A cloud-edge coordination and management platform and method

Abstract

This application discloses a cloud-edge coordinated management and control platform and method, belonging to the field of UAV intelligent inspection and construction modeling technology. The platform includes a UAV equipped with millimeter-wave radar and an infrared multispectral imager, an edge control layer, and a cloud-based management and control center. The UAV collects terrain and heat source data according to a preset path. The edge layer uses a Gaussian splash engine to preprocess the data, generating a rough DSM model which is then uploaded to the cloud. The cloud generates a fine DSM model through error correction, further generating a construction heat map, and optimizing the path and obstacle avoidance strategy template based on historical flight data. This solution possesses stable perception capabilities in complex environments, improving operational stability and modeling accuracy in environments such as fog and dust. Simultaneously, it achieves edge-cloud collaborative processing, effectively reducing data transmission burden and improving system response efficiency. Combined with the output of the fine DSM model and heat map, it supports the construction party in visual management and decision-making regarding earthwork volume changes and work distribution.

Description

Technical Field

[0001] This application relates to the field of cloud-edge collaborative computing, and in particular to a cloud-edge coordination and management platform and method. Background Technology

[0002] Currently, drones are being used more and more widely in scenarios such as building construction, infrastructure operation and maintenance, and earthwork monitoring. In traditional operation modes, drones mainly undertake the tasks of image acquisition and aerial modeling. Combined with onboard visual and infrared sensors, and with the help of ground station processing systems, they generate three-dimensional models (such as 3D projection models or DSMs) for construction progress analysis, quantity calculation, and efficiency evaluation.

[0003] To address the dynamic and complex environment of engineering sites, some solutions have incorporated edge processing modules, such as GPU embedded terminals or lightweight AI processors, to perform some image preprocessing and feature extraction operations, reducing the pressure on data backhaul. Some products are also exploring the use of fixed flight path templates combined with timed inspections to obtain changes in the morphology of the construction area at different points in time, assisting in the calculation of earthwork volume changes and construction efficiency indicators.

[0004] However, the above systems generally suffer from the following bottlenecks: 1. Limited perception methods and poor environmental adaptability: Existing UAVs mostly rely on visible light vision systems and ordinary infrared thermal imaging equipment. When encountering low visibility scenarios such as fog, smoke, or rain, the sensor acquisition quality drops significantly, resulting in problems such as blurred imaging, target recognition failure, false alarms, and missed alarms. 2. Insufficient obstacle avoidance capabilities, posing flight safety risks: Traditional path planning often uses static preset schemes. During UAV flight, there is a lack of real-time perception and avoidance capabilities for high-altitude obstacles (such as tower cranes, steel cables, scaffolding, etc.), especially in low visibility environments, making collisions highly likely. 3. Point cloud modeling is significantly affected by interference, and DSM accuracy is unstable: Although millimeter-wave radar has penetrating capabilities, in environments with high PM2.5 concentrations and high air humidity, signal scattering is severe, resulting in a low echo signal-to-noise ratio. The generated point cloud data is prone to noise and holes, affecting the accuracy of the DSM model. 4. High edge computing load and insufficient real-time performance: While a lightweight AI engine deployed on one side can partially process image data, there are computational conflicts between tasks such as image restoration, anomaly detection, and obstacle avoidance, increasing system response latency and making it difficult to handle complex perception and decision-making collaborative tasks. 5. Rigid path design and lack of learning ability: Most existing path planning templates are statically configured, lacking dynamic learning and self-updating mechanisms. They cannot adaptively adjust the flight path according to changes in obstacles and environmental disturbances, severely limiting the system's intelligence level. Summary of the Invention

[0005] The purpose of this application is to address the problems mentioned above, such as limited perception methods, poor environmental adaptability, insufficient obstacle avoidance capabilities leading to flight safety risks, high edge computing load and insufficient real-time performance, and strong path rigidity and lack of learning ability.

[0006] According to one aspect of this application, a cloud-edge coordination and management platform is provided, comprising:

[0007] The drone equipment is equipped with a radar main unit and an infrared slave unit. The radar main unit is used to scan the terrain of the target area and acquire scan data; the infrared slave unit is used to lock onto the heat source area generated by the mechanical equipment, perform detailed supplementary shooting, and acquire detailed data.

[0008] An edge control layer, set on the drone equipment, has a built-in lightweight Gaussian splash engine for preprocessing the scan data and detail data, marking abnormal areas where the image signal-to-noise ratio is lower than a preset threshold, and generating a DSM coarse model.

[0009] The cloud-based control center receives the DSM rough model and optimizes it using a set of deployed engineering perception data intelligent processing algorithms to generate the DSM fine model. The cloud-based control center further calculates the earthwork volume of the target area based on the DSM fine model and generates a construction heat map.

[0010] Preferably, the radar host is a millimeter-wave radar, used to penetrate suspended particles in the air and sense terrain contour information in haze or low visibility environments; the millimeter-wave radar is combined with an acoustic array system, which includes multiple acoustic transmitting and receiving units for auxiliary calibration of the spatial positioning results of the millimeter-wave radar.

[0011] The infrared slave device is a multispectral imager operating in the 8-14μm band, used to acquire image information in different bands; the multispectral imager has a built-in database of infrared reflectivity of metal materials, used to improve the multispectral imager's ability to identify the reflectivity characteristics of different metal surfaces.

[0012] The PM2.5 sensor, which integrates temperature, humidity, and pressure, is used to detect the concentration of fine particulate matter, temperature, humidity, and air pressure parameters in the environment.

[0013] This invention also provides a cloud-edge coordination and management method, applied to the aforementioned cloud-edge coordination and management platform, comprising:

[0014] Generate and send out the drone's cruise path, control the drone to cruise along the cruise path, perform terrain scanning on the target area, collect terrain data, and generate a rough DSM model.

[0015] The DSM rough model is preprocessed through an edge control layer and then uploaded to the cloud management center;

[0016] The DSM coarse model is corrected for errors by the engineering perception data intelligent processing algorithm deployed in the cloud control center to generate the DSM fine model;

[0017] A construction heat map is generated based on the DSM precision model.

[0018] Preferably, the acquired terrain data includes: obtaining terrain contour data by scanning with a radar host, and obtaining images of the heat source area by scanning with an infrared slave.

