Vehicle-mounted unmanned aerial vehicle cooperative adaptive light supplementing system and control method

Abstract

This invention discloses a vehicle-mounted drone collaborative adaptive lighting system and control method. The system includes a vehicle-mounted integrated processing unit, a drone-mounted intelligent lighting module, a vehicle lighting subsystem, a helipad lighting subsystem, and a vehicle-airport collaborative communication network. The vehicle-mounted integrated processing unit performs multi-source fusion of perceived data to generate fused data frames, performs scene feature extraction and refined scene classification, calculates required illuminance and beam characteristics based on Retinex theory, generates a globally optimal control parameter set through multi-objective optimization, and distributes it hierarchically to the three terminals for collaborative lighting. Finally, it performs closed-loop evaluation and parameter re-optimization based on the task-oriented indicators of images transmitted from the drone's mission camera. This invention achieves unified scheduling and dynamic task adaptation of lighting resources across the vehicle, drone, and helipad, solving the problems of single lighting strategies, lack of collaborative capabilities, insufficient energy efficiency and thermal management, and inadequate visual task orientation in existing technologies.

Description

Technical Field

[0001] This invention belongs to the field of vehicle-mounted drone lighting technology, specifically relating to a vehicle-mounted drone collaborative adaptive supplementary lighting system and control method. Background Technology

[0002] With the rapid development of vehicle-mounted drone applications, the demand for vehicle-mounted drones to perform tasks such as nighttime tracking, wide-area search, and precise landing is increasing. However, existing nighttime lighting solutions are insufficient to meet the actual needs of these tasks, and several prominent technical problems exist.

[0003] First, existing supplementary lighting strategies are relatively simplistic. Current vehicle-mounted drone nighttime supplementary lighting solutions typically use conventional light sources with fixed wavelengths, color temperatures, and beam shapes. These can only provide basic illumination output according to preset brightness and color temperature, lacking the ability to dynamically adjust the supplementary lighting strategy based on real-time factors such as specific mission type, target distance, ambient light, and weather conditions. This limitation makes it difficult for the same lighting hardware to simultaneously meet the long-range, wide-area illumination needs of search missions and the close-range, low-glare illumination needs of tracking missions, resulting in a significant mismatch between supplementary lighting output and specific mission requirements.

[0004] Second, existing technologies lack the ability to coordinate lighting resources at the vehicle, drone, and helipad ends. In current solutions, lighting resources such as vehicle headlights, drone-borne supplementary lights, and helipad lights typically operate independently, each outputting its light field according to its local strategy, without a unified coordination mechanism. This leads to problems including glare artifacts caused by the strong beams of vehicle headlights reflecting off the drone's onboard camera, spatial interference between the drone's supplementary lights and vehicle headlights causing target obstruction or light field conflicts, and the lack of active coordination from helipad lighting during the drone's precise landing phase. This lack of coordination prevents the lighting resources at the vehicle, drone, and helipad ends from working together effectively, instead becoming sources of optical noise that interfere with each other.

[0005] Third, existing technologies face challenges in terms of energy efficiency and thermal management. Current supplemental lighting solutions typically employ constant power or simple brightness feedback driving methods, maintaining high power consumption even in scenarios with low demand, leading to rapid depletion of the drone's onboard battery and excessive temperature rise of the lighting module. Due to the lack of a unified modeling and dynamic balancing mechanism for the relationship between lighting performance, power consumption cost, glare risk, and thermal load, existing solutions struggle to flexibly switch between long-endurance and extreme performance operations, and also find it difficult to autonomously adjust output strategies under real-time constraints such as insufficient battery power or high module temperature, significantly limiting the overall sustainable operation capability of the system.

[0006] Fourth, existing technologies lack effective integration between supplementary lighting control and downstream visual tasks, resulting in insufficient task-oriented guidance. Current solutions typically use generalized indicators such as achieving preset brightness or meeting human visual observation requirements as evaluation criteria for supplementary lighting effects, without considering whether the supplementary lighting output truly meets the needs of specific downstream visual tasks such as visual recognition, object detection, and portrait photography. Due to the lack of real-time closed-loop feedback between supplementary lighting control and visual tasks, the system cannot dynamically adjust supplementary lighting parameters based on the actual performance of the visual task. This easily leads to situations such as bright illumination but unclear images, or sufficient brightness but recognition failure, thus failing to fully realize the practical task support value of the supplementary lighting stage.

[0007] To address the limitations of the four existing technologies mentioned above, there is an urgent need for a vehicle-mounted UAV collaborative adaptive lighting system and its control method that can realize dynamic task-adaptive lighting strategies, has the ability to coordinate lighting at three ends (vehicle, UAV, and helipad), achieve dynamic balance between energy efficiency and thermal management under multi-objective constraints, and form a closed-loop feedback with downstream visual tasks. Summary of the Invention

[0008] The purpose of this invention is to address the shortcomings of the aforementioned background technology by providing a vehicle-mounted UAV collaborative adaptive lighting system and control method, thereby diversifying lighting strategies and enabling UAV lighting, vehicle lighting, and landing site lighting to work collaboratively, thus solving the problem of insufficient visual task guidance.

[0009] The technical solution adopted in this invention is: a vehicle-mounted drone collaborative adaptive lighting system, including a vehicle-mounted integrated processing unit, a drone-mounted intelligent lighting module, a vehicle lighting subsystem, a helipad lighting subsystem, and a vehicle-airport collaborative communication network; The vehicle-mounted integrated processing unit is deployed inside the vehicle and is configured to process the perception data and generate a set of globally optimal control parameters for the coordinated supplementary lighting of the UAV-based intelligent lighting module, the vehicle lighting subsystem, and the helipad lighting subsystem. The perception data includes environmental perception data collected and transmitted back by the UAV-based intelligent lighting module and vehicle-side data obtained by the vehicle-mounted integrated processing unit from the vehicle itself. The intelligent lighting module for the UAV is integrated on the UAV body and is configured to output a light field according to the global optimal control parameter set. The vehicle lighting subsystem is deployed on the vehicle and configured to output a light field according to the globally optimal control parameter set; the helipad lighting subsystem is located on the helipad and configured to output a light field according to the globally optimal control parameter set. The vehicle-airport collaborative communication network has time synchronization capabilities and is used to build a cross-platform synchronous control and data interaction link between the vehicle-mounted integrated processing unit, the UAV-mounted intelligent lighting module, the vehicle lighting subsystem, and the apron lighting subsystem.

[0010] In the above technical solution, the UAV-side intelligent lighting module includes an environmental perception kit, a programmable multispectral lighting matrix, and an airborne lighting controller; The environmental perception kit is used to collect the environmental perception data; The programmable multispectral illumination matrix is ​​used to perform light field output. The spectral output of the programmable multispectral illumination matrix covers a preset color temperature working range and includes at least one specific band for enhancing the sensitivity of specific target recognition. The airborne lighting controller is used to receive the global optimal control parameter set, drive the programmable multispectral lighting matrix, and manage the local thermal sensor and heat dissipation unit.

[0011] In the above technical solution, the vehicle-mounted integrated processing unit is further configured to perform multi-source fusion of the perceived data and unify it into the vehicle coordinate system to generate a fused data frame, and to perform collaborative decision-making and global scheduling on the vehicle lighting subsystem, the UAV-based intelligent lighting module and the helipad lighting subsystem based on the vehicle coordinate system and the fused data frame.

[0012] In the above technical solution, the environmental perception kit includes at least two of the following: a high dynamic range ambient light sensor, a time-of-flight depth camera, a real-time dynamic positioning and inertial measurement fusion unit, and a meteorological sensor. The high dynamic range ambient light sensor is used to collect ambient light illuminance and spectral information; The time-of-flight depth camera is used to acquire target distance and scene 3D point cloud data; The real-time dynamic positioning and inertial measurement fusion unit is used to provide the altitude, three-dimensional attitude and positioning information of the UAV; The meteorological sensor is used to collect at least one of the meteorological parameters, including ambient temperature and humidity, fog concentration, and meteorological visibility.

[0013] In the above technical solution, the vehicle-mounted integrated processing unit further includes a vehicle bus interface, a task instruction parsing unit, and a device status monitoring unit; The vehicle bus interface is used to acquire vehicle driving status data, including vehicle speed, steering wheel angle, current headlight mode, and vehicle GPS location. The task instruction parsing unit is used to receive task instructions from the vehicle or ground control terminal and parse and output a task scenario label indicating the current task scenario type. The task scenario type includes at least one of following task, search task, and precision landing. The following task includes vehicle-following shooting and portrait following shooting. The search task includes wide-area search and detection task for specific targets. The device status monitoring unit is used to monitor the remaining battery power of the UAV and the core temperature of the programmable multispectral illumination matrix.

[0014] In the above technical solution, the programmable multispectral illumination matrix includes a beam control mechanism, which is used to dynamically adjust the shape and spatial illumination angle of the output beam of the programmable multispectral illumination matrix. The beam control mechanism is selected from one or more combinations of electrically tunable metamaterial optical surfaces, microelectromechanical systems digital micromirror arrays, and micro stepper motor driven lens groups.

[0015] In the above technical solution, the beam control mechanism adopts an electrically tunable metamaterial optical surface, and the airborne lighting controller is configured to change the optical properties of the electrically tunable metamaterial optical surface through a program to achieve precise beam shaping and deflection without mechanical movement. The response time of beam shaping and deflection is not greater than a preset response time threshold.

[0016] In the above technical solution, the vehicle-airport cooperative communication network includes a hardware timestamp synchronization clock component. The hardware timestamp synchronization clock component adopts a precise time protocol and compensates for wireless transmission delay through underlying protocol message exchange, so that the time deviation between the vehicle, the drone and the apron where the apron lighting subsystem is located is not greater than a preset time synchronization threshold.

[0017] In the above technical solution, the vehicle-mounted integrated processing unit uses the point cloud data of the fused data frame as a three-dimensional scene approximation to simulate the light propagation of the light sources carried by the vehicle lighting subsystem, the UAV-mounted intelligent lighting module, and the helipad lighting subsystem, calculates the illuminance distribution of each light source in the core lighting area and the glare risk value at the driver's field of vision and the UAV's self-camera, and performs beam avoidance in areas where the glare risk value is greater than a preset glare safety threshold.

[0018] In the above technical solution, the vehicle lighting subsystem includes at least one of an adaptive digital headlight with matrix pixel-level dimming capability and a front auxiliary lighting lamp disposed at the front of the vehicle; the adaptive digital headlight is configured to perform a pixel-level darkening operation in a cone-shaped projection area extending from the drone body as the vertex along the direction from the adaptive digital headlight to the road surface in front of the adaptive digital headlight, forming an anti-glare shadow area, wherein the pixel-level darkening operation reduces the brightness of the corresponding pixels in the cone-shaped projection area to below a preset ratio threshold.

[0019] In the above technical solution, the apron lighting subsystem includes an apron lighting controller and light-emitting devices disposed around the apron where the apron lighting subsystem is located. The apron lighting controller is configured to switch the working mode of the light-emitting devices when the UAV enters the landing sequence. The working modes include a soft light illumination mode for visual positioning and a breathing gradient illumination mode for visual guidance.

[0020] In the above technical solution, the vehicle-mounted integrated processing unit is also configured to start the system and perform system self-test in response to preset trigger conditions. The trigger conditions include the UAV entering night flight mode or the vehicle or ground control terminal issuing night operation instructions. When the system self-test detects any abnormality in the UAV-end intelligent lighting module, the environmental perception kit, or the vehicle-airport cooperative communication network, it reports error information and stops the system startup, while triggering the UAV to switch to a preset safe flight mode.

[0021] This invention provides a vehicle-mounted UAV cooperative adaptive supplementary lighting control method, applied to the vehicle-mounted integrated processing unit in the vehicle-mounted UAV cooperative adaptive supplementary lighting system described above, comprising the following steps: Step 1: Perform multi-dimensional acquisition and spatiotemporal alignment of perception data, acquire heterogeneous data streams containing at least one of ambient light information, target depth information, UAV pose information, and vehicle status information, and generate fused data frames; Step 2: Based on the fused data frame, perform scene feature extraction and task scene recognition, determine the core lighting area, calculate the beam characteristic requirements for the core lighting area, and output a lighting requirement parameter set including task scene label, task priority, the core lighting area, required illuminance, and beam characteristic set, wherein the task priority is determined based on the task scene label and the real-time event status during task execution. Step 3: Perform a multi-objective optimization solution process based on the lighting demand parameter set to generate a globally optimal control parameter set; Step 4: Perform security verification on the global optimal control parameter set, and distribute the global optimal control parameter set to the UAV terminal intelligent lighting module, the vehicle lighting subsystem, and the apron lighting subsystem through the vehicle-airport cooperative communication network, so that the three can coordinate to perform supplementary lighting actions. Step 5: Receive the feedback information returned after the collaborative execution of the supplementary lighting action, and iteratively re-optimize the global optimal control parameter set based on the feedback information to form closed-loop control.

[0022] In the above technical solution, the spatiotemporal alignment in step one includes unifying the heterogeneous data streams to the same time reference based on hardware timestamps, and unifying the spatial coordinate data to the vehicle body coordinate system.

[0023] In the above technical solution, the scene feature extraction includes four types of feature vectors: output spatial relationship features, ambient light semantic features, meteorological perception features, and dynamic context features. The spatial relationship features include the relative distance, azimuth angle, and altitude difference between the UAV and the target, as well as the spatial proximity used to determine whether the UAV is in the glare-sensitive area in front of the vehicle. The ambient light semantic features include a base illuminance level, a light contrast ratio used to determine whether high dynamic range supplemental lighting is needed, and spectral features used to identify external light source interference. The meteorological sensing features include meteorological influencing factors, which are derived from meteorological parameters collected by the meteorological sensor or from image haze detection algorithms. The dynamic context features include the task scenario label, the relative motion state between the drone and the vehicle, and the system constraint state including the remaining battery power and the temperature of the lighting module.

[0024] In the above technical solution, the task scene recognition includes a scene classifier decision sub-step, which takes at least one of the spatial relationship features, the ambient light semantic features, the meteorological perception features and the dynamic context features as input, and outputs refined scene labels based on a preset lightweight hybrid decision model. The refined scene label is obtained by combining the task scene label and the conditional parameters in the feature vector. The refined scene label includes at least one of the following: low-speed close-range tracking scene, medium-to-long-range search and inspection scene, and return and precise landing scene. The low-speed close-range tracking scene corresponds to the feature combination of the relative distance being less than a preset close-range threshold, the relative motion state indicating low speed in the same direction, and the task scene label being the tracking task. The mid-to-long-range search and inspection scenario corresponds to a relative distance greater than the preset short distance threshold and a task scenario label that is a feature combination of the search task. The return-to-home and precision landing scenarios correspond to the UAV's positioning information being close to the helipad and the mission scenario label being a combination of features of the precision landing. The refined scene label serves as the final task scene label determined from the lighting requirement parameter set. The output of the scene classifier decision sub-step covers the original task scene label output by the task instruction parsing unit. Subsequent calculations of the core lighting area, the required illuminance, the beam characteristic set, the dynamic adjustment of the task priority, and the scenario-based evaluation of the feedback information are all performed based on the refined scene label as the task scene label.

[0025] In the above technical solution, the core lighting area in the lighting requirement parameter set is expressed in the form of a three-dimensional spatial boundary in the vehicle coordinate system based on the target detection box, the attitude information of the UAV, and the relative distance, azimuth angle, and height difference in the spatial relationship features; the beam characteristic set includes at least one of beam divergence angle, target color temperature, special band activation indicator, and polarization lighting activation indicator; the beam divergence angle is determined according to the spatial size of the core lighting area and the relative distance of the UAV to the core lighting area in the spatial relationship features. In the tracking task where the relative distance is in a preset close range, the beam divergence angle is set to a preset narrow beam range, and in the search task, the beam divergence angle is set to a preset wide beam range.

[0026] In the above technical solution, the task priority in the lighting demand parameter set is determined jointly by a preset task scenario priority mapping table and the real-time event status during task execution; the preset task scenario priority mapping table assigns an initial level to the task priority based on the task scenario label; the real-time event status is an event status triggered by changes in the perceived data or changes in the target detection result during task execution, including at least one of target detection, target loss, and environmental anomaly; the task priority is dynamically adjusted according to the type of the real-time event status, wherein when a target detection event occurs during the execution of the search task, the task priority is dynamically adjusted to the highest level and serves as one of the triggering conditions for switching the weight coefficients of the multi-objective cost function.

[0027] In the above technical solution, the multi-objective optimization solution process includes a light propagation simulation sub-step. In each optimization iteration, the light propagation simulation sub-step performs light propagation simulation on the light sources carried by the UAV-side intelligent lighting module, the vehicle lighting subsystem, and the helipad lighting subsystem based on the current decision variables, and outputs the illuminance distribution of each light source in the core lighting area and the glare risk value in the driver's field of vision and the UAV's self-camera as intermediate quantities. The light propagation simulation uses the point cloud data of the fused data frame as a three-dimensional scene approximation.

[0028] In the above technical solution, the decision variables in the multi-objective optimization solution process include the UAV terminal vector, the vehicle terminal vector, and the helipad terminal vector; The UAV terminal vector includes the brightness, color temperature, beam divergence angle, horizontal deflection angle, and vertical deflection angle of each sub-module in the programmable multispectral illumination matrix; The vehicle terminal vector includes the vertex coordinates of the polygon of the illumination area of ​​the adaptive digital headlight or the front auxiliary lighting lamp and the average brightness of the area; The helipad terminal vector includes the on / off status and operating mode identifier of the helipad lighting subsystem.

[0029] In the above technical solution, the coordinated execution of supplementary lighting includes enabling the UAV-side intelligent lighting module, the vehicle lighting subsystem, and the helipad lighting subsystem to execute synchronously within a preset time synchronization threshold.

