Vehicle front windshield adjusting method, system and equipment based on deep learning

Abstract

The invention discloses a vehicle front windshield adjusting method, system and device based on deep learning, and the method comprises the steps: recognizing light source feature information according to a vehicle front image, recognizing eye feature information of a driver according to a driver image, and obtaining environment light intensity information and vehicle speed information; according to the light source feature information and the eye feature information, determining the sight line overlapping rate of the eyes of the driver and the light source, according to the environment light intensity information and the vehicle speed information, determining a scene dynamic factor, and generating a fusion feature vector; inputting the fusion feature vector into a pre-trained scene recognition model to obtain a current scene type; according to the current scene type and the environment light intensity information, target light transmittance of the multiple partitions of the front windshield is determined, a corresponding partition adjusting instruction is generated according to the target light transmittance, and then the front windshield is adjusted according to the partition adjusting instruction. The method improves the accuracy and timeliness of vehicle front windshield adjustment, and can be applied to the technical field of vehicle control.

Description

Technical Field

[0001] This invention relates to the field of vehicle control technology, and in particular to a method, system and device for adjusting a vehicle windshield based on deep learning. Background Technology

[0002] The windshield is a crucial component of a vehicle, and its adjustment technology impacts vehicle performance and the user's driving experience. Existing windshield adjustment technologies include the following: 1) Traditional physical adjustment technology: Reduce strong light incidence through mechanical blocking, including manually retractable sunshades and sun visors (with adjustable angle); reduce glare interference from oncoming high beams at night by manually flipping the lens to change the reflection angle; the windshield adopts a single light transmittance design (normal light transmittance 70%-80%), without dynamic adjustment capability, and relies solely on high beams (halogen / LED) to enhance night vision.

[0003] 2) First-generation smart dimming glass technology: It adopts an electrochromic film and triggers adjustment through a single light sensor (detecting ambient light intensity of 0.01-100000 lux) to achieve stepless change of overall light transmittance (0.1%-99.8%). It only judges based on the "ambient light intensity threshold" (such as reducing light transmittance when light intensity is >50000 lux), and has no scene subdivision and local control capabilities.

[0004] 3) Adaptive Headlight System (AFL): It uses a camera to identify oncoming vehicles / pedestrians and dynamically adjusts the headlight illumination range (such as turning off the lights in the area where oncoming vehicles are located) to avoid glare for oncoming drivers. It only optimizes the "light source output end" and does not involve the adjustment of the optical characteristics of the "light transmission medium" of the windshield.

[0005] However, existing windshield adjustment technology has the following drawbacks: 1) Insufficient safety performance: limited visual blind spots and recognition distance.

[0006] (1) Nighttime low light scenario: Traditional high beams have a limited illumination range (only 30-50 meters) and are prone to light spot interference. The driver has a short recognition distance for pedestrians and obstacles on the road edge, and the emergency braking reaction time is insufficient (requiring 0.5-1.2 seconds). (2) Strong light / oncoming scene: Traditional glass cannot specifically absorb glare, causing drivers to have a brief "blind spot" (such as a white blur when the midday sun shines directly on them). When meeting oncoming traffic, the high beams can easily cause "dazzle" and increase the risk of collision.

[0007] 2) Low adjustment precision: lacks scenario-based and local control capabilities.

[0008] 1) Lack of scene recognition: The first generation of smart dimming glass only relies on light intensity sensors and cannot distinguish between specific scenes such as "direct strong light", "oncoming high beams" and "low light road", which is prone to "false adjustment" (such as mistakenly increasing the light transmittance when the light intensity is low in the tunnel, which instead causes the light reflection to be dazzling). 2) Limited control dimensions: Existing solutions only support “overall dimming” or “light source adjustment”, and cannot achieve “precise zone control” of the windshield (such as only blocking the oncoming high beams in the upper right area of ​​the windshield, rather than reducing the overall light transmittance).

[0009] 3) Low response efficiency: lag and high operational load.

[0010] 1) Manual adjustment lag: Physical sunshades and manual anti-glare rearview mirrors require the driver to operate with distraction (such as adjusting the sun visor with one hand), with a response time of more than 2 seconds, which does not comply with the principle of "driving safety first"; 2) Automatic adjustment lag: The first generation of smart dimming glass relies on a single sensor with a response time of >500ms, which cannot cope with "sudden changes in light intensity" such as tunnel entrances and exits and passing vehicles (e.g., the light transmittance is not increased in time when entering the tunnel, resulting in a sudden darkening of the field of vision).

[0011] 4) Lack of synergy: The "light source-medium-vision" closed loop is broken. Adaptive headlights (AFL) only optimize "light source output", smart dimming glass only optimizes "medium transmittance", and traditional physical devices only optimize "mechanical blocking". There is no data interaction and collaborative control among the three, and they cannot form a closed loop system of "external light perception → light source adaptation → medium adjustment → visual optimization", resulting in poor overall lighting adaptation effect.

[0012] The above problems urgently need to be addressed. Summary of the Invention

[0013] The purpose of this invention is to at least partially solve one of the technical problems existing in the prior art.

[0014] Therefore, one objective of this invention is to provide a deep learning-based method for adjusting a vehicle's windshield, which improves the accuracy and timeliness of windshield adjustment, thereby enhancing vehicle driving safety and the user's driving experience.

[0015] Another objective of this invention is to provide a vehicle windshield adjustment system based on deep learning.

[0016] To achieve the above-mentioned technical objectives, the technical solutions adopted in the embodiments of the present invention include: On one hand, embodiments of the present invention provide a method for adjusting a vehicle windshield based on deep learning, comprising the following steps: The system identifies light source characteristics based on the image in front of the vehicle, identifies the driver's eye characteristics based on the image of the driver, and obtains ambient light intensity and vehicle speed information. The driver's eye-to-eye line overlap rate with the light source is determined based on the light source feature information and the eye feature information. The scene dynamic factor is determined based on the ambient light intensity information and the vehicle speed information. A fusion feature vector is generated based on the line-of-eye line overlap rate and the scene dynamic factor. The fused feature vector is input into a pre-trained scene recognition model to obtain the current scene type; The target transmittance of multiple zones of the windshield is determined based on the current scene type and the ambient light intensity information. Corresponding zone adjustment instructions are generated based on the target transmittance, and the windshield is adjusted according to the zone adjustment instructions.

[0017] Furthermore, in one embodiment of the present invention, the step of identifying light source feature information based on the image in front of the vehicle, identifying the driver's eye feature information based on the driver image, and obtaining ambient light intensity information and vehicle speed information specifically includes: The light source contour is extracted from the image in front of the vehicle, and the light source coordinates and spot area are determined based on the light source contour to obtain the light source feature information. The driver's eye image is extracted from the driver's image, and the coordinates of both eyes and the pupil diameter are extracted from the driver's eye image to obtain the eye feature information; Obtain the illuminance values ​​of the four corners of the windshield, and determine the ambient light intensity information based on the average of the illuminance values; The vehicle speed information is obtained by converting the vehicle speed based on the pulse frequency of the vehicle speed sensor.

[0018] Further, in one embodiment of the present invention, the scene dynamic factor includes the light intensity abrupt change rate and the vehicle speed change rate. The steps of determining the eye-to-eye overlap rate between the driver's eyes and the light source based on the light source feature information and the eye feature information, determining the scene dynamic factor based on the ambient light intensity information and the vehicle speed information, and generating a fused feature vector based on the eye-to-eye overlap rate and the scene dynamic factor specifically include: The line-of-sight overlap rate is calculated based on the light source coordinates, the light spot area, the binocular coordinates, and the pupil diameter. The light intensity abrupt change rate is determined based on the ambient light intensity information, and the vehicle speed change rate is determined based on the vehicle speed information; The line-of-sight overlap rate, the light intensity abrupt change rate, and the vehicle speed change rate are time-aligned and time-series fused to obtain the fused feature vector.

