Anti-collision intelligent identification and warning device and method for elevator

Abstract

The application relates to an anti-collision intelligent identification and warning device and method for an elevator, which comprises an intelligent analysis module, a three-dimensional identification and perception module, a hierarchical warning module and a linkage control module; the three-dimensional identification and perception module is used for acquiring initial spatial positioning data of goods in an elevator car and a plurality of multi-source point cloud data of elevator operation; the intelligent analysis module is used for analyzing all the multi-source point cloud data to obtain a collision risk result; the linkage control module is used for linkage control of elevator operation and control of the hierarchical warning module to output different warning signals according to the collision risk result. The device realizes real-time three-dimensional identification, dynamic risk prediction, collision hidden danger avoidance in advance, equipment damage reduction, goods loss reduction and personnel safety risk reduction through the intelligent analysis module, the three-dimensional identification and perception module, the hierarchical warning module and the linkage control module; hierarchical warning and linkage control can guarantee safety without affecting operation efficiency, and have the functions of safety and high efficiency.

Description

Technical Field

[0001] This application relates to the field of elevator technology, and in particular to an intelligent anti-collision identification and warning device and method for elevators. Background Technology

[0002] General-purpose passenger elevators are designed to transport passengers, and the distance between passengers and the elevator car is not detected or controlled. For elevators used for general freight transport, the distance between the goods and the elevator car is visually predictable and controllable by the transport personnel when goods are entering or exiting the car. However, when forklifts transport goods in and out of elevators, they are usually large items, and these goods inevitably obstruct the forklift driver's view, making it impossible to predict and control the distance between the goods and the front, left, right, and top, potentially leading to damage to the goods or elevator car walls during transport.

[0003] There is an urgent need to develop a cargo anti-collision identification device for elevators to solve problems such as damage to the car walls, cargo, and elevator structure caused by cargo deviation, oversized dimensions, misalignment, human error, and blind spots during lifting, lowering, and loading / unloading, as well as personnel safety hazards. The device should be able to detect and warn transport personnel, reduce cargo loss, improve efficiency, and enhance the safety of transport personnel. Summary of the Invention

[0004] This application provides an intelligent anti-collision identification and warning device and method for elevators, which solves the technical problem that existing elevators do not detect and warn of damage to transported goods and / or personnel, resulting in large cargo losses and low safety.

[0005] To achieve the above objectives, this application provides the following technical solution:

[0006] On the one hand, an intelligent anti-collision recognition and warning device for elevators is provided, including an intelligent analysis module and a three-dimensional recognition and perception module, a graded warning module and a linkage control module connected to the intelligent analysis module;

[0007] The three-dimensional recognition and perception module is used to acquire the initial spatial positioning data of the goods in the elevator car and several multi-source point cloud data of the elevator operation. The multi-source point cloud data includes the real-time operation data of the elevator and the real-time three-dimensional data and change data of the goods contained in the elevator car.

[0008] The intelligent analysis module is used to analyze all the multi-source point cloud data to obtain collision risk results;

[0009] The linkage control module is used to control the operation of the elevator and control the graded warning module to output different warning signals based on the collision risk results.

[0010] Optionally, the intelligent analysis module includes a data fusion processing submodule and a risk identification submodule;

[0011] The data fusion processing submodule is used to perform preprocessing, mapping and trend prediction processing on each of the multi-source point cloud data in sequence to obtain distance prediction data between the cargo and the car.

[0012] The risk identification submodule is used to determine the risk by comparing the distance prediction data with a preset safety threshold, and obtain the collision risk result.

[0013] Optionally, the data fusion processing submodule includes a preprocessing unit, a three-dimensional structure unit, a mapping unit, and a prediction and estimation unit;

[0014] The preprocessing unit is used to sequentially perform time synchronization, noise reduction and normalization processing on the multi-source point cloud data to obtain multi-source processed data.

[0015] The three-dimensional structural unit is used to generate a three-dimensional spatial mesh of the cargo based on the cargo's moving speed, attitude change, and spatial position within the car using dynamic structural mapping, thereby obtaining a variable resolution cargo three-dimensional spatial mesh.

[0016] The mapping unit is used to map the contour differences between the cargo and the car onto the three-dimensional spatial grid to obtain a geometric mapping model.

[0017] The prediction estimation unit is used to obtain continuous local geometric mapping data by inputting the continuous multi-source processing data into the geometric mapping model, and to predict the continuous local geometric mapping data to obtain distance prediction data.

[0018] Optionally, the preprocessing unit includes a time unification subunit, a noise reduction subunit, a drift correction subunit, a normalization subunit, a mapping subunit, and a registration optimization subunit;

[0019] The time unification subunit is used to perform time synchronization processing on all the multi-source point cloud data to obtain time-unified processed data after time unification.

[0020] The noise reduction subunit is used to perform high-frequency noise reduction on the time-uniformly processed data using frequency domain filtering to obtain low-frequency processed data.

[0021] The drift correction subunit is used to perform low-frequency offset compensation processing on the low-frequency processed data using surface fitting to obtain drift correction processed data.

[0022] The normalization subunit is used to normalize the spatial scale of the drift correction data according to a preset three-dimensional metric standard to obtain normalized data.

[0023] The mapping subunit is used to perform positioning mapping processing on the initial spatial positioning data and the normalized processing data to obtain standardized positioning processing data.

[0024] The registration optimization subunit is used to perform geometric alignment processing on the standardized positioning data using iterative nearest-point registration to obtain multi-source processing data.

[0025] Optionally, the three-dimensional structural unit includes a matrix generation subunit, a network skeleton subunit, a partition adjustment subunit, a density adjustment subunit, and a verification subunit;

[0026] The matrix generation subunit is used to calculate the velocity vector and attitude quaternion based on the spatial positioning data of the multi-source processing data; and to generate a motion state matrix based on the cargo spatial position, velocity vector and attitude quaternion of all the multi-source processing data.

[0027] The network skeleton sub-unit is used to perform distributed fusion processing based on the initial spatial positioning data and the motion state matrix to obtain a unified three-dimensional space reference frame mesh skeleton.

[0028] The partitioning adjustment subunit is used to perform spatial partitioning transformation on the reference frame mesh skeleton according to the cargo velocity vector, attitude change rate and the relative position between the cargo and the car, to obtain a variable resolution matrix.

[0029] The density adjustment subunit is used to perform mesh density adjustment processing on the reference frame mesh skeleton using the variable resolution matrix to obtain a multi-precision spatial mesh.

[0030] The verification subunit is used to resample and perform consistency verification on the multi-precision spatial grid to obtain an optimized three-dimensional spatial grid.

[0031] Optionally, the mapping unit includes an identifier processing subunit, a separation subunit, a feature extraction subunit, an association mapping subunit, and a region synthesis subunit;

[0032] The identification processing subunit is used to perform region identification processing on each grid cell of the three-dimensional spatial grid to obtain a spatial grid model with structural difference markers.

[0033] The separation subunit is used to separate and extract the contour region of the spatial grid model by point cloud clustering segmentation to obtain contour point cloud data;

[0034] The feature extraction subunit is used to extract features from the contour point cloud data using normal vectors and surface fitting to obtain geometric feature data of each region.

[0035] The association mapping subunit is used to perform association mapping processing on each region geometric feature data and the spatial grid model to obtain a region geometric mapping model corresponding to each region geometric feature data.

[0036] The region synthesis subunit is used to perform multi-region synthesis processing on all the region geometric mapping models to obtain a geometric mapping model.

[0037] Optionally, the prediction estimation unit includes: a sequence matrix subunit, a dynamic extraction subunit, a prediction trend subunit, a distance calculation subunit, and a distance correction subunit;

[0038] The sequence matrix sub-unit is used to extract the geometric features of cargo posture changes in the continuous multi-source processing data according to the geometric mapping model, and obtain a morphological change sequence matrix composed of a unified time series index.

[0039] The dynamic extraction subunit is used to extract dynamic attributes from the morphological change sequence matrix to obtain a dynamic attribute matrix.

[0040] The prediction trend subunit is used to predict the movement trajectory of goods in the three-dimensional space of the geometric mapping model by using the dynamic attribute matrix as the multidimensional time feature of the multivariate long short-term memory network, and to obtain the position sequence matrix of the predicted position of the goods.

[0041] The distance calculation subunit is used to calculate the shortest distance between the predicted location of the goods and the car using Euclidean distance based on the location sequence matrix, and to obtain distance prediction data.

[0042] The distance correction subunit is used to perform weighted correction on the distance prediction data to obtain corrected distance prediction data.

[0043] Optionally, the risk identification submodule is further configured to determine the collision risk result as no collision risk if the distance prediction data is greater than the upper limit of the preset safety threshold; determine the collision risk result as high collision risk if the distance prediction data is less than the lower limit of the preset safety threshold; and determine the collision risk result as low collision risk if the distance prediction data is between the upper and lower limits of the preset safety threshold.

[0044] Optionally, the linkage control module is further configured to control the elevator to operate normally if the collision risk result is no collision risk; control the elevator to operate normally and control the graded warning module to output a cargo adjustment warning signal if the collision risk result is low collision risk; and control the elevator to stop working and control the graded warning module to output cargo adjustment and danger warning signals if the collision risk result is high collision risk.

[0045] On the other hand, a collision avoidance intelligent recognition and warning method for elevators is provided, including the following steps:

[0046] Acquire initial spatial positioning data of goods in the elevator car and several multi-source point cloud data of elevator operation. The multi-source point cloud data includes real-time operation data of the elevator and real-time three-dimensional data and change data of goods contained in the elevator car.

[0047] The collision risk results are obtained by analyzing all the multi-source point cloud data.

[0048] Based on the collision risk results, the elevator operation is controlled and different warning signals are output in a coordinated manner.

[0049] Specifically, each of the multi-source point cloud data is preprocessed, mapped, and trend predicted sequentially to obtain distance prediction data between the cargo and the car; the distance prediction data is then compared with a preset safety threshold to determine the risk and obtain the collision risk result.

[0050] The intelligent collision avoidance identification and warning device and method for the elevator includes an intelligent analysis module and a three-dimensional identification and perception module, a graded warning module, and a linkage control module connected to the intelligent analysis module. The three-dimensional identification and perception module is used to acquire the initial spatial positioning data of the goods in the elevator car and several multi-source point cloud data of the elevator operation. The multi-source point cloud data includes the real-time operation data of the elevator and the real-time three-dimensional data and change data of the goods in the elevator car. The intelligent analysis module is used to analyze all the multi-source point cloud data to obtain the collision risk result. The linkage control module is used to control the operation of the elevator and control the graded warning module to output different warning signals according to the collision risk result.

