Self-adaptive Micro-LED headlamp automatic control system for vehicle

Abstract

The technical scheme of the invention discloses an automatic control system for a self-adaptive Micro-LED headlamp for a vehicle, and the intelligent adjustment of the Micro-LED headlamp is realized through the cooperative work of a sensing module, a calculation module, a control module and an execution module. Through high-frequency multi-sensor data fusion (including a camera, a millimeter-wave radar and a steering sensor) and an innovative adaptive multi-head attention You-only-look-only (AMHA-YOLO) algorithm, real-time accurate recognition of lane width, bending direction and opposite vehicles is realized, the angle of a Micro-LED headlamp is dynamically adjusted, and the brightness of the headlamp is adjusted in a partitioned manner; the limitation of poor lighting effect caused by GPS positioning failure and map missing limitation is overcome, and the driving safety and comfort are improved.

Description

Technical Field

[0001] The present invention relates to an adaptive automatic control system for Micro-LED headlights based on multi-sensor fusion and adaptive multi-head attention YOLO algorithm, belonging to the technical field of automotive electronic control systems. Background Art

[0002] Automobile headlight control technology is an important part of the intelligent development of automobiles and is of great significance for improving driving safety and comfort. Currently, automobile headlight control technology mainly relies on drivers to manually switch between high and low beams or automatically adjust the brightness based on ambient light. However, at night or under complex weather conditions, manual operation will distract the driver's attention, and it is difficult to avoid dazzling oncoming vehicles in real time or fully illuminate the road in curves. In the prior art, such as the Chinese invention patent with the publication number CN103419713A published on December 04, 2013 (hereinafter referred to as "Prior Patent One") and the Chinese invention patent application with the publication number CN111845536A published on October 30, 2020 (hereinafter referred to as "Prior Patent Two"), rely on positioning information and map data to adjust the headlight angle or spacing, but there are significant limitations. Specifically:

[0003] Prior Patent One relies on GPS positioning to judge the road trajectory. However, in signal-blocked areas such as tunnels or viaducts, the GPS positioning error can reach more than 10 meters, resulting in inaccurate headlight adjustment. Especially in curves, it is impossible to perform adaptive steering in advance, affecting the lighting effect.

[0004] Prior Patent Two judges the road width based on static map data. However, static maps cannot respond to dynamic situations such as road construction or temporary lane changes in real time, affecting the real-time performance and accuracy of headlight adjustment. Especially in scenarios where road conditions change frequently, it is difficult to ensure the headlight coverage area.

[0005] In addition, the prior art lacks the ability to achieve millisecond-level headlight angle adjustment through efficient multi-sensor fusion and innovative intelligent algorithms. There is especially a technical gap in the coordinated control of steering prediction and glare avoidance. The prior art cannot achieve real-time and accurate headlight adjustment. Especially under complex road conditions and bad weather conditions, it is difficult to simultaneously take into account the avoidance of dazzling oncoming vehicles and the lighting requirements in curves. Summary of the Invention

[0006] The technical problem to be solved by the present invention is that the existing automobile headlight control technology relies on positioning information and map data to adjust the headlight angle or spacing, and there are significant limitations.

[0007] To solve the above technical problems, the technical solution of the present invention discloses an automatic control system for an adaptive Micro-LED headlamp for vehicles, which is characterized in that through the collaborative work of a sensing module, a calculation module, a control module and an execution module, the intelligent adjustment of the Micro-LED headlamp is realized, wherein:

[0008] The sensing module is used to obtain road images, the distance of oncoming vehicles, relative speed, servo motor feedback information for adjusting the headlamp angle, vehicle yaw angular velocity and the speed of the vehicle itself at a fixed frequency. Among them, the road image is obtained through a camera, and the relative speed is obtained through a millimeter wave radar;

[0009] The calculation module is used to calculate the fused lane angle θ f (t):

[0010] θ f (t) = w I ·θ I (t) + w v ·θ v (t)

[0011] In the formula: t is the current sampling point time;

