Intelligent sunshade control method and system based on deep learning

Abstract

The invention relates to the technical field of sunshade control, in particular to an intelligent sunshade control method and system based on deep learning. The method comprises the following steps: acquiring a user position image and solar azimuth illumination data to form real-time environment perception information; carrying out coordinate positioning on the user by utilizing a pre-trained deep learning model and generating user movement track prediction data; a projection shadow area needing to be shielded is calculated based on the sun angle and the user position, and therefore a sunshade position expectation point is determined; and finally, carrying out flight attitude adjustment and buoyancy control on the intelligent sunshade according to the sunshade position expectation point and the user movement trend to realize dynamic intelligent sunshade for the user. According to the invention, through combination of deep learning identification and adaptive flight control, an intelligent sunshade effect of dynamically adjusting a shielding area according to a user position and sun illumination and stably hovering is realized.

Description

Technical Field

[0001] This invention relates to the field of sunshade control technology, and in particular to a deep learning-based intelligent sunshade control method and system. Background Technology

[0002] With the increasing number of outdoor activities and people's growing demand for comfortable environments, smart sunshade devices that can move with users and continuously provide shade protection are gradually becoming a research hotspot. Most existing sun umbrellas are fixed or require manual adjustment. When the user moves or the sun's angle changes, the shading range cannot automatically adapt, resulting in unstable shading effects and a poor user experience. While some portable sunshade devices have incorporated electric adjustment mechanisms, they still rely on active user operation and lack autonomous sensing and intelligent control capabilities.

[0003] In recent years, the rapid development of deep learning, target recognition, attitude control, and UAV flight control technologies has made intelligent dynamic positioning based on visual recognition and environmental perception possible. However, technologies that combine user positioning, sunlight angle analysis, trajectory prediction, and multi-degree-of-freedom flight control to achieve real-time adaptive shading are still relatively lacking. Furthermore, existing intelligent flight equipment still suffers from performance deficiencies in continuous hovering, energy consumption control, and environmental disturbance stability, making it difficult to achieve long-term, accurate, and safe shadow adjustment control. Summary of the Invention

[0004] Therefore, it is necessary to provide a deep learning-based intelligent sunshade control method and system to solve at least one of the above-mentioned technical problems.

[0005] To achieve the above objectives, a deep learning-based intelligent sunshade control method is applied to an intelligent sunshade comprising a camera module, a light sensor, a main control module, a power rotor assembly, and a buoyancy-adjusting airbag assembly. The method includes the following steps: Step S1: Acquire real-time environmental perception data, which includes user location images captured by the camera module and solar azimuth data captured by the light sensor. Step S2: Based on the pre-trained deep learning recognition model, perform user coordinate localization on the user location image to obtain user location information, and generate user motion trajectory prediction information based on continuous user location information; Step S3: Calculate the shading coverage requirement area based on solar azimuth and illumination data and user location information, and generate the desired shading location points; Step S4: Control the flight attitude and inflation / deflation of the smart sunshade umbrella by using the expected sunshade position and user movement trajectory prediction information to perform smart sunshade umbrella control operations.

[0006] The present invention has the following beneficial effects: I. A deep learning-based recognition model is used to perform real-time identification and coordinate positioning of the user's location in images. Combined with solar azimuth data acquired from a light sensor, this enables dynamic calculation of the shading range and prediction of shadow trajectories. Through user trajectory prediction, solar angle calculation, and a mechanism for generating desired shading positions, the sun umbrella can automatically and precisely provide shading coverage based on changes in user location and sunlight intensity, eliminating the need for manual adjustments. This solution significantly improves shading response speed and algorithm adaptability, enabling a personalized, predictive shading interaction experience and enhancing the intelligence and adaptability of smart shading.

[0007] Second, real-time adjustment of the powered rotor output is achieved through multi-dimensional attitude control commands and a dynamic flight correction mechanism. Dynamic attitude feedback closed-loop control, including rotor speed increase / decrease, differentiated output, and directional compensation, ensures the parachute maintains high stability during flight movement, attitude adjustment, and hovering. Simultaneously, by generating a correction thrust difference based on the attitude deviation direction and amount, rotational runaway, deviation accumulation, and flight jitter are avoided, improving the safety, stability, and positioning accuracy of the intelligent parachute during flight. This invention makes flight control more flexible and efficient, adaptable to wind disturbances, mobile users, and complex environmental interference.

[0008] Third, by introducing a buoyancy-adjustable airbag assembly, altitude control and fine-tuning are performed after the smart parachute reaches the target area. Automatic inflation and deflation achieve buoyancy compensation matching the hovering altitude range. The airbag adjustment rate is controlled by combining altitude change trends and rates, achieving flexible and dynamic altitude adjustment and balance, effectively reducing problems such as increased energy consumption, attitude instability, or increased noise caused by individual adjustment of the powered rotor. This solution improves user comfort, reduces the long-term workload of the power system, and improves the equipment's endurance and lifespan. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating the steps of a deep learning-based intelligent sunshade control method. Figure 2 for Figure 1 A detailed flowchart illustrating the implementation steps of step S4. Figure 3 This is a schematic diagram illustrating a scenario for a deep learning-based intelligent sunshade control method proposed in this application. Figure 4 This is a schematic diagram of the power rotor assembly structure of the intelligent sunshade control method based on deep learning proposed in this application; Figure 5 This is a schematic diagram of the buoyancy adjustment airbag component structure of a deep learning-based intelligent sunshade control method according to this application. 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

[0010] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0011] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0012] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0013] To achieve the above objectives, please refer to Figures 1 to 5 A deep learning-based intelligent sunshade control method is applied to an intelligent sunshade comprising a camera module, a light sensor, a main control module, a power rotor assembly, and a buoyancy adjustment airbag assembly. The method includes the following steps: Step S1: Acquire real-time environmental perception data, which includes user location images captured by the camera module and solar azimuth data captured by the light sensor. In one embodiment, real-time environmental perception data is first acquired. Specifically, the camera module and the light sensor module, mounted on the top of the sunshade bracket, are simultaneously activated and enter continuous sampling mode. The camera module uses a wide-angle imaging camera with a resolution of 1920×1080 and a frame rate of 30fps to acquire real-time images of the user's position within a 3-meter radius around the smart sunshade. To reduce the impact of ambient light changes on the recognition results, the camera uses automatic exposure adjustment and image noise reduction modes to ensure that the user's outline is clear and detectable.

