and regional risk level reflects its tendency to approach high-risk areas;

[0178] Permission conflict frequency: The number of times permission violations or unauthorized access are triggered within a unit of time.

[0179] Through the above analysis, the system can clearly identify the causal triggering path of abnormal behavior, that is, determine whether the abnormality is caused by plan deviation, access control vulnerability or deliberate violation by personnel.

[0180] Step 3: Based on the behavioral state characteristic parameters, calculate the behavioral risk score of non-ticket personnel in a given construction environment. The risk score is based on three dimensions: trajectory deviation, plan deviation, and permission violation.

[0181] Step 4: Based on the correspondence between the risk score and the preset multi-level risk threshold range, generate the behavioral risk judgment result for non-ticket personnel.

[0182] Specifically, after extracting behavioral state features, the system fuses them with trajectory deviation data to generate a behavioral risk score for non-ticket holders. The risk score is calculated considering the following three main dimensions:

[0183] Trajectory deviation dimension: reflects the intensity and persistence of the deviation between the person's travel path and the authorized path, and reflects the degree of abnormality in their physical behavior;

[0184] Deviation from Plan Dimension: Measures the degree of inconsistency between the individual's actual activities and the planned construction schedule and area allocation;

[0185] Access violation level: This reflects the severity of the individual's violation of area security rules and access permissions, such as entering restricted areas or crossing boundaries at night.

[0186] The system uses a weighted fusion mechanism to comprehensively score the three dimensions. For individuals who consistently deviate from or repeatedly violate regulations in high-risk areas, the system will automatically increase their risk score weight; for occasional or accidental entries, the system will reduce the risk score through causal correlation analysis to avoid false alarms.

[0187] Ultimately, the behavioral risk score is a quantitative result ranging from 0 to 100, along with an explanation of the source of the risk, such as "high risk of overstepping boundaries", "medium risk of authority conflict", and "low risk of unplanned entry".

[0188] Furthermore, the formula for the behavioral causal graph model is as follows:

[0189]

[0190] Among them, C anom (u) represents the causal anomaly score for non-ticket recipient u; M represents the number of behavioral outcome variables involved in the behavioral causal graph model; Y i This represents the i-th actually observed behavioral outcome; This represents the i-th expected value obtained from the causal model inference; denoted as the variance of the i-th behavioral outcome in the historical normal ticket holder group; N represents the number of dimensions for comparing causal effects; This represents the observed causal effect of non-ticket holders in the j-th dimension; ρ represents the reference causal effect value for normal ticket holders in the same dimension; L represents the number of detected permission conflict events; k Let ξ represent the risk factor of the k-th permission conflict event; ξ represent the weight factor of the residual term; ν represent the weight factor of the causal effect deviation term; and χ represent the weight factor of the permission conflict term.

[0191] In step 105, a risk level label is generated based on the behavioral risk discrimination result.

[0192] In this invention, a corresponding risk level label is generated based on the behavioral risk assessment result, the alarm threshold and triggering strategy are adaptively adjusted based on a hierarchical alarm mechanism, and the alarm information is pushed to the construction safety management terminal in real time. Specifically, it includes:

[0193] Step 1: Based on the behavioral risk assessment results, construct a multi-level risk label set, which is divided into safety level, suspicious level, serious level and emergency level according to the risk level from low to high.

[0194] Specifically, after extracting behavioral state features, the system fuses them with trajectory deviation data to generate a behavioral risk score for non-ticket holders. The risk score is calculated considering the following three main dimensions:

[0195] Trajectory deviation dimension: reflects the intensity and persistence of the deviation between the person's travel path and the authorized path, and reflects the degree of abnormality in their physical behavior;

[0196] Deviation from Plan Dimension: Measures the degree of inconsistency between the individual's actual activities and the planned construction schedule and area allocation;

[0197] Access violation level: This reflects the severity of the individual's violation of area security rules and access permissions, such as entering restricted areas or crossing boundaries at night.

[0198] The system uses a weighted fusion mechanism to comprehensively score the three dimensions. For individuals who consistently deviate from or repeatedly violate regulations in high-risk areas, the system will automatically increase their risk score weight; for occasional or accidental entries, the system will reduce the risk score through causal correlation analysis to avoid false alarms.

[0199] Ultimately, the behavioral risk score is a quantitative result ranging from 0 to 100, along with an explanation of the source of the risk, such as "high risk of overstepping boundaries", "medium risk of authority conflict", and "low risk of unplanned entry".

