Dynamic dimming method and system based on dual-channel camera illumination sensing

Abstract

The invention belongs to the technical field of illumination control, and discloses a dynamic dimming method and system based on dual-channel camera illumination sensing, which realizes accurate dimming control based on vehicle speed and ambient illumination through tunnel pavement grid division and dual-channel image processing in combination with dynamic target detection and real-time illumination calculation. An independent special illuminance sensor does not need to be deployed for illuminance detection, and the method has the advantages that the accuracy and energy-saving effect of tunnel illumination dimming control are improved, stable illuminance perception under the full-illumination condition is achieved, and hardware deployment and maintenance cost is reduced.

Description

Technical Field

[0001] This application relates to the technical field of lighting control. Specifically, it relates to a dynamic dimming method and system based on dual-channel camera illuminance sensing. Background Art

[0002] In intelligent lighting control systems for scenarios such as highway tunnels, lamp dimming needs to comprehensively consider multi-dimensional parameters such as traffic flow, vehicle speed, external illuminance of the tunnel, and internal illuminance of the tunnel. Current mainstream technologies usually use cameras to detect traffic flow and vehicle speed, and at the same time rely on independently deployed dedicated illuminance sensors (such as photoresistors, silicon photocells, etc.) to detect the illuminance inside and outside the tunnel. However, such dedicated sensors have significant defects: on the one hand, their single-point detection mode is difficult to cover the complex light gradients inside the tunnel (such as the strong light area at the entrance, the middle shadow area, and the exit transition area), which is likely to lead to the formation of local illuminance monitoring blind spots; on the other hand, the installation and maintenance costs of independent sensors are relatively high, and there is hardware redundancy with the video monitoring system, which does not conform to the development trend of the integration of intelligent devices. In fact, as the core device of video monitoring, the imaging pixel value of a camera has a clear linear response relationship with the environmental illuminance, and it fully has the potential to synchronously sense illuminance, but it has not been fully utilized in existing systems.

[0003] Although visible light cameras can provide rich color and detail information, their imaging performance highly depends on the environmental light: in low-illuminance scenarios (such as at night or in the middle of the tunnel), the image signal-to-noise ratio significantly decreases, and even the "black screen" phenomenon occurs; in strong light scenarios (such as direct sunlight at the tunnel entrance), it is easy to cause overexposure due to pixel saturation and lose the illuminance information of key areas. Infrared cameras can achieve clear imaging in a completely dark environment, but the grayscale images they output lack color information, and the detection effect on objects with low infrared light reflectivity (such as dark-colored vehicles) is not good. In the prior art, single-channel cameras (only visible light or only infrared) are limited by the spectral response range and cannot stably sense the target and environmental illuminance under full lighting conditions. For example, the strong light at the tunnel entrance may cause the visible light camera to be overexposed and distorted, while the infrared images in the area without light inside the tunnel are difficult to accurately reflect the actual illuminance level perceived by the human eye. Therefore, dual-channel cameras play an important role in all-weather target detection.

[0004] The prior art also has the following problems: insufficient spatio-temporal alignment accuracy of cross-spectral images leads to illuminance calculation errors; lack of collaborative optimization between dynamic target detection and illuminance perception; low coupling degree between vehicle speed estimation and illuminance adjustment strategies; and inability to achieve refined dimming control based on grid management. These problems severely restrict the intelligent level and energy-saving effect of the tunnel lighting system.

[0005] In view of the above problems, the prior art urgently needs to be improved. Summary of the Invention

[0006] The purpose of this application is to provide a dynamic dimming method and system based on dual-channel camera illuminance sensing, which has the advantages of improving the accuracy and energy-saving effect of tunnel lighting dimming control, realizing stable illuminance perception under full light conditions, and reducing the hardware deployment and maintenance costs.