[0012] w I is the weight of camera data, w v is the weight of millimeter wave radar data, and: under high visibility, w I > w v , under low visibility, w I < w v ;

[0013] d r (t) is the lateral distance between the vehicle and the road boundary, v(t) is the speed of the vehicle itself, and ω(t) is the vehicle yaw angular velocity;

[0014] θ I (t) is the lane angle obtained through the AMHA-YOLO network. The AMHA-YOLO network is based on the YOLO architecture introduced with a multi-head attention mechanism, and takes road images, vehicle yaw angular velocity, vehicle speed itself, and the distance of oncoming vehicles as inputs:

[0015] The backbone network of the AMHA-YOLO network is used to extract multi-scale features of the input road image. In the attention layer of the backbone network, the input vehicle yaw angular velocity, vehicle speed itself and the distance of oncoming vehicles are used as context to modulate the attention weight;

[0016] The neck network of the AMHA-YOLO network aggregates multi-scale features, and a lightweight attention module is newly added to the FPN up path;

[0017] The head network of the AMHA-YOLO network outputs lane detection boxes;

[0018] The calculation module converts the lane detection box at the current moment t into world coordinates through the camera's internal and external parameters, and combines the world coordinates obtained at all moments between the current moment t to fit a curve, and the curvature of the curve is the lane angle θ I (t);

[0019] The control module obtains a control instruction for controlling the brightness of each area of the Micro-LED based on the distance of the oncoming vehicle, and generates a PWM signal for driving the corresponding Micro-LED area to turn on / off or adjust the brightness;

[0020] Meanwhile, the control module uses the fused lane angle θ f (t) and the servo motor feedback information as inputs, and uses a closed-loop control algorithm to generate a servo motor control signal;

[0021] The execution module receives the PWM signal and the servo motor control signal output by the control module, drives all areas of the Micro-LED headlight to adjust the brightness, and drives the servo motor to perform dynamic angle adjustment of the Micro-LED headlight.

[0022] Preferably, the perception module generates synchronous time points t at a fixed frequency f k , t k = t0 + k·Δt, where t0 is the initial moment, Δt = 1 / f, and k is the time step index.

[0023] Preferably, the perception module obtains the servo motor feedback information θ for adjusting the headlight angle at each synchronous time point t k : (t current ), obtains the distance d(t of the oncoming vehicle using the millimeter-wave radar k ) and the lateral distance d between the vehicle and the road boundary k (t), and obtains the vehicle's yaw angular velocity ω(t using the steering sensor r ). k )

[0024] Preferably, the perception module uses the linear interpolation method to align the vehicle's speed to the time axis t k , and obtains the vehicle's speed v(t k ).

[0025] Preferably, the perception module uses the nearest neighbor interpolation to align the road image to the time axis t k , and obtains the road image I(t k ).

[0026] Preferably, the backbone network adopts the CSPDarknet53 structure, which includes multiple layers of convolution and cross-stage partial connections. The attention layer is inserted after the convolution module of the backbone network. Each attention layer contains 4 parallel attention heads, which focus on lane edge texture, oncoming vehicle contour, pedestrian contour, and background noise suppression respectively.

[0027] Preferably, when calculating the lane angle θ I (t), the curve is a quadratic curve.

[0028] Preferably, the calculation module uses Kalman filtering to dynamically correct the fused lane angle, the distance of the oncoming vehicle, and the servo motor feedback information.

[0029] Preferably, the closed-loop control algorithm adopted by the control module is the PID control algorithm. Then, the servo motor control signal is obtained by the following formula:

[0030]

[0031] In the formula: K p 、K i 、K d are PID parameters; e(t) = θ f (t) - θ current (t), θ current (t) is the servo motor feedback information.

