Air flight intelligent control system and method of polyhedral LED display screen

Abstract

The invention discloses an air flight intelligent control system and method for a polyhedral LED display screen. The polyhedral LED display screen comprises a multi-rotor unmanned aerial vehicle and a full-color display screen composed of multiple peripheral LEDs. According to the invention, audience position detection is realized by adopting the RGB-D camera and the millimeter-wave radar, and a multi-level sensing fusion system is constructed by combining the ambient light sensor and the IMU attitude sensor, so that the accuracy and reliability of data acquisition are improved; the adaptive display control architecture is constructed by adopting a Transform attention mechanism, so that the display effect is improved, different application scene requirements are met, and meanwhile, the power consumption is reduced; according to the invention, functions of automatically adjusting the display angle according to the audience position, adjusting the LED brightness according to the ambient brightness and carrying out content anti-shake processing according to the flight attitude are realized; according to the invention, the problem of instruction response delay caused by channel congestion in a multi-device concurrent transmission scene is solved, and the endurance time is prolonged.

Description

Technical Field

[0001] This invention relates to the field of aerial canopy technology, specifically to an intelligent control system and method for a multi-faceted LED display screen in flight. Background Technology

[0002] With the rapid development of drone technology, unmanned aerial vehicles (UAVs) are widely used in aerial photography, agriculture, mini selfies, express delivery, disaster relief, and other fields. Among them, multi-rotor UAVs, due to their small size, maneuverability, and ease of operation, play a vital role in public service sectors, such as urban traffic monitoring and fire patrols. However, existing multi-rotor UAVs still have several shortcomings in certain application scenarios. For example, they are vulnerable to surveillance or attack when performing missions over long distances, and are susceptible to attacks by humans or simple tools when flying at low altitudes. Furthermore, existing multi-rotor UAVs are prone to overheating and rapid discharge when operating in hot weather, leading to excessive power consumption of the remote controller and display, which affects flight safety.

[0003] To address these issues, researchers have begun exploring novel multi-rotor drone designs. For example, some studies have proposed fan-screen-based advertising drones, using LED displays mounted on the drone for mobile advertising. Other research focuses on improving the rain resistance of multi-rotor drones to ensure they can fly normally in rainy weather. However, these improvements still haven't completely solved the problems of multi-rotor drone applications in complex environments. For instance, motors and rotor blades are easily affected by rain, and jet parachute technology doesn't offer ideal wind resistance and rain protection in heavy downpours.

[0004] Therefore, there is an urgent need to develop a novel multi-rotor UAV display system that can ensure flight stability and safety in complex environments while achieving high-quality display effects. Simultaneously, the system should possess intelligent and adaptive capabilities to adapt to different application scenarios, extend flight time, and reduce power consumption. This requires integrating advanced sensors and intelligent control algorithms into multi-rotor UAVs and developing efficient display terminals to achieve real-time fusion processing of multi-source heterogeneous data, thereby improving display stability and display effects, extending flight time, and reducing power consumption.

[0005] Several invention patents have been developed to address the issues of real-time fusion processing of multi-source heterogeneous data, display stability, and battery life.

[0006] For example, CN113232827A discloses a tail-seat VTOL UAV for flight performances and a multi-screen interconnected UAV formation. This patent achieves multi-screen interconnection of the UAV formation through signal receivers and protocol interoperability devices installed in the nose, and alleviates the UAV load problem by using an embedded screen solution. However, this patent still has instability issues in the installation method of the display screen.

[0007] CN115027672A discloses an aerial display system based on a drone array. This patent uses highly mobile drones as pixels instead of traditional LED display pixels, allowing graphics to be displayed in various situations and breaking free from the regional limitations of traditional LED screen displays. However, this patent still has shortcomings in terms of the drone array's anti-disturbance capability.

[0008] The existing technology has the following drawbacks: 1. Existing multi-rotor drones lack flight stability and safety in complex environments, making them vulnerable to surveillance or attacks. Furthermore, operation in hot weather can easily lead to excessively high temperatures and rapid discharge, causing the remote controller and display to consume power too quickly, thus affecting flight safety.

[0009] 2. Existing multi-rotor UAV display systems are insufficient in the real-time fusion processing of multi-source heterogeneous data, making it difficult to achieve high-quality display effects and meet the intelligent needs of different application scenarios.

[0010] 3. Existing multi-rotor drones are insufficient in terms of anti-disturbance capabilities, making it difficult to maintain a stable flight attitude in complex environments, which affects the stability of display effects.

[0011] 4. Existing multi-rotor drones have significant problems in terms of flight time and power consumption, especially when the display is bright, the power consumption is too high, making it difficult to meet the needs of long-term continuous flight.

[0012] 5. Existing multi-rotor drones have insufficient stability in the installation method of the display screen, which affects the stability and reliability of the display screen during flight. Summary of the Invention

[0013] Existing technologies suffer from problems such as poor flight stability of multi-rotor UAVs in complex environments, insufficient real-time fusion capability of multi-source heterogeneous data, short flight time, and high power consumption. Therefore, to address these issues, this invention provides an intelligent control system and method for the aerial flight of a multi-faceted LED display screen.

[0014] A method for intelligent aerial flight control of a multi-faceted LED display screen, comprising a multi-rotor UAV and a full-color display screen composed of peripheral multi-faceted LEDs, wherein the intelligent aerial flight control method includes: RGB-D cameras, millimeter-wave radar, ambient light sensors, and IMU attitude sensors are deployed on a multi-faceted LED display screen to acquire real-time data on audience position, viewing angle, ambient brightness, and flight attitude. The acquired data undergoes fusion processing and time synchronization processing; Assign weights to visual information, radar information, ambient light, and IMU data; Based on the Transformer attention mechanism, a three-tiered adaptive display control architecture of perception layer, decision layer, and execution layer is adopted to realize content layered display and mode switching; By combining an improved Kalman filter algorithm with multi-source data, the optimal display angle is output. Preferably, the RGB-D cameras are located at 45° above the front of the device and 135° below the rear, covering a field of view of 120°; the millimeter-wave radars are located at the front, left, and right of the device, covering a field of view of 120°; the IMU attitude sensor has the main sensor located at the center of gravity and the auxiliary sensor offset by 15mm.

[0015] Preferably, the data fusion and time synchronization processing includes: The Kalman filter algorithm is used to fuse multi-sensor data, and the state vector includes the audience's three-dimensional position, line of sight angle, ambient light intensity, and drone attitude angle; Sensor data acquisition employs a time synchronization mechanism, using GPS time synchronization to align data timestamps. The weighting of visual information, radar information, ambient light, and IMU data includes: Visual information weight 0.3-0.5, with higher values ​​used for core line-of-sight localization when illumination is sufficient (≥1000 lux); radar information weight 0.2-0.4, with higher values ​​used when illumination is insufficient (<1000 lux) / occlusion; ambient light weight 0.1-0.2; IMU data weight 0.1; all weights are dynamically summed to 1.0.

[0016] Preferably, the perception layer generates an environmental state feature vector, including: light intensity L, audience density ρ, audience viewing angle θ, drone attitude angle φ, and audience distance d; Decision layer: Dynamically allocate modal weights through cross-attention modules to achieve unified representation and feature fusion of heterogeneous data; Based on audience distance d and attention concentration Establish a priority evaluation model for illumination intensity L:

[0017] in ,default , =0.5, =0.2, For optimal display brightness, The maximum ambient light level is set to 10000 lux. Adjusting the display mode by setting a threshold: Night mode: Activated when ambient light L < 200 lux, reducing brightness and increasing contrast; High-density mode: Switches when audience density ρ > 5 people / m², font size is enlarged, and the proportion of core information is increased; Execution layer: Content is displayed in layers: the font size of core information is adaptive and stable; the transparency of auxiliary information is dynamically adjusted; the background decoration is automatically matched according to the ambient color temperature; and the content switching uses a fade-in / fade-out effect to achieve a smooth transition.

