An on-vehicle lamp and its intelligent light-emitting control method

Abstract

The present invention relates to the technical field of image processing, and specifically to an in-vehicle lamp and its intelligent light control method, which includes the following steps: an image acquisition and preprocessing module, where a vehicle camera captures continuous image data, formats the image, and generates formatted image data; the formatted image data is optimized through contrast adjustment to generate optimized image data. The present invention captures continuous image data through a vehicle camera, implements formatting and contrast adjustment optimization on the image, improves the accuracy of target detection, and reduces interference factors caused by changes in ambient light; performs target detection on the basis of the optimized image, real-time identifies vehicles and road signs on the road, and evaluates the real-time distance status of the vehicle ahead, making the perception of the road conditions ahead more accurate and improving the intelligence and real-time performance of vehicle light control.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly to an on-vehicle lamp and an intelligent light-emitting control method therefor. Background Art

[0002] The technical field of image processing refers to the technology of using a computer to analyze, enhance, reconstruct, and understand images, specifically including steps such as image acquisition, preprocessing, feature extraction, recognition and classification, image enhancement, target detection and tracking. Through the analysis and processing of image information, this technology can achieve precise perception, recognition, and decision-making of the environment or targets, and is widely applied in fields such as autonomous driving, security monitoring, medical diagnosis, face recognition, and industrial inspection to improve the intelligence, automation, and accuracy of the system.

[0003] In the prior art, when a vehicle is traveling at high speed, if the vehicle in front suddenly brakes urgently or the road conditions deteriorate rapidly, the prior art is restricted by the lag in information analysis, resulting in a slow response of the vehicle lights flashing or untimely adjustment of the light illumination, making the driver unable to obtain accurate and effective visual warning signals at critical moments, and increasing the risk of collision or accident. Therefore, it is necessary to improve the light control. Summary of the Invention

[0004] The purpose of the present invention is to solve the drawbacks existing in the prior art, and to propose an on-vehicle lamp and an intelligent light-emitting control method therefor.

[0005] To achieve the above purpose, the present invention adopts the following technical solution. An intelligent light-emitting control method includes the following steps:

[0006] An image acquisition and preprocessing module, in which a vehicle camera captures continuous image data, formats the image to generate formatted image data; optimizes the formatted image data through contrast adjustment to generate optimized image data;

[0007] An image analysis and parsing module, based on the optimized image data, performs target detection, identifies vehicles and road signs, and obtains the target detection result; analyzes the target detection result, evaluates the distance to the vehicle in front, and generates an analysis result of the state of the vehicle in front;

[0008] A vehicle lamp regulation decision module, based on the analysis result of the state of the vehicle in front, determines whether it is necessary to flash the vehicle lights and generates a vehicle lamp regulation instruction;

[0009] An emergency response and feedback module, monitors the emergency braking signal of the vehicle in front and sudden road conditions, and adjusts the flashing frequency of the vehicle lights according to the monitored emergency situation to obtain an emergency response execution result.

[0010] Preferably, the step of obtaining the formatted image data is as follows: the vehicle camera captures continuous image data, adjusts the ISO sensitivity and aperture size of the camera according to the lighting conditions, and obtains the unprocessed original image data;

[0011] According to the unprocessed original image data, convert the RGB color mode to the YUV color mode to obtain the formatted image data;

[0012] Based on the formatted image data, apply a high-pass filter and edge enhancement to obtain the formatted image data.

[0013] Preferably, the step of obtaining the optimized image data is as follows: according to the formatted image data, calculate the contrast score, and the calculation formula is:

[0014]

[0015] where, is the contrast score, is the brightness value of the pixel at the -th row and the -th column, and are the maximum brightness value and the minimum brightness value in the image respectively, is the overall average brightness of the image, and are the number of rows and the number of columns of the image respectively, and represent the brightness values of the upper pixel and the left pixel adjacent to the current pixel point respectively;

[0016] Based on the contrast score, adjust the gray mapping curve of the image to make the brightness value distribution uniform, and obtain the optimized image data.

[0017] Preferably, the step of obtaining the target detection result is as follows: input the optimized image data into a deep learning model for target region feature extraction, perform layer-by-layer convolution operations on the image patches of the candidate regions using a convolutional neural network, and combine pooling operations to reduce redundant information, calculate the class probability distribution of each candidate region, and convert it into a vector of a fixed dimension through a fully connected layer to generate target recognition result data;

[0018] Based on the target recognition result data, screen the targets whose class confidence meets the set threshold to obtain the target detection result.

[0019] Preferably, the step of obtaining the analysis result of the leading vehicle state is as follows: based on the target detection result, extract the position coordinate information of the leading vehicle, combine the focal length of the camera and the pixel density of the image sensor, calculate the scale ratio of the leading vehicle in the image coordinate system, and convert it into a distance through the perspective projection relationship to obtain the preliminary distance data of the leading vehicle;

[0020] Based on the preliminary distance data of the vehicle ahead, calculate the actual driving distance of the vehicle ahead. The calculation formula is:

[0021]

[0022] where, is the actual driving distance of the vehicle ahead, is the focal length of the camera, is the width of the vehicle ahead, is the pixel width of the target detection frame, is the single-pixel size of the image sensor;

[0023] Based on the actual driving distance of the vehicle ahead, analyze the speed change trend of the vehicle ahead, combine the displacement of the vehicle ahead between consecutive frames, calculate the acceleration change in the time series, and evaluate the driving state of the vehicle ahead according to the dynamic relationship between speed and distance, and generate the analysis result of the vehicle ahead state.

[0024] Preferably, the step of obtaining the headlight control instruction is: based on the analysis result of the vehicle ahead state, extract the acceleration information of the vehicle ahead and the acceleration information of the own vehicle, calculate the relative acceleration change between the two, extract the instantaneous speed data of the vehicle ahead and the instantaneous speed data of the own vehicle, and obtain the preliminary determination data for headlight control;

[0025] According to the preliminary determination data for headlight control, calculate the headlight flashing trigger value. The calculation formula is:

[0026]

[0027] where, is the headlight flashing trigger value, is the current acceleration of the vehicle ahead, is the current acceleration of the own vehicle, is the instantaneous speed of the vehicle ahead, is the instantaneous speed of the own vehicle, is the relative distance between the vehicle ahead and the own vehicle, is the speed of the vehicle ahead at the previous moment, is the speed of the own vehicle at the previous moment;

[0028] Based on the headlight flashing trigger value, set the flashing trigger threshold, perform a conditional judgment. If the headlight flashing trigger value exceeds the trigger threshold, trigger the headlight to flash to obtain the headlight control instruction.

[0029] Preferably, the step of obtaining the emergency response execution result is: monitor the emergency braking signal of the vehicle ahead and sudden situations on the road, including the sudden appearance of obstacles or pedestrians crossing, to obtain the emergency situation data;

[0030] Based on the emergency situation data, calculate the headlight flashing adjustment index, and the calculation formula is:

[0031]

[0032] wherein, is the headlight flashing adjustment index, is the actual distance from the vehicle in front or the obstacle, is the safety distance threshold, is the speed difference, is the reaction time constant;

[0033] Based on the headlight flashing adjustment index, set the flashing frequency to generate the emergency response execution result.