[0019] Furthermore, in one embodiment of the present invention, the scene recognition model is trained through the following steps: Obtain the line-of-sight overlap rate and dynamic factor of the sample scene in the test scenario, and determine the corresponding scene type label through manual annotation; A sample fusion feature vector is generated based on the sample line-of-sight overlap rate and the sample scene dynamic factor. The sample fusion feature vector is used as a training sample, and the scene type label is used as the corresponding sample label to obtain a training dataset. The training samples are input into a pre-built YOLO11 neural network to obtain the predicted scene type; The loss value is determined based on the predicted scenario type and the scenario type label; The parameters of the YOLO11 neural network are updated based on the loss value to obtain the trained scene recognition model; The scene type tags include daytime strong light scenes, nighttime oncoming traffic scenes, tunnel entrance / exit scenes, and evening backlight scenes.

[0020] Furthermore, in one embodiment of the present invention, determining the target light transmittance of multiple zones of the windshield based on the current scene type and the ambient light intensity information specifically includes: Based on the current scene type, obtain the pre-calibrated transmittance adjustment strategy for the corresponding scene; The light intensity-transmittance intensity mapping relationship of each zone of the windshield is determined according to the light transmittance adjustment strategy. The target transmittance of each zone is determined based on the ambient light intensity information and the light intensity-transmittance mapping relationship.

[0021] Furthermore, in one embodiment of the present invention, each section of the windshield is provided with individually adjustable electrochromic glass, and the step of generating corresponding section adjustment commands based on the target light transmittance, and then adjusting the windshield according to the section adjustment commands, specifically involves: The target driving voltage is determined based on the target transmittance, the partition adjustment command is generated based on the partition number of each partition and the corresponding target driving voltage, and the partition adjustment command is sent to the PWM controller; The PWM controller generates PWM pulse signals corresponding to each partition according to the partition adjustment command, and the driver chip converts the PWM pulse signals into DC voltages for the corresponding partitions, thereby driving the electrochromic glass of the corresponding partitions to adjust the light transmittance through the DC voltages.

[0022] On the other hand, embodiments of the present invention provide a vehicle windshield adjustment system based on deep learning, comprising: The multimodal perception module is used to identify light source feature information based on the image in front of the vehicle, identify the driver's eye feature information based on the driver image, and obtain ambient light intensity information and vehicle speed information. The data fusion module is used to determine the line-of-sight overlap rate between the driver's eyes and the light source based on the light source feature information and the eye feature information, determine the scene dynamic factor based on the ambient light intensity information and the vehicle speed information, and generate a fusion feature vector based on the line-of-sight overlap rate and the scene dynamic factor. The scene recognition module is used to input the fused feature vector into a pre-trained scene recognition model to obtain the current scene type; The adjustment module is used to determine the target light transmittance of multiple zones of the windshield based on the current scene type and the ambient light intensity information, generate corresponding zone adjustment instructions based on the target light transmittance, and then adjust the windshield according to the zone adjustment instructions.

[0023] On the other hand, embodiments of the present invention provide an electronic device, the electronic device including a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for implementing communication between the processor and the memory, wherein when the program is executed by the processor, it implements the deep learning-based vehicle windshield adjustment method described above.

[0024] On the other hand, embodiments of the present invention also provide a vehicle, the vehicle including a deep learning-based vehicle windshield adjustment system or electronic device as described above.

[0025] On the other hand, embodiments of the present invention also provide a storage medium, which is a computer-readable storage medium for computer-readable storage. The storage medium stores one or more programs, which can be executed by one or more processors to implement the deep learning-based vehicle windshield adjustment method described above.

[0026] On the other hand, embodiments of the present invention also provide a computer program product, including a computer program that, when executed by a processor, implements the deep learning-based vehicle windshield adjustment method described above.

[0027] The advantages and beneficial effects of the present invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention: This invention identifies light source features from a vehicle-front image, driver eye features from a driver image, and ambient light intensity and vehicle speed. It then determines the line-of-sight overlap between the driver's eyes and the light source based on the light source and eye features, determines scene dynamic factors based on the ambient light intensity and vehicle speed, and generates a fused feature vector based on the line-of-sight overlap and scene dynamic factors. This fused feature vector is input into a pre-trained scene recognition model to obtain the current scene type. Based on the current scene type and ambient light intensity, it determines the target transmittance for multiple zones of the windshield and generates corresponding zone adjustment commands based on the target transmittance. The windshield is then adjusted according to these commands. This invention identifies the current scene type based on the line-of-sight overlap and scene dynamic factors, and adjusts the transmittance of multiple zones of the windshield based on the current scene type and ambient light intensity. This improves the accuracy and timeliness of windshield adjustment, thereby enhancing vehicle driving safety and the user's driving experience. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the embodiments of the present invention are described below. It should be understood that the drawings described below are only for the convenience of clearly describing some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 A flowchart illustrating the steps of a deep learning-based method for adjusting a vehicle windshield, as provided in an embodiment of the present invention. Figure 2 A schematic diagram of the structure of a vehicle windshield adjustment system based on deep learning provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0030] The embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application. It should be noted that although functional modules are divided in the system schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the system schematic diagram or the order in the flowchart. The step numbers in the following embodiments are only set for ease of explanation and do not limit the order between steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0031] In the description of this invention, "multiple" means two or more. The use of "first" and "second" is for distinguishing technical features only and should not be construed as indicating or implying relative importance, the number of indicated technical features, or the order of the indicated technical features. Furthermore, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0032] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards of the relevant countries and regions. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirects to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data for the proper functioning of the embodiments of this application obtained.

[0033] This invention aims to solve the following core problems existing in current driver assistance technologies: 1) Address the issues of "insufficient safety performance: short recognition distance at night, and blind spots in strong light / oncoming traffic". In response to the shortcomings of existing technologies, such as "the recognition distance of high beams at night is only 30-50 meters, and the field of vision is completely white when there is strong light or oncoming traffic", this invention solves the core safety problems of limited recognition distance of drivers for pedestrians and obstacles in low light conditions at night, as well as the temporary blind spots caused by direct strong light and glare from oncoming traffic, through "dual-light intelligent camera (visible light + infrared) + infrared light transmission enhancement design".

[0034] 2) Solve the problem of "low adjustment precision: inability to subdivide scenarios and lack of local control capability". To address the shortcomings of existing technologies, such as "first-generation intelligent dimming relying solely on light intensity thresholds, lacking scene segmentation, and only supporting overall dimming," this invention solves the problems of being unable to accurately determine three core scenarios—"low-light night vision, strong direct light, and oncoming glare"—and being unable to individually dim local areas of the windshield (such as the area where oncoming headlights are incident), by using "an improved YOLO11 scene recognition network + 3×3 zoned electrochromic glass." This avoids erroneous adjustments such as "incorrectly increasing transmittance in tunnels, leading to glare from reflections."

[0035] 3) Solve the problem of "low response efficiency: manual distraction lag and slow automatic response". To address the shortcomings of existing technologies, such as "manual adjustment response > 2 seconds and first-generation intelligent dimming response > 500ms", this invention solves the problems of manual adjustment requiring driver distraction (violating driving safety principles) and automatic adjustment being unable to cope with "instantaneous changes in lighting" in scenarios such as tunnel entrances and exits and oncoming traffic by using "multimodal data parallel preprocessing + deep learning rapid decision-making".

[0036] 4) Addressing the issue of "lack of collaboration: fragmented technology and lack of a closed loop of 'perception-decision-execution'". To address the shortcomings of existing technologies where "adaptive headlights (AFL), dimming glass, and physical obstruction lack data interaction and result in fragmented optimization," this invention solves the problem of the disconnect between "light source output adjustment, windshield medium light transmission, and ambient light perception," which prevents the formation of a complete closed loop of "external light → scene decision → light transmission adjustment → effect feedback," through "CAN bus data interaction + closed-loop feedback mechanism."

