Vehicle driving environment sensing method and anti-dazzling control method of DLP intelligent vehicle lamp

Abstract

The invention belongs to the technical field of vehicle lighting, and particularly relates to a vehicle driving environment sensing method and an anti-dazzling control method of a DLP intelligent vehicle lamp. Transmitting the image as a to-be-detected image to a target detection model deployed on the vehicle-mounted computer in real time; predicting the to-be-detected image through the target detection model to detect a set target and a pixel coordinate position thereof in the to-be-detected image; and calculating the distance from a set target in the to-be-detected image to the binocular camera by adopting a distance measurement algorithm. According to the invention, the road condition information in front of the vehicle is collected and analyzed in real time, the pedestrian and the position thereof are accurately identified in combination with the target detection model, the target distance is calculated by using the stereoscopic vision technology, and the spatial relationship between the target object and the vehicle is rapidly determined to provide an accurate basis for the illumination mode adjustment of the vehicle lamp. Therefore, the dazzling problem to pedestrians and opposite vehicles is effectively avoided.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle lighting, and in particular relates to a vehicle driving environment perception method and a DLP intelligent vehicle lamp anti-glare control method. Background Art

[0002] With the rapid development of intelligent automobile technology, the functions of vehicle systems are gradually changing from traditional passive assistance to active intelligence. Especially in night driving environments, intelligent control of car lights has become one of the key research areas to improve driving safety and driving comfort. Traditional automobile lighting systems mainly rely on fixed light intensity distribution design, which makes it difficult to dynamically adjust the lighting mode according to actual road conditions and target objects such as pedestrians or oncoming vehicles. This fixed light intensity design often leads to the following problems in actual driving:

[0003] When driving at night, when vehicle lights directly shine on pedestrians, the strong light beam can cause temporary blindness or dazzle to pedestrians, especially in city streets or crowded areas, which may cause visual discomfort to pedestrians and even induce traffic accidents; high-intensity high beams can easily cause visual fatigue or temporary disability of the driver when oncoming vehicles come into view, seriously affecting nighttime driving safety; the current lighting modes of vehicle lighting systems mostly rely only on simple switching, such as switching between low beam and high beam, and lack the ability to perceive real-time road conditions, especially in complex driving scenarios, such as multiple pedestrians, curves, ramps, etc., the lights cannot be accurately controlled.

[0004] In the existing technology, some solutions based on monocular cameras or infrared sensors try to achieve local shading or light adjustment by identifying the target area, but these methods have problems such as limited perception range, insufficient accuracy and poor environmental adaptability. For example, the performance of monocular cameras will drop significantly in low-light environments, and infrared sensors have limited recognition capabilities for target shapes and cannot provide sufficient optical shading accuracy. At the same time, these solutions are usually difficult to effectively distinguish different road objects, such as pedestrians and obstacles, resulting in inflexible lighting control strategies, which further limits the actual application effect of the system. Summary of the invention

[0005] The present invention provides a vehicle driving environment perception method and a DLP intelligent vehicle lamp anti-glare control method, which can effectively solve the problems in the background technology.

[0006] To achieve the above object, the technical solution of the present invention is as follows:

[0007] A vehicle driving environment perception method is implemented based on a binocular camera, wherein the binocular camera is installed at a set position on the vehicle body, and the method comprises:

[0008] The road condition information within the shooting range is collected by the binocular camera, and the collected results are transmitted in real time as the image to be detected to the target detection model deployed on the vehicle-mounted computer;

[0009] Predicting the image to be detected by using the target detection model to detect the set target and its pixel coordinate position in the image to be detected;

[0010] A distance measurement algorithm is used to calculate the distance from the set target in the image to be detected to the binocular camera.

[0011] Furthermore, the steps of implementing the ranging algorithm include:

[0012] For each pixel of the image to be detected, a cost value of pixel matching is obtained by fusing the BT cost and the cost processed by the SobelX operator;

[0013] Locally smoothing the cost value of each pixel to obtain a cost matrix;

[0014] Performing multi-path constraint cost aggregation on the cost matrix to obtain an aggregated cost matrix;

[0015] Determining an optimal disparity value for each pixel based on the aggregated cost matrix, and obtaining a disparity map through post-processing optimization;

[0016] Distance information is calculated based on the disparity map.