[0034] The present invention also provides a vehicle-mounted lamp applicable to the intelligent light emission control method in any one of the above, including: a lamp housing with heat dissipation through holes opened on the outer wall, a plug for conducting electricity is arranged at one end of the lamp housing, a first heat dissipation fan and a second heat dissipation fan are correspondingly arranged inside the lamp housing, a light source substrate is horizontally arranged between the first heat dissipation fan and the second heat dissipation fan, and LED lamp chips are arranged on both surfaces of the light source substrate;

[0035] One end of the light source substrate is provided with a control circuit board corresponding to the heat dissipation through hole, a rectifying circuit board electrically connected to the plug is arranged outside the control circuit board, a connection terminal electrically connected to the first heat dissipation fan is arranged on the light source substrate, and the second heat dissipation fan is located between the rectifying circuit board and the control circuit board. The rectifying circuit board is connected to the control circuit board through pin headers, and the second heat dissipation fan is electrically connected to the rectifying circuit board.

[0036] Further, air vents corresponding to the first heat dissipation fan and the second heat dissipation fan are also opened on the outer part of the lamp housing, and an opening corresponding to the light-emitting surface of the LED lamp chip is also opened on the outer part of the lamp housing. Both ends of the air guiding strip hole are communicated with the first heat dissipation fan and the second heat dissipation fan respectively.

[0037] Further, a hot air through hole for hot air to flow through and an installation bayonet for assembling the end of the light source substrate are respectively opened on the control circuit board.

[0038] Compared with the prior art, the advantages and positive effects of the present invention are:

[0039] The present invention captures continuous image data through a vehicle camera, formats and adjusts the contrast of the images to optimize them, improves the accuracy of target detection, and reduces interference factors caused by changes in ambient light. Based on the optimized images, target detection is performed to identify vehicles and road signs on the road in real time, and the real-time distance status of the vehicle ahead is evaluated, making the perception of the road conditions ahead more accurate and improving the intelligence and real-time performance of vehicle lighting control. According to the distance status of the vehicle ahead, a headlight flashing control instruction is generated in a timely manner to effectively prompt the driver of potential risks in advance and reduce the driver's reaction delay. At the same time, the emergency braking signal of the vehicle ahead and sudden road conditions are continuously monitored, and the headlight flashing frequency is dynamically adjusted to respond to sudden conditions more quickly and accurately, improving driving safety and driving stability. The depth intelligence of the vehicle lighting control strategy is realized, effectively improving the driving vision quality and safety guarantee level at night or under complex road conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is a schematic diagram of the steps of the present invention;

[0041] Figure 2 is a schematic diagram of the overall external structure of the vehicle-mounted lamp of the present invention;

[0042] Figure 3 is a schematic diagram of the overall exploded structure of the vehicle-mounted lamp of the present invention;

[0043] Figure 4 is a schematic diagram of the overall internal structure of the vehicle-mounted lamp of the present invention.

[0044] Figure 5 In the present invention Figure 3 is an enlarged schematic diagram of part A.

[0045] Figure 6 In the present invention Figure 3 is an enlarged schematic diagram of part B.

[0046] Reference numerals:

[0047] 1, lamp housing; 3, plug; 4, rectifying circuit board; 5, control circuit board; 10, heat dissipation through hole; 11, light source substrate; 12, air outlet; 13, opening; 21, first heat dissipation fan; 22, second heat dissipation fan; 50, hot air through hole; 51, installation bayonet; 99, pin; 110, LED lamp chip; 210, connection terminal. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] In order to make the purpose, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0049] Please refer to Figure 1 , the present invention provides a technical solution, an intelligent lighting control method, including the following steps:

[0050] Image acquisition and preprocessing module: The vehicle camera captures continuous image data, formats the image to generate formatted image data; optimizes the formatted image data through contrast adjustment to generate optimized image data;

[0051] Image analysis and parsing module: Based on the optimized image data, performs object detection to identify vehicles and road signs, obtains the object detection result; analyzes the object detection result to evaluate the distance to the vehicle ahead and generates the analysis result of the vehicle-ahead state;

[0052] Headlight regulation decision-making module: Based on the analysis result of the vehicle-ahead state, determines whether to flash the headlights and generates a headlight regulation instruction;

[0053] Emergency response and feedback module: Monitors the emergency braking signal of the vehicle ahead and sudden road conditions, adjusts the headlight flashing frequency according to the monitored emergency situation, and obtains the emergency response execution result.

[0054] The steps for obtaining the formatted image data are as follows: The vehicle camera captures continuous image data, adjusts the ISO sensitivity and aperture size of the camera according to the lighting conditions to obtain the unprocessed original image data;

[0055] According to the unprocessed original image data, converts the RGB color mode to the YUV color mode to obtain the formatted image data;

[0056] Based on the formatted image data, applies a high-pass filter and edge enhancement to obtain the formatted image data.

[0057] Specifically, place a camera at the shooting scene, obtain the light intensity through an ambient light intensity sensor and compare it with the range of 0 lx to 30,000 lx. If the measured illuminance is lower than 500 lx, select a higher ISO value within the range of ISO 100 to ISO 800 and set the aperture to f / 2.8 to f / 4. If the illuminance is between 500 lx and 2,000 lx, select ISO within the range of ISO 100 to ISO 400 and set the aperture to f / 5.6 to f / 8. These values such as 500 lx and 2,000 lx are set based on the data collected in multiple shooting environments and combined with empirical statistics. It is also possible to replace 500 lx with 300 lx or 700 lx and replace 2,000 lx with 1,500 lx or 2,500 lx according to the actual situation on site. After each illuminance detection, repeat the process of adjusting ISO and aperture and perform the same operation on each frame of the continuous shooting. When the illuminance exceeds 2,000 lx, the ISO can be controlled between 100 and 200 and the aperture can be set to f / 8 to f / 16. In a darker environment, the ISO can also be appropriately increased to 800 or above and try to decrease the aperture value to below f / 2.8. During this period, the corresponding relationship between the light measurement values and the camera parameters can be recorded in chronological order, and finally the unprocessed raw image data is obtained.

[0058] When reading the unprocessed raw image data obtained, first extract the red channel, green channel, and blue channel values pixel by pixel and record them in a temporary array. Then, construct a conversion matrix according to the ITU-R BT.709 recommendation or other common standards. For example, the formula can be used: , in the above matrix, the required coefficients are obtained by referring to the public literature or by testing and fitting typical image samples, and can be slightly adjusted within the range of 0.001 to 1.000 to adapt to specific image features. Then, perform a linear operation on the RGB components of each pixel in the image and write the calculation result to the corresponding YUV component position. If there is an Alpha channel in the image, keep the Alpha data unchanged. The entire conversion process needs to traverse all the pixel points of the image. When the number of pixels is large, line-by-line processing can be considered, that is, process the next line after the data of each line is calculated. After all lines are processed, combine the converted YUV data into a new image format, and finally obtain the formatted image data.

