Complex road condition pre-judgment lighting system

Abstract

The invention discloses a complex road condition pre-judgment lighting system, and relates to the technical field of automobile lighting, the system comprises an environment sensing module, a positioning and map module, a central control module and an execution module, the environment sensing module collects multi-dimensional environment data; the positioning and map module is used for acquiring a vehicle pose and a high-precision map; the central control module generates an environment state vector based on environment parameter quantization logic, and performs dynamic weighted fusion on the multi-sensor data to generate target state estimation; meanwhile, a preview distance is calculated based on a vehicle motion state and system delay, a virtual prejudgment point is locked in a high-precision map, a feedforward control instruction is generated, and an execution module responds to the instruction and adjusts the brightness of a Micro LED array light source and the focal length of an optical projection lens set. Through cross-modal fusion and map space-time alignment, all-weather accurate perception and zero-delay light beam control are realized, and the problems of perception failure and dynamic illumination response lag in severe weather are solved.

Description

Technical Field

[0001] This application relates to the field of automotive lighting technology, and more particularly to a lighting system for predicting complex road conditions. Background Technology

[0002] With the rapid evolution of driver assistance technologies, automotive lighting systems are gradually shifting from passive lighting to intelligent and proactive systems. Adaptive High Beam (ADB) technology, as a core application of current intelligent lighting, has begun to rapidly penetrate the passenger vehicle market. This technology aims to detect vehicles and pedestrians ahead in real time using onboard cameras and other sensors, and automatically control the zonal illumination of the high beams according to the driving scenario, attempting to improve nighttime driving safety while ensuring the driver's visibility and avoiding glare for other road users.

[0003] While ADB (Adaptive Driving Beam) technology has made some progress in improving nighttime driving safety, the existing technical architecture still faces significant challenges when dealing with complex and ever-changing real-world road environments. Firstly, existing systems have limitations in environmental perception. Current mainstream solutions mostly rely on a combination of cameras and millimeter-wave radar, a perception architecture that is less robust in adverse weather conditions. For example, in heavy rain, dense fog, or heavy snow, optical lenses are easily obstructed or generate noise due to diffuse light reflection, leading to perception failure. Furthermore, conventional millimeter-wave radar lacks sufficient sensitivity to detect non-metallic, low-reflectivity targets such as non-motorized vehicles without reflective markings, dark-colored vehicles, or roadside debris, easily resulting in missed detection of dangerous targets. This can lead to high beams not being timely shielded or enhanced, creating safety hazards.

[0004] Secondly, existing lighting control logic generally suffers from response lag and insufficient accuracy. Traditional ADB systems employ a post-response logic of perception, judgment, and execution, typically resulting in a system delay of hundreds of milliseconds from the sensor capturing the target to the beam switching being completed. In high-speed driving scenarios, the vehicle has already undergone significant displacement within this delay time, causing the illuminated dark areas or obscured areas to fail to align accurately with the fast-moving target in real time, resulting in obscuration lag or brief glare. Simultaneously, traditional matrix LEDs have a limited number of zones, making it difficult to simultaneously achieve avoidance of fine targets and supplemental lighting for surrounding areas in complex road conditions, easily creating a conflict between avoiding glare and ensuring visibility.

[0005] Furthermore, the scenario adaptability and hardware maintenance costs of existing systems are also key factors restricting their widespread adoption. On the one hand, systems based on single visual logic rely excessively on structured features such as lane lines. In rural roads without lane lines, tunnel entrances and exits with sudden changes in lighting, or complex urban intersections, the system often fails to activate or makes incorrect judgments due to a lack of prior information. On the other hand, in pursuit of high integration, existing ADB modules typically integrate precision sensors, drive motors, and light sources into a solid package, resulting in complex structures and high maintenance costs. Once a partial failure occurs, the entire system often needs to be replaced, and the expensive initial hardware cost also limits the application of this technology in economy car models. Summary of the Invention

[0006] The purpose of this invention is to provide a complex road condition prediction lighting system that solves the problems of target misjudgment, glare risk and lighting blind spots caused by insufficient perception and system response lag in existing adaptive high beam systems under adverse weather and high-speed dynamic scenarios.

[0007] This invention provides the following solution:

[0008] This invention provides a complex road condition prediction lighting system, which mainly includes an environmental perception module, a positioning and mapping module, a central control module, and an execution module.

[0009] The environmental perception module is configured to collect environmental data surrounding the vehicle and the vehicle's own status data. This environmental data includes at least image data, millimeter-wave radar data, lidar point cloud data, rainfall data, and illuminance data to achieve comprehensive digital perception of the external environment. The positioning and mapping module is configured to acquire the vehicle's real-time latitude and longitude coordinates, attitude angles, and high-precision map data containing road geometry information, providing the system with prior road information capable of beyond-line-of-sight.

[0010] The central control module, as the core of the system's computation, communicates with both the environmental perception module and the positioning and mapping module. Its working principle is as follows: First, it receives environmental data and high-precision map data, and generates an environmental state vector through environmental parameter quantification logic calculations. Second, it uses the environmental state vector to perform weighted fusion calculations on multi-sensor data to generate a high-confidence target state estimate. Third, to address the control lag problem caused by system delays, it calculates the aiming distance based on the vehicle's motion state and the system's total response delay, determines virtual prediction points in the high-precision map data, and generates feedforward control commands based on the road characteristics at the virtual prediction points.

[0011] The execution module is electrically connected to the central control module and includes a left headlight assembly and a right headlight assembly. Each headlight assembly integrates a Micro LED array light source and an optical projection lens group. The execution module is configured to respond to feedforward control commands by dynamically adjusting the brightness of the pixel units in the Micro LED array light source and the focal length of the optical projection lens group, thereby achieving pixel-level reconstruction of the light pattern and physical adjustment of the beam divergence angle.

[0012] In a preferred embodiment, the system introduces an environmental parameter quantification mechanism. Specifically, the central control module receives signals and image data from an optical rain gauge sensor, calculates and outputs a rainfall intensity coefficient by comparing the normalized signal from the optical rain gauge sensor with the rain line density extracted from the image data; simultaneously, it receives image data, estimates atmospheric light intensity and medium transmittance by calculating the mean and variance of the dark channel brightness in the image data, and then calculates and outputs a fog concentration coefficient. The rainfall intensity coefficient and fog concentration coefficient are used as dynamic input variables of the environmental state vector to quantify the degree of interference of current meteorological conditions on the optical sensor.

[0013] In a preferred embodiment, the system employs dynamic confidence weight calculation logic based on environmental conditions. The central control module pre-stores baseline weights for cameras, millimeter-wave radar, and lidar. A suppression factor for optical sensors is calculated using a formula constructed through an S-shaped function model, combined with rainfall intensity and fog concentration coefficients. Based on this suppression factor, the system reduces the real-time weight of the camera and increases the real-time weight of the millimeter-wave radar by a preset ratio. Finally, the adjusted real-time weights of each sensor are normalized to generate the final weight vector for target state estimation. This logic ensures that in rainy or foggy weather, the system can automatically reduce the weight of severely interfered optical data and increase reliance on radar data with strong penetration capabilities.