[0017] Furthermore, the BT cost calculation step includes:

[0018] Calculate the pixel (x R -n,y) and (x R +n,y) between sub-pixel positions (x R +x,y) grayscale value

[0019] Calculate the right center of the input right eye image (x R +dn,y) and (x R +d+n,y) between the sub-pixel positions (x R +d+x,y) gray value

[0020] The BT cost is calculated by the following formula:

[0021]

[0022] cos=min(cos 1 , cos 2 );

[0023] Wherein, cos is the BT cost, cos 1 and cos 2 is the bidirectional cost value, d is the parallax, 0.1≤n≤0.9, I R (x R , y) is the sub-pixel position in the left image (x R ,y), IT(x R , d, y) is the sub-pixel position (x, d, y) in the right image offset by the parallax d. R +d,y) grayscale value.

[0024] The anti-glare control method of DLP intelligent vehicle lamp includes:

[0025] The vehicle driving environment perception method as described above is used to perceive the environment in front of the vehicle to obtain a perception result, wherein the perception result includes a set target in the image to be detected and its pixel coordinate position, and a distance from the set target to the binocular camera;

[0026] In combination with the optical characteristics of the DLP smart headlight, a corresponding relationship between the set target and the headlight area is established to determine the area where brightness adjustment is required;

[0027] The DMD control chip is used to independently control several micromirrors to adjust the brightness of each area.

[0028] Furthermore, in the process of determining the area that needs to be adjusted in brightness, the determination is made in combination with the intensity of ambient light.

[0029] Furthermore, the DMD control chip is used to independently control a number of micromirrors to adjust the brightness of each area, including:

[0030] Calculate the brightness requirement of each light zone according to the perception result and the ambient light intensity;

[0031] Convert the brightness requirement of each light zone into a corresponding PWM signal to generate a duty cycle;

[0032] The micromirror switch is driven by PWM signal to adjust the brightness of the light zone.

[0033] Furthermore, the ambient light intensity is detected by a light sensor and / or is identified by an image to be detected captured by a binocular camera.

[0034] Furthermore, the ambient light intensity is identified by using the image to be detected collected by the binocular camera, including:

[0035] The left and right images are collected by the binocular camera and aligned in time and space;

[0036] Acquire a set target and its pixel coordinate position detected in the image to be detected by using the target detection model, and remove the set target from the left and right images;

[0037] Convert the remaining parts of the left and right images after elimination into grayscale images;

[0038] Calculate the average grayscale value as the current ambient light intensity.

[0039] Furthermore, the set target is removed from the left and right images, specifically:

[0040] Replacing the pixel value of the area corresponding to the set target with the set value;

[0041] Or, the area corresponding to the set target is marked as invalid and does not participate in the brightness calculation.

[0042] Furthermore, the ambient light intensity is detected by the light sensor and also recognized by the image to be detected collected by the binocular camera, including:

[0043] Based on the detection result of the optical sensor, when the difference between the previous and the next detection results is greater than a set value, the recognition of the image to be detected is started;

[0044] The basis is revised according to the recognition result of the image to be detected to obtain the final ambient light intensity.

[0045] The technical solution of the present invention can achieve the following technical effects:

[0046] The present invention proposes an environment perception method based on a binocular camera, and based on this method, proposes an anti-glare control method for DLP smart car lights; by real-time collection and analysis of road condition information in the direction set by the vehicle, combined with a target detection model, pedestrians and their positions are accurately identified, and the target distance is calculated using stereo vision technology, data support can be provided for the dynamic adjustment of smart car lights; through a distance measurement algorithm based on a disparity map, the present invention can quickly determine the spatial relationship between the target object and the vehicle, and provide an accurate basis for adjusting the lighting mode of the car lights, thereby effectively avoiding the glare problem for pedestrians and oncoming vehicles. Through the control method of the present invention, the DLP smart car lights can perform pixel-level precise control of the lighting area, avoid the limitations of traditional mechanical shading, have a fast reaction speed, can respond to changes in road conditions in real time, and provide the best lighting effect while ensuring safe driving. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0048] Figure 1 A flow chart of a method for perceiving a vehicle's driving environment;

[0049] Figure 2 Flowchart for the range finding algorithm implementation;

[0050] Figure 3 The flowchart of the anti-glare control method of DLP intelligent vehicle lamp;

[0051] Figure 4 A flowchart for adjusting the brightness of each area;