[0037] like Figure 1 The diagram shows a flowchart of a deep learning-based method for adjusting a vehicle windshield according to an embodiment of the present invention. (Refer to...) Figure 1 This invention provides a deep learning-based method for adjusting a vehicle's windshield, specifically including the following steps: S101. Identify light source feature information based on the image in front of the vehicle, identify driver's eye feature information based on the driver image, and obtain ambient light intensity information and vehicle speed information. S102. Determine the line-of-sight overlap rate between the driver's eyes and the light source based on the light source feature information and eye feature information, determine the scene dynamic factor based on the ambient light intensity information and vehicle speed information, and generate a fused feature vector based on the line-of-sight overlap rate and the scene dynamic factor. S103. Input the fused feature vector into the pre-trained scene recognition model to obtain the current scene type; S104. Determine the target transmittance of multiple zones of the windshield based on the current scene type and ambient light intensity information, generate corresponding zone adjustment instructions based on the target transmittance, and then adjust the windshield according to the zone adjustment instructions.

[0038] Specifically, this embodiment of the invention integrates a forward vision module (HDR mode camera for light source positioning and strong light overexposure suppression), a driver's cab vision module (eye positioning and pupil monitoring), four-corner HDR light intensity sensors (OPT3001 for precise acquisition of light intensity data), and a 3×3 zone electrochromic execution module (TLC5940 PWM control, NCP3065 driver). The control module, based on the fusion of multiple data such as "light source-eye position-light intensity", matches a transmittance adjustment strategy (e.g., during midday strong light, only the transmittance of zones 2 and 5 corresponding to the driver's line of sight is adjusted to 15%, while the rest remain at 50%), to achieve precise local dimming of the windshield without manual operation by the driver. This avoids blind spots caused by direct strong light and ensures a clear overall field of vision, ultimately improving driving focus and driving safety.

[0039] Specifically, the implementation of the vehicle windshield adjustment method in this embodiment of the invention relies on a multimodal perception layer, a decision control layer, an execution layer, a collaborative feedback layer, and a power support layer, as detailed below: 1) Multimodal perception layer (the core of data acquisition, providing a high signal-to-noise ratio data source for decision-making) 1.1 Forward Vision Module Core functions: Identifies the outlines of strong light sources such as the sun and oncoming headlights through the Canny operator edge detection algorithm, and outputs the pixel coordinates of the light source and the area of ​​the light spot; automatically switches to HDR mode in strong light scenes (light intensity > 80000 lux), and configures an exposure time of 1 / 2000s and ISO sensitivity of 100 to suppress overexposure; crop the corresponding image range of the windshield (x=500-1500 pixels, y=100-700 pixels) to remove redundant areas and reduce data transmission by 30%.

[0040] 1.2 Cab Vision Module Core functions: Output the physical coordinates of the driver's eyes through the process of "face recognition → facial key point extraction → corneal reflection positioning"; monitor the pupil diameter at a 5Hz sampling rate and send an over-contraction warning when the pupil is <1.5mm under strong light; resample the eye position every 500ms and update the coordinates in real time when the driver's head shifts laterally by more than 5mm to avoid deviation of the dimming area.

[0041] 1.3 Light Intensity Sensor Core functions: The system outputs the light intensity values ​​of the four corners of the windshield at a default sampling rate of 50Hz. In strong light scenarios (single sensor light intensity > 80,000 lux and a decrease of < 5% in three consecutive samplings), it automatically switches to a 100Hz sampling rate. It uses the 3σ criterion to remove extreme values ​​and calculates the average light intensity of the four sensors. When the average light intensity is > 60,000 lux, it sends a "strong light signal" to trigger a rapid response from the decision-making layer.

[0042] 1.4 Vehicle speed sensor Core functions: Calculate real-time vehicle speed using the formula v=0.0015×f based on pulse frequency (60 pulses per revolution); send a "low-speed following signal" when vehicle speed <30km / h to prompt the decision-maker to increase the light transmittance of the taillight area of ​​the vehicle in front; send a fault prompt when there is no pulse signal for 100ms, triggering the decision-maker to switch to the default mode of "overall light transmittance 50%".

[0043] 2) Decision control layer (the core brain of the system, realizing precise mapping of "data-scenario-strategy") 2.1 Data Fusion Unit Core functions: Align multi-sensor data by timestamp (error ≤ 10ms) to avoid judgment bias caused by the asynchrony between light intensity and visual data; remove eye positioning data with confidence < 80% and light intensity mean with deviation > 5%, and supplement with data interpolated from adjacent time points in case of failure; extract "light source-eye overlap rate" and "scene dynamic factors (sudden change rate of light intensity, rate of change of vehicle speed)" to generate standardized feature vectors.

[0044] 2.2 Scene Recognition Unit Core functionality: Based on an improved YOLO11 neural network (embedded with a multi-core attention module), it classifies four scenarios: "daytime strong light (I_env > 60000 lux + overlap rate > 80%)", "nighttime oncoming traffic (I_env < 5000 lux + recognition of oncoming vehicle lights)", "tunnel entrances and exits (sudden change rate of light intensity > 50000 lux / s)", and "evening backlight (I_env 10000-60000 lux + low angle of the sun)". Classification confidence ≥ 95% is considered valid; in composite scenarios, the priority of targets is sorted in order of "anti-glare > following vehicle recognition > field of vision protection".

[0045] 2.3 Strategy Generation Unit Core functions: In strong light scenarios, when the overlap rate is >90%, the transmittance of the target zone is 15-20% (the higher the I_env, the lower the transmittance), and the non-overlapping zone is 50-60%; in oncoming vehicle scenarios, the transmittance of the oncoming vehicle headlight area is 10-15%, and the other areas are 70-80%; in tunnel scenarios, the overall transmittance is 85%, and the transmittance at the entrance and exit is adjusted by 5% every 0.3 seconds; the driving voltage is calculated by querying the mapping table based on the transmittance, and the dimming command is encapsulated as "zone number + target transmittance + driving voltage + execution time limit".

[0046] 2.4 Control Module Core functions: Send dimming commands to the execution layer PWM controller via SPI bus, and send coordination signals to the vehicle AFL system and instrument panel via CAN bus; receive dimming progress from the execution layer, and fine-tune the drive voltage when the actual transmittance deviates from the target by more than 3%; trigger "adjacent zone compensation" when the execution layer feedback voltage is abnormal (e.g., if zone 4 is faulty, adjust the transmittance of zones 6 and 8 to the target value of zone 4).

[0047] 3) Execution layer (action execution end, which converts dimming commands into physical actions) 3.1 PWM Controller Core functions: Receives partition duty cycle commands and generates a corresponding duty cycle PWM wave at a frequency of 1kHz; outputs adjacent partition PWM signals with a 50μs delay to avoid electrode crosstalk; monitors the total output current and cuts off the output to protect the electrochromic film when it exceeds 1.5A.

[0048] 3.2 Driver Chip Core functions: Converts PWM pulse signals into stable DC voltage, and adjusts the voltage division ratio through feedback resistor to achieve 1.5-5V voltage regulation; configures a current limiting register to limit the output current to 150mA (the upper limit of thin film safety); adopts a soft-start mechanism to control the voltage change rate at 0.93V / s to avoid visual discomfort caused by sudden changes in light transmittance.

[0049] 3.3 Multi-zone electrochromic glass Core functions: After receiving the driving voltage, stepless dimming from 0.1% to 99.8% is achieved through Li+ migration in the electrochromic film (rate 0.2μm / s); 9 zone electrodes are independently powered, supporting "local anti-glare + global view"; the all-solid-state structure has no liquid leakage, and the dimming response time only increases by 0.3 seconds at a low temperature of -40℃.

[0050] 3.4 Condition Monitoring Unit Core functions: Sample the electrode voltage of each zone every 50ms, calculate the actual transmittance through the "voltage-transmittance" mapping table; collect the drive current through a 0.1Ω sampling resistor, and determine the overcurrent fault when it is >200mA; send "zone number + current transmittance + target transmittance + fault identifier" to the control module to provide feedback on the dimming progress.

[0051] 4) Collaborative feedback layer (to achieve internal and external linkage of the system and form closed-loop control) 4.1 Vehicle CAN bus module Core functions: Transmitting interactive data between the decision-making and execution layers (coordination signals, fault feedback); linking with the vehicle's AFL system (such as synchronously triggering high beam mode + infrared light transmission enhancement at night); linking with the instrument panel to send mode commands; caching key data such as light intensity, vehicle speed, and dimming commands for nearly 10 seconds, and exporting them for diagnosis in case of faults.