[0014] Meanwhile, the illumination sensor module uses a four-quadrant photosensitive array sensor to collect information on solar illumination intensity distribution and solar azimuth. The sensor sampling frequency is set to 5Hz, and the sampled data undergoes an average filtering process to reduce measurement errors caused by instantaneous changes in illumination. Based on the differences in illumination intensity received in each quadrant of the four-quadrant photosensitive array, the solar azimuth and elevation angles are obtained by calculating the illumination intensity vector. For example, when the illumination intensity in the first region of the illumination sensor is 820 lux, the second region is 412 lux, the third region is 387 lux, and the fourth region is 95 lux, the solar direction vector is calculated as (θ=34.6°, φ=−17.8°) based on the built-in direction analysis model.

[0015] During the acquisition process, by controlling the time synchronization logic built into the motherboard, the user position image acquired by the camera module and the solar azimuth illumination data acquired by the light sensor are stored in a unified timestamp format, such as UTC+time format, and each sampling result is uniformly recorded as a data frame <image→illuminance vector→acquisition time>.

[0016] Step S2: Based on the pre-trained deep learning recognition model, perform user coordinate localization on the user location image to obtain user location information, and generate user motion trajectory prediction information based on continuous user location information; In one embodiment, after acquiring the user location image collected in step S1, the image is input into a pre-trained deep learning recognition model for processing. The deep learning recognition model can be a target detection and human pose estimation fusion network structure trained based on YOLO, HRNet, or MobileNet-Pose frameworks. The detection network is used to identify the human target region of the user in the image and output the human bounding box, while the pose estimation branch is used to extract the coordinates of key points of the human skeleton to improve the positioning accuracy.

[0017] After model recognition is completed, based on the coordinates of the center point of the output user bounding box or the coordinates of the torso center point in the skeleton key points, combined with the intrinsic parameters of the camera module (including focal length, distortion coefficient, pixel ratio coefficient, etc.) and extrinsic calibration parameters, the initial pixel-level coordinates are transformed into user position information in a real-world two-dimensional or three-dimensional coordinate system. The transformed user position information is stored in meters and includes a timestamp for subsequent trajectory analysis.

[0018] In another specific implementation, to improve the stability of location information, continuously acquired user location information is fused using moving average filtering and Kalman filtering to reduce coordinate noise caused by changes in lighting, occlusion, or recognition jitter. When the user's movement direction or speed changes abruptly, the filtering module automatically switches to a fast response mode to avoid delays affecting prediction results.

[0019] Based on the processed time-series location information, the trajectory analysis module is invoked to generate the user's motion trajectory. The trajectory analysis module utilizes a real-time velocity calculation model, a direction change detection algorithm, and a historical path-based trend prediction algorithm to generate predicted user motion trajectory information. This predicted information includes data such as the user's future short-term location, motion direction trend, and velocity change trend, which supports the automatic rotation of the sunshade, adjustment of the shading angle, and the execution of power control strategies.

[0020] In a preferred embodiment, if the user's movement trajectory is detected to exhibit a repetitive pattern (such as periodic movement, circling in place, linear approach or departure, etc.), the behavior pattern recognition submodule is further invoked. By analyzing the trajectory shape, acceleration changes, and number of turns, the user's intention is automatically determined, such as stopping, avoiding light while moving, and stability of walking trajectory. This improves the accuracy of movement trajectory prediction and provides in-depth behavioral support for subsequent intelligent control strategies.

[0021] Step S3: Calculate the shading coverage requirement area based on solar azimuth and illumination data and user location information, and generate the desired shading location points; In one embodiment, the user location information and user movement trajectory prediction information output in step S2 are obtained, and combined with the solar azimuth illumination data collected in step S1, the shading area to be covered is calculated, and the desired shading location points for guiding the movement of the sunshade umbrella are generated.

[0022] Specifically, the solar azimuth and illumination data in this embodiment include the solar azimuth angle, the solar altitude angle, and the real-time illumination intensity value. The solar azimuth angle is used to determine the direction of the sun's projection on the horizontal plane, and the solar altitude angle is used to calculate the tilt angle of the projected light. Before calculating the shading area, the solar projection direction vector Vsun is estimated based on the following formula: Vsun=(cos(θ)⋅cos(ϕ),sin(θ)⋅cos(ϕ),sin(ϕ)); where θ represents the solar azimuth angle and φ represents the solar altitude angle.

[0023] Subsequently, the user's real-time location information is represented as a three-dimensional coordinate point Puser(x,y,z) in the world coordinate system, and a shading demand area model is constructed. In this embodiment, the shading model is defined as a shading projection line segment extending in the opposite direction of sunlight with the center point of the user's head as the target point, and its endpoint is the desired shading coverage point. The desired shading coverage point Ptarget is calculated as follows: Ptarget = Puser − d ⋅ Vsun, where d is the dynamic offset distance calculated based on the solar altitude angle and the radius of the sunshade, satisfying: d = r ÷ tan(ϕ), where r is the effective shading radius of the sunshade (set to 0.8m in this embodiment); φ is the solar altitude angle, which comes from the light sensor or the solar position calculation model inside the system.

[0024] For example, when the user's location point Puser is (2.35, 1.72, 1.58)m, the solar azimuth angle θ=55°, and the solar altitude angle ϕ=30°, then we calculate: Vsun≈ (0.49, 0.70, 0.50); ​​d=r / tan(30°)≈1.385m; substituting into the formula: Ptarget≈(2.35,1.72,1.58)−1.385×(0.49,0.70,0.50), we finally obtain the desired shading point: Ptarget≈(1.67,0.75,0.89)m, and store this desired point in the motion control module to guide the subsequent motion control, step planning, and dynamic trajectory following of the smart sunshade.

[0025] Step S4: Control the flight attitude and inflation / deflation of the smart sunshade umbrella by using the expected sunshade position and user movement trajectory prediction information to perform smart sunshade umbrella control operations.

[0026] In one embodiment, after calculating the desired shading position and predicting the user's movement trajectory, the smart sunshade begins to perform sunshade attitude control and buoyancy adjustment operations so that the sunshade moves with the user and achieves intelligent control to continuously block sunlight.

