Self-adaptive perception control method and system for dual-radar cooperative scheduling

Abstract

The invention relates to the technical field of intelligent control of engineering machinery, and provides a self-adaptive perception control method and system for dual-radar cooperative scheduling. According to the method, the first rotary radar and the second rotary radar are respectively installed on the rotatable support; obstacle distribution data of a tower crane hoisting operation area are obtained in real time through the first rotating radar and the second rotating radar, and real-time stage information of hoisting operation is obtained in real time through the operation state sensing module. Calculating a sensing dead angle position according to the real-time stage information; dynamically generating a first task instruction and a second task instruction according to the priority level, the sensing dead angle position and the obstacle distribution data of the target monitoring area, and controlling the first rotary radar and the second rotary radar to adjust the rotation angle and the sensing parameter, so that the first rotary radar performs emphasized sensing on the target monitoring area with high priority, and the second rotary radar performs emphasized sensing on the target monitoring area with high priority. And the second rotary radar performs compensation sensing on areas corresponding to the sensing blind area and the sensing dead angle of the first rotary radar.

Description

Technical Field

[0001] This application relates to the field of intelligent control technology for engineering machinery, and in particular to an adaptive perception control method and system for dual-radar coordinated scheduling. Background Technology

[0002] In the construction industry, the safety of tower crane hoisting operations depends on real-time perception of obstacles in the work area. Traditional technologies, such as single radar sensors or multi-radar systems arranged at fixed angles, have significant drawbacks: on the one hand, single radars are limited by detection angles and physical obstructions, making it difficult to cover complex work areas and resulting in large-area blind spots; on the other hand, fixed multi-radar systems typically employ preset scanning strategies and cannot adjust monitoring focus according to the dynamic stages of hoisting operations (such as lifting, rotation, and landing), leading to insufficient monitoring accuracy or wasted resources in high-risk areas (such as under the hook and along the boom's rotation path).

[0003] In existing technologies, some solutions attempt to improve sensing performance by increasing the number of radars or fixing the detection area, but they fail to address the following core issues:

[0004] 1. Lack of dynamic correlation between operation phase and monitoring area: The priority of target monitoring area is not dynamically adjusted according to the risk characteristics of different phases such as lifting, rotation, and landing. For example, high-risk areas under the hook are not scanned in the lifting phase, and the path ahead is not monitored in the rotation phase.

[0005] 2. Insufficient compensation mechanism for blind spots: Traditional multi-radar systems simply stitch together the detection range without calculating the blind spots based on the radar installation location and real-time angle. Furthermore, they lack a strategy for coordinated compensation between the main and auxiliary radars, resulting in the inability to effectively monitor obstructed areas or radar beam coverage blind spots.

[0006] 3. Lack of intelligent task allocation strategy: Radar task instructions are not dynamically generated by combining obstacle distribution data and priority rules, making it difficult to achieve a balance between detection accuracy and efficiency under different operating conditions. For example, the scanning frequency is not increased in high-risk areas, while there is redundant detection in low-risk areas.

[0007] Therefore, a method is urgently needed to solve at least one of the above problems. Summary of the Invention

[0008] This application provides an adaptive perception control method and system for dual-radar cooperative scheduling, aiming to solve the problem that existing technologies have not proposed a dual-radar cooperative scheduling method based on operation phase information, target area priority, and perception blind spot compensation, and especially lack a technical solution to achieve perception range coverage optimization and redundant monitoring of high-risk areas through dynamic task allocation strategies.

[0009] In a first aspect, embodiments of this application provide an adaptive perception control method for dual-radar cooperative scheduling; the method includes:

[0010] The first and second rotating radars are respectively mounted on a rotatable bracket to construct a dual-radar module sensing system. The dual radar modules are respectively connected to the sensing task scheduling unit and the operation status sensing module.

[0011] The first and second rotating radars acquire obstacle distribution data in the tower crane hoisting operation area in real time, and the operation status perception module acquires real-time stage information of the hoisting operation, including the lifting stage, rotation stage and landing stage.

[0012] Based on the real-time stage information and combined with the preset target area priority rules, the target monitoring area and the corresponding priority level corresponding to the current operation stage are determined. Based on the installation position and real-time rotation angle of the first and second rotating radars, the blind spot position is calculated. Based on the priority level of the target monitoring area, the blind spot position, and obstacle distribution data, the first task instruction and the second task instruction are dynamically generated according to the preset task allocation strategy.

[0013] According to the first and second task instructions, the first and second rotating radars are controlled to adjust their rotation angles and sensing parameters, so that the first rotating radar focuses on sensing high-priority target monitoring areas, and the second rotating radar compensates for sensing the areas corresponding to the sensing blind spots and dead angles of the first rotating radar, thereby achieving coverage optimization of the sensing range and redundant monitoring of high-risk areas.

[0014] In some embodiments, the real-time acquisition of obstacle distribution data in the tower crane hoisting operation area via the first rotating radar and the second rotating radar includes: performing time synchronization and spatial coordinate calibration on the raw point cloud data collected by the first rotating radar and the second rotating radar respectively; filtering out noise points based on a preset adaptive filtering algorithm; inputting the filtered point cloud data into a preset target classification neural network model to identify obstacle types and label static and dynamic obstacles; and generating structured distribution data containing obstacle location, type, and risk level based on the obstacle's position coordinates, size information, and motion state.

[0015] In some embodiments, the real-time acquisition of hoisting operation stage information through the operation state perception module includes: collecting data from the tower crane's hoisting distance encoder, slewing mechanism angle encoder, and hook height sensor; inputting multi-dimensional sensor signals into a preset state recognition finite state machine model; performing time-series analysis on the action sequence of the hoisting operation through the state machine model; and determining the current operation stage by combining the motion thresholds of each actuator; determining the hoisting stage when the hoisting motor speed changes abruptly from zero and the hook height changes; determining the rotation stage when the slewing mechanism angle change rate exceeds a preset threshold; and determining the landing stage when the hook height tends to stabilize and approaches the target height.

[0016] In some embodiments, determining the target monitoring area and corresponding priority level for the current operation stage based on the real-time stage information and a preset target area priority rule includes: setting a multi-area partitioning model that includes a lifting area, a rotation path area, and a landing point buffer zone; establishing a dynamic priority mapping table for different operation stages; when in the lifting stage, setting a 5-meter radius area below the hook as a first-level priority monitoring area and the area directly below the boom as a second-level priority area; when in the rotation stage, setting a 180-degree fan-shaped area in front of the boom in its current rotation direction as a first-level priority area and the 60-degree fan-shaped areas on both sides as second-level priority areas; and using a priority optimization neural network trained with historical accident data to correct the preset priority rules online and dynamically adjust the priority coefficients of each area according to the real-time obstacle distribution.

[0017] In some embodiments, calculating the location of the blind spot based on the installation position and real-time rotation angle of the first and second rotating radars includes: establishing a three-dimensional coordinate system with the tower crane's rotation center as the origin; obtaining the installation coordinates and orientation angle parameters of the first and second rotating radars; constructing a mathematical model of a fan-shaped detection area based on the radar sensor's detection distance threshold and beam angle range; projecting the detection range corresponding to the current rotation angle of the two radars onto the working area plane using a coordinate transformation algorithm; calculating the union and complement of the two radar detection areas; identifying uncovered blind spots; and, combined with an obstacle occlusion probability model, performing secondary screening on the occluded blind spots to determine the actual location coordinates of the blind spot.