[0059] When reading the formatted image data obtained in the previous step, first perform high-pass filtering on its luminance component (Y). A 3×3 convolution kernel can be selected, for example, the center position is , and the upper, lower, left, and right positions are , and the diagonal positions are 0 matrix. Calculate pixel by pixel: , to highlight the high-frequency components of the image. The coefficients of the convolution kernel are determined after debugging the noise and detail performance in combination with the test image. If it is found that the contrast of a local area after high-pass filtering is too strong, an attenuation coefficient can be adjusted within the range of 0.1 to 1.0, and the filtered result is multiplied by this attenuation coefficient and then added to the original Y component to form a preliminary edge enhancement effect. After that, the U and V components can be differentially calculated in the same way, or only the Y component can be selected for edge processing. If it is necessary to limit the intensity of certain edges, judgment conditions can be added. For example, enhancement processing is only performed when the filtered result is greater than a preset threshold. This threshold can be determined by estimating the contrast of the noise area in the sample image, generally taking a value between 10 and 50, and can be adjusted in steps of 5. After all pixels are processed, the new YUV data is summarized and recorded as the new image output state, and finally the formatted image data is obtained.

[0060] The steps to optimize the acquisition of image data are as follows: According to the formatted image data, calculate the contrast score, and the calculation formula is:

[0061]

[0062] where, is the contrast score, is the brightness value of the pixel at the th row and the th column, and are the maximum brightness value and the minimum brightness value in the image respectively, is the overall average brightness of the image, and are the number of rows and the number of columns of the image respectively, and represent the brightness values of the upper adjacent pixel point and the left adjacent pixel point of the current pixel point respectively;

[0063] Based on the contrast score, adjust the gray mapping curve of the image to make the brightness value distribution uniform, and obtain the optimized image data.

[0064] Specifically, the advantage of the formula is that by multiple measurements of the difference between the pixel point and the extreme values of the image brightness range, the difference between the pixel point and the overall average brightness, and the difference between adjacent pixels, the image brightness distribution, local contrast, and neighborhood change conditions can be comprehensively reflected in a single score, so as to more accurately reflect the overall contrast situation; the acquisition steps of each parameter are described in detail below:

[0065] The acquisition steps of the parameter are as follows: This parameter represents the th row and the The brightness value of the pixel at the column. To obtain this brightness value, it is necessary to first convert the RGB components of each pixel in the image into grayscale values, and the luminance component or other methods can be used. If the traditional grayscale formula is adopted, can be used to combine the values of the three channels to obtain an integer brightness value within the range of 0 to 255. When reading the entire 1920×1080 image, the rows and columns will be scanned in sequence to obtain the R, G, and B values of the corresponding pixels. After calculation using the above formula, is obtained. Taking a certain pixel as an example: if its R = 100, G = 150, and B = 200, then the corresponding is obtained. After obtaining the brightness of all pixels, it can be directly used in the subsequent calculation of the contrast score.

[0066] The steps to obtain the parameter are as follows: This parameter represents the maximum brightness value in the processed image. The acquisition method is to traverse all the pixel brightness data, store it in an array, and then find the highest brightness value in the array as . For example, if the highest brightness value in the set of all statistically obtained brightness values is 247, then . In actual operation, the entire 1920×1080 image will be scanned first, and the brightness of each pixel will be compared item by item. The currently occurring maximum brightness will be recorded and updated in real time. Finally, after scanning the entire image, the clear maximum brightness value can be obtained.

[0067] The steps to obtain the parameter are as follows: This parameter represents the minimum brightness value in the processed image. Similar to the acquisition of , it is necessary to traverse the brightness values of all pixels and record the minimum brightness through comparison. If the brightness value of 12 appears during the complete scan and this is the minimum, then . If there is a lower value in the image, such as 0, it will eventually be updated to 0. It is necessary to ensure that all pixels are accessed during the whole process. According to the actual measured results, assume that the statistical result is .

[0068] The steps to obtain the parameter are as follows: This parameter represents the overall average brightness of the image. Its acquisition process is to accumulate after obtaining the brightness of each pixel, denoted as , and then divide the accumulated result by the total number of image pixels . When the total number of image pixels is 1920×1080, the total number of pixels is approximately 2,073,600. If the sum of pixel brightness is 2.50×10^8, then can be calculated.

[0069] The acquisition steps of and Refers to the number of rows of the image, Refers to the number of columns of the image. Both are the pixel resolution information of the image itself. If the image resolution is 1920×1080, then , . This information is obtained by reading the image file header or metadata. For example, if the width of the image is read as 1920 pixels and the height is 1080 pixels, then and can be determined.

[0070] The steps for obtaining the parameters are as follows: These two parameters respectively represent the brightness of the upper adjacent pixel in the vertical direction and the brightness of the left adjacent pixel in the horizontal direction of the current pixel. The row and column data need to be read once during pixel traversal. Taking a certain pixel coordinate (i, j) as an example, the brightness values of (i-1, j) and (i, j-1) can be read and stored in the cache before accessing the current pixel, or read dynamically during the calculation process. For example, when scanning to the 3rd row and the 10th column, the brightness values of the 2nd row and the 10th column and the 3rd row and the 9th column are taken out from the memory. If the brightness of the upper pixel is 140 and the brightness of the left pixel is 138 at this time, then , .

[0071] Calculation process:

[0072] Assume the image size is 1080 rows × 1920 columns, that is , .

[0073] Statistics show that , , .

[0074] Scan the pixel brightness of the entire image and perform the following cumulative processing on each pixel brightness in the formula:

[0075]

[0076] Here, the three difference terms of each pixel are summed separately and then accumulated into the total sum.

[0077] Record the total sum of the previous step as , and then divide by the total number of pixels .

[0078] Finally, take the square root, that is .

[0079] In the example, if after scanning and statistics is approximately 3.32×10^8, then: ;

[0080] The result shows that the overall image contrast score is 12.65. The higher the value, the more significant the brightness difference. Different pictures may have a range between 6 and 20. By comparing the values of different images, it is possible to identify which image has a greater or smaller contrast, thus providing a reference for subsequent gray mapping or other equalization operations.

[0081] Based on the contrast score , combined with data such as the minimum and maximum brightness and the average brightness in the image, by reading the previously obtained brightness distribution records line by line and selecting specific intervals for targeted analysis, first identify the proportion of current pixels concentrated in the range of 80 to 180 from the previously recorded brightness distribution. If the proportion of the number of brightness pixels in the range of 80 to 180 in the image exceeds 40%, then determine that this part of the range is regarded as the medium brightness area. Subsequently, for the low brightness area, that is, the proportion in the range of 0 to 80, if it is observed to be higher than 30%, then this area is taken as the focus for gray mapping enhancement processing. For the high brightness area, that is, the proportion in the range of 180 to 255, if it is greater than 20%, then this area is compressed or redistributed through a correction curve. During the process, the pixel brightness is mapped to an adjustable curve and recorded in the real-time cache. The curve shape can be defined in segments according to the gain coefficient and threshold information. For example, when the brightness is lower than 80, the segmental gain coefficient is set to 1.2 to 1.4 to increase the brightness of the dark area. When the brightness falls between 180 and 255, the gain coefficient is set to 0.8 to 1.0 to balance the excessive brightness distribution, and an attenuation value, such as 0.05 to 0.1, is set at the connection position of the curve segments so that there will be no excessive brightness jump at the transition. The specific values of the above thresholds and gains are determined by statistical analysis of past image processing experience. For example, 20 images with obvious dark and bright parts are statistically analyzed, and the brightness histogram distribution is analyzed to determine the division boundary between the dark and bright parts. Then, through multiple tests, it is found that the attenuation value in the range of 0.05 to 0.15 is more appropriate, and the gain in the range of 1.2 to 1.5 can significantly increase the brightness of the dark area. Whether in the dark or bright area, segmented processing will be carried out instead of uniformly increasing the gain for the whole image to avoid over-enhancing or reducing local details. After the recording is completed, the pixel brightness distribution after different segmented mappings is summarized to form a new brightness table, and then the updated gray values are generated line by line, and finally the optimized image data is obtained.