[0018] Preferably, an improved Kalman filter is used to fuse multi-sensor data to establish a display effect evaluation function:

[0019] Where: θ is the angle between the line of sight and the normal vector of the display screen, and B is the actual display brightness. For optimal brightness, d represents the audience distance. For the optimal distance, ,default =0.5, =0.3, =0.2; using the gradient descent algorithm, with a step size η=0.01 and a convergence accuracy ε=0.001, the Q value is maximized.

[0020] Preferably, the intelligent flight control method also includes image stabilization compensation: Collect 3-axis angular velocity, 3-axis acceleration, and 3-axis magnetometer data from the IMU attitude sensor, and wind speed, air pressure, and temperature characteristic parameters from the environmental sensing module; A 3-layer LSTM fully connected architecture is adopted. The number of neurons in the input layer matches the 12-dimensional feature parameters. The number of neurons in the hidden layer are 128, 64, and 32 respectively. The activation function is ReLU. The output layer contains 3-dimensional attitude parameters of pitch angle, roll angle, and yaw angle. Based on the pose change predicted by LSTM, a bilinear interpolation algorithm is used for pixel remapping, and the compensation model is as follows:

[0021] In the formula, P is the compensated target pixel value, and w is the interpolation weight, satisfying... P represents the value of four adjacent pixels in the original image.

[0022] Preferably, the intelligent flight control method also includes: The angle adjustment is controlled as a primary control, and the anti-shake compensation is controlled as a secondary control. When the angle adjustment is performed, an active adjustment flag is output to the LSTM model, the model temporarily stores the current prediction result, and the anti-shake compensation is paused. If the angle error is ≤0.08° and lasts for 50ms, the angle adjustment is completed, and the anti-shake compensation is restarted after a 50ms delay.

[0023] Preferably, the intelligent flight control method also includes: By using predictive algorithms to preload display content, response time is optimized and overall power consumption is controlled. A three-level Markov prediction model is used: The first layer is scene state prediction: a state transition matrix is ​​established based on environmental parameters such as illumination L, audience density ρ, and time t. Predict scene type; The second layer is content demand prediction: establishing content call probability matrices for different scenarios. The probability of text content being retrieved is 0.6-0.8, image content 0.3-0.5, and video content 0.1-0.3. The third layer is resource allocation prediction: an improved LRU-K algorithm (K=3) is used to dynamically allocate 32MB of cache space, and a sliding window learning mechanism (window size 100 samples) is combined to optimize cache parameters; Cache hit rate formula:

[0024] Where: H is the cache hit rate, P is the scene prediction accuracy, C is the content prediction accuracy, and R is the resource utilization rate. =0.5、 =0.3、 =0.2.

[0025] Preferably, the intelligent flight control method also includes: Predict power consumption requirements under different scenarios and dynamically adjust LED drive current and display area activation ratio; Predictive load balancing algorithm formula:

[0026] in: The total power consumption of the system is =0.7 is the basic power consumption coefficient, and β=0.3 is the environmental adaptability coefficient. Based on power consumption, This formula is used to predict power consumption in different scenarios and dynamically adjust the LED driving current and the activation ratio of the display area.

[0027] An intelligent control system for aerial flight of a multi-faceted LED display screen is provided to implement the above-mentioned intelligent control method for aerial flight. The intelligent control system for aerial flight includes a sensor acquisition module, a data fusion and synchronization module, a weight allocation module, an adaptive display control algorithm module, an angle adaptive adjustment module, and an anti-shake compensation module. Sensor acquisition module: Real-time acquisition of audience position, line of sight angle, ambient brightness, and flight attitude data based on multiple types of sensors; Data fusion and synchronization module: performs fusion processing and time synchronization processing on the acquired data; Weight allocation module: Assigns weights to visual information, radar information, ambient light, and IMU data; Adaptive display control algorithm module: Based on the Transformer attention mechanism, it adopts a three-level adaptive display control architecture of perception layer - decision layer - execution layer to realize content layered display and mode switching; Angle adaptive adjustment module: Combines improved Kalman filter algorithm with multi-source data to output the optimal display angle; Image stabilization compensation module: Based on the pose change predicted by the LSTM neural network, pixel remapping is performed using a bilinear interpolation algorithm. Beneficial effects

[0028] 1. This invention uses an RGB-D camera and millimeter-wave radar to detect the audience's position. Combined with an ambient light sensor and an IMU attitude sensor, a multi-level perception fusion system is constructed, which greatly improves the accuracy and reliability of data acquisition and overcomes the shortcomings of insufficient multi-modal data acquisition accuracy in the prior art.

[0029] 2. This invention employs a Transformer attention mechanism to construct an adaptive display control architecture, achieving content layering and mode switching through a dynamic threshold and priority model. This intelligent display control strategy significantly improves display quality, adapts to different application scenarios, and reduces power consumption.

[0030] 3. This invention enables the display angle to be adjusted autonomously based on the viewer's position, the LED brightness to be adjusted based on the ambient light, and the content anti-shake processing to be performed based on the flight attitude, thus solving the problem in the prior art that it is difficult to comprehensively consider multiple factors such as ambient light, user interaction behavior, and flight status.

[0031] 4. This invention adopts an edge computing architecture, which significantly reduces response time and improves the real-time performance of the system through efficient hardware design and optimized algorithm flow, while effectively controlling power consumption and meeting the requirements for long battery life.

[0032] 5. This invention achieves intelligent layered display of content by constructing a three-level adaptive display control architecture, including flexible switching of core information, auxiliary information and background decoration, effectively solving the problem of instruction response delay caused by channel congestion in multi-device concurrent transmission scenarios.

[0033] 6. This invention proposes a three-level Markov predictive caching mechanism, which achieves intelligent resource scheduling through scene state prediction, content demand prediction, and resource allocation prediction. This predictive load balancing strategy effectively extends battery life and solves the problem of excessive power consumption caused by high-brightness displays in traditional solutions. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the polyhedral LED display screen of the present invention; Figure 2 This is a schematic diagram of the intelligent flight control method for a polyhedral LED display screen according to the present invention.

[0035] Figure 3 This is a schematic diagram of a three-level adaptive display control architecture.

[0036] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Example

[0038] like Figure 1 , 2 As shown, a multi-faceted LED display screen for intelligent control of aerial flight consists of a multi-rotor drone and an outer multi-faceted LED display screen, which is a full-color display screen capable of playing videos.

[0039] The multi-rotor drone's flight platform adopts a quadcopter / six-rotor configuration (selected according to payload requirements), and the fuselage is made of high-strength carbon fiber composite material, balancing structural strength and lightweight requirements. The power system is equipped with a brushless DC motor paired with a high-efficiency propeller, providing an endurance of ≥40 minutes (under standard load). It supports flight modes such as hovering, path planning, and autonomous obstacle avoidance. Through the linkage communication between the flight control system and the display control module, it ensures real-time matching of flight position accuracy (±0.5m) and displayed angle.

[0040] The multi-faceted LED full-color display terminal adopts a multi-faceted three-dimensional structure that fits the layout of the drone body. The size of each display screen is designed according to the specifications of the drone body, and the whole adopts a modular splicing design, which is convenient for disassembly and maintenance.

[0041] Power supply system: provides power to the display screen driver and control circuit, with a voltage regulation accuracy of ±0.1V to ensure voltage stability.

[0042] Mechanical connection: A customized metal bracket is rigidly connected to the drone body. The bracket has built-in rubber shock-absorbing pads (5mm thick, damping coefficient 0.3) to reduce the structural damage and display interference caused by flight vibration. The vibration attenuation rate is ≥80%.