[0051] As can be seen from the above technical solutions, this application has the following advantages: The intelligent anti-collision recognition and warning device of the elevator achieves real-time three-dimensional recognition and dynamic risk prediction through intelligent analysis module, three-dimensional recognition and perception module, graded warning module and linkage control module, avoiding collision hazards in advance and reducing equipment damage, cargo loss and personnel safety risks; graded warning and linkage control do not affect the operation efficiency and can ensure safety, and have safe and efficient functions; it solves the technical problem that existing elevators do not detect and warn of damage to transported goods and / or transported personnel, resulting in large cargo losses and low safety. Attached Figure Description

[0052] 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 of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 This is a schematic diagram of the frame of the anti-collision intelligent recognition and warning device for the elevator described in the embodiments of this application;

[0054] Figure 2 This is a schematic diagram of the frame of the anti-collision intelligent recognition and warning device for the elevator according to another embodiment of this application;

[0055] Figure 3 This is a schematic diagram of the data fusion processing submodule in the anti-collision intelligent identification and warning device for the elevator described in the embodiments of this application;

[0056] Figure 4 This is a schematic diagram of the preprocessing unit in the anti-collision intelligent identification and warning device for the elevator described in the embodiments of this application;

[0057] Figure 5 This is a schematic diagram of the frame of the three-dimensional structural unit in the anti-collision intelligent recognition and warning device for the elevator described in the embodiments of this application;

[0058] Figure 6 This is a schematic diagram of the mapping unit in the anti-collision intelligent recognition and warning device for the elevator described in the embodiments of this application;

[0059] Figure 7 This is a schematic diagram of the frame of the prediction and estimation unit in the intelligent collision avoidance and warning device for elevators according to an embodiment of this application;

[0060] Figure 8 This is a flowchart illustrating the steps of the intelligent anti-collision recognition and warning method for elevators described in this application embodiment. Detailed Implementation

[0061] To make the inventive objectives, features, and advantages of this application more apparent and understandable, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0062] In the description of the embodiments of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0063] In the embodiments of this application, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in the embodiments of this application according to the specific circumstances.

[0064] This application provides an intelligent anti-collision identification and warning device and method for elevators, which solves the technical problem that existing elevators do not detect and warn of damage to transported goods and / or personnel, resulting in large cargo losses and low safety.

[0065] Example 1:

[0066] Figure 1 This is a schematic diagram of the frame of the intelligent anti-collision recognition and warning device for the elevator described in the embodiments of this application.

[0067] like Figure 1 As shown, this application embodiment provides an intelligent anti-collision recognition and warning device for an elevator, including an intelligent analysis module 10 and a three-dimensional recognition and perception module 20, a graded warning module 30 and a linkage control module 40 connected to the intelligent analysis module 10. The linkage control module 40 is also connected to the graded warning module 30.

[0068] It should be noted that the intelligent anti-collision recognition and warning device of the elevator uses a three-dimensional recognition and perception module 20 to identify multi-dimensional data of the elevator car and the goods inside the car. The intelligent analysis module 10 analyzes the multi-dimensional data identified by the three-dimensional recognition and perception module 20 to obtain the collision risk result of the goods inside the car. The graded warning module 30 and the linkage control module 40 execute prompts and control the elevator to perform avoidance operations according to the collision risk result, thereby improving the operation safety of the elevator and avoiding damage to transported goods and / or transported personnel.

[0069] In the embodiments of this application, the three-dimensional recognition and perception module 20 is used to acquire the initial spatial positioning data of the goods in the elevator car and several multi-source point cloud data of the elevator operation. The multi-source point cloud data includes the real-time operation data of the elevator and the real-time three-dimensional data and change data of the goods contained in the elevator car.

[0070] It should be noted that the 3D recognition and perception module 20 is used to collect multi-source point cloud data, including panoramic depth scanning, lateral laser array and top inertial vision combination, at multiple 3D sensing sampling positions at the elevator entrance and inside the car, and records the spatial positioning label (real-time position data) of each sampling point to form a multi-position difference perception dataset composed of several multi-source point cloud data in time series.

[0071] In the embodiments of this application, the three-dimensional recognition and perception module 20 includes a three-dimensional detection element, a visual detection element, and a position detection element.

[0072] It should be noted that the 3D detection element can be a laser 3D scanner. The laser 3D scanner is installed at the top center and inner sides of the elevator car wall panels (30-50cm from the top of the car). Two to three laser 3D scanners work together, with a scanning frequency of up to 100Hz, accurately capturing the length, width, and height of the cargo's 3D contour data with an error ≤5mm. It can penetrate light dust and is unaffected by light (suitable for nighttime and dimly lit environments), outputting the cargo's 3D coordinate data in real time. The visual detection element can be a high-definition visual camera (with depth sensing). The high-definition visual camera is installed at the top and bottom of the inner side of the car door and at the four corners of the car wall panels. Four to six high-definition visual cameras work in conjunction with the laser 3D scanner to achieve dual laser + visual verification, capturing cargo surface details, offset angles, and dynamic movement trajectories (such as cargo swaying or tilting), as well as cargo change data. Simultaneously, it identifies whether there are people lingering inside the car or foreign objects obstructing the view. The position detection element can be selected as a position sensor. The position sensor is installed at the bottom of the elevator car, the edge of the car door and both sides of the guide rail. It collects parameters such as the elevator's running height, lifting speed, car door opening and closing status (fully open / closed / half open), and car tilt angle in real time as real-time operating data, and transmits them synchronously to the intelligent analysis module 10 to provide data support for risk prediction.

[0073] In the embodiments of this application, the intelligent analysis module 10 is used to analyze all multi-source point cloud data to obtain collision risk results.

[0074] It should be noted that the intelligent analysis module 10 can use an industrial-grade embedded chip (main frequency ≥ 2.0 GHz), equipped with a preset 3D recognition algorithm and collision risk prediction rules to analyze all multi-source point cloud data to obtain collision risk results, so that the intelligent analysis module 10 integrates data receiving, processing, analysis and decision-making functions.

[0075] In the embodiments of this application, the linkage control module 40 is used to control the operation of the elevator and control the graded warning module 30 to output different warning signals based on the collision risk results.

[0076] It should be noted that the graded warning module 30 employs a multi-layered warning system combining sound, light, voice, and visual warnings. These are installed inside the elevator car, on the control panel, and in the waiting area outside the car. Different warning signals are output based on the risk level of the collision, ensuring that operators and relevant personnel can quickly perceive the risk. The linkage control module 40 synchronizes the warnings with the elevator's operation, preventing accidents caused by operators failing to respond to warnings in a timely manner.

[0077] Figure 2 This is a schematic diagram of the frame of the intelligent anti-collision recognition and warning device for an elevator according to another embodiment of this application.

[0078] like Figure 2 As shown in the embodiments of this application, the anti-collision intelligent identification and warning device of the elevator further includes a power supply and protection module 50 for supplying power and protecting the intelligent analysis module 10, the three-dimensional identification and perception module 20, the graded warning module 30 and the linkage control module 40.

[0079] It should be noted that the power supply and protection module 50 includes a power supply component and a protection component. The power supply component adopts a dual power supply mode with a main power supply and a backup power supply. The main power supply is connected to the original power supply system of the elevator (such as AC220V), and the backup power supply is a lithium battery (capacity ≥10000mAh), which can continue to work for ≥4 hours after a power outage, ensuring that the elevator's anti-collision intelligent recognition and warning device can still identify and warn normally in the event of a sudden power outage, avoiding the omission of potential hazards. The protection component is used to ensure that the intelligent analysis module 10, the three-dimensional recognition and perception module 20, the graded warning module 30, and the linkage control module 40 all have sealed shells that are waterproof, dustproof, and vibration-resistant (engineering plastic + silicone rubber sealing ring), suitable for the humid, dusty, and frequently starting and stopping working environment inside the elevator car; the exposed components such as sensors and warning lights are designed to resist collisions (the shell is made of high-strength aluminum alloy) to prevent damage from impacts with goods; and the wiring uses wear-resistant and interference-resistant cables with a hidden wiring design to avoid cable tangling and damage.

[0080] In this embodiment, the installation position of the 3D recognition and perception module 20 (laser scanner, camera, position sensor) is determined according to the dimensions of the elevator car wall panel to ensure no blind spots. The graded warning module 30 is installed inside the car, on the control panel, and in the waiting area outside the car. The intelligent analysis module 10, the linkage control module 40, and the power supply and protection module 50 are installed on the side of the car wall panel (without occupying cargo space). The wiring connecting the power supply and protection module 50, the 3D recognition and perception module 20, the graded warning module 30, and the linkage control module 40 to the intelligent analysis module 10 uses concealed wiring, is securely fixed, and avoids tangling and wear. The linkage control module 40 is connected to the elevator's main control system to ensure normal transmission of commands such as emergency braking and speed adjustment. All exposed components and wiring interfaces of the elevator's anti-collision intelligent recognition and warning device are sealed to ensure compliance with IP65 protection standards. The detection elements of the 3D recognition and perception module and the warning lights of the graded warning module are reinforced against collisions to prevent damage from cargo impacts.

[0081] In this embodiment, the intelligent anti-collision recognition and warning device of the elevator utilizes a 3D recognition and perception module 20 with laser + visual dual-sensing 3D precision recognition technology. This overcomes the limitations of traditional 2D recognition, accurately capturing the 3D contours and dynamic trajectories of goods with an error ≤5mm. It is unaffected by light or dust, has no blind spots, and is suitable for goods of various shapes and sizes (from small, scattered items to large, bulky items), improving the practicality of the elevator's intelligent anti-collision recognition and warning device. The device uses the collision risk prediction rules of the intelligent analysis module 10 combined with cargo position, elevator operating parameters (speed, height), and other multi-source point cloud data for dynamic calculation, predicting collision risks 0.5-1 second in advance, rather than issuing a warning after a collision occurs, allowing operators sufficient adjustment time. The device achieves low-risk alerts and high-risk forced braking through a graded warning module 30 and a linkage control module 40, deeply linking warnings with equipment operation. This avoids excessive intervention affecting operational efficiency while effectively preventing serious accidents, balancing efficiency and safety.

[0082] It should be noted that the intelligent anti-collision recognition and warning device of this elevator does not require large-scale modification of the elevator and can be directly applied to various specifications of vertical elevators (freight elevators, lifting platforms, etc.); it supports automatic calibration and adaptive parameter adjustment, adapting to different cargo types and different operating scenarios, and has strong versatility. This makes the intelligent anti-collision recognition and warning device of this elevator highly self-adaptive and compatible. Through the power supply and protection module 50, the intelligent anti-collision recognition and warning device of this elevator is designed with industrial-grade protection, which is waterproof, dustproof, and vibration resistant, adapting to the harsh working environment of the elevator; the dual power supply mode ensures that the intelligent anti-collision recognition and warning device of this elevator continues to work normally in the event of a sudden power outage, avoiding the omission of potential hazards; the fault self-diagnosis function facilitates rapid troubleshooting and repair, reducing maintenance costs.

[0083] In this embodiment, the anti-collision intelligent identification and warning device for the elevator is applicable to various types of vertical elevators, such as cargo elevators in industrial workshops, suitable for loading and unloading large, heavy, and loose goods, avoiding collisions between goods and car walls and guide rails, and protecting the safety of equipment and goods. For warehouse vertical transport elevators in warehousing and logistics (such as inter-floor cargo transfer), it solves the collision problems caused by cargo stacking misalignment and exceeding size limits, improving logistics transfer efficiency; for elevators transporting building materials (steel bars, cement, boards, etc.) at construction sites, it is suitable for the dusty and complex environment of construction sites, avoiding material collision damage while ensuring the safety of construction personnel; for vertical transport elevators of goods and equipment in supermarkets / office buildings, it avoids goods colliding with elevator car walls and doors, reducing equipment maintenance costs and improving operational safety.