[0052] Figure 5 The flowchart is for identifying the ambient light intensity using the image to be detected collected by the binocular camera. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0054] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0055] Embodiment 1:

[0056] like Figure 1 As shown, the vehicle driving environment perception method is implemented based on a binocular camera, and the binocular camera is installed at a set position on the vehicle body. The method includes:

[0057] S1: The road condition information within the shooting range is collected by the binocular camera, and the collected results are transmitted in real time as the image to be detected to the target detection model deployed on the vehicle computer;

[0058] S2: predicting the image to be detected by the target detection model to detect the set target and its pixel coordinate position in the image to be detected;

[0059] S3: Use the ranging algorithm to calculate the distance from the set target in the image to be detected to the binocular camera.

[0060] During the implementation process, the binocular camera can be installed at the front position of the car body, or at the rear position or other position of the car body. Regardless of the installation position, environmental perception of the corresponding direction can be achieved, and therefore it is within the protection scope of the present invention.

[0061] In this embodiment, a binocular camera-based environment perception method is proposed, in which the binocular camera is installed at the front position of the vehicle as a specific implementation method. In this method, by real-time collection and analysis of road condition information in front of the vehicle, combined with the target detection model, pedestrians and vehicles in corresponding directions and their positions can be accurately identified, and the target distance is calculated using stereo vision technology, which can provide data support for the dynamic adjustment of intelligent headlights; through the distance measurement algorithm based on the disparity map, this embodiment can quickly determine the spatial relationship between the target object and the vehicle, and provide an accurate basis for adjusting the lighting mode of the headlights, thereby effectively avoiding the glare problem for pedestrians and oncoming vehicles.

[0062] During the implementation of step S1, the binocular camera simultaneously collects two synchronous images, i.e., the left image and the right image, which need to be calibrated by time stamps to ensure that the two images are completely aligned in time; image acquisition may be affected by environmental factors, such as light intensity, weather conditions, etc., so pre-processing can be performed during the acquisition process, such as: light balance, using CLAHE to improve image visibility; denoising, reducing image noise through Gaussian filtering or median filtering, etc. The collected images can be transmitted to the on-board computer in standard data formats such as JPEG, PNG or RAW, specifically through a high-speed bus or a dedicated communication protocol to ensure real-time performance.

[0063] The target detection model used in this embodiment can be based on a deep learning algorithm, such as YOLO, SSD or Faster R-CNN. Choosing a suitable model requires balancing real-time performance and accuracy. As a specific implementation method, the yolov9m lightweight target detection model can be specifically used. The yolov9m lightweight target detection model is trained based on a nighttime traffic target data set in this embodiment, covering targets such as motor vehicles, non-motor vehicles, and pedestrians. The trained model predicts the real-time driving picture obtained by the on-board binocular camera. Similarly, taking the binocular camera installed in the front position as an example, the motor vehicle, non-motor vehicle and pedestrian targets in front and their pixel coordinate positions can be detected.

[0064] As a specific implementation method for step S3, Figure 2 As shown, the implementation steps of the ranging algorithm include:

[0065] A1: For each pixel of the image to be detected, the cost value of pixel matching is obtained by fusing the BT cost and the cost after processing by the SobelX operator. By directly calculating the BT cost value for the input image, more original image information is retained, while the cost calculation after processing by the SobelX operator retains more edge and detail information. The cost calculated after processing by the SobelX operator is fused with the cost value directly calculated for the original image to improve the accuracy of the cost value.

[0066] As a specific example of the cost fusion in this step, the cost weight adjustment method can be adopted in the implementation process. For example, the BT cost is the first cost, and its weight is W 1 , the cost after SobelX operator processing is the second cost, and its weight is W 2 In the fusion process of the first cost and the second cost, the weight can be adjusted by experiment or adaptively. The fusion formula is:

[0067] C(p, d) = W 1 ·C BT (p,d)+W 2 ·C Sobel (p, d);

[0068] Among them, C(p,d) is the cost value after fusion, which is used to indicate the matching degree of pixel p on disparity d; C BT (p, d) represents the cost value calculated using the BT method; C Sobel (p, d) represents the cost value after processing with the Sobel operator. Of course, this fusion method is only a specific example, and other fusion methods, such as maximum fusion, minimum fusion, product fusion, geometric mean fusion, etc., are also within the protection scope of the present invention if they can meet the technical purpose of the present invention.