[0014] In a preferred embodiment, the system performs cross-modal spatiotemporal fusion processing. The central control module uses system time as a reference and calculates interpolation or extrapolation results based on the motion velocity vectors of the targets detected by each sensor, unifying heterogeneous data from different timestamps to the same moment. It then uses a pre-calibrated extrinsic parameter matrix to calculate the transformation results of each sensor data to the vehicle coordinate system and determines whether the data from different sensors originate from the same physical target based on the projection overlap. For the same physical target that has been successfully associated, the final weight vector is used to perform a weighted average of the position observations from each sensor to calculate the fused target position vector.

[0015] In a preferred embodiment, the system possesses spatiotemporal alignment and virtual prediction capabilities to eliminate system latency. The central control module calculates the total system response latency by summing the time of perception latency, transmission latency, computation latency, and execution latency; it calculates the vehicle's travel distance within the response cycle as the pre-aiming distance using the vehicle's real-time longitudinal speed and acceleration, the total system response latency, and a preset safety buffer time; and it searches forward along the topological path of the lane centerline in the high-precision map data, locking the position where the path integral length equals the pre-aiming distance as the virtual prediction point.

[0016] In a preferred embodiment, the system generates feedforward control commands based on the road geometry features of the virtual prediction point. Specifically, the road curvature at the virtual prediction point is extracted, and the theoretical deflection angle of the beam center relative to the vehicle's longitudinal axis is calculated using the road curvature. This theoretical deflection angle is then mapped to the lateral index displacement of the pixel column in the Micro LED array light source, enabling the curved beam to follow the direction of travel. Simultaneously, the road slope difference is calculated using the slope values ​​between the virtual prediction point and the vehicle's current position. When the road slope difference is positive and exceeds a threshold, a negative pitch adjustment command is generated; when the road slope difference is negative and its absolute value exceeds a threshold, a positive pitch adjustment command is generated, achieving slope beam pitch compensation.

[0017] In a preferred embodiment, the system employs layer overlay technology to generate the light intensity distribution. The central control module constructs a virtual brightness matrix with the same resolution as the Micro LED array light source: a first global basic lighting layer is generated, whose brightness distribution is determined based on vehicle speed and road type; a second dynamic occlusion layer is generated by mapping the fused target position vector to the Micro LED array plane to calculate the occlusion area, and setting or reducing the pixel brightness value in the corresponding projection area; a third key enhancement layer is generated by increasing the pixel brightness value in the projection area of ​​identified pedestrians or traffic signs to a high brightness threshold. Finally, a light intensity distribution matrix is ​​generated by performing temporal smoothing filtering on the overlay virtual brightness matrix and sent to the execution module.

[0018] In a preferred embodiment, the system possesses a user behavior adaptive learning function. When a driver's manual headlight or height adjustment intervention is detected, an intervention log containing geographic location features and system state features is generated. The intervention log is then clustered geographicly. If the frequency of manual interventions within the same grid coordinate area exceeds a threshold, a high-beam suppression label is added to that grid coordinate area in the high-precision map data, achieving geofencing for blind spots in the algorithm. Furthermore, the average target distance during driver manual intervention is calculated, and the deviation is calculated between the average value and a preset trigger threshold. This deviation is then used to update the following-vehicle occlusion trigger threshold in the lighting control strategy, enabling personalized iteration of the control logic.

[0019] In a preferred embodiment, the system employs a modular design. The execution module includes a lamp housing and a universal mounting bracket located within it. The Micro LED array light source is encapsulated on a replaceable light source module, which is blind-mold connected to the drive control box via a floating connector with triaxial elastic floating margin. The optical projection lens group is encapsulated within a separate lens barrel, which is fixed to the front end of the replaceable light source module via a quick-release structure.

[0020] The above solution achieves the following beneficial technical effects:

[0021] The invention constructs a multimodal perception network by integrating LiDAR, millimeter-wave radar, and HDR night vision cameras, and achieves dynamic weight allocation by combining environmental parameter quantification logic. In complex lighting scenarios such as heavy rain, dense fog, or entering and exiting tunnels, the system can automatically reduce the weight of the interfered optical sensors and instead utilize the characteristic of LiDAR that is unaffected by light and rain / fog for accurate detection. Combined with an AI recognition model trained on complex scene datasets, it effectively solves the problem of missing detection of non-motorized vehicles without reflective markings or dark targets, ensuring perception stability under extreme weather conditions.

[0022] This invention utilizes high-precision map data and vehicle dynamics models to introduce a predictive feedforward control mechanism. By calculating the total system response delay and locking virtual prediction points, the system can extract road curvature and slope information and generate beam adjustment commands in advance before the vehicle physically reaches a curve or slope. This mechanism completely compensates for the inherent lag in sensor acquisition, algorithm calculation, and mechanical execution in the time dimension, eliminating blind spots in dynamic driving.

[0023] This invention combines high-resolution pixel control of Micro LED arrays with an electric zoom mechanism of optical lenses to achieve dual adjustment from the source to the optical path. The system can not only perform micron-level anti-glare masking for oncoming vehicles or pedestrians based on the fusion target position, but also physically adjust the lens position according to vehicle speed to change the beam divergence angle. Thus, in high-speed scenarios, it can physically converge light energy to increase the illumination distance, and in low-speed scenarios, it can expand the field of view to cover roadside risks, taking into account both the needs of long-distance lighting and wide field of view paving. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the system modules of the present invention.

[0025] Figure 2 This is a schematic diagram of the modular structure of the execution module (headlight) of this invention. Detailed Implementation

[0026] The technical solution 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] See attached document Figure 1 This invention provides a complex road condition prediction lighting system, which includes an environmental perception module, a central control module, an execution module, and a positioning and mapping module. These modules are connected via an onboard communication bus to achieve data interaction and command transmission. The onboard communication bus adopts a Controller Area Network (CAN) bus or an onboard Ethernet bus.

[0028] The environmental perception module is configured to collect environmental data surrounding the vehicle and the vehicle's own status data. The environmental perception module includes a forward-facing camera, millimeter-wave radar, lidar, rain sensor, and light sensor. The forward-facing camera is mounted on the upper inner side of the windshield or at the base of the rearview mirror. It is a high dynamic range (HDR) camera configured to collect red, green, and blue (RGB) image data in front of the vehicle, including lane line information, traffic sign information, and the outlines of luminous objects. The millimeter-wave radar is mounted on the front bumper or grille, configured to transmit and receive millimeter-wave signals, outputting the relative distance, relative speed, and azimuth angle data of targets ahead. The lidar is mounted on the roof or front bumper, configured to generate 3D point cloud data of the environment in front of the vehicle by emitting a laser beam and receiving the echo. The 3D point cloud data includes the spatial coordinates of targets and reflection intensity information.