[0084] This application provides an intelligent collision avoidance and warning device for an elevator, comprising an intelligent analysis module and a three-dimensional recognition and perception module, a graded warning module, and a linkage control module connected to the intelligent analysis module; the three-dimensional recognition and perception module is used to acquire initial spatial positioning data of goods in the elevator car and several multi-source point cloud data of the elevator operation, including real-time operating data of the elevator and real-time three-dimensional data and change data of goods contained in the elevator car; the intelligent analysis module is used to analyze all multi-source point cloud data to obtain collision risk results; the linkage control module is used to control the operation of the elevator and control the graded warning module to output different warning signals according to the collision risk results. The intelligent collision avoidance and warning device of this elevator achieves real-time three-dimensional recognition and dynamic risk prediction through intelligent analysis module, three-dimensional recognition and perception module, graded warning module and linkage control module, avoiding collision hazards in advance and reducing equipment damage, cargo loss and personnel safety risks; graded warning and linkage control do not affect the operation efficiency and ensure safety, and have safe and efficient functions; it solves the technical problem that existing elevators have no detection and warning of damage to transported goods and / or personnel, resulting in large cargo losses and low safety.

[0085] It should be noted that the intelligent anti-collision recognition and warning device of this elevator operates fully automatically without manual intervention. It is self-calibrating and self-adaptive, reducing the burden of manual operation and lowering the risk of accidents caused by misoperation, demonstrating a high degree of intelligence. This intelligent anti-collision recognition and warning device is compatible with various elevator specifications, offering strong versatility; it is simple to maintain, features self-diagnosis of faults, and facilitates rapid troubleshooting and repair. Through a modular design of intelligent analysis, 3D recognition and perception, graded warning, and linkage control modules, the device allows for individual module replacement as needed (e.g., module damage), resulting in low maintenance costs. It effectively reduces equipment repair and cargo damage costs caused by collisions, offering high long-term cost-effectiveness. The device achieves industrial-grade protection through power supply and protection modules, being waterproof, dustproof, and vibration-resistant, suitable for harsh working environments, with a long service life (designed service life ≥ 5 years), demonstrating strong environmental adaptability.

[0086] In one embodiment of this application, the intelligent analysis module 10 includes a data fusion processing submodule and a risk identification submodule; the data fusion processing submodule is used to preprocess, map and trend predict each multi-source point cloud data in sequence to obtain distance prediction data between the cargo and the car; the risk identification submodule is used to determine the risk by comparing the distance prediction data with a preset safety threshold to obtain the collision risk result.

[0087] It should be noted that the data fusion processing submodule receives real-time 3D data, change data, and elevator operation data transmitted from the 3D recognition and perception module 20, forming multi-source point cloud data. This data is then denoised, calibrated, and fused to generate a complete geometric mapping model of the 3D linkage between the cargo and the elevator car, eliminating the recognition error of a single sensor. Subsequently, based on the continuously collected multi-source point cloud data, distance prediction data is obtained according to the geometric mapping model. The risk identification submodule compares the distance prediction data output by the data fusion processing submodule with a preset safety threshold to determine the collision risk result, providing control data for the subsequent linkage control module 40.

[0088] Figure 3 This is a schematic diagram of the data fusion processing submodule in the anti-collision intelligent identification and warning device for elevators described in this application embodiment.

[0089] like Figure 3As shown, in one embodiment of this application, the data fusion processing submodule includes a preprocessing unit 110, a three-dimensional structure unit 120, a mapping unit 130, and a prediction estimation unit 140 connected in sequence. The preprocessing unit 110 is used to sequentially perform time synchronization, denoising, and normalization processing on multi-source point cloud data to obtain multi-source processed data; the three-dimensional structure unit 120 is used to generate a structure for the multi-source processed data according to the cargo's moving speed, attitude changes, and spatial position within the car using dynamic structure mapping to obtain a three-dimensional spatial mesh of the cargo with variable resolution; the mapping unit 130 is used to map the contour differences between the cargo and the car onto the three-dimensional spatial mesh to obtain a geometric mapping model; the prediction estimation unit 140 is used to input the geometric mapping model with continuous multi-source processed data to obtain continuous local geometric mapping data, and then predict the continuous local geometric mapping data to obtain distance prediction data.

[0090] It should be noted that the preprocessing unit 110 performs time synchronization processing on the multi-source point cloud data, mapping point cloud data of different sensor types to a unified time reference frame to eliminate morphological deviations caused by acquisition delays. It also performs denoising and normalization preprocessing on each multi-source point cloud data based on the unified time reference frame to eliminate high-frequency interference and low-frequency drift, and maintains consistency in spatial positioning data of each sampling point's changing data for subsequent dynamic structure mapping. The 3D structure unit 120 inputs the preprocessed multi-source data into the dynamic structure mapping algorithm, generating a variable-resolution 3D spatial mesh based on the cargo's moving speed, attitude changes, and spatial position, ensuring that the mesh resolution of different regions adapts to their morphological detail requirements. The mapping unit 130 performs cargo contour difference modeling on the variable-resolution 3D spatial mesh, establishing local geometric mappings for areas near the car walls, ceiling, and floor, respectively, to obtain geometric mapping models reflecting the structural differences at different locations. The prediction estimation unit 140 inputs continuous multi-source processed data into the geometric mapping model to obtain continuous local geometric mapping data that varies over time. It then inputs the continuous local geometric mapping data into the time-series feedback mechanism to predict the motion trend of the continuously collected cargo shape change sequence and outputs the estimated distance changes between the cargo and the car structure in each direction over a certain number of future times. The minimum estimated distance change value is used as the distance prediction data.

[0091] Figure 4 This is a schematic diagram of the preprocessing unit in the anti-collision intelligent identification and warning device for the elevator described in this application embodiment.

[0092] like Figure 4 As shown, in one embodiment of this application, the preprocessing unit 110 includes a time unification subunit 111, a noise reduction subunit 112, a drift correction subunit 113, a normalization subunit 114, a mapping subunit 115, and a registration optimization subunit 116 connected in sequence.

[0093] The time unification subunit 111 is used to perform time synchronization processing on all multi-source point cloud data to obtain time-unified processed data after time unification.

[0094] The noise reduction subunit 112 is used to perform high-frequency noise reduction on time-uniformly processed data using frequency domain filtering to obtain low-frequency processed data.

[0095] The drift correction subunit 113 is used to perform low-frequency offset compensation processing on the low-frequency processed data by surface fitting to obtain drift correction processed data.

[0096] The normalization subunit 114 is used to normalize the spatial scale of the drift correction data according to a preset three-dimensional measurement standard to obtain normalized data.

[0097] The mapping subunit 115 is used to perform positioning mapping processing on the initial spatial positioning data and the normalized processing data to obtain standardized positioning processing data.

[0098] The registration optimization subunit 116 is used to perform geometric alignment processing on the standardized positioning data by using iterative nearest point registration to obtain multi-source processing data.

[0099] It should be noted that the preprocessing unit 110 performs denoising and normalization preprocessing on multi-source point cloud data based on a unified time reference frame through the time unification subunit 111, denoising subunit 112, drift correction subunit 113, normalization subunit 114, mapping subunit 115, and registration optimization subunit 116. The aim is to eliminate high-frequency interference and low-frequency drift in sequence through an orderly multi-level data processing flow, retain effective three-dimensional spatial information, and maintain the consistency of spatial positioning data of each sampling point. This provides accurate and uniformly formatted multi-source processed data for subsequent dynamic structure mapping, which is the key data for constructing a high-precision three-dimensional spatial mesh for cargo and car linkage.

[0100] In this embodiment, the time unification subunit 111 performs time synchronization processing on all multi-source point cloud data, mapping each point cloud data in each multi-source point cloud data to a unified time reference frame to eliminate morphological deviations caused by acquisition delay.

[0101] In this embodiment, the noise reduction subunit 112 uses time-unified processing data based on a unified time reference frame to call a frequency domain filtering algorithm to suppress high-frequency interference in each point cloud data of the time-unified processing data, so as to obtain low-frequency processing data composed of low-noise point cloud data, and provide a stable spatial signal input for subsequent low-frequency drift correction.

[0102] It should be noted that frequency domain filtering algorithms can employ fast Fourier transform filtering, frequency domain smoothing filtering, and other techniques. Frequency domain smoothing filtering is a noise reduction technique in digital image processing that attenuates high-frequency components in the frequency domain. Its principle is based on converting the image to the frequency domain using Fourier transform, then using a low-pass filter to retain low-frequency information, and finally reconstructing a smooth image through inverse transform.

[0103] In this embodiment, the drift correction subunit 113 performs low-frequency drift compensation processing based on a polynomial trend surface on the low-frequency processed data. It corrects the global deformation offset generated during the sampling process through a surface fitting algorithm, so as to output the drift correction processed data after geometric morphology stabilization and establish an accurate geometric reference for normalization processing.

[0104] It should be noted that surface fitting algorithms can use methods such as least squares and moving least squares to correct scattered points.

[0105] In this embodiment, the normalization subunit 114 is based on drift correction processing data. It uses a spatial scale normalization algorithm to adjust the coordinate data of each sampling point in the drift correction processing data according to a preset three-dimensional measurement standard, so as to generate a normalized point cloud matrix with uniform size as normalized processing data, providing a consistent scale reference for the remapping of spatial positioning labels.

[0106] It should be noted that the spatial scale normalization algorithm is to spatially normalize the coordinates of the drift correction data (converting pixel / physical coordinates into relative scales).

[0107] In this embodiment, the mapping subunit 115 is a preset spatial mapping module that inputs the normalized point cloud matrix as normalized processing data. The spatial mapping module performs consistency correction on the spatial positioning data of each sampling point based on the mapping relationship between the original initial spatial positioning data and the coordinate system of the normalized processing data, so as to obtain standardized labeled point cloud data with the positioning data and coordinates aligned, thus laying the labeling foundation for the spatial registration optimization of the registration optimization subunit 116.

[0108] It should be noted that the spatial mapping module is set up according to requirements. The spatial mapping module can be set according to the actual application scenario, as long as it can achieve a one-to-one mapping between the initial data and the processed data.

[0109] In this embodiment, the registration optimization subunit 116 performs multi-source point cloud spatial registration optimization processing based on standardized positioning processing data. It uses the iterative nearest point registration (ICP) algorithm to refine the geometric alignment of different sensor location data, so as to output multi-source processing data with seamless spatial connection and consistent positioning, providing high-precision and consistent input for dynamic structure mapping.

[0110] It should be noted that the ICP algorithm (Iterative Closest Point) is a point set to point set registration method used to calculate the optimal rotation and translation transformation between two point cloud data sets to achieve accurate alignment.

[0111] In this embodiment, the preprocessing unit 110 is also used to perform quality consistency detection on the spatially seamless multi-source processed data. For example, by calculating the point cloud density distribution and noise ratio, quality evaluation parameters are generated, and low-quality regional point sets are filtered out to obtain high-quality multi-source fused point cloud data that meets robustness requirements, providing a reliable data foundation for subsequent dynamic structure mapping with variable resolution.

[0112] Figure 5 This is a schematic diagram of the frame of the three-dimensional structural unit in the anti-collision intelligent identification and warning device for the elevator described in the embodiments of this application.

[0113] like Figure 5 As shown, in one embodiment of this application, the three-dimensional structural unit 120 includes a matrix generation subunit 121, a network skeleton subunit 122, a partition adjustment subunit 123, a density adjustment subunit 124, and a verification subunit 125 connected in sequence.