[0030] In the above technical solution, the beam characteristic requirement calculation adopts a task gain adaptive mathematical model based on Retinex theory. The required illuminance of the core lighting area is pre-calculated using feedforward. The required illuminance is calculated as the sum of four physical quantities: ambient base illuminance, base safety supplementary lighting gain, task-specific gain coefficient and task spectral sensitivity factor, and meteorological compensation gain and meteorological influence factor. The ambient base illuminance is derived from the base illuminance level in the ambient light semantic features. The task-specific gain coefficient and the task spectral sensitivity factor are both determined based on the task scene label. When the task scene label indicates the search task, the task-specific gain coefficient is greater than the baseline gain coefficient, and the task spectral sensitivity factor is greater than the preset sensitivity threshold at a preset specific wavelength.

[0031] In the above technical solution, the beam characteristic requirement calculation further includes a color temperature dynamic decision sub-step. This sub-step determines the color temperature parameters according to a three-layer decision logic: a default adaptive layer, a weather trigger layer, and a task trigger layer. The default adaptive layer adaptively selects the color temperature according to the required illuminance. When the required illuminance is in the high brightness range, the color temperature shifts towards a cool color temperature, and when the required illuminance is in the low brightness range, the color temperature shifts towards a warm color temperature. When the meteorological visibility collected by the meteorological sensor is lower than the preset visibility threshold, the meteorological triggering layer forces the color temperature to be no less than the high color temperature threshold in order to enhance the light beam penetration. When the task scene label indicates that the portrait is being filmed, the task triggering layer sets the color temperature to a preset low to medium color temperature range to optimize skin tone reproduction. The decision priority of the weather triggering layer and the task triggering layer is higher than that of the default adaptive layer.

[0032] In the above technical solution, the multi-objective optimization solution process uses a multi-objective cost function for evaluation, and the global optimal control parameter set that minimizes the multi-objective cost function is solved by an optimization algorithm under the premise of satisfying rigid constraints. The multi-objective cost function uses the illuminance distribution and glare risk value output from the light propagation simulation sub-step as intermediate quantities to calculate four sub-costs: lighting performance error, total power consumption cost, glare risk sub-cost, and heat load prediction value. The four sub-costs are then weighted and summed to obtain the value of the multi-objective cost function. The rigid constraints include total power consumption not exceeding the maximum power limit, predicted heat load not exceeding the maximum temperature limit, glare risk value not exceeding the glare safety threshold, and each decision variable in the global optimal control parameter set being within the physically adjustable range of the corresponding actuator; wherein the maximum power limit is dynamically determined based on the remaining battery power in the dynamic context features, and the maximum temperature limit is dynamically determined based on the lighting module temperature in the dynamic context features.

[0033] In the above technical solution, the weighting coefficients of each sub-item in the multi-objective cost function are dynamically adjusted according to the task scenario label and the task priority. Different weight combinations are configured under different task scenarios to achieve adaptive optimization objective balance of the task scenario.

[0034] In the above technical solution, the security verification includes a triple check mechanism of timeliness verification, security verification, and feasibility verification; The timeliness verification determines whether the global optimal control parameter set is within a preset valid time window by comparing the current time with the generation timestamp of the global optimal control parameter set. The safety verification uses a safety rule engine to check whether there are dangerous parameter combinations in the global optimal control parameter set where the beam angle directly hits the vehicle's cockpit and the brightness exceeds the glare threshold. The feasibility verification is based on the instantaneous battery voltage state to determine whether the total power required by the globally optimal control parameter set exceeds the current maximum output capacity; If any check in the triple check mechanism fails, the globally optimal control parameter set is discarded and the effective control parameter set of the previous cycle is maintained, or the UAV is triggered to switch to a preset safe flight mode.

[0035] In the above technical solution, the feedback information includes the monitoring image transmitted back by the drone-side mission camera after completing synchronous supplementary lighting; The task-oriented indicators are determined based on the scene indicated by the task scene label. The task-oriented indicators are image quality assessment quantities that are directly related to the visual information required by the current task scene, including at least one of image pixel-level statistics and target-level morphological quantities. The task-oriented indicators include contrast and signal-to-noise ratio in the tracking task and target boundary sharpness in the detection task. The task orientation index of the monitored image is evaluated. When the task orientation index is lower than the preset image quality threshold, the evaluation result is fed back to step two to fine-tune the required illumination and trigger the multi-objective optimization solution process for secondary optimization.

[0036] In the above technical solution, the collaborative execution of supplementary lighting in step four also includes an execution status monitoring sub-step. The execution status monitoring sub-step collects at least one of the driving current, temperature, and beam pointing angle of each sub-module of the programmable multispectral illumination matrix in real time, and compares it with the target control signal to form a hardware-level closed-loop feedback. When the comparison result exceeds the preset deviation range, at least one abnormal handling action is executed, such as shutting down the faulty module, switching to the backup illumination unit, reducing the output power, or sending a visual assistance degradation warning to the flight control system.

[0037] In the above technical solution, when the proportion of specular reflection in the scene exceeds a preset reflection threshold, a control command to enable the ring polarization lighting mode is output in the lighting demand parameter set.

[0038] The beneficial effects of this invention are as follows: The vehicle-mounted drone collaborative adaptive supplementary lighting system and its control method provided by this invention achieve the above four comprehensive effects through a combination of these effects, providing a complete, efficient, safe, and intelligent supplementary lighting solution for vehicle-mounted drones operating at night. This significantly improves the operational quality, energy efficiency, and safety margin of vehicle-mounted drones in various nighttime tasks such as tracking, searching, and precise landing, demonstrating significant technological advancement and application promotion value. First, the comprehensive effect of multi-dimensional dynamic shaping of the lighting strategy. This invention, through the combined design of a programmable multispectral lighting matrix and a beam control mechanism, achieves independent controllability of five dimensions—brightness, color temperature, beam shape, beam angle, and spectral band—on a single hardware platform. Combined with a task gain adaptive mathematical model based on Retinex theory, the system can dynamically shape a light field precisely matched to the current task based on real-time changing parameters such as target distance, ambient light intensity, weather conditions, and task type. Taking search tasks and portrait tracking tasks as examples, the two form significantly differentiated configuration schemes in terms of color temperature, spectrum, and beam strategy, significantly solving the problem of a single lighting strategy in existing technologies.

[0039] Secondly, the comprehensive effect of vehicle-airport collaborative lighting. This invention, through a centralized decision-making mechanism between the vehicle-airport collaborative communication network and the vehicle-mounted integrated processing unit, achieves for the first time unified scheduling of lighting resources from three parties: drone lighting, vehicle lights, and apron lighting. The vehicle's adaptive digital headlights perform pixel-level darkening operations within the cone-shaped projection area to create an anti-glare shadow zone, complementing the drone's onboard light source. The apron lighting subsystem switches to a breathing gradient mode during the precise landing phase to actively cooperate with the drone's visual positioning. This three-terminal collaborative lighting network is significantly superior to the existing independent operation mode, fundamentally solving the problem of lack of collaborative capabilities.

[0040] Third, the comprehensive effect of dynamically balancing energy efficiency and thermal management. This invention incorporates four sub-objectives—lighting performance error, total power consumption cost, glare risk value, and thermal load estimation—into a multi-objective optimization function. A dynamic weighting coefficient mechanism allows the system to automatically switch its optimization focus under different mission scenarios. In long-endurance cruise scenarios, the system automatically reduces the light source output power to extend range; in extreme search scenarios, performance weights are extremely high to ensure lighting effectiveness for critical tasks; simultaneously, rigid constraints ensure that total power consumption does not exceed the battery's current output capacity and thermal load does not exceed the module's maximum allowable temperature. This dynamic optimization mechanism based on real-time system status significantly solves the challenges of energy efficiency and thermal management.

[0041] Fourth, the comprehensive effect of the visual task-oriented closed loop. This invention, through the real-time visual feedback mechanism in step five, forms a complete closed loop between supplementary lighting control and downstream visual recognition tasks for the first time. The system not only generates initial supplementary lighting parameters based on task requirements during the feedforward stage, but also continuously evaluates whether the supplementary lighting effect truly meets the needs of the visual task during execution. Through real-time evaluation of task-oriented indicators such as contrast and signal-to-noise ratio for the tracking task and target boundary sharpness for the detection task, the system can determine whether the supplementary lighting truly achieves the task objective. If the evaluation fails to meet the requirements, the system triggers parameter re-optimization and reissues instructions, forming a truly visual task-oriented closed-loop optimization, significantly solving the problem of insufficient visual task orientation.

[0042] Furthermore, the vehicle-mounted drone collaborative adaptive lighting system constructs an integrated lighting architecture of "centralized decision-making and distributed execution." The vehicle-mounted integrated processing unit acts as the decision-making center, uniformly generating a globally optimal set of control parameters. These parameters are then distributed across the drone's intelligent lighting module, the vehicle's lighting subsystem, and the helipad lighting subsystem, supplemented by a vehicle-airport collaborative communication network as a cross-platform synchronization link. This architecture fundamentally solves the problem of the lack of collaborative capability in existing technologies where the lighting resources at the three ends are independent and cannot form a synergy. It provides a system-level architectural foundation for subsequent collaborative lighting, task adaptive adjustment, and closed-loop optimization.

[0043] Furthermore, the UAV-based intelligent lighting module achieves integrated sensing, execution, and control functions on a single hardware platform through a combined design of an environmental perception kit, a programmable multispectral lighting matrix, and an onboard lighting controller. The programmable multispectral lighting matrix's spectral output covers a preset color temperature operating range and includes at least one specific band to enhance the sensitivity of specific target recognition. This allows the supplementary lighting output to flexibly switch color temperature and band strategies according to mission requirements, effectively overcoming the limitations of single supplementary lighting strategies in existing technologies.

[0044] Furthermore, the multi-source data fusion and collaborative decision-making mechanism unifies the perceived data into the vehicle coordinate system to form a fused data frame, giving heterogeneous data from different sensors and coordinate systems a unified spatiotemporal reference framework. This provides a unified and authoritative data foundation for subsequent collaborative decision-making and global scheduling, avoiding collaborative errors caused by inconsistent coordinate systems of perceived data in existing technologies.

[0045] Furthermore, the environmental perception kit, through the combination of four sensors, provides the system with multi-dimensional perception capabilities covering ambient light, scene depth, UAV pose, and weather conditions. Compared with the simple perception schemes in existing technologies that rely on only a single sensor, it significantly improves the system's comprehensive perception capabilities for complex nighttime operating environments.

[0046] Furthermore, the vehicle-mounted integrated processing unit sub-component architecture establishes a complete data link for obtaining control decision-making basis from three dimensions: vehicle driving status, task instructions, and real-time equipment status, through the division of labor between the vehicle bus interface, task instruction parsing unit, and equipment status monitoring unit. This provides necessary input support for subsequent task adaptive parameter generation and real-time constraint determination.

[0047] Furthermore, the beam control mechanism allows for flexible selection of three implementation methods: electrically tunable metamaterial optical surfaces, MEMS micromirror arrays, and micro stepper motor-driven lens groups. This enables the system to have multiple selectable beam shaping and deflection paths. Compared with the existing technology that only uses a single mechanical beam control scheme, this improves the flexibility of hardware selection and the breadth of applicable scenarios.

[0048] Furthermore, the electrically tunable metamaterial optical surface, as a preferred implementation, achieves precise beam shaping and deflection through electronically controlled adjustment without mechanical movement. The response time is no greater than a preset response time threshold, which brings a 3 to 10-fold improvement in response speed compared to the mechanical deflection method driven by a stepper motor, providing a hardware foundation for millisecond-level dynamic scene adaptation.

[0049] Furthermore, the hardware timestamp synchronization clock component ensures that the time deviation between the vehicle, drone, and helipad is no greater than a preset time synchronization threshold through a precise time protocol. This provides a unified time reference at the millisecond level for the three-terminal coordinated execution of supplementary lighting actions, solving the technical problem of coordination failure caused by the asynchronous timing of the three terminals in the existing independent operation scheme.

[0050] Furthermore, the light propagation simulation and beam avoidance mechanism achieves pre-simulation of light propagation from multiple light sources and glare risk assessment by approximating a 3D scene based on fused data frame point cloud data. When the glare risk exceeds a preset threshold, beam avoidance is actively executed, fundamentally eliminating the typical problem of glare artifacts caused by the reverse illumination of drone cameras by vehicle headlights in existing solutions.

[0051] Furthermore, the adaptive digital headlights create an anti-glare shadow area through pixel-level darkening operations within the cone-shaped projection area, enabling the lighting output in front of the vehicle to actively avoid the direction of the drone's camera and form a complementary lighting force with the drone's onboard light source. This is the core innovative embodiment of the vehicle-airport cooperative lighting concept of this invention in the vehicle-side execution stage.

[0052] Furthermore, the helipad lighting subsystem provides a stable lighting environment and directional guidance signals for the visual positioning and precise landing of UAVs by dynamically switching between two working modes: soft light mode and breathing gradient light mode. This solves the defect in existing solutions where helipad lighting cannot actively cooperate with the UAV landing process.

[0053] Furthermore, the system self-check and safe flight mode switching mechanism starts the system and performs core component self-checks by setting preset trigger conditions. When an abnormality is detected in a component, the system reports the error and switches the drone to safe flight mode, thus providing flight safety assurance for nighttime operations from the system startup level.

[0054] Furthermore, the five-stage closed-loop control method establishes an end-to-end control link from multi-dimensional perception to collaborative execution and then to visual closed loop through a complete closed-loop process of "perception-decision-execution-feedback-re-optimization". Compared with the open-loop or only local feedback control schemes in the existing technology, it significantly improves the system's adaptive response capability to dynamic scene changes.

[0055] Furthermore, the spatiotemporal alignment mechanism ensures the spatiotemporal consistency of heterogeneous data streams in the fused data frame through a dual alignment method that unifies the hardware timestamp benchmark and the vehicle coordinate system space, providing a reliable data foundation for all subsequent decision calculations based on the fused data frame.

[0056] Furthermore, the scene feature extraction method based on four types of feature vectors organizes features in four dimensions: spatial relationship, ambient light semantics, meteorological perception, and dynamic context. This expands the system's understanding of nighttime operation scenarios from a single brightness dimension to a multi-dimensional comprehensive cognition, providing rich feature basis for refined scene classification and differentiated parameter calculation.

[0057] Furthermore, the scene classifier decision sub-step outputs refined scene labels through a lightweight hybrid decision model, which overwrites the original task scene labels. This enables all subsequent computational steps of the system to be executed based on more refined scene cognition, significantly improving the scene adaptation accuracy of parameter calculation and optimization.

[0058] Furthermore, the core lighting area and beam divergence angle calculation method explicitly utilizes spatial quantities such as relative distance, azimuth angle, and height difference in spatial relationship characteristics to achieve precise positioning of the lighting area and scene-adaptive configuration of beam width. The tracking task uses a narrow beam while the search task uses a wide beam, so that the beam characteristics and task requirements form a direct correspondence.

[0059] Furthermore, the dynamic task priority adjustment mechanism combines a preset task scenario priority mapping table with real-time event status, enabling the system to automatically raise the priority to the highest level and switch the cost function weight when a suspected target is found in the search task, providing a mechanism to ensure the concentrated allocation of resources during critical tasks.

[0060] Furthermore, the light propagation simulation sub-step, as an internal component of the multi-objective optimization solution process, calculates the illuminance distribution and glare risk value based on the current decision variables as intermediate evaluation quantities in each optimization iteration. This ensures that the optimization process is always based on the real physical model of light propagation, avoiding the accuracy deviation caused by relying solely on empirical formulas for optimization in existing technologies.

[0061] Furthermore, the three-subvector method for constructing decision variables combines three independent subvectors from the UAV end, the vehicle end, and the helipad end, enabling the optimization solution to simultaneously cover all controllable parameters at all three ends, achieving true global unified optimization.

[0062] Furthermore, the synchronous execution mechanism within the time synchronization threshold ensures that the three ends maintain millisecond-level timing coordination when performing supplementary lighting actions, avoiding the destruction of the overall coordination effect caused by timing deviation of any end.

[0063] Furthermore, the adaptive calculation method for required illuminance based on Retinex theory establishes a reproducible mathematical model for the quantitative calculation of required illuminance by summing four physical quantities: ambient base illuminance, base safety supplementary lighting gain, task-specific gain multiplied by sensitivity factor, and meteorological compensation gain multiplied by meteorological influence factor. This is one of the core innovations of this invention.

[0064] Furthermore, the three-layer logic of dynamic color temperature decision-making, through a hierarchical decision-making architecture of default adaptive layer, weather trigger layer, and task trigger layer, enables color temperature parameters to automatically switch to the most suitable value range in different scenarios, optimizing the visual perception effect under different tasks.

[0065] Furthermore, the multi-objective cost function and rigid constraint mechanism establish a comprehensive optimization framework that simultaneously considers four objectives: lighting performance, power consumption, glare, and heat load, through the weighted summation of four sub-costs and the combination of four rigid constraints. The rigid constraints ensure that the optimization solution is always within the physical feasible boundary of the system, thus guaranteeing the engineering feasibility of the optimization results from a mathematical perspective.

[0066] Furthermore, the dynamic adjustment mechanism of the cost function weights enables the system to switch weight combinations in different scenarios such as extreme search, long-endurance cruise, and in-vehicle filming, achieving an adaptive optimization target balance for the task scenario. This is the key implementation carrier of the dynamic balance concept of energy efficiency and thermal management in this invention.

[0067] Furthermore, the triple verification mechanism performs three layers of checks—timeliness verification, security verification, and feasibility verification—to conduct multi-dimensional compliance checks on the control parameter set before it is issued and executed, ensuring that non-compliant parameter sets do not enter the execution channel and providing strong protection for the security of system operation.

[0068] Furthermore, the visual feedback and task-oriented index evaluation mechanism evaluates the supplementary lighting effect in real time by configuring image quality evaluation indicators based on task scene labels, directly connecting supplementary lighting control with downstream visual tasks, fundamentally solving the problem of the disconnect between supplementary lighting and visual tasks in existing technologies.

[0069] Furthermore, the execution status monitoring sub-step forms a hardware-level closed-loop feedback by real-time acquisition and comparison of the driving current, temperature, and beam pointing angle of each sub-module of the lighting matrix. This enables abnormalities in the execution process to be detected immediately and trigger corresponding abnormality handling actions, thereby improving the reliability of system operation.