[0082] The steps to obtain the object detection result are as follows: Input the optimized image data into the deep learning model for object region feature extraction. Use the convolutional neural network to perform layer-by-layer convolutional operations on the image patches of the candidate regions, and combine pooling operations to reduce redundant information. Calculate the class probability distribution of each candidate region, and convert it into a vector of a fixed dimension through the fully connected layer to generate the object recognition result data;

[0083] Based on the target recognition result data, filter out the targets whose class confidence meets the set threshold to obtain the target detection results.

[0084] Specifically, input the optimized image data into a deep learning model for target region feature extraction. First, annotate the existing training samples and record the bounding box coordinates and class information of the target regions in each sample. Subsequently, divide these annotated samples into a training set and a validation set and ensure that the training set contains no less than 10,000 images and covers a variety of scenarios and target classes. Combine the optimized image data of each image in the training set with the corresponding labels and send them into a convolutional neural network, set the learning rate to 0.001, the batch size to 32, and the number of training epochs to 50. When starting the layer-by-layer training, first read the optimized image data at the network input and apply a convolutional kernel of size to perform operations on the image patches. During the convolution process, record the weights of each channel and update the parameters in combination with mean shift. After each convolution, add a pooling operation of size and only retain the maximum value in the local region in the pooling layer to compress redundant data. In each round of iteration during training, calculate the cross-entropy loss and perform backpropagation updates on the convolutional kernel and the parameters of the fully connected layer. Stop the iteration when the cross-entropy loss converges to less than 0.02 on the validation set for 3 consecutive rounds. Subsequently, fix the network parameters and receive new optimized image data as input. Gradually extract edge, texture, and higher-level feature information for the image patches of each candidate region in the convolutional layer and the pooling layer, and convert the feature map into a fixed-dimensional vector in the fully connected layer, so as to calculate the probability distribution of each region belonging to each target class. During the process, set several output neurons according to the number of different classes and constrain the total probability within 1. When the probability value of a certain class is the highest among all outputs, it is regarded as the recognition result of the candidate region. After recording the class distribution and probability score of each region, integrate them into the output structure, and finally generate the target recognition result data.

[0085] Based on the target recognition result data, read the class confidence values of each candidate region and compare them with the best confidence threshold obtained from the statistics of approximately 2000 images before. This threshold was finally set at 0.75 with precision and recall as the evaluation metrics in multiple experiments. The specific approach is to select different confidence thresholds incrementing by 0.05 from 0.5 to 0.9 in sequence for testing and record the detection accuracy and miss rate at each threshold. After repeatedly comparing the same batch of images, select the threshold of 0.75 with the best balance between accuracy and miss rate. Subsequently, when analyzing the current target recognition result data, if the class confidence exceeds 0.75, mark this candidate region as a valid target and append the target class name and confidence value during output. If the class confidence is lower than 0.75, abandon the output of this candidate region. Regardless of the number of candidate regions, repeat the same determination and classify and integrate the finally qualified targets, record the number of targets meeting the threshold and output their positioning coordinates, and finally obtain the target detection result.

[0086] The steps to obtain the analysis result of the leading vehicle's state are as follows: Based on the target detection result, extract the position coordinate information of the leading vehicle, combine the focal length of the camera and the pixel density of the image sensor, calculate the scale ratio of the leading vehicle in the image coordinate system, and convert it to distance through the perspective projection relationship to obtain the preliminary distance data of the leading vehicle;

[0087] According to the preliminary distance data of the leading vehicle, calculate the actual driving distance of the leading vehicle. The calculation formula is:

[0088]

[0089] where, is the actual driving distance of the leading vehicle, is the focal length of the camera, is the width of the leading vehicle, is the pixel width of the target detection box, is the single-pixel size of the image sensor;

[0090] Based on the actual driving distance of the leading vehicle, analyze the speed change trend of the leading vehicle, combine the displacement of the leading vehicle between consecutive frames, calculate the acceleration change in the time series, and evaluate the driving state of the leading vehicle according to the dynamic relationship between speed and distance to generate the analysis result of the leading vehicle's state.

[0091] Specifically, based on the object detection results, read the pixel coordinate range of the vehicle ahead and select a set of reference pixel points at the center position as the basis for measurement. First, confirm the calibration file corresponding to the focal length parameter and the pixel density of the image sensor at the camera installation location and record the focal length value and the physical size corresponding to each unit pixel. Subsequently, segment the rectangular area where the vehicle ahead is located according to the recognized position coordinates of the vehicle ahead and count the pixel width and height of this area. By multiplying the pixel size by the pixel density value registered in the calibration file, the preliminary proportional information of the corresponding length can be obtained. In order to maintain consistency in strong light or night environments, multiple sets of environmental data in the calibration file will be called for comparison before each shot, and the set of contrast correction parameters closest to the current brightness range will be selected. Once the available contrast correction parameters are confirmed, they will be applied to the recognition stage of the camera output image, and the position boundary changes of the vehicle ahead in consecutive frames will be recorded. By comparing the differences in the pixel width and height of the rectangular area of the vehicle ahead in adjacent frames, its scaling trend is judged and compared with the relative position of the optical center of the lens. When both the pixel width and height show an increasing trend and exceed the reference threshold specified in the calibration file, it is determined that the distance between the vehicle ahead and the host vehicle is decreasing, and the pixel increment and the optical transformation relationship are calculated in real time. The reference threshold is determined by obtaining the optimal discrimination when testing various vehicle sizes with a fixed focal length and usually ranges between 5 and 20 pixel increments. If the increment exceeds 10 pixels, it is regarded as a significant change and the perspective projection conversion is performed immediately. The pixel increment is associated with the distance between the lens principal point and the imaging plane to deduce the relative distance change between the vehicle ahead and the host vehicle. Then, combined with parameters such as the camera focal length, the size ratio of the vehicle ahead in the image plane is calculated and matched with the previously registered projection matrix entry. This matrix is calibrated using a fixed-point target during the installation stage of the vehicle detection system. For each pixel ratio, there is a unique corresponding distance information. Finally, the results of multiple-frame comparisons are merged to integrate a relatively stable average distance, which is used as the preliminary distance data of the vehicle ahead.

[0092] The advantage of the formula is that by introducing multi-dimensional elements such as pixel width, focal length, real vehicle width, and single-pixel size of the sensor, the pixel measurement value in the image can be accurately mapped to the distance dimension in the real scene in a relatively concise form, thus providing higher accuracy for vehicle distance measurement. The acquisition steps of each parameter are described as follows:

[0093] The acquisition steps of the parameter are as follows: This parameter represents the focal length value of the camera, which is usually calibrated by the lens manufacturer before leaving the factory or can be determined through a high-precision calibration program. During specific determination, a target is placed at a fixed distance, the camera is aimed at the target, and the distance between the lens center and the imaging plane when the image is clearly captured is recorded and quantified in millimeters. For example, the focal length obtained through actual calibration millimeters.

[0094] The steps for obtaining the parameter are as follows: This parameter refers to the width of the vehicle in front.