[0007] In the first aspect, this application provides a dynamic dimming method based on dual-channel camera illuminance sensing for dimming control of lighting fixtures in a tunnel based on a dual-channel camera with a visible light channel and an infrared channel, including the steps of: A1. Obtain the tunnel road surface grid division result determined in advance according to the coverage range of each lighting fixture in the tunnel; each grid of the grid division result corresponds to at least one lighting fixture; A2. Collect dual-channel images of the tunnel road surface through the dual-channel camera, and perform preprocessing and cross-spectrum spatio-temporal alignment on the dual-channel images to obtain an affine transformation matrix for spatio-temporal alignment and the dual-channel images after spatio-temporal alignment; the dual-channel images include visible light channel images and infrared channel images; A3. Based on the dual-channel images after spatio-temporal alignment, perform dynamic target detection on the vehicles in the tunnel and perform time compensation on the affine transformation matrix; A4. Evaluate the average vehicle speed of each grid according to the dynamic target detection result; A5. Calculate the real-time illuminance values of each grid according to the dual-channel images after spatio-temporal alignment and the affine transformation matrix after time compensation; the real-time illuminance values include visible light illuminance values and infrared illuminance values; A6. Adjust the light intensity of the lighting fixtures corresponding to each grid according to the real-time illuminance value and the average vehicle speed evaluation result.

[0008] Preferably, in step A2, performing cross-spectrum spatio-temporal alignment on the dual-channel images includes: Taking the visible light channel image and the infrared channel image after preprocessing as the target visible light image and the target infrared image, perform SIFT feature point matching on the target visible light image and the target infrared image to obtain a preliminary affine transformation matrix; Estimate the global average motion vector of the target visible light image based on the optical flow method to correct the translation parameter of the preliminary affine transformation matrix to obtain a corrected affine transformation matrix; Taking maximizing the normalized mutual information and minimizing the SIFT feature point matching error as the goal, establish an objective function related to the affine transformation matrix; the normalized mutual information is related to the entropy and mutual information of the target visible light image and the target infrared image; Taking the corrected affine transformation matrix as the initial value of the affine transformation matrix, iteratively solve the objective function by the gradient ascent method to obtain the final affine transformation matrix; Perform an affine transformation on the target infrared image using the final affine transformation matrix, and splice the affine result with the target visible light image to obtain a two-channel image after spatio-temporal alignment.

[0009] Preferably, step A3 includes: A301. Use the improved YOLOv11 model to perform object detection on the two-channel image after spatio-temporal alignment to obtain the object detection bounding box and ReID feature vector of the vehicle; A302. Estimate the average motion vector of the target area enclosed by the object detection bounding box based on the optical flow method to correct the translation parameter of the affine transformation matrix and obtain the affine transformation matrix after time compensation.

[0010] Preferably, step A4 includes: A401. According to the object detection bounding box and the ReID feature vector, use the Deepsort algorithm to perform object tracking on each vehicle to obtain the tracking trajectory; A402. Calculate the speed of each vehicle according to the tracking trajectory; A403. For each grid, perform a weighted average operation on the speeds of the vehicles on all lanes passing through this grid to obtain the average vehicle speed of this grid.

[0011] Preferably, in step A403, the average vehicle speed of the grid is calculated by the following formula: ; ; where, is the average vehicle speed of the grid, is the speed of the i-th vehicle on all lanes of this grid, is 's weight, N is the total number of vehicles on all lanes of this grid, is the average speed of the lanes on all lanes of this grid, is the standard deviation of the speeds of the vehicles on all lanes of this grid.

[0012] Preferably, step A5 includes: A501. For each grid, calculate the visible light illumination value of each grid according to the average pixel value of the visible light channel image in this grid after spatio-temporal alignment, the linear conversion coefficient from visible light pixel value to illumination, and the preset environmental background illumination compensation value; A502. For each grid, perform an affine transformation on the infrared channel image after spatio-temporal alignment using the affine transformation matrix after time compensation, and calculate the infrared illumination value corresponding to the average pixel value of the affine transformation result in this grid to obtain the initial infrared illumination value; A503. For each grid, compare the average pixel value of the visible light channel image after spatio-temporal alignment within this grid with a preset pixel value threshold to determine the dynamic infrared compensation weight, and combine the initial infrared illuminance value to calculate the final infrared illuminance value of each grid.