[0029] The environmental perception module also includes a rain sensor and a light sensor. The rain sensor is installed in the windshield area and connected to the vehicle's wiper control system. It is configured to detect rainfall intensity and output a normalized rainfall signal. The light sensor is installed on the vehicle roof or dashboard area and is configured to detect ambient light intensity and output a illuminance value in lux.

[0030] The positioning and mapping module includes a Global Navigation Satellite System (GNSS) receiver, an Inertial Measurement Unit (IMU), and a high-precision map data storage unit. The GNSS receiver is configured to receive satellite signals to determine the vehicle's latitude and longitude coordinates. The IMU is configured to measure the vehicle's three-axis acceleration and angular velocity. The high-precision map data storage unit stores road geometry information, including lane centerline coordinates, road curvature, slope, tunnel location coordinates, and traffic rule attributes. The positioning and mapping module is configured to match the high-precision map data with the vehicle's real-time location and output road feature data within a preset distance ahead.

[0031] The central control module includes an Electronic Control Unit (ECU). The ECU comprises input interfaces, a processor, memory, and output interfaces. The input interfaces are connected to the environmental perception module and the positioning and mapping module, respectively, and are configured to receive RGB image data, radar data, point cloud data, environmental sensor signals, and positioning map data. The processor is configured to run multi-sensor fusion algorithms and lighting control strategy algorithms to generate lighting control commands. The output interfaces are connected to the execution module and are configured to send lighting control commands.

[0032] The execution module includes a left headlight assembly and a right headlight assembly. Each headlight assembly contains a microlight-emitting diode (MicroLED) array light source, a driver circuit board, and an optical projection lens assembly. The MicroLED array light source is composed of... OK It consists of a column of LED pixel units, in which and All values ​​are positive integers. The brightness of each LED pixel unit can be adjusted independently. The driver circuit board is connected to the central control module and is configured to receive lighting control commands and independently control the on / off state and current magnitude of each pixel unit in the MicroLED array via pulse width modulation (PWM) signals.

[0033] The optical projection lens assembly is located in the light emission path of the MicroLED array light source and is configured to project the light emitted by the MicroLED array onto the road surface in front of the vehicle. The optical projection lens assembly includes an electric focusing mechanism driven by a stepper motor, configured to move the lens position along the optical axis according to control commands, thereby changing the divergence angle of the illumination beam.

[0034] In this embodiment, there is a one-to-one geometric mapping relationship between the pixel distribution on the MicroLED array light source and the three-dimensional space in front of the vehicle. The central control module stores pre-calibrated optical system parameters, including the equivalent focal length of the optical projection lens group, the principal point coordinates, and the rotation matrix and translation vector of the MicroLED array plane relative to the vehicle coordinate system. The central control module is configured to use the aforementioned optical system parameters, based on the principle of perspective projection, to map the three-dimensional spatial position of the obstacle detected by the environmental perception module in the vehicle coordinate system into a two-dimensional pixel coordinate region on the MicroLED array plane. Through this mapping operation, the central control module determines a specific set of LED pixel units corresponding to the obstacle position and generates instructions to turn off or reduce the brightness of the pixel units in that set, thereby forming a dark area at the obstacle.

[0035] The environmental parameter acquisition and quantification method provided by this invention is executed by the environmental perception processing unit in the central control module. This processing unit is configured to perform feature extraction and numerical processing on data from multiple sources and output a standardized environmental state vector.

[0036] Firstly, to quantify the rainfall environment, the system employs a dual verification mechanism combining data from an optical rain sensor and image data from a forward-facing camera. The optical rain sensor, installed on the windshield, consists of an infrared emitter and receiver, operating on the principle of total internal reflection. In a dry state, infrared light undergoes total internal reflection on the windshield's outer surface and is received by the receiver; when raindrops are present on the windshield surface, some light is refracted and lost, leading to attenuation of the received light intensity. The central control module receives the voltage or frequency signal output from the rain sensor, the amplitude of which is negatively correlated with the water coverage on the windshield surface. The central control module internally stores a preset interference threshold table, mapping the received real-time signal values ​​to continuous values ​​between 0 and 1, defined as the rainfall intensity coefficient. Here, 0 represents a dry, rainless state, and 1 represents a heavy rain causing sensor saturation.

[0037] To eliminate false triggering of the rain sensor by wiper operation or dirt such as bird droppings, the central control module simultaneously extracts rain streak features from the image captured by the forward-facing camera. The processor uses an edge detection operator to analyze the high-frequency components of the image, identifying pixel stripes with specific tilt angles and lengths, i.e., rain streaks. The system calculates the rain streak density per unit time. Only when the rainfall intensity coefficient of the optical rain sensor exceeds a preset threshold, and the rain streak density identified in the image increases synchronously, does the system confirm the validity of the rainfall intensity coefficient and use it as the final environmental factor input.

[0038] Secondly, for quantification in hazy or low-visibility environments, the system is entirely based on image signal processing from a forward-looking camera, without relying on an additional visibility meter. The central control module runs an image dehazing analysis algorithm based on Dark Channel Prior. The processor divides each frame of the acquired RGB image into multiple local patches and calculates the minimum pixel value channel for each patch in the non-sky region. Under clear, fog-free conditions, this minimum pixel value approaches zero; however, in foggy conditions, due to atmospheric light scattering, the pixel value of the dark channel increases significantly, and the increase is proportional to the fog concentration.

[0039] The central control module estimates the atmospheric light intensity and medium transmittance of the current scene by statistically analyzing the mean and variance of the dark channel brightness of the entire image. The system normalizes the reciprocal of the medium transmittance to generate a fog concentration coefficient. This coefficient is also quantized into a value between 0 and 1, with a higher value indicating lower visibility and denser fog. In addition, the processor calculates the overall contrast histogram of the image as an auxiliary verification indicator to prevent misjudgments caused by capturing large areas of white objects (such as white walls or snow).

[0040] Finally, for the quantification of ambient light, the system directly collects the illuminance value of the environment using a light sensor. The central control module performs analog-to-digital conversion on the collected analog current signal to obtain the absolute illuminance value in lux. To match the input requirements of subsequent fusion algorithms, the central control module uses piecewise linear or logarithmic transformation to map the wide dynamic range of absolute illuminance values ​​to a normalized illuminance intensity range. This illuminance intensity data is used to determine whether the vehicle is in a tunnel, on a road without streetlights at night, or during the twilight transition period, thereby deciding whether to activate the basic enabling logic of the adaptive high beam system.