[0027] Specifically, the intelligent sunshade flight control system incorporates a quadcopter power module, with four rotors driven by brushless DC motors. The drive power of each rotor is adjusted in real time by the flight control unit (FCU). Every 50ms, the flight control unit receives target sunshade position data and user motion trajectory prediction data from the navigation control module. By judging the user's motion trend, wind direction compensation parameters, and sunshade offset, it dynamically generates flight attitude control commands.

[0028] When a horizontal offset greater than 20cm or an altitude offset greater than 10cm is detected relative to the current position of the sunshade, the flight control unit generates attitude adjustment commands and differential thrust based on the offset direction. For example, when the desired position is located to the right front of the smart sunshade, the output speed of the right front powered rotor decreases, while the output speed of the corresponding rear left powered rotor increases. This generates an attitude tilt and displacement flight maneuver towards the desired position, allowing the sunshade to gradually approach the target area and hover stably again.

[0029] Meanwhile, the buoyancy adjustment airbag assembly of the sunshade automatically compensates based on the current height information. When external wind interference is detected, causing a deviation of ±5cm or more in the vertical height of the sunshade, the control system determines that a buoyancy correction operation is required: when the height of the sunshade is lower than the target height threshold, the buoyancy adjustment module activates the inflation actuator to inflate the buoyancy airbag at a flow rate of approximately 25–40 mL / s, providing additional lift; when the height of the sunshade is higher than the target height threshold, the exhaust actuator opens the exhaust valve to release the gas inside the airbag at a rate of approximately 20–30 mL / s, thereby reducing buoyancy and allowing the sunshade to return to the preset height range.

[0030] During flight control, the attitude stabilization module continuously collects data from the gyroscope, accelerometer, and barometer. If the rotor vibration amplitude is detected to exceed the set attitude stabilization limit (such as ±2° pitch deviation or ±3° roll deviation), an adaptive anti-shake algorithm is executed to fine-tune the motor output. At the same time, the gas regulation strategy is dynamically optimized according to the buoyancy change to keep the sunshade in a stable hovering state in the moving environment.

[0031] Ultimately, through the coordinated operation of the power rotor assembly and the buoyancy adjustment airbag assembly, the smart sunshade can always remain near the desired shading position (within ±10cm error range), and adjust its attitude and maintain its height in real time based on the user's movement trajectory prediction data, thus realizing the dynamic shading operation of the smart sunshade.

[0032] In another embodiment, the structure of the powered rotor assembly is referenced. Figure 4 The intelligent sunshade adopts a quadcopter flight structure, with the power rotor assembly symmetrically installed on both sides of the fuselage to provide lift, directional control and attitude stabilization. The power rotor assembly mainly includes the power rotor (1), brushless DC motor mounting base, motor drive circuit, propeller blades, shock absorption structure and power supply connection module, and is electrically connected to the main control module (7) and lithium battery assembly (5).

[0033] The rotor blades are made of lightweight composite materials and are directly locked to the output shaft of the brushless motor via a central shaft. The brushless motor is fixed to the fuselage frame with screws and a shock-absorbing structure, which reduces vibrations generated during high-speed operation.

[0034] The powered rotor assembly is connected to the main control module and lithium battery. The control signal output from the main control module adjusts the rotor speed, thereby achieving ascent, descent, hovering, and directional adjustment. When flight attitude needs to be adjusted, each powered rotor can increase or decrease its speed according to the control command to create a thrust difference and achieve precise attitude control.

[0035] In another embodiment, the structural reference of the buoyancy regulating airbag assembly Figure 5 The buoyancy-adjusting airbag assembly is located below the intelligent sunshade body and is used to adjust the device's floating height and hovering stability in the air. This assembly includes a main airbag, a secondary airbag, a main control air pump, a secondary control air pump, a main control air valve, a secondary control air valve, and a gas recovery module. All components are connected to the control module through flexible air supply pipes to form a closed gas circulation system.

[0036] The main airbag, located in the middle of the fuselage, has a large internal volume and provides basic buoyancy. The auxiliary airbags are symmetrically positioned on either side of the main airbag, used for fine-tuning flight attitude and lateral balance. The main control air pump and valve regulate the amount of gas inside the main airbag, controlling inflation and deflation to change overall buoyancy. The auxiliary control air pump and valve regulate the gas in the auxiliary airbags, refining the buoyancy difference between the left and right sides, thus achieving precise attitude correction.

[0037] The gas recovery module is located at the end of the gas path to collect and store gas during exhaust, preventing it from being released directly into the environment, improving gas utilization, and reducing flight noise. The entire airbag assembly uses internal pressure sensors to provide real-time feedback on the airbag status and communicates with the main control module to achieve closed-loop regulation.

[0038] As an example of the present invention, reference is made to Figure 2 As shown, step S4 in this example includes: Step S41: Obtain the current flight position of the umbrella and confirm the target area of ​​the parasol's movement using the user's motion trajectory prediction information; Step S42: Calculate the flight correction vector based on the spatial coordinate difference between the desired point of the sunshade position and the current flight position of the parachute, and convert the flight correction vector into multi-dimensional control data containing horizontal displacement, vertical displacement and rotational attitude adjustment. Step S43: Apply parameter constraints to the multidimensional control data using preset flight control constraints to obtain attitude control commands; Step S44: Input attitude control commands into the power rotor assembly to adjust the attitude of the smart umbrella, and perform inflation / deflation control after the smart sunshade reaches the target area of ​​the sunshade movement to execute the smart sunshade control operation.

[0039] In one embodiment, the user's motion trajectory prediction information may be derived from the fusion analysis of accelerometer data from a mobile terminal or wearable device and GPS data. Step S41, obtaining the flight position and predicted target area, includes: activating the umbrella positioning module (such as GPS or IMU) to obtain the current three-dimensional coordinates of the umbrella; inputting the current umbrella coordinates and the user's historical motion trajectory data into the prediction model to generate a predicted area for the user's possible location; delineating the target area for the parasol's motion around the umbrella and recording the center coordinates and boundary range of the area; performing a preliminary validity check on the predicted area to ensure that the area covers the user's possible future location, and adjusting the prediction model parameters if necessary to optimize coverage accuracy.