[0070] Furthermore, the specular reflection-triggered polarized illumination mode mechanism enables the system to automatically activate ring polarized illumination when a large number of specular reflections are detected in the scene, effectively dealing with scenes with strong reflective characteristics such as slippery roads and vehicle glass surfaces, and ensuring the stability of visual recognition. Attached Figure Description

[0071] Figure 1 This is a schematic diagram of the overall control flow of the system described in this invention.

[0072] Figure 2 This is a logic diagram for step one of the present invention.

[0073] Figure 3This is a logic diagram for step two of the present invention.

[0074] Figure 4 This is the logic diagram for step three of the present invention.

[0075] Figure 5 This is the logic diagram for step four of the present invention.

[0076] Figure 6 This is a schematic diagram of the system architecture of the present invention. Detailed Implementation

[0077] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments to facilitate a clear understanding of the present invention, but these descriptions do not constitute a limitation on the present invention.

[0078] The technical solution of the present invention will be further described in detail below with reference to the figures and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0079] The vehicle-mounted drone collaborative adaptive supplementary lighting system and control method provided by this invention address the problems existing in the prior art, such as single lighting strategy, lack of vehicle-airport collaboration capability, lack of energy efficiency and thermal management mechanism, and insufficient visual task orientation. It constructs an integrated supplementary lighting architecture of "centralized decision-making and distributed execution", and realizes task adaptive supplementary lighting in nighttime operation scenarios through dynamic collaborative scheduling of light sources from vehicles, drones and helipads.

[0080] Example 1: Vehicle-mounted UAV Cooperative Adaptive Compensation Lighting System Reference Figure 1 The system control flowchart shown in this embodiment illustrates that the vehicle-mounted UAV collaborative adaptive supplementary lighting system adopts a centralized decision-making and distributed execution architecture. Figure 6 As shown, the system consists of three main components: an onboard integrated processing unit deployed within the vehicle, a drone-based intelligent lighting module integrated onto the drone's body, and a vehicle-airport collaborative communication network connecting the onboard unit, the drone, and the helipad. In addition to these three main components, the system also includes a vehicle lighting subsystem and a helipad lighting subsystem that collaboratively participate in supplementary lighting. The onboard integrated processing unit, as the decision-making core of the entire system, is responsible for fusing multi-source sensing data, intelligently identifying task scenarios, and collaboratively optimizing dynamic light fields, generating a unified set of globally optimal control parameters. The drone-based intelligent lighting module, the vehicle lighting subsystem, and the helipad lighting subsystem, as the execution ends, each output a collaborative light field based on this globally optimal control parameter set, enabling the three lighting resources to form a three-dimensional collaborative supplementary lighting network under the same time reference and coordinate system.

[0081] Figure 1 The complete closed-loop control process of the system operation is presented in the form of a flowchart. After the system starts, it sequentially executes the following steps: Step 1: startup and multi-dimensional data acquisition; Step 2: intelligent recognition and deconstruction of the task scene; Step 3: model-based dynamic light field collaborative calculation; Step 4: hierarchical collaborative light field execution; and Step 5: real-time visual feedback and optimization. The execution result of Step 4 is sent to the UAV lighting module, vehicle lighting system, and helipad lighting system via three parallel command distribution channels. The image quality assessment result of Step 5 constitutes the closed-loop feedback entry point of the system. When the image quality assessment result does not meet the task requirements, the parameter re-optimization process is triggered, and the system re-executes Step 3 for a new round of light field collaborative calculation; when the image quality assessment result meets the task requirements, the task continues to execute and enters the next control cycle.

[0082] The vehicle-mounted integrated processing unit (VRAM), serving as the system's decision-making hub, is deployed within the vehicle's central computing platform or domain controller. This VRAM integrates a high-performance computing module, a multi-source data fusion module, a collaborative decision-making algorithm module, and a vehicle lighting control interface. The high-performance computing module handles computationally intensive tasks such as multi-objective optimization algorithms, real-time ray tracing simplified models, and scene classifier decision models. The multi-source data fusion module performs spatiotemporal alignment and coordinate unification of heterogeneous data streams from different sensors and coordinate systems. The collaborative decision-making algorithm module generates a globally optimal set of control parameters across the three terminals (vehicle, vehicle, and domain controller) based on the current task scenario and device constraints. The vehicle lighting control interface connects to the vehicle's domain controller via the vehicle bus, enabling direct control of the vehicle lighting subsystem.

[0083] The perception data processed by the onboard integrated processing unit comes from two sources. One part is environmental perception data, which is collected by the intelligent lighting module on the drone and transmitted back to the onboard unit via the vehicle-to-aircraft cooperative communication network. The other part is vehicle-side data, which is acquired by the onboard integrated processing unit from the vehicle itself through the vehicle bus interface. Vehicle-side data includes real-time vehicle speed, steering wheel angle, current headlight mode, vehicle GPS location, and other vehicle driving status data. The onboard integrated processing unit processes both types of data uniformly to generate a globally optimal set of control parameters for coordinated supplementary lighting by the three execution subsystems.

[0084] The vehicle-mounted integrated processing unit further includes three functional sub-components: a vehicle bus interface, a task instruction parsing unit, and an equipment status monitoring unit. The vehicle bus interface uses CAN FD or Ethernet bus protocols to acquire the aforementioned vehicle driving status data in real time. The task instruction parsing unit is responsible for receiving task instructions from the vehicle or ground control terminal and parsing and outputting a task scenario label indicating the current task scenario type. This embodiment supports three primary categories of task scenario types: follow-up shooting, search, and precision landing. Follow-up shooting is further subdivided into vehicle-following and portrait-following sub-types; search tasks are further subdivided into wide-area search and target-specific detection sub-types. This hierarchical task scenario organization provides a flexible scenario basis for subsequent calculation of requirement parameters. The equipment status monitoring unit continuously reads the remaining battery power and core temperature of the programmable multispectral illumination matrix on the UAV terminal through the vehicle-airport cooperative communication network, providing real-time data for power consumption and thermal load constraints in subsequent optimization processes.

[0085] The vehicle-mounted integrated processing unit is also configured to start the entire system and perform a system self-test upon receiving a trigger condition. Trigger conditions include the drone entering nighttime flight mode and the vehicle or ground control terminal issuing nighttime operation instructions. The system self-test covers three aspects: the drone's intelligent lighting module, the environmental perception kit, and the vehicle-to-airport collaborative communication network. If any aspect detects an anomaly, the vehicle-mounted integrated processing unit immediately reports the error information and terminates the system startup process, while simultaneously triggering the drone to switch to a preset safe flight mode to ensure flight safety.

[0086] The UAV-based intelligent lighting module is integrated into the UAV body, serving as the system's sensing and execution terminal. The module is configured with two functions: firstly, it collects environmental perception data and transmits it back to the vehicle-mounted integrated processing unit; secondly, it outputs a light field based on the received global optimal control parameter set. The UAV-based intelligent lighting module comprises three components: an environmental perception kit, a programmable multispectral lighting matrix, and an onboard lighting controller.

[0087] The environmental perception suite is responsible for the all-around perception required for nighttime operations. In a typical implementation, it includes at least two of four sensors: a high dynamic range ambient light sensor, a time-of-flight depth camera, a real-time dynamic positioning and inertial measurement fusion unit, and a meteorological sensor. The high dynamic range ambient light sensor continuously samples ambient light illuminance and spectral information to determine whether the current environment is dim, completely dark, or has complex artificial light source interference. Its illuminance output can be further classified into three typical levels: extremely dark (less than 1 lux), dim (1 to 10 lux), and with significant ambient light (greater than 10 lux), to provide a discrete base illuminance level for subsequent scene feature extraction. The time-of-flight depth camera acquires 3D point cloud data of the scene in front of the UAV through active ranging, calculates the precise distance and relative angle between the UAV and the main target of interest, and provides a basis for 3D scene approximation for subsequent light propagation simulation. In an implementation where the time-of-flight depth camera is equipped with a synchronous RGB sensor, the depth camera can also simultaneously output point cloud data with texture information to support visual recognition-based target detection and scene understanding. The real-time dynamic positioning and inertial measurement fusion unit integrates RTK positioning and attitude data from the inertial measurement unit to provide the UAV with high-precision altitude, three-dimensional attitude (pitch angle, roll angle, and yaw angle), and high-precision positioning information. In this embodiment, a weather sensor is configured as an optional component. When the system is equipped with this component, the weather sensor can directly monitor ambient temperature and humidity and output meteorological parameters such as visibility. When the system is not equipped with a separate weather sensor, fog concentration can be indirectly estimated using a fog analysis algorithm based on high dynamic range ambient light sensor data and depth images as an alternative method. The estimated fog concentration and meteorological visibility are used as the basis for subsequent color temperature decisions and weather compensation.

[0088] The programmable multispectral illumination matrix is ​​a core execution component of the UAV, consisting of an array of multiple independently controllable sub-modules. Each sub-module is equipped with independent drive circuitry, a temperature sensor, and optical components, allowing for independent adjustment of brightness and color temperature. The entire matrix's spectral output covers a preset color temperature operating range, typically 2700K to 6500K in a typical implementation, encompassing the full range from warm white light to cool white light. Furthermore, the matrix includes at least one specific wavelength band for enhancing the sensitivity of specific target recognition. For example, in a biological target recognition scenario within a search mission, the matrix can be equipped with a dedicated sub-module operating in the 650nm red light band. This significantly increases the light power output in this band, leveraging the unique red light reflection characteristics of biological tissue to enhance target recognition.

[0089] The programmable multispectral illumination matrix further includes a beam control mechanism for dynamically adjusting the shape and spatial illumination angle of the matrix output beam. The beam control mechanism can be selected from one or more combinations of three implementations: a lens group driven by a microstepper motor, a micromirror array based on microelectromechanical systems (MEMS), and an electrically tunable metamaterial optical surface. In a preferred embodiment using an electrically tunable metamaterial optical surface as the beam control mechanism, the airborne illumination controller changes the microstructure of the metamaterial surface by applying a specific voltage, thereby altering its optical properties to achieve precise beam shaping and deflection without mechanical movement. Since the electro-control response process of the metamaterial surface does not require mechanical component movement, the overall beam shaping and deflection response time is no greater than a preset response time threshold, typically set to within 20 milliseconds, significantly better than the mechanical deflection response speed driven by a stepper motor.

[0090] In a preferred embodiment, the electrically tunable metamaterial optical surface can be realized through one of three physical implementation paths: a liquid crystal embedded metasurface, a dielectric resonant unit metasurface, or an electrically controlled phase transition metasurface. In a typical implementation of this embodiment, a liquid crystal embedded metasurface is preferred as the physical implementation path. Its structure consists of a substrate layer, a subwavelength artificial structural unit array, a nematic liquid crystal layer embedded between the structural units, and a transparent driving electrode layer stacked sequentially from bottom to top. The subwavelength artificial structural units adopt a dielectric or metallic resonant structure, and the characteristic lateral dimension of each structural unit is typically 200 to 500 nanometers to match the operating wavelength range from visible light to near-infrared (400 to 1000 nanometers). The periodic spacing between adjacent structural units is typically 300 to 800 nanometers. The thickness of the nematic liquid crystal layer is typically 1 to 5 micrometers, and its molecular orientation is controlled by an electric field applied to the driving electrode layer. When the driving electric field changes, the orientation change of the liquid crystal molecules causes a change in the equivalent refractive index of the liquid crystal, thereby changing the local phase response of the metasurface unit at that location, realizing the shaping and deflection of the light beam.

[0091] The macroscopic dimensions of the electrically tunable metamaterial optical surface are typically 5 mm to 30 mm square, divided into independently controllable pixel arrays, with typical array sizes ranging from 100 × 100 to 500 × 500. The pixel array's driving interface employs a thin-film transistor active matrix driving architecture (similar to the driving method of liquid crystal display panels) or a row-column scanning passive matrix driving architecture. In a preferred embodiment, the driving voltage for each pixel ranges from 1 to 20 volts, with a typical digital adjustment precision of 8 bits (corresponding to 256 voltage levels) to obtain a smooth optical response; the refresh cycle of the driving circuit is typically 1 to 20 milliseconds, matching the orientation response time characteristics of liquid crystals. The airborne illumination controller is connected to the electrically tunable metamaterial optical surface via a dedicated pixel driving interface circuit, typically using a Serial Peripheral Interface (SPI) or Inter-Integrated Circuit (I2C) bus to send pixel driving configuration data.

[0092] The beam shaping and deflection control mapping relationship of electrically tunable metamaterial optical surfaces is established based on the metasurface phase modulation principle. The system discretizes the ideal phase distribution corresponding to the target beam shape or target deflection angle into a pixel array, obtaining the target phase value for each pixel. Then, using a pre-calibrated voltage-phase response curve, the target phase value is used to back-calculate the specific driving voltage for each pixel via a lookup table. The voltage-phase response curve is obtained during system development by applying a scanning voltage pixel-by-pixel and measuring the corresponding phase response. The calibration results are stored in the non-volatile memory of the airborne lighting controller in the form of a lookup table. During operation, the airborne lighting controller receives beam characteristic parameters such as beam divergence angle, horizontal deflection angle, and vertical deflection angle from the vehicle-mounted integrated processing unit. It first calculates the ideal phase distribution using geometric optics formulas, then quickly obtains the driving voltage configuration for each pixel using a lookup table and linear interpolation, and finally completes the voltage transmission through the pixel driving interface circuit. The calculation time for the entire control mapping is typically no more than 2 milliseconds.

[0093] The beam control performance achievable by electrically tunable metamaterial optical surfaces exhibits the following typical quantization boundaries in the spatial dimension: the controllable range of the horizontal beam deflection angle is typically ±30 degrees, the controllable range of the vertical beam deflection angle is typically ±20 degrees, and the minimum control resolution of the deflection angle is typically 0.5 degrees; the malleable shape library of the beam shape includes basic shapes such as circles, ellipses, rectangles, and rings, as well as composite shapes obtained by combining and superimposing basic shapes; the controllable range of the beam divergence angle is typically 5 to 60 degrees, and the minimum control resolution of the divergence angle is typically 1 degree. These spatial performance boundaries support the actual adjustable range of the decision variables such as the beam divergence angle, horizontal deflection angle, and vertical deflection angle involved in the claims, allowing those skilled in the art to determine the specific values ​​of the system's physical adjustable range constraints.

[0094] In the time dimension, the response time of electrically tunable metamaterial optical surfaces is directly related to the physical implementation path adopted. When using liquid crystal embedded metasurfaces, the response time is mainly limited by the liquid crystal molecule orientation reconstruction process, with a typical response time of 1 to 10 milliseconds. This can be further optimized to within 5 milliseconds by using low-viscosity liquid crystal materials and thin liquid crystal layer thickness design. When using dielectric resonant unit metasurfaces, the response time is mainly limited by the capacitor charge-discharge time constant, with a typical response time in the microsecond to millisecond range. When using electrically controlled phase change metasurfaces, the response time depends on the phase change process of the phase change material, typically in the nanosecond to microsecond range. In this embodiment, all three physical implementation paths can meet the aforementioned preset response time threshold (within 20 milliseconds). Among them, the liquid crystal embedded metasurface scheme has the most advantages in terms of cost, process maturity, and band compatibility with LED light sources, and is therefore recommended as the preferred implementation path.

[0095] The airborne lighting controller receives the globally optimal control parameter set from the on-board integrated processing unit and parses it into specific electrical signals to drive the programmable multispectral lighting matrix. Simultaneously, the airborne lighting controller manages local thermal sensors and heat dissipation units, collecting real-time temperature data from each sub-module. When the temperature exceeds a preset threshold, it triggers the heat dissipation unit to enhance cooling or reduce the output power of the sub-module, ensuring the thermal stability of the entire lighting matrix under long-term operating conditions.

[0096] The vehicle lighting subsystem is deployed on the vehicle and configured to output a light field based on a globally optimal set of control parameters. In this embodiment, the vehicle lighting subsystem includes at least one of an adaptive digital headlight with matrix pixel-level dimming capability and a front auxiliary lighting lamp disposed at the front of the vehicle. The adaptive digital headlight uses digital pixel light source technology, and its illumination area is divided into tens of thousands of independently dimmable pixel units, allowing the headlight's illumination output to be arbitrarily blocked, brightened, or colored in space.

[0097] The adaptive digital headlights feature independent anti-glare functionality. When the drone is in the airspace in front of the vehicle, the adaptive digital headlights are configured to perform pixel-level darkening within a cone-shaped projection area extending from the drone's apex along the direction from the headlight to the road surface in front of it. This cone-shaped projection area geometrically covers the ground projection range from the headlight to the drone and then to the area behind the drone. After darkening the corresponding pixels within this area, the headlight's illumination output is significantly reduced, creating an anti-glare shadow zone. The pixel-level darkening operation reduces the brightness of the corresponding pixels within the cone-shaped projection area to below a preset percentage threshold, typically set to 10%, meaning the pixel brightness within this area does not exceed 10% of the normal lighting brightness. This mechanism is particularly crucial in tracking missions, preventing overexposure or glare artifacts caused by direct headlight beams hitting the drone's lens.

[0098] The apron lighting subsystem is located on the apron and configured to output a light field based on a globally optimal set of control parameters. The apron lighting subsystem includes an apron lighting controller and light-emitting devices positioned around the apron perimeter. The light-emitting devices typically employ high color rendering, flicker-free LED strip arrays, arranged at least on both sides of the apron edge. The apron lighting controller is configured to switch the operating mode of the light-emitting devices when the UAV enters the landing sequence. The operating modes include two preset modes: one is a soft lighting mode for visual positioning, in which the light-emitting devices continuously emit stable, uniform soft light, providing a stable lighting reference for the UAV's visual positioning algorithm; the other is a breathing gradient lighting mode for visual guidance, in which the brightness of the light-emitting devices gradually brightens and dims sinusoidally according to a preset cycle, providing clear directional guidance for the UAV during the precise landing phase.