[0114] The matrix generation subunit 121 is used to calculate the velocity vector and attitude quaternion based on the spatial positioning data of the multi-source processing data; and to generate the motion state matrix based on the cargo spatial position, velocity vector and attitude quaternion of all multi-source processing data.

[0115] The network skeleton subunit 122 is used to perform distributed fusion processing based on the initial spatial positioning data and motion state matrix to obtain a unified three-dimensional space reference frame mesh skeleton.

[0116] The partition adjustment subunit 123 is used to perform spatial partitioning transformation on the reference frame mesh skeleton according to the cargo velocity vector, attitude change rate and the relative position between the cargo and the car, to obtain a variable resolution matrix.

[0117] Density adjustment sub-unit 124 is used to perform mesh density adjustment on the reference frame mesh skeleton using a variable resolution matrix to obtain a multi-precision spatial mesh.

[0118] The verification subunit 125 is used to perform resampling and consistency verification on the multi-precision spatial mesh to obtain an optimized three-dimensional spatial mesh.

[0119] It should be noted that the 3D structural unit 120, through matrix generation subunit 121, network skeleton subunit 122, partition adjustment subunit 123, density adjustment subunit 124, and verification subunit 125, can input preprocessed source data into the 3D structural unit 120, generating a variable-resolution 3D spatial mesh based on the cargo's speed, attitude changes, and spatial position. This forms the basis for constructing a 3D linkage geometric mapping model between the cargo and the car. By dynamically adjusting the mesh resolution of different regions, key information details are preserved while ensuring computational efficiency, providing a high-precision spatial data foundation for subsequent cargo contour difference modeling, motion trend prediction, and collision risk assessment.

[0120] In this embodiment, the matrix generation subunit 121 uses spatial positioning data to calculate the cargo speed and attitude change parameters based on the multi-source processed data after denoising and normalization, so as to generate a motion state matrix containing cargo spatial position, velocity vector and attitude quaternion, providing time and space dual input conditions for the grid resolution adaptation of the mapping unit 130.

[0121] It should be noted that the matrix generation subunit 121 first uses a differential calculation method based on spatial positioning data to extract the position change of the cargo within a unified time reference frame, based on the three-dimensional coordinates and timestamps of continuous sampling points in the multi-source processed data after denoising and normalization. Then, a velocity vector solving algorithm is used to calculate the position change and sampling time interval of the multi-source processed data, realizing the instantaneous velocity vector calculation of the cargo and obtaining vector field data reflecting the dynamic motion direction and speed of the cargo. By using attitude calculation on the spatial positioning data and local surface normal vectors of the point cloud from the multi-source processed data, the attitude angle transformation of the overall attitude of the cargo is realized, and an attitude quaternion matrix covering roll, pitch, and yaw angles is generated. Through a motion state fusion algorithm, the velocity vector, attitude quaternion, and spatial position coordinates of the multi-source processed data are calculated, combining the temporal dynamic characteristics and spatial morphological characteristics into a unified data structure, and obtaining a motion state matrix containing position coordinates, velocity vectors, and attitude quaternions. Through the motion state matrix generation process, the multi-source processed data from the previous step is transformed into dual spatial and temporal input conditions, realizing the basic data for adapting the dynamic structure mapping unit to a 130-grid resolution. Attitude calculation refers to the real-time calculation of attitude angles by maintaining the direction cosine matrix or quaternion parameters using direction cosine matrices, quaternions, and equivalent rotation vectors. The direction cosine method directly updates nine matrix parameters, while the quaternion method converts the data into a direction cosine matrix after calculation using four parameters. The motion state fusion algorithm is a systematic method that uses heterogeneous data—including velocity vectors, attitude quaternions, and spatial position coordinates—obtained from multiple detection elements to perform more accurate and robust estimations of the cargo's position, velocity, attitude, acceleration, and other motion states.

[0122] For example, in a lift entrance scenario, a panoramic depth scanner with a sampling frequency of 20Hz, a lateral laser array with a sampling frequency of 50Hz, and a top inertial vision system with a sampling frequency of 30Hz are configured. After denoising and normalization, the coordinates of the continuous sampling points are (1.200, 0.850, 1.600)m and (1.250, 0.890, 1.605)m, with a sampling interval of 0.05s. The position change is calculated as (0.050, 0.040, 0.005)m using differential calculation. The velocity vector formula for the velocity vector solution algorithm is as follows:

[0123] ;

[0124] In the formula, The change in position is given by Δt, which is the time interval. The result is the velocity vector. (1.000, 0.800, 0.100) m / s. During the attitude quaternion calculation, the local surface normal vector is (0.0, 0.5, 0.866), which, after transformation by the direction cosine matrix, yields a roll angle of 2.5°, a pitch angle of 30°, and a yaw angle of 0°, with quaternions of (0.9659, 0.1294, 0.2241, 0.0). Combining the position coordinates, velocity vector, and attitude quaternions forms a motion state matrix data structure, fulfilling the temporal and spatial conditions required for mesh resolution adaptation. When the cargo approaches the car wall and the velocity vector exceeds 0.8 m / s, dynamic structure mapping increases the local mesh resolution to 1 cm, improving modeling accuracy and collision prediction accuracy to 98.5%.

[0125] In this embodiment, the network skeleton subunit 122 inputs the motion state matrix and initial spatial positioning data into the time and space mapping module. The time and space mapping module uses a distributed fusion computing method to construct a unified three-dimensional spatial reference frame mesh skeleton from the position coordinates, velocity vectors, attitude quaternions and initial spatial positioning data in the motion state matrix. This enables the unified construction of a multi-source point cloud spatial reference, ensuring that different sampling position data of goods in the car can be divided into variable resolution segments under the same spatial reference.

[0126] It should be noted that the distributed fusion computing method includes a multi-coordinate system consistency transformation algorithm, a distributed mesh mapping generation algorithm, a topology correction algorithm, and a node attribute fusion algorithm. The multi-coordinate system consistency transformation algorithm transforms the local coordinate system corresponding to the coordinates of each position in the motion state matrix, achieving accurate transformation of different detection data within a unified 3D spatial reference frame, and obtaining a multi-source point cloud dataset in the global coordinate system. The distributed mesh mapping generation algorithm processes the global coordinate system point cloud coordinates of the motion state matrix using a partitioning rule, mapping the multi-source point cloud data to the initial mesh skeleton nodes and generating a mesh skeleton data structure containing node indices and spatial coordinates. The topology correction algorithm corrects the node coordinates, connection edge sets, and maximum permissible deviation of the mesh skeleton data structure, achieving geometric and topological consistency correction of the node connection relationships in the mesh skeleton, and generating a structurally complete and geometrically conflict-free 3D spatial reference mesh skeleton. Furthermore, the node attribute fusion algorithm processes the dynamic attributes in the motion state matrix and the node indices of the mesh skeleton data structure, annotating dynamic characteristics such as cargo velocity vectors and attitude quaternions to the corresponding mesh skeleton nodes, and generating a unified 3D reference frame mesh skeleton with dynamic attribute annotations. The network skeleton subunit 122 transforms the motion state matrix and initial spatial positioning data from the previous step into a unified three-dimensional spatial reference frame mesh skeleton through distributed fusion computing and topology correction processing. This achieves the expected technical effect of variable resolution partitioning of multi-source acquired data under the same spatial reference. In this embodiment, the multi-coordinate system consistency transformation algorithm refers to a mathematical method that unifies spatial points or attitudes in multiple different reference systems (such as sensor coordinate system, world coordinate system, device local coordinate system, etc.) to the same target coordinate system, achieving precise alignment and consistent expression of spatial position and attitude. The distributed mesh mapping generation algorithm can adopt the mapping method in structured mesh generation or its enhanced variants (such as sub-mapping method, multi-block structured mesh, etc.). The topology correction algorithm can adopt a directed acyclic graph. The node attribute fusion algorithm refers to the organic combination of node attribute information (such as dynamic attributes in the motion state matrix and node indexes in the mesh skeleton data structure) with the three-dimensional spatial reference mesh skeleton (such as edge connection relationships, adjacency patterns) in the three-dimensional spatial reference mesh skeleton.

[0127] For example, in a scene at the entrance of an elevator, a motion state matrix has been generated, containing position coordinates (1.250, 0.890, 1.605) m, velocity vector (1.000, 0.800, 0.100) m / s, and attitude quaternions (0.9659, 0.1294, 0.2241, 0.0). Spatial positioning data includes the entrance panoramic scanner number #001 and the laser array number on the left side of the elevator car number #002. A distributed fusion computing method calculates the local coordinate system transformation matrices of #001 and #002 as T001 and T002 respectively. A multi-coordinate system consistency transformation algorithm is used to perform coordinate system unification, generating a global coordinate system point cloud dataset. A distributed mesh mapping generation algorithm sets the initial side length of the mesh cell to 5cm, maps the global coordinate system data to the mesh nodes, and establishes a node index table and a spatial coordinate table. A topology correction algorithm detects a maximum node connection error of 1.2cm, which is lower than the set maximum allowable deviation of 2cm, completing geometric consistency correction. The node attribute fusion algorithm labels velocity vectors and attitude quaternions to corresponding node indices, outputting a unified 3D reference frame mesh skeleton with dynamic attributes. The fused reference frame mesh skeleton can complete variable resolution partitioning within 0.05s, improving the accuracy of local collision risk prediction to 97.8%.

[0128] In this embodiment, the partition adjustment subunit 123, based on the reference frame mesh skeleton in three-dimensional space, calls the partition adaptation unit of dynamic structure mapping, calculates the dynamic attributes of the nodes, the cargo velocity vector, the attitude quaternion and the relative position coordinates according to the magnitude of the cargo velocity vector, the rate of attitude change and the relative position relationship between the cargo and the car wall panel, ceiling and floor, to obtain the resolution adjustment coefficient of each mesh unit in the reference frame mesh skeleton, so as to form a variable resolution matrix.

[0129] It should be noted that the process of obtaining the variable resolution matrix includes:

[0130] The velocity vector magnitude of the cargo is processed by velocity threshold grading according to the velocity grading boundary to realize the velocity partitioning determination of the dynamic state of the cargo, and to obtain the velocity perturbation factor matrix composed of the velocity perturbation factors corresponding to each grid cell in the reference frame mesh skeleton;

[0131] The algorithm performs attitude change rate analysis on the attitude quaternion time series and time sampling interval of the reference frame mesh skeleton, realizes the quantitative calculation of attitude change frequency, and generates an attitude perturbation factor matrix composed of the attitude perturbation factors corresponding to each mesh cell in the reference frame mesh skeleton.

[0132] The relative positional relationship of the spatial coordinates of the grid nodes of the reference frame mesh skeleton is evaluated using the reference plane equation of the car wall panel / ceiling / floor. This enables the calculation of the shortest Euclidean distance between each node and the car structure, and generates a spatial proximity factor matrix composed of the spatial proximity factors corresponding to each grid unit in the reference frame mesh skeleton.