[0200] Step 2: Driven by the multi-level risk label set, a hierarchical alarm mechanism optimized by reinforcement learning is adopted. The hierarchical alarm mechanism adaptively and dynamically adjusts the alarm threshold based on historical alarm feedback samples and real-time risk evolution characteristics, and adopts a threshold correction strategy based on a reward function. The reward function is used to measure the balance between the safety intervention effect and the false alarm rate under different alarm levels.

[0201] Specifically, based on the generation of risk labels, the system initiates a reinforcement learning-driven hierarchical alarm mechanism to achieve adaptive updates of dynamic alarm thresholds and triggering strategies.

[0202] The system extracts sample data from a historical risk event database, including alarm levels, on-site handling results, false alarm and missed alarm records, and processing time. During the model initialization phase, the system uses these samples to calculate the initial alarm thresholds corresponding to each risk level. For example, the trigger probability for the emergency level is set to above 0.9, and for the suspicious level, it is set between 0.3 and 0.6.

[0203] When a new non-ticket-related risk event is detected, the system immediately extracts real-time features such as risk score, regional potential value change, duration of anomaly, and event density, compares them with historical samples, and inputs them into the reinforcement learning module for online updates.

[0204] The reinforcement learning module dynamically adjusts the thresholds for each risk level based on actual on-site alarm feedback (such as false alarm rate, response delay, and effectiveness of safety intervention). When the system detects an increase in the false alarm rate, it automatically raises the alarm threshold to reduce unnecessary intervention; when actual safety incidents occur frequently but alarms are not triggered in a timely manner, it automatically lowers the threshold to improve sensitivity. This process forms a continuously converging closed-loop learning mechanism, enabling the alarm system to adapt to changes in different construction stages, environmental noise, and pedestrian density.

[0205] The system designs a reward function in the reinforcement learning model to quantify the overall benefits brought by different threshold adjustments: when the alarm response is timely and the on-site intervention is effective, the reward value increases; when the false alarm rate is high or the processing delay is long, the reward value decreases.

[0206] Step 3: According to the hierarchical alarm mechanism, update the triggering strategy corresponding to the risk level label, generate the optimal alarm decision sequence, and push the alarm information to the construction safety management terminal in real time through the wireless communication link. The alarm information includes personnel identity information, risk level label and trajectory deviation index.

[0207] This invention achieves high-precision identification and dynamic risk warning for non-ticketed personnel at construction sites by integrating multi-source heterogeneous data with graph intelligence algorithms. First, by cross-comparing multimodal data from video surveillance, access control card swipes, and positioning sensors, the method effectively compensates for the deficiencies of single data sources in identity recognition, improving the comprehensiveness and accuracy of screening suspected non-ticketed personnel. Second, by constructing a graph structure relationship between non-ticketed and ticketed personnel and introducing a graph regularized sparse representation algorithm, the method can aggregate abnormal features while maintaining the consistency of local node features, thereby achieving adaptive identification of identity anomalies and enhancing the model's robustness to complex scenarios such as concealed entry and impersonation. A graph potential field path constraint algorithm is used to map the trajectories of non-ticketed personnel into the topological space of the construction site. By calculating the deviation between the trajectory and the authorized path, quantitative analysis of abnormal activity paths is achieved, overcoming the problem that traditional Euclidean distance or regional judgment models cannot reflect spatial constraint relationships. Combined with a causal inference-driven anomaly detection model, the system can identify behavior types with potential security threats based on construction plans and regional permission rules, improving the interpretability and traceability of abnormal behavior judgment. Through dynamic adaptive adjustment of risk level labels and graded alarm mechanisms, the system can update alarm thresholds and strategies in real time according to the on-site risk status, realizing a fully intelligent closed loop from personnel identification and trajectory analysis to behavior alarms.

[0208] Figure 4 This is a structural schematic diagram of a non-ticket-based personnel dynamic tracking system 400 for construction sites according to an embodiment of the present invention. Figure 4 As shown, the non-ticketed personnel dynamic tracking system 400 for construction sites provided by the embodiments of the present invention includes: a screening unit 401, an aggregation identification unit 402, a deviation calculation unit 403, a risk judgment unit 404, and an alarm unit 405.

[0209] Preferably, the screening unit 401 is used to collect information data of personnel entering the construction site and compare it with the construction ticket roster to screen out suspected non-ticket personnel.

[0210] Preferably, the aggregation and identification unit 402 is used to establish a heterogeneous graph based on the information data of the suspected non-ticket personnel and the information data in the ticket personnel database, and to aggregate local similar nodes based on the heterogeneous graph in order to identify target non-ticket personnel with abnormal identity characteristics.

