Anti-dazzle dynamic dimming control system and method based on ambient light induction

Abstract

The invention discloses an anti-dazzle dynamic dimming control system and method based on ambient light induction, and particularly relates to the technical field of dimming control. According to the anti-dazzle dynamic dimming control system and method based on ambient light induction, a picture of the environment in front of a vehicle lamp is captured, a video stream collected by a camera is obtained, the picture is processed, and then the type of a target in the picture and the position of a pixel where the target is located are judged. And according to the installation height of the camera, the target bottom pixel ordinate and the vertical pixel offset of the reference line in the image relative to the horizontal axis, and in combination with the focal length of the camera, calculating and adjusting a threshold value. And when the adjustment threshold value exceeds a preset threshold value, a ground hypothesis method constraint formula is adopted to calculate depth information of the target. And generating an anti-dazzle dark block for the normal lighting picture according to the type and the position information of the target. The size and darkness of the generated dark block are dynamically adjusted according to the difference value between the illumination intensity of the headlamp and the environment illumination intensity and the target depth information.

Description

Technical Field

[0001] The present invention relates to the field of dynamic dimming technology, and more specifically, to an anti-glare dynamic dimming control system and method based on ambient light sensing. Background Art

[0002] With the acceleration of urbanization and the development of science and technology, artificial light sources have been widely used in various places. Due to the inappropriate intensity and direction of light, light pollution and glare have occurred, affecting people's quality of life and work. In road lighting, strong light sources can cause discomfort to the eyes, even affect vision, cause visual fatigue, and lead to accidents. Therefore, solving these problems is of great significance to improving environmental comfort and visual health.

[0003] In the prior art, when the road surface is uneven, the calculated target depth information is inaccurate, which affects the subsequent generation of anti-glare dark blocks, resulting in poor anti-glare effect; and the anti-glare dark blocks of fixed size and darkness cannot adapt to different environments. In order to solve the above two defects, a technical solution is now provided. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides an on-site comprehensive maintenance system for electrical equipment in a distribution network to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions: The anti-glare dynamic dimming control method based on ambient light sensing includes the following steps: Capture the environment in front of the headlights, obtain the video stream captured by the camera, process the image and determine the type of target in the image and the pixel position of the target; The adjustment threshold is calculated according to the ratio of the camera installation height and the pixel ordinate of the target bottom, the camera focal length, and the vertical pixel offset of the reference line in the image relative to the horizontal axis. When the adjustment threshold is greater than the preset threshold, the ground assumption method constraint formula is introduced to calculate the target depth information. According to the target type and target position information, anti-glare dark blocks are generated for the normally lit picture. The size and darkness of the generated dark blocks are dynamically controlled according to the difference between the current headlight light intensity and the ambient light intensity and the target depth information.

[0006] In a preferred embodiment, YOLOv5+StrongSORT is used to obtain the video stream captured by the camera, process the picture and determine the type of the target in the picture and the pixel position of the target, and continuously track the target through the StrongSORT algorithm.

[0007] In a preferred embodiment, a monocular depth estimation is performed based on a known set of real heights and corresponding pixel data clues; the depth information of important objects in the picture is calculated using the following formula, which is as follows: ; Where f represents the focal length of the camera; Represents the actual physical height of the target object; Represents the pixel height of the target in the image.

[0008] In a preferred embodiment, the ratio of the camera installation height and the pixel ordinate of the bottom of the target, as well as the camera focal length and the vertical pixel offset of the reference line in the image relative to the horizontal axis are determined; the distance between the vehicle and the target is calculated by the ratio of the camera installation height and the pixel ordinate of the bottom of the target; the inclination of the vehicle is calculated by the camera focal length and the vertical pixel offset of the reference line in the image relative to the horizontal axis. The specific formula is as follows: ; f represents the focal length of the camera; Represents the vertical pixel offset of the reference line in the image relative to the horizontal axis.

[0009] In a preferred embodiment, the ratio of the camera installation height to the pixel ordinate at the bottom of the target and θ are determined, and the adjustment threshold is calculated by weighted summation. When the adjustment threshold is greater than the system preset threshold, the ground hypothesis constraint formula is introduced to calculate the depth information of the important target in the current picture. The formula is as follows: ; represents the camera installation height; f represents the camera focal length; v represents the pixel vertical coordinate of the bottom of the target; Represents the ordinate of the principal point of the image.