[0040] In another embodiment, a visual positioning module (such as a camera or lidar) can be used to assist in obtaining the umbrella's position, and the precise position data of the umbrella can be generated by fusing multiple frames. The predicted area can be dynamically updated according to the user's movement speed and direction. Each time it is updated, the position is verified first to ensure the validity of the target area.

[0041] In one embodiment, step S42 includes: calculating the three-dimensional coordinate difference between the current position of the parachute and the center of the target sunshade position, and generating a flight correction vector; decomposing the flight correction vector into horizontal displacement (X, Y directions), vertical displacement (Z direction), and rotational attitude adjustment (yaw angle, pitch angle, roll angle); converting the correction vector into a multi-dimensional control data format according to the parachute dynamics model for subsequent attitude control; and performing preliminary verification of the multi-dimensional control data to ensure that the data is within the operable range, and correcting it according to constraint rules if it exceeds the limit.

[0042] In another embodiment, the flight correction vector can be weighted and adjusted based on the size of the user's motion prediction area and real-time wind speed data to ensure a smooth and efficient parachute movement path.

[0043] In one embodiment, step S43 includes: setting flight control constraint parameters, including maximum displacement velocity, maximum acceleration, maximum rotational angular velocity, and upper limit of powered rotor load; constraining the multidimensional control data to generate attitude control commands that meet safety and stability requirements; verifying the attitude control commands to ensure that the commands will not cause powered rotor overload or attitude instability; if the verification fails, automatically adjusting the constraint parameters and regenerating the attitude control commands.

[0044] In another embodiment, the constraint parameters can be dynamically adjusted according to weather and environmental factors, such as reducing the maximum displacement speed and rotation speed when the wind speed is high, in order to ensure the stability of the umbrella.

[0045] In one embodiment, step S44 includes: transmitting attitude control commands to the parachute powered rotor control module to drive the parachute to complete horizontal and vertical movement and attitude rotation; monitoring the parachute position and attitude in real time to ensure that the flight path matches the target area, and generating correction control commands immediately if a deviation is detected; after the parachute reaches the target area of ​​the sunshade movement, activating the airbag inflation / deflation component to inflate or deflate the airbag according to the user's sunshade needs; and after completing the sunshade attitude adjustment, recording the final position, attitude, and airbag status of the parachute as reference data for the next flight and control optimization.

[0046] In another embodiment, the effect of umbrella posture adjustment can be verified by multi-sensor feedback (such as height sensor and wind speed sensor) to ensure that the umbrella surface achieves the best sunshade effect after inflation and deflation.

[0047] Preferably, inputting attitude control commands to the powered rotor assembly for intelligent parachute attitude adjustment includes: The attitude control command is converted into a control signal for controlling the powered rotor by the main control module, and the control signal is sent to the drive circuit of the powered rotor. The motor drive circuit of the powered rotor adjusts the working state of the corresponding powered rotor according to the control signal, including adjusting the rotor speed and rotor output direction; The different thrust outputs of each power rotor are used to adjust the attitude of the canopy, so that the canopy can be displaced or its attitude corrected in the direction of the desired point of sunshade. During the adjustment process, the main control module receives real-time attitude feedback information from the powered rotor assembly and makes adaptive fine adjustments to the powered rotor output based on the real-time attitude feedback information to keep the parachute in a stable flight or hovering state.

[0048] In one embodiment, the attitude control commands are first parsed and converted by the smart sunshade main control module to generate specific control signals for controlling the powered rotors. The steps include: converting multi-dimensional control data (horizontal displacement, vertical displacement, and rotational attitude adjustment) into rotor speed and output direction commands; sending the generated control signals to the drive circuits of each powered rotor via wired or wireless interfaces; and verifying the generated control signals to ensure signal integrity and latency are within acceptable limits (e.g., latency less than 50 milliseconds).

[0049] In another embodiment, after receiving the control signal, the motor drive circuit of the powered rotor adjusts the working state of the rotor according to the control signal, including: adjusting the rotation speed of each powered rotor to generate different thrust; adjusting the rotor output direction to change the thrust direction and realize spatial attitude control; each rotor outputs differentiated thrust, which works together to make the parachute undergo the required horizontal and vertical displacement or attitude rotation; and monitoring the rotor working state in real time during the movement to ensure that the rotor output meets the preset performance parameters.

[0050] In another embodiment, the main control module receives real-time attitude feedback information from the powered rotor assembly, including the parachute pitch angle, roll angle, yaw angle, and three-dimensional position coordinates; based on the real-time attitude feedback information, the main control module calculates the deviation from the target attitude; based on the deviation, it generates an adaptive fine-tuning signal and sends it to the powered rotor for fine-tuning; the above feedback and fine-tuning process is repeated until the parachute reaches the target attitude and maintains stable flight or hovering. Optionally, the influence of environmental factors (such as wind speed and air density) on thrust output can be considered during fine-tuning to achieve adaptive attitude control.

[0051] In another embodiment, after the parachute completes attitude adjustment, the following information is recorded as a reference for subsequent optimization: the final rotational speed and output direction data of each rotor; the final attitude and three-dimensional position of the parachute; the attitude deviation and number of fine adjustments during the adjustment process; if the parachute experiences an abnormality during the adjustment process (such as overspeed or attitude oscillation), a safety stop or return to the initial position operation can be triggered.

[0052] Preferably, adaptive fine-tuning of the rotor output based on real-time attitude feedback information to maintain stable flight or hovering of the parachute includes: The acquired information on the umbrella's attitude angle, tilt direction, and height changes is integrated into attitude feedback information, which is then transmitted to the main control module. The main control module compares the attitude feedback information with the target attitude state to determine the attitude offset direction and offset amount. The output of the powered rotor is adjusted according to the direction and amount of attitude deviation. The output adjustment includes increasing or decreasing the rotational speed of the powered rotor to create a thrust difference for correction. The thrust difference is used to make adaptive fine adjustments to the output of the powered rotor so that the parachute can maintain stable flight or hovering.