[0095] The steps for obtaining the parameter are as follows: This parameter represents the pixel width of the detection box of the vehicle in front in the image, and it needs to be obtained by analyzing the object detection results. The specific method is to read the left and right edge position coordinates of the vehicle in front in the image plane and calculate the difference between the two, which is stored in pixels. If the left edge of the image coordinate system is pixels and the right edge is pixels, then . When conducting detection, the vehicle contour area will be marked through the vehicle recognition algorithm and accurate to pixel coordinates, and then this coordinate difference will be statistically calculated as . For example, in an image of 1920×1080, if the left edge of the detection box of the vehicle in front is recorded as 420 pixels and the right edge is 720 pixels, then .

[0096] The steps for obtaining the parameter are as follows: This parameter is the single-pixel size of the image sensor, and it needs to be looked up or calibrated in combination with the specific camera model. When calibrating, a micrometer measurement tool traceable to the international unit can be used to measure the total size of the sensor and calculate the single-pixel size in combination with the number of effective pixels. For example, if the horizontal size of the sensor is 3.456 mm and the number of effective pixels is 1920, then .

[0097] Calculation process:

[0098] First step, list all parameters:

[0099]

[0100] Second step, substitute the above values into the formula:

[0101]

[0102] Third step, first calculate the inside the denominator:

[0103]

[0104] Fourth step, calculate ;

[0105] The value can be obtained by looking up the table or through calculation as radians, approximately equivalent to 89.97 degrees.

[0106] Fifth step, tidy up the denominator and complete the calculation:

[0107]

[0108]

[0109]

[0110] The result shows that the actual driving distance corresponding to the vehicle in front in the current picture is about 3.437 meters. If the increases with time during the subsequent detection process, it indicates that the distance to the vehicle in front is shortening. If it continues to decrease, it means the distance is widening. Therefore, this formula can reflect the change of the vehicle distance in real time during multi-frame calculation.

[0111] Based on the actual driving distance of the vehicle in front, first record the displacement of the vehicle in front in the ground coordinate system in consecutive image frames, and fix the sampling time interval for each frame at 0.04 seconds. When reading the distance difference between the k-th frame and the (k + 1)-th frame of the vehicle in front, if the distance difference falls between 0.01 meters and 0.3 meters, it indicates that the vehicle in front is in a small-range moving state. If it exceeds 0.3 meters, it is considered a rapid displacement. Therefore, this section of displacement will be regarded as the key point of acceleration analysis when calculating the speed. By dividing the distance difference between adjacent frames by 0.04 seconds, the instantaneous speed of the vehicle in front is obtained, and then it is compared again between adjacent frames. If the speed increment is between 0.1 meters per second and 3 meters per second, record this value and add it to the acceleration sequence. If the speed increment is lower than 0.1 meters per second, it means the speed remains basically unchanged and is not included in the acceleration calculation. If the increment exceeds 3 meters per second, it is considered a drastic change and continue to monitor the next section of frame data. During this period, refer to a speed threshold table constructed based on the acquisition results of the previous 600 frames to determine whether there is abnormal fluctuation. This speed threshold table is divided into a low-speed range below 3 meters per second, a normal range from 5 meters per second to 15 meters per second, and a fast range above 15 meters per second according to the urban road traffic conditions, and can be further subdivided for different vehicle types. Subsequently, when analyzing the acceleration sequence, first compare each acceleration value with a reasonable range. For example, compare the acceleration within the range of 0 meters per second squared to 10 meters per second squared. If it exceeds 10 meters per second squared, mark it as an extreme acceleration value and continue to observe the subsequent speed change of this section of video frames. Finally, combine all the speed and acceleration results into a time series and evaluate the driving state of the vehicle in front during the following vehicle process by calculating the average speed and median acceleration of each frame. Then, combined with the distance change trend, distinguish the situations where congestion or unobstructed passage may occur, so as to generate the analysis result of the state of the vehicle in front.

[0112] The steps to obtain the headlight control instruction are as follows: Based on the analysis result of the state of the vehicle in front, extract the acceleration information of the vehicle in front and the acceleration information of the own vehicle, calculate the relative acceleration change between the two, extract the instantaneous speed data of the vehicle in front and the instantaneous speed data of the own vehicle, and obtain the preliminary determination data for headlight control;

[0113] According to the preliminary determination data for headlight control, calculate the headlight flashing trigger value. The calculation formula is:

[0114]

[0115] Among them, is the trigger value for headlight flashing, is the current acceleration of the vehicle ahead, is the current acceleration of the host vehicle, is the instantaneous speed of the vehicle ahead, is the instantaneous speed of the host vehicle, is the relative distance between the vehicle ahead and the host vehicle, is the speed of the vehicle ahead at the previous moment, is the speed of the host vehicle at the previous moment;

[0116] Based on the headlight flashing trigger value, set the flashing trigger threshold, perform condition judgment. If the headlight flashing trigger value exceeds the trigger threshold, then trigger the headlight flashing to obtain the headlight control instruction.

[0117] Specifically, based on the analysis result of the leading vehicle's state, first read the acceleration information of the leading vehicle and the host vehicle collected in the previous stage and check whether the sources of the two sets of acceleration values are complete. The acceleration information of the leading vehicle is calculated from the displacement difference and the corresponding time series identified in consecutive video frames, with the time interval set to 0.04 seconds. The acceleration information of the host vehicle is obtained by accumulating the longitudinal acceleration data output by the vehicle-mounted sensor within the same 0.04-second time step and splitting it into specific moment values for each segment. Subsequently, calculate the relative acceleration change between the leading vehicle and the host vehicle by comparing the acceleration values at the same timestamp, record it as a new data sequence, and mark it as a rapid change interval when the relative acceleration change is greater than the upper limit value of 5 m / s² measured in road tests and conduct a key comparison at the next moment. If the relative acceleration change is less than 0 m / s², mark it as a deceleration interval and continue to track the acceleration in the next frame of data. Then, extract the instantaneous speed data of the leading vehicle and the host vehicle from the vehicle speed dataset. The instantaneous speed data of the leading vehicle is obtained by dividing the distance increment between consecutive frames of the leading vehicle by 0.04 seconds and storing it frame by frame in the speed sequence. The instantaneous speed data of the host vehicle is taken from the vehicle-mounted speed sensor and recorded with the same 0.04-second sampling period. Synchronously read in and record the corresponding speed value at the previous moment according to the numerical positions of the instantaneous speed sequences of the leading vehicle and the host vehicle. To ensure accurate speed matching, compare the timestamp differences and keep the error within 0.01 seconds. When the speed difference between the leading vehicle and the host vehicle is found to be greater than 5 m / s, register it as a high difference interval and continue to confirm whether it remains in the high difference state in the next cycle. If the speed difference persists for 3 or more adjacent moments, output a reminder message indicating a significant difference in the current vehicle speeds. After all these values are correlated, integrate and generate the preliminary determination data for headlight control in chronological order.