[0211] Preferably, the nodes of the heterogeneous graph include personnel identity nodes, entry and exit record nodes, and scene location nodes, and the edges of the heterogeneous graph are used to represent the interactive relationship between personnel and their behavioral trajectories.

[0212] Preferably, the aggregation and identification unit 402 aggregates locally similar nodes based on the heterogeneous graph to identify target non-ticket personnel with abnormal identity characteristics, including:

[0213] The structure of the heterogeneous graph is optimized based on graph regularization constraint mechanism and sparse relation constraint mechanism;

[0214] A node aggregation mechanism based on local subgraphs is used to cluster and align the ticket holder nodes in the optimized heterogeneous graph to generate reference group features;

[0215] The node features of the suspected non-ticket holders are compared with the features of the reference group to identify target non-ticket holders whose identity features are abnormal.

[0216] Preferably, the deviation calculation unit 403 is used to obtain the movement trajectory of the target non-ticket person, calculate the deviation based on the movement trajectory of the non-ticket person, and obtain the trajectory deviation index of the target non-ticket person.

[0217] Preferably, the deviation calculation unit 403 calculates the deviation based on the movement trajectory of the non-ticket holder to obtain the trajectory deviation index of the target non-ticket holder, including:

[0218] Based on the movement trajectory, the spatial location information of the target non-ticket personnel at different times is mapped to the construction site topology model; wherein, the construction site topology model consists of multiple nodes and edges, the nodes correspond to key location units of the construction area, including: entrance and exit nodes, work area nodes and hazard source nodes, and the edges represent the path connection relationship that personnel can walk on;

[0219] In the construction site topology model, the path potential function is modeled for the trajectory of non-ticketed personnel and the authorized path in the construction area based on the graph potential field path constraint algorithm;

[0220] Based on the path potential function, combined with the edge weight constraints between nodes and the spatial topological distance, the trajectory of non-ticket personnel is decomposed and compared segment by segment to generate the corresponding deviation index sequence.

[0221] Extract the cumulative deviation and the maximum single-segment deviation from the deviation index sequence as two feature parameters to quantify the degree of trajectory abnormality of the target non-ticket personnel, and output the trajectory deviation index.

[0222] Preferably, the graph potential field path constraint algorithm includes:

[0223]

[0224] Among them, e risk(p) represents the comprehensive risk potential value of non-ticket holders under trajectory p; p t q represents the spatial location node of a person not represented by a ticket at time t; t This represents the standard location node of the authorized path in the construction area at time t; d(p t ,q t ) indicates the actual location p of the person not represented on the ticket. t With standard node q t Spatial topological distance between them; w(p) t ,q t ) represents the connection node p in the topology graph. t With q t The path edge weights; T represents the total temporal length of the trajectory; This represents the maximum deviation of a single segment in the trajectory; π(a t |s t ) indicates that in state s t The person who took the action of not receiving the ticket a t The policy probability; α represents the weight coefficient of cumulative deviation; β represents the power coefficient of deviation; γ represents the weight coefficient of the maximum single-segment deviation; λ represents the reinforcement learning risk correction factor.

[0225] Preferably, the risk discrimination unit 404 is used to construct a behavioral causal graph model based on the trajectory deviation index, and to judge the behavioral state of the target non-ticket person based on the behavioral causal graph model, thereby generating a behavioral risk discrimination result;

[0226] Preferably, the behavioral causal graph model includes:

[0227]

[0228] Among them, C anom (u) represents the causal anomaly score for non-ticket recipient u; M represents the number of behavioral outcome variables involved in the behavioral causal graph model; Y i This represents the i-th actually observed behavioral outcome; This represents the i-th expected value obtained from the causal model inference; denoted as the variance of the i-th behavioral outcome in the historical normal ticket holder group; N represents the number of dimensions for comparing causal effects; This represents the observed causal effect of non-ticket holders in the j-th dimension; ρ represents the reference causal effect value for normal ticket holders in the same dimension; L represents the number of detected permission conflict events; k Let ξ represent the risk factor of the k-th permission conflict event; ξ represent the weight factor of the residual term; ν represent the weight factor of the causal effect deviation term; and χ represent the weight factor of the permission conflict term.