[0019] Preferably, the radar host includes a millimeter-wave radar and a sonic array system; correspondingly, controlling the UAV to cruise along the cruise path includes:

[0020] The millimeter-wave radar transmits frequency-modulated continuous waves in the target area to acquire terrain contour data;

[0021] The ultrasonic array system emits ultrasonic pulses toward the target area and receives echo signals after encountering obstacles.

[0022] The spatial location of the obstacle is calculated based on the sound wave flight time and temperature and humidity correction model.

[0023] The obstacle position data measured by the acoustic array system is spatiotemporally aligned with the millimeter-wave radar data to calculate the ranging error;

[0024] Based on the ranging error, the positioning error of the millimeter-wave radar is determined and corrected.

[0025] Preferably, the determination and correction of the millimeter-wave radar positioning error includes:

[0026] The obstacle distance data obtained by the millimeter-wave radar and the acoustic array system are given a unified timestamp, and the obstacle spatial position data determined by the acoustic array system in its own coordinate system is transformed into the coordinate system of the millimeter-wave radar through rigid coordinate transformation.

[0027] Calculate the deviation between the obstacle distance data obtained by the millimeter-wave radar and the acoustic array system respectively:

[0028]

[0029] Obtain the PM2.5 concentration value at the corresponding time point and determine whether any of the following conditions are met:

[0030] Δd exceeds the preset error threshold;

[0031] PM2.5 concentrations were higher than the preset standards for dense fog environments;

[0032] If any of the above conditions are met, the compensation mechanism will be triggered:

[0033] If the smog interference is severe, the acoustic coordinates should be used directly to replace the corresponding point cloud coordinates.

[0034] If the interference level is moderate, the radar and acoustic coordinates are weighted and fused according to preset weights to generate corrected point cloud coordinates:

[0035]

[0036] The corrected point cloud data will be used as the coordinate data for navigation.

[0037] Preferably, the preprocessing of the DSM rough model through the edge control layer includes:

[0038] Abnormal region marking: By performing signal-to-noise ratio analysis on the image or point cloud data in the DSM coarse model, regions with a signal-to-noise ratio lower than a preset threshold are identified and marked as abnormal regions.

[0039] Data augmentation processing performs quality optimization operations on the DSM coarse model data, including but not limited to noise filtering, detail sharpening, and contrast adjustment;

[0040] Data compression processing is performed to compress the preprocessed DSM coarse model data.

[0041] Preferably, the intelligent processing algorithm for engineering perception data is a haze scattering error compensation algorithm, and correspondingly, the error correction includes:

[0042] Collect real-time PM2.5 concentration and atmospheric humidity;

[0043] Based on a haze forecasting model, predict the height shift of point clouds;

[0044] The height of the original point cloud is corrected using the height offset of the point cloud, thereby adjusting the height data in the point cloud.

[0045] Preferably, the height offset of the predicted point cloud includes:

[0046] The point cloud height is corrected using the following error correction formula:

[0047]

[0048] In the formula, The dust emission coefficient at the construction site. It is the scattering attenuation factor. The atmospheric humidity, the construction site dust coefficient, and the scattering attenuation factor are obtained through training with historical data.

[0049] Preferably, the method further includes:

[0050] Based on the latest terrain morphology and obstacle spatial distribution information contained in the DSM fine model, potential flight risk areas and obstacles within the construction area are identified.

[0051] The DSM model is compared and analyzed with historical flight trajectories and obstacle avoidance records to form an updated path planning scheme;

[0052] The path planning scheme and corresponding obstacle avoidance strategy are distributed to the drone in template form.

[0053] The cloud-edge coordinated management platform and method proposed in this application enhances environmental adaptability by constructing multimodal perception. This embodiment, through a combined radar host and infrared slave perception mechanism, significantly enhances the UAV's perception capability in complex environments (such as smoke, haze, and obstructions), improving operational stability. It improves modeling accuracy by integrating a Gaussian splash engine into the edge control layer, enabling image quality pre-screening and anomaly area labeling, effectively avoiding the impact of low-quality data on DSM construction. The cloud further optimizes coarse model errors, making the final DSM fine model closer to the actual terrain. It reduces transmission burden and improves real-time performance by processing data at the edge before uploading to the cloud, significantly reducing redundant transmission of original images and sensor data, shortening return latency, and improving the overall system response speed. It provides construction data visualization support; through the generated DSM fine model and construction heat map, this platform can dynamically track the trend of earthwork volume changes, providing intuitive auxiliary decision-making basis for construction parties.

[0054] Furthermore, this solution brings precise and efficient data acquisition and modeling capabilities. Drones can perform missions autonomously without human intervention, quickly completing terrain data acquisition and preliminary modeling, significantly improving inspection efficiency. Stable perception capabilities in complex environments such as fog and haze are enhanced through the collaborative perception of millimeter-wave radar and acoustic arrays, as well as edge image enhancement processing, effectively mitigating the interference of low-visibility environments on modeling accuracy and improving the quality of the DSM rough model. Cloud-based intelligent optimization ensures accuracy; after error correction by the cloud-based control center, the DSM rough model forms a high-precision DSM fine model, providing a solid foundation for the generation of construction heat maps and avoiding data distortion caused by noise errors in traditional modeling. Enhanced data support for construction management decisions is provided; the construction heat map, as a layer intuitively reflecting the intensity and distribution of construction activities, provides reliable data support for construction units to optimize project scheduling, adjust resource allocation, and assess project quantities. Attached Figure Description

[0055] 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 from these drawings without creative effort.

[0056] Figure 1 This is a logic block diagram of the cloud-edge coordination and control method described in one embodiment of this application. Detailed Implementation

[0057] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings. Preferred embodiments of this application are shown in the drawings. However, this application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of this application.

[0058] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein in the specification of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0059] Please refer to Figure 1 One embodiment of this application provides a cloud-edge coordinated management and control platform, including unmanned aerial vehicle (UAV) equipment, an edge control layer, and a cloud management and control center.

[0060] The drone equipment is equipped with a radar main unit and an infrared slave unit. The radar main unit is used to scan the terrain of the target area and acquire scan data; the infrared slave unit is used to lock onto the heat source area generated by the mechanical equipment, take detailed supplementary photos, and acquire detailed data.

