Intelligent detection method and system for passage occupation based on three-dimensional space reconstruction

Abstract

The application discloses a kind of channel occupation intelligent detection method and system based on three-dimensional space reconstruction.Method includes: mobile terminal gathers the visual data of channel, according to visual data, generates the live scene three-dimensional model of channel and uploads to cloud server by three-dimensional reconstruction; cloud server compares live scene three-dimensional model with the standard database in the specification model of channel in advance, to extract spatial geometric difference; cloud server judges whether there is foreign matter occupation according to spatial geometric difference, and sends corresponding audit result to mobile terminal;When there is foreign matter occupation, cloud server marks foreign matter on live scene three-dimensional model, generates channel detection report according to the marking result, and sends channel detection report and abnormal early warning to mobile terminal.The application significantly improves the accuracy of channel occupation identification, reduces artificial inspection and management cost.

Description

Technical Field

[0001] This invention belongs to the fields of computer vision, three-dimensional spatial reconstruction and intelligent detection technology, and more specifically, relates to a channel occupancy intelligent detection method and system based on three-dimensional spatial reconstruction. Background Technology

[0002] Fire lanes are vital lifelines for evacuation and fire truck access during fires. However, due to haphazard parking, disorderly stacking of debris, and illegal building alterations, fire lanes are frequently obstructed, posing a significant threat to public safety. Traditional fire lane inspections rely primarily on regular manual patrols or image recognition from ordinary 2D surveillance cameras. Manual inspections are time-consuming and labor-intensive, and cannot provide continuous, high-frequency monitoring.

[0003] Existing 2D image recognition technologies, such as deep learning-based vehicle or obstacle detection, have significant limitations in practical applications. On one hand, they suffer from limited viewing angles and severe occlusion. 2D images lack depth information, making them prone to missed or false detections when multiple vehicles are parked in a staggered manner or when obstacles are partially obscured by trees or awnings. On the other hand, they cannot accurately determine spatial boundaries. Fire lanes include not only the ground-level boundary line but also legally mandated clearance height and spatial limits. 2D images cannot quantitatively calculate whether an object truly encroaches on the fire lane space in the vertical direction or along the vertical boundary. Therefore, overcoming the limitations of traditional 2D planar vision and achieving high-precision, blind-spot-free, and quantitatively analytical intelligent detection of fire lane occupancy from the perspective of 3D spatial scene perception is a pressing technical problem that needs to be solved in the field of fire safety supervision. Summary of the Invention

[0004] The main objective of this invention is to provide a channel occupancy intelligent detection method and system based on three-dimensional spatial reconstruction, so as to overcome the shortcomings of the prior art.

[0005] To achieve the above-mentioned objectives, the present invention adopts the following technical solution: The first aspect of this invention provides an intelligent channel occupancy detection method based on three-dimensional spatial reconstruction, comprising: S1, a mobile terminal collecting visual data of a channel, performing three-dimensional reconstruction based on the visual data, generating a real-world three-dimensional model of the channel, and uploading it to a cloud server; S2, the cloud server comparing the real-world three-dimensional model with a standard model of the channel in a pre-set standard database to extract spatial geometric differences; S3, the cloud server determining whether there is foreign object occupancy in the channel based on the spatial geometric differences, and sending the corresponding review result to the mobile terminal; when foreign object occupancy is present, the cloud server marking the foreign object on the real-world three-dimensional model, generating a channel detection report based on the marking result, and sending the channel detection report and anomaly warning to the mobile terminal.

[0006] Preferably, the method further includes: when the mobile terminal receives the channel detection report and the abnormal warning, it issues an alarm signal to instruct the person in charge to rectify the channel according to the channel detection report, and after the rectification is completed, it executes S1-S3 again until the channel is free of foreign objects.

[0007] Preferably, before step S1, the method further includes: collecting three-dimensional spatial geometric data, boundary red line data, legal clearance height data, and legal clearance depth data of each channel under standard conditions to construct a standard model for each channel; adding each of the standard models to the standard database, and establishing a correspondence between the standard models and the channels in the standard database.

[0008] Preferably, the passageway includes a fire escape route; and / or, the visual data includes video data or multi-angle image data.

[0009] Preferably, the 3D reconstruction based on the visual data specifically includes: extracting key images from the visual data; using the SFM algorithm or SLAM algorithm to calculate the motion trajectory of the mobile terminal and the sparse point cloud of the scene; and reconstructing the 3D model of the scene based on the motion trajectory and the sparse point cloud.