[0118] The advantage of the formula is that by introducing information such as the acceleration difference, speed difference, and distance between the leading vehicle and the host vehicle, the dynamic relationship between the two vehicles can be measured in a compact expression, providing a more comprehensive reference for headlight flashing determination. The following is a detailed description of the steps for obtaining each parameter:

[0119] The steps for obtaining the parameter are as follows: This parameter represents the current acceleration of the leading vehicle, with the unit of m / s². After obtaining the speed by dividing the displacement change of the leading vehicle between consecutive frames by the frame interval, perform a first-order difference on the speed and divide it by the time step to obtain the acceleration. Specifically, during urban road tests, it is statistically found that the average acceleration of the leading vehicle is approximately in the range of 0 m / s² to 5 m / s². After real-time calculation, an accurate instantaneous acceleration value is formed at each sampling moment. For example, at the current moment, it may be recorded as 2.2 m / s². Then 。

[0120] The steps for obtaining the parameter are as follows: This parameter represents the current acceleration of the vehicle, and the unit is also meters per second squared. It is mainly obtained through the output data of the vehicle body acceleration sensor and the in-vehicle ECU. During specific operations, read the value of the longitudinal acceleration sensor at this moment. In the scenarios of urban expressways or highways, the reference vehicle test results show that this value is mostly distributed between 0 meters per second squared and 4 meters per second squared. When it exceeds this range, a recheck will be performed to confirm whether there is a measurement deviation in the sensor. For example, if the measured acceleration value of the vehicle at this time is 1.5 meters per second squared, then in the operation of this formula 。

[0121] The steps for obtaining the parameter are as follows: This parameter is the instantaneous speed of the leading vehicle, and the unit is selected as meters per second. Its acquisition is similar, and it will also be through the operation of the change in the inter-frame distance of the image and the time interval. On urban roads, the speed of the leading vehicle is approximately in the range of 0 meters per second to 20 meters per second. For example, when the instantaneous speed of the leading vehicle calculated through the image at a certain moment is 15.0 meters per second, then 。

[0122] The steps for obtaining the parameter are as follows: This parameter is the instantaneous speed of the vehicle, and the unit is meters per second. It is obtained from the vehicle's own instrument, ECU or vehicle speed sensor and recorded in the software. According to the benchmark test, the vehicle speed on urban ordinary roads is mainly distributed in the range of 0 meters per second to 25 meters per second. Assuming that the currently read value is 17.5 meters per second, then at this time 。

[0123] The steps for obtaining the parameter are as follows: This parameter is the relative distance between the leading vehicle and the vehicle, and the unit can be meters. It is the net value obtained by subtracting the reference distance from the front of the vehicle to the camera installation point from the previously obtained measurement result of the leading vehicle distance. If the leading vehicle distance data has been obtained by means such as camera ranging or millimeter wave radar in the previous stage, the corresponding value can be directly taken at this moment as 。For the urban driving scenario, L can be judged within the range of 2 meters to 100 meters. For example, if the calibrated distance result shows that the center-to-center distance between the leading vehicle and the vehicle is 30.0 meters, and the distance from the front of the vehicle to the camera is 1.0 meter, then 。

[0124] The steps for obtaining the parameter are as follows: This parameter represents the speed of the leading vehicle at the previous moment, and the unit is also meters per second. The value at the previous time index of the historical stored leading vehicle speed sequence will be taken out as During offline or online processing, the speed value of the previous frame or the previous sampling period can be quickly queried. For example, if the speed of the vehicle in front at the current moment is 15.0 meters per second and the recorded value at the previous moment is 14.0 meters per second, then at this time 。

[0125] The steps for obtaining the parameter are as follows: This parameter is the speed of the vehicle at the previous moment, with the unit of meters per second, and is also obtained through the position of the previous index in the vehicle speed sequence in terms of time. In the urban vehicle usage scenario, the speed of the vehicle often fluctuates within a short period of time, so relatively high-frequency records are maintained in the time series. If the speed at the previous moment is queried to be 17.0 meters per second at this time, then this value is written into 。

[0126] Calculation process:

[0127] First step, select the parameter at a certain moment:

[0128]

[0129] Second step, calculate the numerator :

[0130]

[0131] Third step, calculate the denominator :

[0132]

[0133]

[0134]

[0135] Fourth step, calculate the speed ratio part in the brackets :

[0136]

[0137]

[0138]

[0139]

[0140] Fifth step, combine the above results:

[0141]

[0142]

[0143]

[0144] The results show that when If it is necessary to determine whether to flash the lights in the subsequent steps, this value will be compared with the light flashing trigger threshold. If it exceeds the threshold, the light flashing will be started, and if it is lower than the threshold, the current light state will be maintained.

[0145] Based on the headlight flashing trigger value, first select the current value from the trigger value sequence obtained above and compare it with the headlight flashing trigger threshold. The threshold is set by referring to the vehicle following data for more than 300 hours in the past and selecting boundary points in the curve diagram of driving speed and relative acceleration to determine the critical range. For example, when the distance between the vehicle and the vehicle in front is between 5 meters and 30 meters, continue to observe the speed and acceleration difference and record the distribution of the trigger value. Finally, a range between 0.2 and 0.3 is selected as the feasible threshold range. After testing one by one under multiple working conditions, 0.25 is set as the headlight flashing trigger threshold. If the headlight flashing trigger value calculated at the current moment exceeds 0.25, the headlight flashing trigger value calculated at the current moment will be 0.25. It is determined that the flashing conditions are met and the light output switching frequency is set in the headlight control program. For example, the light flashes within the range of 0.8 to 1.2 seconds. If it does not exceed 0.25, the current headlight state is maintained, and the trigger value at the current moment is recorded together with the time index into the trigger value sequence. The sequence will be extended and monitored later. If the trigger values at five consecutive sampling moments exceed the threshold, the flashing time is extended or the flashing frequency is increased. If the threshold is only exceeded for a short time, the flashing is terminated after the vehicle speed or acceleration drops. All flashing actions must be compared with the brake signal status in the vehicle's electronic system to complete the final output action, and finally settled as a headlight control instruction in the headlight control program.

[0146] The steps for obtaining the emergency response execution results are: monitoring the emergency brake signal of the preceding vehicle and the sudden conditions on the road, including the sudden appearance of obstacles or the crossing of pedestrians, to obtain emergency data;

[0147] According to the emergency data, the headlight flashing adjustment index is calculated using the following formula:

[0148]

[0149] in, Adjust the flashing index for the headlights. is the actual distance to the vehicle ahead or obstacle, is the safety distance threshold, is the speed difference, is the reaction time constant;

[0150] Based on the vehicle light flashing adjustment index, the flashing frequency is set and the emergency response execution result is generated.

[0151] Specifically, it monitors the emergency braking signal of the vehicle ahead and unexpected situations on the road. It synchronously reads the measured information through the front camera and the in-vehicle braking monitoring device, and records the vehicle speed, deceleration, and road surface image data at that moment. The continuous monitoring time interval is set to 0.05 seconds, and within this interval, it repeatedly compares the braking light status of the vehicle ahead with the accelerometer value. If the braking light is highly bright and the corresponding acceleration drops below -4 m / s², it is registered as an emergency deceleration. If a newly appeared obstacle or pedestrian position is detected in the image and the displacement of its coordinates from the original frame to the current frame is large, this situation is marked as a sudden change. When braking or obstacle detection markings occur in three consecutive samples, it is confirmed that an emergency event has occurred, and multiple data of this emergency event are written into the emergency situation data sequence. This sequence includes the braking mark of the vehicle ahead, the pixel coordinate range of the obstacle, and the detected pedestrian speed value, etc. If multiple pedestrians are detected crossing the road during the road monitoring period, the travel distance of each of them from 10 meters to 30 meters is counted one by one, and the number of frames in which they appear in the camera image is multiplied by 0.05 seconds per frame as the continuous walking time. After identifying an emergency situation exceeding their respective thresholds on the lane, it will continue to track the subsequent displacement changes and vehicle acceleration and update them into the data sequence. If a sudden situation appears only in one sampling period, it will be verified in the subsequent period whether it has been resolved. All markings will be included in this emergency situation data for continued use and recorded period by period, so that in the subsequent processing stage, it can accurately locate at which moment an emergency braking or an obstacle appears based on this emergency situation data, and finally obtain the emergency situation data.

