Adaptive headlight control method and adaptive headlight system

Abstract

This invention relates to a control method for adaptive headlights. The method includes: acquiring a target distance in real time; using an optimal angle control model to map the target distance to obtain the corresponding headlight angle and light level; and controlling the headlights according to the headlight angle and light level. The optimal angle control model is a fitting function that maps distance to the headlight angle, and outputs different light levels based on the distance and headlight angle. This adaptive headlight control method, by acquiring the target distance in real time and directly mapping the corresponding light level and optimal headlight angle through the optimal angle control model, instantly adjusts the headlights and rapidly responds to environmental changes, improving the system's dynamic adaptability. Furthermore, this invention can be easily deployed to embedded systems and possesses high real-time performance.

Description

Technical Field

[0001] The invention relates to the field of automobile adaptive headlights, and in particular to a control method for adaptive headlights, an adaptive headlight system, an electronic device and a computer storage medium. Background Art

[0002] Traditional headlights switch between high and low beams and adjust the angle of the headlights manually after the driver observes the surrounding environment. However, in low-light environments, especially when driving at high speeds, when the driver is concentrating on observing the environment, there are certain safety hazards in manually switching between high and low beams and adjusting the angle of the headlights. In addition, improper use of manually adjusted high beams can easily cause glare to oncoming vehicle drivers, pedestrians or cyclists, and improper adjustment of the headlight angle may create blind spots, increasing the driver's driving risks.

[0003] In order to solve the safety hazards or errors caused by the need for manual adjustment of car headlights, the existing technology usually combines deep learning algorithms with car headlight control systems to realize the adaptive front-lighting system (Adaptive Front-lighting System, AFS) of automobile vehicles to solve the problem of manual adjustment of car headlights. However, in practical applications, the use of multi-modal deep learning algorithms or high-precision deep learning algorithms usually has high computational complexity and amount of calculation, resulting in defects such as difficulty in deployment of embedded systems and reduced real-time performance. Summary of the invention

[0004] Based on this, an object of the present invention is to provide a control method for an adaptive headlamp.

[0005] A method for controlling an adaptive headlamp comprises the following steps:

[0006] S1. Get the target distance in real time;

[0007] S2. Use the optimal angle control model to convert the mapping relationship of the target distance to obtain the corresponding headlamp angle and light gear; and control the headlamp according to the headlamp angle and light gear; wherein the optimal angle control model is a fitting function of distance-mapping headlamp angle, and outputs different light gears according to the distance and the headlamp angle, specifically:

[0008] Suppose there are n light levels, then:

[0009] f(x)=θ

[0010] light=α n ,if x∈[d n-1 ,d n ]and θn-1 ≤θ≤θ n

[0011] Where f(x) is the optimal angle control model for the input target distance x; θ is the output optimal angle corresponding to the input x; light = α n is the nth light level; [d n-1 ,d n ] is the light level α n The target distance range used below; θ n-1 ≤θ≤θ n For the light level α n The optimal angle control model outputs the headlamp angle range.

[0012] The control method of the adaptive headlamp described in the present invention directly maps the corresponding light gear and the optimal headlamp angle through the target distance acquired in real time and the optimal angle control model, thereby adjusting the headlamp in real time and responding quickly to environmental changes, thereby improving the dynamic adaptability of the system; at the same time, the optimal angle control model proposed in the present invention is a function model based on mathematical modeling, which has good scalability and can be expanded according to different sensors and data sources to construct a multi-modal control model, thereby adapting to new environments and needs and improving the control flexibility and adaptability of the adaptive headlamp.

[0013] Furthermore, the steps of constructing the optimal angle control model are as follows:

[0014] T1, obtaining video data of headlights with different light intensities but the same travel mode;

[0015] The video data is a front view image of a test vehicle's driving path captured by a camera;

[0016] T2, perform target detection on each frame of the video data to obtain confidence data of several correct detection results;

[0017] T3. Perform data cleaning on the confidence data of a number of correct detection results to obtain a number of cleaned confidence data;

[0018] T4. By fitting a function to a number of confidence data of cleaning, a series of continuous curves are obtained, and according to the same travel mode of the test vehicle, the frame number is converted into distance to obtain the mapping relationship between distance and confidence;

[0019] T5. Screen the distance-confidence mapping relationship according to a confidence threshold to obtain several optimal angle key points;

[0020] T6. Perform function fitting on several optimal angle key points to obtain the optimal angle control model.