[0010] Preferably, the 3D reconstruction based on the visual data specifically includes: pre-constructing a four-plane structured prior for the channel; initializing and optimizing the Gaussian point cloud of the channel in the visual data based on the four-plane structured prior; dividing the optimized Gaussian point cloud into several overlapping sub-segments along the depth direction of the channel, and performing local optimization in each overlapping sub-segment; performing global correction on all overlapping sub-segments based on the positional closure and Gaussian alignment of the overlapping areas between adjacent overlapping sub-segments, with the condition of satisfying the edge continuity constraint; filtering out dynamic interference Gaussians in the corrected Gaussian point cloud and performing planarization to complete the missing areas, thereby obtaining the 3D model of the actual scene.

[0011] Preferably, the on-site real-scene 3D model is compared with the standard model of the channel in the preset standard database. Specifically, this includes: spatially aligning the on-site real-scene 3D model and the standard model in the same 3D coordinate system; calculating the spatial point cloud distance or volume overlap between the spatially aligned on-site real-scene 3D model and the standard model; and determining whether the spatial objects in the on-site real-scene 3D model exceed the boundary red line or occupy the legal clearance height based on the spatial point cloud distance or the volume overlap.

[0012] Preferably, the comparison in S2 includes one or more of the following: compliance anchoring benchmark comparison, three-dimensional geometric refinement comparison, semantic attribute hierarchical comparison, temporal persistence comparison, and multimodal cross-validation comparison.

[0013] A second aspect of this invention provides an intelligent channel occupancy detection system based on three-dimensional spatial reconstruction, comprising: a data acquisition module deployed on a mobile terminal for acquiring visual data of a channel; a three-dimensional reconstruction module deployed on the mobile terminal for performing three-dimensional reconstruction based on the visual data, generating a real-world three-dimensional model of the channel and uploading it to a cloud server; a comparison module deployed on the cloud server for comparing the real-world three-dimensional model with a standard model of the channel in a pre-set standard database to extract spatial geometric differences; a judgment module deployed on the cloud server for judging whether there is foreign object occupancy in the channel based on the spatial geometric differences and sending the corresponding review result to the mobile terminal; and a reporting and early warning module deployed on the cloud server, which, when foreign object occupancy is detected, marks the foreign object on the real-world three-dimensional model, generates a channel detection report based on the marking result, and sends the channel detection report and anomaly warning to the mobile terminal; the intelligent channel occupancy detection system based on three-dimensional spatial reconstruction is used to implement the intelligent channel occupancy detection method based on three-dimensional spatial reconstruction as described above.

[0014] Preferably, the system further includes: a task publishing module, deployed on a cloud server, used to publish inspection tasks to mobile terminals according to a preset timing strategy, the preset timing strategy including a timed inspection strategy and a report triggering strategy; and a closed-loop monitoring module, deployed on the mobile terminal, used to mark the corresponding warning task in the mobile terminal display interface as pending rectification and start the rectification countdown when receiving an abnormal warning.

[0015] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention provides a channel occupancy intelligent detection method and system based on three-dimensional spatial reconstruction. On the one hand, by taking multi-angle shots from a mobile terminal, a complete three-dimensional model of the actual scene is obtained based on three-dimensional (3D) spatial reconstruction, which can perfectly solve the common problems of object occupancy and blind spots in traditional 2D images, and greatly improve the accuracy of occupancy identification. On the other hand, by introducing standardized models from a standard database for comparison, it can not only determine the occupancy of the ground red line, but also calculate the actual three-dimensional size of the object through 3D mesh, accurately identifying illegal space occupancy that is difficult to quantify from a 2D perspective, such as "high-altitude suspended objects and illegal low ground piles". Furthermore, the mobile terminal completes the initial reconstruction with high computational load, and the cloud performs standard comparison of big data and closed-loop supervision of the entire process. The "end-cloud" collaboration is highly efficient, which not only improves the emergency response speed, but also realizes full-process automation, greatly reducing the cost of manual verification and management. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart of a channel occupancy intelligent detection method based on three-dimensional spatial reconstruction provided in an embodiment of the present invention.

[0018] Figure 2 A user interface diagram of a mobile terminal provided in an embodiment of the present invention.