[0010] In a preferred embodiment, based on the target type and target position information, an anti-glare dark block is generated for the normal lighting picture, the difference between the current headlight illumination intensity and the ambient illumination intensity and the target depth information are determined, and the adjustment coefficient is calculated by weighted summation.

[0011] In a preferred embodiment, the darkness of the dark block is adjusted according to the adjustment coefficient; the size of the dark block is negatively correlated with the target depth information.

[0012] In a preferred embodiment, the anti-glare dynamic dimming control system based on ambient light sensing includes the following modules: a video capture preprocessing module, a depth information calculation module, and a projection module; The video capture preprocessing module is used to capture the environment in front of the headlights; then YOLOv5+StrongSORT is used to obtain the video stream captured by the camera, process the image and determine the type of target in the image and the pixel position of the target, output the target frame, frame the target; and continuously track the target through the StrongSORT algorithm; The depth information calculation module is used to perform monocular depth estimation based on a known set of real heights and measured corresponding pixel data clues; calculate the depth information of important targets in the current picture; The depth information calculation module includes a decision module for determining whether to introduce a ground hypothesis constraint to calculate the depth information based on the ratio of the camera installation height to the pixel ordinate at the bottom of the target, the camera focal length, and the vertical pixel offset of the reference line in the image relative to the horizontal axis; The projection module is used to decide the projection mode and projection strategy according to the target type and target location information.

[0013] In a preferred embodiment, the depth information calculation module includes a decision module for determining whether to introduce ground hypothesis constraints to calculate the depth information based on the ratio of the camera installation height and the vertical coordinate of the target bottom pixel, the camera focal length, and the vertical pixel offset of the baseline in the image relative to the horizontal axis.

[0014] In a preferred embodiment, the projection module includes an adjustment module for dynamically controlling the size and darkness of the generated dark block according to the difference between the current headlight illumination intensity and the ambient illumination intensity and the target depth information.

[0015] Technical effects and advantages of the present invention: The present invention firstly captures the environment in front of the headlights through the roof camera; then uses YOLOv5+StrongSORT to obtain the video stream captured by the camera, processes the picture and determines the type of the target in the picture and the pixel position of the target, outputs the target frame, and frames the target. And continuously tracks the target through the StrongSORT algorithm; outputs a specific number and the target frame and its trajectory information under the same number; performs monocular depth estimation based on a known set of real heights and measured corresponding pixel data clues; calculates the camera focal length through known information, and calculates the depth information of important targets in the current picture by combining the typical height information of vehicles, pedestrians, bicycles and motorcycles. When the ground is uneven, the measured height information is inaccurate, which affects the calculation of depth information; introduces the ground hypothesis method on the basis of the similar triangle method for calculation, and comprehensively judges whether it is necessary to introduce the ground hypothesis method to calculate the target depth information according to the ratio of the camera installation height and the vertical coordinate of the pixel at the bottom of the target, as well as the camera focal length and the vertical pixel offset of the reference line in the image relative to the horizontal axis. The ground hypothesis method uses the camera installation height and the vertical coordinate of the pixel at the bottom of the target to derive the target depth information. This makes the calculation of target depth information more accurate, suitable for complex terrain, and facilitates the subsequent generation of dark blocks.