[0041] Through the above processing, the central control module outputs an environmental state vector in real time, which includes rainfall intensity coefficient, fog concentration coefficient and normalized light intensity. This vector serves as the dynamic input variable for the subsequent multi-sensor fusion weight allocation algorithm.

[0042] The dynamic confidence weight calculation method provided by this invention is executed by the fusion algorithm unit within the central control module. This unit is configured to adjust the confidence level, i.e., the weight coefficients, of the camera, millimeter-wave radar, and lidar in the data fusion process in real time based on the environmental state vector output by the environmental perception processing unit.

[0043] The central control module's memory stores a set of baseline weight vectors: Each corresponds to a camera Millimeter-wave radar and lidar Initial confidence values ​​under ideal clear weather conditions. Typically, due to the camera's highest angular resolution and texture recognition capabilities, its baseline weight is [missing information - likely a specific value]. The setting is the highest.

[0044] The fusion algorithm unit first determines the input rainfall intensity coefficient. and fog concentration coefficient The suppression factor for optical sensors is calculated. Considering that sensor performance degradation with environmental deterioration typically exhibits a non-linear S-shaped characteristic (i.e., performance remains stable initially but drops sharply beyond a certain boundary), the processor uses the Logistic function (S-shaped function) to construct the calculation model for the suppression factor. Suppression factor for rainfall environments is also calculated. And inhibitors for foggy environments The calculation formula is as follows:

[0045] ;

[0046] ;

[0047] in, and Defined as a sensitivity coefficient, it is used to control the response rate of an inhibitory factor to changes in environmental parameters; a larger value indicates a lower sensitivity coefficient. The value means that once the environmental parameters exceed the threshold, the inhibition factor will increase rapidly, simulating the steep degradation characteristics of sensor performance. and Defined as environmental transition thresholds, these correspond to the critical rainfall and critical fog concentration at which camera performance begins to decline significantly, respectively. The central control module allows for calibration via an external interface. Value and The values ​​are updated to adapt to the hardware characteristics of different camera models.

[0048] After obtaining the suppression factor, the fusion algorithm unit performs weight redistribution calculation. For the forward-looking camera, its weights... The attenuation decreases as the inhibition factor increases, and the calculation logic is configured to take the more significant effect of either rainfall or fog as the dominant attenuation source:

[0049] ;

[0050] For millimeter-wave radar, due to its longer wavelength, it has an extremely strong ability to penetrate rain and fog, and its weight... Not only must the baseline value be maintained, but a compensatory boost is also needed to take over detection control should the optical sensor fail. The calculation formula includes a compensation gain coefficient.

[0051] ;

[0052] For lidar, it offers superior contour detection capabilities compared to cameras in foggy and low-light conditions, but it is affected by water droplet scattering in heavy rain. Therefore, its weight... The adjustment strategy is configured as follows: in foggy or low light conditions (by... During heavy rain (when the camera is in a dominant position), the weight is increased to compensate for camera loss, while during torrential rain (when the camera is in a dominant position), the weight is increased to compensate for camera loss. When in control, maintain the baseline or slightly reduce it.

[0053] Finally, the fusion algorithm unit normalizes the three original weight values ​​obtained from the above calculations to ensure that the sum of the weights of all sensors is 1. The normalized final weight vector will be directly used in the subsequent target state estimation equation as the coefficient input for the Kalman filter or weighted average algorithm. Through this dynamic calculation process, the system can automatically shift the perception dependence to millimeter-wave radar in heavy rain, or use a combination of lidar and millimeter-wave radar for perception in dense fog, avoiding high beam control failure caused by blinding a single camera.

[0054] For lidar, it offers superior contour detection capabilities compared to cameras in foggy and low-light conditions, but it is affected by water droplet scattering in heavy rain. Therefore, its weight... The adjustment strategy is configured as follows: in foggy or low light conditions (by... During heavy rain (when the camera is in a dominant position), the weight is increased to compensate for camera loss, while during torrential rain (when the camera is in a dominant position), the weight is increased to compensate for camera loss. When in control, maintain the baseline or slightly reduce it.

[0055] Finally, the fusion algorithm unit normalizes the three original weight values ​​obtained from the above calculations to ensure that the sum of the weights of all sensors is 1. The normalized final weight vector will be directly used in the subsequent target state estimation equation as the coefficient input for the Kalman filter or weighted average algorithm. Through this dynamic calculation process, the system can automatically shift the perception dependence to millimeter-wave radar in heavy rain, or use a combination of lidar and millimeter-wave radar for perception in dense fog, avoiding high beam control failure caused by blinding a single camera.

[0056] The cross-modal spatiotemporal fusion processing method provided by this invention is executed by the fusion algorithm unit within the central control module. This processing aims to unify heterogeneous sensor data from different frequencies and spatial coordinate systems under the same spatiotemporal reference. Utilizing the aforementioned dynamic confidence weight calculation results, the output unit is configured to unify the time-asynchronous and spatially independent data from heterogeneous sensors into the vehicle coordinate system, and generate the final target state estimate based on the aforementioned calculated dynamic weights.

[0057] First, the fusion algorithm unit performs time synchronization and motion compensation processing. Due to the different sampling frequencies and data transmission delays of the camera, millimeter-wave radar, and lidar (e.g., camera data is typically 30Hz, millimeter-wave radar is 15-20Hz, and lidar is 10Hz), the arrival timestamps of the data from each sensor are inconsistent. The central control module establishes a unified timeline based on the system time of the computing unit.

[0058] When the system receives the latest frame of image data The processor backtracks to the most recent radar data frame in the cache. and lidar data frames .because and , Since there is a time difference, the fusion algorithm unit uses the velocity vector information carried by each target itself and performs interpolation or extrapolation calculations using a linear motion model to extrapolate the target position coordinates detected by radar and lidar to the current image frame. This step eliminates the temporal ghosting error caused by the high-speed movement of the vehicle and the resulting discrepancies in the data from different sensors.

[0059] Secondly, the fusion algorithm unit performs spatial coordinate transformation and ROI association matching. The raw output data of each sensor is based on its own independent sensor coordinate system. The central control module stores pre-calibrated extrinsic parameter matrices (rotation matrix and translation vector) to describe the rigid body transformation relationship of each sensor relative to the vehicle's rear axle center coordinate system (vehicle coordinate system). The processor uniformly transforms all time-synchronized target position coordinates to the vehicle coordinate system.

[0060] After unifying the coordinate system, the fusion algorithm unit performs data association. The processor maps the 3D target points detected by millimeter-wave radar and lidar onto the 2D image plane through the camera's intrinsic parameter projection matrix. The system calculates the overlap (Intersection over Union, Euclidean distance) or Euclidean distance between the radar or lidar projection points and the visual target detection box (Region of Interest, ROL) identified by the camera. When the projection point falls inside the visual detection box or the distance is less than a preset association threshold, the system determines that the data from the different sensors originates from the same physical target (e.g., the same oncoming vehicle) and establishes a unique global target ID.