[0019] Figure 3 A block diagram of a channel occupancy intelligent detection system based on three-dimensional spatial reconstruction provided in an embodiment of the present invention. Detailed Implementation

[0020] In view of the shortcomings of the prior art, the inventors of this invention, through long-term research and extensive practice, have proposed the technical solution of this invention. The following will further explain and illustrate this technical solution, its implementation process, and its principles.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0022] Furthermore, in the description of this invention, it should be understood that the terms "upper," "lower," "inner," "outer," "horizontal," "vertical," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0023] In the description of this specification, the references to terms such as "an embodiment," "a particular embodiment," or "the embodiment" indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0024] Figure 1 A flowchart illustrating the intelligent channel occupancy detection method based on three-dimensional spatial reconstruction provided in this embodiment of the invention. (See attached document.) Figure 1 , combined Figure 2 This paper provides a detailed description of a channel occupancy intelligent detection method based on three-dimensional spatial reconstruction. The method includes the following operations S1-S3.

[0025] Operation S1: The mobile terminal collects visual data from the channel, performs 3D reconstruction based on the visual data, generates a real-world 3D model of the channel, and uploads it to the cloud server.

[0026] Operation S2 allows the cloud server to compare the on-site 3D model with the standard model of the channel in the pre-set standard database to extract spatial geometric differences.

[0027] When operating S3, the cloud server determines whether there are foreign objects occupying the channel based on spatial geometric differences and sends the corresponding review results to the mobile terminal. When there are foreign objects occupying the channel, the cloud server marks the foreign objects on the on-site 3D model, generates a channel detection report based on the marking results, and sends the channel detection report and anomaly warning to the mobile terminal.

[0028] Compared to traditional 2D single-angle image recognition, this invention obtains a complete on-site spatial model through 3D spatial reconstruction. Inspectors can take multi-angle photos using mobile devices, solving common problems in traditional 2D images such as object overlap and blind spots, significantly improving the accuracy of occupancy identification. By introducing standardized model comparison, it can not only determine ground boundary occupancy but also calculate the actual three-dimensional dimensions of objects through 3D meshes, accurately identifying illegal spatial occupancy that is difficult to quantify from a 2D perspective, such as "suspended objects at height" and "illegally placed low-lying piles." The mobile terminal completes the initial reconstruction with high computational demands, while the cloud performs big data standard comparisons and full-process closed-loop monitoring, not only improving emergency response speed but also achieving full automation, greatly reducing manual verification and management costs.

[0029] The mobile terminal in this embodiment of the invention is a device with shooting function, data processing function and communication function, including but not limited to smartphones, tablets, smartwatches, laptops, 3D laser scanners, etc.

[0030] In a preferred embodiment, the method further includes: when the mobile terminal receives a channel detection report and an anomaly warning, it issues an alarm signal to instruct the person in charge to rectify the channel according to the channel detection report, and after the rectification is completed, it executes operations S1-S3 again until there are no foreign objects occupying the channel.

[0031] This invention establishes a closed-loop tracking mechanism for the entire task lifecycle, monitoring in real time the rectification status of responsible persons regarding abnormal warning information; when an abnormal warning exists, the user interface of the mobile terminal displays the following: Figure 2 As shown. After rectification, operations S1-S3 are re-executed for a second three-dimensional comparison and verification until the occupancy is determined to be lifted and the order is closed.

[0032] In a preferred embodiment, before performing operation S1, the method further includes: collecting three-dimensional spatial geometric data, boundary red line data, legal clearance height data and legal clearance depth data of each channel under standard conditions to construct a standard model of each channel; adding each standard model to the standard database, and establishing a correspondence between the standard model and the channel in the standard database.

[0033] Preferably, the passageway is a fire escape route. It should be noted that the passageway can also be other types of passageways, such as personnel evacuation routes, emergency access routes, vehicle access routes, pipeline access routes, etc.

[0034] Preferably, the visual data includes video data or multi-angle image data. For example, a supervisor may use a handheld mobile device to capture a video covering the entire area of ​​the passageway, or capture multiple images from different angles covering the entire area.

[0035] In a preferred embodiment, the three-dimensional reconstruction based on visual data in operation S1 specifically includes: extracting key images from the visual data; calculating the motion trajectory of the mobile terminal and the sparse point cloud of the scene using the Structure from Motion (SFM) algorithm or the Simultaneous Localization and Mapping (SLAM) algorithm; and reconstructing a three-dimensional model of the scene based on the motion trajectory and the sparse point cloud.