[0021] The present invention performs mathematical modeling by evaluating the influence of headlights with different light intensities on a target detection algorithm based on the confidence of the target detection algorithm, and ensures that the mathematical model is applied and deployed under high confidence, thereby ensuring the reliability and accuracy of the mathematical model.

[0022] Furthermore, the step T5 includes the following sub-steps:

[0023] T51, setting a confidence threshold Cof;

[0024] T52, screening the points whose confidence is greater than or equal to Cof in the distance-confidence mapping relationship as several candidate key points;

[0025] T53. Filter out the lowest angles among several candidate key points as several optimal angle key points, and the optimal angle key points satisfy:

[0026] θ=min{θ|c(d,θ,α)≥Cof}

[0027] Where c(d,θ,α) is used to represent the target detection confidence function at distance d, longitudinal lighting angle θ, and light level α.

[0028] The present invention selects points with confidence (ordinate) greater than or equal to Cof in the distance-confidence relationship curve as candidate key points, and selects key points corresponding to the minimum angle to ensure that the headlamp can provide sufficient detection confidence at different distances while minimizing the glare effect on others.

[0029] Furthermore, the target distance is obtained by using a camera device and a lightweight target detection algorithm, which specifically includes the following steps:

[0030] SA11: Recognize the image through a lightweight target detection algorithm to obtain the pixel coordinate information of the target;

[0031] SA12: locate the target through its pixel coordinate information to obtain the target's distance information;

[0032] SA13: A sensor is used to measure distance and perform data compensation according to the distance information of the target to obtain the target distance.

[0033] The present invention obtains the accurate distance to the target by combining a lightweight target detection algorithm to further improve the intelligence of the adaptive headlamp, and reduces the calculation amount of the target detection algorithm and ensures the accuracy of the target distance through sensor ranging compensation to fully reduce the implementation cost.

[0034] Furthermore, the target distance is acquired by a distance measuring sensor, thereby reducing calculation costs.

[0035] An adaptive headlamp system comprises a headlamp, a motor, a light gear adjustment mechanism, a distance measurement module, and a controller electrically and / or communicatively connected to the motor, the light gear adjustment mechanism, and the distance measurement module;

[0036] The motor is used to control the longitudinal lighting angle of the headlamp according to the angle of the headlamp;

[0037] The light gear adjustment mechanism is used to control the light gear of the headlamp according to the light gear;

[0038] The controller is used to obtain the target distance in real time through the distance measurement module; and use the optimal angle control model to convert the mapping relationship of the target distance to obtain the corresponding headlamp angle and light gear; wherein the optimal angle control model is a fitting function of distance-mapping headlamp angle, and outputs different light gears according to the distance and the headlamp angle, specifically:

[0039] Suppose there are n light levels, then:

[0040] f(x0=θ

[0041] light=α n ,if x∈[d n-1 ,d n ]and θ n-1 ≤θ≤θ n

[0042] Where f(x) is the optimal angle control model for the input target distance x; θ is the output optimal angle corresponding to the input x; light = α n is the nth light level; [d n-1 ,d n ] is the light level α n The target distance range used below; θ n-1 ≤θ≤θ n For the light level α n The optimal angle control model outputs the headlamp angle range.

[0043] Furthermore, the steps for obtaining the optimal angle control model are as follows:

[0044] T1, obtaining video data of headlights with different light intensities but the same travel mode;

[0045] Wherein, the video data is a front view image of a test vehicle's driving path collected by a shooting device;

[0046] T2, perform target detection on each frame of the video data to obtain confidence data of several correct detection results;

[0047] T3. Perform data cleaning on the confidence data of a number of correct detection results to obtain a number of cleaned confidence data;

[0048] T4. By fitting a function to a number of confidence data of cleaning, a series of continuous curves are obtained, and according to the same travel mode of the test vehicle, the frame number is converted into distance to obtain the mapping relationship between distance and confidence;

[0049] T5. Screen the distance-confidence mapping relationship according to a confidence threshold to obtain several optimal angle key points;

[0050] T6. Perform function fitting on several optimal angle key points to obtain the optimal angle control model.