[0229] Preferably, the risk discrimination unit 404, based on the behavioral causal graph model, determines the behavioral state of the target non-ticket holder and generates a behavioral risk discrimination result, including:

[0230] In the behavioral causal graph model, a causal inference-driven anomaly detection mechanism is introduced. Structural equation modeling and comparative intervention methods are used to compare the causal effects of the behavioral sequence of the target non-ticket holder with the baseline behavioral pattern of the normal ticket holder, determine the potential causal triggers of the abnormal behavior, and extract behavioral state feature parameters. Among them, the behavioral state feature parameters include: the duration of the anomaly, the sensitivity of the deviation area, and the frequency of permission conflicts.

[0231] Based on the aforementioned behavioral state characteristic parameters, a behavioral risk score is calculated for the target non-ticket personnel in a preset construction environment; wherein, the risk score is based on three dimensions: trajectory deviation, plan deviation, and permission violation.

[0232] Based on the correspondence between the behavioral risk score and the preset multi-level risk threshold range, the behavioral risk judgment result of the target non-ticket personnel is generated.

[0233] Preferably, the alarm unit 405 is used to generate a risk level label based on the behavior risk judgment result.

[0234] The non-ticketed personnel dynamic tracking system 400 for construction sites according to an embodiment of the present invention corresponds to the non-ticketed personnel dynamic tracking method 100 for construction sites according to another embodiment of the present invention, and will not be described again here.

[0235] According to another aspect of the present invention, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the steps of a method for dynamic tracking of non-ticketed personnel at a construction site.

[0236] According to another aspect of the present invention, the present invention provides an electronic device, comprising:

[0237] The aforementioned computer-readable storage medium; and

[0238] One or more processors for executing a program in the computer-readable storage medium.

[0239] The present invention has been described with reference to a few embodiments. However, it will be apparent to those skilled in the art that other embodiments besides those disclosed above fall equivalently within the scope of the present invention.

[0240] Generally, all terms used in this invention are interpreted according to their ordinary meaning in the art, unless otherwise expressly defined herein. All references to “a / the / the [device, component, etc.]” ​​are openly interpreted as at least one instance of said device, component, etc., unless otherwise expressly stated. The steps of any method disclosed herein need not be performed in the exact order disclosed unless explicitly stated otherwise.

[0241] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0242] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0243] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0244] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0245] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the present invention.

Claims

1. A construction site-oriented non-ticketed person dynamic tracking method, characterized by, The method comprises: Collecting information data of personnel entering the construction site and comparing with the construction ticket roster to screen suspected non-ticket personnel; Based on the information data of the suspected non-ticket personnel and the information data in the ticket personnel database, a heterogeneous graph is established, and local similar nodes are aggregated based on the heterogeneous graph to identify target non-ticket personnel with abnormal identity characteristics; Obtain the motion trajectory of the target non-ticket personnel, calculate the deviation degree based on the motion trajectory of the non-ticket personnel, and obtain the trajectory deviation degree index of the target non-ticket personnel; Based on the trajectory deviation degree index, a behavior causal diagram model is constructed, and the behavior state of the target non-ticket personnel is judged based on the behavior causal diagram model to generate a behavior risk judgment result; Based on the behavior risk judgment result, a risk level label is generated.

2. A method for tracking non-ticketed personnel on a construction site according to claim 1, wherein, The nodes of the heterogeneous graph include personnel identity nodes, entry and exit record nodes, and scene location nodes, and the edges of the heterogeneous graph are used to represent the interaction relationship between personnel and their behavior trajectory.

3. A method for tracking non-ticketed personnel in a construction site according to claim 1, wherein, Based on the heterogeneous graph, local similar nodes are aggregated to identify target non-ticket personnel with abnormal identity characteristics, comprising: Optimize the structure of the heterogeneous graph based on graph regularization constraint mechanism and sparse relationship constraint mechanism; Using a node aggregation mechanism based on local subgraph, the ticket personnel nodes in the optimized heterogeneous graph are clustered and aligned to generate reference group features; Compare the node features of the suspected non-ticket personnel with the reference group features to identify target non-ticket personnel with abnormal identity characteristics.

4. A method for tracking non-ticketed personnel in a construction site according to claim 1, wherein Based on the motion trajectory of the non-ticket personnel, the trajectory deviation degree index of the target non-ticket personnel is calculated, comprising: Based on the motion trajectory, the spatial position information of the target non-ticket personnel at different times is mapped into a construction site topology graph model; wherein the construction site topology graph model is composed of multiple nodes and edges, and the nodes correspond to key position units of the construction area, including: entrance and exit nodes, work area nodes and hazard source nodes, and the edges represent the path connection relationship that personnel can walk; In the construction site topology graph model, the trajectory of the non-ticket personnel and the authorized path of the construction area are modeled based on a graph potential field path constraint algorithm; Based on the path potential function, combined with the edge weight constraint and spatial topology distance between nodes, the non-ticket personnel trajectory is decomposed and compared segment by segment to generate a corresponding deviation degree index sequence; Extract two feature parameters, cumulative deviation degree and maximum single segment deviation degree, from the deviation degree index sequence as quantitative basis for the abnormal degree of the target non-ticket personnel trajectory, and output the trajectory deviation degree index.