[0061] The edge control layer is set on the drone equipment and has a built-in lightweight Gaussian splash engine to preprocess the scan data and detail data, mark abnormal areas where the image signal-to-noise ratio is lower than a preset threshold, and generate a DSM coarse model.

[0062] The cloud-based control center receives the DSM rough model and optimizes it using a set of intelligent processing algorithms for engineering perception data to generate the DSM fine model. The cloud-based control center then calculates the earthwork volume of the target area based on the DSM fine model and generates a construction heat map.

[0063] In this embodiment, it should be noted that the cloud-edge coordinated management and control platform consists of three main components: drone equipment, edge control layer, and cloud management and control center. They work together to achieve high-precision perception and dynamic modeling of the construction site.

[0064] Specifically, the drone equipment is equipped with two types of sensing units:

[0065] The radar unit is preferably a combination system integrating millimeter-wave radar and acoustic array, used for terrain scanning of the target area during flight. Millimeter-wave radar has strong penetration capabilities and can identify the outline of ground features in environments with slight obstruction or low visibility; the acoustic array is used to assist in obstacle location correction and improve ranging accuracy.

[0066] Infrared slave devices are used to locate areas with high heat sources, and are especially suitable for identifying continuously operating construction machinery such as tower cranes, excavators, and concrete mixing plants, thereby assisting in detailed modeling or heat map analysis.

[0067] The edge control layer, serving as the local computing unit for the drone device, integrates a lightweight Gaussian splash engine. This module possesses edge AI image processing capabilities, enabling it to complete processing before data upload.

[0068] Image quality assessment and anomaly region labeling, such as image frames with a signal-to-noise ratio below a set threshold;

[0069] Data compression and enhancement preprocessing, including preliminary feature extraction and filtering for noise reduction;

[0070] Based on the above processing, a DSM (Digital Surface Model) is generated, which serves as the basis for uploading structured data to the cloud.

[0071] After receiving the DSM coarse model data from the edge control layer, the cloud-based control center uses a set of deployed engineering-aware data intelligent processing algorithms to further optimize error and anomaly areas in the DSM model, generating a DSM fine model that meets the accuracy requirements of actual terrain. The cloud platform also features:

[0072] The ability to perform high-precision earthwork volume calculation on the DSM fine model, and to quantify the inflow and outflow volumes by combining historical model changes;

[0073] Construction heat maps are generated based on terrain data and heat source distribution to assist in construction progress management, resource scheduling, and efficiency analysis.

[0074] The technical solution implemented in this embodiment has the following technical effects: Enhanced environmental adaptability: This embodiment significantly enhances the UAV's perception capability in complex environments (such as smoke, haze, and obstructions) through a combined radar host and infrared slave perception mechanism, improving operational stability. Improved modeling accuracy: The edge control layer integrates a Gaussian splashing engine to achieve image quality pre-screening and abnormal area labeling, effectively avoiding the impact of low-quality data on DSM construction; the cloud further optimizes coarse model errors, making the final DSM fine model closer to the actual terrain. Reduced transmission burden and improved real-time performance: Data is processed at the edge before being uploaded to the cloud, significantly reducing redundant transmission of original images and sensor data, shortening return latency, and improving the overall system response speed. Construction data visualization support: Through the generated DSM fine model and construction heat map, this platform can dynamically track the trend of earthwork volume changes, providing construction parties with intuitive auxiliary decision-making basis.

[0075] In summary, this embodiment improves modeling efficiency and accuracy by establishing a closely coordinated perception and processing link between the UAV, the edge layer, and the cloud, and also enhances the system's adaptability and responsiveness to complex construction environments.

[0076] Furthermore, the radar host is a millimeter-wave radar, used to penetrate suspended particles in the air and perceive terrain contour information in haze or low visibility environments; the millimeter-wave radar is combined with an acoustic array system, which includes multiple acoustic transmitting and receiving units, used to assist in calibrating the spatial positioning results of the millimeter-wave radar.

[0077] The infrared slave unit is a multispectral imager operating in the 8-14μm band, used to acquire image information in different bands; the multispectral imager has a built-in database of infrared reflectivity of metal materials, which is used to improve the multispectral imager's ability to identify the reflectivity characteristics of different metal surfaces;

[0078] The PM2.5 sensor, which integrates temperature, humidity, and pressure, is used to detect the concentration of fine particulate matter, temperature, humidity, and air pressure parameters in the environment.

[0079] In this embodiment, it should be noted that the sensing components carried by the UAV equipment of the cloud-edge coordinated management and control platform have been further designed with multimodal integration, specifically including: millimeter-wave radar, acoustic array system, multispectral imager (infrared slave), and PM2.5 and temperature, humidity and pressure integrated sensor, so as to achieve stronger environmental adaptability and multi-source information fusion sensing effect.

[0080] Specifically:

[0081] Millimeter-wave radar, as the main unit of the radar host, operates in the frequency band between 30 GHz and 300 GHz. It has good penetration performance and is especially suitable for detecting terrain contours in low visibility or hazy environments. Millimeter-wave radar can construct preliminary point cloud data of the target area in real time by transmitting frequency modulated continuous wave (FMCW) and receiving echo signals, which is used to generate the terrain basis for the DSM coarse model.

[0082] The acoustic array system works in conjunction with millimeter-wave radar, consisting of multiple ultrasonic transmitting and receiving units deployed at different locations. The system calculates the spatial location of obstacles by emitting ultrasonic pulses and using the time-of-flight (TOF) of the echoes. It further incorporates a temperature and humidity compensation model to correct for changes in sound speed, thereby improving positioning accuracy. When the radar echo signal-to-noise ratio decreases due to environmental interference, the acoustic array can serve as an auxiliary reference for calibrating the biased point cloud results of the millimeter-wave radar.

[0083] The infrared slave unit is a multispectral imager operating in the 8–14 μm band, covering the long-wave infrared region. It can acquire thermal imaging images in different bands under low-light or no-light conditions. This imager incorporates a database of infrared reflectivity for metallic materials, used to compare the reflectivity characteristics of different metals (such as steel, aluminum, and copper) in different bands. This improves the accuracy of identifying the surface of mechanical equipment, especially in scenarios where temperature differences are not significant but material contrasts are pronounced.