[0036] In another preferred embodiment, operation S1 involves 3D reconstruction based on visual data, specifically including the following sub-operations S11-S15.

[0037] Sub-operation S11 pre-constructs the four-plane structured prior of the channel.

[0038] The main geometry of passageway scenes can be abstracted as a cylindrical space enclosed by four dominant planes: the ground, the top, the left side wall, and the right side wall. The "four-plane prior" is essentially a simplified Manhattan world assumption for passageway scenes. Initial equations for the four planes can be obtained in advance through initial point cloud plane fitting, homography estimation, and manually inputting cross-sectional parameters, serving as benchmark constraints for subsequent full-process optimization. The four-plane structured prior injects deterministic geometric constraints into the reconstruction process, solving the core problems of pose estimation drift, point cloud dispersion, and uneven planes in general reconstruction methods under weak or repetitive texture environments in passageways.

[0039] Sub-operation S12 initializes and optimizes the Gaussian point cloud of the channels in the visual data based on the four-plane structured prior.

[0040] Gaussian point cloud initialization and coplanar constraint optimization can correct the scattering problem of 3D Gaussian sputtering (3DGS) at its source. Specifically, the initial Gaussian points are preferentially distributed near the four planes, and the initial normals of the Gaussian points in the planar regions are aligned with the plane normals, which significantly reduces the optimization search space, accelerates the convergence speed, and reduces the probability of divergence. In addition to the original rendering loss of 3DGS, a planar regularization term is added to constrain the distance from the Gaussian center belonging to the same plane to the corresponding plane to be minimized, and the principal direction of the Gaussian points to be consistent with the plane normal, thus forcing the Gaussian points in the planar regions to maintain coplanarity.

[0041] Sub-operation S13 divides the optimized Gaussian point cloud into several overlapping sub-segments along the channel depth direction, and performs local optimization in each overlapping sub-segment.

[0042] Sub-operation S14, based on satisfying the edge continuity constraint, performs global correction on all overlapping sub-segments according to the positional closure and Gaussian alignment of the overlapping regions between adjacent overlapping sub-segments.

[0043] As the camera moves along the depth, the pose estimation error accumulates frame by frame, eventually leading to channel scale deformation, planar misalignment, and structural line bending. Sub-operations S13-S14 are used to achieve depth segmented gradient correction and edge continuity constraints, thus solving the problem of cumulative drift in long channels.

[0044] The segmented gradient correction is as follows: the channel is divided into several overlapping sub-segments along the depth direction. Local high-precision optimization is performed within each segment, and global correction is performed between segments by closure of the pose of the overlapping area and Gaussian alignment. This avoids the unidirectional accumulation of errors along the depth direction, which is equivalent to controlling the "gradient" amplification of the pose error.

[0045] The specific boundary line continuity constraint is as follows: the intersection lines of the four planes (wall-ground intersection line, wall-top intersection line) are the rigid structural lines of the passage, constraining the intersection lines of adjacent segments to be continuous and smooth in space, further locking the consistency of the global structure and preventing misalignment during segment splicing.

[0046] Sub-operation S15 filters out dynamic interference Gaussians in the corrected Gaussian point cloud and performs planarization to complete the missing areas, thus obtaining a 3D model of the actual scene.

[0047] Dynamic objects such as pedestrians, vehicles, and temporary equipment within the passageway generate redundant outlier Gaussians. Dynamic interference filtering can be achieved by identifying and removing dynamic Gaussians through multi-view consistency checks, planar outlier determination, and temporal motion saliency, retaining only the static structure. For reconstruction holes caused by occlusion or weak textures, a four-plane prior is used to fill the missing areas with Gaussians that conform to the planar equation, ensuring the integrity of the main structure of the passageway and outputting a complete static 3D model (a real-world 3D model).

[0048] The 3D reconstruction method defined by sub-operations S11-S15 is a customized and optimized 3DGS 3D reconstruction scheme for "narrow and regular passage scenarios". Its core idea is to use the structured geometry prior of the passage to constrain the optimization process of 3DGS. In regular cross-section passage scenarios such as integrated utility tunnels, subway tunnels, mine roadways, and building corridors, it outperforms unconstrained general 3DGS. The above 3D reconstruction method can specifically solve the technical problems commonly found in general 3D reconstruction technology in narrow and structured passages, such as geometric drift in weak texture areas, attenuation of depth direction accuracy, artifacts of dynamic objects, and insufficient geometric dimensional accuracy. Finally, it outputs a high-precision, high-flatness static 3D benchmark model of the passage, providing a reliable spatial benchmark for subsequent occupancy detection and comparison.