[0152] The advantage of the formula is that it quantitatively characterizes the actual distance and speed difference between the vehicle and the obstacle or the vehicle ahead by combining the exponential decay and exponential growth characteristics, so as to comprehensively reflect multiple risk factors in the same expression. The following will explain the acquisition steps of each parameter one by one:

[0153] The acquisition steps of the parameter are as follows: This parameter represents the actual distance from the vehicle ahead or the obstacle, which can be obtained through a vehicle ranging radar or an image perspective ranging method, and the unit is meters. The measured distance is updated once every 0.05 seconds or 0.1 seconds for each sampling interval. To ensure accuracy, it can be calibrated by combining the camera focal length and the known obstacle size comparison list. When the actual distance between vehicles detected on the spot falls within the range of 2 meters to 100 meters, it is regarded as the normal following distance or the general obstacle distance range. If it is less than 2 meters, it is marked as an extremely close situation and noted in the ranging record. The following is an example: After measurement, the distance record from the obstacle at this moment is meters, and then it is brought into the subsequent calculation.

[0154] The steps to obtain the parameter are as follows: This parameter is the safety distance threshold, with the unit of meters, and needs to be determined with reference to the road driving regulations and vehicle braking test data before the vehicle hits the road. Usually, it can be set between 10 meters and 30 meters in an urban environment. If the driving speed is relatively high or the road environment is special, a higher value will be taken, such as 25 meters or 30 meters. During specific operations, the emergency braking tests of multiple vehicles can be simulated at different speeds on a dry road surface, and the appropriate can be selected by measuring the braking distance and combining the requirements of laws and regulations for the safe vehicle distance. In this example, take meters.

[0155] The steps to obtain the parameter are as follows: This parameter represents the speed difference, which is generally determined by the speed difference between the leading vehicle and the own vehicle or the relative motion speed difference between the vehicle and the obstacle, with the unit of meters per second. The speed readings at that moment can be taken and compared to obtain it. If the speed of the leading vehicle is 50 kilometers per hour and the own vehicle is 40 kilometers per hour, the difference is 10 kilometers per hour, which is converted to meters per second and recorded as .

[0156] The steps to obtain the parameter are as follows: This parameter is the reaction time constant, with the unit of seconds, and is used to characterize the response delay of the vehicle to an emergency. The average time required for the driver to decelerate effectively from the moment of discovering the obstacle can be statistically obtained in multiple road tests, or the delay amount from the vehicle attitude change read by the on-vehicle control system to the execution of the braking signal can be measured. If about 500 person-times of evaluations in the urban driving environment are referred to, and the most common human-machine reaction time falls within the range of 0.8 seconds to 1.2 seconds, then seconds can be selected as the practical value.

[0157] Calculation process:

[0158] First step, list each parameter:

[0159]

[0160] Second step, calculate :

[0161]

[0162]

[0163] Third step, calculate the denominator :

[0164]

[0165]

[0166]

[0167] Step 4, substitute into the formula:

[0168]

[0169] The result shows that when the current distance is 10 meters and the speed difference is about 2.78 meters per second, the headlight flashing adjustment index is 0.676. A higher value means that the headlights can perform flashing actions at a medium frequency in the current situation. If the value continues to rise to 0.9 or above, it can be determined that the proximity to the vehicle in front or the obstacle and the relative speed difference have further intensified, and a higher frequency of light signals is required for prompting.

[0170] Based on the headlight flashing adjustment index, first compare the calculated index value with the flashing frequency distribution range statistically obtained from three groups of research data. The research data is based on 300 hours of urban road and 200 hours of highway driving tests, and multiple bins are selected according to the speed difference and distance. The flashing frequencies are listed in the range of 1 flash per second to 4 flashes per second at a step of 0.5 flashes per second. When the headlight flashing adjustment index reaches between 0.6 and 0.7, it corresponds to the range of 2 flashes per second to 2.5 flashes per second. If the index rises to 0.8 to 0.9, it is in the range of 3 to 3.5 flashes per second. If the index exceeds 0.9, it is in the range of 4 flashes per second. Subsequently, record the change of the instantaneous index during the vehicle operation and continuously compare. If the flashing adjustment index remains near 0.8 to 0.9 at multiple consecutive sampling points, fix the flashing frequency at 3 flashes per second and determine whether it continues to increase at subsequent moments. If the index is less than 0.2, it is determined as a low-risk situation and the flashing is terminated or only the taillight is turned on in a constant-on manner. During the whole process, the frequency settings at each sampling moment will be covered in real time to ensure that the later results can cover the previous settings. When the index suddenly drops or re-enters another interval, refresh the flashing frequency to the lower limit of the corresponding value interval and record the actual frequency value selected at each moment. After a period of time, the system integrates these frequency control actions to obtain the final lighting control behavior and outputs it as the emergency response execution result.

[0171] Please refer to Figures 2 - 6 , the present invention also provides a vehicle-mounted lamp, which is applicable to the intelligent light emission control method in the above embodiments, including: a lamp housing 1 with heat dissipation through holes 10 opened on the outer wall, a plug 3 for conducting electricity is arranged at one end of the lamp housing 1, a first heat dissipation fan 21 and a second heat dissipation fan 22 are correspondingly arranged inside the lamp housing 1, a light source substrate 11 is horizontally arranged between the first heat dissipation fan 21 and the second heat dissipation fan 22, and LED lamp chips 110 are arranged on both sides of the light source substrate 11;

[0172] One end of the light source substrate 11 is provided with a control circuit board 5 corresponding to the heat dissipation through hole 10. An external rectifying circuit board 4 electrically connected to the plug 3 is provided outside the control circuit board 5. A connection terminal 210 electrically connected to the first heat dissipation fan 21 is provided on the light source substrate 11. The second heat dissipation fan 22 is located between the rectifying circuit board 4 and the control circuit board 5. The rectifying circuit board 4 is connected to the control circuit board 5 through a pin 99. The second heat dissipation fan 22 is electrically connected to the rectifying circuit board 4.

[0173] Regarding the above description, it should be noted that the lamp housing 1 is also provided with grooves for the first heat dissipation fan 21 and the second heat dissipation fan 22 to be placed. During installation, the first heat dissipation fan 21 can be electrically connected to the light source substrate 11, and the second heat dissipation fan 22 can be electrically connected to the rectifying circuit board 4, thereby realizing the power supply to the first heat dissipation fan 21 and the second heat dissipation fan 22.

[0174] Regarding the above description, it should be noted that a hot air through hole 50 and an installation bayonet 51 are respectively formed on the control circuit board 5. The hot air through hole 50 is used for the hot air to flow through, and the installation bayonet 51 is adapted to the end of the light source substrate 11.