[0061] Finally, for the same target that has been successfully associated, the fusion algorithm unit performs weighted state fusion estimation. This is to obtain the most accurate position information (lateral distance) of the target in the vehicle coordinate system. and longitudinal distance The system utilizes the calculated dynamic weights. , , The observations from each sensor are weighted and fused.

[0062] Let the target position vector observed by the camera, transformed into the vehicle coordinate system, be... The millimeter-wave radar observation vector is The lidar observation vector is The final fused target position vector The calculation formula is as follows:

[0063] ;

[0064] The physical meaning of this formula is that in clear weather, high-weighted camera data dominates the fusion result, ensuring high resolution for vehicle headlight contour recognition; while in heavy rain or dense fog, as... The suppressed factor decreased, and Improved, fusion result The system will automatically and smoothly transition to radar-based data to ensure that the output target position coordinates do not change or are lost due to visual failure. The fused target position vector... The data is then fed into the lighting control strategy unit to calculate the ADB matrix pixel areas that need to be masked. For independent targets that cannot be associated (e.g., unlit obstacles detected only by radar), the system directly retains the high-confidence data from that single sensor and treats it as an independent obstacle.

[0065] The pre-aiming distance and spatiotemporal alignment method provided by this invention is executed by the lighting control strategy unit within the central control module. This method aims to eliminate the inherent physical delays in the system caused by sensor exposure, data transmission, algorithm calculations, and actuator actions, ensuring that beam adjustment commands take effect synchronously when the vehicle actually arrives at a specific road feature point.

[0066] The central control module first performs calibration and calculation of the total system response delay. (Total system response delay) Defined as the time difference between a change in the environment (or a change in vehicle position) and the actual change in the MicroLED light pattern. This delay is not a single value, but rather the sum of the times consumed by each component in the hardware chain. The central control module stores a system delay parameter table, which includes:

[0067] Perceived delay This includes camera exposure time, ISP processing time, and radar point cloud generation time.

[0068] Transmission delay The time it takes for data to be transmitted to the central control module via the vehicle's Ethernet or CAN bus;

[0069] computation delay The time consumed by the central processing unit to run the fusion algorithm and control strategy;

[0070] Execution delay The time it takes for the drive circuit to respond to the PWM signal and for the MicroLED to complete current build-up.

[0071] Total system response delay The calculation is the sum of the delays of the above-mentioned individual items. During the vehicle's factory calibration phase, this value is written into non-volatile memory.

[0072] The lighting control strategy unit acquires real-time longitudinal motion data of the vehicle via the vehicle's CAN bus, including real-time vehicle speed. (unit: meters per second) and longitudinal acceleration (Unit: m / s²). To ensure control safety redundancy, the system also incorporates a preset safety buffer time. This time is used to compensate for distance estimation errors that may be caused by slippery road surfaces or tire slippage.

[0073] Based on kinematic principles, the processor calculates the physical distance the vehicle will travel within the next response cycle, i.e., the pre-aiming distance. To improve prediction accuracy under acceleration and deceleration conditions, the computational model employs second-order equations of motion that include acceleration terms:

[0074] ;

[0075] This formula ensures that when a vehicle decelerates rapidly on a highway (such as when entering a ramp) or accelerates rapidly (such as when overtaking), the system's calculated advance distance dynamically matches the actual driving trajectory, rather than being a linear inference based solely on the current speed.

[0076] Obtain the pre-aiming distance Then, the central control module performs a spatiotemporal alignment operation, which involves finding virtual prediction points in the spatial coordinate system of the high-precision map. The processor uses the vehicle's current GNSS absolute coordinates. Starting from the center line of the lane in the high-precision map, the system searches forward along the topological path. The system calculates the path integral length; when the cumulative length along the road's geometric trajectory equals... At that time, the location point is locked as a virtual prediction point.

[0077] Subsequently, the positioning and mapping module extracts virtual prediction points. The road geometry attribute data at the location, not the current vehicle position. The extracted attribute data includes the road curvature radius, road slope, presence of tunnel entrance markers, and whether the location is within a no-high-beam zone at the predicted point.

[0078] Based on the aforementioned spatiotemporal alignment model, the system establishes a future time window. The lighting control commands generated by the central control module are based on virtual prediction points. The instruction is generated based on road features, but the timing of its delivery is strictly controlled. The processor stores the instruction in the execution queue and, according to... A countdown is triggered. When the countdown ends, the vehicle has physically traveled to... Position, at this time the MicroLED beam is adjusted synchronously (such as the beam deflection in the curve or the beam pressure in front of the tunnel), thus realizing zero-delay spatial synchronous control at the physical level.

[0079] The curve and terrain prediction method provided by this invention is executed by the lighting control strategy unit within the central control module. This method is configured to utilize the road geometry attributes output by the positioning and mapping module to calculate the theoretical beam adjustment amount and generate feedforward control commands before the vehicle enters a curve or undulating road section. The lighting control strategy unit first executes horizontal curve prediction control. The processor extracts virtual prediction points from high-precision map data. Road curvature value at the location Curvature value Defined as the radius of the road curve at that point. The reciprocal of (i.e.) To ensure the driver's line of sight always covers the extended trajectory inside the curve, the system needs to calculate the theoretical deflection angle of the beam center relative to the vehicle's longitudinal axis.

[0080] Based on geometric relationships, the processor uses the following mathematical model to calculate the theoretical deflection angle:

[0081] ;

[0082] in, This refers to the aiming distance. The physical meaning of this formula is: as the vehicle speed increases (leading to...) (Increase) or the road curvature becomes sharper (leading to) As the beam deflection angle increases, the required beam deflection angle exhibits a non-linear growth.

[0083] Calculated The central control module then compares the optical coverage of the current execution module with that of the current execution module. The MicroLED array in the execution module is divided into multiple vertical pixel columns in the horizontal direction, each column corresponding to a specific horizontal azimuth angle. The processor then continuously adjusts the deflection angles... Discretization is mapped to the pixel column index displacement of the MicroLED array. For example, if the calculation shows that the beam needs to be deflected 15 degrees to the left, the processor will not rely on mechanical rotation, but will directly generate instructions to light up the high-brightness pixel area on the left edge of the MicroLED array, while turning off the right edge area, thereby achieving electronic virtual deflection of the beam's center of gravity in milliseconds. This electronic deflection eliminates the physical delay of traditional mechanical adaptive front lighting (AFS).

[0084] Secondly, the lighting control strategy unit performs vertical terrain slope prediction control. The processor extracts the road slope of the current vehicle position from the high-precision map. Road slope at virtual prediction point A positive slope value indicates an uphill slope, and a negative value indicates a downhill slope. The system focuses on the rate of change of the road's longitudinal profile, i.e., the slope difference. .