[0069] Before implementing this step, the image can be first subjected to illumination equalization and denoising to improve the robustness of the cost calculation;

[0070] A2: Locally smooth the cost value of each pixel to obtain the cost matrix;

[0071] As a specific smoothing method, the obtained cost value can be processed in blocks. In specific implementation, the cost value of each pixel can be replaced by the sum of the cost values ​​of the surrounding neighborhood to improve the robustness of matching; and the neighborhood window size can be dynamically adjusted according to the texture complexity of the pixel area, such as 3×3, 5×5, etc. It is preferred to use a large window in a sparse texture area and a small window in a dense texture area;

[0072] A3: Perform multi-path constraint cost aggregation on the cost matrix to obtain an aggregated cost matrix; this step can achieve an effect similar to minimizing the global energy function of the global stereo matching algorithm. In the implementation process, in order to balance the accuracy and real-time performance of the algorithm matching, four-way aggregation can be performed on the cost matrix, namely horizontal, vertical and two diagonal paths;

[0073] A4: Determine the optimal disparity value for each pixel based on the aggregated cost matrix, and obtain the disparity map through post-processing optimization; specifically, the disparity corresponding to the minimum cost value can be selected as the optimal disparity value among the cost values ​​of all disparities of the pixel. The disparity optimization includes confidence detection, sub-pixel difference and left-right consistency detection;

[0074] A5: By calculating the distance information based on the disparity map, you can get the distance information between the binocular camera and the set target.

[0075] As a preferred embodiment of the above embodiment, the step of calculating the BT cost includes:

[0076] Calculate the pixel (x R -n,y) and (x R +n,y) between sub-pixel positions (x R +x,y) grayscale value

[0077] Calculate the right center of the input right eye image (x R +dn,y) and (x R +d+n,y) between the sub-pixel positions (x R +d+x,y) gray value

[0078] The BT cost is calculated by the following formula:

[0079]

[0080] cos=min(cos 1 , cos 2 );

[0081] Wherein, cos is the BT cost, cos 1 and cos 2 is the bidirectional cost value, d is the parallax, 0.1≤n≤0.9, I R (x R , y) is the sub-pixel position in the left image (x R ,y) gray value, I T (x R +d, y) is the sub-pixel position (x + d, y) in the right image that is offset by the parallax d. R +d,y) grayscale value.

[0082] In this preferred solution, the bidirectional cost value cos is calculated based on the sub-pixel grayscale value extraction of the left and right images. 1 and cos 2 This method further improves the accuracy of pixel matching while retaining image details and global features, especially in scenes with sparse or repeated textures, which helps to alleviate disparity calculation errors and generate higher quality disparity maps. In the implementation process, the grayscale value difference of the target pixel in the left and right images is calculated, combined with the bidirectional matching cost cos 1 and cos 2 , the minimum value of the substitution cost is used as the final BT cost, which effectively reduces the influence of noise and mismatch. In low-light environments, complex texture backgrounds or dynamic targets, the scheme can match image features more stably and ensure the reliability of the ranging algorithm. Even in dynamic scenes with large changes in illumination, the refined grayscale value interpolation calculation can enhance the algorithm's ability to capture target details, thereby reducing the mismatch problem caused by illumination changes. The disparity optimization strategy based on the selection of the optimal cost value in the scheme can complete the entire process of cost calculation at a low computational cost, which is suitable for the high real-time requirements of the vehicle-mounted system. At the same time, the calculation method of the BT cost maintains high accuracy while having moderate computational complexity, which is conducive to real-time operation on embedded hardware platforms.

[0083] Embodiment 2

[0084] like Figure 3 As shown, the anti-glare control method of DLP intelligent vehicle lamp includes:

[0085] H1: Using the vehicle driving environment perception method as in Example 1, the environment in front of the vehicle is perceived to obtain a perception result, wherein the perception result includes a set target in the image to be detected and its pixel coordinate position, and a distance from the set target to the binocular camera;

[0086] H2: Based on the optical characteristics of DLP smart headlights, a correspondence between the set target and the light zone of the headlight is established to determine the area that needs brightness adjustment; the identified set target is associated with the light area of ​​the headlight to facilitate the subsequent dynamic adjustment of the light beam distribution of the headlight; specifically, the target pixel coordinates and distance information obtained in step H1 are used to determine the position of the target object in the actual space, and the light zone range where the target is located, that is, the area that needs to be shielded, is determined in combination with the light zone correspondence in step H2; during the implementation process, the shading area needs to be updated in real time according to the movement of the target object, such as oncoming vehicles and pedestrians. The shading area needs to change with the vehicle speed and the relative speed of the target to ensure that the position that may cause glare is always covered. The update process is realized based on the real-time perception of the vehicle's driving environment;

[0087] H3: The DMD control chip is used to independently control several micromirrors to adjust the brightness of each area.