[0053] In one embodiment, the attitude control of the smart sunshade includes inputting attitude control commands to the powered rotor assembly and achieving adaptive fine-tuning through real-time attitude feedback, thereby maintaining stable flight or hovering of the umbrella. The attitude control commands are first parsed and converted by the umbrella's main control module to generate control signals for each powered rotor. These control signals include the rotational speed and output direction commands for each rotor, and are sent to the powered rotor drive circuit via a wired or wireless interface. Upon receiving the control signals, the drive circuit adjusts the rotational speed and output direction of the corresponding powered rotor according to the commands, utilizing the differentiated thrust of each rotor to achieve horizontal and vertical movement and rotational attitude adjustment of the umbrella, causing the umbrella to shift or correct its attitude along the desired point of the sunshade position.

[0054] During parachute adjustment, built-in sensors acquire information on changes in pitch, roll, yaw, tilt direction, and altitude, integrating this data into complete attitude feedback information which is then transmitted to the main control module. The main control module compares this feedback information with a preset target attitude, determines the direction and amount of attitude deviation, and makes adaptive fine-tuning adjustments to the rotor output based on the deviation. This fine-tuning process involves increasing or decreasing the rotor speed to generate the thrust difference required for correction, thereby achieving precise adjustment of the parachute attitude. Through cyclical feedback and fine-tuning, the parachute can maintain stable flight or hovering, and in complex environments, it can dynamically optimize fine-tuning parameters to enhance flight stability.

[0055] Once the parachute reaches the target area, the main control module activates the built-in airbag inflation / deflation system to adjust the parachute's posture and achieve final sunshade control. Based on the user's location, sunshade requirements, and parachute posture, the airbags are inflated or deflated to adjust the parachute's height, tilt, and stability, ensuring coverage of the target area. After inflation / deflation, the final rotor speed, posture, and airbag status are recorded for future flight path planning and control optimization. Furthermore, if any posture or rotor output abnormalities occur during adjustment, safety protection mechanisms are triggered, such as automatic hovering or returning to the initial position.

[0056] Preferably, the output of the powered rotor is adjusted according to the attitude deviation direction and amount, wherein the output adjustment includes increasing or decreasing the rotational speed of the powered rotor to form a thrust difference for correction, including: When the attitude deviation direction is consistent with the location of a certain powered rotor, it is determined that the parachute has a tendency to tilt in that direction, and the rotation speed of the powered rotor on the side of the tilt direction is reduced to reduce the thrust output in that direction. At the same time, the rotational speed of the powered rotor located on the opposite side of the tilt direction is increased to cause the parachute to return to its normal position in the opposite direction; When an upward lifting trend is detected in the parachute or the flight altitude exceeds the preset hovering altitude range, the rotation speed of all powered rotors is reduced synchronously. When a descent trend is detected in the parachute or its height is below the preset hovering height range, the synchronous rotation speed of all powered rotors is increased.

[0057] In one embodiment, the output of the powered rotor is adjusted according to the attitude deviation direction and amount, including increasing or decreasing the rotor speed to create a thrust difference for correction, thereby achieving parachute attitude recovery and altitude stability. The output adjustment process includes the following operations: When the parachute's attitude deviation direction aligns with the location of a certain powered rotor, it is determined that the parachute is tilting in that direction. To correct the tilt, the main control module reduces the rotational speed of the powered rotor on the side of the tilt, thereby reducing the thrust output on that side. Simultaneously, it increases the rotational speed of the powered rotor on the opposite side of the tilt, generating a relatively increased thrust on the other side of the parachute, causing the parachute to return to its correct position in the opposite direction, thus achieving attitude correction. Through this differentiated thrust output, the parachute can complete tilt correction in a short time and maintain a horizontal and stable state.

[0058] In another embodiment, when an upward trend is detected in the parachute or the flight altitude exceeds the preset hovering altitude range, the main control module will synchronously reduce the rotational speed of all powered rotors to reduce overall lift, causing the parachute to gradually descend to the target hovering altitude range. Conversely, when a downward trend is detected in the parachute or the flight altitude is below the preset hovering altitude range, the rotational speed of all powered rotors will be synchronously increased to increase overall lift, causing the parachute to rise to the target hovering altitude.

[0059] During the aforementioned fine-tuning operations, the main control module continuously receives feedback information on the output status of each rotor and the parachute attitude, verifying the output adjustment effect in real time. If insufficient or excessive correction is detected, the rotational speed of each rotor or the direction of fine-tuning can be further adjusted to ensure that the parachute maintains a stable flight or hovering state in terms of attitude and altitude. In addition, the rotor speed changes and attitude correction data during the adjustment process can be recorded for subsequent control optimization and safety monitoring.

[0060] Preferably, the inflation / deflation control after the smart sunshade reaches the target area for its movement includes: After the smart sunshade reaches the target area of ​​the sunshade movement, the main control module obtains the current height data of the sunshade and compares the current height data of the sunshade with the preset hovering height range to determine whether the height fine-tuning needs to be performed. When the current height of the sun umbrella exceeds the upper limit of the preset hovering height range, the buoyancy adjustment airbag assembly is controlled to perform an air deflation operation, so that the buoyancy is gradually reduced, thereby causing the umbrella to descend to the target hovering height range. When the current height of the parasol is lower than the lower limit of the preset hovering range, the buoyancy adjustment airbag assembly is controlled to inflate, causing the airbag to expand and increase buoyancy, so that the parasol rises to the target hovering height range.

[0061] In one embodiment, when the smart sunshade reaches the target area, the height of the umbrella is fine-tuned through inflation / deflation control to ensure that the umbrella surface is within a preset hovering height range to meet the sunshade requirements. After reaching the target area, the main control module first obtains the current height data of the umbrella and compares this height data with the preset hovering height range to determine whether a height fine-tuning operation is needed.

[0062] In another embodiment, when the current height of the parachute is detected to be higher than the upper limit of the preset hovering height range, the main control module controls the buoyancy regulating airbag assembly to perform an air deflation operation. During the deflation process, the buoyancy inside the airbag gradually decreases, and the parachute tends to sink until the height returns to within the target hovering height range. The air deflation operation can be performed in stages, and the height change is monitored in real time to ensure a smooth descent and avoid overshoot or oscillation.

[0063] Conversely, when the parachute's current altitude is below the lower limit of the preset hovering height range, the main control module controls the airbag to inflate. During inflation, the airbag expands, increasing buoyancy and causing the parachute to gradually rise to the target hovering height range. The inflation operation can also be implemented in stages, with fine-tuning based on real-time altitude feedback information to achieve a smooth ascent and maintain hovering stability.

