A dynamic face recognition device for smart construction site based on a BIM model

Abstract

The application relates to the field of face recognition technology and discloses a dynamic face recognition device for a smart construction site based on a BIM model, which comprises a perception terminal integrated on a standard safety helmet and comprises an internal miniature camera, an external wide-angle camera and a controller. The internal miniature camera is fixed to the inner side of the brim of the safety helmet and captures stable eyebrow-eye and forehead area images of a wearer in a non-invasive manner, so as to collect a first biological feature image of a worker. The external wide-angle camera is installed on the forehead of the safety helmet and is used for collecting a first environment image of the visual angle of the worker. The first environment image is collected by the external wide-angle camera to determine the space-time anchor point of the worker, a BIM real-time interaction engine infers an expected worker based on the space-time anchor point, and then the unobstructed local first biological feature image of the eyebrows and eyes is collected by the internal miniature camera, and the first biological feature image is compared with a preset biological feature model of the expected worker, so that the stability and accuracy of identity recognition are ensured.

Description

A dynamic facial recognition device for smart construction sites based on BIM models Technical Field

[0001] This invention relates to the field of facial recognition technology, specifically a dynamic facial recognition device for smart construction sites based on BIM models. Background Technology

[0002] The construction of smart construction sites places higher demands on the refined and automated management of on-site personnel. Existing facial recognition technology, as a key entry point, plays a crucial role in static scenarios such as access control and attendance. Currently, facial recognition applications in the smart construction site field are mainly solidified into several typical forms: fixed access management systems, deploying facial recognition gates or access control at the construction site gate, dormitory building, or key material warehouse entrances. When personnel pass through, the system captures facial images and compares them with a whitelist in the background to achieve attendance and basic area access control. Shallow linkage with BIM models, using the BIM model as a static database of personnel information. After the facial recognition system confirms the person's identity, it queries the BIM database for the person's job type, work group, and preset access permissions to determine whether they can enter a specific area. Handheld terminal inspection applications, where managers use an app on their mobile phones or tablets to take photos of on-site workers, manually or semi-automatically verifying their identities, and associating the inspection results with components or location points in the BIM model.

[0003] However, once personnel enter the vast, dynamic, and harsh environment of a construction site, its application suffers from numerous shortcomings: complex environmental factors such as strong light, backlighting, shadows, rain, fog, and dust severely interfere with image acquisition quality and reduce recognition accuracy. Due to safety requirements, workers must wear personal protective equipment such as helmets, masks, and safety glasses, resulting in the loss of key facial features over large areas, rendering traditional recognition algorithms ineffective. While existing technologies have solved the "access" problem, they cannot dynamically and accurately correlate a person's identity with their real-time location in the three-dimensional physical space represented by the BIM model and the construction tasks they undertake. The application of BIM models remains at a superficial level of static data querying. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a dynamic facial recognition device for smart construction sites based on BIM models, which solves the bottleneck problem of difficult identity recognition in complex construction site environments.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a dynamic facial recognition device for smart construction sites based on BIM models, comprising:

[0006] The sensing terminal, integrated into a standard safety helmet, includes an inner miniature camera, an outer wide-angle camera, and a controller. The inner miniature camera is fixed inside the brim of the safety helmet to capture stable images of the wearer's eyebrows, eyes, and forehead in a non-invasive manner, for collecting the worker's first biometric image. The outer wide-angle camera is installed on the forehead of the safety helmet to collect the worker's first environmental image from their perspective. The controller is communicatively connected to both the inner miniature camera and the outer wide-angle camera.

[0007] The BIM real-time interactive engine communicates with the controller and matches the first environmental image captured by the external wide-angle camera with the preset BIM model to determine the spatiotemporal anchor point of the sensing terminal in the BIM model. Based on the spatiotemporal anchor point and the construction plan data contained in the BIM model, it determines the expected workers associated with the spatiotemporal anchor point. It obtains the preset biometric model of the expected workers and performs a 1:1 comparison between the first biometric image captured by the internal miniature camera and the preset biometric model to complete the identification of the workers.