[0032] Preferably, when adjusting the brightness of all areas of the Micro-LED headlight, the brightness B i (t) of the i-th Micro-LED area at the current moment t is obtained according to the following formula:

[0033]

[0034] In the formula, B max is the maximum brightness;

[0035] w i (t) is the weight, and there is:

[0036]

[0037] Among them: θ i is the central angle of the i-th Micro-LED area, σ is the width controlling the weight distribution; is the horizontal angle of the oncoming vehicle.

[0038] The present invention proposes an adaptive automatic control method for Micro-LED headlights. Through high-frequency multi-sensor data fusion (including cameras, millimeter-wave radars, and steering sensors) and an innovative Adaptive Multi-Head Attention You-Only-Look-Once (AMHA-YOLO) algorithm, it realizes real-time and accurate identification of lane width, bending direction, and oncoming vehicles, dynamically adjusts the angles of Micro-LED headlights and adjusts the brightness of the headlights in zones, overcomes the limitations of poor lighting effects caused by GPS positioning failures and map missing limitations, and improves driving safety and comfort. Brief Description of the Drawings

[0039] Figure 1 It is the architecture diagram of the AMHA-YOLO network;

[0040] Figure 2 It is the structure diagram of Micro-LED zones;

[0041] Figure 3 It schematically shows the system architecture and data flow. Detailed Embodiments

[0042] The following further elaborates the present invention in combination with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and not to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms also fall within the scope defined by the appended claims of this application.

[0043] An embodiment of the present invention discloses an automatic control system for vehicle-mounted adaptive Micro-LED headlights, which realizes the intelligent adjustment of Micro-LED headlights through the coordinated work of a sensing module, a computing module, a control module, and an execution module.

[0044] The sensing module integrates multiple sensors to form high-frequency and synchronous data acquisition and fusion, and is used to obtain the lane width, bending direction, oncoming vehicle information, and the state of the vehicle in real time.

[0045] The hardware of the sensing module further includes a camera, a millimeter-wave radar, a steering sensor, and a vehicle speed sensor, and obtains the feedback information θ current (t k ) of the servo motor for adjusting the headlight angle, where t k is the synchronous time point generated at a fixed frequency f (in an embodiment of the present invention, a preferred embodiment is that the fixed frequency is 100 hz, that is, one synchronous point every 10 ms), and t k= t0 + k·Δt, where t0 is the initial time, Δt = 1 \f(In a preferred embodiment of the present invention, Δt = 10 ms), k = 1, 2, 3, ……, n, which is the time step index. In the embodiments of the present invention, the camera acquires high-resolution road images I(t k )(In a preferred embodiment of the present invention, the road image I(t k ) has a resolution of 640×480 and a format of RGB), providing visual information of lane lines, oncoming vehicles, and pedestrians.

[0046] The millimeter-wave radar obtains the distance d(t k )(unit: m) of the oncoming vehicle and the lateral distance d r (t) between the vehicle and the road boundary, which is applicable to low visibility environments.

[0047] The steering sensor provides the vehicle's yaw rate ω(t k )(unit: rad / s), reflecting the steering dynamics.

[0048] The vehicle speed sensor measures the speed v(t k )(unit: m / s) of the vehicle itself, which is used to dynamically adjust the forward attention distance.

[0049] The camera, millimeter-wave radar, steering sensor, and vehicle speed sensor transmit the collected data to the data synchronization unit through the CAN bus. The data synchronization unit is a software unit that uses timestamp alignment technology to align heterogeneous data to the unified time axis t k , eliminating time deviation and achieving multi-source data fusion. The fused data includes image features, distance vectors, and vehicle states, and is output to the calculation module in a unified format (In a preferred embodiment of the present invention, the fused data adopts the JSON format or can also be in a custom protocol), ensuring low latency and high consistency.

[0050] In the embodiments of the present invention, for sensor data with different sampling frequencies, the data synchronization unit uses linear interpolation or nearest neighbor interpolation methods to align the data to the same time point.