[0043] Electrical connection: The waterproof aviation plug (IP67 protection rating) integrates the power cord and data cable to realize bidirectional communication between the flight control system and the main control of the display screen. The data transmission rate is ≥100Mbps, which meets the safety requirements for use in complex outdoor environments.

[0044] The polyhedral LED display can be any polyhedron. The following description uses a cubic LED display, which has six faces. The hexahedral LED display can have LED displays on all six faces, or it can be open at the top and not have an LED display, with LED displays only on the bottom and four sides. In this invention, it is preferred that the top not have an LED display.

[0045] The system of this invention consists of three rigidly integrated parts: a multi-rotor UAV flight platform, a multi-faceted LED full-color display terminal, and an intelligent sensing and control module. Through mechanical vibration damping and electrical linkage communication, it constructs a full-link collaborative control system of "flight stability - accurate sensing - adaptive display - controllable power consumption". The core interaction logic of each module is as follows: The flight vehicle outputs real-time attitude, position, and endurance status data to the control module; The sensing module outputs fused data such as audience position / status and ambient brightness to the control module; The control module outputs path fine-tuning instructions to the flight vehicle and brightness, angle, and content adjustment instructions to the display terminal. The control module uses a load balancing algorithm to coordinate the power consumption distribution between the flight vehicle and the display terminal.

[0046] This invention constructs a multi-sensor fusion intelligent sensing system, which provides accurate input for adaptive display control through spatiotemporal synchronization and feature fusion of heterogeneous sensor data, and clarifies the functional boundaries and data flow of each sensor.

[0047] Sensor configuration and layout: RGB-D camera: Combining binocular stereo vision and structured light technology, it acquires audience depth information with a detection accuracy of ±0.3m. It is positioned diagonally at 45° above the front and 135° below the rear of the camera body, covering a field of view of 120°. It is used for audience position detection and gaze direction recognition, and works with millimeter-wave radar to achieve 360° coverage without blind spots.

[0048] Millimeter-wave radar: 360° all-round audience position detection, ranging from 0.5 to 50m, located at the front of the aircraft, 60° to the left, and 60° to the right, in a three-point layout, used for audience distribution density calculation and positional blind spot filling when visual occlusion occurs.

[0049] Ambient light sensor: It adopts a photodiode + ADC converter to output 16-bit digital brightness data, with a detection accuracy of ±5 lux. It is distributed in four quadrants, and each LED display surface is equipped with an independent sensor. It has the functions of adaptive adjustment of display brightness and night mode triggering.

[0050] The IMU attitude sensor detects the instantaneous rotational angular velocity ωt and acceleration of the UAV with a measurement accuracy of ±0.01°. It features a dual-redundant central arrangement, with the main sensor located at the center of gravity and the auxiliary sensor offset by 15mm.

[0051] Data fusion and synchronization strategy: Fusion algorithm: The Kalman filter algorithm is used to fuse multi-sensor data. The state vector includes the audience's three-dimensional position, line of sight angle, and UAV attitude angle, so as to achieve high-precision estimation of audience position, ambient brightness, and flight attitude, with a positioning error ≤0.3m.

[0052] Time synchronization: Sensor data acquisition adopts a time synchronization mechanism, which uses GPS time synchronization to align data timestamps with a synchronization accuracy of <1ms, avoiding detection and control errors caused by data time difference.

[0053] Weighting: Application scenarios of modal weights in the cross-attention module: Visual information weight 0.3-0.5, with higher values ​​taken when the lighting is sufficient (≥1000 lux) for core gaze localization; Radar information weight 0.2-0.4, with higher values ​​taken when the lighting is insufficient (<1000 lux) / occlusion; Ambient light weight 0.1-0.2; IMU data weight 0.1; All weights are dynamically summed to 1.0.

[0054] Adaptive display control core algorithm: This invention is based on the Transformer attention mechanism and constructs a three-level adaptive display control architecture of "perception layer - decision layer - execution layer".

[0055] Perception layer: Multimodal feature extraction A multi-head attention mechanism (8 attention heads, each with 64 dimensions) is used to extract features from multimodal data from RGB-D cameras, millimeter-wave radar, ambient light sensors, and IMUs, generating an environmental state feature vector with a dimension of 512, which includes core parameters: illumination intensity L, audience density ρ, audience line of sight angle θ, drone attitude angle φ, and audience distance d.

[0056] Decision-making level: Dynamic threshold and priority model Multimodal data fusion: By dynamically allocating modal weights through a cross-attention module, a unified representation and feature fusion of heterogeneous data can be achieved.

[0057] Based on audience distance d and attention concentration Establish a priority evaluation model for illumination intensity L:

[0058] in: ,default , =0.5, =0.2, For optimal display brightness, The maximum ambient brightness is set to 10000 lux; the larger the P value, the higher the priority, guiding the content to be displayed in layers.

[0059] Dynamic mode switching: Setting a threshold to trigger display mode adjustment: 1. Night Mode: Activated when ambient light L < 200 lux, reducing brightness and increasing contrast; 2. High-density mode: Switch when audience density ρ > 5 people / m², font size is increased by 1.5 times, core information ratio is increased to 70%.

[0060] Execution layer: Content and angle adaptive control Content is displayed in layers: the font size of core information adapts to 20-60 pixels, the transparency of auxiliary information is dynamically adjusted from 0.3 to 0.8, and the background decoration is automatically matched according to the ambient color temperature of 2700K-6500K; content switching uses a 200ms fade-in and fade-out effect to achieve a smooth transition.

[0061] Display angle closed-loop control: Constructing a visual servo closed-loop system and clarifying the participation path of millimeter-wave radar data: When the RGB-D camera is unobstructed, the audience position is calculated by fusing visual data (weight 0.6) as the primary data source and radar data (weight 0.4) as the secondary data source. When visual obstruction (such as strong light or obstacles) is detected, the system switches to radar data as the primary data source (weight 0.6) and visual data as the secondary data source (weight 0.4). By combining an improved Kalman filtering algorithm with multi-source data, the optimal display angle is output; a display effect evaluation function is established.

[0062] Where: θ is the angle between the line of sight and the normal vector of the display screen ( The closer to 1, the more upright the line of sight; B represents the actual display brightness. For optimal brightness, d represents the audience distance. The optimal distance is 10m. + =1 (default) =0.5, =0.3, =0.2); by maximizing the Q value through the gradient descent algorithm (step size η=0.01, convergence accuracy ε=0.001), the angle can be precisely adjusted.

[0063] To resolve the contradiction between ±0.1° natural attitude fluctuation and ±0.08° display angle adjustment accuracy during multi-rotor UAV flight (including hovering), this embodiment adopts an integrated design of "hardware adaptation upgrade + algorithm-hardware collaborative compensation" to ensure that the accuracy target can be achieved. The specific hardware adaptation and collaborative logic is as follows: 1. Hardware adaptation and upgrade solution: (1) High-precision power control unit: A brushless DC motor with an incremental photoelectric encoder (encoder resolution ≥ 1024 lines) is selected and paired with a high-precision ESC (control accuracy ±0.01°) to achieve fine adjustment of motor speed and provide a power basis for attitude stability; In response to hovering attitude fluctuations, the motor PID control parameters (proportional coefficient Kp=5.2, integral coefficient Ki=0.8, derivative coefficient Kd=1.5) are optimized to compress the hovering attitude fluctuation range from ±0.1° to within ±0.05°, thereby reducing the impact of attitude disturbances on the display angle from the power end.

[0064] (2) High-frequency sampling sensor array: The sampling frequency of the IMU attitude sensor is increased to 200Hz (originally 100Hz), and a high-precision MEMS IMU (angle measurement accuracy ±0.005°) is selected. With dual redundancy arrangement (the main sensor is located at the center of gravity, and the auxiliary sensor is offset by 15mm), the high-frequency and high-precision acquisition of attitude data is realized, ensuring that the attitude fluctuation signal can be accurately captured; the sampling frequency of the RGB-D camera and the millimeter-wave radar is synchronously increased to 100Hz to ensure the real-time data of the audience's position and line of sight angle, and to provide accurate input for angle adjustment.