[0008] Preferably, the controller includes a multi-source positioning and attitude sensing unit and a wireless communication module; wherein, the multi-source positioning and attitude sensing unit integrates an inertial measurement unit to assist in sensing the wearer's head posture and integrates other positioning signals as auxiliary; the wireless communication module is used to exchange data with the BIM real-time interaction engine.

[0009] Preferably, the BIM real-time interactive engine includes:

[0010] The visual positioning and matching service module is used to perform matching between the first environmental image and the BIM model to determine spatiotemporal anchor points;

[0011] A personnel information and feature database is used to store the preset biometric model;

[0012] The collaborative verification and decision-making logic center is used to determine the expected operators and schedule the 1:1 comparison based on the spatiotemporal anchor points.

[0013] Preferably, the first biometric image is a partial image of the worker's face that is not obscured when the worker is wearing personal protective equipment, and the preset biometric model is a model pre-established based on this partial image of the worker's face.

[0014] Preferably, the BIM real-time interaction engine is specifically used for:

[0015] The first environmental image is processed into an environmental fingerprint;

[0016] The environmental fingerprint is matched with a preset virtual environmental fingerprint library generated based on the BIM model rendering to determine the spatiotemporal anchor point.

[0017] Preferably, the BIM real-time interaction engine is further configured as follows:

[0018] When it is determined that the spatiotemporal anchor points of two sensing terminals are in a proximity state, peer-to-peer cross-validation is triggered.

[0019] The peer-to-peer cross-validation includes: instructing one of the sensing terminals to use its external wide-angle camera to capture a second environmental image containing the wearer of the other sensing terminal, and validating the second environmental image based on a preset biometric model of the wearer of the other sensing terminal.

[0020] Preferably, based on the results of the peer-to-peer cross-validation, a spatiotemporal trust chain is established for multiple sensing terminals that have successfully verified each other, thereby improving the credibility of their identity recognition results.

[0021] Preferably, after completing identity recognition, the BIM real-time interaction engine is further configured to:

[0022] Based on the spatiotemporal anchor points and the construction plan data in the BIM model, the key work objects corresponding to the current tasks of the expected workers are determined;

[0023] The visual features of the key work object are used as attention focus instructions and fed forward to the sensing terminal, so that the sensing terminal can determine whether the operator's field of view is effectively focused on the key work object through the external wide-angle camera.

[0024] Preferably, the controller further includes an edge computing unit configured to receive the attention focus instruction and analyze the image captured by the external wide-angle camera to determine whether the operator's field of view is effectively focused on the key work object.

[0025] Preferably, the sensing terminal also includes a power supply, which is detachably installed on both sides of the safety helmet and electrically connected to the internal miniature camera, the external wide-angle camera, and the controller, respectively.

[0026] This invention provides a dynamic facial recognition device for smart construction sites based on BIM models. It has the following beneficial effects:

[0027] 1. This invention reliably identifies workers wearing full personal protective equipment (PPE), cleverly avoiding the industry problem of recognition failure due to facial occlusion. Its implementation does not seek to reconstruct or identify a face under occlusion, but rather through a logical transformation: the device's external wide-angle camera first captures an initial environmental image to determine the worker's spatiotemporal anchor point, and the BIM real-time interactive engine infers the expected worker based on this. This process transforms an open-ended "search" problem into a targeted "verification" problem; that is, it only needs to capture an unobstructed partial biometric image (such as eyebrows and eyes) using the internal miniature camera and compare it 1:1 with the expected worker's preset biometric model. Because the facial area required for verification remains exposed when wearing the equipment, the stability and accuracy of identity recognition are ensured without violating safety regulations.