[0088] The DMD control chip of DLP smart car lights divides the area covered by the white laser light source into millions of independently controllable pixels. The DMD chip is composed of millions of independently controllable micromirrors, each of which can quickly switch between + / -12 degrees. Brightness adjustment is achieved by controlling the flip timing of the micromirrors.

[0089] Through the above control, DLP smart car lights can perform pixel-level precise control of the lighting area, avoiding the limitations of traditional mechanical shading, with a fast reaction speed and real-time response to changes in road conditions, providing the best lighting effect while ensuring safe driving.

[0090] As a preference of the above embodiment, in the process of determining the area requiring brightness adjustment, the determination is made in combination with the intensity of ambient light.

[0091] As a further optimization method of step H3, the brightness of each area can be adjusted by independently controlling a plurality of micromirrors through a DMD control chip, such as Figure 4 As shown, including:

[0092] H31: Calculate the brightness requirement of each light zone based on the perception results and ambient light intensity;

[0093] H32: Convert the brightness requirement of each light zone into a corresponding PWM signal and generate a duty cycle;

[0094] H33: Drive the micromirror switch through PWM signal to adjust the brightness of the light zone.

[0095] In the above process, the brightness requirements of each light zone are accurately calculated based on the perception results and ambient light intensity, which can effectively avoid glare interference to oncoming vehicles, pedestrians or other road users. The precise dimming range can avoid unnecessary over-illumination, while ensuring the driver's clear field of vision and improving the safety of night driving; by real-time analysis of ambient light intensity and target detection results, the system can dynamically adjust the brightness of the light zone to adapt to changes in road conditions and lighting conditions. In the implementation process, the timing control coordinates all micro-mirror actions through a precise timing controller, which is usually implemented using a dedicated ASIC or FPGA, and the refresh rate needs to reach more than 60Hz to avoid flickering; the micro-mirror actions in the same area are based on a unified duty cycle to ensure uniform brightness of the light zone, and the micro-mirrors in different areas work independently with their own duty cycles while maintaining the overall synchronization of the PWM cycle.

[0096] As a preference of the above embodiment, the ambient light intensity is detected by a light sensor, and / or is identified by an image to be detected captured by a binocular camera.

[0097] In this preferred embodiment, the light sensor is specifically used to measure the intensity of ambient light, usually as an overall brightness value in lux, and can quickly and directly provide the current ambient light intensity value, without relying on other perception systems, and can independently complete the ambient light measurement; the image obtained by the binocular camera can be used to indirectly deduce the ambient light intensity through an image processing algorithm; in this preferred embodiment, the above two methods can be used simultaneously to assist each other, or they can be used alone, both of which are within the scope of protection of the present invention.

[0098] As a specific way, Figure 5 As shown, when the ambient light intensity is identified by the image to be detected collected by the binocular camera, it includes:

[0099] B1: The left and right images are collected by binocular cameras and aligned in time and space; thus, a pair of views is generated to ensure that the scenes captured by the two images are consistent. In the specific implementation process, epipolar correction can be performed to adjust the corresponding scene points in the left and right images to the same line to ensure that the parallax exists only in the horizontal direction during stereo matching. To ensure the implementation effect, the left and right images can be distorted to remove the deformation caused by the optical lens;

[0100] B2: Obtain the set target and its pixel coordinate position detected in the image to be detected using the target detection model, and remove the set target from the left and right images;

[0101] Eliminating the abnormal local brightness area can avoid its interference with the evaluation of ambient light intensity. This step can avoid the occurrence of additional costs with the help of the results of the previous steps, and can prevent misjudgment caused by dynamic targets, such as moving vehicles or pedestrians, and abnormal light sources;