[0099] The vehicle-to-airport collaborative communication network connects the vehicle-mounted integrated processing unit, the drone-mounted intelligent lighting module, the vehicle lighting subsystem, and the apron lighting subsystem, establishing a cross-platform synchronous control and data interaction link between them. This network consists of a high-bandwidth, low-latency wireless data link and a wired communication link. Data transmission between the vehicle-mounted unit and the drone-mounted unit uses 5G or a dedicated millimeter-wave link; between the vehicle-mounted unit and the vehicle lighting subsystem, an in-vehicle CANFD or vehicle Ethernet bus is used; and between the vehicle-mounted unit and the apron lighting subsystem, a private ultra-wideband link or a wired connection is used.

[0100] The vehicle-airport cooperative communication network possesses time synchronization capabilities, the core of which is a hardware timestamp synchronization clock component. This hardware timestamp synchronization clock component employs the Precision Time Protocol (PTP), compensating for wireless transmission latency through underlying protocol message exchange to achieve high-precision time synchronization among the vehicle, the drone, and the helipad lighting subsystem. In a typical implementation, the time deviation among the three is no greater than a preset time synchronization threshold, which can be set to within 1 millisecond. High-precision time synchronization is fundamental for the three-way collaborative execution of supplementary lighting actions; only under a unified time reference can the light fields of the three parties achieve true coordination.

[0101] The onboard integrated processing unit is further configured to perform multi-source fusion of environmental perception data and unify it into the vehicle coordinate system to generate a fused data frame. The vehicle coordinate system has the vehicle's center of mass as its origin, with the X-axis pointing forward, the Y-axis pointing to the left, and the Z-axis pointing upward. Multi-source fusion is completed through two steps: timestamp synchronization and coordinate transformation, mapping environmental perception data from the UAV, vehicle driving status data from the vehicle, and positioning information from RTK / INS to a unified vehicle coordinate system. The fused data frame is the system's unique and authoritative description of the external environment and internal state at the current moment.

[0102] Based on the vehicle coordinate system and fused data frames, the on-board integrated processing unit performs collaborative decision-making and global scheduling for the vehicle lighting subsystem, the drone-side intelligent lighting module, and the helipad lighting subsystem. Specifically, the on-board integrated processing unit uses the point cloud data from the fused data frames as a 3D scene approximation to simulate the light propagation of the light sources carried by each of the three subsystems, calculating the illuminance distribution of each light source in the core lighting area and the glare risk value at the driver's field of vision and the drone's self-camera. The light propagation simulation uses a simplified ray tracing algorithm, treating each light source as a point source or surface source with a specific radiation pattern. Sampled rays are emitted from the light source location along multiple directions, calculating the propagation, attenuation, and reflection of the rays in the 3D scene, and finally virtually imaging the target area and the driver's field of vision to obtain the illuminance distribution and glare spots. To achieve millisecond-level calculation speed, the simulation simplifies sampling in non-critical areas, performing dense sampling only in the core lighting area and glare-sensitive areas. In areas where the glare risk value of the simulated light propagation output exceeds the preset glare safety threshold, the vehicle-mounted integrated processing unit performs beam avoidance operation, which blocks or darkens the output of the corresponding light source in that area.

[0103] Example 2: Cooperative Adaptive Illumination Control Method for Vehicle-Mounted Unmanned Aerial Vehicles The vehicle-mounted UAV cooperative adaptive lighting control method described in this embodiment is applied to the vehicle-mounted integrated processing unit in the vehicle-mounted UAV cooperative adaptive lighting system described in Embodiment 1. This method is built upon a closed-loop concept of perception-decision-execution-evaluation, and the complete process includes five steps: startup and multi-dimensional data acquisition, intelligent task scene recognition and deconstruction, model-based dynamic light field cooperative calculation, hierarchical cooperative light field execution, and real-time visual feedback and optimization. The following section combines... Figures 1 to 5 The specific implementation of each step will be explained in turn.

[0104] Step 1: Start and collect multidimensional data.

[0105] Reference Figure 2 The logic diagram of step S1 is as follows: step one unfolds in three stages: triggering and initialization, parallel multi-source data acquisition, and spatiotemporal alignment and fusion.

[0106] Step 1.1: Triggering and initialization.

[0107] Figure 2 The trigger node at the top represents the entry point for system startup. The trigger condition can be the UAV autonomously entering night flight mode, or a night operation command issued by the ground station or vehicle system. When either trigger condition is met, the supplementary lighting system is activated and enters the startup process.

[0108] The startup process begins with a system self-check to verify that the drone's intelligent lighting module, various sensors, and vehicle-to-airport communication links are all functioning correctly. Figure 2 The diamond-shaped decision node "System self-test passed" indicates that the check node is in progress. If the self-test fails, the system reports an error and stops the supplementary lighting system from starting, while switching the flight control system to the preset safe flight mode to ensure the safety of the UAV flight; if the self-test passes, the system enters the parallel multi-source data acquisition stage.

[0109] Step 1.2: Parallel multi-source data acquisition.

[0110] After the self-test passes, the triggering and initialization sub-step sends an activation signal to the vehicle-airport cooperative communication network, synchronizing the data acquisition synchronization clock in the network to a unified time base. The sampling trigger signal is then broadcast to the three data sources—airborne perception layer, task and state layer, and vehicle perception layer—via clock broadcast. Upon receiving the trigger signal, all three data sources simultaneously initiate sampling, ensuring that the raw data within the same sampling period has a unified timestamp starting point.

[0111] Figure 2 The central parallel multi-source data acquisition frame comprises three parallel data source layers: the airborne perception layer, the task and status layer, and the vehicle-mounted perception layer. These three layers are uniformly controlled by the aforementioned data acquisition synchronization clock, simultaneously activating and continuously executing sampling actions. It should be noted that the above three-layer division is organized according to the logical hierarchy of data sources, rather than according to physical deployment location; the components of the airborne perception layer are physically deployed in the UAV-side intelligent lighting module, corresponding to the environmental perception kit described in the system embodiment; the components of the task and status layer are physically deployed in the vehicle-mounted integrated processing unit, corresponding to the task instruction parsing unit and equipment status monitoring unit described in the system embodiment; the data from the vehicle-mounted perception layer is acquired from the vehicle itself through the vehicle bus interface in the vehicle-mounted integrated processing unit. Considering the differences in the native sampling frequencies of different sensors, typically, high dynamic range ambient light sensors can reach 100Hz to 200Hz, time-of-flight depth cameras are usually 30Hz, real-time dynamic positioning and inertial measurement fusion units can reach 200Hz, and vehicle bus data is usually 50Hz to 100Hz. After completing the original sampling, the three data sources enter their respective time buffers for alignment buffering, waiting for the spatiotemporal alignment and fusion sub-steps to read them uniformly.

[0112] The airborne perception layer comprises four components: a high dynamic range ambient light sensor (hereinafter referred to as the ambient light sensor), a time-of-flight depth camera (hereinafter referred to as the depth camera), a real-time dynamic positioning and inertial measurement fusion unit (hereinafter referred to as the RTK / INS unit), and a weather sensor. The ambient light sensor continuously samples ambient light illuminance and spectral information, outputting the illuminance level of the current scene, indicating whether it is extremely dark (less than 1 lux), dimly lit (1 to 10 lux), or has significant ambient light (greater than 10 lux). The depth camera outputs point cloud data containing the scene's 3D coordinates and texture information, from which the precise distance (denoted as target distance) and relative angle between the UAV and the main target are calculated. The RTK / INS unit fuses satellite positioning and inertial measurement data, outputting the UAV's altitude (denoted as UAV altitude), 3D attitude, and high-precision positioning information. The weather sensor is an optional component; when equipped, it can directly output weather visibility data; when not equipped, fog concentration can be indirectly estimated through a fog analysis algorithm based on ambient light sensor data and depth images.

[0113] The task and state layer comprises two components: a task instruction parsing unit and a device status monitoring unit. The task instruction parsing unit reads task instructions from the vehicle or ground control terminal and outputs task scenario tags. These task scenario tags employ a two-level hierarchical structure. The first-level category includes three types: follow-up shooting, search, and precision landing. The second-level subcategories are further subdivided under follow-up shooting into vehicle-following and portrait-following, and under search, into wide-area search and target-specific detection. The device status monitoring unit reads the remaining battery power of the drone and the core temperature of the programmable multispectral illumination matrix (hereinafter referred to as the illumination module core temperature in the subsequent description of the method embodiment) in real time, serving as inputs for power consumption and thermal constraints during subsequent optimization.

[0114] The vehicle-mounted perception layer acquires vehicle status data from the vehicle itself via the vehicle bus interface, which uses either CAN FD or automotive Ethernet protocol. The acquired vehicle status data includes real-time vehicle speed, steering wheel angle, current headlight mode (high beam / low beam / off), and the vehicle's own GPS location. The vehicle's own GPS location is used to establish the spatial relationship between the vehicle and the drone, and the steering wheel angle is used to predict the vehicle's next trajectory so that the light field can be adjusted in advance.

[0115] Step 1.3 Spatiotemporal alignment and fusion.

[0116] Once enough data for one synchronization cycle has accumulated in the buffers of the three data sources, the spatiotemporal alignment and fusion sub-step reads all the raw data from the most recent sampling cycle from each buffer and enters... Figure 2The lower section consists of three processing nodes arranged sequentially: timestamp synchronization, spatial coordinate system one, and generation of fused data frames. Timestamp synchronization is performed based on the hardware timestamp synchronization clock component in the vehicle-airport cooperative communication network, using Precise Time Protocol (PTP) or GPS timing to unify the timestamps of all data streams to the same time base. For data streams with a native sampling frequency higher than the system synchronization period, nearest neighbor sampling is used to retain the sampling point closest to the synchronization period timestamp; for data streams with a native sampling frequency lower than the system synchronization period, timestamp interpolation or zero-order hold methods are used for resampling to ensure that the time alignment accuracy of all data sources in the fused data frame is not lower than the preset time synchronization threshold. Spatial coordinate system one transforms all spatial data from different sensors and reference frames into a unified vehicle coordinate system with the vehicle's center of mass as the origin, making the positions of the UAV relative to the vehicle and the target relative to the vehicle clear and definite geometric quantities.

[0117] After spatiotemporal alignment is completed, the system packages all synchronized data into a structured fused data frame. The fused data frame is the system's sole authoritative description of the current external environment and internal state, serving as the direct input for intelligent scene recognition and deconstruction in step two.

[0118] Step 2: Intelligent recognition and deconstruction of task scenarios.

[0119] Reference Figure 3 The logic diagram for step S2 is as follows: step two unfolds sequentially in three stages: scene feature extraction, requirement parameter deconstruction and calculation, and output of structured instructions.

[0120] Step 2.1: Scene feature extraction.

[0121] Figure 3 The first stage of scene feature extraction comprises three parallel feature parsing sub-blocks: spatial relationship parsing, ambient light semantic analysis, and dynamic context understanding. This embodiment further refines these three into four types of feature vectors: spatial relationship features, ambient light semantic features, meteorological perception features, and dynamic context features.

[0122] Spatial relationship features are calculated based on a unified vehicle coordinate system and include the relative distance, azimuth, and altitude difference between the UAV and the target, as well as spatial proximity to determine whether the UAV is within the glare-sensitive zone in front of the vehicle. The relative distance, denoted as target distance, is the straight-line distance between the UAV and the main subject or region of interest. The azimuth reflects the horizontal and vertical angles of the target in the UAV or vehicle coordinate system. The altitude difference reflects the relative altitude between the UAV and the subject or the ground. Spatial proximity is a Boolean or continuous value indicator used to determine whether the target is within the glare-sensitive zone in front of the vehicle.

[0123] In a typical implementation, the geometric range of the aforementioned glare-sensitive area in front of the vehicle is defined as a wedge-shaped field of view extending along the vehicle's forward direction with the driver's eye position as the apex. Specifically, this wedge-shaped field of view starts from the driver's eye position, extends horizontally to the left and right with a preset horizontal half-angle (typically 30 degrees), vertically upwards and downwards with a preset vertical half-angle (typically 15 degrees), and extends radially along the vehicle's forward direction for a preset sensitive distance (typically 15 meters). When light emitted by a drone or other light source enters the aforementioned wedge-shaped field of view, the system determines that there is a glare risk. The spatial proximity is calculated as follows: taking the drone's current position as the evaluation point, if the evaluation point is within the wedge-shaped field of view, the spatial proximity is 1; otherwise, it is 0 (in Boolean output mode), or a continuous value between 0 and 1 is taken according to the normalized distance of the evaluation point from the central axis of the wedge-shaped field of view (in continuous output mode). This geometric definition also serves as the basis for specifying the glare risk value GlareRisk(X) in the sensitive area S set in formula (3c).

[0124] Ambient light semantic features are derived from the analysis of ambient light sensor and depth image information. They include a base illuminance level, light contrast for determining whether high dynamic range (HDR) lighting is needed, and spectral features for identifying external light source interference. Typical base illuminance levels include extremely dark, low light, and ambient light. Light contrast is represented by the ratio of the brightest to the darkest area in the scene. When this ratio exceeds a preset threshold, it indicates a large dynamic range, requiring the activation of an HDR lighting strategy. Spectral features are used to identify the presence of specific light sources in the scene, such as sodium lamps and LED billboards. These sources may interfere with subsequent visual tasks and need to be compensated for in color temperature decisions.

[0125] Meteorological perception features include meteorological influencing factors. These factors are derived from meteorological parameters collected by meteorological sensors or image fog detection algorithms, and range from 0 to 1, where 0 indicates good weather conditions (sunny) and 1 indicates severe weather conditions (dense fog or heavy rain).

[0126] Dynamic context features integrate task instructions, device status, and vehicle dynamic information, including task scenario labels, the relative motion state between the UAV and the vehicle, and system constraint states such as remaining battery power and lighting module temperature. The task scenario label indicates the type of the current task and is output by the task instruction parsing unit. The relative motion state reflects typical motion relationships between the UAV and the vehicle, such as stationary, low-speed in the same direction, or high-speed following. The system constraint states, including the current remaining battery power and the core temperature of the lighting module, serve as the real-time basis for power consumption and thermal constraints in subsequent optimization models.

[0127] After the feature extraction sub-step is completed, the four types of feature vectors are uniformly packaged into a structured scene feature data package. This data package contains four fields corresponding to the four types of features, and each field stores the specific feature items in the form of key-value pairs. The scene feature data package, as the sole output of the scene feature extraction sub-step, is passed to the next sub-step through the internal data bus for the scene classifier decision sub-step to read as needed.

[0128] Step 2.2: Scene classifier decision.

[0129] Figure 3 The diamond-shaped decision nodes in the middle of the scene classifier correspond to the scene classifier decision sub-step. The scene classifier is a lightweight hybrid decision model that takes at least one of the four feature vectors extracted in the previous stage as input and outputs refined scene labels. In typical implementations, the lightweight hybrid decision model can employ decision trees, lightweight random forests, or pre-trained shallow neural networks, with computational complexity sufficient to meet the millisecond-level decision-making requirements on automotive embedded platforms.

[0130] The refined scene labels are obtained by combining the task scene labels with the conditional parameters in the feature vector. In this embodiment, the refined scene labels include, but are not limited to, low-speed close-range tracking scenes, medium-to-long-range search and inspection scenes, and return-to-home and precise landing scenes. Specifically, low-speed close-range tracking scenes correspond to a combination of features where the relative distance is less than a preset close-range threshold, the relative motion state indicates low speed in the same direction, and the task scene label is a tracking task. The preset close-range threshold is typically set to 15 meters. Medium-to-long-range search and inspection scenes correspond to a combination of features where the relative distance is greater than the preset close-range threshold, the target exhibits an uncertain or large-scale distribution characteristic, and the task scene label is a search task. The uncertain or large-scale distribution characteristic of the target is evaluated based on the deep point cloud data of the fused data frame: when the number of candidate targets detected based on the point cloud exceeds a preset target number threshold (typically 3) or the spatial span of the candidate targets in the horizontal direction exceeds a preset span threshold (typically 20 meters), the current scene is determined to meet the target uncertain or large-scale distribution characteristic. The return-to-home and precision landing scenarios are characterized by the drone's location information being close to the helipad (typically, the three-dimensional distance is less than 5 meters) and the mission scenario being labeled as precision landing.

[0131] After the refined scene labels are generated, they serve as the final task scene labels determined within the lighting requirement parameter set. The output of the scene classifier decision sub-step overwrites the original task scene labels output by the task instruction parsing unit. This means that in all subsequent calculations referencing the task scene labels, the refined scene labels obtained through the scene classifier are actually used. This overwriting mechanism ensures that all downstream decisions of the system are based on a more accurate and granular understanding of the scene.

[0132] The specific configuration of the scene classifier is further explained below. In this embodiment, the scene classifier can be implemented using one of three typical methods: rule-based decision tree, statistical learning-based random forest, or pre-trained shallow neural network.

[0133] In the rule-based decision tree implementation, the scene classifier executes decisions in a hierarchical order: first, it judges the task scene label; then, it judges the spatial relationship conditions; and finally, it judges the dynamic context conditions. The first-level root node of the decision tree is divided into three branches based on the original task scene label: the tracking task branch, the search task branch, and the precise landing branch. Under the tracking task branch, the second-level node judges based on the relative distance in the spatial relationship features: if the relative distance is less than a preset close-range threshold (typically 15 meters), it enters the third-level node, which further judges based on the relative motion state in the dynamic context features; if the relative motion state indicates low speed in the same direction (typically defined as the relative speed between the drone and the vehicle being less than 20% of the vehicle speed and in the same direction), it outputs a refined scene label of "low-speed close-range tracking scene"; otherwise, it outputs a general tracking scene label. Under the search task branch, the second-level node judges based on whether the relative distance is greater than the preset close-range threshold; if it is greater, it outputs "medium-to-long-range search and inspection scene". Under the precision landing branch, the second-layer node determines the location based on the Euclidean distance between the UAV's positioning information and the helipad coordinates in the dynamic context features. If the distance is less than the helipad proximity threshold (typically 5 meters), it outputs "Return to Home and Precision Landing Scenario". The branch judgment thresholds of the above decision tree can all be adjusted through the parameter configuration file to adapt to different vehicle models and different application requirements.