[0052] 4.2 Instrument Panel Display Core functions: Receive CAN commands and display the current dimming mode such as "strong light anti-glare" and "tunnel mode", and use color to distinguish priorities (red for high priority, blue for normal); in strong light scenes, the backlight is reduced to 30% to avoid reflection, and in tunnel scenes, it is increased to 70% to adapt to low light; when the execution layer reports a fault, it displays prompts such as "Zone dimming failure (zone 4), already compensated" and is accompanied by an adjustable volume buzzer alarm.

[0053] 5) Power supply support layer (provides a stable and low-consumption power supply for the entire system) Power Management Unit Core functions: Converts the on-board 12V power supply to 3.3V (for the perception layer, decision layer, and coordination layer) and 1.5V (for the execution layer thin film), with an output ripple of ≤50mV; after the execution layer dimming is completed, the ADP150-1.5 converter switches to sleep mode, and the output current drops to 10μA to achieve zero power consumption maintenance; when the input voltage is >15V, the Zener diode breaks down and discharges voltage, and when the total current is >5A, the fuse blows to protect downstream modules.

[0054] The dual-light intelligent camera used in this embodiment of the invention is a dual-mode acquisition system for visible light (1920×1080@30fps) and infrared (850nm band, 1280×720). The HDR light intensity sensor is distributed at the four corners of the windshield and has a measurement range of 0.01-100000 lux. The preprocessing unit uses wavelet transform noise reduction (db4 basis, 3-level decomposition) and CLAHE feature enhancement.

[0055] The electrochromic glass used in this embodiment of the invention consists of, from the outside in, ultra-white float glass (2.1 mm), an all-solid-state electrochromic film (500 nm), an ITO transparent electrode (sheet resistance ≤ 15 Ω / □), and a PET flexible substrate (0.1 mm). The ITO electrode is divided into 3×3 independent regions. The driving unit uses a TLC5940 chip to generate a 0-5V PWM signal, and a current-limiting resistor (20 Ω / 2 W) is connected in series in the electrode circuit. It can be recognized that the embodiments of the present invention identify the current scene type based on the line-of-sight overlap rate and scene dynamic factors, and adjust the light transmittance of multiple sections of the windshield according to the current scene type and ambient light intensity information, thereby improving the accuracy and timeliness of the vehicle's windshield adjustment, and thus improving the vehicle's driving safety and the user's driving experience.

[0056] As a further optional implementation, the system identifies light source feature information based on the image in front of the vehicle, identifies the driver's eye feature information based on the driver image, and obtains ambient light intensity information and vehicle speed information, specifically including: S201. Extract the light source contour from the image in front of the vehicle, determine the light source coordinates and spot area based on the light source contour, and obtain the light source feature information; S202. Extract the driver's eye image from the driver's image, and extract the coordinates of both eyes and the pupil diameter from the driver's eye image to obtain eye feature information; S203. Obtain the illuminance values ​​of the four corners of the windshield and determine the ambient light intensity information based on the average value of the illuminance values. S204. Calculate the vehicle speed based on the pulse frequency of the vehicle speed sensor to obtain the vehicle speed information.

[0057] Specifically, after the system is powered on, it simultaneously starts four perception modules: the front vision module (windshield camera), the driver's cab vision module (rearview mirror eye camera), the four corner light intensity sensors of the windshield, and the drive wheel speed sensor, and completes hardware self-test and parameter initialization (such as camera frame rate and sensor sampling rate) to prepare for data collection.

[0058] Forward vision module: Acquires images from the front, identifies the coordinates and area of ​​the light source (sun / car headlights), and automatically switches to HDR mode in strong light scenes (light intensity > 80000 lux) to output data at 30fps; The driver's cab vision module uses facial recognition to locate the coordinates of the driver's eyes (accuracy ±2mm), monitors the pupil diameter, and updates the position every 500ms. Light intensity sensor: Collects light intensity data from 4 channels, calculates the average value (I_env), and increases the sampling rate to 100Hz and removes outliers during periods of strong light. Vehicle speed sensor: Calculates real-time vehicle speed (error ±0.5km / h) by pulse frequency, outputs every 100ms, and simultaneously determines low speed / fault status.

[0059] As a further optional implementation, the scene dynamic factors include the rate of sudden change in light intensity and the rate of change in vehicle speed. The overlap rate of the driver's line of sight with the light source is determined based on light source characteristic information and eye characteristic information. The scene dynamic factors are determined based on ambient light intensity information and vehicle speed information. A fused feature vector is generated based on the line-of-sight overlap rate and the scene dynamic factors, specifically including: S301. Calculate the visual overlap rate based on the light source coordinates, light spot area, binocular coordinates, and pupil diameter. S302. Determine the light intensity abrupt change rate based on ambient light intensity information, and determine the vehicle speed change rate based on vehicle speed information; S303. Time alignment and temporal fusion are performed on the line-of-sight overlap rate, light intensity abrupt change rate, and vehicle speed change rate to obtain a fused feature vector.

[0060] Specifically, the data fusion unit aligns the data from multiple modules according to timestamps (error ≤ 10ms), cleans low-confidence data (such as eye positioning confidence < 80%) and supplements interpolation; at the same time, it extracts core features: the overlap rate between the light source and the eye's line of sight, the rate of sudden change in light intensity / vehicle speed, and other scene dynamic factors to generate a fused feature vector.

[0061] Using a scene recognition model trained on an improved YOLO11 neural network, and combining fused feature vectors, four core scene categories are identified: daytime strong light scene, nighttime oncoming traffic scene, tunnel entrance / exit scene, and evening backlight scene.

[0062] As an optional implementation, the scene recognition model is trained through the following steps: S401. Obtain the sample line-of-sight overlap rate and sample scene dynamic factor in the test scenario, and determine the corresponding scene type label through manual annotation; S402. Generate a sample fusion feature vector based on the sample line-of-sight overlap rate and sample scene dynamic factors. Use the sample fusion feature vector as training samples and the scene type label as the corresponding sample label to obtain the training dataset. S403. Input the training samples into the pre-built YOLO11 neural network to obtain the predicted scene type; S404. Determine the loss value based on the predicted scene type and scene type label; S405. Update the parameters of the YOLO11 neural network based on the loss value to obtain the trained scene recognition model; The scene type tags include daytime strong light scenes, nighttime passing scenes, tunnel entrance / exit scenes, and evening backlight scenes.

[0063] Specifically, the eye-to-eye overlap rate and scene dynamics factor of the test scenario are obtained, and the corresponding scene type labels are determined by manual annotation. A sample fusion feature vector is generated based on the eye-to-eye overlap rate and scene dynamics factor. The sample fusion feature vector is used as the training sample, and the scene type label is used as the corresponding sample label to obtain the training dataset. The training sample is input into a pre-built YOLO11 neural network to obtain the predicted scene type. The loss value is determined based on the predicted scene type and scene type label. The parameters of the YOLO11 neural network are updated based on the loss value to complete one round of iterative training. When the number of iterations reaches a preset threshold, or the loss value is lower than the preset threshold, training is stopped, and the trained scene recognition model is obtained.

[0064] As a further optional implementation, the target light transmittance of multiple zones of the windshield is determined based on the current scene type and ambient light intensity information, specifically including: S501. Obtain the pre-calibrated transmittance adjustment strategy for the corresponding scene based on the current scene type; S502. Determine the light intensity-transmittance intensity mapping relationship of each zone of the windshield according to the transmittance adjustment strategy. S503. Determine the target transmittance of each zone based on ambient light intensity information and the light intensity-transmittance strong mapping relationship.

[0065] Specifically, a corresponding transmittance adjustment strategy is determined based on the current scene type. This strategy defines a strong mapping relationship between light intensity and transmittance for nine zones of the windshield. Then, based on this strong mapping relationship and the real-time ambient light intensity, the target transmittance for each of the nine zones is determined. For example, in a strong light scene, when the ambient light intensity is I1, the transmittance of the target zone with an overlap rate >90% is 15-20% (the higher the I_env, the lower the transmittance), and the non-overlapping zone is 50-60%. In a meeting scene, when the ambient light intensity is I2, the transmittance of the oncoming headlight area is 10-15%, and the other areas are 70-80%. In a tunnel scene, when the ambient light intensity is I3, the overall transmittance is 85%, and the transmittance at the entrance / exit is adjusted by 5% every 0.3 seconds.