[0051] For the vehicle speed sensor, there is:

[0052] Suppose the vehicle speed sensor acquires data v1 and v2 at time t1 and time t2 respectively. The data synchronization unit uses the following formula to estimate the synchronized time point t k of the data v(t k ):

[0053]

[0054] For the camera, there is:

[0055] The camera captures images at 30fps and cannot directly interpolate pixel values. Therefore, the data synchronization unit uses nearest neighbor interpolation to estimate the road image I(t k ):

[0056] I(t k ) = I(t s )

[0057] where t s is the timestamp of the image frame closest to t k in time,

[0058] After time synchronization, the data of each sensor is aligned to the same time point t k , forming a time-step data packet D(t k ) = {i(t k ), v(t k ), ω(t k ), d(t k ), d r (t k )}.

[0059] The calculation module integrates the aligned data D(t k ) through a weighted fusion algorithm to calculate the fused lane angle θ f (t k ):

[0060] θ f (t k ) = w I ·θ I (t k ) + w v ·θ v (t k )

[0061] where: w I is the weight of the camera data, w v is the weight of the millimeter-wave radar data. In an embodiment of the present invention, a preferred implementation is that, under high visibility, the weight w i of the camera data is higher (for example, it can be taken as 0.7), while the weight w v of the millimeter-wave radar data is lower (for example, it can be taken as 0.3); in low visibility situations such as foggy days, the weight w v of the millimeter-wave radar data is increased. For example, w v can be increased to 0.8;

[0062] L is the wheelbase of the vehicle and can be taken as 2.5m;

[0063] θ I (t k ) is the lane angle extracted by AMHA - YOLO provided by the present invention.

[0064] In the present invention, the AMHA - YOLO adopted by the calculation module is the innovative Adaptive Multi - Head Attention YOLO. The hardware is based on automotive - grade embedded GPUs (such as NVIDIA Orin series) or high - performance SoCs (such as Qualcomm Snapdragon Ride), supporting real - time deep learning and motion prediction. AMHA - YOLO is optimized through the following lightweight technologies:

[0065] Network pruning: Remove redundant convolutional layers, reducing the number of parameters by about 30%;

[0066] Quantization optimization: Adopt INT8 quantization, convert FP32 weights and activation values into 8 - bit integers, reducing the computational overhead by about 50%;

[0067] Multi - head attention mechanism: Dynamically allocate feature weights, giving priority to focusing on oncoming vehicle contours and lane line edges, improving detection accuracy in low - light and foggy environments.

[0068] Compared with the conventional YOLO scheme, the AMHA - YOLO scheme disclosed in the present invention optimizes the attention allocation for specific targets in automotive scenarios (lane edges, oncoming vehicles, pedestrians, etc.). By introducing a customized multi - head attention mechanism, it dynamically focuses on lane line edges and oncoming vehicle contours, improving the detection accuracy in complex weather (such as low - light, foggy, rainy scenarios, etc.).

[0069] The multi - head attention mechanism of AMHA - YOLO dynamically allocates feature weights through parallel attention heads, giving priority to focusing on lane line edges (high - gradient regions) and oncoming vehicle contours (high - contrast regions). Each attention head is optimized for specific features (such as edge texture, vehicle shape), significantly improving the detection accuracy in night and foggy scenarios. For example, when the distance of the oncoming vehicle is < 100m, the attention mechanism can quickly lock on the headlight area of the vehicle, triggering glare avoidance adjustment.

[0070] AMHA - YOLO is used to identify oncoming vehicles, pedestrians, and lane lines, output target coordinates and confidence levels, predict the position of the lane center line 3 seconds later based on the vehicle kinematic model, and calculate the on - off and angle adjustment requirements of the Micro - LED headlight area.