[0013] Preferably, step A6 includes: A601. For each grid, calculate the dimming intensity of each grid according to the preset reference dimming intensity, the ratio of the average vehicle speed of this grid to the maximum designed vehicle speed of the tunnel, the preset speed influence coefficient, the preset tunnel lighting required illuminance, and the maximum value among the visible light illuminance value and the infrared illuminance value of this grid; A602. Adjust the light intensity of the lighting fixtures corresponding to each grid according to the dimming intensity of each grid.

[0014] Preferably, in step A601, calculate the dimming intensity of each grid according to the following formula: ; where, is the dimming intensity of the j-th grid, is the reference dimming intensity, is the average vehicle speed of the j-th grid, is the maximum designed vehicle speed of the tunnel, is the speed influence coefficient, is the tunnel lighting required illuminance, is the visible light illuminance value of the j-th grid, is the infrared illuminance value of the j-th grid.

[0015] Preferably, in step A501, set the visible light illuminance value of the grid containing the vehicle with its headlights on to be invalid; In step A601, if the visible light illuminance value of a grid is invalid, then the maximum value among the visible light illuminance value and the infrared illuminance value of this grid is the infrared illuminance value.

[0016] In a second aspect, the present application provides a dynamic dimming system based on dual-channel camera illuminance sensing, including a dual-channel camera with a visible light channel and an infrared channel, a plurality of lighting fixtures arranged above the tunnel road surface, and a control terminal; The dual-channel camera is used to collect dual-channel images of the tunnel road surface and upload them to the control terminal; the dual-channel images include visible light channel images and infrared channel images; The control terminal is used to execute: Obtain the tunnel road surface grid division result determined in advance according to the coverage range of each lighting fixture in the tunnel; each grid of the grid division result corresponds to at least one lighting fixture; Preprocess the dual-channel image and perform cross-spectral spatio-temporal alignment to obtain an affine transformation matrix for spatio-temporal alignment and the dual-channel image after spatio-temporal alignment; the dual-channel image includes a visible light channel image and an infrared channel image; Based on the dual-channel image after spatio-temporal alignment, perform dynamic target detection on the vehicles in the tunnel and perform time compensation on the affine transformation matrix; Evaluate the average vehicle speed of each grid according to the dynamic target detection result; Calculate the real-time illuminance value of each grid according to the dual-channel image after spatio-temporal alignment and the affine transformation matrix after time compensation; the real-time illuminance value includes visible light illuminance value and infrared illuminance value; Adjust the light intensity of the lighting fixtures corresponding to each grid according to the real-time illuminance value and the average vehicle speed evaluation result.

[0017] Beneficial effects: The dynamic dimming method and system based on dual-channel camera illuminance sensing provided by this application, through tunnel pavement grid division and dual-channel image processing, combined with dynamic target detection and real-time illuminance calculation, realizes precise dimming control based on vehicle speed and ambient light, without deploying independent dedicated illuminance sensors for illuminance detection, has the advantages of improving the accuracy and energy-saving effect of tunnel lighting dimming control, realizing stable illuminance perception under all lighting conditions, and reducing the hardware deployment and maintenance costs. Description of the Drawings

[0018] Figure 1 It is a flowchart of the dynamic dimming method based on dual-channel camera illuminance sensing provided by an embodiment of this application.

[0019] Figure 2 It is a schematic structural diagram of the dynamic dimming system based on dual-channel camera illuminance sensing provided by an embodiment of the application.

[0020] Label description: 1. Dual-channel camera; 2. Lighting fixture; 3. Control terminal. Detailed Embodiment

[0021] Next, the technical solutions in the embodiments of this application will be clearly and completely described in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. Usually, the components of the embodiments of this application described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the drawings is not intended to limit the scope of the claimed application, but only represents the selected embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of this application.

[0022] It should be noted that: Similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of the present application, terms such as "first" and "second" are only used for distinguishing descriptions and cannot be construed as indicating or implying relative importance.