[0051] Furthermore, the step T5 includes the following sub-steps:

[0052] T51, setting a confidence threshold Cof;

[0053] T52, screening the points whose confidence is greater than or equal to Cof in the distance-confidence mapping relationship as several candidate key points;

[0054] T53. Filter out the lowest angles among several candidate key points as several optimal angle key points, and the optimal angle key points satisfy:

[0055] θ=min{θ|c(d,θ,α)≥Cof}

[0056] Where c(d,θ,α) is used to represent the target detection confidence function at distance d, longitudinal lighting angle θ, and light level α.

[0057] Furthermore, the distance measurement module is a camera device and a lightweight target detection algorithm, which obtains the target distance through the following steps:

[0058] SA11: Recognize the image through a lightweight target detection algorithm to obtain the pixel coordinate information of the target;

[0059] SA12: locate the target through its pixel coordinate information and obtain the target's distance information;

[0060] SA13: Based on the distance information of the target, a sensor is used to measure the distance and perform data compensation to obtain the target distance.

[0061] Furthermore, the distance measuring module is a distance measuring sensor.

[0062] For better understanding and implementation, the present invention is described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a simplified structural schematic diagram of the adaptive headlight system of the present invention;

[0064] Figure 2 for Figure 1 A simplified flowchart of a corresponding adaptive headlamp control method;

[0065] Figure 3 This is a flow chart for constructing the optimal angle control model of the present invention;

[0066] Figure 4 A schematic diagram of a simulation example of multiple distance-confidence curves;

[0067] Figure 5 Based on Figure 4 Schematic diagram of curve trend simulation example of the optimal angle control model;

[0068] Figure 6 The figure is a simplified schematic diagram of the implementation effect of the adaptive headlight system of the present invention. DETAILED DESCRIPTION

[0069] In order to solve the problems of high complexity, difficult embedded system deployment and low real-time performance of high-precision deep learning algorithms in the prior art, the present invention provides an adaptive headlight system, which includes an improved method for controlling the headlight position and angle. The control method adopts an optimal angle control model fitted by a function to convert the acquired target distance in real time to obtain the optimal angle and light position corresponding to the target distance, thereby realizing the adaptive headlight adjustment of the vehicle, reducing the algorithm complexity of the prior art, simplifying the embedded system deployment and improving the real-time performance of the adaptive headlight adjustment.

[0070] See also Figure 1 The adaptive headlamp system of the present invention includes a headlamp, a motor, a light gear adjustment mechanism, a distance measurement module, and a controller electrically and / or communicatively connected to the motor, the light gear adjustment mechanism, and the distance measurement module.

[0071] The motor may be a stepper motor, which controls the longitudinal lighting angle of the headlamp according to a given corresponding optimal headlamp angle (desired deflection angle), thereby achieving precise control of the headlamp angle.

[0072] The light level adjustment mechanism may be a pulse width modulation (PWM) to control the light level of the headlamp according to a given light level, specifically by adjusting the pulse width to change the light average current, so as to adjust the lighting intensity of the headlamp. In this embodiment, the light level includes at least a high beam and a low beam. However, since the headlamp control method corresponds to different scene requirements (fast response, accuracy or cost control, etc.), the selection of the control means is different, and accordingly, the present invention does not specifically limit the control means.

[0073] In one embodiment, the ranging module is a ranging sensor for measuring the distance to the target. Laser ranging, laser point cloud, ultrasonic sensor, etc. can be selected, or multiple sensors can be combined to obtain a more accurate target distance. Accordingly, the present invention does not specifically limit the selection of sensors.

[0074] In another embodiment, the distance measurement module is a camera device and a lightweight target detection algorithm, and the target image captured in real time by the camera device is processed by the lightweight target detection algorithm to obtain the target distance.

[0075] See also Figure 2 The controller obtains the light gear position and the optimal headlight angle corresponding to the target through the following steps.

[0076] S1: Obtain the target distance in real time through the distance measurement module.

[0077] In one embodiment, when the distance measurement module is used as a camera device and a lightweight target detection algorithm as a distance measurement module, the steps for obtaining the target distance are as follows:

[0078] SA11: Recognize the image through a lightweight target detection algorithm to obtain the pixel coordinate information of the target.