[0071] The present invention implements AMHA - YOLO based on the improvement of the YOLOv5 architecture, using the road image I(t k), vehicle yaw angular velocity ω(t k ), the vehicle speed v(t k ), the distance d(t k ) is the input, and the network structure is as follows Figure 1 As shown, it contains the following key components:

[0072] (a) Backbone network: It uses CSPDarknet53 structure, including multi-layer convolution and cross-stage partial connection (CSP), to extract the road image I(t k )’s multi-scale features. The multi-head attention layer is inserted after the convolutional module of the backbone network (layers 3, 6, and 9). Each attention layer contains 4 parallel attention heads, focusing on lane edge texture, oncoming vehicle outline, pedestrian outline, and background noise suppression. The attention mechanism calculates feature weights through Scaled Dot-Product Attention. k ),ω(t k )、d(t k ) as context to modulate the attention weight W of the attention layer {注意力} =σ(f(v,ω,d)).

[0073] (b) Neck Network: Utilizing a Path Aggregation Network (PAN) to aggregate multi-scale features and adding a lightweight attention module (in the FPN upstream path), the Neck Network further enhances the representation of lane edges and vehicle outlines. Three different features in the Neck Network are optimized for three scales (small, medium, and large objects), adapting to lane lines and vehicles at varying distances.

[0074] (c) Head network: Outputs lane line detection box, category and confidence.

[0075] The lane detection frame is converted to world coordinates using camera intrinsic and extrinsic parameters.

[0076] In the embodiment of the present invention, a preferred implementation is that the lane line detection frame [x i ,y i ,x i+1 ,y i+1 ] is converted to world coordinates through the following process:

[0077]

[0078] Where, f x =f y =600, c x =320, c y =240;

[0079] [Xw,Yw,0] = R[x n Zc,y n Zc,Zc] + T

[0080] where Zc ≈ d(t k ), T represents the translation vector, indicating the spatial displacement of the origin of the camera coordinate system relative to the origin of the world coordinate system, and R represents the rotation matrix, indicating the rotational transformation of the camera relative to the world coordinate system.

[0081] Using all the world coordinates obtained before the current time t k to fit a quadratic curve, y(x) = ax 2 + bx + c, and calculating the curvature κ(t k ) which is θ I (t k ), and there is:

[0082]

[0083] where y′(x) = 2ax + b and y″(x) = 2a.

[0084] To further improve the data accuracy, the calculation module uses Kalman filtering to dynamically correct the fused lane angle θ f (t k ), compensating for the dynamic blur error of the camera.

[0085] In an embodiment of the present invention, a preferred implementation is that the Kalman filtering specifically uses the following formula:

[0086]

[0087] where: x k is the state vector, θ represents the smoothed lane angle θ f (t k ), represents the change rate of the lane angle,

[0088]

[0089] w k represents the process noise, which follows a Gaussian distribution, w k ~N(0,Q), Q = diag(0.01,0.01);

[0090] z k represents the observed value, that is, the measured value θ f (t k ) directly obtained from the sensor fusion data;

[0091] H = [1 0];

[0092] v k represents the observation noise, sensor measurement error, v k ~N(0, R), where R = 0.05.

[0093] The hardware of the control module is an automotive-grade microcontroller (MCU, such as the NXP S32K series) or an electronic control unit (ECU), which runs a real-time decision-making algorithm. This algorithm is based on a mature closed-loop control logic (in an embodiment of the present invention, a preferred implementation is that the closed-loop control logic adopts a PID control algorithm). According to the fused lane angle θ f (t k ), the distance d(t k ) of the oncoming vehicle, and the servo motor feedback information θ current (t k ), two types of control instructions are generated:

[0094] The first type of control instruction is the Micro-LED area brightness control instruction. PWM signals are allocated to 10 Micro-LED areas to adjust the brightness from 0 to 100% to dynamically distribute the brightness of each area based on the distance of the oncoming vehicle (glare avoidance is triggered when <100m), ensuring glare avoidance while maintaining road lighting.

[0095] In an embodiment of the present invention, the brightness B k of the i-th Micro-LED area at the current moment t i (t k )(0 - 100%) is obtained according to the following formula:

[0096]

[0097] where B max is the maximum brightness. When d(t k ) < 50m, B i (t k ) = 0 corresponds to completely turning off the corresponding area.