[0065] (3) Precision angle adjustment actuator: A miniature electric gimbal (adjustment accuracy ±0.001°) is added between the display screen and the UAV body as an independent angle adjustment actuator with fast response capability (response time ≤10ms). The gimbal adopts a direct drive motor design with no transmission backlash, which can realize independent and high-precision fine adjustment of the display angle to compensate for the accuracy deviation caused by the attitude fluctuation of the UAV body.

[0066] 2. Algorithm-hardware collaborative compensation logic: Based on high-frequency acquired IMU attitude data, a collaborative control system of "body attitude fluctuation prediction + active display angle compensation" is achieved by improving the Kalman filter algorithm (adding body attitude fluctuation to the state vector): First, the body attitude fluctuation of the UAV is calculated in real time using IMU data. Then, the impact of this fluctuation on the display angle is predicted. Finally, a miniature motorized gimbal actively outputs a reverse compensation angle to counteract the impact of the body fluctuation. For example, when a pitch fluctuation of +0.05° is detected, the gimbal immediately outputs a pitch compensation angle of -0.05° to ensure the stability of the actual display angle.

[0067] Meanwhile, the closed-loop control of angle adjustment is deeply integrated with the gimbal control: the optimal display angle command output by the visual servo closed-loop system is first corrected by the body attitude fluctuation compensation module before being sent to the gimbal for execution, forming a full-link closed loop of "audience state perception - optimal angle calculation - body fluctuation compensation - gimbal precise execution", ensuring that the final display angle adjustment accuracy is stable within ±0.08°.

[0068] Eye tracking and local enhancement: Detect the position of the viewer's pupils to calculate the focal point of the gaze, and enhance the brightness and contrast of the focal area; set a speed threshold (viewer movement speed < 2m / s), and turn off local enhancement when the threshold is exceeded to prioritize angular stability.

[0069] Flight attitude prediction and content stabilization To address the display jitter problem caused by drone flight attitude disturbances, an intelligent anti-shake scheme combining LSTM neural network prediction and bilinear interpolation compensation is proposed. At the same time, the collaborative logic between the anti-shake scheme and the display angle adjustment is clarified to avoid mutual interference.

[0070] LSTM pose prediction model Input features: Collect 3-axis angular velocity, 3-axis acceleration, and 3-axis magnetometer data from the IMU attitude sensor, as well as wind speed, air pressure, and temperature from the environmental perception module, for a total of 12 feature parameters, with a sampling frequency of 100Hz.

[0071] Network structure: A 3-layer LSTM fully connected architecture is adopted. The number of neurons in the input layer matches the 12-dimensional features. The number of neurons in the hidden layer are 128, 64 and 32 respectively. The activation function is ReLU. The output layer contains 3-dimensional attitude parameters of pitch angle, roll angle and yaw angle.

[0072] Prediction performance: Based on 10,000 hours of labeled data from multiple scenarios (including extreme scenarios such as strong wind, low temperature, and occlusion), it can predict the attitude within 200ms in the future with a prediction error of ≤0.5° and a prediction accuracy of ≥95%.

[0073] Adaptive pixel compensation algorithm Based on the pose change predicted by LSTM, a bilinear interpolation algorithm is used for pixel remapping, and the compensation model is as follows:

[0074] In the formula, P is the compensated target pixel value, and w is the interpolation weight (satisfying... The value of P is calculated from the spatial relationship between the target pixel and the original pixel, where P is the value of the four adjacent pixels of the original image, achieving pixel-level precise image stabilization without image stretching or distortion.

[0075] Coordination logic with display angle adjustment Priority definition: Display angle adjustment is a first-level control (high priority), and anti-shake compensation is a second-level control (low priority). When performing angle adjustment, an "active adjustment flag" is output to the LSTM model. The model temporarily stores the current prediction result and pauses anti-shake compensation to avoid judging the active adjustment as a posture disturbance.

[0076] Timing coordination: After the angle adjustment is completed (convergence accuracy reaches ε=0.001), the anti-shake compensation is restarted after a 50ms delay to ensure that the adjusted attitude is stable before jitter correction is performed; the timing scheduler of the control module realizes the seamless connection between the two strategies.

[0077] Edge computing and real-time performance assurance An edge AI chip integrating 4 TOPS computing power (CPU+NPU heterogeneous architecture) localizes the entire process of data preprocessing, model inference, and compensation calculation. Through task scheduling optimization, it achieves end-to-end processing latency of ≤30ms (5ms for preprocessing + 18ms for inference + 7ms for compensation), meeting the real-time requirements of human vision.

[0078] Image stabilization performance evaluation The 16% power consumption increment of the image stabilization function refers to the increase in power consumption between the high-precision operating mode (1.45W) and the basic operating mode (1.25W) (0.2W ÷ 1.25W × 100% = 16%), not the increase in power consumption relative to the total system power consumption.

[0079] Define the stability coefficient S to quantize and display stability:

[0080] In the formula, Let θ represent the attitude angle deviation in the i-th frame (in rad), θ be the maximum deviation threshold (0.1 rad), and n be the number of frames counted (n=1000 frames in the test). In strong wind conditions (wind speed ≥8m / s), the stability coefficient S of this solution is ≥0.92, which is 40% higher than that of traditional image stabilization solutions, with a power consumption increase of only 8%.

[0081] Intelligent caching and power optimization strategies Three-tier Markov prediction caching mechanism Scene state prediction: Based on environmental parameters (lighting L, audience density ρ, time t), a state transition matrix is ​​established to predict 8 scene types (indoor / outdoor / nighttime / crowd gathering, etc.).

[0082] Content demand prediction: Establish a content call probability matrix for different scenarios, with text content having a call probability of 0.6-0.8, image content 0.3-0.5, and video content 0.1-0.3.

[0083] Resource allocation prediction: An improved LRU-K algorithm (K=3) is used to dynamically allocate 32MB of cache space, and the cache parameters are optimized by combining a sliding window learning mechanism (window size 100 samples).

[0084] Cache hit rate formula:

[0085] Where: H is the cache hit rate, P is the scene prediction accuracy, C is the content prediction accuracy, and R is the resource utilization rate. =0.5、 =0.3、 =0.2 (the sum of weights is 1); the goal is to achieve P>85%, C>75%, R>90%, and ultimately H>82%.

[0086] Predictive load balancing and power control Predictive load balancing algorithm formula:

[0087] in: The total system power consumption is represented by α=0.7, which is the basic power consumption coefficient, and β=0.3 is the environmental adaptability coefficient. Based on base power consumption (8W). The power consumption is adaptive to the environment (positively correlated with illumination and display brightness); this formula is used to predict power consumption requirements in different scenarios and dynamically adjust the LED driving current and the activation ratio of the display area.

[0088] It adopts a low-power GPU chip (computing power requirement <15TOPS) and combines a dynamic computing power allocation strategy—when there is no audience movement or the line of sight is stable, the computing power output is reduced to 5TOPS, further reducing ineffective power consumption.

[0089] Coordinated optimization of battery life and power consumption increment The standard load mode features all core system functions enabled, including anti-shake, adaptive display adjustment, basic sensing, and data fusion.