[0085] when When the value is positive and exceeds a preset threshold (e.g., when a vehicle is about to enter an uphill section or exit a downhill section), the front of the vehicle tends to tilt upwards relative to the road surface, causing the high beam cutoff line to rise, which may dazzle oncoming vehicles at a distance. At this time, the processor generates a negative pitch adjustment command, driving the focusing motor in the optical projection lens group to fine-tune the optical axis of the lens downwards, or controlling the MicroLED array to cut off the top row of pixels, actively lowering the cutoff line height.

[0086] Conversely, when When the value is negative and its absolute value exceeds the threshold (e.g., when a vehicle is about to pass the crest of a hill and enter a downhill section, i.e., a convex road section), the front of the vehicle tilts down relative to the road surface, resulting in a shortened headlight illumination distance. The processor generates a positive pitch adjustment command to drive the lens optical axis upward or illuminate the pixels in the higher rows of the MicroLED array to compensate for the illumination distance loss caused by the terrain, ensuring that the road surface on the downhill section is still illuminated when the vehicle passes the crest of a hill.

[0087] Finally, the central control module packages the calculated horizontal deflection and vertical pitch commands into a feedforward control package. This control package is not immediately sent to the drive circuit board; instead, it is marked with a precise effective timestamp. This effective timestamp equals the current system time plus a pre-calculated countdown. When the system clock reaches this effective timestamp, the vehicle has physically reached the entry point of a curve or the point of change of slope, and the execution module synchronously performs beam adjustment. Through this spatiotemporally locked predictive logic, the system completes pre-compensation for the beam pattern before the vehicle's attitude actually changes, ensuring absolute synchronization between the illumination field of view and the road direction.

[0088] The user behavior adaptive learning method provided by this invention is executed by a machine learning unit within a central control module. This method is configured to capture the driver's manual intervention operations on the automatic lighting control system, converting the driver's personalized preferences into adjustments to the system's control parameters, thereby addressing the adaptability bias of general algorithms in specific scenarios or user habits.

[0089] The central control module first captures and structures the features of human intervention events. When the adaptive high beam (ADB) system is in automatic activation mode, the processor monitors the status signals of the vehicle's combination switch (dimming lever) and the human-machine interface (headlight height adjustment knob or touchscreen menu) in real time. Once the system detects that the driver has manually turned off the high beams, manually flashed the headlights, or manually adjusted the beam height, it determines that a negative feedback intervention event has occurred.

[0090] The processor immediately freezes all system state data at the current moment and generates an intervention log with a timestamp. This log contains the following features:

[0091] Geographic location characteristics: the vehicle's current latitude and longitude coordinates, road ID, and direction of travel;

[0092] Perceive environmental characteristics: number of currently identified targets, target distance, ambient light level, and weather conditions (rain / fog coefficient);

[0093] System status characteristics: The beam matrix status and headlight pitch angle values ​​output by the system at the moment of intervention. The above intervention logs are stored in a user behavior database on non-volatile memory.

[0094] Secondly, the machine learning unit performs scene memory learning based on geofencing. This logic aims to address algorithm failures in specific road segments (e.g., highly reflective road signs at intersections that are easily misidentified as vehicles). The processor periodically scans the user behavior database and performs cluster analysis on geographic location features. The system divides the map into grids of a preset size (e.g., 10m x 10m).

[0095] If the number of manually switched-off high beam events recorded by the system exceeds a preset frequency threshold (e.g., 3 times) within the same grid coordinate area, the processor will mark a high beam suppression tag in the grid data of the high-precision map. When the vehicle travels back to the same grid coordinate area in the same direction as recorded, regardless of whether the perception system detects an obstacle, the lighting control strategy unit will prioritize reading the map tag and forcibly execute high beam switching or beam avoidance. This logic implements empirical patching to fix specific algorithm blind spots.

[0096] Finally, the machine learning unit performs online adaptive adjustments to the control parameters. This logic is designed to accommodate the different psychological tolerance levels of drivers to glare distance. Taking a following scenario as an example, the general algorithm is set to activate beam blocking 200 meters from the vehicle in front. If a driver is accustomed to manually turning off the high beams 300 meters from the vehicle in front, the system will record the perceived distance parameters corresponding to this behavior.

[0097] The processor uses a moving average algorithm or a Kalman filter algorithm to calculate the average target distance during each manual intervention by the driver, and calculates the deviation between this average and a preset threshold of the system. When the deviation is statistically significant (i.e., the variance is less than the preset convergence value), the processor generates a global correction coefficient and updates the following occlusion trigger threshold in the lighting control strategy.

[0098] Similarly, for headlight height adjustment, if the user frequently manually lowers the headlight height under specific loads or speeds, the system will learn the pitch angle offset under those conditions. In subsequent operation, once the vehicle enters the same load or speed conditions, the execution module will automatically add this learned offset to the basic control commands, pre-adjusting the headlight beam to a comfortable height for the user. Through this process, the system achieves an iterative evolution from factory-preset logic to user-personalized logic.

[0099] The beam matrix generation and execution method provided by this invention is executed collaboratively by the lighting control strategy unit within the central control module and the drive circuit board of the execution module. This process is configured to convert the abstract data output from environmental perception and map prediction into specific drive current values ​​for each pixel in the MicroLED array, achieving high-precision beam pattern reconstruction.

[0100] The central control module first performs a projection mapping from three-dimensional space to a two-dimensional pixel plane. The processor receives a list of fused target states (including the vehicle coordinates, size, and type of the target) from the fusion algorithm unit and road geometry parameters from the prediction logic unit. Using pre-stored intrinsic parameter matrices (focal length, principal point) and extrinsic parameter matrices (installation position, rotation angle) of the optical system, the processor calculates the projection area of ​​each spatial target on the MicroLED array plane through perspective projection transformation operations.

[0101] To eliminate projection jitter caused by system errors and vehicle bumps, the processor superimposes a dynamic safety margin around the calculated projection area. The pixel width of this safety margin is positively correlated with the vehicle's current speed and the degree of road bumpiness (calculated from IMU vertical acceleration data). For example, when the vehicle speed is high or the road bumpiness is significant, the processor automatically expands the pixel range of the safety margin to ensure that the target remains within the shaded or illuminated area.

[0102] Subsequently, the lighting control strategy unit performs the synthesis calculation of the light intensity distribution matrix. The processor constructs a matrix in memory that matches the physical resolution of the MicroLED array. A consistent virtual brightness matrix. Each element in the matrix corresponds to the target brightness value of a pixel unit (grayscale level is typically 0-255 or 0-1024). The generation of the light intensity distribution matrix follows the layer overlay logic of global base lighting, dynamic occlusion, and focused enhancement:

[0103] The first layer is the global base lighting layer: the processor generates a basic high beam pattern based on the current vehicle speed and road type (provided by map data). In high-speed scenes, the processor increases the brightness value of pixels in the center area of ​​the matrix and narrows the horizontal illumination range to increase the illumination distance; in urban or curved scenes, it reduces the center brightness and illuminates the pixels on both sides of the matrix to widen the field of view.