[0133] The variable resolution calculation formula, derived from multi-factor fusion computation, processes the velocity perturbation factor matrix, attitude perturbation factor matrix, spatial proximity factor matrix, and fusion weight coefficients to achieve unified calculation of resolution adjustment coefficients and output a variable resolution matrix. Through multi-factor fusion calculation using partitioned adaptation units, the dynamic attributes of the previous 3D reference frame mesh skeleton are transformed into variable resolution matrices for different spatial partitions, achieving the expected technical effect of intelligent adaptation of mesh accuracy. The variable resolution calculation formula is as follows:

[0134] Radj(i,j,k)=wv×Fv(i,j,k)+wa×Fa(i,j,k)+wp×Fp(i,j,k);

[0135] Where Radj(i,j,k) is the resolution adjustment coefficient of grid cell (i,j,k), Fv(i,j,k), Fa(i,j,k), and Fp(i,j,k) are the velocity perturbation factor, attitude perturbation factor, and spatial proximity factor of grid cell (i,j,k), respectively, and wv, wa, and wp are the corresponding fusion weight coefficients.

[0136] For example, in a cargo-carrying scenario of a lifting platform, the dynamic attributes of the nodes in the generated reference frame mesh skeleton include a velocity vector of (1.200, 0.900, 0.100) m / s, an attitude quaternion of (0.9659, 0.1294, 0.2241, 0.0), and a node spatial coordinate distance of 0.18m from the left wall panel. The velocity threshold grading boundaries are set to 0.5m / s and 1.0m / s, and the velocity modulus is calculated as follows: The corresponding velocity perturbation factor is Fv=1.0. The attitude change rate is calculated by dividing the modulus of the difference between adjacent sampled quaternions by the sampling time interval of 0.05s, resulting in 0.08 rad / s, corresponding to an attitude perturbation factor Fa=0.4. The spatial proximity factor is set with a distance threshold of 0.2m and a current distance of 0.18m, corresponding to a proximity factor Fp=0.9. The fusion weight coefficients are set to wv=0.5, wa=0.3, and wp=0.2, respectively. Substituting these values ​​into the variable resolution calculation formula, the resolution adjustment coefficient Radj=0.5×1.0+0.3×0.4+0.2×0.9=0.79 is obtained. Based on this resolution adjustment coefficient, the side length of the mesh unit in this area is reduced from the original 5cm to 2.5cm, improving the modeling accuracy and increasing the collision warning accuracy of this area to 99.1%.

[0137] In this embodiment, the density adjustment subunit 124 uses a variable resolution matrix to adjust the mesh density of the reference frame mesh skeleton in three-dimensional space. The mesh density is increased in areas where the cargo contour changes drastically or is near the car structure, and decreased in areas where the shape is stable or far from the structure, thereby generating a multi-precision spatial mesh. This can be understood as the density adjustment subunit 124 using a mesh density adjustment algorithm with a variable resolution matrix to process the reference frame mesh skeleton in three-dimensional space, achieving density optimization of each mesh unit in the reference frame mesh skeleton.

[0138] It should be noted that the process of obtaining multi-precision spatial grids includes:

[0139] High-density region recognition processing is performed on the resolution adjustment coefficient threshold of the variable resolution matrix to achieve the recognition of areas with drastic changes in cargo outline or adjacent to the car structure, and to generate a high-density region marker set corresponding to the variable resolution matrix.

[0140] Local mesh refinement is performed on the side length of the mesh cells in the high-density region marker set and the reference frame mesh skeleton to proportionally reduce the side length of the mesh cells in the high-density region and obtain the refined local mesh dataset.

[0141] Low-density region identification processing is performed on the variable resolution matrix and the set lower limit threshold of the resolution adjustment coefficient to achieve region identification that is morphologically stable or far from the car structure, and to generate a low-density region marker set corresponding to the variable resolution matrix.

[0142] Local mesh coarsening is performed on the side lengths of the grid cells in the low-density region marker set and the reference frame mesh skeleton to proportionally enlarge the side lengths of the grid cells in the low-density region and obtain the coarsened local mesh dataset.

[0143] The system performs mesh density synthesis processing on all mesh datasets, including the refined mesh dataset, the coarse mesh dataset, and the reference frame mesh skeleton, to achieve global synthesis of multi-precision spatial meshes and output multi-precision spatial meshes composed of spatial data with mixed local high-density and low-density structures.

[0144] For example, in a cargo-carrying scenario of a lift, nodes with a resolution adjustment factor greater than 0.8 in the variable resolution matrix are identified as high-density regions, and nodes with a resolution factor less than 0.3 are identified as low-density regions. In high-density regions, the reference frame mesh skeleton has a mesh cell side length of 5cm, which is reduced to 2cm through local mesh refinement; in low-density regions, the reference frame mesh skeleton has a mesh cell side length of 5cm, which is enlarged to 8cm through local mesh coarsening. During local mesh refinement, the high-density region marker set is input into the local mesh refinement process, and the formula for proportional reduction is:

[0145] ;

[0146] In the formula, Lnew is the side length of the scaled-down mesh element, Lorig is the original side length of the mesh element in the reference frame mesh skeleton, and α is the scaling ratio, which, when set to 0.6, results in a side length of 2 cm. During local mesh coarsening, the low-density region marker set is input into the local mesh coarsening process, and the formula for scaling up is as follows:

[0147] Lnew = Lorig × (1 + β);

[0148] In the formula, β is the magnification ratio, which is set to 0.6 to obtain a side length of 8cm. The mesh density synthesis process is to fuse the refined and coarsened mesh data to generate a global multi-precision spatial mesh, which can be rendered within 0.04s. It improves the contour recognition details in the high-density area of ​​the reference frame mesh skeleton by 15%, reduces the modeling computation in the low-density area by 40%, and keeps the overall collision risk prediction accuracy within 99%.

[0149] In this embodiment, the verification subunit 125 performs data resampling and topology consistency verification on the multi-precision spatial mesh to ensure that the node connection relationships between various mesh cells in the multi-precision spatial mesh are consistent with the geometric continuity of the original point cloud data (such as multi-source point cloud data), and outputs a structure-optimized variable-resolution 3D spatial mesh, providing a high-precision and computationally efficient spatial data foundation for subsequent cargo contour difference modeling. The verification subunit 125 uses multi-resolution resampling based on node position to process the node coordinate set and variable-resolution matrix of the multi-precision spatial mesh, realizing the unified processing of node sampling frequency within mesh cells of different resolutions, and outputting the mesh node resampled coordinate set.

[0150] It should be noted that the process of constructing a three-dimensional spatial mesh includes:

[0151] Interpolation reconstruction is performed on the resampled node coordinate set of the multi-precision spatial grid, the grid cell side length and interpolation weight of the reference frame grid skeleton, so as to realize the interpolation reconstruction of the spatial coordinates of the resampled nodes in the global unified coordinate system and generate resampled grid structure data with enhanced geometric continuity.

[0152] A topology consistency verification process is performed on the resampled grid structure node set, connection edge set, and maximum topology error threshold of the multi-precision spatial grid to determine the geometric consistency and topology integrity of the grid node connection relationship and generate a consistency verification state matrix.

[0153] The system performs topology repair on the consistency check state matrix and the node coordinates and connection edge set of the multi-precision spatial grid, enabling automatic repair of node connections that do not meet the consistency conditions, including edge length adjustment, node coordinate fine-tuning and connection reconstruction, and outputs the grid node set corresponding to the multi-precision spatial grid after topology repair.

[0154] Boundary constraint preservation processing is applied to the original point cloud boundary conditions of the mesh node set and the reference frame mesh skeleton to align the repaired mesh boundary with the original point cloud boundary, ensuring the consistency between the mesh and the original geometric contour, and generating a structure-optimized variable-resolution 3D spatial mesh. Specifically, through data resampling and topology consistency verification, the multi-precision spatial mesh generated by density adjustment sub-unit 124 is transformed into a 3D spatial mesh with a spatial data structure possessing high precision, geometric continuity, and topological integrity. This achieves the expected technical effect of providing high-precision and computationally efficient input data for subsequent cargo contour difference modeling.

[0155] For example, in a cargo-carrying scenario of an elevator, the multi-precision spatial mesh contains 15,000 nodes. The cell side length is 2cm in high-density areas and 8cm in low-density areas. The multi-resolution resampling algorithm sets a uniform sampling frequency of at least 5 nodes per cell side length and outputs a resampled coordinate set. Interpolation reconstruction can use cubic spline interpolation, with interpolation weights calculated based on Euclidean distance distribution. The average node spacing error after reconstruction is 0.35mm. Topology consistency verification sets a maximum topology error threshold of 1cm, detecting 420 node connections that do not meet the criteria. Topology repair performs side length adjustments on these connections using the following formula:

[0156] ;

[0157] Where Ladj is the adjusted mesh side length, Lorig is the original mesh side length, Ltarget is the target mesh side length, and γ is the adjustment coefficient. When set to 0.5, the corresponding repaired side length error is reduced to 0.4mm. Boundary constraint preservation aligns the repaired nodes with the original point cloud contour, improving the boundary coincidence from 98.2% to 99.7%. The output structurally optimized variable-resolution 3D spatial mesh reduces modeling time to 82% of the original in subsequent cargo contour difference modeling, and improves collision risk prediction accuracy to 99.3%.

[0158] Figure 6 This is a schematic diagram of the mapping unit in the anti-collision intelligent identification and warning device for the elevator described in this application embodiment.

[0159] like Figure 6As shown, in one embodiment of this application, the mapping unit 130 includes an identification processing subunit 131, a separation subunit 132, a feature extraction subunit 133, an association mapping subunit 134, and a region synthesis subunit 135 connected in sequence.

[0160] The identification processing subunit 131 is used to perform region identification processing on each grid cell of the three-dimensional spatial grid to obtain a spatial grid model with structural difference markers;

[0161] Separation subunit 132 is used to separate and extract the contour region of the spatial grid model by point cloud clustering segmentation to obtain contour point cloud data;

[0162] The feature extraction subunit 133 is used to extract features from the contour point cloud data using normal vectors and surface fitting to obtain geometric feature data of each region.

[0163] The association mapping subunit 134 is used to perform association mapping processing between the geometric feature data of each region and the spatial grid model to obtain the region geometric mapping model corresponding to the geometric feature data of each region.

[0164] The region synthesis subunit 135 is used to perform multi-region synthesis processing on all region geometric mapping models to obtain a geometric mapping model.

[0165] It should be noted that the mapping unit 130 performs cargo contour difference modeling on a variable resolution three-dimensional spatial mesh. The aim is to construct independent local geometric mapping models for different locations such as car wall panels, ceiling and floor based on the output of the preceding dynamic structural mapping, so as to reflect the structural difference characteristics of each region, thereby providing refined and location-related three-dimensional morphological data support for subsequent time-series feedback and motion trend prediction.

[0166] In this embodiment, the identification processing subunit 131 performs region identification processing on the grid cells of the three-dimensional spatial grid based on a variable resolution, in order to identify the spatial partition labels of the car wall panels, ceiling, and floor, and obtain a spatial grid model with structural difference markers, providing a clear spatial location basis for subsequent local geometric feature extraction. The identification processing subunit 131 can perform spatial region identification on the global coordinates of the grid nodes of the three-dimensional spatial grid, realize the spatial location determination of the grid cells, and generate a preliminary region label matrix.

[0167] It should be noted that the process of obtaining the spatial grid model includes:

[0168] The global coordinates of the grid nodes in the three-dimensional spatial mesh are processed by plane distance determination based on the set reference plane equation and distance threshold. This enables the calculation of the nearest Euclidean distance between each node and the corresponding car structure. Based on the distance threshold, the nodes are assigned to wall panel, ceiling or floor area labels to obtain a refined area label matrix corresponding to the three-dimensional spatial mesh.