[0049] In another preferred embodiment, operation S2 compares the on-site real-scene 3D model with the standard model of the channel in the preset standard database. Specifically, this includes: spatially aligning the on-site real-scene 3D model and the standard model in the same 3D coordinate system, for example, using the Iterative Closest Point (ICP) algorithm or a feature point-based spatial registration algorithm for the spatial alignment; calculating the spatial point cloud distance or volume overlap between the spatially aligned on-site real-scene 3D model and the standard model; and determining whether spatial objects in the on-site real-scene 3D model exceed the boundary red line or occupy the legal clearance height based on the spatial point cloud distance or volume overlap.

[0050] Preferably, the comparison method in operation S2 includes, but is not limited to, one or more of the following: compliance anchoring benchmark comparison, three-dimensional geometric refinement comparison, semantic attribute hierarchical comparison, temporal continuity comparison, and multimodal cross-validation comparison. The comparison method in this invention can also be other than the five types mentioned above. Various comparison methods can be implemented independently or combined to improve the comparison accuracy of occupancy detection and ensure channel passage safety.

[0051] By using compliant anchored benchmarks to compare and anchor the legally mandated boundaries of fire lanes, invalid comparison areas are reduced at the source, and irrelevant interference outside the lanes is shielded, effectively reducing misjudgments caused by light and shadow offsets and edge objects. Through refined 3D geometric comparison, spatial-level precision is achieved, improving the accuracy of clearance dimension detection to the centimeter level, directly aligning with the quantitative judgment requirements for fire compliance. Semantic attribute hierarchical comparison distinguishes target categories according to the degree of fire impact, eliminating invalid differences such as pedestrians and fallen debris, significantly reducing invalid alarms. Through continuous temporal comparison combined with dwell time and dynamic / static characteristics, temporary passage and long-term illegal occupation are accurately distinguished, conforming to actual fire control rules. Multimodal cross-validation comparison, integrating multi-source sensor data for cross-validation, adapts to extreme conditions such as dense smoke, heavy fog, and nighttime, ensuring all-weather detection stability. Ultimately, from multiple levels—spatial boundaries, geometric accuracy, target attributes, temporal dimension, and environmental adaptation—the comparison accuracy and scenario robustness of occupancy detection are systematically improved, effectively ensuring safe passage.

[0052] Based on the same inventive concept, embodiments of the present invention also provide a channel occupancy intelligent detection system based on three-dimensional spatial reconstruction, see below. Figure 3 The intelligent channel occupancy detection system 300 based on three-dimensional spatial reconstruction includes a data acquisition module 310, a three-dimensional reconstruction module 320, a comparison module 330, a judgment module 340, and a reporting and early warning module 350.

[0053] The acquisition module 310, deployed on a mobile terminal, is used to acquire visual data of the channel. The 3D reconstruction module 320, also deployed on the mobile terminal, performs 3D reconstruction based on the visual data, generating a real-world 3D model of the channel and uploading it to the cloud server. The comparison module 330, deployed on the cloud server, compares the real-world 3D model with the standard model of the channel in a pre-set standard database to extract spatial geometric differences. The judgment module 340, deployed on the cloud server, determines whether there are foreign objects occupying the channel based on spatial geometric differences and sends the corresponding review results to the mobile terminal. The reporting and early warning module 350, deployed on the cloud server, marks the foreign object on the real-world 3D model when it is found, generates a channel detection report based on the marking results, and sends the channel detection report and anomaly warning to the mobile terminal.

[0054] The intelligent channel occupancy detection system 300 based on three-dimensional spatial reconstruction is used to implement the above-mentioned intelligent channel occupancy detection method based on three-dimensional spatial reconstruction.

[0055] In a preferred embodiment, the intelligent channel occupancy detection system 300 based on three-dimensional spatial reconstruction further includes a task publishing module and a closed-loop monitoring module. The task publishing module, deployed on a cloud server, is used to publish inspection tasks to mobile terminals according to preset timing strategies, including timed inspection strategies and report-triggered strategies. The closed-loop monitoring module, deployed on the mobile terminal, is used to mark the corresponding warning task in the mobile terminal's display interface as pending rectification and start a rectification countdown when an abnormal warning is received.