[0134] In the implementation based on pre-trained shallow neural networks, the typical network structure of the scene classifier is a multilayer perceptron structure consisting of an input layer, one or two fully connected hidden layers, and an output layer. The number of nodes in the input layer is the same as the total dimension of the four-class feature vectors after standardization, typically ranging from 16 to 32. The number of nodes in each hidden layer is typically configured to be 1.5 to 2 times the input dimension, using ReLU or Leaky ReLU activation functions to balance nonlinear expressiveness and computational efficiency. The output layer is a Softmax layer, with the number of nodes equal to the number of categories for refined scene labels (3 in this example, corresponding to three typical refined scenes; this is only an example, and the labels and their number can be set according to actual needs). The network training data comes from a set of labeled samples collected by the vehicle during actual nighttime operations, supplemented by data augmentation. Training uses the cross-entropy loss function and the Adam optimizer for end-to-end training. The trained network model is deployed to the vehicle-mounted embedded platform in quantized compressed form, with a typical single inference time of no more than 1 millisecond.

[0135] In the random forest-based implementation, the scene classifier consists of an ensemble of multiple decision trees, typically 10 to 30 trees. The maximum depth of each tree is limited to 4 to 6 layers to avoid overfitting and control inference time. The random forest's decision-making is based on the majority vote of all trees, resulting in stronger robustness and generalization ability compared to a single decision tree.

[0136] Regardless of the implementation method used, the scene classifier provides a unified input / output interface after deployment: the input is a scene feature data package, and the output is refined scene labels and their confidence values. When the confidence value is lower than a preset threshold (typically 0.6), the system can further trigger an auxiliary judgment process or temporarily maintain the original task scene labels to avoid misclassification leading to incorrect decisions.

[0137] Step 2.3: Deconstruction and calculation of demand parameters.

[0138] After the scene classifier decision sub-step is completed, the refined scene labels cover the original task scene labels as task scene labels in the lighting requirement parameter set, and then proceed to the requirement parameter deconstruction and calculation sub-step. Figure 3 The second phase, which involves deconstructing and calculating the required parameters, presents three sub-blocks: core lighting area calculation, required illuminance calculation, and beam characteristic calculation. Figure 3 (Presented as D1 structure groups). All three are executed based on the covered task scene labels as the task scene basis. Here, the task scene labels actually refer to the refined scene labels output by the scene classifier, and the same applies below.

[0139] The three computational sub-blocks are not entirely independent but have clear data dependencies. The computational output of the core lighting area is jointly used by the required illuminance calculation and the beam characteristic calculation as a benchmark for the spatial extent; the required illuminance L... req The calculation results are further referenced in the color temperature dynamic decision sub-step of beam characteristic calculation as the basis for color temperature selection of the default adaptive layer.

[0140] The core lighting area calculation is based on the target detection bounding box, UAV attitude information, and relative distance, azimuth, and height differences in spatial relationship features. It represents the 3D area that needs to be preferentially illuminated in the current scene in the form of a 3D spatial boundary within the vehicle coordinate system. The specific process of core lighting area calculation is further explained below to enable those skilled in the art to implement it.

[0141] The calculation of the core illumination area unfolds sequentially through four sub-steps: target spatial location backprojection, coordinate system transformation, 3D spatial boundary construction, and illumination area expansion. In the target spatial location backprojection sub-step, the system combines the point cloud data output by the depth camera with the 2D boundary of the target detection box. Through perspective backprojection of the pinhole camera model, the 3D position of the target in the depth camera coordinate system is obtained. In a typical implementation, the median of the depth values ​​corresponding to all pixels within the detection box is taken as the representative distance of the target. The four vertices and the center point of the detection box are then backprojected to obtain the boundary vertices and the center point of the target point cloud.

[0142] In the coordinate system transformation sub-step, the system sequentially transforms the target spatial position from the depth camera coordinate system to the vehicle coordinate system through three sets of pre-calibrated transformation matrices: first, it transforms to the UAV body coordinate system through the extrinsic parameter matrix of the depth camera on the UAV; then, it transforms to the geographic inertial coordinate system by combining the UAV attitude information (including pitch angle, roll angle, and heading angle) and positioning information; finally, it transforms to the vehicle coordinate system with the vehicle's own GPS position and heading angle as the origin, thus obtaining the three-dimensional center point coordinates and boundary vertex coordinates of the target in the vehicle coordinate system.

[0143] In the 3D space boundary construction sub-step, the system uses the 3D center point of the target in the vehicle coordinate system as the centroid, and combines the image size of the target detection box with the relative distance to infer the approximate length, width, and height of the target in 3D space. Then, it constructs a 3D axis-aligned cuboid bounding box with the target center as the centroid and its boundary aligned with the coordinate axes of the vehicle coordinate system as the target space boundary. In cases where the target shape is significantly non-aligned with the coordinate axes of the vehicle coordinate system, a directed cuboid bounding box can be further constructed using azimuth information to more accurately fit the target posture.

[0144] In the illumination area expansion sub-step, the system expands outward from the target space boundary according to a preset expansion rule to obtain the three-dimensional spatial boundary of the core illumination area. The expansion rule is configured according to the task scenario label: for the tracking task, it expands uniformly in all directions around the target to a preset tracking expansion coefficient (typically 1.5 to 2 times the target size); for the search task, it expands horizontally around the target to a preset search coverage radius (typically 10% to 30% of the relative distance) while maintaining the target size vertically; for the precision landing task, it expands with the center of the apron as the centroid, covering the entire apron area plus a preset safety margin (typically 1 to 2 meters). After expansion, the three-dimensional spatial boundary of the core illumination area is output as a set of coordinates of eight vertices or as a compact parameter form of the center point coordinates plus length, width, height, and attitude angle, serving as the specific value of the core illumination area field in the illumination requirement parameter set.

[0145] The required illuminance is calculated using a task gain adaptive mathematical model based on Retinex theory, with feedforward pre-calculation of the required illuminance in the core lighting area. The required illuminance is denoted as Lreq, and its calculation formula is as follows: (1) The meanings of each physical quantity in formula (1) are as follows. L env This represents the ambient baseline illuminance, derived from the baseline illuminance level in the ambient light semantic features, with a typical value range of 1 to 100 lux. base This is the basic safety supplemental lighting gain, used to ensure the minimum illumination requirements for nighttime operations, typically ranging from 20 to 50 lux. task A task-specific gain coefficient is determined based on the task scenario label. When the task scenario label indicates a search task, G... task The value is set greater than the baseline gain coefficient to provide sufficient illumination for the search task, thereby enhancing the ability to identify specific materials or biological targets. Sens task The task spectral sensitivity factor is also determined based on the task scenario label. When the task scenario label indicates a search task, Sens... task If the value exceeds a preset sensitivity threshold at a specific wavelength (such as the 650 nm red light band), the system is guided to increase the optical power output in that wavelength band. weather This is a meteorological compensation gain, used to compensate for light scattering losses under adverse weather conditions such as fog, rain, and snow. F weather The meteorological influencing factor is derived from meteorological parameters collected by meteorological sensors or image haze detection algorithms, and its value ranges from 0 to 1.

[0146] The beam characteristic calculation output includes at least one of the following: beam divergence angle, target color temperature, special band activation indicator, and polarized illumination activation indicator. The beam divergence angle is determined based on the spatial dimensions of the core illumination area and the relative distance between the UAV and the core illumination area in the spatial relationship characteristics. In tracking tasks where the relative distance is within a preset close-range condition, the beam divergence angle is set to a preset narrow beam range (typically 5 to 15 degrees); in search tasks, the beam divergence angle is set to a preset wide beam range (typically 30 to 60 degrees). The target color temperature is determined through a color temperature dynamic decision sub-step, which is implemented according to a three-layer decision logic: default adaptive layer, weather trigger layer, and task trigger layer. The default adaptive layer adjusts the illuminance L according to requirements. req Adaptive color temperature selection. When L req When in the high brightness range (e.g., greater than 200 lux), the color temperature shifts towards a cooler color temperature (e.g., above 5500K) to enhance visual sharpness; when L reqWhen in a low-brightness range (e.g., less than 50 lux), the color temperature shifts towards a warmer color temperature (e.g., below 3500K) to reduce visual fatigue. When the visibility data collected by the weather sensor is lower than a preset visibility threshold (e.g., less than 500 meters), the weather-triggered layer forces the color temperature to be no less than a high color temperature threshold (e.g., 6000K), utilizing the stronger penetrating power of cool white light in fog to enhance target recognition. When the task-triggered layer indicates portrait tracking, it sets the color temperature to a preset low-to-medium color temperature range (e.g., 3000K to 4000K) to optimize the visual reproduction of skin tones. The decision priority of the weather-triggered and task-triggered layers is higher than that of the default adaptive layer; that is, when one of the trigger conditions of the two layers is met, its decision result overrides the output of the default adaptive layer.

[0147] The polarized illumination enable flag is used to handle situations with a large amount of specular reflection in the scene (such as slippery roads or vehicle glass surfaces). When the system detects that the proportion of specular reflection in the scene exceeds a preset reflection threshold, it outputs a control command to enable the ring polarized illumination mode in the illumination requirement parameter set. The special band enable flag is used to enable a preset specific band (such as the red light band in a search task) in scenarios where it is necessary to enhance the sensitivity of specific target recognition.

[0148] Step 2.4: Output structured instructions.

[0149] Figure 3 The third-stage output structured instruction box includes four outputs: task metadata (scene tags and priority), core lighting region (coordinates and range), and required illuminance L. req (Unit: lux), beam characteristic set (divergence angle, color temperature, polarization, etc.). The system packages the above calculation results into a structured lighting requirement parameter set, using a key-value pair nested data structure (in typical implementations, JSON objects or Protocol Buffers encoding can be used). Task metadata is a first-level field, containing task scene tags (here, refined scene tags) and task priority; the core lighting area is a first-level field, representing the spatial boundary in the vehicle coordinate system as a three-dimensional vertex coordinate array; the required illuminance is a first-level field, storing the Lreq value as a floating-point number; the beam characteristic set is a first-level field, containing four sub-fields: beam divergence angle, target color temperature, special band activation flag, and polarization lighting activation flag. This lighting requirement parameter set is accompanied by a parameter set timestamp generated by a hardware timestamp synchronization clock component, serving as the sole explicit input for the model-based dynamic light field collaborative calculation in step three.

[0150] The task priorities in the lighting requirement parameter set are determined jointly by a preset task scenario priority mapping table and real-time event states during task execution. The preset task scenario priority mapping table assigns initial priority levels to tasks based on task scenario labels; for example, the initial priority of return-to-home and precision landing scenarios is higher than that of search scenarios. Real-time event states are event states triggered by changes in sensing data or target detection results during task execution, including target detection, target loss, and environmental anomalies. Task priorities are dynamically adjusted based on the type of real-time event state. A typical example is: when a target detection event occurs during the execution of a search task (i.e., the system detects a suspected search target for the first time), the task priority is dynamically adjusted to the highest level and serves as one of the trigger conditions for switching the weight coefficients of the multi-objective cost function.

[0151] Step 3: Model-based dynamic light field collaborative computation.

[0152] Step 3 receives the lighting demand parameter set from Step 2 as the sole input, referring to... Figure 4 The logic diagram for step S3 unfolds sequentially through three stages: optimization model initialization and construction, core optimization solution loop, and result verification and output. Fields in the lighting demand parameter set, such as task scene label, task priority, core lighting area, required illuminance, and beam characteristic set, are referenced at different stages of step three: the task scene label and task priority serve as the basis for dynamically adjusting the weight coefficients α, β, γ, and δ in the multi-objective optimization function F; the core lighting area serves as the target area for light propagation simulation and the evaluation space for illuminance distribution; the required illuminance Lreq serves as the reference benchmark for the lighting performance error sub-objective PerfErr; and the beam characteristic set serves as a reference for the initial value estimation and constraint conditions of the decision variable X.

[0153] Step 3.1: Optimize model initialization and construction.

[0154] Figure 4 The first stage consists of four sequential nodes: defining decision variables X (parameter matrices of each light source), loading real-time system states (power / temperature / location), constructing a multi-objective optimization function F, and setting physical and safety constraints. These are explained below.

[0155] The decision variable X is defined as a vector of all controllable light source parameters in the entire system, consisting of three parts: (2) The subvectors in formula (2) are constructed as follows. X uav This is a terminal vector for the UAV, containing the brightness I, color temperature T, beam divergence angle θ, and horizontal deflection angle φ of each sub-module in the programmable multispectral illumination matrix. h With vertical deflection angle φ v Xvehicle For vehicle terminals, X represents the vertex coordinates and average brightness of the polygon representing the illumination area of ​​the adaptive digital headlights or front auxiliary lighting. pad This is the terminal vector for the helipad, containing the on / off status and operating mode identifier of the helipad lighting subsystem.

[0156] The system loads the latest real-time system state from the data buffer as the basis for constraints, including the drone's real-time battery power (which determines the upper limit P of the total available power). max ), Current temperature of the lighting module (determines the maximum allowable heat load T) max (and updated spatial relationships).

[0157] The multi-objective optimization function F is constructed as a weighted summation: (3) The meanings of the four sub-objective functions in formula (3) are as follows. Perf Err (X) represents the lighting performance error; calculate the actual illuminance distribution L in the target area. actual (X) and required illuminance L req The differences between them are the most crucial sub-objectives in the optimization process. (Pwr) Cost (X) represents the total power consumption cost, which is the sum of the instantaneous total power consumption of all activated light sources. Glare Risk (X) represents the glare risk value, which quantifies the risk of direct or strong reflection of light hitting the driver's field of vision or the drone's camera. Thermal Load (X) represents the estimated heat load, which is a prediction of the temperature rise of the lighting module based on the current power consumption and heat dissipation model.

[0158] The weighting coefficients α, β, γ, and δ are dynamically adjusted based on the refined scene labels and task priorities identified in step two: In extreme search tasks, α is set to a very high value and β to a medium value, with the system pursuing the best lighting effect regardless of power consumption; in long-endurance cruise inspection tasks, β and δ are set to higher values, with the system prioritizing battery life and cooling; in tasks where someone is filming inside the vehicle, γ is set to the highest value to ensure absolute safety against glare.

[0159] The specific calculation methods for the four sub-objective functions are further given below. The lighting performance error PerfErr(X) is measured by the average of the sum of squares of the differences between the actual illuminance and the required illuminance at each sampling point within the core lighting area. The specific calculation formula is as follows: (3a) In formula (3a), Nroi is the total number of sampling points in the core lighting area, and L actual,i (X) represents the actual illuminance at the i-th sampling point under the decision variable X, and L req,iLet N be the required illuminance for the i-th sampling point. Sampling points within the core lighting area are divided into a uniform grid, with a typical grid spacing of 0.1 meters to 0.5 meters, and the total number of sampling points is N. roi The value is adaptively determined based on the size of the core lighting area, with a typical range of 100 to 2000.

[0160] Total power consumption cost Pwr Cost (X) represents the sum of the instantaneous power consumption of all activated light sources in the system, calculated using the following formula: (3b) In formula (3b), K is the set of indices of all activated light sources under the current decision variable X (covering UAV, vehicle, and helipad ends). Let I be the power consumption function of the k-th light source under brightness Ik and color temperature Tk. The typical power consumption function of an LED light source can be represented by a linear model P. k =P 0,k ·I k Fit, where P 0,k Let X be the rated power consumption of the k-th light source at full brightness. It should be noted that the power consumption is mainly determined by the driving current and thus reflected in the brightness and color temperature parameters. The geometric parameters such as beam divergence angle, horizontal deflection angle, and vertical deflection angle in the decision variable X are controlled by the optical shaping mechanism or the optical surface of the electrically tunable metamaterial. This control process itself does not change the power consumption of the light source. Therefore, the power consumption function in formula (3b) only explicitly depends on the brightness and color temperature components.

[0161] Glare risk value Risk (X) is a weighted sum of the ratios of the direct or strongly reflected luminous flux received in the glare-sensitive area to a reference luminous flux threshold. The specific calculation formula is as follows: (3c) In formula (3c), S is the set of indices for glare-sensitive areas, typically including the driver's field of view and the UAV's self-camera field of view; Φ j (X) represents the sum of direct and strongly reflected light flux received by the j-th sensitive region under decision variable X, output by the light propagation simulation sub-step; Φ ref,j The reference luminous flux threshold for the j-th sensitive area is calibrated based on the glare perception threshold, with a typical value of 0.1 lumens per square meter; w j The weighting coefficient for the j-th sensitive area reflects the severity of glare in that area to personal or task safety, with a typical value of 0.5 to 1.0.

[0162] Thermal load estimate Load (X) The predicted temperature rise of the lighting module is calculated based on the instantaneous power consumption and a simplified thermal resistance-capacity network model. The specific calculation formula is as follows: (3d) In formula (3d), Tcurrent is the current temperature of the lighting module, which is collected in real time by a thermistor; Rth is the equivalent thermal resistance from the lighting module to the environment, typically ranging from 1 to 5 Kelvin per watt; C th The equivalent heat capacity of the lighting module is typically 5 to 20 joules per Kelvin; Δt is the prediction time window length, typically 500 milliseconds to 2 seconds. These thermal resistance and heat capacity parameters are obtained through thermal simulation or actual measurement calibration during the development phase of the lighting module and are used as constants during operation.

[0163] Furthermore, to ensure that the four sub-objectives—lighting performance error, total power consumption cost, glare risk value, and heat load estimation—are reasonably weighted in the same cost function, each sub-objective is normalized before entering formula (3): PerfErr divided by the square of Lreq yields the dimensionless relative error; PwrCost divided by Pmax yields the relative power consumption; GlareRisk is already a dimensionless ratio; and ThermalLoad divided by Tmax yields the relative temperature rise. All four normalized sub-objectives fall within the range of 0 to 1.

[0164] Physical and safety constraints set hard boundaries for the optimization problem, including the following four rigid constraints: (4) (5) (6) Formula (4) is the total power consumption constraint. The total power consumption must not exceed the maximum power limit P determined by the current power level. max P max The remaining battery capacity is dynamically determined based on dynamic context features. Formula (5) is a thermal constraint; the predicted heat load value must not exceed the highest temperature limit T predicted by the current temperature and historical temperature rise. max T max The temperature of the lighting module is dynamically determined based on the dynamic context features. Formula (6) is the glare safety constraint; the glare risk value must not exceed the glare safety threshold G set by regulations or experience. th Furthermore, the brightness and angle of each light source have an adjustable physical range, constituting a fourth type of rigid constraint, namely, each component of the decision variable X must be within the physically adjustable range of the corresponding actuator.