[0084] An integrated PM2.5 sensor with temperature, humidity, and pressure is used to supplement environmental sensing. It can monitor environmental parameters such as the concentration of fine particulate matter (PM2.5), temperature, humidity, and air pressure in the air along the flight path in real time. This data is used to correct the sound velocity calculation error in the sound wave detection process, and also provides environmental factor support for error judgment and filtering compensation of subsequent point cloud data.

[0085] The technical solution implemented in this embodiment has the following technical effects:

[0086] Significantly improving radar positioning stability and anti-interference capability, millimeter-wave radar has the ability to penetrate fog and haze, and can stably output terrain contour information under low visibility conditions, overcoming the shortcoming of visual sensors being prone to failure in bad weather.

[0087] To enhance obstacle recognition and avoidance capabilities, the platform uses an acoustic array system to perform spatiotemporal alignment and error correction on the scanning data from millimeter-wave radar. This enables multimodal complementary perception, allowing for more accurate identification of slender metal obstacles such as tower crane cables and scaffolding, thus improving dynamic obstacle avoidance capabilities.

[0088] By enhancing the accuracy of infrared image recognition and leveraging the infrared database matching capabilities of a multispectral imager, the system can still identify the material characteristics of equipment even when the heat source area is not obvious or the temperature difference is low, thereby improving the precise locking of construction machinery and the effect of image re-capture.

[0089] The dynamic environmental parameter closed-loop correction mechanism uses PM2.5 concentration, temperature, humidity and pressure data to reverse-calibrate the environmental variable dependence in millimeter-wave radar and acoustic ranging, enabling the platform to have dynamic adaptive calibration capabilities and improve data reliability and scenario generalization capabilities.

[0090] In one specific embodiment, the present invention also provides a cloud-edge coordination and management method, applied to the aforementioned cloud-edge coordination and management platform, comprising:

[0091] S100: Generate and distribute the drone's cruise path, control the drone to cruise along the path, and perform terrain scanning of the target area to collect terrain data and generate a rough DSM model. In this step, it should be noted that the cruise path is generated by the cloud control center based on factors such as historical flight paths, construction stage characteristics, regional construction frequency, and risk level, and is automatically distributed to the drone system controlled by the edge control layer. During the cruise mission, the onboard millimeter-wave radar and infrared slave unit work together. The millimeter-wave radar continuously scans the area to acquire terrain contour data, while the infrared slave unit performs targeted detail photography of areas with high heat source characteristics. All perceived data is initially filtered and fused at the edge end and then used to construct the DSM (Digital Surface Model) rough model.

[0092] S200: The DSM coarse model is preprocessed through the edge control layer and uploaded to the cloud management center. In this step, it should be noted that the edge control layer incorporates a lightweight Gaussian splash image enhancement engine and an anomaly detection module. These are used to enhance the generated DSM coarse model, improving image clarity and boundary features, and marking abnormal areas in the image with a signal-to-noise ratio below a preset threshold (such as point cloud holes or blurred boundaries caused by haze). Simultaneously, this step also includes registration, coordinate unification, and format standardization of multi-source sensing data to ensure the structural integrity of the uploaded data and algorithm compatibility.

[0093] S300: The DSM coarse model is error-corrected using an intelligent processing algorithm for engineering perception data deployed in the cloud control center to generate a DSM fine model. In this step, it's important to note that the cloud control center, equipped with this intelligent processing algorithm, first performs restoration interpolation on abnormal areas after receiving the DSM coarse model uploaded from the edge. Then, it combines the ranging results from acoustic waves and millimeter-wave radar to estimate errors and compensate for point cloud shifts or noise distortion caused by dense fog, reflections, or occlusion. This step also includes further refining the surface structure of the 3D model using historical datasets, multi-time period comparison results, and sensor intrinsic parameter models to achieve high-precision DSM fine model reconstruction.

[0094] S400. Generate a construction heat map based on the DSM high-resolution model. In this step, it's important to note that the construction heat map is generated based on the DSM high-resolution model. By comparing the volumetric differences between the current model and historical models, the earthwork variation is calculated. This is then combined with spatiotemporal overlay analysis of heat source distribution (areas with high-frequency mechanical activity) to ultimately visualize the regional construction intensity distribution map. This heat map provides support for subsequent construction scheduling, efficiency analysis, and risk warning.

[0095] The technical solution implemented in this embodiment has the following technical effects:

[0096] With its precise and efficient data acquisition and modeling capabilities, the drone can perform missions without human intervention, autonomously scanning paths and quickly completing terrain data acquisition and preliminary modeling, significantly improving inspection efficiency.

[0097] The stable perception capability in complex environments such as haze, through the coordinated perception of millimeter-wave radar and acoustic arrays, as well as edge image enhancement processing, effectively mitigates the interference of low visibility environment on modeling accuracy and improves the quality of DSM coarse model.

[0098] Cloud-based intelligent optimization ensures accuracy. After error correction by the cloud-based control center, the DSM coarse model is transformed into a high-precision DSM fine model, providing a solid foundation for the generation of construction heat maps and avoiding data distortion caused by noise errors in traditional modeling.

[0099] Enhanced data support for construction management decisions: Construction heat maps, as a layer that intuitively reflects the intensity and distribution of construction activities, provide reliable data support for construction units to optimize project scheduling, adjust resource allocation, and assess project quantities.

[0100] Furthermore, the acquisition of terrain data includes: obtaining terrain contour data through radar mainframe scanning, and obtaining images of heat source areas through infrared slave scanning.

[0101] In this embodiment, it should be noted that the acquired terrain data not only includes conventional terrain contour information but also further integrates heat source area image information to achieve a more comprehensive three-dimensional perception and scene understanding of the target area. Specifically, the radar host actively transmits millimeter-wave signals to perform high-precision detection of the surface contour, acquiring structural features such as terrain boundaries and undulations; simultaneously, the infrared slave scans the target area in a multispectral manner, capturing the thermal radiation information generated by moving targets such as machinery or personnel, thereby locating the heat source area within the construction site. This multimodal perception mechanism breaks through the limitations of traditional single-vision or radar perception, enabling the platform to maintain good terrain reconstruction and dynamic object recognition capabilities even in low-visibility environments (such as fog, haze, and dust).