[0018] In some embodiments, the step of dynamically generating a first task instruction and a second task instruction according to the priority level of the target monitoring area, the location of the blind spot, and the obstacle distribution data, and in accordance with a preset task allocation strategy, includes: constructing a radar task parameter space containing detection accuracy, scanning frequency, and angle adjustment step size; establishing an objective function with priority level as the weight; using an improved particle swarm optimization algorithm, with the constraints of covering the first-priority area, compensating for the blind spot, and balancing the radar load, to optimize and solve the task parameters of the two sets of radars; when the priority difference of the target monitoring area exceeds a preset threshold, assigning the first rotating radar to perform a high-frequency fine scanning task in the high-priority area, and the second rotating radar to perform a reciprocating blind spot filling scanning task in the blind spot area; when the priority difference is low, generating a cooperative scanning strategy through a task scheduling neural network, so that the two radars cover the area according to a preset phase difference.

[0019] In some embodiments, controlling the first and second rotating radars to adjust their rotation angles and sensing parameters according to the first and second task instructions includes: establishing a kinematic model of the radar servo system and converting the target angle in the task instructions into a motor control pulse sequence; using a PID control algorithm with feedforward compensation to perform closed-loop control of the radar rotation axis and correct angle deviations in real time; dynamically configuring the transmit and receive module parameters through the radar hardware driver interface according to the sensing parameters in the task instructions; and during the adjustment process, monitoring the angular velocity and acceleration of the rotating mechanism in real time through a radar self-test sensor and triggering a safety speed limit mechanism when abnormal vibration is detected; wherein the sensing parameters include detection distance, resolution, and pulse repetition frequency.

[0020] In some embodiments, the method further includes: establishing a historical operation data training set, including operation stages, obstacle distribution, radar scheduling strategies, and collision risk event labels; training a radar scheduling strategy network using a deep reinforcement learning algorithm, with the current operation status, target area priority, and blind spot information as state inputs, the combination of primary and secondary rotating radar task parameters as action outputs, and maximizing risk warning accuracy and minimizing scheduling delay as reward functions; and after each operation cycle is completed, inputting the deviation data between the actual scheduling effect and the preset strategy into the strategy network for online updates, thereby achieving adaptive optimization of the task allocation strategy.

[0021] In some embodiments, the method further includes: constructing a multi-dimensional risk assessment model that includes obstacle trajectory prediction, priority area risk level, and radar perception blind zone; using a long short-term memory network to model the position sequence of dynamic obstacles and predict their trajectory within the next 5 seconds; performing spatial overlap calculation between the predicted trajectory and the target monitoring area; and triggering an emergency scheduling mechanism when the predicted trajectory enters a first-level priority area and the current radar perception frequency is below a safety threshold: forcing the first rotating radar to switch to the full-time tracking and scanning mode of the target area, and simultaneously adjusting the second rotating radar to the forward warning scanning mode in the trajectory prediction direction.

[0022] Secondly, this application provides an adaptive perception control system with dual radar cooperative scheduling, applied to a controller, the system comprising:

[0023] The system construction unit is used to install the first rotating radar and the second rotating radar on a rotatable bracket to build a dual radar module sensing system. The dual radar modules are respectively connected to the sensing task scheduling unit and the operation status sensing module.

[0024] The data acquisition unit is used to acquire obstacle distribution data in the tower crane hoisting operation area in real time through the first rotating radar and the second rotating radar, and to acquire real-time stage information of the hoisting operation through the operation status perception module. The real-time stage information includes the hoisting stage, the rotation stage and the landing stage.

[0025] The instruction generation unit is used to determine the target monitoring area and its corresponding priority level corresponding to the current operation stage based on the real-time stage information and in combination with the preset target area priority rules; calculate the blind spot position based on the installation position and real-time rotation angle of the first and second rotating radars; and dynamically generate the first task instruction and the second task instruction according to the priority level of the target monitoring area, the blind spot position and obstacle distribution data, in accordance with the preset task allocation strategy.

[0026] The perception and monitoring unit is used to control the first rotating radar and the second rotating radar to adjust their rotation angles and perception parameters according to the first task instruction and the second task instruction. This enables the first rotating radar to focus on sensing high-priority target monitoring areas, and the second rotating radar to compensate for the sensing blind spots and dead angles of the first rotating radar, thereby achieving coverage optimization of the sensing range and redundant monitoring of high-risk areas.

[0027] This application provides an adaptive perception control method and system based on dual-radar collaborative scheduling. By acquiring real-time information on the lifting operation stages (lifting, rotation, and landing point), and dynamically determining the target monitoring area based on preset priority rules, the radar resources are focused on the current high-risk area (e.g., focusing on monitoring below the hook during the lifting stage, and prioritizing coverage of the path ahead during the rotation stage), solving the monitoring lag problem of traditional fixed strategies. Based on the radar installation position and real-time angle calculation of the perception blind spots, a collaborative strategy of primary and secondary rotating radars (the primary rotating radar focuses on perceiving high-priority areas, and the secondary rotating radar compensates for blind spots and dead angles) is adopted to achieve blind-spot-free coverage of the operation area and dual monitoring of high-risk areas, significantly reducing the risk of collisions. By dynamically generating primary and secondary rotating radar commands through a preset task allocation strategy, the detection accuracy (e.g., scanning frequency, angle step size) is matched with the monitoring requirements, improving safety performance while avoiding resource waste. Compared with traditional fixed parameter settings, this method has higher efficiency and flexibility.

[0028] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0029] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a schematic flowchart illustrating the steps of an adaptive perception control method with dual radar cooperative scheduling provided in an embodiment of this application;

[0031] Figure 2 This is a schematic diagram of the structure of a dual radar module sensing system provided in an embodiment of this application;

[0032] Figure 3 This is a schematic block diagram of an adaptive perception control system with dual radar cooperative scheduling provided in an embodiment of this application;

[0033] Figure 4 This is a schematic block diagram of the controller provided in one embodiment of this application.

[0034] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Detailed Implementation

[0035] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0036] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0037] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and the terms "first" and "second" are not necessarily different.

[0038] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0039] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0040] In the construction industry, the safety of tower crane hoisting operations depends on real-time perception of obstacles in the work area. Traditional technologies, such as single radar sensors or multi-radar systems arranged at fixed angles, have significant drawbacks: on the one hand, single radars are limited by detection angles and physical obstructions, making it difficult to cover complex work areas and resulting in large-area blind spots; on the other hand, fixed multi-radar systems typically employ preset scanning strategies and cannot adjust monitoring focus according to the dynamic stages of hoisting operations (such as lifting, rotation, and landing), leading to insufficient monitoring accuracy or wasted resources in high-risk areas (such as under the hook and along the boom's rotation path).

[0041] In existing technologies, some solutions attempt to improve sensing performance by increasing the number of radars or fixing the detection area, but they fail to address the following core issues:

[0042] 1. Lack of dynamic correlation between operation phase and monitoring area: The priority of target monitoring area is not dynamically adjusted according to the risk characteristics of different phases such as lifting, rotation, and landing. For example, high-risk areas under the hook are not scanned in the lifting phase, and the path ahead is not monitored in the rotation phase.

[0043] 2. Insufficient compensation mechanism for blind spots: Traditional multi-radar systems simply stitch together the detection range without calculating the blind spots based on the radar installation location and real-time angle. Furthermore, they lack a strategy for coordinated compensation between the main and auxiliary radars, resulting in the inability to effectively monitor obstructed areas or radar beam coverage blind spots.