[0165] Step 3.2: Core optimization solution loop.

[0166] After the optimization model is built, the system enters the core optimization solution loop. Before the loop starts, the initial value of the decision variable X is selected according to the following rules: if the current time is the first startup of the system or the value of X in the previous cycle...optimal If the value has expired, the initial parameters (typical color temperature, typical divergence angle, etc.) indicated by the beam characteristic set in the lighting demand parameter set will be used as the initial value of X; if X from the previous period... optimal Within the effective window, X from the previous period optimal As the initial value for the current period X, the continuity of the optimization results is used to accelerate convergence. Figure 4 The second stage presents the iterative process of the core optimization solution loop, including five nodes: running real-time light field simulation based on the current X, calculating the values ​​of each sub-objective function (performance / energy consumption / glare / heat load), aggregating and calculating the total objective function value F, checking whether all constraints are satisfied and the F value converges, and updating the decision variable X with the optimization algorithm (such as SQP / particle swarm optimization algorithm).

[0167] The multi-objective optimization solution process includes a light propagation simulation sub-step. In each optimization iteration, based on the current decision variable X, the light propagation simulation sub-step simulates the light propagation of the light sources carried by the UAV-side intelligent lighting module, the vehicle lighting subsystem, and the helipad lighting subsystem, and outputs the illuminance distribution L of each light source in the core lighting area. actual Glare risk values ​​in the pilot's field of vision and at the drone's own camera. Risk As an intermediate variable, the ray propagation simulation uses point cloud data from fused data frames as an approximation of the 3D scene. To achieve millisecond-level computation speed, the simulation employs a highly optimized algorithm and simplifies sampling for non-critical regions.

[0168] The specific algorithm for ray propagation simulation is explained in further detail below. Ray propagation simulation consists of five stages: light source modeling, ray emission, propagation attenuation, occlusion determination, and region accumulation.

[0169] In the light source modeling stage, each sub-module in the UAV-side programmable multispectral lighting matrix, each pixel unit of the adaptive digital headlight in the vehicle lighting subsystem, and each LED emitter in the helipad lighting subsystem are modeled as surface light sources with Lambertian radiation characteristics. To facilitate the handling of the three-end light sources in a unified mathematical expression, all activated light sources at the three ends are uniformly indexed with k, corresponding to X, in the order of UAV-side, vehicle-side, and finally helipad-side. uav X vehicle X pad The order of the components in the three sub-vectors. The radiant intensity of each light source is proportional to the corresponding luminance component Ik in the decision variable X, and the radiant direction distribution is determined by the beam divergence angle θ in the current decision variable X. k Horizontal deflection angle φ h,k With vertical deflection angle φ v,k Decision. The expression for the Lambert radiation model is: (7) In formula (7), I0 is the peak radiation intensity of the light source along the normal direction, and θ is the angle between the observation direction and the normal of the light source. This model can be further modified into a directional model with a Gaussian distribution when the divergence angle of the light source is small, so as to more accurately reflect the radiation pattern of actual LED and laser light sources.

[0170] The light emission stage starts from the position of each activated light source and emits sampled rays according to a preset angular density. A typical emission strategy employs layered uniform sampling, that is, generating sampling directions according to a latitude and longitude grid within the solid angle covered by the divergence angle of the light source. The typical number of sampled rays for each light source is 100 to 500. The total number of sampled rays increases linearly with the total number of light sources. The total number of rays simulated by the entire system in each iteration is typically 5,000 to 20,000, ensuring both simulation accuracy and controlling computation time.

[0171] The propagation attenuation mechanism calculates the decrease in light intensity with propagation distance based on the inverse square law. That is, in the absence of meteorological attenuation, the radiant intensity of light reaching a distance *d* is the peak intensity at the light source divided by the square of *d*. In the case of meteorological influence, this is further multiplied by the Beer-Lambert attenuation factor, expressed as: (8) In formula (8), E(d) is the irradiance after the light travels a distance d, and σ is the meteorological attenuation coefficient, the value of which is related to the meteorological influence factor F. weather Positive correlation, typically σ = k·F weather , where k is the calibration coefficient, typically ranging from 0.01 to 0.1 per meter. This attenuation model can effectively simulate the scattering loss of light energy under adverse weather conditions such as fog, rain, and snow.

[0172] The occlusion detection step uses the 3D point cloud data from the fused data frames to determine whether each sampled ray is occluded by obstacles in the scene (such as vehicles, road bumps, trees, etc.) along its propagation path. To accelerate computation, the point cloud data is preprocessed into an octree spatial index structure before entering the ray propagation simulation, and the intersection test between the ray and the point cloud is completed within the logarithmic time complexity of the octree. Once a ray is determined to be occluded, its subsequent propagation energy is set to zero, and it no longer contributes to the target area.

[0173] The regional accumulation step weights and accumulates the contributions of all unobstructed sampled light rays at each sampling point in the core illumination area to form the actual illuminance L at that sampling point. actual,i(X). Glare risk value is also calculated using an accumulation method, by statistically analyzing the direct light flux or strongly reflected light flux within the driver's field of view and the drone's self-camera field of view, thus obtaining GlareRisk(X). The light propagation simulation is implemented in parallel computing, with simulations of light sources independent of each other. This fully utilizes the parallel capabilities of the GPU or NPU within the vehicle's integrated processing unit, with a typical simulation time of 3 to 8 milliseconds.

[0174] The multi-objective cost function uses the illuminance distribution and glare risk value output from the light propagation simulation sub-step as intermediate quantities to calculate four sub-costs: lighting performance error, total power consumption cost, glare risk sub-cost, and heat load prediction value. Since the four sub-costs have different dimensions (illuminance, power, dimensionless risk value, and temperature), each sub-cost is first normalized using its corresponding benchmark value before weighted summation, mapping its numerical range to a dimensionless interval between 0 and 1. The normalized four sub-costs are then weighted and summed using weight coefficients α, β, γ, and δ to obtain the value of the multi-objective cost function. The weight coefficients are dynamically adjusted according to the task scenario label and task priority, typically ranging from 0 to 1 with α+β+γ+δ=1, to achieve adaptive optimization objective balance for the task scenario.

[0175] The optimization solution employs algorithms suitable for real-time computation, typically Sequential Quadratic Programming (SQP) or Constrained Particle Swarm Optimization (PSO). When using SQP, the algorithm approximates the objective function quadratically and the constraints linearly in each iteration. The iterative update direction of the decision variables is obtained by solving the quadratic programming subproblems. Typically, SQP has a maximum of 20 iterations, with a time budget of 5 to 15 milliseconds for a single optimization. When using PSO, the algorithm maintains a swarm of particles searching for the optimal solution in the decision variable space. A typical particle count is 20 to 50, with inertia weights typically ranging from 0.7 to 0.9, and cognitive and social coefficients typically ranging from 1.5 to 2.0. The maximum number of iterations is 30, with a time budget of 10 to 30 milliseconds for a single optimization, ensuring the entire system completes the solution within milliseconds. The two algorithms are dynamically selected based on the real-time requirements of the mission scenario: for high real-time scenarios such as precise landing, the SQP algorithm is preferred to take advantage of its deterministic convergence characteristics; for scenarios with multiple local optima such as wide-area search, the PSO algorithm is used to take advantage of its global search capability.

[0176] The PSO algorithm employs a dynamic penalty function method when dealing with the rigid constraints listed in equations (4) to (6). The dynamic penalty function method extends the objective function F(X) into an augmented objective function F that considers constraint violations. augThe augmented objective function (X) is constructed as follows: For each constraint inequality g(X) ≤ 0, the square of the violation amount max(0, g(X)) is added to the objective function with a penalty factor that increases with the iteration number. The mathematical expression of the augmented objective function is: (9) In formula (9), m is the total number of constraints (in this embodiment, m=4 corresponds to four rigid constraints), g j (X) is the violation function of the j-th constraint, λ j Let λ be the penalty factor for the j-th constraint. The penalty factor increases with the number of iterations during the optimization process to strengthen the constraint-driven effect. A typical dynamic adjustment rule for the penalty factor is λ. j (k+1) = ρ·λ j (k), where k is the current iteration number, ρ is the penalty factor increment coefficient, typically ranging from 1.2 to 1.5, and the initial penalty factor λ. j (0) Typical values ​​range from 10 to 100. Through this dynamic penalty function mechanism, the particle swarm can explore the decision variable space more loosely in the early stage to find the neighborhood of a high-quality solution, and gradually strengthen the constraint drive in the later stage to make the particles converge to the optimal solution in the feasible solution space. In addition to the dynamic penalty function method, alternative constraint handling methods include constraint repair strategies—that is, after each iteration, particles that violate the constraints are pulled back into the feasible solution space by correcting them according to the boundary projection or gradient direction; this embodiment prioritizes the dynamic penalty function method to avoid the computational overhead caused by frequent repair operations.

[0177] The optimization algorithm intelligently generates a new set of potentially better decision variables X based on the current F(X) value and constraint violations. new The algorithm then returns to the real-time light field simulation node, re-simulates and evaluates the light field based on Xnew, and enters the next iteration loop. Convergence is determined according to the following rules: If the current iteration's X simultaneously satisfies all rigid constraints, and the rate of change of the objective function F value relative to the previous iteration is less than a preset convergence threshold (typically 1%), and this condition is met for a preset number of consecutive iterations (typically 3), then convergence is determined, the loop exits, and X at this point is the global optimal solution X at the current moment. optimal If convergence is not achieved after reaching the maximum number of iterations, then the X with the smallest current F value that satisfies all rigid constraints is taken as X for this period. optimal This ensures that the optimization process always returns a feasible solution within the time budget.

[0178] Step 3.3: Result verification and output.

[0179] Figure 4 The third stage includes three nodes: the global optimal solution X. optimal Decomposition of X optimalThis provides the control instruction set for each subsystem and outputs the globally optimal control parameter set. The globally optimal solution X. optimal Given a parameter vector arranged in the order of decision variable definition, decompose it into three independent sub-instruction sets according to a preset protocol: X optimal The fields such as brightness, color temperature, beam divergence angle, horizontal deflection angle, and vertical deflection angle of each sub-module of the corresponding programmable multispectral illumination matrix are extracted and encapsulated into UAV-side control sub-instructions, and X is used to... optimal The fields such as vertex coordinates and average brightness of the illumination area polygon corresponding to the adaptive digital headlights or front auxiliary lighting are extracted and encapsulated into vehicle-side control sub-instructions, and X is used to... optimal Fields such as the on / off status and working mode identifier of the corresponding helipad lighting subsystem are extracted and encapsulated into helipad-side control sub-instructions. These three sub-instruction sets are related to X. optimal Together, they form a globally optimal set of control parameters, which are then uniformly stamped with a high-precision generated timestamp T by a hardware timestamp synchronization clock component. generation This is then used as input for the fourth step of the hierarchical collaborative light field execution.

[0180] Step 4: Hierarchical collaborative light field execution.

[0181] Step four receives the global optimal control parameter set and its generation timestamp T from step three. generation As input. (Refer to...) Figure 5 The logic diagram for step S4 is as follows: step four unfolds sequentially in three stages: instruction reception, verification and distribution, multi-channel synchronous execution, and execution status monitoring and feedback.

[0182] Step 4.1: Command reception, verification and distribution.

[0183] Figure 5 The first phase includes nodes such as receiving instructions and timestamps, pre-execution comprehensive verification, verification pass / fail, instruction decoupling, and hierarchical processing. The onboard integrated processing unit first receives the globally optimal control parameter set from step three; this data packet is accompanied by a high-precision generated timestamp T. generation .

[0184] The pre-execution comprehensive verification includes a triple-check mechanism of timeliness verification, security verification, and feasibility verification. These triple checks are performed sequentially in the order of timeliness verification, security verification, and feasibility verification. If a previous level of verification fails, subsequent verifications are immediately suspended and the process enters an exception handling branch to avoid performing unnecessary subsequent checks on parameter sets already determined to be invalid. The specific check contents of the triple verification under a typical preferred implementation are given below. It should be understood that the listed check contents are preferred embodiments and not an exhaustive limitation of the triple verification concept. Those skilled in the art can expand or adjust the specific check items of each verification according to specific application needs without departing from the inventive concept.

[0185] In the preferred embodiment, timeliness verification is performed by comparing the current time T. now With T generation The system determines whether the globally optimal control parameter set is within a preset valid time window, typically 50 milliseconds. Parameter sets exceeding this window are deemed expired and discarded. In addition to the typical checks mentioned above, timeliness verification can selectively include extended checks such as data link delay compensation verification, timing sequence consistency verification, and parameter set version number increment verification to address timing anomalies in special communication environments.

[0186] In the preferred embodiment, the safety verification uses a safety rule engine to check whether there are dangerous parameter combinations in the global optimal control parameter set where the beam angle directly hits the vehicle's cockpit and the brightness exceeds the glare threshold. In addition to the typical checks mentioned above, the safety verification can selectively include extended checks such as glare risk checks for other road users, direct beam risk checks for the UAV's own mission camera, and potential interference checks for strong light on road reflective markings or traffic signal recognition, to cover a wider range of safety scenarios.

[0187] In the preferred embodiment, the feasibility verification determines whether the total power required by the globally optimal control parameter set exceeds the current maximum output capacity based on the instantaneous battery voltage state. In addition to the typical checks mentioned above, the feasibility verification can selectively include extended checks such as single-module maximum current limit verification, heat dissipation margin verification, communication bandwidth occupancy verification, and actuator mechanical limit verification to address various hardware-level feasibility constraints.

[0188] If any check in the triple check mechanism fails, the system discards the current global optimal control parameter set, maintains the effective control parameter set of the previous cycle to ensure the continuity of supplemental lighting, or triggers the drone to switch to the preset safe flight mode to ensure flight safety. Figure 5 The node that triggers the emergency safety policy corresponds to this processing branch. This branch points to the end state of step four and sends the feedback signal back to step three to trigger the optimization solution of the next cycle. The multi-channel synchronous execution phase of the current cycle will no longer be executed.

[0189] After successful verification, the system enters the instruction decoupling and hierarchical stage, decoupling the unified global optimal control parameter set into three independent sub-instructions according to the protocol. The three sub-instructions are sent to different receiving ends and carry different control information: the UAV-side control sub-instruction sent to the UAV airborne lighting controller contains specific parameters such as brightness, color temperature, beam divergence angle, horizontal deflection angle, and vertical deflection angle of each sub-module in the programmable multispectral lighting matrix; the vehicle-side control sub-instruction sent to the vehicle body domain controller is transmitted through the vehicle bus and contains vehicle lighting area control information such as the vertex coordinates of the polygon of the lighting area of ​​the adaptive digital headlights or front auxiliary lighting lights and the average brightness of the area; the apron-side control sub-instruction sent to the apron lighting controller contains start / stop and mode instructions such as the start / stop status and working mode identifier of the apron lighting subsystem.

[0190] Step 4.2: Multi-channel synchronous execution.

[0191] Figure 5 The second stage involves generating a synchronization trigger signal and distributing it in parallel to the three execution channels. The onboard unit first generates a hardware synchronization pulse signal (or aligns it using precise software timestamps) to ensure that the start time deviation of the three execution channels is less than 1 millisecond. This ensures that the collaborative execution of the supplementary lighting action enables the UAV-side intelligent lighting module, the vehicle lighting subsystem, and the helipad lighting subsystem to start synchronously within a preset time synchronization threshold. The hardware synchronization pulse signal is suitable for the vehicle's internal CAN FD bus and wired connection to the helipad (a typical scenario is a UAV docked in an onboard hangar), providing sub-millisecond-level start synchronization accuracy. The software timestamp alignment is suitable for UAVs connected via a wireless link (a typical scenario is a UAV that has taken off and moved away from the vehicle), achieving millisecond-level start synchronization by compensating for wireless transmission delays through a precise time protocol.

[0192] In addition to synchronization at startup, the system maintains timing synchronization of the three channels throughout the entire illumination operation. The controllers of each execution channel use a unified time reference provided by the hardware timestamp synchronization clock component as their internal clock source, and execute dynamic changes in parameters such as brightness, color temperature, and angle sequentially according to the timing sequence specified in the global optimal control parameter set, so that the three light fields remain in a coordinated state throughout the entire execution cycle.

[0193] The precision requirements for drone lighting execution channels are the highest. The onboard lighting controller precisely controls the output of each LED or laser sub-module via PWM (Pulse Width Modulation) or constant current drive chips, based on the brightness (I) and color temperature (T) values ​​in the drone's control sub-commands. The typical operating frequency for PWM drive is 2kHz to 20kHz. This frequency range is far higher than the flicker frequency threshold perceptible to the human eye (avoiding visible flicker artifacts) but not so high as to cause excessive heat generation in the drive chip. The digital adjustment precision of the PWM duty cycle is typically configured as 10-bit (corresponding to 1024 brightness levels) or 12-bit (corresponding to 4096 brightness levels) to meet the visual requirements of smooth, stepless dimming and the dynamic range demands of visual tasks.

[0194] In implementations using constant current driver chips, a typical approach is analog dimming or a hybrid dimming method combining analog and PWM dimming. Analog dimming adjusts the LED luminous flux by directly changing the amplitude of the drive current, with a typical digital adjustment precision of 10 bits or more. Considering the color temperature drift of LEDs under different drive currents, the hybrid dimming method balances the dynamic range of luminous flux and color temperature stability by using analog dimming in the high current range and switching to PWM dimming in the low current range. Color temperature adjustment is achieved by simultaneously controlling the PWM duty cycle of both cool white and warm white LEDs; the ratio of their duty cycles determines the final output color temperature. Since the actual output color temperature and duty cycle ratio are not strictly linear due to factors such as the change of LED luminous efficacy with current and the nonlinearity of color coordinate interpolation, a typical implementation uses a pre-calibrated lookup table to establish the correspondence between the dual LED duty cycle ratio and the actual output color temperature. That is, during the development phase, the actual output color temperature under different duty cycle combinations is measured point by point to establish a correspondence table and stored in the non-volatile memory of the airborne lighting controller. During operation, after receiving the target color temperature command, the airborne lighting controller uses the lookup table and linear interpolation to back-calculate the dual LED duty cycle configuration required to achieve the target color temperature.