[0090] The power consumption increment of the image stabilization function refers to the increment of the "high-precision operation mode of the image stabilization algorithm (such as prediction accuracy improved to 97% in strong wind conditions)" relative to the "basic operation mode of image stabilization under standard load," rather than the increment relative to the total power consumption of the entire system. Specifically: the basic operation power consumption of image stabilization is 1.25W, the high-precision operation power consumption is 1.45W, and the incremental power consumption is 0.2W. The corresponding power consumption increment ratio of the image stabilization function itself is 8% (0.2W ÷ 2.5W × 100%). This increment is compensated by reducing the power consumption of the sensing module through a sensor time-division activation strategy (50Hz in hovering state, 100Hz in flight state, and 200Hz in emergency obstacle avoidance state). Ultimately, this ensures that the overall total power consumption of the system under standard load remains stable in the range of 12W-12.5W, and the battery life is shortened by ≤3% (meeting the design requirement of ≥38.8 minutes of battery life under standard load, which meets the design requirement of ≥40 minutes).

[0091] Performance indicators The multi-faceted LED display screen of this invention includes, but is not limited to, three-dimensional structures such as cubes, disks, spheres, and cylinders. All technical solutions that adopt the flight vehicle design, multi-sensor fusion perception method, adaptive display control algorithm, and attitude anti-shake strategy of this invention are within the protection scope of this invention. Example 2

[0092] The multi-faceted LED display screen has an automatic return-to-home charging function when the battery is low. This function is achieved through the collaboration of the Battery Management System (BMS) and the Flight Control System, featuring multi-level power monitoring and intelligent decision-making mechanisms. Battery monitoring: Built-in high-precision voltage and current sensors collect parameters such as battery voltage, current, and remaining power in real time, with a sampling frequency ≥10Hz and a remaining power error ≤2%. Three-level power threshold settings: warning threshold, return-to-home threshold, and forced landing threshold.

[0093] Return-to-home decision: When the battery level drops to the return-to-home threshold, the BMS sends a return-to-home signal to the flight control system. The flight control system automatically initiates the return-to-home procedure, prioritizing the location of the nearest charging base station (obtained through pre-stored base station GPS coordinates or real-time communication), planning the optimal return-to-home path, and avoiding obstacles (in conjunction with obstacle avoidance sensor data).

[0094] Automatic charging: The charging base station adopts a fixed-point docking design, equipped with wireless charging or contact charging interfaces. The base station has a built-in positioning guidance module (dual guidance of infrared positioning and visual positioning). When the drone returns to a position 1m above the base station, it achieves precise landing by visually recognizing the base station's positioning marker, with a landing error of ≤5cm. After landing, the charging interface docking is automatically triggered. The charging system adopts intelligent fast charging technology, with a charging efficiency of ≥90% and a full charge time of ≤60 minutes. During charging, the BMS monitors the battery status in real time and has overcharge, over-temperature, and overcurrent protection functions. Autonomous obstacle avoidance and environmental perception capabilities through multi-sensor fusion:

[0095] The intelligent agent incorporates multiple types of sensors and uses data fusion algorithms to achieve autonomous obstacle avoidance and real-time environmental parameter acquisition. The specific configuration and implementation logic are as follows: Obstacle avoidance sensors: A total of 6 sets of lidar sensors (detection range 0.1-10m, accuracy ≤2cm) and 4 sets of ultrasonic sensors (detection range 0.05-5m) are deployed around the fuselage and on its top and bottom, forming a 360° detection range with no blind spots. When an obstacle is detected ahead (distance ≤1.5m), the sensors transmit data to the flight control system in real time. The flight control system automatically adjusts the flight trajectory through path planning algorithms to achieve obstacle avoidance flight or emergency hovering.

[0096] Ambient light sensor: A digital light sensor (measurement range 0-100000 lux, accuracy ±5%) is installed on the top of the display screen to collect ambient light intensity data in real time and transmit it to the main control system of the display screen to automatically adjust the brightness of the display screen (adjustment range 300-1500 cd / ㎡) to ensure a balance between display effect and energy saving requirements under different lighting conditions.

[0097] Temperature and humidity sensor: A high-precision temperature and humidity module (temperature measurement range -20℃~60℃, accuracy ±0.5℃; humidity measurement range 0-100%RH, accuracy ±3%RH) is selected and installed inside the UAV fuselage to monitor the temperature and humidity of the operating environment of the fuselage's electronic components in real time. The data is transmitted wirelessly to the ground control terminal. When the temperature and humidity exceed the safety threshold (temperature >50℃ or <0℃, humidity >85%RH), an alarm signal is triggered to remind ground personnel to deal with it in time.

[0098] Data fusion: Data from each sensor is transmitted to the central processing unit via the CAN bus. The Kalman filter algorithm is used to reduce noise in the data to ensure its accuracy. The fused environmental data and obstacle avoidance data work together to support the autonomous decision-making of the intelligent agent.

[0099] Remote real-time control function

[0100] It adopts a control architecture of "wireless communication + cloud collaboration", supports remote control from both computer and mobile terminals, and realizes real-time adjustment of flight attitude and displayed content: Communication Module: Equipped with an industrial-grade 4G / 5G module and a WiFi module (supporting the 802.11ac protocol), achieving dual-mode communication redundancy and ensuring communication stability in complex environments. The 4G / 5G module supports nationwide data transmission with a transmission rate ≥10Mbps; the WiFi module supports short-range (≤100m) high-definition data transmission for local control in environments without network coverage. Communication latency ≤200ms ensures real-time response to control commands.

[0101] Flight attitude control: The ground control terminal (using dedicated control software on the computer and an APP on the mobile phone) can display the drone's flight parameters (altitude, speed, position, battery level, etc.) in real time, supporting manual control (joystick control) and switching between autonomous modes (path planning, hovering, etc.). Control commands are transmitted in encrypted form to prevent tampering or hijacking, ensuring flight safety.

[0102] Display content control: Supports remote uploading and switching of display content via ground control terminal, and supports real-time live broadcast push (achieved through 4G / 5G module). The control software / APP has built-in content editing function, which can edit text, pictures and videos, set the playback order and playback duration, and the display screen content update delay is ≤1 second, realizing flexible control of the displayed content.

[0103] To ensure personal safety during on-site operations, a comprehensive protective structure is designed for the drone's wings. The core technical details are as follows: Protective Structure Design: A dual protective structure of a ring-shaped protective ring and a mesh protective net is adopted. The protective ring is arranged circumferentially around each rotor and is made of high-strength engineering plastic (ABS + fiberglass) with a thickness of ≥5mm. The diameter of the protective ring is 10-15cm larger than the rotor diameter to ensure that the rotor does not directly contact external objects during rotation. A metal mesh protective net (mesh size ≤2cm×2cm) is installed inside the protective ring to further prevent small objects from entering the rotor area, while not affecting the airflow.

[0104] Buffer energy absorption design: The connection between the protective ring and the fuselage adopts an elastic buffer structure with a built-in spring shock absorber. When a collision occurs, it can absorb the impact force (withstanding a maximum impact force of 5kg), reducing damage to the fuselage and rotor, while preventing rotor fragments from flying due to the collision.

[0105] Safety redundancy design: The wing protection structure features quick-disassembly for easy maintenance and transportation, while its weight is kept to ≤1kg, ensuring no impact on the drone's flight performance. Furthermore, the flight control system incorporates collision detection; upon detecting a severe collision, it automatically cuts off motor power to prevent the rotors from continuing to rotate and causing secondary damage.

[0106] System collaborative workflow 1. Startup Phase: The ground control unit sends a start command, the UAV completes a self-check (status of sensors, battery, communication, etc.), the display screen initializes and loads preset display content; 2. Flight Phase: The UAV completes flight according to control commands (manual / autonomous). Obstacle avoidance sensors work in real time to ensure flight safety; environmental sensors collect temperature, humidity, and brightness data in real time and automatically adjust the display screen brightness. 3. Display Phase: The ground control terminal can update the display content in real time, and the display screen synchronously displays high-definition videos / images; 4. Low battery stage: When the battery level drops to the return threshold, the return procedure is automatically initiated, and the vehicle lands precisely at the charging base station and begins automatic charging; 5. Emergency Phase: In the event of a collision or abnormality, the motor power will be automatically cut off, an alarm will be triggered, and an abnormal signal will be sent to the ground control terminal. Example

[0107] This embodiment discloses another method for autonomously adjusting the display angle based on the viewer's position.