[0102] Implementing the technical solution of this embodiment can effectively improve the integrity and perception accuracy of terrain data. On the one hand, millimeter-wave radar has the advantage of not being affected by visible light interference in the perception of structural information, ensuring that a clear terrain outline can still be obtained under complex weather conditions. On the other hand, the infrared slave device supplements the ability to identify heat sources, which is particularly suitable for focused detection and detailed imaging in high-intensity operation areas. It helps to capture the location of construction machinery with high heat intensity, enrich the semantic information of the DSM rough model, and provide important input for subsequent construction heat map generation and flight path optimization.

[0103] Furthermore, the radar main unit includes a millimeter-wave radar and a sonic array system. Accordingly, controlling the UAV to cruise along the cruise path in the S100 includes:

[0104] S110. The millimeter-wave radar transmits frequency-modulated continuous waves (FMCW) over the target area to acquire terrain contour data. In this step, it should be noted that the millimeter-wave radar operates in the frequency band above 30 GHz, possessing good penetration capabilities and enabling it to acquire coarse-grained contour data of the target area even in low-visibility environments (such as fog, dust, and low light). The radar periodically transmits FMCW and receives the echo signals, measuring the relative distance and shape changes to obstacles. The obtained echoes are pre-processed to construct a basic terrain point cloud, forming a preliminary contour of the DSM coarse model.

[0105] S120. Ultrasonic pulses are emitted towards the target area via an acoustic array system, and echo signals are received after an obstacle is encountered. In this step, it should be noted that the acoustic array consists of multiple transmitting and receiving units, employing directional transmission and multi-point reception to achieve small-scale, high-sensitivity obstacle identification. This system is particularly suitable for detecting metallic obstacles (such as tower crane cables, steel structures, and scaffolding). Even in areas where millimeter waves fail due to multipath scattering or attenuation, the acoustic system can still stably identify reflectors, significantly improving the overall robustness of obstacle identification.

[0106] S130. Calculate the spatial location of the obstacle based on the sound wave time-of-flight (TOF) correction model and temperature / humidity adjustment. In this step, it should be noted that to obtain high-precision sound wave ranging results, the system is based on the Time-of-Flight (TOF) principle, combined with ambient temperature and humidity correction formulas:

[0107] Where T represents ambient temperature (°C) and H represents relative humidity (%), to correct for the speed of sound propagation c. Then, according to the distance calculation formula... It can accurately obtain the acoustic ranging value, thereby calculating the position of each obstacle point in space, and is especially suitable for blind zone compensation in millimeter wave scanning where there is obstruction or error.

[0108] S140. The obstacle position data measured by the acoustic array system is spatiotemporally aligned with the millimeter-wave radar data, and the ranging error is calculated. In this step, it should be noted that the system timestamps the data from the acoustic array and the millimeter-wave radar together and establishes a coordinate transformation relationship in geometric space to achieve spatiotemporal alignment of the two sets of sensor data.

[0109] S150. Based on the ranging error, determine and correct the positioning error of the millimeter-wave radar. In this step, it should be noted that the system dynamically compensates for the positioning of the millimeter-wave radar based on the ranging error value. This adaptive filtering mechanism ensures that the millimeter-wave point cloud data still has good continuity and accuracy under adverse weather conditions, overcoming the defects of traditional radar systems that are prone to point cloud holes or jumps in foggy and hazy environments.

[0110] The technical solution implemented in this embodiment can realize a multimodal collaborative sensing mechanism, combining the advantages of millimeter-wave radar and acoustic array system, and complementary sensing under different physical interference conditions to ensure the integrity of obstacle recognition and anti-interference capability.

[0111] To improve ranging accuracy and positioning robustness, the accuracy is enhanced through acoustic ranging and environmental compensation models, while a dynamic error correction mechanism is introduced to achieve real-time optimization processing of millimeter-wave radar point cloud data.

[0112] To enhance operational stability under adverse weather conditions, the system is specially designed for typical meteorological environments at construction sites, such as fog and high humidity, significantly reducing the impact of the external environment on the modeling accuracy and obstacle avoidance performance of the UAV.

[0113] This provides reliable foundational data for the subsequent construction of the DSM fine model. The corrected high-confidence point cloud lays the data foundation for the accuracy of the digital surface model, thereby improving the accuracy of subsequent construction heat maps and volume calculations.

[0114] In this embodiment, determining and correcting the positioning error of the millimeter-wave radar includes:

[0115] S151. The obstacle distance data obtained by the millimeter-wave radar and the acoustic array system are timestamped, and the obstacle spatial position data determined by the acoustic array system in its own coordinate system is transformed to the coordinate system of the millimeter-wave radar using a rigid coordinate transformation. It should be noted that, due to the temporal and spatial coordinate system differences between the data acquired by the millimeter-wave radar and the acoustic array system, time synchronization is required for effective alignment. The system uses the UAV's unified clock system or synchronization signal to assign precise timestamps to the two modal data. Subsequently, based on the installation position of the acoustic array relative to the millimeter-wave radar (including installation angle and baseline distance), a rigid coordinate transformation (e.g., Euler angle or quaternion transformation) is performed to uniformly map the data in the acoustic coordinate system to the coordinate system of the millimeter-wave radar, achieving spatial alignment of the two modal data.

[0116] S152. Calculate the deviation between the obstacle distance data obtained by the millimeter-wave radar and the acoustic array system respectively:

[0117]

[0118] In this step, it should be noted that the system calculates the difference between the distance measured by the millimeter-wave radar and the distance measured by the acoustic array system. By performing difference analysis on key obstacle points or continuous point cloud segments, it can be determined whether the millimeter-wave radar is affected by false alarms, missing data, or abrupt changes in a specific area, thus forming a dynamic evaluation basis for the radar ranging accuracy.