[0079] The lightweight target detection algorithm can be selected as YOLOv8 (You only look once Version 8), which identifies the target in the current frame and determines the bounding box of the target to obtain the pixel coordinate information of the target. The YOLOv8 target detection algorithm includes a feature extraction module (Cross Stage Partial Network, CSPNet), a feature fusion module (Path Aggregation Network, PANet) and a detection module, which are respectively used to extract features, fuse features and detect judgments, and output target categories, target bounding boxes and their corresponding confidence data.

[0080] The present invention takes pedestrians, traffic signs and vehicles as the primary detection targets, but the primary detection targets can be further increased or changed according to different environments and requirements, such as when applied to autonomous mobile devices, drones, etc., the target types preset by the user are detected. Therefore, the present invention does not specifically limit the target types to be detected.

[0081] SA12: Locate the target through its pixel coordinate information to obtain the target's distance information.

[0082] The positioning may be optionally monocular ranging, in which the bounding box of the target is identified according to the target detection algorithm, and the distance corresponding to the current target bounding box size is estimated in combination with the calibration data (size and focal length of the corresponding shooting device), thereby estimating the world coordinate position of the target as the target distance.

[0083] Furthermore, the method further includes step SA13: using a sensor to perform distance measurement and data compensation according to the distance information of the target to obtain the target distance, so as to obtain the target distance with higher accuracy. The data compensation can be performed by summing up the distances obtained by combining the distances measured by different sensors with the distance information of the current target and taking the average.

[0084] S2: Use the optimal angle control model to convert the input target distance into a mapping relationship to obtain the corresponding headlight angle and light level.

[0085] Specifically, the optimal angle control model is a fitting function of the distance mapping headlight angle, and has a light gear condition. Different light gears will be output according to the relationship between the distance and the headlight angle. For example, if there are n light gears, the following expression can be used as a reference:

[0086] f(x)=θ

[0087] light=α n ,if x∈[d n-1 ,d n ]and θ n-1 ≤θ≤θ n

[0088] Where f(x) is the optimal angle control model for the input target distance x; θ is the output optimal angle corresponding to the input x; α n is the nth light level; [d n-1 ,d n ] is to use the current light level α n The target distance range can be set to multiple target distance intervals according to different headlamp angles; θ n-1 ≤θ≤θ n For the light level α n The optimal angle control model outputs the headlamp angle range.

[0089] The present invention adopts a function model based on mathematical modeling to obtain the control parameters of the headlights, so that the control of the adaptive headlights can quickly respond to environmental changes, thereby improving driving safety and comfort. In addition, the function model based on mathematical modeling has good scalability and can be expanded according to different sensors and data sources to build a multi-modal control model to adapt to new environments and requirements, thereby improving the control flexibility and adaptability of the adaptive headlights.

[0090] In order to ensure that the optimal headlamp angle and light level output by the optimal angle control model can effectively reduce the glare effect on other vehicles or pedestrians, and ensure that the driver is in the maximum visual range in a low-light environment, the construction steps of the optimal angle control model are as follows:

[0091] T1: Acquire video data of headlights with different light intensities but the same travel pattern.

[0092] Specifically, the video data is a front view image of the test vehicle's driving path captured by a camera, and the test vehicle starts from a fixed starting point and drives along the same route and travel mode (such as speed, acceleration, etc.). The light intensity is a combination of different illumination angles and light levels of the test vehicle's headlights.

[0093] In order to ensure the consistency and comparability of the collected video data, the present invention only adjusts the light intensity as the only variable, and all video data have the same resolution and frame rate to evaluate the impact of different headlights on the driver's visibility.

[0094] In addition to using light intensity as the only variable, video data can be further collected under multiple weather conditions or through dynamic lighting simulation to simulate the impact of more variables on the light intensity of the test vehicle's headlights.

[0095] T2: Perform target detection on each frame of the video data to obtain confidence data of several correct detection results.

[0096] Specifically, the video data is grouped according to the light intensity and synchronously framed to obtain a plurality of groups of video frames for maintaining temporal and spatial consistency between frames.

[0097] Target detection is performed on several groups of video frames to identify the target categories in the frames, such as vehicles, pedestrians or traffic signs, and the incorrectly identified target categories are filtered out to obtain the confidence data of several groups of correct detection results.

[0098] The confidence data is a numerical value output by the target detection model, which is used to indicate the similarity between the detected category and the actual category of the target in the current frame, thereby representing the visibility of the frame.