[0056] The intelligent channel occupancy detection method and system based on three-dimensional spatial reconstruction provided in this invention starts with the input of basic data including a standardized model, goes through task issuance and execution, intelligent analysis and early warning, and finally achieves closed-loop management of task supervision and verification. Taking a building fire escape route as an example, the specific implementation steps of this invention are as follows.

[0057] Step 1 involves the basic data entry for the standardized model.

[0058] Before the system is put into use, technicians use high-precision 3D laser scanners, portable multi-view reconstruction equipment, or mobile phones to conduct initial 3D scans of the fire escape routes of each building that are completely cleared and in compliance with regulations.

[0059] The collected data includes the ground width boundary of the fire lane, the turning radius, and the vertical fire clearance height (e.g., the 4-meter clearance height required by regulations). This data is imported into the system to establish a standard database and generate a digital twin "standard model" for each fire lane.

[0060] Step 2, Task Issuance and Execution.

[0061] The system automatically sends inspection tasks to the mobile app of grid-based fire inspection personnel according to preset time-series strategies (such as daily routine inspections or based on reported clues).

[0062] After arriving at the designated fire exit, inspection personnel use their mobile phones to record a short video of 15 to 30 seconds around the exit, ensuring that the video covers all angles of the exit. The lightweight edge-side 3D reconstruction engine embedded in the mobile app extracts video frames in real time and performs spatial geometric calculations while the video is being recorded, quickly generating a realistic 3D model of the scene on the phone. This model is then uploaded to the cloud server without loss of quality via 5G / Wi-Fi network.

[0063] Step 3, Intelligent Analysis and Early Warning.

[0064] After receiving the 3D model of the actual scene, the cloud server automatically retrieves the corresponding "standard model" for that channel using the intelligent spatial comparison engine. Utilizing the ICP iterative nearest-point algorithm, the actual scene model and the standard model are aligned by translation and rotation within the same 3D Cartesian coordinate system. After alignment, the system performs difference calculations on the spatial meshes of the two models.

[0065] Taking a illegally parked car in a fire lane as an example, the 3D model of the scene will show prominent grid data or point cloud data at the corresponding location. By comparing it with the open space in the standard model, the system can accurately extract the 3D outline of the car and calculate that it occupies 2.1 meters of the fire lane width, which is considered illegal occupation. The system automatically renders this part of the foreign object data representing the car in the 3D model of the scene as a highlighted red for differentiation. At the same time, the system automatically generates a fire lane occupancy detection report (lane detection report) with a 3D intuitive view, and pushes abnormal warning notifications and rectification tasks to the mobile phones of property managers, grid workers, and car owners in the area based on the Internet of Things communication protocol.

[0066] Step 4: Task monitoring and verification.

[0067] The system backend marks the warning task as "pending rectification" and starts a countdown. Upon receiving the warning, the responsible person immediately goes to the site to persuade vehicles to leave or clear debris. After rectification, the responsible person or grid worker reopens the mobile app, re-captures the video at the original location, and uploads a second 3D reconstruction. The cloud server compares the reconstructed model with the "standard model" again. When the system determines that the spatial geometric difference between the two is within the allowable error range, proving that the passage is no longer obstructed by foreign objects, the system automatically determines that the hazard has been eliminated and changes the task status to "closed," completing the closed-loop process of the entire regulatory verification.

[0068] It should be understood that the above embodiments are merely illustrative of the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent changes or modifications made in accordance with the spirit and essence of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A channel occupancy intelligent detection method based on three-dimensional spatial reconstruction, characterized in that, include: S1, the mobile terminal collects visual data of the channel, performs three-dimensional reconstruction based on the visual data, generates a real-world three-dimensional model of the channel, and uploads it to the cloud server. S2, the cloud server compares the on-site real-scene 3D model with the standard model of the channel in the pre-set standard database to extract spatial geometric differences; S3, the cloud server determines whether there is any foreign object occupying the channel based on the spatial geometric difference, and sends the corresponding review result to the mobile terminal; when there is a foreign object occupying the channel, the cloud server marks the foreign object on the on-site real-world 3D model, generates a channel detection report based on the marking result, and sends the channel detection report and anomaly warning to the mobile terminal.

2. The intelligent channel occupancy detection method based on three-dimensional spatial reconstruction according to claim 1, characterized in that, The method further includes: When the mobile terminal receives the channel detection report and the abnormal warning, it issues an alarm signal to instruct the person in charge to rectify the channel according to the channel detection report. After the rectification is completed, S1-S3 are executed again until the channel is free of foreign objects.