[0044] 3. Lack of intelligent task allocation strategy: Radar task instructions are not dynamically generated by combining obstacle distribution data and priority rules, making it difficult to achieve a balance between detection accuracy and efficiency under different operating conditions. For example, the scanning frequency is not increased in high-risk areas, while there is redundant detection in low-risk areas.

[0045] Existing technologies have not proposed a dual-radar collaborative scheduling method based on operational phase information, target area priority, and perception blind spot compensation. In particular, they lack technical solutions for achieving perception range coverage optimization and redundant monitoring of high-risk areas through dynamic task allocation strategies.

[0046] To resolve the above issues, please refer to... Figure 1 This application provides an adaptive perception control method based on dual-radar cooperative scheduling, applied to a controller. It should be noted that all information involved in the method provided in this application is extracted with the authorization of the relevant user and in accordance with relevant regulations, and will not infringe on user privacy.

[0047] The provided adaptive perception control method for dual-radar cooperative scheduling includes steps S101 to S104. Details are as follows:

[0048] Step S101. Install the first rotating radar and the second rotating radar on the rotatable bracket respectively to construct a dual radar module sensing system. The dual radar modules are respectively connected to the sensing task scheduling unit and the operation status sensing module.

[0049] Specifically, two sets of rotatable radar sensors are installed at key locations on the tower crane (such as under the trolley). The radar angle is dynamically adjusted via a rotatable support (such as an electric pan-tilt head), thus constructing a dual-radar collaborative sensing hardware system with angle adjustment capabilities (such as...). Figure 2 (As shown). The two sets of radars are defined as the first rotating radar and the second rotating radar (not fixed main and auxiliary, but dynamically switchable). Both establish bidirectional communication with the sensing task scheduling unit (such as an industrial-grade PLC or edge computing controller) and the operation status sensing module (integrating multiple types of sensors) to realize data interaction and control command transmission.

[0050] The selected radar is a 120° field-of-view lidar (such as the TF-ALS-LIDAR-01 radar, with a single radar detection range ≥150 meters, accuracy ≤3 cm, and angular resolution ≤0.05°), meeting the obstacle detection requirements in complex environments. The two radars scan synchronously at a frequency of 10Hz. The first rotating radar is installed on the top of the tower crane cab, and the second rotating radar is installed below the front end of the boom, forming a staggered three-dimensional monitoring layout to reduce structural obstruction (such as the boom blocking the radar beam).

[0051] The radar and dispatch unit are connected via Ethernet (TCP / IP) to transmit point cloud data (including distance, angle, and speed information) in real time; the operation status perception module transmits operation stage signals (such as lifting / rotation / landing point status switch signals) via CAN bus or RS485 interface.

[0052] Step S102. Obstacle distribution data of the tower crane hoisting operation area are acquired in real time through the first rotating radar and the second rotating radar. Real-time stage information of the hoisting operation is acquired in real time through the operation status perception module. The real-time stage information includes the lifting stage, the rotation stage and the landing stage.

[0053] Specifically, the system uses dual radars to scan the work area in real time (within a radius of 50-100 meters) and generate point cloud data containing the location, size, and movement status of obstacles. At the same time, the system collects tower crane operating parameters through the work status perception module to identify the current stage of the hoisting operation (lifting, rotation, landing point) and provides input for dynamic task scheduling.

[0054] Obstacle data acquisition includes simultaneous scanning by dual radars at a frequency of 10Hz. The first rotating radar covers the 180° area in front of the tower crane (the rotation range of the boom) by default, and the second rotating radar covers the blind spots behind and on both sides by default. The raw point cloud data is filtered by Kalman to remove noise and generate a dynamic obstacle list (including coordinates X, Y, Z, velocity V, and category label).

[0055] During the operation phase identification, the operation status perception module integrates multiple types of sensors: Lifting phase: The hook height sensor (laser rangefinder) detects whether the hook is off the ground (height change rate > 5cm / s and load sensor value > 10% of rated load); Rotation phase: The slewing mechanism encoder (accuracy 0.5°) detects the boom rotation angle change rate > 1° / s; Landing phase: The hook height sensor detects that the hook is approaching the target height (error ±50cm) and the load sensor value decreases (releasing the load). Multi-sensor data fusion: A finite state machine (FSM) algorithm is used, combined with a time window (e.g., lasting 5 seconds) to confirm phase transitions, avoiding misjudgments.

[0056] Step S103. Based on the real-time stage information and combined with the preset target area priority rules, determine the target monitoring area and corresponding priority level corresponding to the current operation stage. Calculate the blind spot location based on the installation position and real-time rotation angle of the first and second rotating radars. Based on the priority level of the target monitoring area, the blind spot location, and obstacle distribution data, dynamically generate the first and second task instructions according to the preset task allocation strategy.

[0057] Specifically, in conjunction with the real-time operation phase, based on preset priority rules (such as prioritizing monitoring the 30° fan-shaped area below the hook during the lifting phase), the target monitoring area and its priority (high / medium / low) are determined; the perception blind spots under the current radar angle are calculated through geometric modeling (such as the blind spots of overlapping two radar beams, the area blocked by the tower crane structure); finally, based on the priority, blind spot location and obstacle distribution, the control commands such as the angle, scanning frequency, and detection accuracy of the primary and secondary rotating radars are generated through the task allocation algorithm.

[0058] The target area priority division includes: Lifting phase: High priority area is the 20-meter diameter circular area directly below the hook (where personnel are likely to gather), and the 10-meter range below the boom (to prevent collisions); Medium priority area is the 30-meter range around the tower crane; Rotation phase: High priority area is the 50-meter fan-shaped area in front of the boom (in front of the rotation path), and the 15° blind spots on both sides; Landing phase: High priority area is the 10-meter rectangular area around the target landing point (to prevent accidental collisions with obstacles). Priority rules are stored in the scheduling unit database and can be customized by the user.

[0059] The blind spot calculation is performed by establishing a three-dimensional model of the tower crane (including the boom, counterweight boom, tower body and other obstructing components). Based on the current radar angle (θ1, φ1) and beam range (120° horizontally, 15° vertically), the area obstructed by the tower crane structure is calculated using a ray tracing algorithm. The non-overlapping area of ​​the dual radar detection range (i.e., the single radar blind spot) is calculated and marked as the "compensation area".

[0060] The task allocation strategy includes: First rotating radar task: Prioritize scanning high-priority areas, adopting a "key area encrypted scanning" strategy (e.g., increasing the scanning frequency of high-priority areas to 100Hz and the angular resolution to 0.5°); Second rotating radar task: Cover the blind spots and perception dead zones of the first rotating radar, adopting a "dynamic blind spot filling scanning" strategy (e.g., performing reciprocating scanning on obscured areas at a scanning frequency of 50Hz); Based on the priority queue and ant colony optimization algorithm, radar resources are dynamically allocated to ensure a balance between detection accuracy and efficiency in high-risk areas.

[0061] Step S104. According to the first task instruction and the second task instruction, control the first rotating radar and the second rotating radar to adjust the rotation angle and sensing parameters, so that the first rotating radar focuses on sensing the high-priority target monitoring area, and the second rotating radar compensates for the sensing blind spot and the area corresponding to the sensing dead angle position of the first rotating radar, so as to realize the coverage optimization of the sensing range and the redundant monitoring of high-risk areas.