[0195] If the instruction includes angle adjustment, the controller drives a micro stepper motor or MEMS micromirror to deflect the beam to the target angle within 10 to 20 milliseconds; if equipped with an electrically tunable metamaterial surface, its optical properties are dynamically changed by altering the applied voltage, achieving inertial-free switching of the beam shape (such as from a circle to an ellipse).

[0196] The vehicle lighting control channel is completed through the vehicle network, which is further subdivided into two sub-steps: sending control messages and executing area lighting control.

[0197] In the step of sending control messages, the on-board processing unit encodes the vehicle-side control sub-instructions into standard vehicle network messages (such as CAN FD or on-board Ethernet messages) and sends them to the body domain controller through the vehicle's internal on-board network.

[0198] In the execution of the area lighting control sub-step, the vehicle domain controller receives and parses the network message, extracting the pixel-level dimming instructions for the adaptive digital headlights (including the vertex coordinates of the pixel areas to be darkened and the corresponding pixel brightness ratios). It then drives the matrix pixel light sources of the adaptive digital headlights to perform area-specific dimming according to the instructions. For example, in a following shooting scenario, a typical instruction received by the vehicle domain controller is "maintain high beam illumination, but reduce pixel brightness to below 10% within a cone-shaped spatial area with the drone as the vertex." Based on this, the vehicle domain controller maintains the overall high beam output of the headlights while precisely locating the corresponding pixel units within the cone-shaped spatial area and reducing the brightness of these pixels to the target ratio. This creates an anti-glare shadow area in front of the vehicle, preventing glare or image overexposure artifacts caused by the strong headlight light to the drone's lens.

[0199] The helipad lighting execution channel is triggered when the UAV enters the landing sequence. The vehicle-mounted unit sends helipad control sub-commands to the helipad controller via a low-latency wireless link (such as a private ultra-wideband link) or a wired connection. Upon receiving the command, the helipad controller drives the surrounding LED light strips to switch to a high color rendering index, flicker-free soft lighting mode or a breathing gradient mode, according to the working mode indicator specified in the command. The soft lighting mode provides a stable and friendly lighting environment for the UAV's visual positioning algorithm, reduces image noise, and ensures lighting consistency so that the UAV camera can accurately identify helipad markings and boundaries. The breathing gradient mode provides directional guidance and distance perception assistance for the UAV's precise landing with periodic brightness gradient lighting characteristics, enabling the UAV to determine the landing timing and position by recognizing the rhythm of the light brightness changes during the final approach phase.

[0200] Step 4.3: Execution status monitoring and feedback.

[0201] Figure 5 The third phase involves parallel acquisition of actual status feedback from each subsystem, status comparison and confirmation, and checking whether execution errors are within tolerance limits. The system collects data from the UAV lighting controller, body controller, and landing pad controller in parallel and compares it with the target values ​​of the issued commands.

[0202] Specifically, the collaborative execution of supplementary lighting in step four also includes an execution status monitoring sub-step. This sub-step collects execution status feedback from three terminals in real time and in parallel: It reads back at least one of the following from the airborne lighting controller: drive current, temperature, and beam pointing angle of each sub-module of the programmable multispectral lighting matrix. The beam pointing angle can be obtained through a Hall sensor, magnetic encoder, or similar angle sensor. It also reads back the status confirmation message of the adaptive digital headlights from the vehicle body domain controller, confirming that the pixel-level dimming command of the headlights has been executed according to the specified pixel area and brightness ratio. Finally, it reads back the lighting mode confirmation signal from the helipad lighting controller, confirming that the helipad light-emitting devices have switched to the target working mode as instructed. The system compares the actual status feedback values ​​collected from these three terminals in parallel with the issued command target values, forming a hardware-level closed-loop feedback. If the comparison result is within the tolerance range (typically, brightness error less than 5% and angle error less than 0.5 degrees), it sends a "successful execution" signal to the upper layer, completing step four and entering the "light field has been reshaped as instructed" completion state. The system then proceeds to step five, waiting for the task camera to acquire monitoring images. If the comparison result exceeds the preset deviation range, the system first records the corresponding fault code in the fault log of the vehicle-mounted integrated processing unit. At the same time, it reports the abnormality flag, fault type, location, deviation value, and other specific information as an abnormality feedback signal to the vehicle-mounted integrated processing unit. The vehicle-mounted integrated processing unit then triggers the corresponding abnormality handling action based on this information. Typical types of execution abnormalities include failure of a group of LED drivers, jamming of the stepper motor or MEMS micromirror of the beam pointing mechanism, failure of the vehicle's adaptive digital headlights to respond to commands, and communication interruption of the helipad lighting controller.

[0203] The exception handling actions are divided into three response levels according to the severity of the exception.

[0204] The first-level response corresponds to minor abnormal situations. When a slight deviation is detected in a single sub-module (such as a brightness error between 5% and 15% or a core temperature that reaches the warning threshold but not the critical threshold), the system will only shut down the single faulty module or reduce its output power to a preset safety level (typically 50% of the rated power). At the same time, the remaining normal sub-modules will work together to compensate and maintain the continuity of the overall lighting effect.

[0205] The second-level response corresponds to moderate abnormal situations. When multiple sub-modules are detected to fail simultaneously or the comparison deviation exceeds the tolerance within multiple consecutive sampling cycles (such as brightness error exceeding 15% for more than three sampling cycles), the system enables the backup lighting unit to replace the faulty module and reports the abnormality to the vehicle-mounted integrated processing unit to trigger the adjustment of optimization parameters in the next cycle.

[0206] The third-level response corresponds to severe anomalies. When the core temperature of the lighting module exceeds the critical temperature threshold (typically 85 degrees Celsius), the beam pointing angle deviation exceeds the safety boundary (typically greater than 2 degrees), or the battery voltage drops sharply below the warning level, the system immediately sends a visual assistance degradation warning to the flight control system, triggering the UAV to switch to a preset safe flight mode to prioritize flight safety. This tiered response strategy enables the system to take appropriate actions when facing anomalies of varying severity, avoiding performance loss due to over-response while ensuring system safety margins in severe situations. After anomaly handling is completed, step four of the current cycle ends, feedback is sent to the onboard integrated processing unit, and step three is triggered to re-execute the optimization solution in the next control cycle.

[0207] Step 5: Real-time visual feedback and optimization.

[0208] After step four is completed, the UAV's mission camera performs image acquisition according to the exposure period required for the current mission, typically ranging from 30 milliseconds to 100 milliseconds. Once acquisition is complete, the monitoring image is transmitted back to the vehicle-mounted integrated processing unit via the vehicle-to-airport cooperative communication network, serving as the input trigger signal for step five. (Refer to...) Figure 1 The "Real-time Visual Feedback and Optimization" node at the end of the overall flowchart shown below, and the diamond-shaped decision node "Image Quality Assessment" below it, execute the vision-based closed-loop feedback process in step five.

[0209] Step five receives feedback information after the coordinated illumination action and iteratively re-optimizes the globally optimal control parameter set based on this feedback information, forming a complete closed-loop control. The feedback information includes monitoring images transmitted by the UAV's mission camera after completing synchronous illumination. The vehicle-mounted integrated processing unit runs a lightweight image quality evaluator to determine mission-oriented indicators based on the scene indicated by the mission scene label. It should be noted that the hardware-level feedback generated in step four (execution status monitoring sub-step) and the visual feedback generated by the mission camera in step five follow two independent feedback paths: the hardware-level feedback closes within step four and triggers exception handling during the execution phase; the visual feedback is processed by the image quality evaluator in step five and triggers parameter re-optimization in step three. Together, they constitute the system's hierarchical feedback mechanism.

[0210] Task-oriented metrics are image quality assessment quantities directly related to the visual information required for the current task scenario, including at least one of image pixel-level statistics and target-level morphological quantities. In tracking tasks, task-oriented metrics include contrast and signal-to-noise ratio (whether the target is clearly and noiselessly recorded in the image), while in detection tasks, they include target boundary sharpness (whether the target can be reliably identified by subsequent target detection algorithms). The system evaluates the task-oriented metrics of the monitored image; when the task-oriented metrics fall below a preset image quality threshold, the system executes a parameter re-optimization process.

[0211] The parameter re-optimization process operates sequentially between steps two and three: the evaluation results are first fed back to step two, where relevant parameters in the lighting demand parameter set, such as the required illuminance Lreq, are fine-tuned based on the new task-oriented index to generate an updated lighting demand parameter set; subsequently, the updated lighting demand parameter set triggers the multi-objective optimization solution process in step three for secondary optimization. Figure 1 The arrow pointing from the "Trigger Parameter Re-optimization" node to step three corresponds to this sequential relationship—the arrow essentially represents the complete path of "updating the lighting requirement parameter set and then re-entering step three," with the fine-tuning action in step two occurring as an intermediate step in the parameter update within this path. This closed loop ensures that the system can adapt to dynamically changing environmental conditions, such as sudden vehicle turns, rapid target movement, and sudden changes in external lighting.

[0212] This invention achieves independent controllability across five dimensions—brightness, color temperature, beam shape, beam angle, and spectral band—on a single hardware platform through a combined design of a programmable multispectral illumination matrix and a beam control mechanism. Combined with a task gain adaptive mathematical model based on Retinex theory, the system can dynamically shape a light field precisely matched to the current task based on real-time changing parameters such as target distance, ambient light intensity, weather conditions, and task type. Taking search and portrait tracking tasks as examples, the color temperature strategies employ 6000K cool white light (enhancing penetration) and 3500K warm white light (optimizing skin tone reproduction), respectively; the spectral strategies utilize the 650nm red light band (enhancing biological target recognition) and a full-spectrum wide-band light source (ensuring natural color reproduction), respectively; and the beam strategies employ wide-beam floodlight (covering a wide area) and narrow-beam focused light (focusing on the subject), respectively. This highly task-oriented illumination strategy significantly improves the adaptability of supplementary lighting to specific tasks, solving the problem of single illumination strategies in existing technologies.

[0213] This invention, through a centralized decision-making mechanism combining a vehicle-airport collaborative communication network and an onboard integrated processing unit, achieves for the first time unified scheduling of lighting resources from three parties: drone lighting, vehicle lights, and helipad lighting. Taking a tracking shot as an example, in traditional solutions, drone fill lights and vehicle headlights operate independently, often resulting in glare artifacts caused by strong headlight beams reflecting off the drone's lens, or obscuring the subject due to conflicting illumination directions between the drone's fill lights and the headlights. This invention, through adaptive digital headlights performing pixel-level darkening operations within a cone-shaped projection area to create an anti-glare shadow zone, actively avoids the drone's camera direction when the vehicle's lighting output is directed, forming a complementary lighting synergy with the drone's onboard light source. In precision landing scenarios, the helipad lighting subsystem switches from standby mode to a breathing gradient lighting mode, providing a stable and clear guidance signal for the drone's visual positioning. This three-terminal collaborative lighting network is significantly superior to the independent lighting modes of existing technologies, solving the problem of lack of collaborative capabilities.

[0214] This invention incorporates four sub-objectives—lighting performance error, total power consumption cost, glare risk value, and thermal load estimation—into a multi-objective optimization function, and uses a dynamic weighting coefficient mechanism to automatically switch the optimization focus under different mission scenarios. In long-endurance cruise and inspection scenarios, the weights of power consumption and thermal load are increased, and the system automatically reduces the light source output power, reduces the heat dissipation burden, and extends the UAV's endurance. In extreme search scenarios, performance has extremely high weights, and the system is willing to increase instantaneous power consumption to ensure the lighting effect for critical tasks, while rigid constraints ensure that the total power consumption does not exceed the current output capacity of the battery and the thermal load does not exceed the maximum allowable temperature of the module. This dynamic optimization mechanism based on real-time system status is significantly superior to existing solutions that use fixed power or simple brightness feedback, solving the challenges of energy efficiency and thermal management.

[0215] This invention, through a real-time visual feedback mechanism in step five, establishes a complete closed loop between supplementary lighting control and subsequent visual recognition tasks for the first time. The system not only generates initial supplementary lighting parameters based on task requirements during the feedforward phase, but also continuously evaluates whether the supplementary lighting effect truly meets the needs of the visual task during execution. By real-time evaluation of task-oriented indicators such as contrast and signal-to-noise ratio for the tracking task and target boundary sharpness for the detection task, the system can determine whether the supplementary lighting has truly achieved the goals of "focused shooting" and "clear vision," rather than merely achieving "sufficient brightness." When the evaluation result does not meet the preset threshold, the system triggers parameter re-optimization and reissues instructions, forming a truly visual task-oriented closed-loop optimization. This mechanism is significantly superior to existing solutions where supplementary lighting control is disconnected from the visual task, solving the problem of insufficient visual task orientation.

[0216] This invention innovatively introduces an electrically tunable metamaterial optical surface as an optional method for beam shaping and deflection in the beam control mechanism. Compared with the mechanical beam control mechanisms widely used in the prior art, such as micro-stepper motor-driven lens groups or MEMS micromirror arrays, it brings four significant technical effects. First, the response speed is greatly improved: traditional mechanical beam control mechanisms are limited by the mechanical inertia of stepper motors or MEMS micromirrors, with a typical response time of 20 to 50 milliseconds; the electrically tunable metamaterial optical surface can dynamically adjust the optical properties by changing the applied voltage through a program, without the need for any mechanical parts to move. The response time is no greater than the preset response time threshold (typically within 20 milliseconds, and within 5 milliseconds in preferred embodiments), which is 3 to 10 times faster than the mechanical solution, enabling the system to adjust the beam according to scene changes at the millisecond level. Secondly, there is a significant improvement in beam shape shaping capability: mechanical mechanisms typically only allow for linear adjustment of beam direction and focal length, with limited flexibility in switching beam shapes (such as circular, elliptical, and stripe shapes). Electrically tunable metamaterial optical surfaces, by altering the electrical response modes of their microstructural units, can achieve arbitrary beam shape shaping and inertia-free switching; for example, they can seamlessly switch from circular to elliptical beams to match the geometric features of different targets. Thirdly, there is an improvement in long-term reliability: mechanical mechanisms suffer from long-term reliability risks such as bearing wear, motor aging, and fatigue failure of MEMS micromirrors, with a typical reliable lifespan of hundreds of thousands of cycles. Electrically tunable metamaterial optical surfaces have no moving mechanical parts; their reliability mainly depends on the electrical stability of the semiconductor substrate, with a typical reliable lifespan exceeding tens of millions of cycles, meeting the requirements for long-term reliable operation in automotive environments. Fourthly, there is an improvement in system integration: the thickness of electrically tunable metamaterial optical surfaces is typically only on the order of millimeters, compared to the centimeter-sized dimensions of traditional mechanical mechanisms. This allows for the integration of more beam control units within the same UAV body space, further enhancing the overall spatial controllability of the programmable multispectral illumination matrix. The combined effect of the above four factors makes the electrically tunable metamaterial optical surface a key innovation of this invention in beam control, providing a fundamental hardware support for improving the overall performance of the system.

[0217] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Equivalent substitutions, combinations, or parameter adjustments made by those skilled in the art based on the technical concept of the present invention, such as reasonable changes to the specific number of sub-modules of the programmable multispectral illumination matrix, the specific endpoint values ​​of the color temperature working range, the specific values ​​of various thresholds, and the specific implementation form of the optimization algorithm, should all fall within the scope of protection of the present invention.

[0218] The contents not described in detail in this specification are existing technologies known to those skilled in the art.

Claims

1. A vehicle-mounted unmanned aerial vehicle (UAV) cooperative adaptive supplementary lighting system, characterized in that, It includes an on-board integrated processing unit, a drone-side intelligent lighting module, a vehicle lighting subsystem, a helipad lighting subsystem, and a vehicle-airport collaborative communication network; The vehicle-mounted integrated processing unit is deployed inside the vehicle and is configured to process the perception data and generate a set of globally optimal control parameters for the coordinated supplementary lighting of the UAV-based intelligent lighting module, the vehicle lighting subsystem, and the helipad lighting subsystem. The perception data includes environmental perception data collected and transmitted back by the UAV-based intelligent lighting module and vehicle-side data obtained by the vehicle-mounted integrated processing unit from the vehicle itself. The intelligent lighting module for the UAV is integrated on the UAV body and is configured to output a light field according to the global optimal control parameter set. The vehicle lighting subsystem is deployed on the vehicle and configured to output a light field according to the globally optimal control parameter set; the helipad lighting subsystem is located on the helipad and configured to output a light field according to the globally optimal control parameter set. The vehicle-airport collaborative communication network has time synchronization capabilities and is used to build a cross-platform synchronous control and data interaction link between the vehicle-mounted integrated processing unit, the UAV-mounted intelligent lighting module, the vehicle lighting subsystem, and the apron lighting subsystem.

2. The vehicle-mounted UAV cooperative adaptive supplementary lighting system according to claim 1, characterized in that, The UAV-side intelligent lighting module includes an environmental perception kit, a programmable multispectral lighting matrix, and an onboard lighting controller. The environmental perception kit is used to collect the environmental perception data; The programmable multispectral illumination matrix is ​​used to perform light field output. The spectral output of the programmable multispectral illumination matrix covers a preset color temperature working range and includes at least one specific band for enhancing the sensitivity of specific target recognition. The airborne lighting controller is used to receive the global optimal control parameter set, drive the programmable multispectral lighting matrix, and manage the local thermal sensor and heat dissipation unit.

3. The vehicle-mounted UAV cooperative adaptive supplementary lighting system according to claim 1, characterized in that, The vehicle-mounted integrated processing unit is further configured to perform multi-source fusion of the perceived data and unify it into the vehicle coordinate system to generate a fused data frame, and to perform collaborative decision-making and global scheduling on the vehicle lighting subsystem, the UAV-based intelligent lighting module, and the helipad lighting subsystem based on the vehicle coordinate system and the fused data frame.