[0028] 2. This invention enables continuous and dynamic tracking and management of all personnel within the construction site, expanding the scope of identity perception from isolated entrance "points" to the entire "volume" of the three-dimensional construction space. This is achieved through the continuous acquisition of initial environmental images by the sensing terminal's wide-angle camera, which are then matched with the BIM model by the backend BIM real-time interaction engine, generating a continuous spatiotemporal anchor point trajectory. This mechanism allows managers to monitor the precise distribution and historical movement of each worker in the three-dimensional space of the construction site in real time, providing an unprecedented global dynamic perspective for resource scheduling, safety management, and emergency response.

[0029] 3. This invention significantly deepens the application level of BIM models, transforming them from static 3D information dashboards into a proactive "digital twin" engine capable of cognition, reasoning, and decision-making. When the BIM real-time interactive engine acquires a spatiotemporal anchor point, it not only passively displays that location but also actively queries its associated construction plan data to infer the expected workforce at that time and space. BIM serves as the core of this logical reasoning, providing context for on-site events and proactively initiating the identity verification process. This deep integration allows BIM to truly participate in the smart construction site management system.

[0030] 4. This invention significantly enhances the reliability and anti-spoofing capability of the entire system by introducing a collaborative verification mechanism among groups. When the system determines that the spatiotemporal anchor points of multiple sensing terminals are adjacent to each other, a peer-to-peer cross-verification procedure is triggered, whereby the operator's devices become temporary verification nodes for each other's identities. The device group that successfully completes cross-verification forms a highly reliable spatiotemporal trust chain. This decentralized verification network can effectively identify and isolate erroneous information caused by individual device failures, signal drift, or malicious video spoofing, establishing a solid "firewall" for the stable operation of the entire system.

[0031] 5. This invention elevates the application value of identity management from confirming "personnel presence" to supervising "task execution." After confirming the worker's identity and location, the BIM real-time interaction engine can further determine the key work objects that the worker should currently focus on based on the construction plan and issue attention focus instructions. The sensing terminal analyzes the images from the external wide-angle camera to determine whether the worker's gaze is focused on the key object. This allows managers not only to know "who is where" but also, to a certain extent, whether "they are doing the work correctly," providing a new and refined data support for quality control and process compliance management of key processes. Attached Figure Description

[0032] Figure 1 is a flowchart of the operation of the device of the present invention;

[0033] Figure 2 is a top-view three-dimensional schematic diagram of the sensing terminal in this invention;

[0034] Figure 3 is a three-dimensional view of the sensing terminal in this invention from a low angle.

[0035] The components include: 1. Safety helmet; 2. Internal miniature camera; 3. External wide-angle camera; 4. Controller. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] Please refer to Figures 1-3. This embodiment of the invention provides a dynamic facial recognition device for smart construction sites based on BIM models, comprising:

[0038] The sensing terminal, integrated into a standard safety helmet 1, includes an inner miniature camera 2, an outer wide-angle camera 3, and a controller 4. The inner miniature camera 2 is fixed to the inside of the helmet 1's brim, capturing stable images of the wearer's eyebrows, eyes, and forehead area in a non-invasive manner. This is used to collect the worker's first biometric image, which is a partial image of the worker's face not obscured when wearing personal protective equipment. The outer wide-angle camera 3 is mounted on the forehead of the helmet 1, capturing the worker's first environmental image from their perspective. The controller 4 is communicatively connected to both the inner miniature camera 2 and the outer wide-angle camera 3. The sensing terminal also includes a power supply, which is detachably mounted on both sides of the helmet 1 and electrically connected to the inner miniature camera 2, the outer wide-angle camera 3, and the controller 4. The controller 4 includes a multi-source positioning and attitude sensing unit and a wireless communication module. The multi-source positioning and attitude sensing unit integrates an inertial measurement unit to assist in sensing the wearer's head posture and integrates other positioning signals as aids. The wireless communication module is used for data exchange with the BIM real-time interaction engine. The controller 4 also includes an edge computing unit, which is configured to receive attention focus instructions and analyze the images captured by the external wide-angle camera 3 to determine whether the operator's field of view is effectively focused on the critical work object.