[0119] S153. Obtain the PM2.5 concentration value at the corresponding time point and determine whether any of the following conditions are met: The error exceeds a preset error threshold or the PM2.5 concentration exceeds a preset dense fog environmental standard. In this step, it should be noted that, to improve the adaptability and intelligence of the compensation mechanism, the system not only relies on the ranging error between sensors but also introduces ambient air quality indicators as supplementary judgment conditions. Specifically, when the error... If the measured PM2.5 concentration exceeds a dynamically set threshold (e.g., 0.3m), or exceeds the dense fog level (e.g., >250), the system considers the current radar ranging unreliable and initiates a compensation process. This judgment mechanism, combining environmental parameters and sensor status, constitutes an intelligent judgment model with dynamic environmental perception capabilities.

[0120] S154. If any of the above conditions are met, the compensation mechanism is triggered:

[0121] If the smog interference is severe, the acoustic coordinates should be used directly to replace the corresponding point cloud coordinates.

[0122] If the interference level is moderate, the radar and acoustic coordinates are weighted and fused according to preset weights to generate corrected point cloud coordinates:

[0123] In this step, it should be noted that the system differentiates between different levels of interference and handles them accordingly:

[0124] In scenarios with severe interference (such as PM2.5 > 400 or...) Under these conditions, the radar echo suffers from severe scattering distortion, so the system directly replaces the original millimeter-wave data points with the spatial coordinate points output by the acoustic array.

[0125] Under moderate interference, the system adopts a fusion strategy and calculates the corrected coordinates as follows:

[0126]

[0127] in, The fusion weights are dynamically set (e.g., 0.6~0.8). This method preserves the overall distribution characteristics of millimeter-wave radar point clouds while incorporating the local accuracy of acoustic ranging, significantly improving the model's precision and coherence.

[0128] In one specific embodiment, The value is 0.7.

[0129] S155. Use the corrected point cloud data as the coordinate data for cruise. In this step, it's important to note that after fusing data from the millimeter-wave radar and acoustic array system, and correcting errors based on PM2.5 concentration and ranging errors, the resulting high-precision terrain point cloud data is no longer just used as modeling input, but further serves as the base coordinate set for the UAV's subsequent flight path. This coordinate data possesses stronger terrain fit and obstacle avoidance capabilities, and can be directly used for path replanning, flight navigation, and dynamic obstacle avoidance strategy generation.

[0130] The technical solution implemented in this embodiment can realize a highly robust adaptive calibration mechanism, integrate multi-source sensing capabilities of millimeter-wave radar and acoustic ranging, and achieve highly reliable correction and compensation of millimeter-wave ranging errors through spatiotemporal alignment of point cloud data and dynamic environmental state judgment.

[0131] This solution specifically addresses the error accumulation problem caused by the degradation of millimeter-wave sensing performance in low-visibility environments. It sets up a dual judgment mechanism with PM2.5 concentration threshold and ranging deviation threshold. Once a sensing anomaly is detected, an acoustic wave-dominated compensation strategy is triggered, thereby significantly enhancing the system's stability and adaptability in complex environments.

[0132] By directly using the corrected and optimized point cloud data as the basis for cruise coordinates, spatial deviation problems caused by inaccurate ranging or abnormal modeling can be effectively avoided, ensuring that the UAV has higher path accuracy and obstacle avoidance capabilities in subsequent flight missions. This improves the spatial consistency and structural integrity of the DSM coarse model from the source, thereby maximizing flight safety during UAV cruise and reducing collision risks.

[0133] In an optional embodiment, the preprocessing of the DSM rough model via the edge control layer in S200 includes:

[0134] S210, Abnormal Region Marking: By performing signal-to-noise ratio (SNR) analysis on the image or point cloud data in the DSM coarse model, regions with an SNR below a preset threshold are identified and marked as abnormal regions. In this step, it should be noted that the edge control layer utilizes a built-in Gaussian splash engine to first perform pixel-level or point cloud-level analysis on the DSM coarse model transmitted from the drone, extracting its signal strength to background noise ratio (SNR). The system sets a lower limit threshold for SNR (e.g., SNR < 10dB) as the judgment criterion, automatically identifying and marking regions below this threshold. Such abnormal regions are often caused by factors such as fog / haze obstruction, sensor angle deviation, or unstable device posture. Early identification of these regions facilitates subsequent image enhancement or reshooting task planning.

[0135] S220. Data augmentation processing: Perform quality optimization operations on the DSM coarse model data, including but not limited to noise filtering, detail sharpening, and contrast adjustment. In this step, it should be noted that the system employs various image / point cloud augmentation algorithms to improve quality for labeled regions or the entire DSM coarse model image. For example, for image-based DSM, non-local mean filtering (NLM) can be used to remove granular noise, and the Laplace operator can be applied to edge regions to enhance edge sharpness; for point cloud-based DSM, voxel grid filtering can be applied to reduce outliers, and curvature smoothing can be used to reconstruct the true terrain surface. The enhanced DSM data retains structural details while significantly reducing the interference of measurement noise on subsequent modeling.

[0136] S230. Data Compression Processing: The preprocessed DSM coarse model data is compressed. In this step, it's important to note that to reduce data upload bandwidth load and cloud processing latency, the system performs lossy or lossless compression on the preprocessed DSM coarse model data. For image-based DSMs, compression algorithms supporting progressive transmission, such as JPEG2000, can be used; for point cloud data, Octree encoding or point index compression methods from PCL (Point Cloud Library) can be employed. This compression process ensures efficient data transmission and rapid cloud uploading under limited resource conditions, while guaranteeing that the key features required for model reconstruction are not weakened.

[0137] The technical solution implemented in this embodiment can improve the reliability and availability of data input. By actively marking abnormal areas through signal-to-noise ratio analysis, it can ensure that the system can respond to low-quality data in a targeted manner and improve the overall modeling accuracy.

[0138] To enhance the visual clarity and geometric accuracy of DSM data, image and point cloud enhancement techniques are used to optimize the details of the original coarse model data, significantly improving image contrast and edge clarity, and reducing noise interference.

[0139] To improve edge computing efficiency and cloud processing smoothness, a reasonable data compression mechanism can significantly reduce the size of uploaded data, decrease dependence on communication links, and further improve the efficiency of edge-cloud collaborative processing.