[0098] Area division: The Micro-LED headlamp is divided into 5 areas, covering -30° to 30° along the horizontal direction, with each area being 12°. The vertical direction is not considered (currently, horizontal direction control can meet effective glare avoidance). Central angle:

[0099]

[0100] Then the weight where:

[0101] σ is the width of the control weight distribution;

[0102] is the horizontal angle of the oncoming vehicle, c x is the image center of the road image I(t k ), x i is the abscissa of the oncoming vehicle output by AMHA - YOLO, f x is the horizontal focal length of the camera.

[0103] The control module generates a PWM signal for driving the corresponding Micro - LED area to turn on / off or adjust the brightness based on the calculated B i (t k ).

[0104] The second type of control instruction is the headlight angle adjustment control instruction. The control module generates a servo motor control signal based on the headlight angle adjustment control instruction to adjust the overall headlight angle and ensure that the light beam tracks the lane curvature.

[0105] In the embodiment of the present invention, the control module runs a closed - loop control algorithm to generate a servo motor control signal:

[0106]

[0107] where: K p , K i , K d are PID parameters. In a preferred embodiment of the present invention, K p , K i , K d have parameter ranges as follows:

[0108] K p : 0.5–2.0, controlling the response speed to avoid overshoot;

[0109] K i : 0.01–0.1, eliminating the steady - state error and adapting to different vehicle speeds;

[0110] K d : 0.1–0.5, suppressing the angle oscillation to ensure smooth adjustment;

[0111] e(t) = θ f (t)-θ current (t).

[0112] Cooperatively optimize glare avoidance and lighting coverage according to the target coordinates and lane curvature.

[0113] The control module generates PWM signals and servo signals by analyzing the distance of oncoming vehicles and the lane curvature in real time, and collaboratively optimizes glare avoidance and lighting coverage to ensure the balance between glare avoidance and road lighting. The instructions of the calculation module are processed by the MCU and converted into PWM signals (brightness control) and servo signals (angle adjustment), which are transmitted to the execution module through the CAN bus.

[0114] The execution module receives the PWM signal (driving the Micro-LED area to turn on and off or adjust the brightness) and the servo signal (adjusting the headlight angle) from the control module to achieve precise beam control and dynamic angle adjustment. It drives the brightness adjustment of 10 areas of the Micro-LED headlight, and each area supports 0-100% brightness adjustment. Combined with the angle adjustment mechanism (the reflector bowl or gimbal driven by the servo motor), precise beam control is achieved. In the embodiment of the present invention, the PWM signal controls the Micro-LED array through a dedicated drive circuit, and the servo signal adjusts the reflector bowl / gimbal through the motor controller, with a response time <10ms.

Claims

1. An automatic control system for an adaptive Micro-LED headlight for vehicles, characterized in that, Through the collaborative work of the sensing module, computing module, control module and execution module, the intelligent adjustment of the Micro-LED headlight is realized, where: The sensing module is used to obtain road images, the distance to oncoming vehicles, relative speed, servo motor feedback information for adjusting the headlight angle, vehicle yaw rate, and the vehicle speed at a fixed frequency. Among them, the road image is obtained through a camera, and the relative speed is obtained through a millimeter-wave radar; A calculation module for calculating the fused lane angle θ f (t): θ f (t) = w I ·θ I (t) + w v ·θ v (t) Where: t is the time of the current sampling point; w I is the weight of camera data, w v is the weight of millimeter-wave radar data, and: under high visibility, w I > w v , and under low visibility, w I < w v ; d r (t) is the lateral distance between the vehicle and the road boundary, v(t) is the speed of the host vehicle, and ω(t) is the yaw rate of the vehicle; θ I (t) is the lane angle obtained through the AMHA-YOLO network, and the AMHA-YOLO network is based on the YOLO architecture introducing a multi-head attention mechanism, with a road image, the yaw angular velocity of the vehicle, the speed of the host vehicle, and the distance to the oncoming vehicle as inputs: The backbone network of the AMHA-YOLO network is used to extract multi-scale features of the input road image. At the attention layer of the backbone network, the input vehicle yaw rate, vehicle speed, and distance to oncoming vehicles are used as context to modulate the attention weights; The neck network of the AMHA-YOLO network aggregates multi-scale features and adds a lightweight attention module to the FPN upsampling path; The head network of the AMHA-YOLO network outputs lane detection boxes; The calculation module converts the lane detection box at the current moment t into world coordinates through the internal and external camera parameters, and fits a curve by combining the world coordinates obtained at all moments between the current moment t. The curvature of the curve is the lane angle θ I (t); The control module obtains a control instruction for controlling the brightness of each area of the Micro-LED based on the distance to oncoming vehicles and generates a PWM signal for driving the corresponding Micro-LED area to turn on / off or adjust the brightness; Meanwhile, the control module uses the fused lane angle θ f (t) and the servo motor feedback information as inputs, and adopts a closed-loop control algorithm to generate a servo motor control signal; The execution module receives the PWM signal and servo motor control signal output by the control module, drives all areas of the Micro-LED headlight to adjust the brightness, and drives the servo motor to perform dynamic angle adjustment of the Micro-LED headlight.