[0064] Preferably, after the smart sunshade reaches the target area for its movement and performs inflation / deflation control, the following steps are also included: When controlling the buoyancy regulating airbag assembly to perform deflation or inflation operations, the speed and trend of height change of the smart umbrella are monitored in real time. The inflation and deflation rate of the buoyancy regulating airbag assembly is adjusted according to the speed and trend of height change of the smart umbrella.

[0065] In one embodiment, after the smart parasol reaches the target area for its movement and completes basic inflation / deflation control to achieve fine-tuning of its height, the system further monitors and dynamically adjusts changes in the parasol's height in real time. Specifically, during the inflation or deflation of the buoyancy-adjusting airbag assembly, the main control module continuously acquires information on the rate and trend of changes in the parasol's height and performs real-time analysis of these changes.

[0066] In another embodiment, the inflation / deflation rate of the buoyancy regulating airbag assembly is dynamically adjusted based on the rate and trend of the parachute's altitude change. When the altitude change rate is too fast or the trend indicates a possible overshoot, the main control module appropriately reduces the inflation or deflation rate to prevent the parachute from shaking or exceeding the preset hovering altitude range. Conversely, when the altitude change rate is too slow or the trend is insufficient to quickly reach the target altitude, the inflation or deflation rate can be appropriately increased to accelerate altitude adjustment efficiency.

[0067] Preferably, step S3 includes the following steps: Step S31: Confirm the solar altitude angle and solar azimuth angle based on the solar azimuth data; Step S32: Identify the user contour in the user location image and calculate the two-dimensional projected area of ​​the user contour on the reference plane; Step S33: Calculate the shadow projection trajectory of the two-dimensional projected area using the solar altitude angle and solar azimuth angle, and mark the calculated shadow projection trajectory area as the area requiring shading coverage; Step S34: Generate desired shading locations based on the shading coverage area.

[0068] In one embodiment, the solar altitude angle and solar azimuth angle are determined based on solar illumination data. This solar illumination data can be obtained from a light sensor, GPS information, and an astronomical computing module. The steps include: acquiring real-time light intensity, solar azimuth angle, and solar altitude angle, and performing calculations based on the current geographical location and time to obtain accurate solar position parameters. Optionally, the acquired solar altitude angle and azimuth angle are filtered or smoothed to eliminate the influence of sensor noise or short-term fluctuations on the shading calculation.

[0069] Subsequently, the user's location image is processed using a pre-trained image recognition model to identify the user's contour and calculate the two-dimensional projected area of ​​the contour on a reference plane. Image recognition can use deep learning segmentation networks (such as U-Net or Mask R-CNN) to ensure accurate identification of the user's contour even in complex backgrounds. When calculating the two-dimensional projected area, the user's height, pose, and the mapping relationship between the image and the reference plane must be considered to generate data on the projected area of ​​the space actually occupied by the user on the ground or the reference plane.

[0070] Based on the solar altitude angle and solar azimuth angle obtained in step S31, and the two-dimensional projected area obtained in step S32, the shadow trajectory of the user's projection is calculated. The calculation process includes projecting the two-dimensional projected area along the direction of sunlight to obtain the position and range of the shadow on the reference plane. The area of ​​the projected shadow trajectory is marked as the area requiring shading coverage, and the area can be optionally expanded or its boundaries smoothed to ensure more sufficient actual shading coverage and avoid shading dead angles.

[0071] Finally, desired shading locations are generated based on the shading coverage area. Generation methods may include selecting the center point of the coverage area as the target point, or selecting the optimal shading point to cover the user's activity path based on user movement trajectory prediction. Optionally, the generated desired points can be fine-tuned in conjunction with umbrella flight constraints (such as maximum horizontal displacement range and obstacle avoidance requirements) to ensure that the parasol can safely and efficiently reach the target point and provide effective shading.

[0072] Preferably, step S34 includes: Based on the spatial location of the area requiring shading coverage, determine the projection extension direction of the area under the current direction of solar illumination, and use it as the positioning reference for shading coverage. By using the solar azimuth and solar altitude angles to determine the directional offset of the shadow point based on the shading coverage positioning benchmark, the directional offset is made consistent with the angle of sunlight, thus forming a shading position offset reference line. Spatially correlate the shading position offset reference line with the center point of the shading coverage area to obtain the target range where the smart sun umbrella needs to hover, and calculate the effective shading coverage range that matches the umbrella diameter or radius. Within the effective shading coverage area, a set of spatial points that meet the shading coverage requirements are selected as the desired shading locations.

[0073] In one embodiment, based on the spatial location of the area requiring shading coverage, the projected extension direction of that area under the current direction of sunlight is determined and used as the shading coverage positioning reference. The projected extension direction can be calculated using the solar altitude angle and solar azimuth angle, and can optionally be fine-tuned in conjunction with the user's silhouette projection shape to ensure that the coverage area is consistent with the actual shading requirements.

[0074] Subsequently, the directional offset of the shadow's landing point is determined using the solar azimuth and solar altitude angles as a shading coverage positioning reference. This directional offset aligns with the angle of sunlight illumination, forming a shading position offset reference line. This reference line guides the smart sunshade umbrella in adjusting its position in space to ensure that the generated shading area matches the target shadow direction, thereby optimizing the shading effect.

[0075] Next, the reference line for the shading position offset is spatially correlated with the center point of the area requiring shading coverage to obtain the target range where the smart sunshade needs to hover. Based on the target range and the physical dimensions of the smart sunshade (such as the diameter or radius of the umbrella), the effective shading coverage range is further calculated to ensure that the hovering position of the umbrella can cover all or most of the required area.

[0076] Finally, within the effective shading coverage area, a set of spatial points that meet the shading coverage requirements is selected as the desired shading location points. Optionally, the desired points can be dynamically adjusted based on user movement trajectory prediction to achieve continuous or dynamic shading, ensuring that the umbrella provides optimal shading coverage at different times and locations.