[0104] The second layer is the dynamic masking layer: For any detected oncoming vehicle or following target, the processor forcibly sets the element values ​​in the corresponding projection area (including the safety boundary) of the virtual brightness matrix to zero or sets it to low-beam level. This operation is equivalent to removing the pixels that precisely correspond to the vehicle's position from the complete high beam spot, forming a moving dark area channel.

[0105] The third layer is the ROIHighlightLayer: For pedestrians, non-motorized vehicles, or roadside traffic signs, the processor identifies their projection coordinates in the matrix and sets the pixel brightness value in that area to a higher than the high brightness threshold of the basic lighting layer (e.g., increasing it to 120% of the rated brightness).

[0106] After completing the three-layer overlay calculation, the processor performs temporal smoothing filtering on the final generated brightness matrix. To prevent beam flickering caused by abrupt changes in target detection, the processor compares the difference between the current frame matrix and the previous frame matrix, limiting the rate of brightness change of the same pixel between consecutive frames. That is, the brightening and dimming process of a pixel is not a step change, but follows a preset ramp function to achieve a smooth visual transition.

[0107] Finally, the execution module performs physical drive execution based on the processed brightness matrix. The driver circuit board receives a serial data stream from the central control module. The multi-channel constant current driver chip on the circuit board modulates the duty cycle of the pulse width modulation (PWM) signal output to the MicroLED unit according to the brightness value corresponding to each pixel.

[0108] For the electrically adjustable focusing lens positioned in front of the MicroLED array, the central control module synchronously outputs stepper motor control pulses. When the vehicle speed exceeds a preset high-speed threshold (e.g., 100km / h), the drive motor adjusts the lens position to bring it closer to the MicroLED array plane, reducing the light divergence angle and thus converging light energy at the physical optics level to enhance the center light intensity to meet the needs of ultra-long-distance lighting. When the vehicle speed is below the medium-speed threshold (e.g., 40km / h) or a large-curvature curve is detected, the drive motor reverses the lens position to increase the divergence angle, achieving wide-angle road lighting at close range.

[0109] Through the combination of the aforementioned pixel-level control and zoom lens, the system can simultaneously achieve multi-target parallel processing of anti-glare for oncoming vehicles (pixel deactivation), high-brightness warning for pedestrians on the roadside (pixel enhancement), and curve lighting compensation (lens zoom and edge pixel illumination).

[0110] See attached document Figure 2 The modular and maintainable structural design provided by this invention aims to achieve compatibility and interchangeability of hardware with different configuration levels through standardized mechanical and electrical interfaces, and allows for independent replacement and maintenance of core light-emitting modules and optical components, thereby reducing the total life cycle cost.

[0111] The physical structure of the execution module mainly consists of a lamp housing, a universal mounting bracket, replaceable light source modules, replaceable optical modules, a drive control box, and a heat dissipation system. The lamp housing, serving as the external protective boundary, is injection molded from polypropylene (PP) or polycarbonate (PC), and has a removable maintenance cover at the rear. The maintenance cover is connected to the housing via a rubber sealing ring and bolts, ensuring that the internal waterproof and dustproof rating of the lamp meets automotive-grade standards (such as IP67).

[0112] The universal mounting bracket, located inside the luminaire housing and made of die-cast aluminum or magnesium-aluminum alloy, serves as structural support and aids in heat dissipation. This bracket features a pre-installed array of standardized locating pin holes and threaded holes. Regardless of whether the system is configured with a basic matrix LED module (dozens of pixels) or a flagship MicroLED module (tens of thousands of pixels), the mounting hole positions and dimensions of the physical substrate adhere to a unified mechanical interface standard, thus achieving hardware compatibility on the same luminaire platform.

[0113] The replaceable light source module includes a printed circuit board (PCB) and its packaged light-emitting chip. The back of the PCB is coated with a phase-change thermally conductive material and is secured to the thermally conductive surface of a universal mounting bracket by screws. For electrical connections, the light source module does not use traditional soldered leads but instead features a board-to-board floating connector. When the light source module is pushed into its mounting position and locked, the floating connector automatically engages with the female connector extending from the drive control box in a blind-mating manner. This connector has elastic floating margins in the X, Y, and Z axes to absorb stresses caused by vehicle vibrations and thermal expansion and contraction, preventing poor contact.

[0114] The replaceable optical module comprises a primary collimating lens and a secondary projection lens group. The entire optical module is encapsulated within a separate dustproof lens barrel. The bottom of the lens barrel features a quick-release clip structure or a precision threaded adjustment mechanism, aligned with the front of the light source module. For the high-end version with electric focusing, the stepper motor and transmission gears are integrated inside the optical module's lens barrel. The central control module controls this motor via a separate LIN bus interface or a dedicated drive harness. If the optical module suffers lens scratches or motor failure, maintenance personnel only need to open the rear cover of the headlight fixture and unscrew the retaining ring to remove and replace the optical component individually, without needing to scrap the entire headlight assembly.

[0115] The cooling system is designed with a split architecture combining active and passive cooling. The back of the universal mounting bracket is thermally coupled to an Ethernet heat pipe or vapor chamber, which extends to the heat sink fins on the outside of the luminaire housing. In the high-power operation mode of the MicroLED module, a waterproof fan located behind the heat sink fins activates, creating forced convection. The fan unit is also designed as a modular plug-in, fixed to a dedicated air duct opening on the outside of the luminaire housing via a slot, supporting independent disassembly and maintenance.

[0116] To enable automatic identification and configuration of different hardware modules, the driver control box's circuit interface includes a hardware coding circuit. This circuit consists of a set of voltage-limiting resistors or an erasable programmable memory (EEPROM). When a new light source module or optical module is connected to the system, the central control module reads the voltage value of a specific pin or the ID code in the memory. Based on the read hardware ID, the processor automatically retrieves the corresponding drive parameters (such as maximum allowable current and voltage threshold) and optical calibration parameters (such as a pixel mapping table) from the database. If it is detected that a module of a different specification has been replaced (e.g., upgrading from a low-end to a high-end module), the system automatically updates the control strategy and unlocks the corresponding advanced functions (such as higher resolution projection), achieving plug-and-play functionality at the hardware level.

[0117] The structural design of this embodiment transforms the traditional one-piece packaged automotive headlight into a split assembly. In the event of a minor collision causing lens breakage, or prolonged use leading to excessive LED light decay, only the damaged sub-module needs to be replaced. This design reduces maintenance costs and resource waste, while also reserving physical and electrical interfaces for future hardware upgrades.