[0023] Please refer to Figure 1 , a dynamic dimming method based on dual-channel camera illuminance sensing in some embodiments of the present application, which is used to perform dimming control on the lighting fixtures in the tunnel based on a dual-channel camera with a visible light channel and an infrared channel (preferably a near-infrared channel), including the steps of: A1. Obtain the tunnel road surface grid division result determined in advance according to the coverage range of each lighting fixture in the tunnel; each grid of the grid division result corresponds to at least one lighting fixture; A2. Collect a dual-channel image of the tunnel road surface through the dual-channel camera, and perform preprocessing and cross-spectrum spatio-temporal alignment on the dual-channel image to obtain an affine transformation matrix for spatio-temporal alignment and the dual-channel image after spatio-temporal alignment; the dual-channel image includes a visible light channel image and an infrared channel image; A3. Based on the dual-channel image after spatio-temporal alignment, perform dynamic target detection on the vehicles in the tunnel and perform time compensation on the affine transformation matrix; A4. Evaluate the average vehicle speed of each grid according to the dynamic target detection result; A5. Calculate the real-time illuminance value of each grid according to the dual-channel image after spatio-temporal alignment and the affine transformation matrix after time compensation; the real-time illuminance value includes a visible light illuminance value and an infrared illuminance value; A6. Adjust the light intensity of the lighting fixtures corresponding to each grid according to the real-time illuminance value and the average vehicle speed evaluation result.

[0024] In some embodiments, the method can be based on Figure 2 the dynamic dimming system based on dual-channel camera illuminance sensing shown.

[0025] Among them, the tunnel road surface grid division refers to dividing the tunnel road surface into multiple grid units according to the coverage range of each lighting fixture in the tunnel. Each grid unit corresponds to the control area of at least one lighting fixture, and it can be specifically implemented by a spatial division method based on the installation position and coverage radius of the lighting fixture, such as a rectangular grid or a hexagonal grid. This division realizes the regional pertinence of dimming control and eliminates the single-point detection blind area.

[0026] Among them, cross-spectrum spatio-temporal alignment refers to performing geometric transformation and time synchronization processing on visible light and infrared dual-channel images. Specifically, it can be achieved by using the affine transformation matrix optimization method of SIFT feature point matching combined with the optical flow method. This operation solves the spatial misalignment problem caused by the spectral response difference and sampling time difference of the dual-channel camera, providing a spatio-temporal consistency basis for multi-spectral data fusion.

[0027] Among them, dynamic target detection refers to identifying the position and motion state of moving vehicles in the tunnel. Specifically, it can be achieved by using the improved YOLOv11 model combined with the Deepsort tracking algorithm. This detection provides target motion trajectory data for vehicle speed evaluation, supporting the calculation of dynamic dimming parameters.

[0028] Among them, average vehicle speed evaluation refers to statistically calculating the weighted average of the vehicle speeds related to each grid. Specifically, it can be achieved by using the speed calculation based on the tracking trajectory combined with the standard deviation weight adjustment method. This evaluation reflects the dynamic changes in the lighting requirements for vehicle passage and serves as the basis for adjusting the dimming intensity.

[0029] Among them, real-time illuminance value calculation refers to separately extracting the image data of the visible light and infrared channels and converting them into physical illuminance values. Specifically, it can be achieved by using the linear conversion of pixel values combined with the optimization method of dynamic infrared compensation weights. This calculation utilizes the spectral complementary characteristics of the dual channels to preferentially select effective illuminance data in overexposed or low-illuminance scenarios.

[0030] Among them, dimming intensity adjustment refers to dynamically controlling the output light intensity of the lamps according to the illuminance and vehicle speed parameters. Specifically, it can be achieved by using the formulaic calculation method of combining the reference intensity with the speed proportional coefficient and illuminance deviation compensation. This adjustment realizes the real-time matching of the lighting intensity with the vehicle passage requirements, replacing the function of an independent illuminance sensor.

[0031] The core innovation of this application lies in synchronously solving the problems of hardware redundancy, single-channel perception limitations, and local monitoring blind spots in traditional tunnel lighting systems through the collaborative perception of dual-channel cameras and the integration of dynamic dimming strategies. Specifically, it is manifested as: realizing all-weather illuminance perception by using the spectral complementarity of the visible light and infrared channels, eliminating the dual-channel data deviation through cross-spectrum spatio-temporal alignment and dynamic compensation, combining grid-based area control and vehicle speed dynamic parameter optimization of the dimming strategy to form an integrated intelligent lighting control scheme that does not rely on independent sensors.