[0016] Based on the three-dimensional information of the target, that is, the two-dimensional pixel coordinates and depth information, the pixel coordinate system is substituted into the camera coordinate system formula to obtain the three-dimensional spatial position information of the target relative to the camera coordinate system; according to the target type and target position information, the projection mode and projection strategy are determined. That is, for the normal lighting picture, an anti-glare dark block is generated to block the upper body of the pedestrian and avoid direct projection to achieve an anti-glare effect. The size and darkness of the generated dark block can be dynamically controlled according to the difference between the current headlight light intensity and the ambient light intensity and the target depth information. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to facilitate understanding by those skilled in the art, the present invention is further described below in conjunction with the accompanying drawings; Figure 1 It is a flow chart of the anti-glare dynamic dimming control method based on ambient light sensing of the present invention; Figure 2 It is a structural schematic diagram of the anti-glare dynamic dimming control system based on ambient light sensing of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] Example 1 The present invention provides an anti-glare dynamic dimming control method based on ambient light sensing, such as Figure 1 As shown, the following steps are included: First, the environment in front of the headlight is captured by the roof camera; it is determined that the image captured by the camera includes the influence range of the vehicle headlight; Then use YOLOv5+StrongSORT to obtain the video stream captured by the camera, process the image and determine the type of target in the image and the pixel position of the target, output the target frame, and frame the target. And continue to track the target through the StrongSORT algorithm; output a specific number and the target frame and its trajectory information under the same number; Monocular depth estimation is performed based on a known set of real heights and corresponding pixel data clues. The camera focal length is calculated based on the known information, and the depth information of important targets in the current picture is calculated based on the typical height information of vehicles, pedestrians, bicycles and motorcycles.

[0020] When the ground is uneven, the measured height information is inaccurate, which affects the calculation of depth information. Based on the similar triangle method, the ground assumption method is introduced for calculation. According to the ratio of the camera installation height and the vertical coordinate of the pixel at the bottom of the target, the camera focal length and the vertical pixel offset of the reference line in the image relative to the horizontal axis, it is comprehensively judged whether the ground assumption method needs to be introduced to calculate the target depth information. The ground assumption method uses the camera installation height and the vertical coordinate of the pixel at the bottom of the target to derive the target depth information. This makes the calculation of target depth information more accurate, suitable for complex terrain, and convenient for the subsequent generation of dark blocks.

[0021] Based on the three-dimensional information of the target, that is, the two-dimensional pixel coordinates and depth information, substitute the pixel coordinate system into the camera coordinate system formula to obtain the three-dimensional spatial position information of the target relative to the camera coordinate system; The projection mode and projection strategy are determined based on the target type and target position information. That is, for the normal lighting picture, an anti-glare dark block is generated to block the upper body of pedestrians and avoid direct projection to achieve an anti-glare effect. The size and darkness of the generated dark block can be dynamically controlled based on the difference between the current headlight intensity and the ambient light intensity and the target depth information. The greater the difference between the headlight intensity and the ambient light intensity and the smaller the target depth information, the darker the dark block will be. The size of the dark block is negatively correlated with the target depth information.

[0022] Specifically, First, the camera on the roof captures the environment in front of the headlights. Make sure that the image captured by the camera includes the range of influence of the vehicle headlights. Avoid blind spots that affect the acquisition of video streams. Then use YOLOv5+StrongSORT to get the video stream captured by the camera. YOLOv5 is a popular target detection algorithm, which is an iterative version in the YOLO series. The design concept of YOLOv5 is to provide high accuracy while maintaining high detection speed. It is widely used in various computer vision tasks. The network architecture of YOLOv5 includes multiple versions, such as YOLOv5s (small), YOLOv5m (medium), YOLOv5l (large), and YOLOv5x (extra large) to adapt to different performance requirements and resource constraints. These versions provide different trade-offs between model size, speed, and accuracy.

[0023] The network structure of YOLOv5 is the basis of its efficient object detection performance. This architecture inherits and develops the design concept of the YOLO series, aiming to quickly and accurately detect multiple objects in an image through a single forward propagation. The network structure of YOLOv5 consists of four key parts: input layer, backbone network, neck, and detection head.

[0024] Input layer: The input layer is responsible for receiving the input image and performing preprocessing, including resizing the image to match the dimensions required by the network. YOLOv5 supports multi-scale input, which means it can handle images of different resolutions, thereby improving the adaptability and detection efficiency of the model.

[0025] Backbone network: The backbone network is the core part of the model, responsible for extracting high-level features from the input image. YOLOv5 uses CSPDarknet53 as its backbone network, which is a deep convolutional neural network that reduces the amount of computation and improves the efficiency of information flow by introducing the CrossStage Partial Network structure. CSPDarknet53 contains multiple convolutional layers, residual connections, and downsampling operations, designed to capture rich features from shallow to deep layers.