[0039] The BIM real-time interactive engine communicates with the controller 4 and matches the first environmental image captured by the external wide-angle camera 3 with the preset BIM model to determine the spatiotemporal anchor point of the sensing terminal in the BIM model. Based on the spatiotemporal anchor point and the construction plan data contained in the BIM model, it determines the expected workers associated with the spatiotemporal anchor point. It obtains the preset biometric model of the expected workers, which is a model pre-built based on the local facial image of the workers. It also performs a 1:1 comparison between the first biometric image captured by the internal miniature camera 2 and the preset biometric model to complete the identification of the workers.

[0040] The BIM real-time interactive engine includes:

[0041] The visual positioning and matching service module is used to perform matching between the first environmental image and the BIM model to determine spatiotemporal anchor points;

[0042] The personnel information and feature database is used to store preset biometric models; the collaborative verification and decision-making logic center is used to determine the expected personnel for operation based on spatiotemporal anchor points and schedule 1:1 comparisons.

[0043] The BIM real-time interactive engine is specifically used to: process the first environmental image into an environmental fingerprint; and match the environmental fingerprint with a preset virtual environmental fingerprint library generated based on the BIM model rendering to determine the spatiotemporal anchor point.

[0044] When it is determined that the spatiotemporal anchor points of two sensing terminals are in a proximity state, peer-to-peer cross-verification is triggered. Peer-to-peer cross-verification includes: instructing the wide-angle camera 3 of one sensing terminal to acquire a second environmental image containing the wearer of the other sensing terminal, and verifying the second environmental image based on the preset biometric model of the wearer of the other sensing terminal. Based on the results of peer-to-peer cross-verification, a spatiotemporal trust chain is established for multiple sensing terminals that have successfully verified each other, thereby improving the credibility of their identity recognition results.

[0045] After completing identity recognition, the BIM real-time interaction engine is also configured as follows:

[0046] Based on the spatiotemporal anchor points and construction plan data in the BIM model, the key work objects corresponding to the current tasks of the expected workers are determined; the visual characteristics of the key work objects are used as attention focus instructions and fed forward to the sensing terminal, so that the sensing terminal can determine whether the field of view of the workers is effectively focused on the key work objects through the external wide-angle camera 3.

[0047] The above embodiments provide a dynamic facial recognition device for smart construction sites based on BIM models. In implementation, its overall architecture manifests as a distributed system where front-end perception and back-end cognition work collaboratively. Logically, this system consists of two core entities: a sensing terminal configured on workers for on-site perception, and a BIM real-time interaction engine serving as the system's cognitive decision-making center. These two entities exchange data in real-time and bidirectionally via a wireless communication network, together forming a complete dynamic recognition and verification closed loop.

[0048] In a preferred embodiment, the sensing terminal is physically a lightweight, low-power intelligent module integrated or mounted on the safety helmet 1 worn by construction workers. This deployment method ensures that the device can move freely with the worker throughout the construction site and continuously acquire their environmental and biometric information from a first-person perspective. Combining the sensing terminal with the safety helmet 1, a standardized personal protective equipment, not only ensures the device's universal applicability but also makes its deployment and use in real-world work scenarios natural and non-intrusive, thus providing a physical carrier for the full-process, dynamic tracking and identification of workers within the construction site.

[0049] The BIM real-time interaction engine is typically deployed as a backend server system. Depending on specific project requirements and network conditions, this engine can be deployed on a cloud server with powerful computing and storage capabilities to support concurrent access and data processing on ultra-large-scale construction sites; or it can be deployed as a local server in the site's computer room to achieve lower data transmission latency and stronger local data management capabilities. Regardless of the deployment method, this engine serves as the "digital brain" of the entire system, responsible for performing the most complex computational tasks, including processing and parsing all data uploaded by front-end sensing terminals, engaging in deep interaction with the massive BIM model, and ultimately making identification, verification, and decisions.