[0099] Wherein, the target detection can be performed by a model based on deep learning, such as YOLOv8 (You Only LookOnce version 8). The model includes a feature extraction module, a feature fusion module and a detection module, which are respectively used to extract features, fuse features and detect judgments, and output target categories and their confidence data. The present invention uses confidence data to evaluate the impact of different light intensities on target detection. However, since the choice of target detection models is diverse and can be adjusted according to needs (different accuracy), the present invention does not limit the specific model.

[0100] T3: Perform data cleaning on the confidence data of several correct detection results to obtain several cleaned confidence data.

[0101] Specifically, by processing the confidence data of several groups of correct detection results, several groups of confidence scatter plots are obtained. The horizontal axis of the scatter plot is the number of frames, and the vertical axis is the confidence data of the correct detection results.

[0102] Next, outlier detection is performed on the several groups of confidence scatter plots to obtain abnormal data.

[0103] Data cleaning is performed on several groups of confidence scatter plots according to abnormal data to obtain several cleaned confidence data.

[0104] Among them, the outlier detection can be selected based on algorithms such as distance, similarity, probability or clustering, such as: Local Outlier Factor (LOF), Isolation Forest, Clustering Algorithm (K-means) or Proximity Algorithm (K Nearest Neighbor), etc. Accordingly, the present invention does not specifically limit the means of outlier detection.

[0105] Due to accidental factors or environmental interference, points that deviate from the overall data distribution pattern may appear in the confidence scatter plot. These abnormal points need to be removed through the data cleaning process to ensure that the data reflects the general rules in common scenarios.

[0106] T4: Perform function fitting on several cleaned confidence data to obtain the mapping relationship between distance and confidence.

[0107] Specifically, a series of continuous curves are obtained by performing function fitting on a number of cleaned confidence data, wherein the curves are used to represent the trend of the confidence changing with the frame number under different light intensities.

[0108] Next, according to the fixed travel pattern of the test vehicle, the number of frames is converted into distance, so as to obtain the relationship curve between distance and confidence, that is, the mapping relationship between distance and confidence.

[0109] Since the test vehicle travels at a constant speed and a fixed route, the distance corresponding to the target in each frame of the image can be calculated based on the speed and time. The function fitting can be selected as polynomial fitting or other fitting methods to generate a continuous confidence curve. The present invention does not specifically limit the means of function fitting.

[0110] T5: Filter the distance-confidence mapping relationship according to a confidence threshold to obtain several optimal angle key points.

[0111] Specifically, a confidence threshold Cof is set to control the confidence of the detection results to be greater than Cof.

[0112] In the mapping relationship between screening distance and confidence, the points whose ordinates are greater than or equal to Cof are selected as candidate key points.

[0113] And select the lowest angle among the candidate key points as several optimal angle key points. The expression can be referred to as:

[0114] θ=min{θ|c(d,θ,α)≥Cof}

[0115] Among them, c(d,θ,α) is used to represent the target detection confidence function at distance d, longitudinal lighting angle θ and light level α.

[0116] The Cof value may be selected as 0.8, which is the accuracy closest to human eye recognition. However, due to different road conditions, detection accuracy, safety level and other requirements, different confidence thresholds may be selected to control the confidence of the detection result to always be greater than the threshold. Therefore, the present invention does not specifically limit the selection of the Cof value.

[0117] Since each light level and longitudinal lighting angle has a continuous distance-confidence relationship curve, and these curves are in the same coordinate graph, forming a cluster of curves with the same trend, please refer to Figure 4 ,in, Figure 4 The example is drawn by simulating sampling data, which is not real data and is for reference only. However, within a certain distance range, there will be multiple curves with confidence higher than the Cof threshold. By selecting the light level and longitudinal lighting angle corresponding to the curve with the lowest angle value, it is ensured that the headlamp can provide sufficient detection confidence at different distances while minimizing the glare to others.

[0118] T6: Perform function fitting on several optimal angle key points to obtain the optimal angle control model.

[0119] Specifically, the optimal angle key points corresponding to the distance are fitted and drawn to obtain a continuous distance-optimal angle curve, which is the optimal angle control model. The specific trend of the curve can be referred to Figure 5 .

[0120] Among them, the optimal angle control model is consistent with step S2, and the present invention will not repeat it here.