[0026] Neck: The neck part located between the backbone network and the detection head is responsible for deeply integrating and processing feature information at different levels. To achieve this goal, YOLOv5 uses the structural design of feature pyramid network and path aggregation network. FPN enhances the transmission of high-level semantic features at different scales through a top-down path, while PANet enhances the utilization of low-level features through a bottom-up path. This two-way feature fusion strategy enables the model to effectively detect objects of different sizes.

[0027] Detection head: The detection head is the last part of the YOLOv5 network structure, responsible for generating the final detection results based on the fused feature maps. This includes the location (bounding box), category, and confidence of the target. The detection head of YOLOv5 uses multiple parallel convolutional layers to predict target information at different scales. The prediction of each scale is processed by a specific convolutional layer, thereby achieving a wide range of detection for small objects to large objects.

[0028] In general, the network structure of YOLOv5 achieves fast and accurate detection of targets through a carefully designed backbone network, efficient feature fusion strategy, and multi-scale detection capabilities. The optimization of this structure aims to balance detection performance and computational efficiency, making YOLOv5 not only suitable for high-performance servers, but also able to run effectively on edge devices.

[0029] The loss function of YOLOv5 is the core of its learning process. It is carefully designed to ensure the high performance of the model on the object detection task. This function combines three key parts: bounding box localization loss, object classification loss, and confidence loss. These parts are explained in detail below: Bounding box positioning loss: The bounding box positioning loss uses CIoU loss. This method not only focuses on the position and size accuracy of the bounding box, but also comprehensively considers the consistency of the bounding box shape and the distance difference of the center point. The expression of CIoU loss is as follows: ; Among them, IoU measures the overlap between the predicted box and the real box. is the Euclidean distance between the predicted box and the center point of the real box, c is the length of the diagonal of the minimum closure area covering the two boxes, v considers the consistency of the aspect ratio, and α is an adjustment parameter to balance the impact of the aspect ratio. This composite loss ensures that the model fully considers the size, shape, and position of the bounding box during the learning process.

[0030] Target classification loss: Target classification loss uses the cross entropy loss function, which is a standard method for evaluating the consistency between the model's predicted category and the true category. Its purpose is to minimize the difference between the predicted category distribution and the true category label. The formula is as follows: ; here, represents the one-hot encoding vector of the true category, and Represents the category probability predicted by the model. By optimizing this loss, YOLOv5 effectively improves the accuracy of identifying various types of targets.

[0031] Confidence loss: Confidence loss focuses on the model's prediction accuracy of whether the target exists or not, and uses binary cross entropy to quantify the difference between the predicted confidence and the actual situation. The specific expression is: ; Here, y is the true value of whether the target exists (1 if it exists, 0 otherwise), is the probability of existence predicted by the model. This part of the loss helps the model distinguish the background from the foreground more accurately and reduce false positives and missed positives.

[0032] In summary, the loss function of YOLOv5 achieves an excellent balance in target detection, classification, and confidence prediction through a carefully designed combination, which is the key to achieving efficient and accurate target detection in diverse and complex environments.

[0033] The core technical architecture of the SORT method is based on the collaborative mechanism of motion state prediction and data association. Specifically, the Kalman filter dynamically predicts the motion trajectory of the object between consecutive video frames by establishing a target motion model; at the same time, the Hungarian algorithm, as an optimization matching tool, is responsible for optimally associating the predicted trajectory with the target bounding box detected in the current frame. The limitations of this method are mainly reflected in the single dimension of feature modeling. Since the apparent feature information such as texture and color of the target is not integrated, when complex situations such as target occlusion and intensive cross-motion occur in the scene, the tracking system is prone to problems such as trajectory breakage or identity confusion.

[0034] DeepSORT is a multi-target tracking algorithm based on deep learning, which is widely used in video surveillance, autonomous driving, behavior analysis and other fields. DeepSORT is an extension of the SORT algorithm; DeepSORT introduces appearance features based on SORT, extracts the appearance information of the target through a deep learning model, and significantly reduces the ID switching problem caused by occlusion or temporary disappearance.