[0077] Of particular importance are the training methods for pre-trained deep learning recognition models, including: The user location images collected by the camera module are divided into a dataset to obtain a model training set and a model test set. Image enhancement processing is performed on the model training set to obtain the enhanced model training set; The user key points or user bounding boxes in the enhanced model training set are labeled, and the YOLO model algorithm is used for supervised learning of the model to obtain a preliminarily trained deep learning recognition model. The pre-trained deep learning recognition model is optimized by gradient iteration, and the performance of the optimized deep learning recognition model is evaluated based on the model test set, thus obtaining the pre-trained deep learning recognition model.

[0078] In one embodiment, the camera module in the augmented reality display terminal collects user location images under different environmental conditions, including sunlight, shadows, backlighting, outdoor wind interference, changes in lighting, and changes in walking posture. The number of images obtained is no less than 5000, and they are divided into a dataset of 80%:20%, with 80% used as the model training set and 20% as the model test set, to ensure the model has data generalization ability.

[0079] Subsequently, image enhancement processing is performed on the training set data. Enhancement methods include, but are not limited to, random brightness perturbation (range ±30%), contrast adjustment, rotation (±25°), random cropping, horizontal flipping, background noise superposition, and spot simulation processing. The amount of enhanced data is expanded to 2 to 4 times that of the original training set to ensure that the deep learning training phase can cover more possible changes in the external scene.

[0080] After image enhancement, user features are annotated frame-by-frame on the enhanced training set. Annotations include the user's bounding box coordinates (x, y, w, h), and, if used for pose recognition, further annotations of key user points such as the shoulder, head center, and torso center, forming a standardized annotation file format (e.g., a YOLO annotation format TXT file). After annotation, a YOLO-based deep learning algorithm is used for supervised model training. The initial learning rate is set to 0.001, the batch size to 32, and the training runs at least 100 epochs. An early stopping strategy is implemented; training is terminated when the accuracy improvement is less than 0.1% after 15 consecutive epochs, resulting in a preliminary trained model.

[0081] After initial training, the pre-trained deep learning recognition model underwent gradient iterative optimization. During optimization, the Adam optimizer was enabled, and parameters of the convolutional layers and attention feature layers were fine-tuned to improve the model's sensitivity to small-scale human shifts or pose changes. After optimization, the model's test set was used as validation data input to evaluate recognition accuracy, recall, inference speed (FPS), and the overall detection performance metric mAP.

[0082] If the test results meet the conditions shown in Table 1, then the final pre-trained model will be output: Table 1 Model Test Indicators If the test results do not reach the preset performance threshold, the process returns to the previous training stage for further fine-tuning until the termination condition is met. The final model serves as the pre-trained deep learning recognition model for the smart sunshade and is used for subsequent real-time user coordinate localization and motion trajectory prediction tasks.

[0083] Of particular importance, step S32 includes: Separate the background region from the user's location image to obtain the main visual region; Extract the boundary contours of continuous pixel blocks within the main visual region, and determine the actual pixel range corresponding to the user contour based on the degree of closure of the boundary contours in the user position image to obtain the user contour. The pixel range of the user contour is mapped to a preset reference plane, and a corresponding two-dimensional projection area is generated according to the scaling ratio and pose change characteristics of the user contour under the imaging view. Pixel area statistics are performed on the two-dimensional projection area to obtain the two-dimensional projection area used to describe the proportion of the user's body shape.

[0084] In one embodiment, the user location image, initially located by a deep learning recognition model, is input into an image segmentation processing module. A preset background separation algorithm is used to remove background regions from the image. Background separation employs color segmentation, contrast difference analysis, and feature texture matching to filter out non-human areas, thereby obtaining the main visual region for subsequent analysis. This method effectively reduces interference from environmental noise factors; background elements such as ground textures, curtains, tree shadows, and light spots will not participate in subsequent contour extraction.

[0085] Subsequently, pixel blocks are extracted within the main visual region based on continuous pixel association rules, and edge tracking is performed on each pixel block to form a complete boundary contour. For the boundary contour, its degree of closure is used to determine whether it corresponds to a human target region. When the boundary exhibits continuous, unbroken characteristics, the true range of the user contour corresponding to that pixel region is confirmed. If the contour has local gaps caused by overlapping, clothing movement, or backlighting, a contour repair algorithm is used to interpolate and complete the broken areas, ensuring the final contour integrity.

[0086] After obtaining the user's complete pixel outline, the user outline region is mapped onto a preset reference plane in the system. During the mapping process, the outline shape is scaled according to the focal length of the imaging device, the sensor size, and the shooting distance. At the same time, the mapped shape is aligned and corrected based on possible changes in the user's posture (such as torso tilt, body rotation, side posture, etc.), and finally a two-dimensional projection area corresponding to the real human posture is generated.

[0087] Finally, a pixel area statistical operation is performed on the formed two-dimensional projection area. The statistical method is based on the pixel value ratio calculation to obtain two-dimensional projection area data that describes the proportion occupied by the user's body posture. This two-dimensional projection area can be used to assess the size of the user's field of vision, adjust the priority of sunshade coverage, and serve as a basis for dynamic behavior judgment in the subsequent user movement behavior analysis module.

[0088] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0089] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A deep learning-based intelligent sunshade control method, characterized in that, The method, applied to a smart sunshade umbrella comprising a camera module, a light sensor, a main control module, a power rotor assembly, and a buoyancy-adjusting airbag assembly, includes the following steps: Step S1: Acquire real-time environmental perception data, which includes user location images captured by the camera module and solar azimuth data captured by the light sensor. Step S2: Based on the pre-trained deep learning recognition model, perform user coordinate localization on the user location image to obtain user location information, and generate user motion trajectory prediction information based on continuous user location information; Step S3: Calculate the shading coverage requirement area based on solar azimuth and illumination data and user location information, and generate the desired shading location points; Step S4: Control the flight attitude and inflation / deflation of the smart sunshade umbrella by using the expected sunshade position and user movement trajectory prediction information to perform smart sunshade umbrella control operations.

2. The intelligent sunshade control method based on deep learning according to claim 1, characterized in that, Step S4 includes the following steps: Step S41: Obtain the current flight position of the umbrella and confirm the target area of ​​the parasol's movement using the user's motion trajectory prediction information; Step S42: Calculate the flight correction vector based on the spatial coordinate difference between the desired point of the sunshade position and the current flight position of the parachute, and convert the flight correction vector into multi-dimensional control data containing horizontal displacement, vertical displacement and rotational attitude adjustment. Step S43: Apply parameter constraints to the multidimensional control data using preset flight control constraints to obtain attitude control commands; Step S44: Input attitude control commands into the power rotor assembly to adjust the attitude of the smart umbrella, and perform inflation / deflation control after the smart sunshade reaches the target area of ​​the sunshade movement to execute the smart sunshade control operation.