[0121] In addition to obtaining the distance-correct target confidence curve through the target detection model and obtaining the optimal angle control model by mathematical modeling, the present invention can also use multi-sensor data as the input of the multimodal fusion target detection model, so that the distance-correct target confidence curve obtained by the fusion detection method can be expanded as data in the mathematical modeling process to replace or supplement the distance-optimal angle control model used in the present invention, thereby improving the scalability and accuracy of the optimal angle control model.

[0122] Compared with the prior art, the present invention simplifies the deployment process of the adaptive headlight system by combining the target detection method with mathematical modeling, improves the accuracy and reliability of data processing, and ensures driving safety and the response speed of the adaptive headlight system. In addition, the present invention isolates the target detection model by mathematical modeling, so that users can choose a multi-modal target detection model with high computational load according to different needs to adapt to more complex scenarios and needs, thereby providing good expansion capabilities to cope with a wider range of application scenarios and needs.

[0123] At the same time, the optimal angle control model described in the present invention is deployed to the adaptive headlight system in an offline manner to ensure that the vehicle is in areas such as tunnels and suburbs that lack signal amplifiers and base stations, and can also maintain the adaptive headlight system to provide high-precision, high-real-time adaptive adjustment of the headlights.

[0124] In addition, with the development of target detection algorithms, multimodal fusion target detection methods with multi-sensor data as input have gradually become mainstream. Based on this, the distance-correct target confidence curve obtained by fusing multiple different visual target detection methods can be easily expanded as data in the modeling process, and used to replace or supplement the current optimal angle model, that is, the distance-optimal angle curve, to fully improve the scalability and accuracy of the function model.

[0125] Based on the same inventive concept, the present application also provides an electronic device, which may be a terminal device such as a server, a desktop computing device or a mobile computing device (e.g., a laptop computing device, a handheld computing device, a tablet computer, a netbook, etc.). The device includes one or more processors and a memory, wherein the processor is used to execute a program to implement the control method of the adaptive headlight of the embodiment of the present invention; and the memory is used to store a computer program executable by the processor.

[0126] Based on the same inventive concept, the present application also provides a computer-readable storage medium, corresponding to the embodiment of the aforementioned method for controlling an adaptive headlight, wherein the computer-readable storage medium stores a computer program thereon, and when the program is executed by a processor, the steps of the method for controlling an adaptive headlight recorded in any of the aforementioned embodiments are implemented.

[0127] The present application may take the form of a computer program product implemented on one or more storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing program code. Computer-usable storage media include permanent and non-permanent, removable and non-removable media, and information storage can be achieved by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include but are not limited to: phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, tape disk storage or other magnetic storage devices or any other non-transmission medium that can be used to store information that can be accessed by a computing device.

[0128] The above-mentioned embodiments only express several implementation methods of the present invention, and the description is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, and the present invention is also intended to include these modifications and modifications.

Claims

1. A method for controlling an adaptive headlamp, characterized in that: The following steps are involved: S1. Get the target distance in real time; S2. Using the optimal angle control model to perform mapping conversion on the target distance to obtain the corresponding headlamp angle and light level; The headlight is controlled according to the headlight angle and the light level; wherein the optimal angle control model is a fitting function of the distance-mapped headlight angle, and different light levels are output according to the distance and the headlight angle, specifically: Suppose there are n light levels, then: f(x)=θ light=α n ,if x∈[d n-1 ,d n ]andθ n-1 ≤θ≤θ n Where f(x) is the optimal angle control model for the input target distance x; θ is the output optimal angle corresponding to the input x; light = α n is the nth light level; [d n-1 ,d n ] is the light level α n The target distance range used below; θ n-1 ≤θ≤θ n In the light position Δ n Under the optimal angle control model, the headlamp angle range outputted; The steps for obtaining the optimal angle control model are as follows: T1, obtaining video data of headlights with different light intensities but the same travel mode; Wherein, the video data is a front view image of a test vehicle's driving path collected by a shooting device; T2, perform target detection on each frame of the video data to obtain confidence data of several correct detection results; T3. Perform data cleaning on the confidence data of a number of correct detection results to obtain a number of cleaned confidence data; T4. By fitting a function to a number of confidence data of cleaning, a series of continuous curves are obtained, and according to the same travel mode of the test vehicle, the frame number is converted into distance to obtain the mapping relationship between distance and confidence; T5. Screening the distance-confidence mapping relationship according to a confidence threshold to obtain a number of optimal angle key points; wherein the step T5 includes the following sub-steps: T51, setting a confidence threshold Cof; T52, screening the points whose confidence is greater than or equal to Cof in the distance-confidence mapping relationship as several candidate key points; T53. Filter out the lowest angles among several candidate key points as several optimal angle key points, and the optimal angle key points satisfy: θ=min{θ|c(d,θ,α)≥Cof} Where c(d,θ,α) is used to represent the target detection confidence function at distance d, longitudinal lighting angle θ and light level α; T6. Perform function fitting on several optimal angle key points to obtain the optimal angle control model.