[0032] The working process and principle of this application are as follows: First, obtain the tunnel pavement grid division result determined in advance according to the coverage range of each lighting fixture in the tunnel. Each grid corresponds to at least one lighting fixture. This grid division method ensures the regional pertinence of dimming control and avoids the blind area problem of single-point detection. Next, collect the dual-channel images of the tunnel pavement through a dual-channel camera, including visible light channel images and infrared channel images. Preprocess these images and perform cross-spectral spatio-temporal alignment to obtain the affine transformation matrix for spatio-temporal alignment and the dual-channel images after spatio-temporal alignment. This step eliminates the spatio-temporal deviation caused by spectral differences and provides a geometrically consistent dual-channel data basis for subsequent fusion analysis. Based on the dual-channel images after spatio-temporal alignment, perform dynamic target detection on the vehicles in the tunnel, and at the same time perform time compensation on the affine transformation matrix. This improves the spatio-temporal consistency of vehicle detection and avoids tracking deviation caused by dynamic errors of the camera. Then, evaluate the average vehicle speed of each grid according to the dynamic target detection results. This uses the vehicle speed as an important dynamic parameter of the dimming intensity, reflecting the real-time changes in the lighting requirements for vehicle passage in different regions. According to the dual-channel images after spatio-temporal alignment and the affine transformation matrix after time compensation, calculate the real-time illuminance values of each grid, including visible light illuminance values and infrared illuminance values. This dual-channel real-time illuminance calculation utilizes the spectral complementary characteristics of the visible light and infrared channels, and preferentially selects effective illuminance values respectively in strong light or low illuminance scenarios, overcoming the imaging defects of a single channel. Finally, adjust the light intensity of the lighting fixtures corresponding to each grid according to the real-time illuminance values and the average vehicle speed evaluation results. This realizes the refined matching of the lighting intensity in the tunnel with the vehicle passage requirements, improves the real-time performance and adaptability of lighting control while eliminating the dependence on independent sensors.

[0033] Through the above solution, this application realizes the illuminance perception of the entire tunnel area based on a dual-channel camera, and overcomes the problem of local illuminance monitoring blind areas caused by the single-point detection mode. By coupling grid-based illuminance perception with a dynamic dimming strategy, the accuracy and real-time performance of tunnel lighting control are improved. At the same time, the reuse of camera functions eliminates hardware redundancy and reduces the system deployment and maintenance costs. The cross-spectral fusion of dual-channel images improves the robustness of illuminance perception, and reliable illuminance data can be obtained in both strong light and low illuminance scenarios. The introduction of dynamic target detection and vehicle speed evaluation enables lighting control to better adapt to traffic flow changes and improves the intelligent level of the system. Overall, this solution significantly improves the tunnel lighting effect, enhances driving safety, and realizes the efficient utilization of energy.

[0034] Preferably, the preprocessing in step A2 includes data synchronization of the visible light channel image and the infrared channel image, median filtering and CLAHE enhancement processing of the visible light channel image, and Laplacian of Gaussian sharpening processing of the infrared channel image.

[0035] Among them, data synchronization is achieved by a hardware trigger signal combined with a timestamp calibration method (for example, calibrated by interpolation). The preprocessing sequence follows a logical process of denoising first and then enhancement. Data synchronization establishes a timing reference for subsequent processing, and the differential processing of the visible light and infrared channels forms a complementarity for their respective spectral characteristics.

[0036] Through the above technical solutions, the present application realizes effective preprocessing of dual-channel images. Data synchronization eliminates the time misalignment problem caused by hardware delay or transmission difference, and establishes a unified time reference for subsequent spatio-temporal alignment. Median filtering and CLAHE enhancement processing improve the noise resistance and local contrast of visible light images, adapting to the brightness differences in different regions of the tunnel. Gaussian-Laplacian sharpening processing enhances the edge details of infrared images while suppressing thermal noise. The combined application of these preprocessing steps enables the visible light and infrared channel images to achieve complementary optimization in three dimensions of noise reduction, contrast, and detail clarity, providing high-quality input data for subsequent cross-spectral spatio-temporal alignment, thereby improving the accuracy of spatio-temporal alignment and the accuracy of real-time illuminance value calculation.