[0169] Based on the refined region label matrix and the set of node connection edges of the 3D spatial mesh, the continuity of the region boundary is checked to achieve the continuity and integrity of the node connection relationship within the same region, and a boundary consistency state set corresponding to the 3D spatial mesh is generated. Inconsistent node labels are corrected.

[0170] Based on the boundary consistency state set and the original region label matrix composed of multi-source point cloud data, label mapping optimization is performed to achieve consistent mapping between corrected node labels and the global coordinate system, and an optimized region label mesh model with structural difference markers is output.

[0171] Based on the optimized region-labeled mesh model and the variable-resolution mesh dataset of the 3D spatial mesh, region label embedding processing is performed to obtain a spatial mesh model with structural difference markers. This embeds the structural difference markers into the mesh cell data structure, providing a clear spatial location basis for subsequent local geometric feature extraction. Specifically, through region labeling processing, the variable-resolution 3D spatial mesh obtained by optimizing the 120 3D structural units is transformed into a spatial mesh model with labeled mesh data for the car wall panels, ceiling, and floor, achieving the expected technical effect of precise spatial location positioning required for local geometric feature extraction.

[0172] For example, in a cargo-carrying scenario of a lifting platform, the structure-optimized variable-resolution 3D spatial mesh contains 15,000 nodes. The wall panel reference plane equations are x=0.000 and x=2.500m, the ceiling reference plane equation is z=2.300m, and the floor reference plane equation is z=0.000m. The distance threshold is set to 0.05m. Using Euclidean distance, the distance from node (0.040, 1.200, 1.600)m to the left wall panel at x=0.000m is calculated to be 0.040m, and it is classified as a wall panel area. The distance from node (1.200, 0.850, 2.280)m to the ceiling plane is 0.020m, and it is classified as a ceiling area. The distance from node (1.250, 1.000, 0.030)m to the floor plane is 0.030m, and it is classified as a floor area. The boundary continuity check detected 15 locally broken node sets in the wall panel area. Labels were optimized and corrected through label mapping while maintaining alignment with the global coordinate system. The final output spatial mesh model of the area labels achieved a regional positioning error of less than 2mm in subsequent local geometric feature extraction, improving the local geometric mapping accuracy to 98.9%.

[0173] In this embodiment of the application, the separation subunit 132 uses a point cloud clustering segmentation algorithm to separate the cargo outline point cloud in each target area for the spatial grid model with structural difference markers, so as to obtain the outline point cloud data composed of the cargo outline point cloud set in the area, thereby providing a clear boundary and range for modeling the local geometric features of the area.

[0174] It should be noted that the separation sub-unit 132 processes the global coordinates of nodes, the region label matrix, and the number of cluster categories of the spatial grid model using a point cloud clustering segmentation algorithm to achieve spatial density analysis and category generation of the cargo outline point cloud within the region. The process of obtaining the outline point cloud data includes:

[0175] Based on the local density values ​​of the point cloud in the spatial grid model and the preset density threshold, the density threshold determination process is performed within the region to realize the density compliance check of the cluster category and filter the point cloud subset that meets the contour features.

[0176] Based on the region label matrix of the spatial grid model, the coordinates of the point cloud subsets, and the set boundary conditions, boundary constraint segmentation is performed to achieve boundary clipping of the point cloud subsets, remove irrelevant points that cross the region boundary, and generate a set of contour point clouds composed of point cloud subsets within the boundary range that meet the set boundary conditions.

[0177] Connected components are extracted based on the contour point cloud set and the set connection determination threshold, thereby enhancing the connectivity of the contour point cloud set and reconstructing the region integrity, and obtaining a region contour point cloud set composed of connected component contour point cloud data with connected component labels.

[0178] Based on the connected component contour point clouds of the regional contour point cloud set, and the set smoothing weight coefficients and iteration numbers, contour boundary smoothing is performed to achieve geometric continuity adjustment of the contour boundaries, reduce boundary noise fluctuations, and output contour point cloud data composed of all regional contour point cloud data that match geometric consistency and structural differences. Specifically, through point cloud clustering and boundary connectivity processing, the spatial grid model with structural difference markings obtained from the identification processing subunit 131 is transformed into contour point cloud data with clear boundaries and ranges. The contour point cloud data includes cargo contour point cloud data and car contour point cloud data. The car contour point cloud data includes contour point cloud data of the wall panel area, ceiling area, and floor area, achieving the expected technical effect of data accuracy and spatial isolation required for local geometric feature modeling.

[0179] For example, in a cargo-carrying scenario of an elevator, the spatial mesh model with structural difference markers has 15,000 nodes. The region label matrix labels the number of nodes in the wall panel region as 4,500, the ceiling region as 3,000, and the floor region as 7,500. For instance, the separation sub-unit 132 uses the wall panel region as the basis for obtaining the contour point cloud data of the wall panel. During the point cloud clustering and segmentation process, the number of cluster categories is set to 3. The point cloud of the wall panel region is judged for density according to the set density threshold of 0.015 points / mm³, and sparse categories below the density threshold are removed. During the boundary constraint segmentation process, the boundary conditions of the wall panel region are set to x=0.000m and x=2.500m. Irrelevant points crossing the boundary are clipped, and the number of contour point clouds retained in the wall panel region is 4,200. During the connected component extraction process, the connection determination threshold is set to 5mm, and a continuous and complete set of regional contour point clouds is extracted. The average node spacing within the connected component is 3.8mm. During the contour boundary smoothing process, with a smoothing weight coefficient of 0.3 and 5 iterations, the average change in boundary curvature after smoothing is reduced to 12% of the original value. The output contour point cloud data of the wall panel area has complete and continuous boundaries and a clear spatial range. Similarly, the curvature calculation accuracy of the goods in the elevator in the subsequent local geometric feature modeling is improved to 99.0%.

[0180] In this embodiment, the feature extraction subunit 133 extracts local geometric feature parameters, including curvature distribution, surface normal direction, and local concavity / convexity, based on the contour point cloud data within the elevator area and employing normal vector calculation and surface fitting algorithms. This generates a regional local geometric feature dataset composed of geometric feature data from various regions, providing quantified morphological features for structural difference mapping. Specifically, a normal vector set is obtained by calculating the normal vectors of the contour point cloud data based on the point cloud coordinates and the local neighborhood radius of the point cloud, thus realizing the extraction function of the local surface normal direction for each point. Further, curvature estimation is performed based on the normal vectors of the normal vector set and the local neighborhood radius of the contour point cloud data to calculate the curvature distribution of the local geometric shape and obtain the curvature matrix as a geometric feature index.

[0181] It should be noted that the process of obtaining geometric feature data for each region includes:

[0182] A set of normal vectors is obtained by calculating the normal vectors of the point cloud coordinates and the local neighborhood radius of the point cloud data.

[0183] Curvature estimation is performed based on the normal vectors of the normal vector set and the local neighborhood radius of the point cloud data of the contour point cloud data to obtain a curvature matrix composed of various curvatures.

[0184] Local surface fitting is performed based on the contour point cloud data of each region and the set order of the fitting polynomial to obtain the fitting residuals and surface equation coefficients corresponding to the contour point cloud data of each region; surface morphology fitting is achieved, and a fitting feature dataset consisting of several fitting residuals and surface equation coefficients is generated.

[0185] The concavity and convexity are quantified based on the curvature of each curvature of the curvature matrix and the fitting residual of the fitted feature dataset to obtain the concavity and convexity corresponding to each curvature; the numerical calculation of local concavity and convexity features is realized to obtain a concavity and convexity matrix composed of several concavity and convexity, with one concavity and convexity corresponding to one point cloud node.

[0186] Based on the normal vector set, curvature matrix, and concavity / convexity matrix, feature integration processing is performed on each region to obtain the corresponding regional geometric feature data. This achieves unified integration processing of multiple local geometric features and outputs a regional local geometric feature dataset composed of all regional geometric feature data as the input basis for structural difference mapping. Through normal vector calculation and surface fitting processing, the contour point cloud data of all regions obtained from the separated subunit 132 are transformed into quantified morphological feature data with curvature distribution, surface normal direction, and local concavity / convexity, achieving the expected technical effect of data preparation for structural difference mapping.

[0187] For example, in a cargo-carrying scenario of a lift, if the number of nodes in the contour point cloud data of the wall panel region is 4200, and the local neighborhood radius of the point cloud is set to 5mm during the normal vector calculation process, the normal direction of each node is extracted and normalized. The curvature calculated during curvature estimation ranges from 0.005 to... This forms a curvature matrix. The surface fitting algorithm uses a second-order polynomial for fitting, with a mean fitting residual of 0.35 mm. The equation coefficient set is used for morphological feature analysis. In the process of convexity quantification, the concavity / convexity calculation formula is:

[0188] B i =k i ×(1+e i );

[0189] Among them, B i Let k be the concavity / convexity of the i-th node. i Let e ​​be the curvature value of the i-th node. i Let be the fitting residual ratio of the i-th node; if the curvature value The fitting residual is 0.35, and the fitting residual ratio = local neighborhood radius of the point cloud / fitting residual = 0.35 / 5 = 0.07, thus obtaining the concavity / convexity. In the feature integration process, the normal vector set, curvature matrix, and concavity / convexity matrix are mapped to a unified data structure. The output geometric feature data of each region can achieve a curvature error of less than 1.5% and a direction vector error of less than 2° in the subsequent structural difference mapping, thereby improving the collision warning recognition accuracy to 99.2%.

[0190] In this embodiment, the association mapping subunit 134 performs association mapping processing on the geometric feature data of each region and the structural difference markers of the spatial grid model to construct a regional geometric mapping model (including a wall panel regional geometric mapping model, a ceiling regional geometric mapping model, and a floor regional geometric mapping model) corresponding to the car wall panel, ceiling, and floor, so as to reflect the spatial relationship between the structural constraints of each region and the morphological characteristics of the cargo.

[0191] In this embodiment, the regional synthesis subunit 135 performs multi-region synthesis processing on the geometric mapping models of each region, and merges the local mapping results at different locations into a cargo contour difference modeling output that has both global consistency and local differences, so as to obtain a geometric mapping model to provide time-series feedback for subsequent motion trend prediction and distance change estimation.

[0192] Figure 7 This is a schematic diagram of the frame of the prediction and estimation unit in the intelligent collision avoidance and warning device for elevators according to an embodiment of this application.

[0193] like Figure 7As shown, in one embodiment of this application, the prediction estimation unit 140 includes a sequence matrix subunit 141, a dynamic extraction subunit 142, a prediction trend subunit 143, a distance calculation subunit 144, and a distance correction subunit 145 connected in sequence.

[0194] The sequence matrix subunit 141 is used to extract the geometric features of cargo posture changes in continuous multi-source processing data according to the geometric mapping model, and obtain a morphological change sequence matrix composed of a unified time series index.

[0195] Dynamic extraction subunit 142 is used to extract dynamic attributes from the morphological change sequence matrix to obtain a dynamic attribute matrix;

[0196] The prediction trend subunit 143 is used to predict the movement trajectory of goods in the three-dimensional space of the geometric mapping model by using the dynamic attribute matrix as the multidimensional time feature of the multivariate long short-term memory network, and to obtain the position sequence matrix of the predicted position of the goods.