[0050] The device's overall operation is characterized by a clear division of labor and collaboration. The front-end sensing terminal focuses on its core sensing and data acquisition tasks, acting like the "eyes" and "sensors" of the human body, responsible for capturing raw images and data. The back-end BIM real-time interaction engine acts as the "brain," receiving sensory information from the front end and using its powerful cognitive and reasoning capabilities to transform it into a profound understanding of personnel identity, spatial location, and behavioral status. The sensing terminal sends the collected data stream to the BIM real-time interaction engine via the 5G or Wi-Fi wireless networks covering the construction site; after completing analysis, reasoning, and decision-making, the engine then sends necessary instructions or verification requests back to the sensing terminal. This architectural design cleverly balances the portability and battery life of the front-end device with the powerful data processing capabilities of the back-end system.

[0051] As a highly integrated module, the sensing terminal contains multiple interconnected units. First is the two-way sensing camera system, consisting of an external camera (external wide-angle camera 3) and an internal camera (internal miniature camera 2). The external camera uses a wide-angle lens to capture the widest possible first-person perspective of the worker; the initial environmental image it captures serves as the primary basis for subsequent visual positioning. The internal camera, a miniature camera, is precisely mounted inside the brim of the safety helmet. Its lens is specially designed to stably capture the area above the worker's face, such as the brow ridge and forehead, which remains clearly visible even when wearing personal protective equipment. The initial biometric image acquired from this area provides the foundation for subsequent local feature comparison.

[0052] The controller 4 within the sensing terminal also houses an edge computing and positioning unit. At its core is a low-power artificial intelligence (AI) processor responsible for local preprocessing before data upload. For example, it directly processes the raw video stream captured by the external wide-angle camera 3, transforming it into a small-data-volume, feature-rich environmental fingerprint by running specific feature extraction algorithms, rather than transmitting a large video file. Simultaneously, it can also execute 1:1 comparison commands issued by the BIM real-time interaction engine. This unit also integrates an inertial measurement unit (IMU) for real-time sensing of the wearer's head posture changes. This posture data serves as auxiliary information for visual positioning, improving the stability and accuracy of positioning.

[0053] To enable connectivity with the backend, the controller 4 in the sensing terminal also includes a wireless communication module. This module supports 5G or high-density Wi-Fi networks deployed on the construction site and is responsible for establishing a low-latency, high-bandwidth data link between the sensing terminal and the BIM real-time interaction engine, ensuring reliable uploading of environmental fingerprints and timely reception of backend commands.

[0054] The BIM real-time interactive engine, serving as the central hub for system cognition and decision-making, comprises multiple collaborative software functional modules. Among them, the BIM model and construction plan database forms the system's knowledge foundation. This database not only stores detailed 3D geometric models of buildings but, more importantly, integrates 4D construction plan data. This data precisely binds specific construction tasks, execution times, responsible personnel, and other information to specific components or spatial locations within the BIM model.

[0055] The visual positioning and matching service module is the core of the engine's spatial cognition. During system initialization, this service automatically renders the BIM model to generate a virtual environmental fingerprint database covering the entire construction site area, containing tens of thousands of fingerprints with precise 3D coordinates and orientation labels. During system runtime, after receiving real-time environmental fingerprints uploaded from the front end, this service uses efficient image retrieval technology to quickly match them within the virtual fingerprint database. Once a match is successful, the current spatiotemporal anchor point of the sensing terminal can be immediately determined.

[0056] The personnel information and biometric database manages personnel data. It stores information on all authorized workers entering the construction site, and each person is associated with an encrypted, pre-defined biometric model based on a partial image of their face. This model serves as the basis for subsequent 1:1 identity verification.

[0057] Finally, the Collaborative Verification and Decision-Making Logic Center is the "commander" of the entire engine. This module is responsible for orchestrating and scheduling the work of all other modules. It receives the spatiotemporal anchor points determined by the visual positioning service, then queries the BIM database to infer the expected personnel, and then retrieves the corresponding preset biometric model from the personnel database and issues a 1:1 verification command. In addition, all high-level intelligent decisions, such as triggering peer-to-peer cross-verification, establishing and managing spatiotemporal trust chains, and issuing attention focus commands, are all uniformly completed by this logic center. It organically connects the data and functions of each module, forming the unique, progressive cognitive and verification process of this invention.