[0108] A visual servo closed-loop control system was constructed, which uses an airborne camera to capture the audience’s facial orientation and gaze direction in real time, and combines deep learning algorithms to identify the audience’s gaze point.

[0109] This system adopts a visual servoing closed-loop architecture of "perception-decision-control-feedback" and integrates an edge computing architecture to achieve efficient data processing and low-latency response. The overall architecture is divided into three layers: Perception layer: The core components are airborne cameras and multi-sensor arrays, which are responsible for collecting audience-related data in real time, including basic data such as audience facial images, location information, and ambient brightness, providing raw input for subsequent processing; Decision computing layer: Built on an edge computing architecture, it integrates a deep learning inference module, a data fusion module, and an optimization decision module. It is responsible for real-time processing, analysis, and decision-making on the data collected by the perception layer, and outputting the optimal display angle adjustment command. Execution control layer: Includes angle adjustment actuator and power management module, responsible for responding to the adjustment command of the decision layer, accurately completing the display angle adjustment, and achieving low power operation through the power management module.

[0110] The core advantages of the architecture design are: localized data processing through edge computing to avoid data transmission latency, while adopting a low-power design strategy to strictly control power consumption while ensuring computing power requirements, making it suitable for mobile or portable display scenarios.

[0111] The system captures real-time image data of the audience's facial orientation and gaze direction using an onboard camera. This data is then input into a deep learning algorithm model to accurately identify the audience's gaze point. The identification result serves as a closed-loop feedback signal, which is transmitted to the decision-making and computation layer. This signal is compared with the preset display effect target. If a deviation is found, an angle adjustment command is triggered. After the adjustment is completed by the execution mechanism, audience status data is collected again via the camera for feedback verification until the optimal display effect is achieved, forming a complete closed loop of "acquisition-identification-decision-adjustment-verification".

[0112] Multi-sensor data fusion using an improved Kalman filter The state vector includes the viewer's position, line of sight, and movement speed.

[0113] The current system state vector contains three core parameters: audience position (three-dimensional coordinates), viewing angle (angle with the normal vector of the display screen), and movement speed (linear velocity and angular velocity). A is the state transition matrix, which describes the relationship between the state at the previous time step and the state at the current time step; B is the control input matrix, which characterizes the influence of external control variables on the system state; The input quantity was controlled in the previous time step; This is process noise, which follows a Gaussian distribution, and is used to compensate for sensor measurement errors and environmental interference.

[0114] By employing a prediction-update mechanism using Kalman filtering, data from multiple sources such as cameras, position sensors, and speed sensors can be effectively fused to filter out noise interference and improve the accuracy and stability of state parameter estimation.

[0115] Establish a function to evaluate the display effect: , Q: Display performance evaluation index; the higher the value, the better the display performance. The angle between the viewer's line of sight and the normal vector of the display screen. The higher the value, the closer the line of sight is to the front of the display screen, and the better the visual experience; Display brightness dynamically adapts to ambient light levels to ensure a clear and non-glaring image. Viewing distance between the audience and the display screen The impact of distance on visual experience is represented by the fact that the more appropriate the distance, the more reasonable the contribution value of this factor. α, β, γ: Weighting coefficients, which are adaptively adjusted according to the actual application scenario (such as indoor / outdoor, static / dynamic viewing) to ensure reasonable weighting of each evaluation dimension.

[0116] The optimal display angle that maximizes the evaluation function Q is found using the gradient descent algorithm, i.e.: To balance the adjustment speed and accuracy, the iteration step size η=0.01 and the convergence accuracy ε=0.001 are set. When the change in Q during the iteration process is less than ε, the iteration stops and the current angle is output as the optimal adjustment angle.

[0117] Eye tracking and local enhancement module Module positioning: Based on the angle adjustment, further optimize the local effects of the displayed content to enhance the audience's perception of key content.

[0118] Implementation details: A high-precision eye-tracking algorithm is integrated, using an onboard camera to detect changes in the viewer's pupil position in real time. Combined with facial feature point matching, the coordinates of the viewer's gaze focus are calculated. Based on these coordinates, the corresponding area of ​​the displayed content undergoes local highlighting (enhancing local brightness and contrast) and detail enhancement (optimizing resolution and sharpness), making the content in the viewer's gaze area clearer and more prominent.

[0119] Stability control: For scenes with moving viewers, a speed threshold is set (moving speed <2m / s). Within this threshold range, the system maintains stable eye tracking and local enhancement effects. When the moving speed exceeds the threshold, the local enhancement function is temporarily turned off to prioritize the stability of angle adjustment and avoid tracking deviations caused by rapid movement.

[0120] Low-power and edge computing adaptation module: Module positioning: Ensures the system operates efficiently with low power consumption and adapts to the application needs of mobile or portable display devices.

[0121] Implementation strategy: By adopting an edge computing architecture, all data processing (deep learning inference, Kalman filtering, gradient descent solution) is completed locally, avoiding the latency and additional power consumption caused by uploading data to the cloud; Hardware computing power adaptation: Low-power GPU chips are selected, and the GPU computing power requirement is strictly controlled to <15TOPS. At the same time, through a dynamic computing power allocation strategy, the computing power output is reduced when there is no audience movement or the line of sight is stable, further saving power consumption. Power consumption management: By combining power management chips with software energy-saving algorithms, the overall power consumption of the system is controlled within the specified range.

[0122] To ensure the system's usability and user experience, the following core performance indicators are set, covering key dimensions such as response speed, adjustment accuracy, power consumption, and computing power: System response latency: <80ms (the entire process from the camera capturing the change in the viewer's state to the actuator completing the angle adjustment); Angle adjustment accuracy: ±0.08° (verified by professional angle measurement equipment to ensure the accuracy of display angle adjustment); GPU computing power requirement: <15TOPS (adapts to low-power hardware solutions, reducing equipment costs and heat dissipation pressure); System power consumption: meets the battery life requirements of mobile devices and is suitable for battery-powered scenarios; Stable tracking speed threshold: Maintain stable eye tracking and angle adjustment when the audience's movement speed is <2m / s.

[0123] This invention employs a three-layer Markov prediction model to implement an intelligent caching mechanism. The first layer, scene state prediction, establishes a state transition matrix based on environmental parameters, predicting the transition probabilities of eight scene states. The second layer, content demand prediction, establishes a content call probability matrix for each scene. The access probability distribution of different content types is quantified. The third-layer resource allocation prediction uses an improved LRU-K algorithm (K=3), combined with a sliding window learning mechanism to dynamically optimize the allocation of 32MB cache space; a segmented caching strategy is adopted for large video files. The overall cache hit rate is calculated using the formula... Calculation, where For scene prediction accuracy, For content prediction accuracy, To improve resource utilization, the goal is to achieve a hit rate of over 82% and a content loading latency of 35ms.

[0124] This invention adopts an edge computing architecture and a low-power design, with GPU computing power requirement of <15TOPS, power consumption controlled within 12W, system response latency of <80ms, and angle adjustment accuracy of ±0.08°. Example

[0125] A multi-faceted LED display screen for intelligent control of aerial flight includes a multi-rotor UAV flight carrier, a multi-faceted LED full-color display terminal, and an intelligent sensing and control module.

[0126] The multi-rotor UAV's flight platform adopts a quadcopter configuration, with the fuselage made of high-strength carbon fiber composite material. The power system is equipped with four brushless DC motors. It supports flight modes such as hovering, path planning, and autonomous obstacle avoidance. Through the linkage and communication between the flight control system and the display control module, it ensures that the flight position accuracy is maintained within ±0.5 meters and matches the displayed angle in real time.