[0037] In some embodiments, in step A2, performing cross-spectral spatio-temporal alignment on the dual-channel images includes: Taking the preprocessed visible light channel image and infrared channel image as the target visible light image and target infrared image, performing SIFT feature point matching on the target visible light image and target infrared image to obtain a preliminary affine transformation matrix; Estimating the global average motion vector of the target visible light image based on the optical flow method to correct the translation parameter of the preliminary affine transformation matrix to obtain a corrected affine transformation matrix; Taking maximizing the normalized mutual information and minimizing the SIFT feature point matching error as the goal, establishing an objective function related to the affine transformation matrix; the normalized mutual information is related to the entropy and mutual information of the target visible light image and target infrared image; Taking the corrected affine transformation matrix as the initial value of the affine transformation matrix, iteratively solving the objective function by the gradient ascent method to obtain the final affine transformation matrix; Performing an affine transformation on the target infrared image using the final affine transformation matrix, and splicing the affine result with the target visible light image to obtain the dual-channel image after spatio-temporal alignment.

[0038] Specifically, first, the SIFT algorithm is used to extract the feature points of visible light and infrared images, overcoming the matching difficulties caused by the texture differences of cross-spectrum images and generating a preliminary geometric transformation relationship. Subsequently, the optical flow method is introduced to analyze the global motion trend of the visible light image, dynamically correcting the translational deviation caused by camera jitter or vehicle movement. For example, in a tunnel scenario, the image smear error generated by the rapid passing of a vehicle can be corrected. When constructing the multi-objective optimization function, the normalized mutual information effectively quantifies the statistical correlation between spectra by calculating the ratio of the information entropy and joint entropy of two images. For example, in the strong light area at the tunnel entrance, the overexposed area of the visible light image can supplement the statistical features through the high dynamic range information of the infrared image. The gradient ascent method iteratively updates through the gradient direction in the parameter space. For example, based on the corrected affine transformation matrix, the rotation, scaling, and translation parameters of the affine transformation matrix are adjusted simultaneously in each iteration, enabling the objective function to achieve a balance between minimizing geometric error and maximizing statistical correlation. The finally generated affine transformation matrix accurately maps the infrared image to the visible light coordinate system, realizing pixel-level alignment of the two-channel images and providing a geometrically consistent two-channel data basis for subsequent illuminance calculation.

[0039] Among them, the SIFT feature point matching process is a prior art and will not be elaborated here. The obtained preliminary affine transformation matrix can be expressed as: ; where is the preliminary affine transformation matrix, , are the scaling factors of the preliminary affine transformation matrix, , are the shear factors of the preliminary affine transformation matrix, , are the translation parameters of the preliminary affine transformation matrix.

[0040] Among them, when estimating the global average motion vector of the target visible light image based on the optical flow method, the global average motion vector can be calculated according to the continuous M frames of the target visible light image (M can be set according to actual needs) using the following formula: ; ; where is the horizontal component of the global average motion vector, is the vertical component of the global average motion vector, is the inter-frame horizontal displacement of the (x, y) pixel point, is the inter-frame vertical displacement of the (x, y) pixel point.

[0041] where and It can be obtained in the following way: First, use the Sobel operator to calculate the spatial gradient and temporal gradient between adjacent-frame target visible light images, construct the spatial gradient matrix A and the temporal gradient matrix b, and then calculate the inter-frame horizontal displacement matrix and the inter-frame vertical displacement matrix according to the following formula, and extract and specifically, extract the (x, y)-numbered elements in the inter-frame horizontal displacement matrix and the inter-frame vertical displacement matrix to obtain and : ; where is the inter-frame horizontal displacement matrix, is the inter-frame vertical displacement matrix.

[0042] Among them, the corrected affine transformation matrix can be expressed as: .

[0043] Optionally, the objective function can be: ; is the objective function, represents the affine transformation matrix, is the target visible light image, is the target infrared image, represents the affine transformation result of performing an affine transformation on using , is and the global similarity between, is the SIFT feature point of the target visible light image, is 's SIFT feature point, is the weight coefficient, which is used to balance the statistical dependence and geometric consistency.