[0035] The target position is predicted by using Kalman filtering, and the similarity between the prediction and the detection box is measured by Mahalanobis distance. The cosine similarity of the deep feature vector is combined to ensure the consistency of the target identity. The matching robustness is improved by weighted fusion of motion and appearance information (such as 0.6 weight for appearance).

[0036] Prioritize matching the most recently appeared targets to avoid newly detected targets being mistakenly associated with long-lost tracks, thereby solving the tracking problem in frequent occlusion scenarios.

[0037] StrongSORT further improves the feature extraction and data association parts of DeepSORT. It uses a more advanced feature extraction network and a more complex data association strategy to improve the accuracy and robustness of tracking. StrongSORT especially strengthens the ability to handle occlusion and interaction in complex scenes.

[0038] Optimization algorithm of apparent feature branch: In terms of the apparent feature branch of the StrongSORT algorithm, two key optimizations are implemented: first, the performance of the feature extractor is enhanced; second, the exponential moving average (EMA) is adopted as the feature update mechanism to replace the feature library method used in the DeepSORT algorithm.

[0039] Optimization algorithm of motion model branch: In the motion model branch of the StrongSORT algorithm, two key optimizations were performed: first, compensation for camera motion was achieved by adopting the enhanced correlation coefficient (ECC) algorithm; second, the NSA (nonlinear robust adaptive) Kalman filter algorithm was introduced to replace the traditional Kalman filter algorithm, thereby improving the accuracy and stability of tracking.

[0040] StrongSORT model construction, first, the detection results of the YOLOv5s algorithm provide the initial input for the entire tracking process. Based on the output of the current video frame, the StrongSORT algorithm generates a list of all detected targets. Then, the NSA Kalman filter algorithm is used to predict the possible position and state of each target in the new frame based on the tracking data of the previous frame. In order to match the tracking data with the new detection results, the algorithm combines the apparent features of the target and uses a global linear matching mechanism based on the Hungarian algorithm for association. When the initial association is not completely successful, the algorithm further adopts an IOU matching strategy to increase the matching success rate of the tracked target. This series of operations ensures that the algorithm can identify which tracked targets match the detection results, which are still unmatched, and identify new unmatched detection objects. After completing the matching process, the StrongSORT algorithm uses the NSA Kalman filter again to update the tracking data. This step is to optimize the tracking results and improve their accuracy and robustness in the current frame. The whole process not only ensures the continuity and stability of tracking, but also enhances the algorithm's adaptability to occlusion and dynamic changes, thereby achieving efficient and accurate target tracking in real-time tracking tasks.

[0041] Monocular depth estimation is performed based on a known set of real heights and corresponding pixel data clues. The focal length of the camera is determined, and the depth information of important objects in the current image is calculated by combining the typical height information of vehicles, pedestrians, bicycles and motorcycles.

[0042] Furthermore, the depth information of important objects in the picture is calculated using the following formula: ; where f represents the focal length of the camera, which can be calculated offline using the chessboard calibration method; Represents the real physical height of the target object, which is dynamically updated (e.g., through continuous tracking optimization) by presetting a typical target height library; Represents the pixel height of the target in the image, which is calculated by the bounding box output by detection models such as YOLO.

[0043] Furthermore, the target needs to be upright and highly stable (such as a pedestrian standing or a vehicle not overturned). If the target is bent or tilted, Need to be corrected according to the joint angle (such as when bending down to 1.3m).

[0044] Determine the ratio of the camera installation height to the pixel ordinate at the bottom of the target, as well as the camera focal length and the vertical pixel offset of the reference line in the image relative to the horizontal axis; roughly estimate the distance between the vehicle and the target through the ratio of the camera installation height to the pixel ordinate at the bottom of the target; calculate the vehicle inclination through the camera focal length and the vertical pixel offset of the reference line in the image relative to the horizontal axis. The specific formula is as follows: ; f represents the focal length of the camera; Represents the vertical pixel offset of the reference line in the image relative to the horizontal axis.

[0045] Furthermore, the threshold G is adjusted by weighted summation calculation, and the formula is as follows: G=c*BZ+d*θ; wherein BZ represents the ratio of the camera installation height to the pixel ordinate of the target bottom; c and d are the weight coefficients of the ratio of the camera installation height to the pixel ordinate of the target bottom and θ, respectively, and both are greater than zero.