[0102] B3: Convert the remaining parts of the left and right images after the removal into grayscale images; extract brightness information from the remaining parts of the image after removing the local interference area to ensure that the evaluation result can better reflect the real ambient light intensity; specifically, convert the left and right images after the target is removed into grayscale images, extract the brightness component, and remove the interference of color information; as a preferred method, the grayscale image can be divided into several small blocks, such as 8×8 or 16×16 pixels. The block division can localize the brightness analysis and reduce the impact of single-point brightness changes on the global judgment;

[0103] B4: Calculate the average grayscale value as the current ambient light intensity. Taking the above-mentioned block implementation as an example, this step can calculate the average grayscale value of each block, obtain the brightness information of each block, and then calculate the average value; after eliminating the set target, the global brightness statistical result can better reflect the actual ambient light intensity. The global brightness intensity, as the core parameter of the headlight adjustment, can provide a reliable basis for the brightness control of the intelligent headlight.

[0104] Among them, the set target is removed from the left and right images, specifically:

[0105] The pixel values ​​of the area corresponding to the set target are replaced with the set values, such as the average gray value of the whole image, or other specific values; or, the area corresponding to the set target is marked as invalid and does not participate in the brightness calculation.

[0106] As another implementation, when the ambient light intensity is detected by both the light sensor and the image to be detected captured by the binocular camera, it includes:

[0107] C1: Based on the detection result of the optical sensor, when the difference between the previous and the next detection results is greater than the set value, the recognition of the image to be detected is started;

[0108] C2: Revise the basis according to the recognition result of the image to be detected to obtain the final ambient light intensity.

[0109] In the above process, triggering image recognition on demand can reduce power consumption and computing resource usage, flexibly cope with complex lighting conditions, and dual data sources improve system reliability. As a specific way to revise the basis according to the recognition result of the image to be detected, the weight corresponding to the recognition result of the image to be detected can be obtained by the average grayscale value obtained in step B4 of the above embodiment, thereby calculating the weight corresponding to the detection result of the photoelectric sensor, and revising the detection result of the light sensor. By assigning weights and summing them, the efficient fusion of the two detection methods can be achieved, giving full play to the global detection capability of the light sensor and the local analysis capability of the binocular camera, and providing reliable guarantee for the accurate assessment of the ambient light intensity. In the implementation process, since the grayscale value of the image reflects the brightness of the light, but the unit is different from the lux value detected by the light sensor, the grayscale value can be converted into a weight factor comparable to the lux value through a mapping function.

[0110] The following is an example of revision by weight. Of course, other methods that can achieve the above revision purpose are also within the scope of protection of the present invention:

[0111] L final =W sensor ·L sensor +W camera ·L cauera (1);

[0112] W sensor =1-W camera (2);

[0113] Among them, L final is the final revised value of ambient light intensity;

[0114] L sensorThe light intensity value detected by the light sensor;

[0115] L camera is the light intensity value obtained by mapping the image gray value;

[0116] W sensor and W camera are weight factors corresponding to the detection result of the light sensor and the recognition result of the image to be detected respectively.

[0117] Alternatively, as another method in which the light sensor and the binocular camera jointly participate in the confirmation of the ambient light intensity, the ratio of the area of ​​the set target to be removed from the left and right images to the total area of ​​the left and right images can be calculated, and W in the above formula (1) can be adjusted. sensor and W camera When the culling ratio is high, increase W camera The size of the ambient light intensity can be more accurately integrated to obtain a more reliable ambient light intensity assessment result.

[0118] During the implementation process, dynamic environment classification and light intensity change compensation strategies can also be implemented, specifically dynamically classifying the ambient light intensity scenes based on the elimination area ratio:

[0119] Low rejection ratio (<10%): The scene lighting is uniform and the detection results of the light sensor can be used directly;

[0120] Medium rejection ratio (10% 50%): The local brightness is abnormal, but the proportion is not high. The light sensor results are revised by slightly adjusting the weight factor, such as changing W sensor The decline is controlled within 10% to 20%;

[0121] High rejection ratio (>50%): Local brightness anomalies occupy a large area. The light sensor results are revised by significantly adjusting the weight factors, such as changing W sensor The decline is controlled between 20% and 50%.