[0066] As a further optional implementation, each section of the windshield is equipped with individually adjustable electrochromic glass. A corresponding section adjustment command is generated based on the target light transmittance, and the windshield is then adjusted according to the section adjustment command. Specifically: S601. Determine the corresponding target driving voltage based on the target transmittance, generate a partition adjustment command based on the partition number of each partition and the corresponding target driving voltage, and send the partition adjustment command to the PWM controller. S602: The PWM controller generates PWM pulse signals corresponding to each partition according to the partition adjustment command, and the driver chip converts the PWM pulse signals into DC voltages corresponding to the partitions, thereby driving the electrochromic glass of the corresponding partitions to adjust the light transmittance through the DC voltages.

[0067] Specifically, the strategy generation unit outputs a customized solution based on the scenario: first, it calculates the light transmittance of each zone (e.g., in a strong light scenario, zones 2 and 5 corresponding to the driver's line of sight are adjusted to 15%, and the remaining zones to 50%); then, it matches the driving voltage through a "light transmittance-voltage" mapping table; and finally, it encapsulates the solution into an instruction set of "zone number + light transmittance + voltage + execution time limit".

[0068] After receiving the command, the control module sends the dimming command to the PWM controller via the SPI bus; at the same time, it sends a coordination signal via the CAN bus: linking the vehicle AFL system (adaptive headlights) to adjust the lighting mode, and synchronously controlling the instrument panel to display the current dimming mode (such as "high beam anti-glare") and adapt to the backlight brightness.

[0069] The PWM controller generates a 1kHz pulse signal (duty cycle as instructed); the driver chip (NCP3065) converts the pulse signal into a stable DC voltage (current limited to 150mA to prevent film burn-out); the 3×3 zoned electrochromic glass uses Li-line transistors within the thin film... + Migration allows for stepless dimming (range 0.1%-99.8%) within 2 seconds, enabling localized anti-glare or field-of-view adaptation.

[0070] In some optional embodiments, the state monitoring unit (STM809) samples the voltage of the electrochromic glass electrode every 50ms and calculates the actual transmittance: if the deviation from the target is ≤3%, it feeds back "dimming normal"; if the deviation exceeds 3%, it triggers the control module to fine-tune the drive voltage to ensure dimming accuracy, forming a "command-execution-feedback" closed loop.

[0071] The AFL system adjusts the lights according to the coordination signals (such as turning off high beams when in strong light / when meeting oncoming traffic, and turning on low beams in tunnels); the instrument panel displays the mode status in sync and adapts the backlight to the ambient light (such as brightening to 70% in tunnels and dimming to 30% under strong light) to ensure driver comfort.

[0072] This invention utilizes "dual-light vision + multi-sensor fusion + deep learning scene recognition" to accurately capture high-risk scenarios such as strong daytime sunlight, glare from oncoming traffic at night, and sudden changes in light intensity in tunnels. It completes localized dimming (e.g., only obscuring the driver's field of vision in strong sunlight) within 1.2-2 seconds, eliminating the problems of "blurred vision due to overall dimming" and "distraction from manual operation" inherent in traditional dimming systems. This increases the target recognition distance by 3-5 times (e.g., from 20 meters to 60 meters in a tunnel), reducing the risk of traffic accidents by 35%-40%. No manual adjustment of sun visors or anti-glare rearview mirrors is required; the system automatically completes the entire process of "light intensity acquisition → scene judgment → dimming execution → dynamic calibration." It is particularly suitable for complex road conditions such as morning rush hour and long-distance nighttime driving, improving driver focus by 40%-50% and avoiding the distraction risk caused by manual operation, meeting the core design requirement of "driving safety first" in vehicles. The use of 3×3 zone electrochromic glass (0.1%-99.8% stepless dimming) differs from traditional "one-size-fits-all" dimming solutions, enabling... The current solution combines "partial glare reduction with overall clarity" (e.g., only the oncoming headlights are blocked when meeting oncoming traffic, while the rest maintains 70% light transmittance). This avoids direct glare into the eyes without affecting the observation of road markings, taillights of vehicles ahead, and pedestrians, solving the industry dilemma of "the trade-off between glare reduction and visibility." The power management unit achieves "low power consumption (≤5W) during adjustment and zero power consumption during maintenance," with an average daily energy consumption of only 2.4Wh (far lower than the 12Wh of traditional sunshades), having a negligible impact on the range of new energy vehicles. At the same time, it is compatible with more than 90% of existing mass-produced vehicles through the CAN bus, eliminating the need for large-scale modifications to the vehicle wiring harness, reducing the implementation costs for automakers, and possessing high mass production feasibility. By improving the YOLO11 network to recognize four core in-vehicle scenarios (daytime strong light, nighttime oncoming traffic, tunnel entrances and exits, and evening backlight), and combining them with dynamic speed adjustment strategies (e.g., shortening decision delay at high speeds and enhancing following visibility at low speeds), it adapts to the full range of vehicle usage needs from urban commuting to long-distance highways, breaking the limitations of traditional single-scenario glare reduction solutions.

[0073] The present invention will be further described below with reference to two specific embodiments.

[0074] 1. Scene 1: Daytime midday direct sunlight scene; Time: 18:45 in autumn evening (sunset, backlight angle 30°, light intensity decreases over time: from 35000 lux to 12000 lux).

[0075] 1) Forward vision module Triggering conditions: The light intensity sensor reports a light intensity of >80000 lux, and the camera captures a brightness value of >240 (RGB color gamut) in the central area of ​​the image (x=800-1200 pixels, y=200-400 pixels).

[0076] Specific operation: Automatically switch to "HDR High Dynamic Range Mode" (write to register 0x07 via I2C bus, configure exposure time 1 / 2000s, ISO sensitivity 100) to suppress overexposure of strong light; start "Light Source Positioning Submodule", identify the sun outline (circular light spot, diameter > 50 pixels) through edge detection algorithm, and output the sun's coordinates in the image (x=1000 pixels, y=300 pixels); crop the "Light Source-Windshield Mapping ROI Area" (only retain the image range corresponding to the windshield x=500-1500 pixels, y=100-700 pixels), and send positioning data (format: sun coordinates + light spot area) to the control module once every 30ms.

[0077] 2) Light intensity sensor Triggering conditions: After initialization, the default sampling rate is 50Hz. When the light intensity sampled by the A1 (top left) and A2 (top right) sensors is greater than 80000 lux for 3 consecutive times.

[0078] Specific operations: Increase the sampling rate to 100Hz (modify the sampling period to 10ms via configuration register 0x03), and simultaneously start "outlier removal" (remove sampling points with a deviation from the mean of >5%, such as a single sample of 85000 lux with a mean of 82000 lux, which is judged as outlier and discarded); calculate the mean value of the 4 sensors (82000 lux), and send a "strong light signal" (data frame: 0x01 + 82000 lux hexadecimal value 0x14010) to the control module via the I2C bus, updating once every 20ms.

[0079] 3) Driver's cab vision module Triggering condition: The vehicle starts by default after power-on, when the control module sends a "strong light scene trigger signal" (CANID0x18F00501).

[0080] Specific operation: Start "Fast Face Recognition" (prioritize locating the driver's facial contour, time ≤100ms), then locate the coordinates of both eyes (x=350mm, y=180mm) and (x=420mm, y=180mm) through corneal reflection, with a coordinate accuracy of ±2mm; resample the position of both eyes every 500ms. If the lateral displacement of both eyes is found to be >5mm (such as the driver's head adjustment), immediately send an "eye position update frame" (format: new coordinates of the left eye + new coordinates of the right eye) to the control module; turn on "Pupil Monitoring" (sampling rate 5Hz) to provide real-time feedback on the pupil diameter (default diameter 2mm under strong light, if the diameter <1.5mm, send a "pupil constriction warning").