[0058] When a worker registers and receives the sensing terminal for the first time, the system executes an identity initialization and local biometric modeling process. In this process, the worker must face the miniature camera 2 inside the sensing terminal, which captures multiple frames of first biometric images of the area above their face from multiple angles. This area is chosen because it remains stably and unobstructedly exposed even after the worker has worn safety helmets and masks according to regulations. After receiving these images, the BIM real-time interaction engine processes them using feature extraction algorithms into a compact, robust, and uniquely representative biometric model. This model is then encrypted and bound to the worker's identity information (such as name, job title, and permissions) within the BIM system, stored in the personnel information and biometric database, providing a benchmark for subsequent 1:1 verification.

[0059] When workers wearing the sensing terminal move around the construction site, the system enters a continuous spatiotemporal anchor point positioning process based on environmental fingerprints. The sensing terminal's external wide-angle camera 3 captures first-person environmental images from the worker's perspective in real time. The edge computing unit built into the controller 4 does not directly transmit the raw video, but instead runs an efficient feature point extraction algorithm to extract key points and descriptors that remain stable under changes in lighting and viewing angle from the image, encodes them into a small environmental fingerprint, and uploads it via a wireless module. Upon receiving this environmental fingerprint, the visual positioning and matching service module of the BIM real-time interaction engine immediately performs high-speed retrieval and matching in a pre-built BIM virtual environmental fingerprint database. Once the best-matching virtual fingerprint is found, the precise three-dimensional coordinates and orientation information attached to the virtual fingerprint are determined as the spatiotemporal anchor point of the sensing terminal at the current moment.

[0060] Following this, the system seamlessly transitions to a 1:1 identity verification process based on BIM proactive reasoning. This process is the core of this invention's solution to the occlusion recognition problem. The instant the spatiotemporal anchor point is established, the collaborative verification and decision-making logic center of the BIM real-time interaction engine immediately uses this spatiotemporal information as an index to query the BIM model and construction plan database. For example, the query result indicates that at the current time, the construction task corresponding to this 3D coordinate point is "pipeline welding in area A of the second floor," which should be performed by welder "Zhang San" according to the plan. Based on this, the system makes proactive reasoning, identifying "Zhang San" as the expected worker at this location. Subsequently, the engine retrieves "Zhang San's" preset biometric model from the database and sends it as a 1:1 verification command to the sensing terminal at that location. Upon receiving the command, the terminal's internal miniature camera 2 instantly captures the wearer's current first biometric image and compares it locally with the received model. The comparison result (yes / no) is sent back to the engine, thus completing identity verification with extremely high efficiency and accuracy.

[0061] To further enhance the overall credibility of the system, this invention also includes a peer-to-peer cross-verification process based on group perception. When the BIM real-time interaction engine determines, through spatiotemporal anchor point data, that multiple sensing terminals have physically clustered together (e.g., a work group working on the same platform), it will proactively trigger this process. The engine will instruct the first sensing terminal to use its external wide-angle camera 3 to locate the wearer of the second sensing terminal within its field of view, and use a preset biometric model obtained from the engine, belonging to the wearer of the second sensing terminal, to perform a temporary identity verification. This process occurs within the group. Once multiple terminals within an area have successfully verified each other's identities, the engine will establish a highly credible spatiotemporal trust chain for this temporary group. Any individual that cannot be confirmed by neighboring verified nodes will have its location and identity credibility automatically reduced by the system, triggering attention.

[0062] Finally, after the identities and locations of the workers have been confirmed with high credibility, the system can extend the verification to the task execution level, that is, initiate the attention focus verification process based on the BIM task. If the current spatiotemporal anchor point corresponds to a key process in the BIM construction plan, the BIM real-time interaction engine will extract the core operation object of that process from the model, such as a specific high-voltage switch or a prefabricated component requiring precise installation, and define its 3D model or visual features as the key operation object. Subsequently, the engine feeds forward the visual features of this object as an attention focus instruction to the worker's perception terminal. The edge computing unit on the terminal analyzes the video stream of its external wide-angle camera 3 to determine whether the wearer's field of view center has stayed and focused on the designated key operation object for an effective period of time. This completes the final closed-loop verification of "the right person, in the right place, focusing on the right thing".