[0062] Specifically, according to the mission instructions, the first rotating radar adjusts its rotation angle to the target monitoring area (e.g., aligning with the area below the hook during the lifting phase) and improves the sensing parameters (e.g., increasing the transmission power and reducing the beamwidth); the second rotating radar simultaneously adjusts its angle to perform compensatory scanning on blind spots not covered by the first rotating radar (e.g., behind the crane arm) and calculated blind angle areas, forming a collaborative mechanism of "first rotating radar focusing on monitoring + second rotating radar filling in the blind spots and redundancy" to ensure that high-risk areas are double-covered (the detection ranges of the main and second rotating radars overlap by ≥30%).

[0063] The actuator control includes: the sensing task scheduling unit sends angle commands to the radar gimbal (e.g., the horizontal angle θ=45° and the elevation angle φ=10° of the first rotating radar), and achieves precise angle positioning (error ≤0.5°) through PID control algorithm; dynamically adjusts radar sensing parameters: "high-precision mode" is enabled in high-priority areas (pulse repetition frequency increases by 20%, signal-to-noise ratio is improved by 10dB), and "energy-saving mode" is enabled in low-priority areas (scanning frequency is reduced to 30Hz).

[0064] The collaborative compensation mechanism works by having the second rotating radar automatically cover the blind spot behind the first rotating radar when the first rotating radar monitors a high-priority area (e.g., the first rotating radar monitors 180° in front and the second rotating radar monitors 180° behind, with an overlap of 30°). If a blind spot is detected (e.g., a 20° blind spot caused by tower obstruction), the second rotating radar immediately starts a "blind spot compensation scan," performing a ±15° reciprocating scan centered on that area until the blind spot is eliminated (e.g., the tower crane rotates to change the obstruction angle).

[0065] Redundancy monitoring verification includes: cross-checking the detection data of the two radars in high-risk areas (e.g., if the first rotating radar detects obstacle A, the second rotating radar simultaneously confirms that the position error is ≤10cm) to ensure data reliability; when any radar fails, the other radar automatically switches to full-range scanning mode (covering a 180° field of view) and enters emergency backup mode.

[0066] In some embodiments, the real-time acquisition of obstacle distribution data in the tower crane hoisting operation area via the first rotating radar and the second rotating radar includes: performing time synchronization and spatial coordinate calibration on the raw point cloud data collected by the first rotating radar and the second rotating radar respectively; filtering out noise points based on a preset adaptive filtering algorithm; inputting the filtered point cloud data into a preset target classification neural network model to identify obstacle types and label static and dynamic obstacles; and generating structured distribution data containing obstacle location, type, and risk level based on the obstacle's position coordinates, size information, and motion state.

[0067] By performing spatiotemporal calibration, noise filtering, and target classification on the raw point cloud data from dual radars, structured data containing location, type, and risk level is generated, providing accurate input for subsequent task scheduling.

[0068] Spatiotemporal synchronization calibration includes: Time synchronization: Nanosecond-level calibration of dual radar sampling clocks is performed using PTP (Precise Time Protocol) to ensure that the timestamp error of point cloud data is <1μs; Spatial calibration: The transformation matrix (rotation matrix R, translation vector T) between the radar coordinate system and the tower crane global coordinate system is established through hand-eye calibration, with a calibration accuracy of ≤2cm.

[0069] The adaptive filtering uses an adaptive Kalman filter, which dynamically adjusts the process noise covariance matrix Q and the measurement noise covariance matrix R based on the environmental noise, and filters out non-target points such as raindrops and birds (retaining points with a signal-to-noise ratio > 15dB); for static clutter (such as fixed buildings), statistical filtering is used, and points that have not moved for 3 consecutive frames are judged as static background and removed.

[0070] Target Classification and Structuring: Neural Network Model: A lightweight 3D point cloud classification model (such as PointNet-Lite) is used. Input is a local point cloud with ≥100 points per frame, and output is the obstacle type (personnel / vehicles / scaffolding / slings, etc., with a classification accuracy ≥95%); Risk Level Labeling: Risk level labels are generated based on the distance between the obstacle and the tower crane (<10 meters is high risk, 10-30 meters is medium risk, >30 meters is low risk) and the movement speed (>5m / s is dynamic high risk); Structured Data Format: JSON format data is generated, containing {obstacle ID, coordinates (X,Y,Z), type, risk level, dimensions (length×width×height), velocity vector (Vx,Vy,Vz)}.

[0071] In some embodiments, the real-time acquisition of hoisting operation stage information through the operation state perception module includes: collecting data from the tower crane's hoisting distance encoder, slewing mechanism angle encoder, and hook height sensor; inputting multi-dimensional sensor signals into a preset state recognition finite state machine model; performing time-series analysis on the action sequence of the hoisting operation through the state machine model; and determining the current operation stage by combining the motion thresholds of each actuator; determining the hoisting stage when the hoisting motor speed changes abruptly from zero and the hook height changes; determining the rotation stage when the slewing mechanism angle change rate exceeds a preset threshold; and determining the landing stage when the hook height tends to stabilize and approaches the target height.

[0072] By using multi-sensor fusion and a finite state machine (FSM) model, the lifting, rotation, and landing stages of hoisting operations can be identified in real time, solving the problem of misjudgment caused by traditional single sensors.

[0073] The sensor configuration includes: hoisting motor speed: a magnetoelectric speed sensor (accuracy ±0.5%) is installed to collect motor encoder pulse signals in real time (resolution 1024 pulses / revolution); slewing mechanism angle: an absolute encoder (accuracy ±0.1°) is used to record the real-time azimuth angle of the boom; hook height: a laser rangefinder (range 200 meters, accuracy ±1 cm) is used to measure the vertical distance from the hook to the reference plane at the bottom of the tower crane.

[0074] The state machine model construction includes: State transition conditions: Lifting phase: rotation speed > 0 rpm for 2 seconds, and hook height change rate > 2 cm / s (to prevent misjudgment of idle rotation); Rotation phase: angle change rate > 0.5° / s for 3 seconds (excluding fine-tuning actions); Landing phase: height change rate < 1 cm / s and distance from the target height (preset by the operation planning system) error < 50 cm. Timing analysis: A sliding time window (5 seconds) is used to verify state continuity and avoid instantaneous signal interference (such as sudden changes in rotation speed caused by brief current fluctuations).

[0075] In some embodiments, determining the target monitoring area and corresponding priority level for the current operation stage based on the real-time stage information and a preset target area priority rule includes: setting a multi-area partitioning model that includes a lifting area, a rotation path area, and a landing point buffer zone; establishing a dynamic priority mapping table for different operation stages; when in the lifting stage, setting a 5-meter radius area below the hook as a first-level priority monitoring area and the area directly below the boom as a second-level priority area; when in the rotation stage, setting a 180-degree fan-shaped area in front of the boom in its current rotation direction as a first-level priority area and the 60-degree fan-shaped areas on both sides as second-level priority areas; and using a priority optimization neural network trained with historical accident data to correct the preset priority rules online and dynamically adjust the priority coefficients of each area according to the real-time obstacle distribution.

[0076] By establishing a multi-regional priority model and dynamically optimizing the priority of monitoring areas based on historical accident data, the problem that traditional fixed priorities cannot adapt to complex working conditions can be solved.

[0077] The preset area division and priority mapping include: a multi-area model: Lifting area: a cylinder with a radius of 5 meters centered on the hook (Level 1 priority, focusing on monitoring personnel intrusion); Rotation path area: a fan-shaped area with a radius of 50 meters and a 180° angle in front of the boom (Level 1 priority, preventing collisions with obstacles); Landing point buffer zone: a rectangular area of ​​10 meters × 10 meters centered on the target landing point (Level 1 priority, preventing the suspended object from touching surrounding structures). A dynamic mapping table stores the area priorities for each stage (e.g., the 60° fan-shaped areas on both sides during the rotation stage are Level 2 priority, focusing on monitoring adjacent tower cranes).