[0046] When G is greater than the preset threshold of the system, it means that the ground is uneven. The simple similarity triangulation method cannot accurately calculate the target depth information. It is necessary to introduce the ground hypothesis constraint formula to calculate the depth information of important targets in the current picture. The formula is as follows: ; represents the camera installation height (vertical distance from the ground), which needs to be measured after installation and needs to be calibrated regularly (for example, tire wear causes ±0.1m change); f represents the camera focal length, which is obtained in the same way as above; v represents the pixel vertical coordinate of the bottom of the target, that is, the coordinate of the center point at the bottom of the target detection frame; Represents the ordinate of the principal point of the image (center of the optical axis).

[0047] Furthermore, the camera pitch angle is monitored in real time through the angle sensor. When the camera pitch angle deviates by 1°, the depth error at 10m is about 0.3m, which requires dynamic compensation with IMU. Instantaneous changes need to be corrected by the suspension system motion model.

[0048] Based on the three-dimensional information of the target, that is, the two-dimensional pixel coordinates and depth information, substitute the pixel coordinate system into the camera coordinate system formula to obtain the three-dimensional spatial position information of the target relative to the camera coordinate system; The projection mode and projection strategy are determined based on the target type and target location information. That is, for the normal lighting screen, an anti-glare dark block is generated to block the upper body of pedestrians and avoid direct projection, thus achieving an anti-glare effect.

[0049] Furthermore, the size and darkness of the generated dark block can be dynamically controlled according to the difference between the current headlight illumination intensity and the ambient illumination intensity and the target depth information.

[0050] Specifically, the adjustment coefficient is calculated by weighted summation, and the formula is as follows: T=a*cz+b*Z; wherein T represents the adjustment coefficient; cz represents the difference between the current headlight intensity and the ambient light intensity; Z represents the target depth information; a and b are the difference between the current headlight intensity and the ambient light intensity and the weight coefficient of the target depth information, respectively; wherein the greater the difference between the headlight intensity and the ambient light intensity and the smaller the target depth information, the darker the dark block, and the darkness of the dark block is adjusted according to the adjustment coefficient; the size of the dark block is negatively correlated with the target depth information.

[0051] The current headlight illumination intensity can be obtained through the photodiode array integrated behind the headlight, and the ambient light intensity can be obtained through the light intensity sensor installed on the roof.

[0052] Example 2 The present invention is based on the anti-glare dynamic dimming control system of ambient light sensing, such as Figure 2 As shown, it includes the following modules: video capture preprocessing module, depth information calculation module, and projection module; The video capture preprocessing module is used to capture the environment in front of the headlights; then YOLOv5+StrongSORT is used to obtain the video stream captured by the camera, process the image and determine the type of target in the image and the pixel position of the target, output the target frame, frame the target; and continuously track the target through the StrongSORT algorithm; The depth information calculation module is used to perform monocular depth estimation based on a known set of real heights and measured corresponding pixel data clues; calculate the depth information of important targets in the current picture; The depth information calculation module includes a decision module for determining whether to introduce a ground hypothesis constraint to calculate the depth information based on the ratio of the camera installation height to the pixel ordinate at the bottom of the target, the camera focal length, and the vertical pixel offset of the reference line in the image relative to the horizontal axis; The projection module is used to decide the projection mode and projection strategy according to the target type and target location information; The projection module includes an adjustment module for dynamically controlling the size and darkness of the generated dark block according to the difference between the current headlight illumination intensity and the ambient illumination intensity and the target depth information.

[0053] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0054] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0055] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0056] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0057] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. Anti-glare dynamic dimming control method based on ambient light sensing, characterized in that: The following steps are involved: Capture the environment in front of the headlights, obtain the video stream captured by the camera, process the image and determine the type of target in the image and the pixel position of the target; The adjustment threshold is calculated according to the ratio of the camera installation height and the pixel ordinate of the target bottom, the camera focal length, and the vertical pixel offset of the reference line in the image relative to the horizontal axis. When the adjustment threshold is greater than the preset threshold, the ground assumption method constraint formula is introduced to calculate the target depth information. According to the target type and target position information, anti-glare dark blocks are generated for the normally lit picture. The size and darkness of the generated dark blocks are dynamically controlled according to the difference between the current headlight light intensity and the ambient light intensity and the target depth information.