[0063] To more intuitively demonstrate the collaborative working method of the various technical components in the device of this invention, a typical application scenario is described below. This scenario describes the complete process of a worker entering the construction site and performing a key procedure.

[0064] Assuming an electrician's identity information and corresponding pre-registered biometric model are already in the system, the device of this invention begins operation when the worker enters the construction site wearing an activated safety helmet integrated with a sensing terminal. Its wide-angle camera 3 automatically captures a first-person view of the environment from the worker's perspective. The edge computing unit within the terminal processes this image in real-time into a series of continuous environmental fingerprints, which are then uploaded to the backend BIM real-time interaction engine via a wireless network.

[0065] Upon receiving the first environmental fingerprint, the engine immediately matches it against the vast BIM virtual environment fingerprint database, quickly determining the worker's initial spatiotemporal anchor point, such as being located inside the main entrance of the construction site. Next, the engine's decision logic center, based on this spatiotemporal anchor point and the current time, queries the BIM construction plan database to infer that the entry behavior matches the planned task of "electrician team entry," thus confirming the worker's expected identity. A 1:1 verification command containing their preset biometric model is then sent to their sensing terminal.

[0066] After receiving the instruction, the sensing terminal uses its internal miniature camera 2 to capture the first biometric image of the worker's eyebrow and eye area, and compares it locally with the issued model. If the comparison matches, the result is transmitted back, and the worker's identity is confirmed for the first time. Subsequently, as the worker moves within the construction site to their work point on the fifth floor of the building, the aforementioned "environmental positioning-identity reasoning-verification" process is continuously and silently executed at a very high frequency, forming a real-time three-dimensional movement trajectory accurate down to the BIM model.

[0067] When the worker arrived at the fifth-floor work area, his spatiotemporal anchor point indicated that he was in proximity to another worker installing pipes nearby. After recognizing this spatial relationship, the decision logic center of the BIM real-time interaction engine proactively triggered a peer-to-peer cross-verification procedure. The engine sent the other worker's preset biometric model to their respective sensing terminals. The electrician's sensing terminal captured the plumber's image through its external wide-angle camera 3 and completed the verification, and vice versa. After two successful verification results were reported, the system established a temporary spatiotemporal trust chain for them, further solidifying the credibility of their respective location and identity data.

[0068] Finally, the electrician arrived at the distribution box specified in the construction plan. His spatiotemporal anchor point perfectly matched the coordinates of the distribution box in the BIM model, and his identity was reconfirmed. At this point, the BIM real-time interaction engine's verification level was raised from identity confirmation to task confirmation. Based on the construction plan, the engine determined that the core operation object of his current task was a specific circuit breaker within the distribution box; this circuit breaker was thus defined as the critical operation object.

[0069] The engine feeds forward the circuit breaker's visual characteristics, such as its shape, color, and label text, as an attention focus instruction to the electrician's sensing terminal. When the worker opens the distribution box door and focuses their gaze on locating and operating the specific circuit breaker, the edge computing unit of the sensing terminal analyzes the video stream from the external wide-angle camera 3 to determine whether the center of their field of view has effectively stopped on the designated key work object. Once confirmed, the entire verification process forms a complete closed loop: from "the right person" to "the right place," and then to "focusing on the right thing," each step is confirmed with data support.