[0078] Priority online correction includes: Training data: collecting obstacle distribution data in historical collision accidents (e.g., 80% of accidents occur within 3 meters below the hook), constructing a priority optimization neural network (input: obstacle type + distance + historical accident weights, output: priority coefficient adjustment value); Real-time adjustment: when a high-risk obstacle (e.g., a moving person) is detected in a certain area 3 times in a row, the priority of that area is automatically increased by 1 level (up to 3 levels) and maintained for 10 minutes.

[0079] In some embodiments, calculating the location of the blind spot based on the installation position and real-time rotation angle of the first and second rotating radars includes: establishing a three-dimensional coordinate system with the tower crane's rotation center as the origin; obtaining the installation coordinates and orientation angle parameters of the first and second rotating radars; constructing a mathematical model of a fan-shaped detection area based on the radar sensor's detection distance threshold and beam angle range; projecting the detection range corresponding to the current rotation angle of the two radars onto the working area plane using a coordinate transformation algorithm; calculating the union and complement of the two radar detection areas; identifying uncovered blind spots; and, combined with an obstacle occlusion probability model, performing secondary screening on the occluded blind spots to determine the actual location coordinates of the blind spot.

[0080] By using a three-dimensional coordinate system and geometric modeling, the blind zones of dual radar detection and the areas obstructed by tower crane structures are accurately calculated, solving the coverage gap problem of traditional fixed-segmentation radar. The coordinate system and detection model include:

[0081] Global coordinate system: O is the tower crane's slewing center, the X-axis points to the initial azimuth of the boom, the Y-axis is horizontal and perpendicular to the X-axis, and the Z-axis is vertically upward; Radar detection model: The detection range of a single radar is a fan-shaped area, mathematically expressed as:

[0082] ;

[0083] Where Rmax = 100 meters, horizontal beam angle α = 120°, and vertical beam angle β = 15°. The current horizontal / tilt angle. The blind spot identification algorithm includes: Projection transformation: Projecting the 3D detection area onto the working plane (Z=0) to generate a 2D fan-shaped coverage area; Blind spot calculation: Calculating the union and complement of the dual radar coverage areas (i.e., the uncovered area) through polygon Boolean operations, and combining it with the tower crane structure CAD model (such as the area obstructed by the crane boom), using the ray casting method to determine whether the area is obstructed; Secondary filtering: Marking blind spots with an obstruction probability >70% as "actual perception blind spots", and outputting a coordinate list (e.g., the blind spot 20-30 meters behind the tower crane caused by the counterweight boom obstruction).

[0084] In some embodiments, the step of dynamically generating a first task instruction and a second task instruction according to the priority level of the target monitoring area, the location of the blind spot, and the obstacle distribution data, and in accordance with a preset task allocation strategy, includes: constructing a radar task parameter space containing detection accuracy, scanning frequency, and angle adjustment step size; establishing an objective function with priority level as the weight; using an improved particle swarm optimization algorithm, with the constraints of covering the first-priority area, compensating for the blind spot, and balancing the radar load, to optimize and solve the task parameters of the two sets of radars; when the priority difference of the target monitoring area exceeds a preset threshold, assigning the first rotating radar to perform a high-frequency fine scanning task in the high-priority area, and the second rotating radar to perform a reciprocating blind spot filling scanning task in the blind spot area; when the priority difference is low, generating a cooperative scanning strategy through a task scheduling neural network, so that the two radars cover the area according to a preset phase difference.

[0085] By constructing a radar mission parameter optimization model and combining particle swarm optimization and neural networks, dynamic mission allocation is achieved, balancing detection accuracy and efficiency.

[0086] The task parameter space and objective function include: Parameter space: detection accuracy (range resolution Δd, angular resolution Δθ), scanning frequency f (30-100Hz), angle adjustment step size Δs (0.1°-5°); Objective function includes:

[0087] ;

[0088] Where wi is the regional priority weight, Pi is the regional coverage, and λ is the load balancing coefficient to avoid excessive energy consumption caused by high-frequency scanning of dual radars.

[0089] The optimized algorithm implementation includes: improved particle swarm optimization algorithm: introducing dynamic adjustment of inertial weights (weight 0.8 in the lifting phase to improve convergence speed; weight 0.5 in the landing phase to enhance local search), and constraints including: coverage of the first priority area ≥95%, radar load difference ≤20%; task scheduling neural network: input the current priority difference, obstacle density, and number of blind spots, and output the phase difference between the main and second rotating radars (e.g., a phase difference of 60° to achieve alternating scanning and reduce data redundancy).

[0090] In some embodiments, controlling the first and second rotating radars to adjust their rotation angles and sensing parameters according to the first and second task instructions includes: establishing a kinematic model of the radar servo system and converting the target angle in the task instructions into a motor control pulse sequence; using a PID control algorithm with feedforward compensation to perform closed-loop control of the radar rotation axis and correct angle deviations in real time; dynamically configuring the transmit and receive module parameters through the radar hardware driver interface according to the sensing parameters in the task instructions; and during the adjustment process, monitoring the angular velocity and acceleration of the rotating mechanism in real time through a radar self-test sensor and triggering a safety speed limit mechanism when abnormal vibration is detected; wherein the sensing parameters include detection distance, resolution, and pulse repetition frequency.

[0091] The radar angle is precisely adjusted by using kinematic models and PID control, and sensing parameters are dynamically configured. At the same time, a safety monitoring mechanism is integrated to prevent mechanical failures.

[0092] Servo control includes: Kinematic model: Establish the relationship between the number of motor pulses and the angle of the two axes of the gimbal (horizontal H, pitch P) (e.g., 1 pulse = 0.05°), and convert the target angle θtar to the number of pulses N = θtar / 0.05; PID control: Use position loop PID + speed feedforward compensation, where the angle tracking error is ≤0.3° and the response time is <1 second.

[0093] The sensing parameter configuration includes: Hardware driver: Configure radar transmission module parameters via SPI interface, such as setting the pulse repetition frequency (PRF) to 20kHz in high-priority areas (to improve distance resolution to 3cm) and PRF to 10kHz in low-priority areas (to reduce power consumption); Safety mechanism: Install vibration sensor (accuracy ±0.1g). When vibration acceleration >1.5g is detected, a speed limiting mechanism (angular velocity ≤5° / s) is triggered to prevent the gimbal from shaking violently.

[0094] In some embodiments, the method further includes: establishing a historical operation data training set, including operation stages, obstacle distribution, radar scheduling strategies, and collision risk event labels; training a radar scheduling strategy network using a deep reinforcement learning algorithm, with the current operation status, target area priority, and blind spot information as state inputs, the combination of primary and secondary rotating radar task parameters as action outputs, and maximizing risk warning accuracy and minimizing scheduling delay as reward functions; and after each operation cycle is completed, inputting the deviation data between the actual scheduling effect and the preset strategy into the strategy network for online updates, thereby achieving adaptive optimization of the task allocation strategy.

[0095] By leveraging deep reinforcement learning (DRL) to learn the optimal scheduling strategy from historical data, the adaptive evolution of task allocation strategies is achieved, solving the problem that preset rules cannot cover complex working conditions.