[0127] The multi-faceted LED full-color display terminal adopts a cubic structure. The display screen uses a direct LED chip driving solution, with a color gamut ≥ 72% NTSC and a response time ≤ 5ms. The mechanical connection of the display screen uses a customized metal bracket to rigidly connect with the drone body. The bracket has built-in rubber shock-absorbing pads (5 mm thick, damping coefficient 0.3) to reduce structural damage and display interference caused by flight vibrations, with a vibration attenuation rate ≥ 80%.

[0128] Power supply system: Provides power to the display screen driver and control circuits, with a voltage regulation accuracy of ±0.1V to ensure voltage stability. Electrical connections utilize waterproof and dustproof aviation connectors (IP67 protection rating) integrating power and data cables, enabling bidirectional communication between the flight control system and the display screen's main controller. Data transmission rate ≥100 Mbps meets safety requirements in complex outdoor environments.

[0129] The intelligent sensing and control module integrates multiple sensors, including an RGB-D camera, three millimeter-wave radar sensors, four ambient light sensors, and two IMU attitude sensors.

[0130] The RGB-D camera combines binocular stereo vision and structured light technology to acquire audience depth information with a detection accuracy of ±0.3 meters. It is positioned diagonally at 45° above the front and 135° below the rear of the drone, covering a field of view of 120°. The millimeter-wave radar sensor provides 360° all-around audience position detection with a ranging range of 0.5-50 meters. It is positioned at three points: the front, left, and right of the drone. The ambient light sensor uses a photodiode and ADC converter to output 16-bit digital brightness data with a detection accuracy of ±5 lux. The IMU attitude sensor detects the drone's instantaneous rotational angular velocity and acceleration with a measurement accuracy of ±0.01°, and is arranged in a dual-redundant central configuration.

[0131] Multi-sensor data is transmitted bidirectionally via a data transmission module. A multi-head attention mechanism is employed to extract features from the multimodal data, generating a 512-dimensional environmental state feature vector. A Kalman filter algorithm is used to fuse the multi-sensor data. The state vector includes the audience's 3D position, line-of-sight angle, and UAV attitude angle, achieving high-precision estimation of audience position, ambient brightness, and flight attitude with a positioning error ≤0.3 meters. Simultaneously, a time synchronization mechanism is used, based on GPS time synchronization, to achieve data timestamp alignment with a synchronization accuracy of <1 millisecond.

[0132] An adaptive display control architecture is constructed based on the Transformer attention mechanism. The perception layer extracts multimodal features, the decision layer implements dynamic threshold and priority models, and the execution layer implements adaptive control of content and angle. By establishing a display effect evaluation function, a gradient descent algorithm is used to achieve precise angle adjustment, ensuring optimal display results.

[0133] To address the display jitter caused by drone flight attitude disturbances, an intelligent image stabilization scheme combining LSTM neural network prediction and bilinear interpolation compensation is proposed. The LSTM attitude prediction model achieves attitude prediction within 200 milliseconds, with a prediction error ≤0.5° and a prediction accuracy ≥95%. A bilinear interpolation algorithm is used for pixel remapping, achieving pixel-level precise image stabilization without image stretching distortion.

[0134] Intelligent resource scheduling is achieved through a three-level Markov predictive caching mechanism, combined with a sliding window learning mechanism to optimize cache parameters. An improved LRU-K algorithm (K=3) is used to dynamically allocate 32 megabytes of cache space, achieving a cache hit rate of ≥82%. Predictive load balancing algorithm is used to achieve power allocation, ensuring that the overall system power consumption meets requirements under both normal and extreme operating conditions. Example

[0135] A multi-faceted LED display screen for intelligent control of aerial flight includes a six-axis multi-rotor UAV flight carrier, a six-sided LED full-color display terminal, and an intelligent sensing and control module.

[0136] The hexagonal multi-rotor UAV adopts a hexagonal multi-rotor configuration, and the fuselage is made of high-strength carbon fiber composite material. The power system is equipped with six brushless DC motors. It supports flight modes such as hovering, path planning, and autonomous obstacle avoidance. Through the linkage and communication between the flight control system and the display control module, it ensures that the flight position accuracy is maintained within ±0.3 meters and matches the displayed angle in real time.

[0137] The hexahedral LED full-color display terminal adopts a hexahedral structure, and the display screen uses a direct LED chip driving solution. The mechanical connection of the display screen adopts a customized metal bracket that is rigidly connected to the drone body. The bracket has built-in rubber shock-absorbing pads (8 mm thick, damping coefficient 0.4) to reduce the structural damage and display interference of flight vibration to the display screen, with a vibration attenuation rate of ≥85%.

[0138] Power supply system: Provides power to the display screen driver and control circuits, with a voltage regulation accuracy of ±0.05V to ensure voltage stability. Electrical connections utilize waterproof and dustproof aviation connectors (IP68 protection rating) integrating power and data cables, enabling bidirectional communication between the flight control system and the display screen's main controller. Data transmission rate is ≥200 Mbps, meeting the safety requirements for use in complex outdoor environments.

[0139] The intelligent sensing and control module integrates multiple sensors, including an RGB-D camera, four millimeter-wave radar sensors, five ambient light sensors, and three IMU attitude sensors.

[0140] The RGB-D camera combines binocular stereo vision and structured light technology to acquire audience depth information with a detection accuracy of ±0.2 meters. It is positioned diagonally at 50° above and 130° below the front of the drone, covering a field of view of 140°. The millimeter-wave radar sensor provides 360° all-around audience position detection with a ranging range of 0.3-60 meters. It is positioned at four points: in front of the drone, 70° to the left, and 70° to the right. The ambient light sensor uses a photodiode and ADC converter to output 16-bit digital brightness data with a detection accuracy of ±3 lux. The IMU attitude sensor detects the drone's instantaneous rotational angular velocity and acceleration with a measurement accuracy of ±0.005°, and is arranged in a dual-redundant central configuration.

[0141] Multi-sensor data is transmitted bidirectionally via a data transmission module. A multi-head attention mechanism is employed to extract features from the multimodal data, generating a 1024-dimensional environmental state feature vector. A Kalman filter algorithm is used to fuse the multi-sensor data. The state vector includes the audience's 3D position, line-of-sight angle, and UAV attitude angle, achieving high-precision estimation of audience position, ambient brightness, and flight attitude with a positioning error ≤0.2 meters. Simultaneously, a time synchronization mechanism is used, based on GPS time synchronization, to align data timestamps with a synchronization accuracy of <0.5 milliseconds.

[0142] An adaptive display control architecture is constructed based on the Transformer attention mechanism. The perception layer extracts multimodal features, the decision layer implements dynamic threshold and priority models, and the execution layer implements adaptive control of content and angle. By establishing a display effect evaluation function, a gradient descent algorithm is used to achieve precise angle adjustment, ensuring optimal display results.

[0143] To address the display jitter caused by drone flight attitude disturbances, an intelligent image stabilization scheme combining LSTM neural network prediction and bilinear interpolation compensation is proposed. The LSTM attitude prediction model achieves attitude prediction within 300 milliseconds, with a prediction error ≤0.3° and a prediction accuracy ≥97%. A bilinear interpolation algorithm is used for pixel remapping, achieving pixel-level precise image stabilization without image stretching distortion.

[0144] Intelligent resource scheduling is achieved through a three-level Markov predictive caching mechanism, combined with a sliding window learning mechanism to optimize cache parameters. An improved LRU-K algorithm (K=4) is used to dynamically allocate 64 megabytes of cache space, achieving a cache hit rate of ≥85%. Predictive load balancing algorithm is used to achieve power allocation, ensuring that the overall system power consumption is within the specified range.