[0081] 4) Control Module Data reception and fusion: Receive the sun coordinates (x=1000 pixels) from the front vision module, and convert them into the projection area of ​​the sun on the physical front windshield (upper half, sections 2 and 5) using the "image-physical coordinate mapping algorithm" (pre-stored front windshield pixel-actual position correspondence table); Receive the average light intensity of 82000 lux and the coordinates of both eyes, substitute them into the "light source-eye matching model", calculate the overlap rate between the light path and the line of sight of both eyes (92%), and determine that sections 2 and 5 need to be occluded.

[0082] Strategy generation and command transmission: Call the "transmittance-voltage lookup table" to calculate the driving voltage of 1.8V corresponding to 15% transmittance in zones 2 and 5 (15% transmittance → 15% PWM duty cycle), and the voltage of 3.2V corresponding to 50% transmittance in the remaining zones (40% duty cycle); send commands to the electrochromic module via the SPI bus (frame format: zone number + duty cycle, such as "2,15;5,15;1,40;3,40;…"), and at the same time send the "strong light anti-glare mode on" display command to the instrument panel via the CAN bus.

[0083] 5) Zoned electrochromic anti-glare module PWM controller (TLC5940) operation: Within 10ms after receiving the command, write the duty cycle registers of each partition (Registration 2 register 0x12=0x0F, Registration 5 register 0x15=0x0F, Registration 1 register 0x11=0x28) to generate a PWM signal with a frequency of 1kHz; start "dead time control" (the PWM signals of adjacent partitions are delayed by 50μs to avoid crosstalk).

[0084] Driver chip (NCP3065) operation: Converts PWM signal to DC voltage, outputs 1.8V in zones 2 and 5 (by adjusting feedback resistors R1=2kΩ and R2=1kΩ to achieve voltage division), current is limited to 150mA (configured current limit register 0x07=0x96); completes a smooth transition of voltage from 3.2V (default) to 1.8V within 1.5 seconds (voltage change rate 0.93V / s, to avoid sudden changes in light transmittance).

[0085] Status monitoring and feedback: The STM809 voltage monitoring chip samples the electrode voltages of zones 2 and 5 every 50ms, calculates the actual transmittance (e.g., 1.8V corresponds to 15%), and sends the "dimming progress" to the control module via SPI (format: zone 2 15% + zone 5 14% → target 15%). If the voltage of zone 2 is detected to be >2.0V (corresponding to transmittance >18%), a "dimming deviation signal" is immediately sent to the control module to trigger duty cycle fine-tuning (from 15% to 14%).

[0086] 6) Scene Judgment The forward vision module detected that the light source (sun) was located at the upper left of the windshield. The light intensity dropped from 35,000 lux to 12,000 lux within 1 minute, with an incident angle of 30°. The driver's vision module captured the driver's head making slight adjustments (eye coordinates from x=355mm→365mm, y=180mm→178mm). The line of sight still overlapped with the backlight source, which was determined to be a "dynamic backlight at dusk" scene (both light intensity and eye position changed).

[0087] 7) Strategy matching and execution The control module outputs a dynamic strategy: the light transmittance of the windshield zones 1 and 2 decreases with light intensity: 18% at 35000 lux → 35% at 12000 lux; simultaneously, the occlusion range of zones 1 and 2 is finely adjusted according to the eye coordinates (lateral offset of 5mm).

[0088] 8) Perform the action The transmittance is adjusted every 10 seconds based on the light intensity data (in 5% increments) to avoid sudden changes in transmittance that could cause visual discomfort; the driver's cab vision module updates the eye coordinates every 200ms to calibrate the obstructed area in real time.

[0089] In this specific embodiment, the light transmittance dynamically adapts to the light intensity, ensuring that the driver's vision is never "blinded" and that the recognition rate of road targets (such as decelerating vehicles ahead) remains at 98%. It adapts to slight eye movements, with an occlusion accuracy error of ≤3mm. It eliminates the need for frequent manual adjustments to the sun visor or windows, reducing the driver's workload.

[0090] 2. Scenario 2: Sudden change in light intensity at tunnel entrance and exit during morning rush hour; Time: 7:40 am during spring morning rush hour (entrance of urban expressway tunnel, outdoor light intensity 65,000 lux on a sunny day, light intensity 800 lux inside the tunnel).

[0091] 1) Dual-light intelligent camera 50 meters before the tunnel entrance: Automatically switch to "visible light + infrared dual-mode acquisition", turn on 850nm supplementary light in the infrared band, temporarily increase the frame rate from 30fps to 40fps, and maintain the image resolution of 1920×1080; When the tunnel entrance outline is captured: activate "Region of Interest (ROI) cropping" to retain only the image of "tunnel entrance + 30 meters in front" (removing redundant sky areas) to reduce data transmission volume; When identifying oncoming headlights: the "light source suppression algorithm" is triggered to perform local exposure control on the headlight pixel area (exposure time is reduced from 1 / 1000s to 1 / 2000s) to avoid overexposure.

[0092] 2) HDR light intensity sensor Outdoor phase (light intensity 65000 lux): A1-A4 sensors output light intensity data at a sampling rate of 50 Hz and send it to the decision module once every 20 ms; When the light intensity drops to 40,000 lux (5 meters from the tunnel entrance): Sensor A2 (upper right) detects the sudden drop first, triggering a "rapid response mechanism". The sampling rate is temporarily increased to 100 Hz, and a "sudden change in light intensity interruption signal" is sent to the decision module. In the tunnel stage (light intensity 800 lux): A1-A4 sampling rate is restored to 50 Hz, and "low light gain mode" is automatically switched (gain value is adjusted from 1× to 8×) to ensure data accuracy.

[0093] 3) Vehicle speed sensor Detect vehicle speed drops to 35km / h: Triggering condition: 5 consecutive pulse cycles (each cycle corresponds to 1 / 60 of a wheel rotation) calculate vehicle speed < 35km / h; Execution action: Send a CAN message (ID0x18F00201, data segment "0023" represents 35km / h) to the decision module, and at the same time pull up the "low speed flag" (GPIO pin PA5 is set high).

[0094] 4) CAN bus data Decision module → Dashboard: Execute action: Send CAN message (ID0x18F00301, data segment "0146", "01" represents tunnel mode, "46" represents backlight 70% (hexadecimal 46 = decimal 70). Execution Module → Decision Module: Execution Action: Send a "status code frame" (ID0x18F00401) every 100ms. For example, "00010101" represents normal operation in zones 1-3, and "000110101" represents abnormal voltage in zone 4.

[0095] 5) Scene Judgment The decision module integrates ROI images from dual-light cameras (tunnel entrance + oncoming headlights + vehicle ahead), HDR sensor data on sudden changes (65000 lux → 22000 lux), and a vehicle speed of 40 km / h. Within 100 ms, it determines the scenario as a composite scene of "sudden drop in light intensity at tunnel entrance + meeting oncoming traffic + following oncoming traffic". Based on voltage feedback, it confirms the dimming progress of zones 4 and 7 and dynamically corrects the PWM duty cycle (e.g., if the transmittance difference in zone 4 is 3%, the duty cycle is reduced by another 1%).

[0096] 6) Strategy matching and execution The decision module outputs dynamic strategies: (1) Dimming stage (0-1.2 seconds): Zones 4 and 7 have 25% transmittance (anti-glare), and the rest have 85% (low light adaptation); (2) Stable stage inside the tunnel (after 1.2 seconds): Zones 4 and 7 recover to 85%, and the overall transmittance remains at 85%; 3. Exiting the tunnel (light intensity > 30000 lux): The overall transmittance gradually decreases from 85% to 55% (5% decrease every 0.3 seconds).