[0197] The distance calculation subunit 144 is used to calculate the shortest distance between the predicted location of the goods and the car based on the Euclidean distance using the position sequence matrix, and to obtain distance prediction data.

[0198] The distance correction subunit 145 is used to perform weighted correction on the distance prediction data to obtain the corrected distance prediction data.

[0199] It should be noted that the prediction and estimation unit 140 maps the geometric mapping model into the time-series feedback mechanism to predict the motion trend of the cargo attitude change sequence of continuously acquired multi-source processing data, and outputs the distance change estimates between the cargo and the car structure in each direction for several future time steps as distance prediction data. Specifically, the prediction and estimation unit 140 can also be understood as mapping the geometric mapping model into the time-series feedback mechanism, performing time-series modeling and dynamic parameter analysis on the cargo attitude change sequence of continuously acquired multi-source processing data, and combining the acquired dynamic attributes such as the cargo's spatial position, velocity vector, and attitude quaternions to predict the distance change trend between the cargo and the car structure in each direction for several future time steps. The prediction and estimation unit 140, based on the variable resolution three-dimensional spatial mesh and the local geometric features of the geometric mapping model, uses a time-series prediction algorithm to quantify potential collision risks in advance, providing high-precision dynamic distance change estimates for subsequent risk assessment and early warning execution, thus serving as the prediction link for collision prevention.

[0200] In the embodiments of this application, the sequence matrix subunit 141 is based on the local geometric feature data of the region generated by the geometric mapping model and the corresponding initial spatial positioning data. It performs time index rearrangement and structured caching on the cargo posture data in several continuously collected multi-source processing data to form a morphological change sequence matrix with a unified time series index, providing a time-consistent input data structure for subsequent dynamic parameter extraction.

[0201] In the embodiments of this application, the dynamic extraction subunit 142 performs dynamic attribute extraction processing on the morphological change sequence matrix, calls the existing velocity vector analysis algorithm and attitude quaternion calculation algorithm to calculate the three-dimensional coordinate difference vector and attitude change amount between consecutive sampling times, and outputs a dynamic attribute matrix containing position change amount, velocity vector sequence and attitude quaternion sequence, providing accurate motion state input for time series modeling.

[0202] In the embodiments of this application, the trend prediction subunit 143 inputs the dynamic attribute matrix into the temporal modeling unit. The temporal modeling unit uses a multivariate long short-term memory network (LSTM) for prediction. The training parameters of the LSTM incorporate the spatial position change of the cargo, the velocity vector sequence, and the attitude quaternion sequence as multidimensional time step features. It performs trend modeling on the motion trajectory of the cargo in three-dimensional space, generates a future predicted position sequence matrix, and provides temporal prediction results for distance change calculation.

[0203] In the embodiments of this application, the distance calculation subunit 144 is based on the future predicted position sequence matrix and the reference plane equations of the car wall panels, ceiling and floor. It uses three-dimensional Euclidean distance to calculate the shortest spatial distance between the predicted position at each future time step and each car structure as distance prediction data, and generates a future distance change sequence matrix to reflect the spatial approach trend in different directions.

[0204] In the embodiments of this application, the distance correction subunit 145 performs dynamic distance gradient analysis on the future distance change sequence matrix, uses a differential gradient calculation algorithm to extract the distance change rate of continuous sampling time, and combines the velocity vector direction of the corresponding sampling time in the dynamic attribute matrix to perform weighted correction on the distance shortening rate in different directions, and outputs the direction-sensitive corrected distance prediction data. The corrected future distance change sequence matrix is ​​recorded as the future distance change estimation matrix, which serves as the direct input data for the subsequent risk assessment step.

[0205] It should be noted that the differential gradient calculation algorithm is a numerical method that uses the differential approximation derivative to estimate the gradient vector (a vector composed of partial derivatives in each direction) of a function (especially a discrete or non-differentiable function) at a certain point.

[0206] In one embodiment of this application, the risk identification submodule is further configured to determine the collision risk result as no collision risk if the distance prediction data is greater than the upper limit of the preset safety threshold; determine the collision risk result as high collision risk if the distance prediction data is less than the lower limit of the preset safety threshold; and determine the collision risk result as low collision risk if the distance prediction data is between the upper and lower limits of the preset safety threshold.

[0207] It should be noted that the risk identification submodule determines risk based on distance prediction data and a preset safety threshold. If the distance prediction data in a certain direction rapidly decreases within the preset safety threshold, a collision risk assessment signal is generated. In this embodiment, the risk identification submodule can also determine a high collision risk based on abnormal situations such as the car door not being closed during lifting or lowering, or people remaining inside the car.

[0208] In other embodiments, the risk identification submodule can also collect the tilt angle and / or sway amplitude of the goods and compare them with corresponding preset angle thresholds and preset sway amplitude thresholds to obtain a collision risk result; for example: if the tilt angle is greater than the preset angle threshold and / or the sway amplitude is greater than the preset sway amplitude threshold, the collision risk result is high collision risk; if the tilt angle is less than the preset angle threshold and / or the sway amplitude is less than the preset sway amplitude threshold, the collision risk result is low collision risk; if the tilt angle is less than 1° and / or the sway amplitude is less than 4cm, the collision risk result is no collision risk. The preset angle threshold can be selected as 5°. The preset sway amplitude threshold can be 10cm.

[0209] In one embodiment of this application, the linkage control module 40 is further configured to control the elevator to operate normally if the collision risk result is no collision risk; control the elevator to operate normally if the collision risk result is low collision risk, and control the graded warning module to output an adjustment cargo warning signal; and control the elevator to stop working if the collision risk result is high collision risk, and control the graded warning module 30 to output an adjustment cargo and danger warning signal.

[0210] It should be noted that the linkage control module 40 inputs the collision risk results into the elevator control execution unit, triggering corresponding multi-level early warning strategies, including audible and visual alerts, automatic door opening, and elevator speed adjustment, to prevent collisions between goods and the elevator car. In this embodiment, the elevator's intelligent collision avoidance recognition and warning device, during execution, uses collision risk results, distance prediction data, and raw multi-source point cloud data to form a collision risk event dataset for offline analysis and adaptive optimization. The linkage control module 40 can also adaptively optimize the collision risk event dataset, readjusting the dynamic structure mapping and variable resolution mesh parameters according to different locations and structural differences, to improve the accuracy and algorithm robustness of the three-dimensional spatial mesh for future three-dimensional linkage between goods and the elevator car.

[0211] In the embodiments of this application, if the collision risk result is no collision risk, the linkage control module 40 controls the warning light of the graded warning module 30 to be constantly green and there is no voice prompt. The operation panel of the linkage control module 40 shows normal operation, and the visualization screen (if present) displays the three-dimensional model of the cargo and the safe distance.

[0212] In the embodiments of this application, if the collision risk result is low (the cargo has slightly shifted, and the distance prediction data is close to the preset safety threshold), the linkage control module 40 controls the warning light of the graded warning module 30 to flash yellow (frequency 1 time / second), and provides a voice prompt that the cargo has slightly shifted and should be adjusted. The operation panel of the linkage control module 40 displays the low risk and the direction of the shift, and the yellow warning light in the waiting area outside the car flashes synchronously. The linkage control module 40 controls the elevator to reduce its lifting speed (to 50% of the normal speed), restricts the car door from closing, and automatically restores the normal operating speed and car door function after the operator adjusts the cargo position and eliminates the potential hazard.

[0213] In the embodiments of this application, the linkage control module 40, based on a collision risk result indicating a high collision risk (clear collision hazard, cargo exceeding limits, abnormal shaking), controls the warning light of the graded warning module 30 to flash rapidly red (3 times / second), and repeatedly prompts a high risk, an impending collision, and requests immediate cessation of operation and adjustment of cargo position, etc. The operation panel of the linkage control module 40 displays the high risk and hazard type, the red warning light in the waiting area outside the car flashes, and a buzzer (volume adjustable, ≥80dB) is triggered. The linkage control module 40 automatically sends an emergency braking signal to the elevator control system, forcibly stopping the elevator operation (lifting action stops, car door locks), and operation can only resume after the hazard is eliminated and the door is manually unlocked.

[0214] It should be noted that the linkage control module 40 can also provide real-time feedback on the operating status (normal, low risk, high risk, fault) of the elevator's intelligent anti-collision recognition and warning device to the elevator's main control system and on-site monitoring platform, facilitating remote monitoring and troubleshooting. If the elevator's intelligent anti-collision recognition and warning device malfunctions (such as a faulty detection element or module abnormality), it will automatically send a fault prompt (the warning light will remain on orange, and a voice prompt will indicate a device malfunction and request repair), without affecting the elevator's manual emergency operation.

[0215] In this embodiment, the intelligent collision avoidance warning device of the elevator can incorporate AI deep learning algorithms into the intelligent analysis module 10. Through long-term accumulated operational data, it automatically optimizes risk prediction and improves recognition accuracy. It supports remote monitoring and debugging; managers can view the operating status of the intelligent collision avoidance warning device in real time via a mobile app or computer, and receive warnings and fault alerts. The intelligent collision avoidance warning device can also add a cargo weight recognition function, combining 3D data to determine if the cargo is overloaded, and triggering the elevator's overload protection system. It also incorporates personnel entry prevention recognition to avoid personnel accidentally entering the car wall panels during cargo lifting, further enhancing safety. The intelligent collision avoidance warning device can also be adapted to intelligent logistics systems, linking with warehouse management systems and elevator control systems to achieve automated scheduling of cargo lifting, early warning of collision risks, and improve the intelligence level of logistics transfer.

[0216] Example 2:

[0217] Figure 8 This is a flowchart illustrating the steps of the intelligent anti-collision recognition and warning method for elevators described in this application embodiment.

[0218] like Figure 8 As shown in the figure, this application provides an intelligent collision avoidance warning method for elevators, including the following steps:

[0219] S1. Acquire the initial spatial positioning data of the goods in the elevator car and several multi-source point cloud data of the elevator operation. The multi-source point cloud data includes the real-time operation data of the elevator and the real-time three-dimensional data and change data of the goods in the elevator car.

[0220] S2. Analyze all multi-source point cloud data to obtain collision risk results;

[0221] S3. Based on the collision risk results, the elevator operation is controlled in conjunction with the output of different warning signals;

[0222] The process involves preprocessing, mapping, and trend prediction of various multi-source point cloud data to obtain distance prediction data between the cargo and the car; the distance prediction data is then compared with a preset safety threshold to determine the collision risk result.

[0223] It should be noted that the content of the method steps in Embodiment 2 corresponds to the content of the module in the anti-collision intelligent identification and warning device of the elevator in Embodiment 1. The module in the anti-collision intelligent identification and warning device of the elevator in Embodiment 1 has been described in Embodiment 1, and the content of the method steps of the anti-collision intelligent identification and warning device of the elevator will not be described again in this embodiment.