[0096] The training framework includes: State space: S = {Operation stage, priority area list, blind spot coordinates, obstacle risk distribution} (discreteized, e.g., stage encoding is 0-2, area priority is normalized to [0,1]); Action space: A = {First rotating radar parameter combination, second rotating radar parameter combination} (a total of 12 parameter combinations, e.g., {Angle 1, Frequency 1, Accuracy 1}); Reward function: R = 0.6 Risk warning accuracy − 0.3 Scheduling delay − 0.1; Radar energy consumption. Where, risk warning accuracy = number of correct warnings / total number of risk events, scheduling delay = time from identification to execution (seconds).

[0097] The PPO (Proximal Policy Optimization) algorithm is used, and the policy network is updated every 50 job cycles (about 2 hours). The experience replay buffer has a capacity of 100,000 entries. In the initial stage, actions are generated using preset rules, and after accumulating 200 valid data entries, the algorithm switches to a reinforcement learning policy.

[0098] In some embodiments, the method further includes: constructing a multi-dimensional risk assessment model that includes obstacle trajectory prediction, priority area risk level, and radar perception blind zone; using a long short-term memory network to model the position sequence of dynamic obstacles and predict their trajectory within the next 5 seconds; performing spatial overlap calculation between the predicted trajectory and the target monitoring area; and triggering an emergency scheduling mechanism when the predicted trajectory enters a first-level priority area and the current radar perception frequency is below a safety threshold: forcing the first rotating radar to switch to the full-time tracking and scanning mode of the target area, and simultaneously adjusting the second rotating radar to the forward warning scanning mode in the trajectory prediction direction.

[0099] By using LSTM to predict the trajectory of dynamic obstacles and combining it with a risk assessment model to trigger emergency dispatch, the ability to respond to sudden risks is improved.

[0100] The risk assessment model includes: LSTM trajectory prediction: Input the obstacle coordinate sequence of the past 10 time steps (200ms), output the position prediction for the next 5 seconds (25 steps) (root mean square error ≤ 5cm), model structure: 2 layers of LSTM (128 units each) + fully connected layer; Spatial overlap calculation: Perform polygon overlap judgment between the predicted trajectory and the first priority area. When the overlap area is greater than 30% of the area and the current radar scanning frequency is less than 80Hz, emergency dispatch is triggered.

[0101] The emergency dispatch mechanism includes: First rotating radar: forcibly switching to full-time tracking and scanning of the target area (frequency 100Hz, angular resolution 0.5°), suspending other low-priority tasks; Second rotating radar: turning to a position 5 meters ahead of the predicted trajectory to conduct a forward warning scan (scanning range ±30°, frequency 60Hz), forming a dual protection of "tracking + warning".

[0102] In some embodiments, by abstracting two groups of radars as agents, a distributed multi-agent reinforcement learning (MARL) system is constructed, enabling the radars to autonomously optimize their detection strategies in dynamic cooperation, solving the communication delay and load imbalance problems of traditional centralized scheduling, and achieving real-time collaborative efficiency improvement.

[0103] The modeling of intelligent agents includes: State space: Each agent's state includes its own detection parameters (angle, frequency), global obstacle distribution (compressed into a grid heatmap), and the opponent's load status (CPU utilization, scan queue length); Action space: Includes angle adjustment (±5° step size), frequency switching (3 levels: low / medium / high), and precision configuration (2 levels: normal / fine), for a total of 12 combined actions.

[0104] The reward function design includes: joint reward Rjoint = 0.5Rcoverage + 0.3Rbalance − 0.2RCollision; where Rcoverage is the first-level area coverage (reward +1 for every 1% improvement), Rbalance is the load balancing (penalty of -5 when the frequency difference between the two radars is >20Hz), and Rcollision is the penalty of -10 when the overlap rate of the detection area is >30%. The QMIX (Quantile Mixing) algorithm is used to handle non-stationary cooperative environments. A central coordination network fuses the local Q-values ​​of each agent, and an experience replay buffer supports cross-agent data sharing. In the early stages of training, rule-guided exploration is used (ε=0.3), and after 3000 episodes, convergence to a stable policy is achieved, reducing the scanning blind zone by 60% in dual-radar cooperation.

[0105] Please see Figure 3 As shown, Figure 3This is a schematic diagram of the structure of the adaptive perception control system 200 with dual-radar cooperative scheduling provided in this application embodiment. The adaptive perception control system 200 with dual-radar cooperative scheduling is used to execute the steps of the adaptive perception control method with dual-radar cooperative scheduling shown in the above embodiments. The adaptive perception control system 200 with dual-radar cooperative scheduling can be a single server or a server cluster, or it can be a terminal, such as a handheld terminal, a laptop computer, a wearable device, or a robot.

[0106] like Figure 3 As shown, the adaptive perception control system 200 with dual radar coordinated scheduling includes:

[0107] System building unit 201 is used to install the first rotating radar and the second rotating radar on a rotatable bracket to build a dual radar module sensing system. The dual radar modules are respectively connected to the sensing task scheduling unit and the operation status sensing module.

[0108] The data acquisition unit 202 is used to acquire obstacle distribution data in the tower crane hoisting operation area in real time through the first rotating radar and the second rotating radar, and to acquire real-time stage information of the hoisting operation through the operation status perception module. The real-time stage information includes the hoisting stage, the rotation stage and the landing stage.

[0109] The instruction generation unit 203 is used to determine the target monitoring area and the corresponding priority level corresponding to the current operation stage based on the real-time stage information and in combination with the preset target area priority rules; calculate the sensing blind spot position based on the installation position and real-time rotation angle of the first rotating radar and the second rotating radar; and dynamically generate the first task instruction and the second task instruction according to the priority level of the target monitoring area, the sensing blind spot position and obstacle distribution data, and in accordance with the preset task allocation strategy.

[0110] The perception and monitoring unit 204 is used to control the first rotating radar and the second rotating radar to adjust their rotation angles and perception parameters according to the first task instruction and the second task instruction. This enables the first rotating radar to focus on sensing high-priority target monitoring areas, and the second rotating radar to compensate for the sensing blind spots and dead angles of the first rotating radar, thereby achieving coverage optimization of the sensing range and redundant monitoring of high-risk areas.

[0111] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the adaptive perception control system and its modules described above for dual-radar coordinated scheduling can be found in the corresponding embodiments of the adaptive perception control method for dual-radar coordinated scheduling, and will not be repeated here.

[0112] The aforementioned adaptive perception control method based on dual-radar cooperative scheduling can be implemented as a computer program, which can be used in, for example... Figure 3 It runs on the device shown.

[0113] Please see Figure 4 , Figure 4 This is a schematic block diagram of the controller provided in an embodiment of this application. The controller includes a processor, a memory, and a network interface connected via a device bus, wherein the memory may include a storage medium and internal memory.

[0114] The storage medium can store operating devices and computer programs. The computer program includes program instructions that, when executed, cause the processor to perform any adaptive perception control method based on dual-radar cooperative scheduling.

[0115] The processor provides computing and control capabilities to support the operation of the entire controller.

[0116] The internal memory provides an environment for the execution of computer programs in non-volatile storage media. When the computer program is executed by the processor, it enables the processor to execute any adaptive perception control method of dual radar cooperative scheduling.

[0117] This network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the terminal to which the present application is applied. The specific controller may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0118] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be 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. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0119] In one embodiment, the processor is configured to run a computer program stored in memory to perform the following steps:

[0120] The first and second rotating radars are respectively mounted on a rotatable bracket to construct a dual-radar module sensing system. The dual radar modules are respectively connected to the sensing task scheduling unit and the operation status sensing module.