[0097] 7) Perform the action During the tunnel entrance dimming phase (0-1.2 seconds): the electrochromic module adjusts the duty cycle of zones 4 and 7 to 15% (25% transmittance) and the remaining zones to 65% (85% transmittance) according to the PWM command. The STM809 provides real-time progress feedback, and the decision module corrects the duty cycle of zone 4 to 14% based on the feedback. Stable phase inside the tunnel (1.2 seconds - 80 seconds): The dual-light camera turns off infrared illumination (power is cut off within 100ms after receiving command 0x31), the HDR sensor sampling rate changes from 100Hz to 50Hz, the electrochromic module maintains the transmittance of each zone, and a "stable state frame" is sent once every 2 seconds via the CAN bus. During the tunnel exit phase (80-80.6 seconds): The electrochromic module receives the "gradual dimming command" and adjusts the PWM duty cycle every 0.3 seconds (65%→60%→55%). After each adjustment, the STM809 samples and confirms that the transmittance deviation is ≤1%. The decision module simultaneously sends a "Tunnel mode exiting" message to the instrument panel (ID0x18F00301, data segment "0237", "37" represents backlight 55%). Full-process coordinated operation: The vehicle speed sensor updates the vehicle speed every 100ms. If the speed is ≤35km / h, the duty cycle of zones 5 and 8 is temporarily increased to 67%. The CAN bus module forwards signals from each device in real time to ensure no data loss.

[0098] In this specific embodiment, dual-light camera ROI cropping reduces decision latency by 20%, electrochromic dimming completes dimming in 1.2 seconds, eliminating the "tunnel black hole effect," and the recognition distance for vehicles ahead increases from 20 meters to 60 meters; dimming accuracy error is ≤2%, anti-glare in zones 4 and 7 is not too dark, and following distance in zones 5 and 8 is not blurry; drivers do not need to manually adjust windows / sun visors, improving focus by 50%, and reducing the risk of tunnel accidents during morning rush hour by 40%.

[0099] In this embodiment of the invention, a multi-source data acquisition architecture consisting of a dual-light intelligent camera (visible light + infrared) + 4-channel distributed HDR light intensity sensors + vehicle speed sensor is adopted, along with preprocessing logic of wavelet transform noise reduction + CLAHE feature enhancement + 3σ outlier removal, to solve the shortcomings of existing single light sensors that "cannot capture low-light details and glare locations." A multi-core attention feature fusion module is embedded on the YOLO11 platform (spatial attention focuses on glare / target areas, channel attention enhances infrared / visible light features), combined with dual outputs of "scene classification + target localization," and a transmittance decision model using a fully connected neural network to achieve "scene-location-transmittance." Precise mapping of "light rate"; the windshield is divided into 9 independent control zones, using "all-solid-state electrochromic film (WO3 / LiAlO2 / NiO) + ITO transparent electrode + PWM drive (TLC5940 chip)", supporting stepless light transmission adjustment from 0.1% to 99.8%, and has a built-in fault mechanism of "voltage / current monitoring + adjacent zone compensation"; data interaction is realized through the CAN bus to "multimodal perception module → deep learning decision module → electrochromic execution module → vehicle AFL system", forming a complete closed loop of "ambient light acquisition → scene decision → light transmission adjustment → AFL collaboration → status feedback", rather than the fragmented optimization of existing technologies.

[0100] The method steps of the embodiments of the present invention have been described above. It is understood that the embodiments of the present invention identify the current scene type based on the line-of-sight overlap rate and scene dynamic factors, and adjust the transmittance of multiple sections of the windshield according to the current scene type and ambient light intensity information, thereby improving the accuracy and timeliness of the vehicle's windshield adjustment, and thus improving vehicle driving safety and the user's driving experience.

[0101] Compared with the prior art, the embodiments of the present invention have the following advantages: 1) Significantly improved safety performance 1. Leap in Nighttime Recognition Capability: Relying on the infrared image acquisition of the dual-light intelligent camera and the design of "30% enhanced infrared transmittance" of electrochromic glass, the nighttime road target (pedestrian, obstacle) recognition distance has been increased from 30-50 meters in the existing technology to 150 meters, with sufficient buffer time reserved for emergency braking reaction time, and the risk of light-related accidents has been reduced by more than 30%.

[0102] 2. Glare interference is completely eliminated: By “scene recognition and glare area location + zone dimming”, the visual blind spot elimination rate in strong light / oncoming scenarios is ≥95%, and the incidence of driver “dazzle” is reduced to below 0.3%, especially solving the problem of strong light interference from oncoming vehicles’ illegal high beams and modified headlights.

[0103] 2) Optimization of adjustment accuracy and adaptability 1. High scene recognition accuracy: The improved YOLO11 architecture embeds a multi-core attention module, achieving a recognition accuracy of ≥98.2% for three types of scenes: "low light / strong light / passing traffic". This is far superior to the fuzzy judgment mode of the first generation of intelligent dimming that "only judges the light intensity threshold" and the false adjustment rate is <1.8%.

[0104] 2. Controllable dimming precision in different zones: The 3×3 zone electrochromic glass supports stepless light transmission adjustment from 0.1% to 99.8%, with local dimming precision reaching 0.1%. It can achieve differentiated control of "10%-20% light transmittance in the direct light area and 50%-60% in the non-direct light area", taking into account both anti-glare needs and overall field of vision, and avoiding the problem of darkening of the field of vision caused by overall dimming.

[0105] 3) Improved response efficiency and ease of operation 1. Fast decision-making and execution speed: Multimodal data preprocessing (wavelet transform noise reduction) and deep learning decision module work together to achieve a response time of ≤100ms for scene recognition and transmittance calculation; With PWM drive technology, the dimming time of electrochromic glass is ≤2 seconds, which is 5 times faster than the first generation of intelligent dimming (>500ms decision), and can respond to transient scenarios such as tunnel entrances and exits and passing vehicles in real time.

[0106] 2. No manual intervention required: The fully automated "perception-decision-execution" process eliminates the distraction of manually adjusting the sun visor and anti-glare rearview mirror, improving driving focus by 40%, which is in line with the development trend of "zero operational burden" in automotive intelligent cockpits.

[0107] 4) The advantages of system synergy and energy efficiency are highlighted. 1. Forming a complete technology closed loop: Data interaction with adaptive headlights (AFL) is realized through CAN bus. When the system determines "nighttime low light scene", "AFL high beam mode + windshield infrared light transmission enhancement" is triggered simultaneously, forming a dual optimization of "light source-medium", and the light adaptation effect is improved by 60% compared with fragmented technology.

[0108] 2. Significantly reduced energy consumption: The all-solid-state electrochromic film has the characteristics of "power consumption ≤5W during the adjustment phase and zero power consumption during the maintenance phase". Compared with the traditional sunshade motor drive (average daily power consumption of 12Wh), the average daily energy consumption is reduced to below 2.4Wh, saving 80% of energy. At the same time, it is compatible with fuel vehicles and new energy vehicles, and can be driven by a 12V vehicle power supply, with strong adaptability.

[0109] like Figure 2 The diagram shown is a structural schematic of a deep learning-based vehicle windshield adjustment system provided in an embodiment of the present invention. (Refer to...) Figure 2 This invention provides a deep learning-based vehicle windshield adjustment system, comprising: The multimodal perception module is used to identify light source feature information based on the image in front of the vehicle, identify the driver's eye feature information based on the driver image, and obtain ambient light intensity information and vehicle speed information. The data fusion module is used to determine the line-of-sight overlap rate between the driver's eyes and the light source based on light source feature information and eye feature information, determine the scene dynamic factor based on ambient light intensity information and vehicle speed information, and generate a fused feature vector based on the line-of-sight overlap rate and scene dynamic factor. The scene recognition module is used to input the fused feature vector into a pre-trained scene recognition model to obtain the current scene type; The adjustment module is used to determine the target transmittance of multiple zones of the windshield based on the current scene type and ambient light intensity information, and generate corresponding zone adjustment instructions based on the target transmittance, and then adjust the windshield according to the zone adjustment instructions.

[0110] It is understood that the content of the above method embodiments is applicable to this system embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0111] This invention also provides an electronic device, comprising: a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for communication between the processor and the memory. When the program is executed by the processor, it implements the aforementioned deep learning-based vehicle windshield adjustment method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0112] like Figure 3 The diagram shown is a hardware structure schematic of an electronic device provided in an embodiment of the present invention. (Refer to...) Figure 3 This invention provides an electronic device, comprising: The processor 301 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present invention. The memory 302 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 302 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 302 and is called and executed by the processor 301 to implement the deep learning-based vehicle windshield adjustment method of the embodiments of this invention. Input / output interface 303 is used to implement information input and output; The communication interface 304 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 305 transmits information between various components of the device (e.g., processor 301, memory 302, input / output interface 303, and communication interface 304); The processor 301, memory 302, input / output interface 303, and communication interface 304 are connected to each other within the device via bus 305.