[0145] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0146] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for intelligent aerial flight control of a polyhedral LED display screen, the polyhedral LED display screen comprising a multi-rotor drone and a full-color display screen composed of peripheral multi-faceted LEDs, characterized in that, Intelligent control methods for airborne flight include: RGB-D cameras, millimeter-wave radar, ambient light sensors, and IMU attitude sensors are deployed on a multi-faceted LED display screen to acquire real-time data on audience position, viewing angle, ambient brightness, and flight attitude. The acquired data undergoes fusion processing and time synchronization processing; Assign weights to visual information, radar information, ambient light, and IMU data; Based on the Transformer attention mechanism, a three-tiered adaptive display control architecture of perception layer, decision layer, and execution layer is adopted to realize content layered display and mode switching; By combining an improved Kalman filter algorithm with multi-source data, the optimal display angle is output.

2. The intelligent flight control method for a polyhedral LED display screen according to claim 1, characterized in that, The RGB-D cameras are located at 45° above the front of the device and 135° below the rear, covering a field of view of 120°. Millimeter-wave radar: located at the front of the fuselage, 60° to the left, and 60° to the right, respectively, with a field of view covering 120°; IMU attitude sensor: The main sensor is located at the center of gravity, and the auxiliary sensor is offset by 15mm.

3. The intelligent flight control method for a polyhedral LED display screen according to claim 2, characterized in that, The data fusion and time synchronization processes include: The Kalman filter algorithm is used to fuse multi-sensor data, and the state vector includes the audience's three-dimensional position, line of sight angle, ambient light intensity, and drone attitude angle; Sensor data acquisition employs a time synchronization mechanism, using GPS time synchronization to align data timestamps. The weighting of visual information, radar information, ambient light, and IMU data includes: Visual information weights are 0.3-0.5, with higher values ​​used for core line-of-sight localization when there is sufficient light; radar information weights are 0.2-0.4, with higher values ​​used when there is insufficient light or occlusion; ambient light weights are 0.1-0.2; IMU data weights are 0.1; all weights are dynamically summed to 1.

0.

4. The intelligent flight control method for a polyhedral LED display screen according to claim 3, characterized in that, Perception layer: Generate an environmental state feature vector, including: light intensity L, audience density ρ, audience viewing angle θ, drone attitude angle φ, and audience distance d; Decision-making level: By dynamically allocating modal weights through a cross-attention module, a unified representation and feature fusion of heterogeneous data can be achieved. Based on audience distance d and attention concentration Establish a priority evaluation model for illumination intensity L: in: ,default , =0.5, =0.2, For optimal display brightness, The maximum ambient light level is set to 10000 lux. Adjusting the display mode by setting a threshold: Night mode: Activated when ambient light L < 200 lux, reducing brightness and increasing contrast; High-density mode: Switches when audience density ρ > 5 people / m², font size is enlarged, and the proportion of core information is increased; Execution layer: Content is displayed in layers: the font size of core information is adaptive and stable; the transparency of auxiliary information is dynamically adjusted; the background decoration is automatically matched according to the ambient color temperature; and the content switching uses a fade-in / fade-out effect to achieve a smooth transition.

5. The intelligent flight control method for a polyhedral LED display screen according to claim 4, characterized in that, By combining an improved Kalman filter algorithm with multi-source data, the optimal display angles are output, including: By fusing data from multiple sensors, a function for evaluating display performance is established. Where θ is the angle between the line of sight and the normal vector of the display screen, and B is the actual display brightness. For optimal brightness, d represents the audience distance. For the optimal distance, ,default =0.5, =0.3, =0.2; using the gradient descent algorithm, with a step size η=0.01 and a convergence accuracy ε=0.001, the Q value is maximized.

6. The intelligent flight control method for a polyhedral LED display screen according to claim 1, characterized in that, Intelligent flight control methods also include image stabilization compensation: An environmental sensing sensor is also installed on the multi-faceted LED display screen; Collect 3-axis angular velocity, 3-axis acceleration, and 3-axis magnetometer data from the IMU attitude sensor, and wind speed, air pressure, and temperature characteristic parameters from the environmental perception sensor; A 3-layer LSTM fully connected architecture is adopted. The number of neurons in the input layer matches the 12-dimensional feature parameters. The number of neurons in the hidden layer are 128, 64, and 32 respectively. The activation function is ReLU. The output layer contains 3-dimensional attitude parameters of pitch angle, roll angle, and yaw angle. Based on the pose change predicted by LSTM, a bilinear interpolation algorithm is used for pixel remapping, and the compensation model is as follows: In the formula, P is the compensated target pixel value, and w is the interpolation weight, satisfying... P represents the value of four adjacent pixels in the original image.

7. The intelligent flight control method for a polyhedral LED display screen according to claim 6, characterized in that, Intelligent flight control methods also include: The angle adjustment is controlled as a primary control, and the anti-shake compensation is controlled as a secondary control. When the angle adjustment is performed, an active adjustment flag is output to the LSTM model, the model temporarily stores the current prediction result, and the anti-shake compensation is paused. If the angle error is ≤0.08° and lasts for 50ms, the angle adjustment is completed, and the anti-shake compensation is restarted after a 50ms delay.

8. The intelligent flight control method for a polyhedral LED display screen according to claim 7, characterized in that, Intelligent flight control methods also include: By using predictive algorithms to preload display content, response time is optimized and overall power consumption is controlled. A three-level Markov prediction model is used: The first layer is scene state prediction: a state transition matrix is ​​established based on environmental parameters such as illumination L, audience density ρ, and time t. Predict scene type; The second layer is content demand prediction: establishing content call probability matrices for different scenarios. The probability of text content being retrieved is 0.6-0.8, image content 0.3-0.5, and video content 0.1-0.

3. The third layer is resource allocation prediction: a modified LRU-K algorithm is used to dynamically allocate 32MB of cache space, and a sliding window learning mechanism is combined to optimize cache parameters; Cache hit rate formula: Where: H is the cache hit rate, P is the scene prediction accuracy, C is the content prediction accuracy, and R is the resource utilization rate. =0.5、 =0.3、 =0.

2.

9. The intelligent flight control method for a polyhedral LED display screen according to claim 8, characterized in that, Intelligent flight control methods also include: Predict power consumption requirements in different scenarios and dynamically adjust LED driving current and display area activation ratio; Predictive load balancing algorithm formula: in: The total power consumption of the system is =0.7 is the basic power consumption coefficient, and β=0.3 is the environmental adaptability coefficient. Based on power consumption, This formula is used to predict power consumption in different scenarios and dynamically adjust the LED driving current and the activation ratio of the display area.

10. An intelligent control system for aerial flight of a multi-faceted LED display screen, used to implement the intelligent control method for aerial flight as described in any one of claims 1-9, characterized in that: The intelligent flight control system includes a sensor acquisition module, a data fusion and synchronization module, a weight allocation module, an adaptive display control algorithm module, an angle adaptive adjustment module, and an anti-shake compensation module. Sensor acquisition module: Real-time acquisition of audience position, line of sight angle, ambient brightness, and flight attitude data based on multiple types of sensors; Data fusion and synchronization module: performs fusion processing and time synchronization processing on the acquired data; Weight allocation module: Assigns weights to visual information, radar information, ambient light, and IMU data; Adaptive display control algorithm module: Based on the Transformer attention mechanism, it adopts a three-level adaptive display control architecture of perception layer - decision layer - execution layer to realize content layered display and mode switching; Angle adaptive adjustment module: Combines improved Kalman filter algorithm with multi-source data to output the optimal display angle; Image stabilization compensation module: Based on the pose change predicted by the LSTM neural network, pixel remapping is performed using a bilinear interpolation algorithm.

Images