[0140] In one specific embodiment, the intelligent processing algorithm for engineering perception data is a haze scattering error compensation algorithm, and correspondingly, the error correction in S300 includes:

[0141] S310. Collect real-time PM2.5 concentration and atmospheric humidity. In this step, it's important to note that after the cloud-based control center receives the DSM coarse model uploaded by the edge control layer, it first invokes the environmental sensing module deployed in the monitoring system to collect real-time PM2.5 concentration values ​​and atmospheric humidity parameters corresponding to the target area, in order to achieve accurate error compensation. The collected data sources may include: environmental sensors mounted on the UAV (such as an integrated temperature, humidity, pressure, and fine particulate matter monitoring module), and interfaced regional meteorological services (such as a local meteorological station or a cloud-based meteorological API). This data serves as a crucial input variable for subsequent haze modeling and point cloud error estimation.

[0142] S320. Based on the haze forecasting model, predict the height shift of the point cloud. Further, the predicted height shift of the point cloud in S320 includes:

[0143] The point cloud height is corrected using the following error correction formula:

[0144]

[0145] In the formula, The dust emission coefficient at construction sites is used to characterize the long-term statistical characteristics of suspended particulate concentration in construction scenarios. This is the scattering attenuation factor, reflecting the energy attenuation of millimeter-wave signals under different concentrations of particulate matter; Atmospheric humidity, construction site dust coefficient, and scattering attenuation factor are obtained through training with historical data, possessing a certain degree of regional adaptability and model generalization ability. In this step, it should be noted that the system incorporates a haze impact prediction model trained based on historical point cloud offsets. This model uses input environmental factors such as PM2.5 concentration and atmospheric humidity as variables and outputs a point cloud height offset estimate. The model can be constructed using multinomial regression, random forest regression, or lightweight neural networks, and is suitable for predicting the systematic errors caused by haze conditions on the propagation distance of millimeter-wave radar signals. This offset reflects the terrain point cloud height estimation error caused by the prolonged radar signal propagation time due to haze.

[0146] S330. Correct the original point cloud height using the point cloud height offset, adjusting the height data in the point cloud. In this step, it's important to note that the cloud control center corrects each terrain point cloud data point in the DSM coarse model point-by-point based on the predicted point cloud height offset. Correction methods include direct linear compensation or weighted adjustment. For example, if the predicted average height deviation due to haze in a certain area is -12cm, the system adds 12cm to the height values ​​of all point clouds in that area. For areas with uneven impact, the system uses differential grid correction to ensure the continuity of the terrain model and its match with the real landform. The corrected point cloud data is then used as input data for generating the DSM fine model, significantly improving the geometric accuracy of the modeling.

[0147] The technical solution implemented in this embodiment can enhance the robustness of the model to environmental disturbances. By integrating PM2.5 and humidity monitoring, the system can dynamically sense the impact of atmospheric conditions on radar ranging, ensuring the environmental adaptability of data processing.

[0148] To achieve accuracy compensation for point cloud modeling under hazy conditions, a prediction model is used to quantitatively predict point cloud deviations, thus overcoming the shortcoming of uncontrollable errors in traditional modeling under low visibility conditions.

[0149] Improving the geometric accuracy and reliability of the DSM model, the corrected point cloud data is closer to the actual terrain, ensuring the reliability of heat map generation and earthwork calculation results, and providing a highly reliable data foundation for subsequent UAV scheduling and engineering construction assessment.

[0150] Furthermore, in one specific embodiment, the method further includes:

[0151] S500 identifies potential flight risk areas and obstacles within the construction area based on the latest terrain morphology and obstacle spatial distribution information contained in the DSM high-precision model. In this step, it should be noted that by analyzing the high-precision terrain height field and edge features reconstructed in the DSM high-precision model, combined with indicators such as point cloud density, structural contour changes, and thermal region concentration, abrupt terrain changes, areas of significant elevation differences, and dynamic heat sources (such as operating construction machinery) at the construction site are clustered and labeled, thereby automatically identifying flight risk areas and obstacles that may interfere with the low-altitude cruise of the UAV.

[0152] In addition, abnormal fluctuations in heat sources (captured by infrared slave devices) can be used to help identify mobile devices with high heat radiation, avoiding path conflicts caused by moving obstacles.

[0153] S600. The DSM (Discrete Path Management System) model is compared and analyzed with historical flight paths and obstacle avoidance records to form an updated path planning scheme. In this step, it should be noted that a multi-dimensional comparative analysis is performed between the constructed DSM model and the historical flight paths, obstacle avoidance point sets, and flight interruption records stored in the database. By constructing a multi-objective constraint graph (with terrain feasibility, flight efficiency, energy consumption, and safety redundancy as objective functions), path optimization calculations based on graph search (such as the A algorithm or RRT tree) are performed to output the optimal path under the current spatiotemporal conditions.

[0154] At the same time, the system automatically assesses the likelihood of historical obstacle avoidance events recurring in the current environment, plans alternative avoidance strategies in advance, and marks them as "foreseeable risk segments" in the path, dynamically monitoring the risk level of these segments during subsequent flights.

[0155] S700: The path planning scheme and corresponding obstacle avoidance strategy are distributed to the UAV in template form. In this step, it should be noted that the path planning results and obstacle avoidance scheme are encapsulated in the form of a "flight strategy template," including but not limited to: target coordinates, altitude constraints, flight speed, emergency avoidance commands, trigger thresholds, and other information for each flight segment. This template structure facilitates direct reading and execution by the UAV at the edge control layer, improving runtime response efficiency.

[0156] Meanwhile, the template supports real-time cloud distribution or offline embedding into the flight control system, and can automatically switch according to the on-site network status, enhancing the platform's adaptability to the field.

[0157] The technical solution implemented in this embodiment can dynamically construct flight safety maps and identify flight risk areas through high-precision DSM models. Compared with static map solutions, it has stronger real-time performance and accuracy, and is particularly suitable for dynamic construction site environments.

[0158] By achieving intelligent coordination between path and obstacle avoidance strategies and combining historical data to form an experience-based flight optimization model, the obstacle avoidance response latency can be significantly reduced while ensuring flight efficiency.

[0159] The platform enhances operational autonomy and deployment flexibility. Flight strategy templates can be standardized for transmission and local operation, exhibiting high versatility and low coupling, which facilitates rapid migration and deployment of the system across different UAV models.