2. The automatic control system for a vehicle adaptive Micro-LED headlamp according to claim 1, wherein, The sensing module generates synchronous time points t at a fixed frequency f k , t k = t0 + k·Δt, where t0 is the initial time, Δt = 1 / f, and k is the time step index.

3. The automatic control system for a vehicle adaptive Micro-LED headlamp according to claim 2, characterized in that The perception module at each synchronization time point t k : Obtain the servo motor feedback information θ for adjusting the headlight angle current (t k ). Use the millimeter-wave radar to obtain the distance d(t k ) to the oncoming vehicle and the lateral distance d r (t) between the vehicle and the road boundary. Use the steering sensor to obtain the vehicle yaw rate ω(t k ).

4. The automatic control system for an adaptive Micro-LED headlight for vehicles according to claim 3, characterized in that, The perception module aligns the vehicle speed to the time axis t using the linear interpolation method k , and obtains the vehicle speed v(t k ).

5. The automatic control system for a vehicle adaptive Micro-LED headlamp according to claim 3, characterized in that, The perception module aligns the road image to the time axis t using nearest neighbor interpolation k , and obtains the road image I(t k ).

6. The automatic control system of a vehicle adaptive Micro-LED headlight according to claim 1, characterized in that, The backbone network adopts the CSPDarknet53 structure, which includes multiple layers of convolution and cross-stage partial connection. The attention layer is inserted after the convolution module of the backbone network. Each attention layer contains 4 parallel attention heads, which focus on lane edge texture, oncoming vehicle contour, pedestrian contour, and background noise suppression respectively.

7. The automatic control system of an adaptive Micro-LED headlamp for vehicles according to claim 1, characterized in that, Calculate the lane angle θ I When (t), the curve is a quadratic curve.

8. The automatic control system of a vehicle adaptive Micro-LED headlamp according to claim 1, characterized in that, The computing module uses Kalman filtering to dynamically correct the fused lane angle, the distance to oncoming vehicles, and the servo motor feedback information.

9. The automatic control system for a vehicle adaptive Micro-LED headlamp according to claim 1, wherein, If the closed-loop control algorithm adopted by the control module is the PID control algorithm, then the servo motor control signal is obtained by the following formula: Where: K p , K i , K d are PID parameters; e(t) = θ f (t) - θ current (t), θ current (t) is the feedback information of the servo motor.

10. The automatic control system of an adaptive Micro-LED headlamp for vehicles according to claim 1, characterized in that, When adjusting the brightness of all areas of the Micro-LED headlamp, the brightness B i (t) of the i-th Micro-LED area at the current moment t is obtained according to the following formula: where B max is the maximum brightness; w i (t) is the weight, and there is: Where: θ i is the central angle of the i-th Micro-LED region, σ is the width for controlling the weight distribution; is the horizontal angle of the oncoming vehicle.

Images