3. The intelligent channel occupancy detection method based on three-dimensional spatial reconstruction according to claim 1, characterized in that, Before step S1, the method further includes: Collect three-dimensional spatial geometric data, boundary line data, legal clearance height data, and legal clearance depth data of each channel under standard conditions to construct a standard model of each channel; Each of the specified standard models is added to the standard database, and a correspondence between the specified standard models and the channels is established in the standard database.

4. The intelligent channel occupancy detection method based on three-dimensional spatial reconstruction according to claim 1, characterized in that, The passageway includes fire escape routes; And / or, the visual data includes video data or multi-angle image data.

5. The intelligent channel occupancy detection method based on three-dimensional spatial reconstruction according to claim 1, characterized in that, The three-dimensional reconstruction based on the visual data specifically includes: Extract key images from the visual data; Using the SFM algorithm or SLAM algorithm, calculate the motion trajectory of the mobile terminal and the sparse point cloud of the scene; Based on the motion trajectory and the sparse point cloud, a 3D model of the actual scene is reconstructed.

6. The intelligent channel occupancy detection method based on three-dimensional spatial reconstruction according to claim 1, characterized in that, The three-dimensional reconstruction based on the visual data specifically includes: A four-plane structured prior is pre-constructed for the channel; Based on the four-plane structured prior, the Gaussian point cloud of the channels in the visual data is initialized and optimized for coplanar constraints. The optimized Gaussian point cloud is divided into several overlapping sub-segments along the channel depth, and local optimization is performed in each overlapping sub-segment. To satisfy the boundary continuity constraint, global correction is performed on all overlapping segments based on the positional closure and Gaussian alignment of the overlapping areas between adjacent overlapping segments; Dynamic interference Gaussians in the corrected Gaussian point cloud are filtered out and the missing areas are filled in using planarization to obtain the actual scene 3D model.

7. The intelligent channel occupancy detection method based on three-dimensional spatial reconstruction according to claim 1, characterized in that, The on-site 3D model is compared with the standard model of the channel in the pre-set standard database, specifically including: The on-site 3D model and the standard model are spatially aligned in the same 3D coordinate system. Calculate the spatial point cloud distance or volume overlap between the spatially aligned on-site 3D model and the standard model; Based on the spatial point cloud distance or the volume overlap, determine whether the spatial objects in the on-site real-scene 3D model exceed the boundary red line or occupy the legal clearance height.

8. The intelligent channel occupancy detection method based on three-dimensional spatial reconstruction according to claim 7, characterized in that, The comparisons in S2 include one or more of the following: compliance anchoring benchmark comparison, three-dimensional geometric refinement comparison, semantic attribute hierarchical comparison, temporal persistence comparison, and multimodal cross-validation comparison.

9. A channel occupancy intelligent detection system based on three-dimensional spatial reconstruction, characterized in that, include: The acquisition module, deployed on a mobile terminal, is used to acquire visual data from the channel. The 3D reconstruction module, deployed on a mobile terminal, is used to perform 3D reconstruction based on the visual data, generate a real-world 3D model of the passage, and upload it to the cloud server. The comparison module, deployed on a cloud server, is used to compare the on-site 3D model with the standard model of the channel in a pre-set standard database to extract spatial geometric differences. The judgment module, deployed on a cloud server, is used to determine whether there is any foreign object occupying the channel based on the spatial geometric differences, and to send the corresponding review result to the mobile terminal. The reporting and early warning module is deployed on a cloud server. When there is a foreign object occupying the space, the reporting and early warning module is used to mark the foreign object on the 3D model of the scene, generate a channel detection report based on the marking results, and send the channel detection report and the abnormal warning to the mobile terminal. The intelligent channel occupancy detection system based on three-dimensional spatial reconstruction is used to implement the intelligent channel occupancy detection method based on three-dimensional spatial reconstruction as described in any one of claims 1-8.

10. The intelligent channel occupancy detection system based on three-dimensional spatial reconstruction according to claim 9, characterized in that, The system also includes: The task publishing module, deployed on a cloud server, is used to publish inspection tasks to mobile terminals according to a preset timing strategy, which includes a timed inspection strategy and a report triggering strategy. The closed-loop monitoring module, deployed on a mobile terminal, is used to mark the corresponding warning task in the mobile terminal display interface as pending rectification when an abnormal warning is received, and to start the rectification countdown.

Images