[0118] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A complex road condition pre-judgment lighting system, characterized by, The application relates to a vehicle headlight control system, comprising: an environment perception module configured to collect environment data of a vehicle periphery and state data of the vehicle itself, the environment data at least including image data, millimeter wave radar data, laser radar point cloud data, rainfall data and illumination data; a positioning and map module configured to obtain real-time latitude and longitude coordinates, attitude angles and high-precision map data containing road geometric information of the vehicle; a central control module in communication connection with the environment perception module and the positioning and map module, configured to receive the environment data and the high-precision map data, calculate an environment state vector through environment parameter quantization logic, generate a target state estimation through weighted fusion calculation of multi-sensor data by using the environment state vector, calculate a preview distance through vehicle motion state and system delay, determine a virtual preview point in the high-precision map data, and generate a feedforward control instruction according to road features of the virtual preview point; an execution module in electrical connection with the central control module, comprising a left front headlight assembly and a right front headlight assembly, each of the front headlight assemblies comprising a micro light-emitting diode (Micro LED) array light source and an optical projection lens group, and configured to adjust brightness of a pixel unit in the Micro LED array light source and focal length of the optical projection lens group in response to the feedforward control instruction.

2. The complex road condition pre-judgment lighting system according to claim 1, characterized in that, The central control module is configured to perform the environment parameter quantization and specifically configured to: receive signals of an optical rainfall sensor and the image data, calculate an output rainfall intensity coefficient by comparing a normalized signal of the optical rainfall sensor with rain line density extracted from the image data; receive the image data, calculate atmospheric light intensity and medium transmission rate through mean value and variance of dark channel brightness in the image data, and further calculate an output fog concentration coefficient; use the rainfall intensity coefficient and the fog concentration coefficient as dynamic input variables of the environment state vector.

3. The complex road condition pre-judgment lighting system according to claim 2, characterized in that, The central control module is configured to perform dynamic confidence weight calculation based on the environment state vector and specifically configured to: prestore reference weights of a camera, reference weights of a millimeter wave radar and reference weights of a laser radar; calculate an inhibition factor for optical sensors through a formula constructed by an S-shaped function model and the rainfall intensity coefficient and the fog concentration coefficient; reduce a real-time weight of the camera according to the inhibition factor, and increase a real-time weight of the millimeter wave radar by a preset proportion; calculate a final weight vector for the target state estimation through normalization processing of the adjusted real-time weights of each sensor.

4. The complex road condition pre-judgment lighting system according to claim 3, characterized in that, The central control module is configured to perform cross-modal spatio-temporal fusion processing by using the final weight vector and specifically configured to: take system time as a reference, calculate interpolation or extrapolation results through a motion speed vector of a target detected by each sensor, and unify heterogeneous data of different time stamps to the same time; calculate conversion results of sensor data to a vehicle body coordinate system through a pre-calibrated external parameter matrix, and determine whether different sensor data are derived from the same physical target based on projection overlap degree. The same physical target is associated successfully, and a fused target position vector is calculated by using the maximum weight vector to perform weighted average calculation on the position observation values of each sensor.

5. The complex road condition pre-judgment lighting system according to claim 1, wherein, The central control module is configured to perform space-time alignment and determine the virtual pre-judgment point, and specifically configured to: The system total response delay is calculated by summing up the perception delay, transmission delay, calculation delay and execution delay; The driving distance of the vehicle in the response period is calculated as the pre-look distance by using the real-time longitudinal speed and longitudinal acceleration of the vehicle, the system total response delay and the preset safety buffer time; The virtual pre-judgment point is locked by searching forward along the topological path of the lane center line in the high-precision map data, and the integral length of the path is equal to the pre-look distance.

6. The complex road condition pre-judgment lighting system according to claim 5, wherein, The central control module is configured to generate the feedforward control instruction according to the road features of the virtual pre-judgment point, and specifically configured to: The road curvature at the virtual pre-judgment point is extracted, the theoretical deflection angle of the beam center relative to the vehicle longitudinal axis is calculated by using the road curvature, and the theoretical deflection angle is mapped to the lateral index displacement amount of the pixel column in the Micro LED array light source; The road slope difference is calculated by using the slope value of the virtual pre-judgment point and the current position of the vehicle, and when the road slope difference is positive and exceeds a threshold value, a negative pitch adjustment instruction is generated, and when the road slope difference is negative and the absolute value exceeds a threshold value, a positive pitch adjustment instruction is generated.

7. The complex road condition pre-judgment lighting system according to claim 4, wherein, The central control module is configured to construct a virtual brightness matrix consistent with the resolution of the Micro LED array light source, and perform layer superposition calculation, and specifically configured to: A first layer global basic lighting layer is generated, and the brightness distribution thereof is determined according to the vehicle speed and the road type; A second layer dynamic shielding layer is generated, which is configured to calculate a shielding area by mapping the fused target position vector to the Micro LED array plane, and set the brightness value of the pixels in the corresponding projection area to zero or reduce; A third layer key enhancement layer is generated, which is configured to increase the brightness value of the pixels in the projection area of the recognized pedestrian or traffic sign to a highlight threshold value; An optical intensity distribution matrix is generated by performing time domain smoothing filtering calculation on the superimposed virtual brightness matrix and sent to the execution module.

8. The complex road condition pre-judgment lighting system according to claim 1, wherein, The central control module is also configured to perform user behavior adaptive learning: When manual dimming or height adjustment intervention operations of the driver are detected, an intervention log containing geographical location features and system state features is generated; The intervention log is clustered by geographical location, and if the frequency of manual intervention in the same grid coordinate area exceeds a threshold value, a high beam suppression label is marked in the high-precision map data in the grid coordinate area; The average target distance when the driver manually intervenes is counted, the deviation amount is calculated by using the average value and the system preset trigger threshold value, and the following vehicle shielding trigger threshold value in the lighting control strategy is updated by using the deviation amount.

9. The complex road condition pre-judgment lighting system according to claim 6, wherein, The optical projection lens group comprises an electric focusing mechanism; the execution module is configured to: According to the lateral index displacement amount of the pixel column in the Micro LED array light source, the on-off state of the corresponding pixel unit is independently controlled by using a pulse width modulation signal; According to the negative pitch adjustment instruction or the positive pitch adjustment instruction, the electric focusing mechanism is driven to move the lens position in the optical axis direction to change the divergence angle of the illumination light beam.

10. The complex road condition pre-judgment lighting system according to claim 1, wherein, The execution module adopts a modular structure design, including: A lamp shell and a general mounting bracket inside; A replaceable light source module, the Micro LED array light source is packaged on the replaceable light source module, the replaceable light source module is connected with the driving control box through the floating connector, the floating connector has a three-axis direction elastic floating allowance; An independent lens barrel, the optical projection lens group is packaged in the independent lens barrel, and the independent lens barrel is fixed to the front end of the replaceable light source module through a quick release structure.

Images