[0122] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A vehicle driving environment perception method, characterized in that: Based on a binocular camera, the binocular camera is installed at a set position on the vehicle body, and the method includes: The road condition information within the shooting range is collected by the binocular camera, and the collected results are transmitted in real time as the image to be detected to the target detection model deployed on the vehicle-mounted computer; Predicting the image to be detected by using the target detection model to detect the set target and its pixel coordinate position in the image to be detected; A distance measurement algorithm is used to calculate the distance from the set target in the image to be detected to the binocular camera.

2. The vehicle driving environment perception method according to claim 1, characterized in that: The implementation steps of the ranging algorithm include: For each pixel of the image to be detected, a cost value of pixel matching is obtained by fusing the BT cost and the cost processed by the SobelX operator; Locally smoothing the cost value of each pixel to obtain a cost matrix; Performing multi-path constraint cost aggregation on the cost matrix to obtain an aggregated cost matrix; Determining an optimal disparity value for each pixel based on the aggregated cost matrix, and obtaining a disparity map through post-processing optimization; Distance information is calculated based on the disparity map.

3. The vehicle driving environment perception method according to claim 2, characterized in that: The steps of calculating the BT cost include: Calculate the pixel (x R -n,y) and (x R +n,y) between sub-pixel positions (x R +x,y) grayscale value x∈[-n,n]; Calculate the right center of the input right eye image (x R +dn,y) and (x R +d+n,y) between the sub-pixel positions (x R +d+x,y) gray value x∈[-n,n]; The BT cost is calculated by the following formula: cos = min(cos1, cos2); Where cos is the BT cost, cos1 and cos2 are the bidirectional cost values, d is the disparity, 0.1≤n≤0.9, I R (x R , y) is the sub-pixel position in the left image (x R ,y) gray value, I T (x R +d, y) is the sub-pixel position (x + d, y) in the right image that is offset by the parallax d. R +d,y) grayscale value.

4. The anti-glare control method of DLP intelligent vehicle lamp is characterized in that: include: Using the vehicle driving environment perception method according to any one of claims 1 to 3, the environment in front of the vehicle is perceived to obtain a perception result, wherein the perception result includes a set target in the image to be detected and its pixel coordinate position, and a distance from the set target to the binocular camera; In combination with the optical characteristics of the DLP smart headlight, a corresponding relationship between the set target and the headlight area is established to determine the area where brightness adjustment is required; The DMD control chip is used to independently control several micromirrors to adjust the brightness of each area.

5. The anti-glare control method of DLP intelligent vehicle lamp according to claim 4, characterized in that: In the process of determining the area that needs to be adjusted in brightness, the determination is made in combination with the ambient light intensity.

6. The anti-glare control method of DLP intelligent vehicle lamp according to claim 5, characterized in that: The DMD control chip is used to independently control several micromirrors to adjust the brightness of each area, including: Calculate the brightness requirement of each light zone according to the perception result and the ambient light intensity; Convert the brightness requirement of each light zone into a corresponding PWM signal to generate a duty cycle; The micromirror switch is driven by PWM signal to adjust the brightness of the light zone.

7. The anti-glare control method of DLP intelligent vehicle lamp according to claim 4, characterized in that: The ambient light intensity is detected by a light sensor and / or is identified by an image to be detected captured by a binocular camera.

8. The anti-glare control method of DLP intelligent vehicle lamp according to claim 7, characterized in that: The ambient light intensity is identified by using the image to be detected collected by the binocular camera, including: The left and right images are collected by the binocular camera and aligned in time and space; Acquire a set target and its pixel coordinate position detected in the image to be detected by using the target detection model, and remove the set target from the left and right images; Convert the remaining parts of the left and right images after elimination into grayscale images; Calculate the average grayscale value as the current ambient light intensity.

9. The anti-glare control method of DLP intelligent vehicle lamp according to claim 8, characterized in that: The set target is removed from the left and right images, specifically: Replacing the pixel value of the area corresponding to the set target with the set value; Or, the area corresponding to the set target is marked as invalid and does not participate in the brightness calculation.

10. The anti-glare control method of DLP intelligent vehicle lamp according to claim 7, characterized in that: The ambient light intensity is detected by the light sensor and also recognized by the image to be detected collected by the binocular camera, including: Based on the detection result of the optical sensor, when the difference between the previous and the next detection results is greater than a set value, the recognition of the image to be detected is started; The basis is revised according to the recognition result of the image to be detected to obtain the final ambient light intensity.

Images