4. The vehicle-mounted UAV cooperative adaptive supplementary lighting system according to claim 2, characterized in that, The environmental perception kit includes at least two of the following: a high dynamic range ambient light sensor, a time-of-flight depth camera, a real-time dynamic positioning and inertial measurement fusion unit, and a meteorological sensor. The high dynamic range ambient light sensor is used to collect ambient light illuminance and spectral information; The time-of-flight depth camera is used to acquire target distance and scene 3D point cloud data; The real-time dynamic positioning and inertial measurement fusion unit is used to provide the altitude, three-dimensional attitude and positioning information of the UAV; The meteorological sensor is used to collect at least one of the meteorological parameters, including ambient temperature and humidity, fog concentration, and meteorological visibility.

5. The vehicle-mounted UAV cooperative adaptive supplementary lighting system according to claim 1, characterized in that, The vehicle-mounted integrated processing unit also includes a vehicle bus interface, a task instruction parsing unit, and a device status monitoring unit. The vehicle bus interface is used to acquire vehicle driving status data, including vehicle speed, steering wheel angle, current headlight mode, and vehicle GPS location. The task instruction parsing unit is used to receive task instructions from the vehicle or ground control terminal and parse and output a task scenario label indicating the current task scenario type. The task scenario type includes at least one of following task, search task, and precision landing. The following task includes vehicle-following shooting and portrait following shooting. The search task includes wide-area search and detection task for specific targets. The device status monitoring unit is used to monitor the remaining battery power of the UAV and the core temperature of the programmable multispectral illumination matrix.

6. The vehicle-mounted UAV cooperative adaptive supplementary lighting system according to claim 2, characterized in that, The programmable multispectral illumination matrix includes a beam control mechanism, which is used to dynamically adjust the shape and spatial illumination angle of the output beam of the programmable multispectral illumination matrix. The beam control mechanism is selected from one or more combinations of electrotunable metamaterial optical surfaces, microelectromechanical systems digital micromirror arrays, and micro stepper motor driven lens groups.

7. The vehicle-mounted UAV cooperative adaptive supplementary lighting system according to claim 6, characterized in that, The beam control mechanism employs an electrically tunable metamaterial optical surface. The airborne illumination controller is configured to change the optical properties of the electrically tunable metamaterial optical surface through a program, thereby achieving precise beam shaping and deflection without mechanical movement. The response time for beam shaping and deflection is not greater than a preset response time threshold.

8. The vehicle-mounted UAV cooperative adaptive supplementary lighting system according to claim 1, characterized in that, The vehicle-airport collaborative communication network includes a hardware timestamp synchronization clock component. The hardware timestamp synchronization clock component adopts a precise time protocol and compensates for wireless transmission delay through underlying protocol message exchange, so that the time deviation between the vehicle, the drone and the apron where the apron lighting subsystem is located is not greater than a preset time synchronization threshold.

9. The vehicle-mounted UAV cooperative adaptive supplementary lighting system according to claim 3, characterized in that, The vehicle-mounted integrated processing unit uses the point cloud data of the fused data frame as a three-dimensional scene approximation to simulate the light propagation of the light sources carried by the vehicle lighting subsystem, the UAV intelligent lighting module, and the helipad lighting subsystem. It calculates the illuminance distribution of each light source in the core lighting area and the glare risk value at the driver's field of vision and the UAV's self-camera. In areas where the glare risk value is greater than a preset glare safety threshold, it performs beam avoidance.

10. The vehicle-mounted UAV cooperative adaptive supplementary lighting system according to claim 9, characterized in that, The vehicle lighting subsystem includes at least one of an adaptive digital headlight with matrix pixel-level dimming capability and a front auxiliary lighting lamp disposed at the front of the vehicle; the adaptive digital headlight is configured to perform a pixel-level darkening operation within a cone-shaped projection area extending from the drone body as the vertex along the direction from the adaptive digital headlight to the road surface in front of the adaptive digital headlight, forming an anti-glare shadow area, wherein the pixel-level darkening operation reduces the brightness of the corresponding pixels within the cone-shaped projection area to below a preset proportional threshold.

11. The vehicle-mounted UAV cooperative adaptive supplementary lighting system according to claim 1, characterized in that, The apron lighting subsystem includes an apron lighting controller and light-emitting devices located around the apron where the apron lighting subsystem is located. The apron lighting controller is configured to switch the working mode of the light-emitting devices when the UAV enters the landing sequence. The working modes include a soft light illumination mode for visual positioning and a breathing gradient illumination mode for visual guidance.

12. The vehicle-mounted UAV cooperative adaptive supplementary lighting system according to claim 2, characterized in that, The vehicle-mounted integrated processing unit is also configured to start the system and perform a system self-test in response to preset trigger conditions. The trigger conditions include the UAV entering night flight mode or the vehicle or ground control terminal issuing a night operation command. When the system self-test detects any abnormality in the UAV's intelligent lighting module, the environmental perception kit, or the vehicle-airport cooperative communication network, it reports an error message and stops the system startup, while simultaneously triggering the UAV to switch to a preset safe flight mode.

13. A method for coordinated adaptive supplementary lighting control of a vehicle-mounted unmanned aerial vehicle (UAV), applied to the vehicle-mounted integrated processing unit in the coordinated adaptive supplementary lighting system of a vehicle-mounted UAV as described in any one of claims 1 to 12, characterized in that, Includes the following steps: Step 1: Perform multi-dimensional acquisition and spatiotemporal alignment of perception data, acquire heterogeneous data streams containing at least one of ambient light information, target depth information, UAV pose information, and vehicle status information, and generate fused data frames; Step 2: Based on the fused data frame, perform scene feature extraction and task scene recognition, determine the core lighting area, calculate the beam characteristic requirements for the core lighting area, and output a lighting requirement parameter set including task scene label, task priority, the core lighting area, required illuminance, and beam characteristic set, wherein the task priority is determined based on the task scene label and the real-time event status during task execution. Step 3: Perform a multi-objective optimization solution process based on the lighting demand parameter set to generate a globally optimal control parameter set; Step 4: Perform security verification on the global optimal control parameter set, and distribute the global optimal control parameter set to the UAV terminal intelligent lighting module, the vehicle lighting subsystem, and the apron lighting subsystem through the vehicle-airport cooperative communication network, so that the three can coordinate to perform supplementary lighting actions. Step 5: Receive the feedback information returned after the collaborative execution of the supplementary lighting action, and iteratively re-optimize the global optimal control parameter set based on the feedback information to form closed-loop control.

14. The vehicle-mounted unmanned aerial vehicle (UAV) cooperative adaptive supplementary lighting control method according to claim 13, characterized in that, The spatiotemporal alignment in step one includes unifying the heterogeneous data streams to the same time reference based on hardware timestamps, and unifying the spatial coordinate data to the vehicle body coordinate system.

15. The vehicle-mounted unmanned aerial vehicle (UAV) cooperative adaptive supplementary lighting control method according to claim 13, characterized in that, The scene feature extraction includes four types of feature vectors: output spatial relationship features, ambient light semantic features, meteorological perception features, and dynamic context features. The spatial relationship features include the relative distance, azimuth angle, and altitude difference between the UAV and the target, as well as the spatial proximity used to determine whether the UAV is in the glare-sensitive area in front of the vehicle. The ambient light semantic features include a base illuminance level, a light contrast ratio used to determine whether high dynamic range supplemental lighting is needed, and spectral features used to identify external light source interference. The meteorological sensing features include meteorological influencing factors, which are derived from meteorological parameters collected by the meteorological sensor or from image haze detection algorithms. The dynamic context features include the task scenario label, the relative motion state between the drone and the vehicle, and the system constraint state including the remaining battery power and the temperature of the lighting module.

16. The vehicle-mounted unmanned aerial vehicle (UAV) cooperative adaptive supplementary lighting control method according to claim 15, characterized in that, The task scene recognition includes a scene classifier decision sub-step, which takes at least one of the spatial relationship features, the ambient light semantic features, the meteorological perception features and the dynamic context features as input, and outputs refined scene labels based on a preset lightweight hybrid decision model. The refined scene label is obtained by combining the task scene label and the conditional parameters in the feature vector. The refined scene label includes at least one of the following: low-speed close-range tracking scene, medium-to-long-range search and inspection scene, and return and precise landing scene. The low-speed close-range tracking scene corresponds to the feature combination of the relative distance being less than a preset close-range threshold, the relative motion state indicating low speed in the same direction, and the task scene label being the tracking task. The mid-to-long-range search and inspection scenario corresponds to a relative distance greater than the preset short distance threshold and a task scenario label that is a feature combination of the search task. The return-to-home and precision landing scenarios correspond to the UAV's positioning information being close to the helipad and the mission scenario label being a combination of features of the precision landing. The refined scene label serves as the final task scene label determined from the lighting requirement parameter set. The output of the scene classifier decision sub-step covers the original task scene label output by the task instruction parsing unit. Subsequent calculations of the core lighting area, the required illuminance, the beam characteristic set, the dynamic adjustment of the task priority, and the scenario-based evaluation of the feedback information are all performed based on the refined scene label as the task scene label.

17. The vehicle-mounted unmanned aerial vehicle (UAV) cooperative adaptive supplementary lighting control method according to claim 15, characterized in that, The core lighting area in the lighting requirement parameter set is expressed in the form of a three-dimensional spatial boundary in the vehicle coordinate system based on the target detection box, the attitude information of the UAV, and the relative distance, azimuth angle, and height difference in the spatial relationship features. The beam characteristic set includes at least one of the following: beam divergence angle, target color temperature, special band activation indicator, and polarized lighting activation indicator. The beam divergence angle is determined according to the spatial size of the core lighting area and the relative distance of the UAV to the core lighting area in the spatial relationship features. In the tracking task where the relative distance is in a preset close range, the beam divergence angle is set to a preset narrow beam range. In the search task, the beam divergence angle is set to a preset wide beam range.

18. The vehicle-mounted unmanned aerial vehicle (UAV) cooperative adaptive supplementary lighting control method according to claim 13, characterized in that, The task priority in the lighting demand parameter set is determined by a preset task scenario priority mapping table and the real-time event status during task execution; the preset task scenario priority mapping table assigns an initial level to the task priority based on the task scenario label; The real-time event state is an event state triggered by changes in the perceived data or changes in the target detection result during task execution, including at least one of target detection, target loss, and environmental anomaly; the task priority is dynamically adjusted according to the type of the real-time event state, wherein when a target detection event occurs during the execution of the search task, the task priority is dynamically adjusted to the highest level and serves as one of the triggering conditions for switching the weight coefficients of the multi-target cost function.

19. The vehicle-mounted unmanned aerial vehicle (UAV) cooperative adaptive supplementary lighting control method according to claim 13, characterized in that, The multi-objective optimization solution process includes a light propagation simulation sub-step. In each optimization iteration, the light propagation simulation sub-step performs light propagation simulation on the light sources carried by the UAV-side intelligent lighting module, the vehicle lighting subsystem, and the helipad lighting subsystem based on the current decision variables, and outputs the illuminance distribution of each light source in the core lighting area and the glare risk value in the driver's field of vision and the UAV's self-camera as intermediate quantities. The light propagation simulation uses the point cloud data of the fused data frame as a three-dimensional scene approximation.

20. The vehicle-mounted unmanned aerial vehicle (UAV) cooperative adaptive supplementary lighting control method according to claim 13, characterized in that, The decision variables in the multi-objective optimization solution process include the UAV terminal vector, the vehicle terminal vector, and the helipad terminal vector; The UAV terminal vector includes the brightness, color temperature, beam divergence angle, horizontal deflection angle, and vertical deflection angle of each sub-module in the programmable multispectral illumination matrix; The vehicle terminal vector includes the vertex coordinates of the polygon of the illumination area of ​​the adaptive digital headlight or the front auxiliary lighting lamp and the average brightness of the area; The helipad terminal vector includes the on / off status and operating mode identifier of the helipad lighting subsystem.

21. The vehicle-mounted unmanned aerial vehicle (UAV) cooperative adaptive supplementary lighting control method according to claim 13, characterized in that, The coordinated execution of supplemental lighting includes synchronizing the UAV-side intelligent lighting module, the vehicle lighting subsystem, and the helipad lighting subsystem within a preset time synchronization threshold.

22. The vehicle-mounted unmanned aerial vehicle (UAV) cooperative adaptive supplementary lighting control method according to claim 15, characterized in that, The beam characteristic requirement calculation adopts a task gain adaptive mathematical model based on Retinex theory. The required illuminance of the core lighting area is pre-calculated using feedforward. The required illuminance is calculated as the sum of four physical quantities: ambient base illuminance, base safety supplementary lighting gain, task-specific gain coefficient and task spectral sensitivity factor, and meteorological compensation gain and meteorological influence factor. The ambient base illuminance is derived from the base illuminance level in the ambient light semantic features. The task-specific gain coefficient and the task spectral sensitivity factor are both determined based on the task scene label. When the task scene label indicates the search task, the task-specific gain coefficient is greater than the baseline gain coefficient, and the task spectral sensitivity factor is greater than the preset sensitivity threshold at a preset specific wavelength.

23. The vehicle-mounted unmanned aerial vehicle (UAV) cooperative adaptive supplementary lighting control method according to claim 22, characterized in that, The beam characteristic requirement calculation also includes a color temperature dynamic decision sub-step. This sub-step determines the color temperature parameters according to a three-layer decision logic: a default adaptive layer, a weather-triggered layer, and a task-triggered layer. The default adaptive layer adaptively selects the color temperature according to the required illuminance. When the required illuminance is in the high brightness range, the color temperature shifts towards a cool color temperature, and when the required illuminance is in the low brightness range, the color temperature shifts towards a warm color temperature. When the meteorological visibility collected by the meteorological sensor is lower than the preset visibility threshold, the meteorological triggering layer forces the color temperature to be no less than the high color temperature threshold in order to enhance the light beam penetration. When the task scene label indicates that the portrait is being filmed, the task triggering layer sets the color temperature to a preset low to medium color temperature range to optimize skin tone reproduction. The decision priority of the weather triggering layer and the task triggering layer is higher than that of the default adaptive layer.

24. The vehicle-mounted unmanned aerial vehicle (UAV) cooperative adaptive supplementary lighting control method according to claim 19, characterized in that, The multi-objective optimization solution process uses a multi-objective cost function for evaluation, and uses an optimization algorithm to solve for the global optimal control parameter set that minimizes the multi-objective cost function under the premise of satisfying rigid constraints; The multi-objective cost function uses the illuminance distribution and glare risk value output from the light propagation simulation sub-step as intermediate quantities to calculate four sub-costs: lighting performance error, total power consumption cost, glare risk sub-cost, and heat load prediction value. The four sub-costs are then weighted and summed to obtain the value of the multi-objective cost function. The rigid constraints include total power consumption not exceeding the maximum power limit, predicted heat load not exceeding the maximum temperature limit, glare risk value not exceeding the glare safety threshold, and each decision variable in the global optimal control parameter set being within the physically adjustable range of the corresponding actuator; wherein the maximum power limit is dynamically determined based on the remaining battery power in the dynamic context features, and the maximum temperature limit is dynamically determined based on the lighting module temperature in the dynamic context features.

25. The vehicle-mounted unmanned aerial vehicle (UAV) cooperative adaptive supplementary lighting control method according to claim 24, characterized in that, The weighting coefficients of each sub-item in the multi-objective cost function are dynamically adjusted according to the task scenario label and the task priority. Different weight combinations are configured for different task scenarios to achieve adaptive optimization objective balance for the task scenario.

26. The vehicle-mounted unmanned aerial vehicle (UAV) cooperative adaptive supplementary lighting control method according to claim 13, characterized in that, The security verification includes a triple check mechanism of timeliness verification, security verification, and feasibility verification. The timeliness verification determines whether the global optimal control parameter set is within a preset valid time window by comparing the current time with the generation timestamp of the global optimal control parameter set. The safety verification uses a safety rule engine to check whether there are dangerous parameter combinations in the global optimal control parameter set where the beam angle directly hits the vehicle's cockpit and the brightness exceeds the glare threshold. The feasibility verification is based on the instantaneous battery voltage state to determine whether the total power required by the globally optimal control parameter set exceeds the current maximum output capacity; If any check in the triple check mechanism fails, the globally optimal control parameter set is discarded and the effective control parameter set of the previous cycle is maintained, or the UAV is triggered to switch to a preset safe flight mode.

27. The vehicle-mounted unmanned aerial vehicle (UAV) cooperative adaptive supplementary lighting control method according to claim 13, characterized in that, The feedback information includes the monitoring images transmitted back by the drone-side mission camera after completing synchronous illumination; The task-oriented indicators are determined based on the scene indicated by the task scene label. The task-oriented indicators are image quality assessment quantities that are directly related to the visual information required by the current task scene, including at least one of image pixel-level statistics and target-level morphological quantities. The task-oriented indicators include contrast and signal-to-noise ratio in the tracking task and target boundary sharpness in the detection task. The task orientation index of the monitored image is evaluated. When the task orientation index is lower than the preset image quality threshold, the evaluation result is fed back to step two to fine-tune the required illumination and trigger the multi-objective optimization solution process for secondary optimization.

28. The vehicle-mounted unmanned aerial vehicle (UAV) cooperative adaptive supplementary lighting control method according to claim 13, characterized in that, The coordinated execution of supplementary lighting in step four also includes an execution status monitoring sub-step. The execution status monitoring sub-step collects at least one of the driving current, temperature, and beam pointing angle of each sub-module of the programmable multispectral illumination matrix in real time, and compares it with the target control signal to form a hardware-level closed-loop feedback. When the comparison result exceeds the preset deviation range, at least one abnormal handling action is executed, such as shutting down the faulty module, switching to the backup illumination unit, reducing the output power, or sending a visual assistance degradation warning to the flight control system.

29. The vehicle-mounted unmanned aerial vehicle (UAV) cooperative adaptive supplementary lighting control method according to claim 13, characterized in that, When the proportion of specular reflection in the scene exceeds a preset reflection threshold, a control command to enable the ring polarization lighting mode is output from the lighting requirement parameter set.

Images