[0044] Among them, , is and the mutual information between (the calculation method of mutual information is the prior art and will not be elaborated here), is 's entropy (the calculation method of the entropy of an image is the prior art and will not be elaborated here), is 's entropy.

[0045] Among them, Adaptive values can be obtained based on the feature point density, for example, calculated by the following formula: , is a preset reference value (which can be adjusted according to actual needs, for example, 0.5), is and the number of successfully matched SIFT feature points between, is and the total number of SIFT feature points of.

[0046] When iteratively solving the objective function by the gradient ascent method, iterative updates are performed based on the corrected affine transformation matrix. The gradient of the objective function with respect to the parameters of the affine transformation matrix (scaling factor, shear factor, translation parameter) is calculated for each iteration. The product of the gradient and the preset learning efficiency is added to the parameters of the affine transformation matrix in the previous step to obtain the affine transformation matrix for this iteration update. Iterate in this way until the objective function converges to obtain the final affine transformation matrix.

[0047] Among them, the target infrared image is affine-transformed using the final affine transformation matrix, and the affine result is stitched with the target visible light image to obtain a spatio-temporally aligned dual-channel image, which can be expressed by the following formula: , where represents the spatio-temporally aligned dual-channel image obtained by stitching, is the stitching function, is the final affine transformation matrix, represents using the final affine transformation matrix for the affine transformation result of the affine transformation.

[0048] In some embodiments, step A3 includes: A301. Use the improved YOLOv11 model to perform object detection on the spatio-temporally aligned dual-channel image to obtain the object detection bounding box and ReID feature vector of the vehicle; A302. Estimate the average motion vector of the target area enclosed by the object detection bounding box based on the optical flow method to correct the translation parameter of the affine transformation matrix (referring to the final affine transformation matrix in the previous text) to obtain the time-compensated affine transformation matrix.

[0049] Through the above technical solutions, the present application realizes accurate vehicle detection and tracking under complex lighting conditions. The improved YOLOv11 model makes full use of dual-channel image information, improving the robustness of detection. Based on the optical flow method for motion estimation and affine transformation matrix compensation, it effectively eliminates the spatio-temporal deviation caused by vehicle motion and camera jitter, ensuring the accuracy of subsequent illuminance calculation. The introduction of ReID feature vectors provides a reliable data association basis for multi-object tracking, enhancing the overall performance of the system.

[0050] Among them, in the input end of the YOLOv11 backbone network of the improved YOLOv11 model, a dynamic spectral adaptive weight (DSAW) module is added to achieve: quantifying the global response intensity of each channel through global average pooling; dynamically allocating the weights of the visible light channel and the infrared channel through a fully connected layer combined with a non-linear activation function; ensuring weight normalization through Sigmoid output to avoid a single channel dominating or weakening; finally outputting a weighted fusion feature map; expressed by the formula as: ; where H is the image height, W is the image width, c is the channel index (c = 1 for the visible light channel, c = 2 for the infrared channel), is the pixel value of the (i,j) pixel point of the image of channel c in the dual-channel image after spatio-temporal alignment, represents the image of channel c in the dual-channel image after spatio-temporal alignment, is the Sigmoid function, and are the weight matrices of the first fully connected layer and the second fully connected layer respectively, is the non-linear activation function, is the global response intensity of channel c, is the fusion weight of channel c. Through the above operations, it can adaptively adjust the contributions of different spectra. In low illuminance scenarios, the infrared weight tends to 1, relying on infrared data to enhance details. In normal lighting scenarios, the visible light weight tends to 1, retaining color information; is the weighted fusion feature map. This feature map is input into the YOLOv11 backbone network for object detection to obtain the object detection bounding box and ReID feature vector of the vehicle.

[0051] Among them, in step A302, for the detected target region, the average motion vector of this target region can be calculated according to the following formula: ; ; where, represents the i-th target region, is the horizontal component of the average motion vector of the i-th target region, is the vertical component of the average motion vector of the i-th target region.