[0121] The first and second rotating radars acquire obstacle distribution data in the tower crane hoisting operation area in real time, and the operation status perception module acquires real-time stage information of the hoisting operation, including the lifting stage, rotation stage and landing stage.

[0122] Based on the real-time stage information and combined with the preset target area priority rules, the target monitoring area and the corresponding priority level corresponding to the current operation stage are determined. Based on the installation position and real-time rotation angle of the first and second rotating radars, the blind spot position is calculated. Based on the priority level of the target monitoring area, the blind spot position, and obstacle distribution data, the first task instruction and the second task instruction are dynamically generated according to the preset task allocation strategy.

[0123] According to the first and second task instructions, the first and second rotating radars are controlled to adjust their rotation angles and sensing parameters, so that the first rotating radar focuses on sensing high-priority target monitoring areas, and the second rotating radar compensates for sensing the areas corresponding to the sensing blind spots and dead angles of the first rotating radar, thereby achieving coverage optimization of the sensing range and redundant monitoring of high-risk areas.

[0124] In some embodiments, the real-time acquisition of obstacle distribution data in the tower crane hoisting operation area via the first rotating radar and the second rotating radar includes: performing time synchronization and spatial coordinate calibration on the raw point cloud data collected by the first rotating radar and the second rotating radar respectively; filtering out noise points based on a preset adaptive filtering algorithm; inputting the filtered point cloud data into a preset target classification neural network model to identify obstacle types and label static and dynamic obstacles; and generating structured distribution data containing obstacle location, type, and risk level based on the obstacle's position coordinates, size information, and motion state.

[0125] In some embodiments, the real-time acquisition of hoisting operation stage information through the operation state perception module includes: collecting data from the tower crane's hoisting distance encoder, slewing mechanism angle encoder, and hook height sensor; inputting multi-dimensional sensor signals into a preset state recognition finite state machine model; performing time-series analysis on the action sequence of the hoisting operation through the state machine model; and determining the current operation stage by combining the motion thresholds of each actuator; determining the hoisting stage when the hoisting motor speed changes abruptly from zero and the hook height changes; determining the rotation stage when the slewing mechanism angle change rate exceeds a preset threshold; and determining the landing stage when the hook height tends to stabilize and approaches the target height.

[0126] In some embodiments, determining the target monitoring area and corresponding priority level for the current operation stage based on the real-time stage information and a preset target area priority rule includes: setting a multi-area partitioning model that includes a lifting area, a rotation path area, and a landing point buffer zone; establishing a dynamic priority mapping table for different operation stages; when in the lifting stage, setting a 5-meter radius area below the hook as a first-level priority monitoring area and the area directly below the boom as a second-level priority area; when in the rotation stage, setting a 180-degree fan-shaped area in front of the boom in its current rotation direction as a first-level priority area and the 60-degree fan-shaped areas on both sides as second-level priority areas; and using a priority optimization neural network trained with historical accident data to correct the preset priority rules online and dynamically adjust the priority coefficients of each area according to the real-time obstacle distribution.

[0127] In some embodiments, calculating the location of the blind spot based on the installation position and real-time rotation angle of the first and second rotating radars includes: establishing a three-dimensional coordinate system with the tower crane's rotation center as the origin; obtaining the installation coordinates and orientation angle parameters of the first and second rotating radars; constructing a mathematical model of a fan-shaped detection area based on the radar sensor's detection distance threshold and beam angle range; projecting the detection range corresponding to the current rotation angle of the two radars onto the working area plane using a coordinate transformation algorithm; calculating the union and complement of the two radar detection areas; identifying uncovered blind spots; and, combined with an obstacle occlusion probability model, performing secondary screening on the occluded blind spots to determine the actual location coordinates of the blind spot.

[0128] In some embodiments, the step of dynamically generating a first task instruction and a second task instruction according to the priority level of the target monitoring area, the location of the blind spot, and the obstacle distribution data, and in accordance with a preset task allocation strategy, includes: constructing a radar task parameter space containing detection accuracy, scanning frequency, and angle adjustment step size; establishing an objective function with priority level as the weight; using an improved particle swarm optimization algorithm, with the constraints of covering the first-priority area, compensating for the blind spot, and balancing the radar load, to optimize and solve the task parameters of the two sets of radars; when the priority difference of the target monitoring area exceeds a preset threshold, assigning the first rotating radar to perform a high-frequency fine scanning task in the high-priority area, and the second rotating radar to perform a reciprocating blind spot filling scanning task in the blind spot area; when the priority difference is low, generating a cooperative scanning strategy through a task scheduling neural network, so that the two radars cover the area according to a preset phase difference.

[0129] In some embodiments, controlling the first and second rotating radars to adjust their rotation angles and sensing parameters according to the first and second task instructions includes: establishing a kinematic model of the radar servo system and converting the target angle in the task instructions into a motor control pulse sequence; using a PID control algorithm with feedforward compensation to perform closed-loop control of the radar rotation axis and correct angle deviations in real time; dynamically configuring the transmit and receive module parameters through the radar hardware driver interface according to the sensing parameters in the task instructions; and during the adjustment process, monitoring the angular velocity and acceleration of the rotating mechanism in real time through a radar self-test sensor and triggering a safety speed limit mechanism when abnormal vibration is detected; wherein the sensing parameters include detection distance, resolution, and pulse repetition frequency.

[0130] In some embodiments, the method further includes: establishing a historical operation data training set, including operation stages, obstacle distribution, radar scheduling strategies, and collision risk event labels; training a radar scheduling strategy network using a deep reinforcement learning algorithm, with the current operation status, target area priority, and blind spot information as state inputs, the combination of primary and secondary rotating radar task parameters as action outputs, and maximizing risk warning accuracy and minimizing scheduling delay as reward functions; and after each operation cycle is completed, inputting the deviation data between the actual scheduling effect and the preset strategy into the strategy network for online updates, thereby achieving adaptive optimization of the task allocation strategy.

[0131] In some embodiments, the method further includes: constructing a multi-dimensional risk assessment model that includes obstacle trajectory prediction, priority area risk level, and radar perception blind zone; using a long short-term memory network to model the position sequence of dynamic obstacles and predict their trajectory within the next 5 seconds; performing spatial overlap calculation between the predicted trajectory and the target monitoring area; and triggering an emergency scheduling mechanism when the predicted trajectory enters a first-level priority area and the current radar perception frequency is below a safety threshold: forcing the first rotating radar to switch to the full-time tracking and scanning mode of the target area, and simultaneously adjusting the second rotating radar to the forward warning scanning mode in the trajectory prediction direction.

[0132] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An adaptive sensing control method for dual-radar cooperative scheduling, characterized in that, include: The first and second rotating radars are respectively mounted on a rotatable bracket to construct a dual-radar module sensing system. The dual radar modules are respectively connected to the sensing task scheduling unit and the operation status sensing module. The first and second rotating radars acquire obstacle distribution data in the tower crane hoisting operation area in real time, and the operation status perception module acquires real-time stage information of the hoisting operation, including the lifting stage, rotation stage and landing stage. Based on the real-time stage information and combined with the preset target area priority rules, the target monitoring area and the corresponding priority level corresponding to the current operation stage are determined. Based on the installation position and real-time rotation angle of the first and second rotating radars, the blind spot position is calculated. Based on the priority level of the target monitoring area, the blind spot position, and obstacle distribution data, the first task instruction and the second task instruction are dynamically generated according to the preset task allocation strategy. According to the first and second task instructions, the first and second rotating radars are controlled to adjust their rotation angles and sensing parameters, so that the first rotating radar focuses on sensing high-priority target monitoring areas, and the second rotating radar compensates for sensing the areas corresponding to the sensing blind spots and dead angles of the first rotating radar, thereby achieving coverage optimization of the sensing range and redundant monitoring of high-risk areas.