5. A construction site-oriented non-ticketed person dynamic tracking method according to claim 4, characterized in that, The graph potential field path constraint algorithm comprises: wherein, E risk (p) represents the comprehensive risk potential value of the off-ticket personnel under the track p; p t represents the spatial position node of the off-ticket personnel at time t; q t represents the standard position node of the construction area authorized path at time t; d(p t ,q t ) represents the spatial topological distance between the actual position p t of the off-ticket personnel and the standard node q t ; w(p t ,q t ) represents the path edge weight value connecting the nodes p t and q t in the topological graph; T represents the total time sequence length of the track; represents the maximum deviation degree of a single segment in the track; π(a t |s t ) represents the strategy probability of the off-ticket personnel taking action a t in state s t ; α represents the weight coefficient of the cumulative deviation degree; β represents the power index coefficient of the deviation degree; γ represents the weight coefficient of the maximum single segment deviation degree; and λ represents the reinforcement learning risk correction factor.

6. A method for tracking non-ticketed personnel on a construction site according to claim 1, wherein, The behavior causal diagram model comprises: where C anom (u) represents the causal anomaly score for non-ticket personnel u; M represents the number of behavioral outcome variables involved in the behavioral causal graph model; Y i represents the ith actual observed behavior outcome; represents the ith expected value inferred by the causal model; represents the variance of the ith behavior outcome corresponding to the historical normal ticket personnel group; N represents the number of dimensions of the causal effect comparison; represents the observed causal effect of non-ticket personnel in the jth dimension; represents the reference causal effect value of normal ticket personnel in the same dimension; L represents the number of detected privilege conflict events; p k represents the risk factor of the kth privilege conflict event; ξ represents the weight factor of the residual term; v represents the weight factor of the causal effect deviation term; χ represents the weight factor of the privilege conflict term.

7. A method for tracking non-ticketed personnel on a construction site according to claim 1, wherein, Based on the behavior causal diagram model, the behavior state of the target non-ticket personnel is judged, and a behavior risk judgment result is generated, comprising: In the behavior causal graph model, a causal inference driven anomaly detection mechanism is introduced, structural equation modeling and comparative intervention methods are used to compare the behavior sequence of the target non-ticket personnel with the baseline behavior pattern of normal ticket personnel, determine the potential causal trigger factors of abnormal behavior occurrence, and extract behavior state characteristic parameters; wherein the behavior state characteristic parameters include: abnormal duration, deviation area sensitivity, and permission conflict frequency; Based on the behavior state characteristic parameters, the behavior risk score of the target non-ticket personnel in the preset construction environment is calculated; wherein the risk score is based on three dimensions of trajectory deviation degree, plan deviation degree, and permission violation degree; According to the corresponding relationship between the behavior risk score and the preset multi-level risk threshold interval, the behavior risk discrimination result of the target non-ticket personnel is generated.

8. A non-ticketed person dynamic tracking system for a construction site, characterized by, The system comprises: A screening unit is configured to collect information data of personnel entering a construction site and compare the information data with a construction ticket roster to screen suspected non-ticket personnel. An aggregation and identification unit is configured to establish a heterogeneous graph based on information data of the suspected non-ticket personnel and information data in a ticket personnel database, and to aggregate local similar nodes based on the heterogeneous graph to identify target non-ticket personnel with abnormal identity characteristics. A deviation degree calculation unit is configured to obtain a motion trajectory of the target non-ticket personnel, calculate a deviation degree based on the motion trajectory of the non-ticket personnel, and obtain a trajectory deviation degree index of the target non-ticket personnel. A risk discrimination unit is configured to construct a behavior causal graph model based on the trajectory deviation degree index, discriminate a behavior state of the target non-ticket personnel based on the behavior causal graph model, and generate a behavior risk discrimination result. An alarm unit is configured to generate a risk level label based on the behavior risk discrimination result.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The program is executed by a processor to implement the steps of the method of any one of claims 1-7.

10. An electronic device, comprising: Comprises: The computer readable storage medium of claim 9; And One or more processors for executing programs in the computer readable storage medium.

Images