2. The method for controlling an adaptive headlamp according to claim 1, characterized in that: The target distance is obtained by a camera device and a lightweight target detection algorithm, which specifically includes the following steps: SA11: Recognize the image through a lightweight target detection algorithm to obtain the pixel coordinate information of the target; SA12: locate the target through its pixel coordinate information to obtain the target's distance information; SA13: A sensor is used to measure distance and perform data compensation according to the distance information of the target to obtain the target distance.

3. The method for controlling an adaptive headlamp according to any one of claims 1 to 2, characterized in that: The target distance is obtained by a distance measuring sensor.

4. An adaptive headlight system, characterized in that: It includes a headlamp, a motor, a light gear adjustment mechanism, a distance measurement module, and a controller electrically and / or communicatively connected to the motor, the light gear adjustment mechanism, and the distance measurement module; The motor is used to control the longitudinal lighting angle of the headlamp according to the angle of the headlamp; The light gear adjustment mechanism is used to control the light gear of the headlamp according to the light gear; The controller is used to obtain the target distance in real time through the distance measurement module; and use the optimal angle control model to convert the mapping relationship of the target distance to obtain the corresponding headlamp angle and light gear; wherein the optimal angle control model is a fitting function of distance-mapping headlamp angle, and outputs different light gears according to the distance and the headlamp angle, specifically: Suppose there are n light levels, then: f(x)=θ light=α n ,if x∈[d n-1 ,d n ]andθ n-1 ≤θ≤θ n Where f(x) is the optimal angle control model for the input target distance x; θ is the output optimal angle corresponding to the input x; light = α n is the nth light level; [d n-1 ,d n ] is the light level Δ n The target distance range used below; θ n-1 ≤θ≤θ n For the light level α n Under the optimal angle control model, the headlamp angle range outputted; The steps for obtaining the optimal angle control model are as follows: T1, obtaining video data of headlights with different light intensities but the same travel mode; Wherein, the video data is a front view image of a test vehicle's driving path collected by a shooting device; T2, perform target detection on each frame of the video data to obtain confidence data of several correct detection results; T3. Perform data cleaning on the confidence data of a number of correct detection results to obtain a number of cleaned confidence data; T4. By fitting a function to a number of confidence data of cleaning, a series of continuous curves are obtained, and according to the same travel mode of the test vehicle, the frame number is converted into distance to obtain the mapping relationship between distance and confidence; T5. Screening the distance-confidence mapping relationship according to a confidence threshold to obtain a number of optimal angle key points; wherein the step T5 includes the following sub-steps: T51, setting a confidence threshold Cof; T52, screening the points whose confidence is greater than or equal to Cof in the distance-confidence mapping relationship as several candidate key points; T53. Filter out the lowest angles among several candidate key points as several optimal angle key points, and the optimal angle key points satisfy: θ=min{θ|c(d,θ,α)≥Cof} Where c(d,θ,α) is used to represent the target detection confidence function at distance d, longitudinal lighting angle θ and light level α; T6. Perform function fitting on several optimal angle key points to obtain the optimal angle control model.

5. The adaptive front lighting system according to claim 4, characterized in that: The distance measurement module is a camera device and a lightweight target detection algorithm, which obtains the target distance through the following steps: SA11: Recognize the image through a lightweight target detection algorithm to obtain the pixel coordinate information of the target; SA12: locate the target through its pixel coordinate information and obtain the target's distance information; SA13: Based on the distance information of the target, a sensor is used to measure the distance and perform data compensation to obtain the target distance.

6. The adaptive headlight system according to any one of claims 4 to 5, characterized in that: The distance measuring module is a distance measuring sensor.

Images