[0175] Regarding the above description, it should be particularly noted that during the installation and use process, air vents 12 corresponding to the first heat dissipation fan 21 and the second heat dissipation fan 22 are also formed outside the lamp housing 1, and an opening 13 corresponding to the LED lamp chip 110 is also formed outside the lamp housing 1. Both ends of the opening 13 are respectively communicated with the first heat dissipation fan 21 and the second heat dissipation fan 22. During the heat dissipation process, the rotation directions of the first heat dissipation fan 21 and the second heat dissipation fan 22 are made consistent, so that the hot air flows in the same direction, and the hot air flows through the opening 13 to achieve a diversion effect.

[0176] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any person skilled in the art may use the disclosed technical content to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention still belong to the protection scope of the technical solution of the present invention.

Claims

1. An intelligent light-emitting control method, characterized in that It includes the following steps: An image acquisition and preprocessing module, where a vehicle camera captures continuous image data, formats the image, and generates formatted image data; optimizes the formatted image data through contrast adjustment to generate optimized image data; An image analysis and parsing module, which performs object detection based on the optimized image data, identifies vehicles and road signs, and obtains object detection results; Analyzes the object detection results, evaluates the distance to the vehicle ahead, and generates an analysis result of the state of the vehicle ahead; A headlight control decision module, which determines whether to flash the headlights based on the analysis result of the state of the vehicle ahead and generates a headlight control instruction; An emergency response and feedback module, which monitors the emergency braking signal of the vehicle ahead and sudden road conditions, adjusts the headlight flashing frequency according to the monitored emergency situation, and obtains an emergency response execution result; The step of obtaining the optimized image data is: calculating a contrast score based on the formatted image data, and the calculation formula is: Among them, is the contrast score, is the th row and and are the maximum brightness value and the minimum brightness value in the image respectively, is the overall average brightness of the image, and are the number of rows and the number of columns of the image respectively, and represent the brightness values of the upper pixel and the left pixel adjacent to the current pixel point respectively; Based on the contrast score, adjust the gray mapping curve of the image to make the brightness value distribution uniform, and obtain optimized image data.

2. The intelligent light-emitting control method according to claim 1, wherein The step of obtaining the formatted image data is: a vehicle camera captures continuous image data, adjusts the ISO sensitivity and aperture size of the camera according to the lighting conditions, and obtains unprocessed original image data; According to the unprocessed original image data, convert the RGB color mode to the YUV color mode to obtain formatted image data; Based on the formatted image data, apply a high-pass filter and edge enhancement to obtain formatted image data.

3. The intelligent light-emitting control method according to claim 1, characterized in that, The step of obtaining the object detection result is: input the optimized image data into a deep learning model for object region feature extraction, perform layer-by-layer convolution operations on the image patches in the candidate regions using a convolutional neural network, and combine pooling operations to reduce redundant information, calculate the class probability distribution of each candidate region, and convert it into a vector of a fixed dimension through a fully connected layer to generate object recognition result data; Based on the object recognition result data, filter out the objects whose class confidence meets the set threshold to obtain the object detection result.

4. The intelligent light-emitting control method according to claim 1, wherein The step of obtaining the analysis result of the state of the vehicle ahead is: based on the object detection result, extract the position coordinate information of the vehicle ahead, combine the focal length of the camera and the pixel density of the image sensor, calculate the scale ratio of the vehicle ahead in the image coordinate system, and convert it into a distance through perspective projection relationship to obtain the preliminary distance data of the vehicle ahead; According to the preliminary distance data of the vehicle ahead, calculate the actual driving distance of the vehicle ahead, and the calculation formula is: Among them, is the actual driving distance of the vehicle ahead, is the focal length of the camera, is the width of the vehicle ahead, is the pixel width of the target detection frame, is the single pixel size of the image sensor; Based on the actual driving distance of the vehicle ahead, analyze the speed change trend of the vehicle ahead, combine the displacement of the vehicle ahead between consecutive frames, calculate the acceleration change in the time series, and evaluate the driving state of the vehicle ahead according to the dynamic relationship between speed and distance to generate an analysis result of the state of the vehicle ahead.

5. The intelligent light-emitting control method according to claim 1, wherein, The step of obtaining the headlight control instruction is: based on the analysis result of the state of the vehicle ahead, extract the acceleration information of the vehicle ahead and the acceleration information of the own vehicle, calculate the relative acceleration change between the two, extract the instantaneous speed data of the vehicle ahead and the instantaneous speed data of the own vehicle to obtain preliminary headlight control determination data; Calculate the headlight flashing trigger value based on the preliminary judgment data of headlight regulation. The calculation formula is: Among them, is the trigger value for the headlight to flash, is the current acceleration of the vehicle in front, is the current acceleration of this vehicle, is the instantaneous speed of the vehicle in front, is the instantaneous speed of this vehicle, is the relative distance between the vehicle in front and this vehicle, is the speed of the vehicle in front at the previous moment, is the speed of this vehicle at the previous moment; Based on the headlight flashing trigger value, set the flashing trigger threshold and perform condition judgment. If the headlight flashing trigger value exceeds the trigger threshold, trigger the headlight to flash to obtain the headlight regulation instruction.

6. The intelligent light-emitting control method according to claim 1, wherein The steps for obtaining the execution result of the emergency response are as follows: Monitor the emergency braking signal of the vehicle ahead and sudden situations on the road, including the sudden appearance of obstacles or pedestrians crossing, to obtain emergency situation data. Calculate the headlight flashing adjustment index according to the emergency situation data. The calculation formula is: Among them, is the headlight flashing adjustment index, is the actual distance from the vehicle in front or the obstacle, is the safety distance threshold, is the speed difference, is the reaction time constant; Based on the headlight flashing adjustment index, set the flashing frequency to generate the execution result of the emergency response.

7. A vehicle-mounted lamp, applicable to the intelligent light-emitting control method described in any one of claims 1-6, characterized in that, Including: A lamp housing (1) with heat dissipation through holes (10) opened on the outer wall. One end of the lamp housing (1) is provided with a plug (3) for conducting electricity. A first heat dissipation fan (21) and a second heat dissipation fan (22) are correspondingly arranged inside the lamp housing (1). A light source substrate (11) is horizontally arranged between the first heat dissipation fan (21) and the second heat dissipation fan (22). LED lamp chips (110) are arranged on both sides of the light source substrate (11). One end of the light source substrate (11) is provided with a control circuit board (5) corresponding to the heat dissipation through holes (10). An external rectifying circuit board (4) electrically connected to the plug (3) is arranged outside the control circuit board (5). A connection terminal (210) electrically connected to the first heat dissipation fan (21) is arranged on the light source substrate (11). The second heat dissipation fan (22) is located between the rectifying circuit board (4) and the control circuit board (5). The rectifying circuit board (4) is connected to the control circuit board (5) through pin headers (99). The terminal of the second heat dissipation fan (22) is connected to the rectifying circuit board (4).

8. The vehicle-mounted lamp according to claim 7, wherein Air vents (12) corresponding to the first heat dissipation fan (21) and the second heat dissipation fan (22) are also opened on the outer part of the lamp housing (1). An opening (13) corresponding to the light-emitting surface of the LED lamp chips (110) is also opened on the outer part of the lamp housing (1). Both ends of the opening (13) are communicated with the first heat dissipation fan (21) and the second heat dissipation fan (22) respectively.

9. The vehicle-mounted lamp according to claim 7, wherein Hot air through holes (50) for hot air to flow through and mounting bays (51) for the end part of the light source substrate (11) to be assembled are respectively opened on the control circuit board (5).

Images