3. The intelligent sunshade control method based on deep learning according to claim 2, characterized in that, Inputting attitude control commands into the powered rotor assembly for intelligent parachute attitude adjustment includes: The attitude control command is converted into a control signal for controlling the powered rotor by the main control module, and the control signal is sent to the drive circuit of the powered rotor. The motor drive circuit of the powered rotor adjusts the working state of the corresponding powered rotor according to the control signal, including adjusting the rotor speed and rotor output direction; The different thrust outputs of each power rotor are used to adjust the attitude of the canopy, so that the canopy can be displaced or its attitude corrected in the direction of the desired point of sunshade. During the adjustment process, the main control module receives real-time attitude feedback information from the powered rotor assembly and makes adaptive fine adjustments to the powered rotor output based on the real-time attitude feedback information to keep the parachute in a stable flight or hovering state.

4. The intelligent sunshade control method based on deep learning according to claim 3, characterized in that, Based on real-time attitude feedback information, adaptive fine-tuning of the rotor output is performed to maintain the parachute in a stable flight or hovering state, including: The acquired information on the umbrella's attitude angle, tilt direction, and height changes is integrated into attitude feedback information, which is then transmitted to the main control module. The main control module compares the attitude feedback information with the target attitude state to determine the attitude offset direction and offset amount. The output of the powered rotor is adjusted according to the direction and amount of attitude deviation. The output adjustment includes increasing or decreasing the rotational speed of the powered rotor to create a thrust difference for correction. The thrust difference is used to make adaptive fine adjustments to the output of the powered rotor so that the parachute can maintain stable flight or hovering.

5. The intelligent sunshade control method based on deep learning according to claim 4, characterized in that, The output of the powered rotor is adjusted according to the direction and amount of attitude deviation. This adjustment includes increasing or decreasing the rotor speed to create a thrust difference for correction. When the attitude deviation direction is consistent with the location of a certain powered rotor, it is determined that the parachute has a tendency to tilt in that direction, and the rotation speed of the powered rotor on the side of the tilt direction is reduced to reduce the thrust output in that direction. At the same time, the rotational speed of the powered rotor located on the opposite side of the tilt direction is increased to cause the parachute to return to its normal position in the opposite direction; When an upward lifting trend is detected in the parachute or the flight altitude exceeds the preset hovering altitude range, the rotation speed of all powered rotors is reduced synchronously. When a descent trend is detected in the parachute or its height is below the preset hovering height range, the synchronous rotation speed of all powered rotors is increased.

6. The intelligent sunshade control method based on deep learning according to claim 2, characterized in that, The inflation / deflation control of the smart sunshade after it reaches the target area for its movement includes: After the smart sunshade reaches the target area of ​​the sunshade movement, the main control module obtains the current height data of the sunshade and compares the current height data of the sunshade with the preset hovering height range to determine whether the height fine-tuning needs to be performed. When the current height of the sun umbrella exceeds the upper limit of the preset hovering height range, the buoyancy adjustment airbag assembly is controlled to perform an air deflation operation, so that the buoyancy is gradually reduced, thereby causing the umbrella to descend to the target hovering height range. When the current height of the parasol is lower than the lower limit of the preset hovering range, the buoyancy adjustment airbag assembly is controlled to inflate, causing the airbag to expand and increase buoyancy, so that the parasol rises to the target hovering height range.

7. The intelligent sunshade control method based on deep learning according to claim 6, characterized in that, After the smart sunshade reaches its target area and undergoes inflation / deflation control, the following steps are also included: When controlling the buoyancy regulating airbag assembly to perform deflation or inflation operations, the speed and trend of height change of the smart umbrella are monitored in real time. The inflation and deflation rate of the buoyancy regulating airbag assembly is adjusted according to the speed and trend of height change of the smart umbrella.

8. The intelligent sunshade control method based on deep learning according to claim 1, characterized in that, Step S3 includes the following steps: Step S31: Confirm the solar altitude angle and solar azimuth angle based on the solar azimuth data; Step S32: Identify the user contour in the user location image and calculate the two-dimensional projected area of ​​the user contour on the reference plane; Step S33: Calculate the shadow projection trajectory of the two-dimensional projected area using the solar altitude angle and solar azimuth angle, and mark the calculated shadow projection trajectory area as the area requiring shading coverage; Step S34: Generate desired shading locations based on the shading coverage area.

9. The intelligent sunshade control method based on deep learning according to claim 7, characterized in that, Step S34 includes: Based on the spatial location of the area requiring shading coverage, determine the projection extension direction of the area under the current direction of solar illumination, and use it as the positioning reference for shading coverage. By using the solar azimuth and solar altitude angles to determine the directional offset of the shadow point based on the shading coverage positioning benchmark, the directional offset is made consistent with the angle of sunlight, thus forming a shading position offset reference line. Spatially correlate the shading position offset reference line with the center point of the shading coverage area to obtain the target range where the smart sun umbrella needs to hover, and calculate the effective shading coverage range that matches the umbrella diameter or radius. Within the effective shading coverage area, a set of spatial points that meet the shading coverage requirements are selected as the desired shading locations.

10. A deep learning-based intelligent sunshade control system, characterized in that, For executing the deep learning-based intelligent sunshade control method as described in claim 1, the deep learning-based intelligent sunshade control system includes: The perception module is used to acquire real-time environmental perception data, which includes user location images captured by the camera module and solar azimuth data captured by the light sensor. The recognition module is used to locate user coordinates in user location images based on a pre-trained deep learning recognition model, obtain user location information, and generate user motion trajectory prediction information based on continuous user location information. The generation module is used to calculate the area requiring shading coverage based on solar azimuth data and user location information, and generate the desired shading location points. The control module is used to control the flight attitude and inflation / deflation of the smart sunshade based on the desired sunshade position and the user's movement trajectory prediction information, so as to perform the smart sunshade control operation.

Images