[0224] In this embodiment, the intelligent collision avoidance and warning method for the elevator has wide applicability in multiple industries and application scenarios, especially in industrial manufacturing, logistics and transportation, warehouse management, large equipment handling, and high-end building facility management. In these application scenarios, the goods are complex in shape and large in size, the operating environment is space-constrained, and the transportation process is often accompanied by obstructed vision and multi-directional collision risks. Traditional methods relying on human experience or simple geometric inference are difficult to ensure safety and efficiency. For transportation tasks with irregularly shaped, large, or specially constructed goods, the intelligent collision avoidance and warning method for the elevator enables adaptive 3D modeling and collision warning of forklift goods and elevator car, providing comprehensive, real-time, and accurate geometric modeling and risk prediction under conditions of obstructed vision and space constraints.

[0225] It should be noted that the application value of this intelligent collision avoidance and warning method for the elevator lies in its significant improvement in the accuracy of collision prediction, as well as the safety and efficiency in the transportation of large and irregularly shaped goods. Through the fusion of multi-source point cloud data, dynamic structure mapping, and variable resolution mesh processing, the elevator can update the cargo outline model in real time and predict changes in the relative distance between the cargo and the elevator car walls, ceiling, and floor. This not only effectively reduces the probability of collisions between cargo and the elevator structure but also avoids production stoppages, equipment damage, and cargo loss caused by accidental collisions. Simultaneously, this intelligent collision avoidance and warning method reduces reliance on manual operating experience, maintaining stable and reliable risk assessment when facing complex cargo shapes and changing transportation environments.

[0226] For example, a computer program can be divided into one or more modules / units, one or more of which are stored in memory and executed by a processor to complete this application. One or more modules / units can be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in a terminal device.

[0227] Terminal devices can be computing devices such as desktop computers, laptops, handheld computers, and cloud servers. Terminal devices may include, but are not limited to, processors and memory. Those skilled in the art will understand that this does not constitute a limitation on the terminal device, which may include more or fewer components than illustrated, or combinations of certain components, or different components. For example, a terminal device may also include input / output devices, network access devices, buses, etc.

[0228] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0229] Memory can be an internal storage unit of a terminal device, such as a hard drive or RAM. Memory can also be an external storage device, such as a plug-in hard drive, SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card. Furthermore, memory can include both internal and external storage units. Memory is used to store computer programs and other programs and data required by the terminal device. Memory can also be used to temporarily store data that has been output or will be output.

[0230] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0231] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0232] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0233] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0234] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0235] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A collision avoidance intelligent identification and warning device for an elevator, characterized in that, It includes an intelligent analysis module and a three-dimensional recognition and perception module, a graded warning module, and a linkage control module connected to the intelligent analysis module; The three-dimensional recognition and perception module is used to acquire the initial spatial positioning data of the goods in the elevator car and several multi-source point cloud data of the elevator operation. The multi-source point cloud data includes the real-time operation data of the elevator and the real-time three-dimensional data and change data of the goods contained in the elevator car. The intelligent analysis module is used to analyze all the multi-source point cloud data to obtain collision risk results; The linkage control module is used to control the operation of the elevator and control the graded warning module to output different warning signals based on the collision risk results. The intelligent analysis module includes a data fusion processing submodule, which is used to preprocess, map, and predict trends of each of the multi-source point cloud data in sequence to obtain distance prediction data between the cargo and the car. The data fusion processing submodule includes a preprocessing unit and a three-dimensional structure unit. The three-dimensional structure unit is used to generate a structure of the multi-source processed data output by the preprocessing unit according to the cargo's moving speed, attitude change, and spatial position in the car, using dynamic structure mapping, to obtain a three-dimensional spatial mesh of the cargo with variable resolution. The three-dimensional structural unit includes a matrix generation subunit, a network skeleton subunit, a partition adjustment subunit, a density adjustment subunit, and a verification subunit. The matrix generation subunit is used to calculate the velocity vector and attitude quaternion based on the spatial positioning data of the multi-source processing data; and to generate a motion state matrix based on the cargo spatial position, velocity vector and attitude quaternion of all the multi-source processing data. The network skeleton sub-unit is used to perform distributed fusion processing based on the initial spatial positioning data and the motion state matrix to obtain a unified three-dimensional space reference frame mesh skeleton. The partitioning adjustment subunit is used to perform spatial partitioning transformation on the reference frame mesh skeleton according to the cargo velocity vector, attitude change rate and the relative position between the cargo and the car, to obtain a variable resolution matrix. The density adjustment subunit is used to perform mesh density adjustment processing on the reference frame mesh skeleton using the variable resolution matrix to obtain a multi-precision spatial mesh. The verification subunit is used to resample and perform consistency verification on the multi-precision spatial grid to obtain an optimized three-dimensional spatial grid.

2. The intelligent anti-collision recognition and warning device for elevators according to claim 1, characterized in that, The intelligent analysis module also includes a risk identification submodule; the risk identification submodule is used to determine the risk by comparing the distance prediction data with a preset safety threshold to obtain a collision risk result.

3. The intelligent anti-collision recognition and warning device for elevators according to claim 1, characterized in that, The preprocessing unit is used to sequentially perform time synchronization, noise reduction and normalization on the multi-source point cloud data to obtain multi-source processed data. The preprocessing unit includes a time unification subunit, a noise reduction subunit, a drift correction subunit, a normalization subunit, a mapping subunit, and a registration optimization subunit; The time unification subunit is used to perform time synchronization processing on all the multi-source point cloud data to obtain time-unified processed data after time unification. The noise reduction subunit is used to perform high-frequency noise reduction on the time-uniformly processed data using frequency domain filtering to obtain low-frequency processed data. The drift correction subunit is used to perform low-frequency offset compensation processing on the low-frequency processed data using surface fitting to obtain drift correction processed data. The normalization subunit is used to normalize the spatial scale of the drift correction data according to a preset three-dimensional metric standard to obtain normalized data. The mapping subunit is used to perform positioning mapping processing on the initial spatial positioning data and the normalized processing data to obtain standardized positioning processing data. The registration optimization subunit is used to perform geometric alignment processing on the standardized positioning data using iterative nearest-point registration to obtain multi-source processing data.

4. The intelligent anti-collision recognition and warning device for elevators according to claim 1, characterized in that, The data fusion processing submodule also includes a mapping unit and a prediction and estimation unit; The mapping unit is used to map the contour differences between the cargo and the car onto the three-dimensional spatial grid to obtain a geometric mapping model. The prediction estimation unit is used to obtain continuous local geometric mapping data by inputting the continuous multi-source processing data into the geometric mapping model, and to predict the continuous local geometric mapping data to obtain distance prediction data.

5. The intelligent anti-collision recognition and warning device for elevators according to claim 4, characterized in that, The mapping unit includes an identifier processing subunit, a separation subunit, a feature extraction subunit, an association mapping subunit, and a region synthesis subunit; The identification processing subunit is used to perform region identification processing on each grid cell of the three-dimensional spatial grid to obtain a spatial grid model with structural difference markers. The separation subunit is used to separate and extract the contour region of the spatial grid model by point cloud clustering segmentation to obtain contour point cloud data; The feature extraction subunit is used to extract features from the contour point cloud data using normal vectors and surface fitting to obtain geometric feature data of each region. The association mapping subunit is used to perform association mapping processing on each region geometric feature data and the spatial grid model to obtain a region geometric mapping model corresponding to each region geometric feature data. The region synthesis subunit is used to perform multi-region synthesis processing on all the region geometric mapping models to obtain a geometric mapping model.

6. The intelligent anti-collision recognition and warning device for elevators according to claim 4, characterized in that, The prediction estimation unit includes: a sequence matrix subunit, a dynamic extraction subunit, a prediction trend subunit, a distance calculation subunit, and a distance correction subunit; The sequence matrix sub-unit is used to extract the geometric features of cargo posture changes in the continuous multi-source processing data according to the geometric mapping model, and obtain a morphological change sequence matrix composed of a unified time series index. The dynamic extraction subunit is used to extract dynamic attributes from the morphological change sequence matrix to obtain a dynamic attribute matrix. The prediction trend subunit is used to predict the movement trajectory of goods in the three-dimensional space of the geometric mapping model by using the dynamic attribute matrix as the multidimensional time feature of the multivariate long short-term memory network, and to obtain the position sequence matrix of the predicted position of the goods. The distance calculation subunit is used to calculate the shortest distance between the predicted location of the goods and the car using Euclidean distance based on the location sequence matrix, and to obtain distance prediction data. The distance correction subunit is used to perform weighted correction on the distance prediction data to obtain corrected distance prediction data.

7. The intelligent anti-collision recognition and warning device for elevators according to claim 2, characterized in that, The risk identification submodule is further configured to: if the distance prediction data is greater than the upper limit of the preset safety threshold, then the collision risk result is no collision risk; if the distance prediction data is less than the lower limit of the preset safety threshold, then the collision risk result is high collision risk. If the distance prediction data is between the upper and lower limits of the preset safety threshold, then the collision risk result is low collision risk.

8. The intelligent anti-collision recognition and warning device for elevators according to claim 1, characterized in that, The linkage control module is also used to control the elevator to operate normally if the collision risk result is no collision risk; to control the elevator to operate normally and control the graded warning module to output a cargo adjustment warning signal if the collision risk result is low collision risk; and to control the elevator to stop working and control the graded warning module to output a cargo adjustment and danger warning signal if the collision risk result is high collision risk.

9. A collision avoidance intelligent recognition and warning method for elevators, characterized in that, Includes the following steps: Acquire initial spatial positioning data of goods in the elevator car and several multi-source point cloud data of elevator operation. The multi-source point cloud data includes real-time operation data of the elevator and real-time three-dimensional data and change data of goods contained in the elevator car. The collision risk results are obtained by analyzing all the multi-source point cloud data. Based on the collision risk results, the elevator operation is controlled and different warning signals are output in a coordinated manner. Specifically, each of the multi-source point cloud data is preprocessed, mapped, and trend predicted sequentially to obtain distance prediction data between the cargo and the car; the distance prediction data is then compared with a preset safety threshold to determine the risk of collision. The preprocessing, mapping, and trend prediction processes for each of the aforementioned multi-source point cloud data include: The multi-source point cloud data is sequentially processed by time synchronization, denoising, and normalization to obtain multi-source processed data. Dynamic structure mapping is used to generate a structure from the multi-source processed data based on the cargo's moving speed, attitude changes, and spatial position within the car, resulting in a variable-resolution three-dimensional spatial mesh for the cargo. Contour difference mapping between the cargo and the car is applied to the three-dimensional spatial mesh to obtain a geometric mapping model. Continuous local geometric mapping data is obtained by inputting the continuous multi-source processed data into the geometric mapping model, and distance prediction data is obtained from this continuous local geometric mapping data. The resulting three-dimensional spatial mesh for the variable-resolution cargo includes: Based on the spatial positioning data of the multi-source processed data, velocity vectors and attitude quaternions are calculated; a motion state matrix is ​​generated based on the cargo spatial position, velocity vectors, and attitude quaternions of all the multi-source processed data. Distributed fusion processing is performed based on the initial spatial positioning data and the motion state matrix to obtain a unified three-dimensional spatial reference frame mesh skeleton; The reference frame mesh skeleton is spatially partitioned and transformed according to the cargo velocity vector, attitude change rate and relative position between the cargo and the car to obtain a variable resolution matrix. The reference frame mesh skeleton is subjected to mesh density adjustment using the variable resolution matrix to obtain a multi-precision spatial mesh. The multi-precision spatial mesh is resampled and subjected to consistency verification to obtain an optimized three-dimensional spatial mesh.

Images