2. The anti-glare dynamic dimming control method based on ambient light sensing according to claim 1, characterized in that: Use YOLOv5+StrongSORT to obtain the video stream captured by the camera, process the picture and determine the type of target in the picture and the pixel position of the target, and continue to track the target through the StrongSORT algorithm.

3. The anti-glare dynamic dimming control method based on ambient light sensing according to claim 1, characterized in that: Monocular depth estimation is performed based on a known set of true heights and corresponding pixel data clues; Use the following formula to calculate the depth information of important targets in the picture: ; Where f represents the focal length of the camera; Represents the actual physical height of the target object; Represents the pixel height of the target in the image.

4. The anti-glare dynamic dimming control method based on ambient light sensing according to claim 1, characterized in that: Determine the ratio of the camera installation height and the pixel ordinate of the bottom of the target, as well as the camera focal length and the vertical pixel offset of the reference line in the image relative to the horizontal axis; calculate the distance between the vehicle and the target through the ratio of the camera installation height and the pixel ordinate of the bottom of the target; calculate the vehicle inclination through the camera focal length and the vertical pixel offset of the reference line in the image relative to the horizontal axis. The specific formula is as follows: ; f represents the focal length of the camera; Represents the vertical pixel offset of the reference line in the image relative to the horizontal axis.

5. The anti-glare dynamic dimming control method based on ambient light sensing according to claim 4, characterized in that: Determine the ratio of the camera installation height to the pixel ordinate at the bottom of the target and θ, and calculate the adjustment threshold by weighted summation. When the adjustment threshold is greater than the system preset threshold, introduce the ground hypothesis constraint formula to calculate the depth information of important targets in the current picture. The formula is as follows: ; represents the camera installation height; f represents the camera focal length; v represents the vertical coordinate of the bottom pixel of the target; Represents the ordinate of the principal point of the image.

6. The anti-glare dynamic dimming control method based on ambient light sensing according to claim 1, characterized in that: According to the target type and target position information, an anti-glare dark block is generated for the normal lighting picture, the difference between the current headlight light intensity and the ambient light intensity and the target depth information are determined, and the adjustment coefficient is calculated through weighted summation.

7. The anti-glare dynamic dimming control method based on ambient light sensing according to claim 6, characterized in that: The darkness of the dark block is adjusted according to the adjustment coefficient; the size of the dark block is negatively correlated with the target depth information. 8.Anti-glare dynamic dimming control system based on ambient light sensing, characterized in that: It includes the following modules: video capture preprocessing module, depth information calculation module, and projection module; The video capture preprocessing module is used to capture the environment in front of the headlights; then YOLOv5+StrongSORT is used to obtain the video stream captured by the camera, process the image and determine the type of target in the image and the pixel position of the target, output the target frame, frame the target; and continuously track the target through the StrongSORT algorithm; The depth information calculation module is used to perform monocular depth estimation based on a known set of real heights and measured corresponding pixel data clues; calculate the depth information of important targets in the current picture; The depth information calculation module includes a decision module for determining whether to introduce a ground hypothesis constraint to calculate the depth information based on the ratio of the camera installation height to the pixel ordinate at the bottom of the target, the camera focal length, and the vertical pixel offset of the reference line in the image relative to the horizontal axis; The projection module is used to decide the projection mode and projection strategy according to the target type and target location information.

9. The anti-glare dynamic dimming control system based on ambient light sensing according to claim 8, characterized in that: The depth information calculation module includes a decision module for determining whether to introduce ground hypothesis constraints to calculate depth information based on the ratio of the camera installation height and the target bottom pixel ordinate, the camera focal length, and the vertical pixel offset of the reference line in the image relative to the horizontal axis.

10. The anti-glare dynamic dimming control system based on ambient light sensing according to claim 8, characterized in that: The projection module includes an adjustment module for dynamically controlling the size and darkness of the generated dark block according to the difference between the current headlight illumination intensity and the ambient illumination intensity and the target depth information.

Images