[0113] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0114] This invention also provides a vehicle that includes the electric drive assembly of the aforementioned deep learning-based vehicle windshield adjustment system or electronic device.

[0115] The vehicle can be a private car, such as a sedan, SUV, MPV, or pickup truck. It can also be a commercial vehicle, such as a van, bus, small truck, or large semi-trailer. The vehicle must have an electric motor capable of outputting power or acting as a generator to store mechanical energy. When the vehicle is a new energy vehicle, it can be a hybrid or a pure electric vehicle.

[0116] Since the vehicle applies all the technical solutions of the above-mentioned deep learning-based vehicle windshield adjustment system or electronic device, it has at least all the beneficial effects brought about by the technical solutions of the above embodiments, which will not be repeated here.

[0117] This invention also provides a storage medium, which is a computer-readable storage medium for computer-readable storage. The storage medium stores one or more programs, which can be executed by one or more processors to implement the above-described deep learning-based vehicle windshield adjustment method.

[0118] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0119] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0120] This invention also provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, causing the computer device to perform... Figure 1 The method shown.

[0121] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0122] The embodiments described in this invention are for the purpose of more clearly illustrating the technical solutions of the embodiments of this invention, and do not constitute a limitation on the technical solutions provided by the embodiments of this invention. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this invention are also applicable to similar technical problems.

[0123] The terms "first," "second," "third," "fourth," etc. (if present) in the specification and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0124] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the aforementioned blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this invention are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and sub-operations described as part of a larger operation are executed independently.

[0125] Furthermore, although the invention has been described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the aforementioned functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the invention. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional skill of an engineer. Therefore, those skilled in the art can implement the invention as set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of the invention, which is determined by the full scope of the appended claims and their equivalents.

[0126] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0127] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0128] More specific examples (a non-exhaustive list) of computer-readable media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the aforementioned program can be printed, because the aforementioned program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0129] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0130] In the foregoing description of this specification, references to terms such as "one embodiment," "another embodiment," or "some embodiments" indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of the present invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0131] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0132] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A method for adjusting a vehicle windshield based on deep learning, characterized in that, Includes the following steps: The system identifies light source characteristics based on the image in front of the vehicle, identifies the driver's eye characteristics based on the image of the driver, and obtains ambient light intensity and vehicle speed information. The driver's eye-to-eye line overlap rate with the light source is determined based on the light source feature information and the eye feature information. The scene dynamic factor is determined based on the ambient light intensity information and the vehicle speed information. A fusion feature vector is generated based on the line-of-eye line overlap rate and the scene dynamic factor. The fused feature vector is input into a pre-trained scene recognition model to obtain the current scene type; The target transmittance of multiple zones of the windshield is determined based on the current scene type and the ambient light intensity information. Corresponding zone adjustment instructions are generated based on the target transmittance, and the windshield is adjusted according to the zone adjustment instructions.

2. The method for adjusting a vehicle windshield based on deep learning according to claim 1, characterized in that, The process of identifying light source feature information based on the image in front of the vehicle, identifying driver eye feature information based on the driver image, and obtaining ambient light intensity information and vehicle speed information specifically includes: The light source contour is extracted from the image in front of the vehicle, and the light source coordinates and spot area are determined based on the light source contour to obtain the light source feature information. The driver's eye image is extracted from the driver's image, and the coordinates of both eyes and the pupil diameter are extracted from the driver's eye image to obtain the eye feature information; Obtain the illuminance values ​​of the four corners of the windshield, and determine the ambient light intensity information based on the average of the illuminance values; The vehicle speed information is obtained by converting the vehicle speed based on the pulse frequency of the vehicle speed sensor.

3. The method for adjusting a vehicle windshield based on deep learning according to claim 2, characterized in that, The scene dynamic factors include the light intensity abrupt change rate and the vehicle speed change rate. The process involves determining the eye-to-eye overlap rate between the driver's eyes and the light source based on the light source feature information and the eye feature information, determining the scene dynamic factors based on the ambient light intensity information and the vehicle speed information, and generating a fused feature vector based on the eye-to-eye overlap rate and the scene dynamic factors. Specifically, this includes: The line-of-sight overlap rate is calculated based on the light source coordinates, the light spot area, the binocular coordinates, and the pupil diameter. The light intensity abrupt change rate is determined based on the ambient light intensity information, and the vehicle speed change rate is determined based on the vehicle speed information; The line-of-sight overlap rate, the light intensity abrupt change rate, and the vehicle speed change rate are time-aligned and time-series fused to obtain the fused feature vector.

4. The method for adjusting a vehicle windshield based on deep learning according to claim 1, characterized in that, The scene recognition model is trained through the following steps: Obtain the line-of-sight overlap rate and dynamic factor of the sample scene in the test scenario, and determine the corresponding scene type label through manual annotation; A sample fusion feature vector is generated based on the sample line-of-sight overlap rate and the sample scene dynamic factor. The sample fusion feature vector is used as a training sample, and the scene type label is used as the corresponding sample label to obtain a training dataset. The training samples are input into a pre-built YOLO11 neural network to obtain the predicted scene type; The loss value is determined based on the predicted scenario type and the scenario type label; The parameters of the YOLO11 neural network are updated based on the loss value to obtain the trained scene recognition model; The scene type tags include daytime strong light scenes, nighttime oncoming traffic scenes, tunnel entrance / exit scenes, and evening backlight scenes.

5. The method for adjusting a vehicle windshield based on deep learning according to claim 1, characterized in that, The step of determining the target light transmittance of multiple zones of the windshield based on the current scene type and the ambient light intensity information specifically includes: Based on the current scene type, obtain the pre-calibrated transmittance adjustment strategy for the corresponding scene; The light intensity-transmittance intensity mapping relationship of each zone of the windshield is determined according to the light transmittance adjustment strategy. The target transmittance of each zone is determined based on the ambient light intensity information and the light intensity-transmittance mapping relationship.

6. A method for adjusting a vehicle windshield based on deep learning according to any one of claims 1 to 5, characterized in that, Each section of the windshield is equipped with individually adjustable electrochromic glass. The process involves generating corresponding section adjustment commands based on the target light transmittance, and then adjusting the windshield according to these commands. Specifically: The target driving voltage is determined based on the target transmittance, the partition adjustment command is generated based on the partition number of each partition and the corresponding target driving voltage, and the partition adjustment command is sent to the PWM controller; The PWM controller generates PWM pulse signals corresponding to each partition according to the partition adjustment command, and the driver chip converts the PWM pulse signals into DC voltages for the corresponding partitions, thereby driving the electrochromic glass of the corresponding partitions to adjust the light transmittance through the DC voltages.

7. A vehicle windshield adjustment system based on deep learning, characterized in that, include: The multimodal perception module is used to identify light source feature information based on the image in front of the vehicle, identify the driver's eye feature information based on the driver image, and obtain ambient light intensity information and vehicle speed information. The data fusion module is used to determine the line-of-sight overlap rate between the driver's eyes and the light source based on the light source feature information and the eye feature information, determine the scene dynamic factor based on the ambient light intensity information and the vehicle speed information, and generate a fusion feature vector based on the line-of-sight overlap rate and the scene dynamic factor. The scene recognition module is used to input the fused feature vector into a pre-trained scene recognition model to obtain the current scene type; The adjustment module is used to determine the target light transmittance of multiple zones of the windshield based on the current scene type and the ambient light intensity information, generate corresponding zone adjustment instructions based on the target light transmittance, and then adjust the windshield according to the zone adjustment instructions.

8. An electronic device, characterized in that, The electronic device includes a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for enabling communication between the processor and the memory. When the program is executed by the processor, it implements the deep learning-based vehicle windshield adjustment method as described in any one of claims 1 to 6.

9. A vehicle, characterized in that, The vehicle includes the deep learning-based vehicle windshield adjustment system as described in claim 7 or the electronic device as described in claim 8.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the deep learning-based vehicle windshield adjustment method as described in any one of claims 1 to 6.

Images