[0070] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A dynamic facial recognition device for smart construction sites based on BIM models, characterized in that, include: The sensing terminal, integrated into a standard safety helmet (1), includes an internal miniature camera (2), an external wide-angle camera (3), and a controller (4). The internal miniature camera (2) is fixed to the inside of the brim of the safety helmet (1) to capture stable images of the wearer's eyebrows, eyes, and forehead in a non-invasive manner, for collecting the first biometric image of the worker. The external wide-angle camera (3) is installed on the forehead of the safety helmet (1) to collect the first environmental image from the worker's perspective. The controller (4) is communicatively connected to the internal miniature camera (2) and the external wide-angle camera (3). The BIM real-time interaction engine is communicatively connected to the controller (4) and is based on the first biometric image collected by the external wide-angle camera (3). An environmental image is matched with a preset BIM model to determine the spatiotemporal anchor point of the sensing terminal in the BIM model; based on the spatiotemporal anchor point and the construction plan data contained in the BIM model, the expected workers associated with the spatiotemporal anchor point are determined; a preset biometric model of the expected workers is obtained, and a 1:1 comparison is performed between the first biometric image collected by the internal miniature camera (2) and the preset biometric model to complete the identification of the workers; the BIM real-time interaction engine is specifically used to: process the first environmental image into an environmental fingerprint; match the environmental fingerprint with a preset virtual environmental fingerprint library generated according to the BIM model rendering to determine the spatiotemporal anchor point.

2. The dynamic face recognition device for smart construction sites based on BIM models according to claim 1, characterized in that, The controller (4) includes a multi-source positioning and attitude sensing unit and a wireless communication module; wherein, the multi-source positioning and attitude sensing unit integrates an inertial measurement unit to assist in sensing the wearer's head posture and integrates other positioning signals as an aid; the wireless communication module is used to exchange data with the BIM real-time interactive engine.

3. The dynamic face recognition device for smart construction sites based on BIM models according to claim 1, characterized in that, The BIM real-time interaction engine includes: a visual positioning and matching service module, used to perform matching between the first environmental image and the BIM model to determine spatiotemporal anchor points; a personnel information and feature database, used to store the preset biometric model; and a collaborative verification and decision-making logic center, used to determine the expected workers based on the spatiotemporal anchor points and schedule the 1:1 comparison.

4. A dynamic face recognition device for smart construction sites based on BIM models according to claim 1, characterized in that, The first biometric image is a partial image of the worker's face that is not obscured when the worker is wearing personal protective equipment, and the preset biometric model is a model pre-established based on this partial image of the worker's face.

5. A dynamic face recognition device for smart construction sites based on BIM models according to claim 1, characterized in that, The BIM real-time interaction engine is also configured to trigger peer-to-peer cross-validation when it is determined that the spatiotemporal anchor points of two sensing terminals are in a proximity state. The peer-to-peer cross-validation includes: instructing one of the sensing terminals to use its external wide-angle camera (3) to capture a second environmental image containing the wearer of the other sensing terminal, and verifying the second environmental image based on the preset biometric model of the wearer of the other sensing terminal.

6. A dynamic face recognition device for smart construction sites based on BIM models according to claim 5, characterized in that, Based on the results of the peer-to-peer cross-validation, a spatiotemporal trust chain is established for multiple sensing terminals that have successfully verified each other, thereby improving the credibility of their identity recognition results.

7. A dynamic face recognition device for smart construction sites based on BIM models according to claim 2, characterized in that, After the BIM real-time interaction engine completes identity recognition, it is further configured to: determine the key work object corresponding to the current task of the expected worker based on the spatiotemporal anchor point and the construction plan data in the BIM model; feed the visual features of the key work object as attention focus instructions to the sensing terminal so that the sensing terminal can determine whether the field of view of the worker is effectively focused on the key work object through the external wide-angle camera (3).

8. A dynamic face recognition device for smart construction sites based on BIM models according to claim 7, characterized in that, The controller (4) further includes an edge computing unit configured to receive the attention focus instruction and analyze the image captured by the external wide-angle camera (3) to determine whether the operator's field of view is effectively focused on the key work object.

9. A dynamic face recognition device for smart construction sites based on BIM models according to claim 1, characterized in that, The sensing terminal also includes a power supply, which is detachably installed on both sides of the safety helmet (1) and electrically connected to the internal miniature camera (2), the external wide-angle camera (3), and the controller (4).

Images