2. The method according to claim 1, characterized in that, The real-time acquisition of obstacle distribution data in the tower crane hoisting operation area via the first and second rotating radars includes: The raw point cloud data collected by the first and second rotating radars are synchronized in time and calibrated in space coordinates, and noise points are filtered out based on a preset adaptive filtering algorithm. The filtered point cloud data is input into a preset target classification neural network model to identify obstacle types and label static and dynamic obstacles; Based on the location coordinates, size information, and motion status of obstacles, structured distribution data containing the location, type, and risk level of obstacles is generated.

3. The method according to claim 1, characterized in that, The real-time acquisition of hoisting operation stage information through the operation status sensing module includes: Collect data from the tower crane's hoisting sensors, slewing mechanism angle encoders, and hook height sensors, and input the multi-dimensional sensor signals into a preset state recognition finite state machine model; The timing analysis of the action sequence of the hoisting operation is performed by using a state machine model, and the current operation stage is determined by combining the motion thresholds of each actuator. When a continuous change in hook height is detected, it is determined to be the lifting stage; when the rate of change of the slewing mechanism angle exceeds a preset threshold, it is determined to be the rotation stage; and when the hook height tends to stabilize and approaches the target height, it is determined to be the landing stage.

4. The method according to claim 1, characterized in that, Based on the real-time stage information and combined with preset target area priority rules, the target monitoring area and corresponding priority level corresponding to the current operation stage are determined, including: A multi-area partitioning model is pre-defined, including a lifting area, a rotation path area, and a landing point buffer zone, and a dynamic priority mapping table is established for different operation stages; During the lifting phase, the area within a 5-meter radius below the hook is designated as the first-priority monitoring zone, and the area directly below the boom is designated as the second-priority zone. When in the rotation phase, the 180-degree sector area in front of the current rotation direction of the crane boom is set as the first priority, and the 60-degree sector areas on both sides are set as the second priority. A priority optimization neural network trained using historical accident data is used to correct preset priority rules online and dynamically adjust the priority coefficients of each area based on the real-time obstacle distribution.

5. The method according to claim 1, characterized in that, Based on the installation positions and real-time rotation angles of the first and second rotating radars, the location of the blind spot is calculated, including: Establish a three-dimensional coordinate system with the tower crane's rotation center as the origin, and obtain the installation coordinates and orientation angle parameters of the first and second rotating radars; Based on the detection range threshold and beam angle range of the radar sensor, a mathematical model of the sector detection area is constructed; through a coordinate transformation algorithm, the detection range corresponding to the current rotation angle of the two radars is projected onto the working area plane. The union and complement of the detection areas of the two radars are calculated to identify the uncovered blind areas. Combined with the obstacle occlusion probability model, the obscured detection blind areas are further filtered to determine the coordinates of the actual sensing blind spots.

6. The method according to claim 1, characterized in that, The first task instruction and the second task instruction are dynamically generated according to the priority level of the target monitoring area, the location of blind spots, and the distribution data of obstacles, in accordance with a preset task allocation strategy, including: Construct a radar mission parameter space that includes detection accuracy, scanning frequency, and angle adjustment step size, and establish an objective function with priority level as the weight; An improved particle swarm optimization algorithm was used to optimize the mission parameters of the two sets of radars under the constraints of covering the first-priority area, compensating for blind spots, and balancing radar load. When the priority difference of the target monitoring area exceeds the preset threshold, the first rotating radar is assigned to perform a high-frequency fine scanning task of the high-priority area, and the second rotating radar performs a reciprocating blind spot scanning task of the blind spot area. When the priority difference is low, a collaborative scanning strategy is generated through a task scheduling neural network, enabling the two radars to cover the area according to a preset phase difference.

7. The method according to claim 1, characterized in that, The step of controlling the first rotating radar and the second rotating radar to adjust their rotation angles and sensing parameters according to the first task instruction and the second task instruction includes: A kinematic model of the radar servo system is established, and the target angle in the mission command is converted into a motor control pulse sequence. A PID control algorithm with feedforward compensation is used to perform closed-loop control of the radar rotation axis and correct the angle deviation in real time. According to the sensing parameters in the mission command, the parameters of the transmit and receive modules are dynamically configured through the radar hardware driver interface. During the adjustment process, the angular velocity and acceleration of the rotating mechanism are monitored in real time by radar self-test sensors. When abnormal vibration is detected, a safety speed limiting mechanism is triggered. The sensing parameters include detection distance, resolution, and pulse repetition frequency.

8. The method according to claim 1, characterized in that, The method further includes: Establish a training set of historical operation data, including operation phases, obstacle distribution, radar scheduling strategies, and collision risk event labels; The radar scheduling strategy network is trained by deep reinforcement learning algorithm. The current operation status, target area priority, and blind spot information are used as state inputs, the combination of primary and secondary rotating radar mission parameters is used as action output, and the reward function is to maximize the risk warning accuracy and minimize the scheduling delay. After each job cycle is completed, the deviation data between the actual scheduling effect and the preset strategy is input into the strategy network for online updates, thereby achieving adaptive optimization of the task allocation strategy.

9. The method according to claim 1, characterized in that, The method further includes: Construct a multi-dimensional risk assessment model that includes obstacle trajectory prediction, priority area risk level, and radar perception blind spot; A long short-term memory network is used to model the position sequence of dynamic obstacles and predict their trajectory within the next 5 seconds. The predicted trajectory is spatially overlapped with the target monitoring area. When the predicted trajectory enters the first-priority area and the radar's current sensing frequency is below the safety threshold, an emergency dispatch mechanism is triggered: the first rotating radar is forced to switch to the full-time tracking and scanning mode of the target area, and the second rotating radar is simultaneously adjusted to the forward warning scanning mode of the trajectory prediction direction.

10. An adaptive sensing control system with dual radar cooperative scheduling, characterized in that, include: The system construction unit is used to install the first rotating radar and the second rotating radar on a rotatable bracket to build a dual radar module sensing system. The dual radar modules are respectively connected to the sensing task scheduling unit and the operation status sensing module. The data acquisition unit is used to acquire obstacle distribution data in the tower crane hoisting operation area in real time through the first rotating radar and the second rotating radar, and to acquire real-time stage information of the hoisting operation through the operation status perception module. The real-time stage information includes the hoisting stage, the rotation stage and the landing stage. The instruction generation unit is used to determine the target monitoring area and its corresponding priority level corresponding to the current operation stage based on the real-time stage information and in combination with the preset target area priority rules; calculate the blind spot position based on the installation position and real-time rotation angle of the first and second rotating radars; and dynamically generate the first task instruction and the second task instruction according to the priority level of the target monitoring area, the blind spot position and obstacle distribution data, in accordance with the preset task allocation strategy. The perception and monitoring unit is used to control the first rotating radar and the second rotating radar to adjust their rotation angles and perception parameters according to the first task instruction and the second task instruction. This enables the first rotating radar to focus on sensing high-priority target monitoring areas, and the second rotating radar to compensate for the sensing blind spots and dead angles of the first rotating radar, thereby achieving coverage optimization of the sensing range and redundant monitoring of high-risk areas.

Images