[0160] This embodiment further expands the application scenarios of DSM precision model in path planning and flight safety, forming a closed-loop data perception-graph construction-path optimization-command execution process, which significantly improves the UAV's operational capabilities in complex environments.

[0161] The embodiments described above are merely illustrative of several implementations of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the scope of protection of this application. Therefore, the scope of protection of this patent application should be determined by the appended claims.

Claims

1. A cloud-edge coordination and management platform, characterized in that, include: The drone equipment is equipped with a radar main unit and an infrared slave unit. The radar main unit is used to scan the terrain of the target area and acquire scan data; the infrared slave unit is used to lock onto the heat source area generated by the mechanical equipment, perform detailed supplementary shooting, and acquire detailed data. An edge control layer, set on the drone equipment, has a built-in lightweight Gaussian splash engine for preprocessing the scan data and detail data, marking abnormal areas where the image signal-to-noise ratio is lower than a preset threshold, and generating a DSM coarse model. The cloud-based control center receives the DSM rough model and optimizes it using a deployed engineering-aware data intelligent processing algorithm to generate a DSM fine model. The cloud-based control center further calculates the earthwork volume of the target area based on the DSM fine model and generates a construction heat map.

2. The cloud-edge coordination and management platform according to claim 1, characterized in that, The radar host is a millimeter-wave radar, used to penetrate suspended particles in the air and perceive terrain contour information in haze or low visibility environments; the millimeter-wave radar is combined with an acoustic array system, which includes multiple acoustic transmitting and receiving units, used to assist in calibrating the spatial positioning results of the millimeter-wave radar. The infrared slave device is a multispectral imager operating in the 8-14μm band, used to acquire image information in different bands; the multispectral imager has a built-in database of infrared reflectivity of metal materials, used to improve the multispectral imager's ability to identify the reflectivity characteristics of different metal surfaces. The PM2.5 sensor, which integrates temperature, humidity, and pressure, is used to detect the concentration of fine particulate matter, temperature, humidity, and air pressure parameters in the environment.

3. A cloud-edge coordination and management method, applied to the cloud-edge coordination and management platform as described in any one of claims 1-2, characterized in that, include: Generate and send out the drone's cruise path, control the drone to cruise along the cruise path, perform terrain scanning on the target area, collect terrain data, and generate a rough DSM model. The DSM rough model is preprocessed through an edge control layer and then uploaded to the cloud management center; The DSM coarse model is corrected for errors by using the engineering perception data intelligent processing algorithm deployed in the cloud control center to generate the DSM fine model; A construction heat map is generated based on the DSM precision model.

4. The cloud-edge coordinated management and control method according to claim 3, characterized in that, The acquired terrain data includes: terrain contour data obtained by scanning with a radar host, and images of the heat source area obtained by scanning with an infrared slave device.

5. The cloud-edge coordinated management and control method according to claim 4, characterized in that, The radar host includes a millimeter-wave radar and an acoustic array system. Accordingly, controlling the UAV to cruise along the cruise path includes: The millimeter-wave radar transmits frequency-modulated continuous waves in the target area to acquire terrain contour data; The ultrasonic array system emits ultrasonic pulses toward the target area and receives echo signals after encountering obstacles. The spatial location of the obstacle is calculated based on the sound wave flight time and temperature and humidity correction model. The obstacle position data measured by the acoustic array system is spatiotemporally aligned with the millimeter-wave radar data to calculate the ranging error; Based on the ranging error, the positioning error of the millimeter-wave radar is determined and corrected.

6. The cloud-edge coordinated management and control method according to claim 5, characterized in that, The process of determining and correcting the positioning error of the millimeter-wave radar includes: The obstacle distance data obtained by the millimeter-wave radar and the acoustic array system are given a unified timestamp, and the obstacle spatial position data determined by the acoustic array system in its own coordinate system is transformed into the coordinate system of the millimeter-wave radar through rigid coordinate transformation. Calculate the deviation between the obstacle distance data obtained by the millimeter-wave radar and the acoustic array system respectively: Obtain the PM2.5 concentration value at the corresponding time point and determine whether any of the following conditions are met: The error exceeds the preset threshold. PM2.5 concentrations were higher than the preset standards for dense fog environments; If any of the above conditions are met, the compensation mechanism will be triggered: If the smog interference is severe, the acoustic coordinates should be used directly to replace the corresponding point cloud coordinates. If the interference level is moderate, the radar and acoustic coordinates are weighted and fused according to preset weights to generate corrected point cloud coordinates: The corrected point cloud data will be used as the coordinate data for navigation.

7. The cloud-edge coordinated management and control method according to claim 3, characterized in that, The preprocessing of the DSM rough model through the edge control layer includes: Abnormal region marking: By performing signal-to-noise ratio analysis on the image or point cloud data in the DSM coarse model, regions with a signal-to-noise ratio lower than a preset threshold are identified and marked as abnormal regions. Data augmentation processing performs quality optimization operations on the DSM coarse model data, including but not limited to noise filtering, detail sharpening, and contrast adjustment; Data compression processing is performed to compress the preprocessed DSM coarse model data.

8. The cloud-edge coordinated management and control method according to claim 3, characterized in that, The intelligent processing algorithm for engineering sensing data is a haze scattering error compensation algorithm, and correspondingly, the error correction includes: Collect real-time PM2.5 concentration and atmospheric humidity; Based on a haze forecasting model, predict the height shift of point clouds; The height of the original point cloud is corrected using the height offset of the point cloud, thereby adjusting the height data in the point cloud.

9. The cloud-edge coordinated management and control method according to claim 8, characterized in that, The height offset of the predicted point cloud includes: The point cloud height is corrected using the following error correction formula: In the formula, The dust emission coefficient at the construction site. It is the scattering attenuation factor. The atmospheric humidity, the construction site dust coefficient, and the scattering attenuation factor are obtained through training with historical data.

10. The cloud-edge coordinated management and control method according to claim 3, characterized in that, The method further includes: Based on the latest terrain morphology and obstacle spatial distribution information contained in the DSM fine model, potential flight risk areas and obstacles within the construction area are identified. The DSM model is compared and analyzed with historical flight trajectories and obstacle avoidance records to form an updated path planning scheme; The path planning scheme and corresponding obstacle avoidance strategy are distributed to the drone in template form.

Images