[0052] In step A302, the translation parameters of the affine transformation matrix are corrected to obtain the affine transformation matrix after time compensation. The means of for each target region and for each target region are respectively added to the two translation parameters in the final affine transformation matrix. It is expressed by the formula: ; where, is the affine transformation matrix after time compensation, , are the scaling factors of the final affine transformation matrix, , are the shear factors of the final affine transformation matrix, , are the translation parameters of the final affine transformation matrix, is the mean of for each target region, is the mean of for each target region.

[0053] In some embodiments, step A4 includes: A401. Using the Deepsort algorithm to perform target tracking on each vehicle according to the target detection box and the ReID feature vector to obtain the tracking trajectory; A402. Calculating the speed of each vehicle according to the tracking trajectory; A403. For each grid, performing a weighted average operation on the speeds of the vehicles on all lanes passing through the grid to obtain the average vehicle speed of the grid.

[0054] Specifically, in the target tracking stage, the target detection boxes between consecutive frames are input into the Deepsort algorithm, and the target position in the next frame is predicted through Kalman filtering. When calculating the speed, the time interval of the tracking trajectory is calibrated to the actual frame rate of the dual-channel camera, and the speed value can be calculated by the difference method. During the weighted average process, the weights of each lane can be dynamically configured according to the tunnel design specifications, and the abnormal speed data with a standard deviation exceeding the set threshold is automatically excluded. By fusing geometric tracking and appearance feature matching, the trajectory continuity of vehicles under complex lighting conditions is guaranteed, and at the same time, the lane-differentiated weighted strategy makes the grid average vehicle speed more conform to the actual traffic flow state, providing accurate input parameters for subsequent dimming control.

[0055] Among them, the application of the Deepsort algorithm is based on Kalman filtering for target position prediction, and the detection boxes are associated with the predicted trajectories through the Hungarian algorithm. The ReID feature vectors are generated by the improved YOLOv11 model and contain deep semantic information about the appearance of the vehicles. During the trajectory association process, the ReID feature vectors are used to calculate the similarity between targets to improve the matching accuracy. The un-matched target detection boxes are initialized as new trajectories, and a survival period (such as 30 frames) is set for the lost trajectories and they are deleted after exceeding the survival period.

[0056] Among them, the speed of the vehicle can be calculated based on the tracking trajectory using the difference method, which is a prior art and will not be elaborated here.

[0057] Through the above technical solutions, the present application realizes stable tracking of vehicles in complex scenarios, effectively solves the problem of tracking trajectory breakage caused by target occlusion or light change. At the same time, by fusing the ReID feature vectors, the ability to distinguish vehicles with similar appearances is improved, and the error of vehicle identity swapping is reduced. In addition, the speed calculation method based on the tracking trajectory provides more continuous and stable speed information compared with the speed estimation of single-frame detection. Finally, the weighted average strategy is used to calculate the average vehicle speed of the grid, taking into account the characteristics of different lanes, making the speed evaluation result more accurate and reliable, and providing accurate input parameters for subsequent dimming control.

[0058] In some preferred embodiments, in step A403, the average vehicle speed of the grid is calculated by the following formula: ; ; Among them, is the average vehicle speed of the grid, is the speed of the i-th vehicle on all lanes of this grid, is 's weight, N is the total number of vehicles on all lanes of this grid, is the average speed of the lanes on all lanes of this grid, is the standard deviation of the speeds of the vehicles on all lanes of this grid.

[0059] Through the above technical solution, the present application can effectively improve the calculation accuracy and robustness of the grid average vehicle speed. Specifically, by introducing a dynamic weight mechanism, vehicles with speeds close to the overall average level are given higher weights in the calculation, while the influence of those vehicles with speeds significantly deviating from the average level (such as vehicles with rapid acceleration or deceleration) is appropriately reduced. This processing method can effectively suppress the interference of abnormal speed values on the overall evaluation result, thereby more accurately reflecting the true movement state of the vehicle flow within the grid. Further, by introducing the standard deviation as a parameter for weight calculation, the method realizes the adaptive adjustment of the vehicle speed distribution characteristics. When the speed distribution within the lane is relatively concentrated, the weight is more sensitive to speed differences, enabling a finer distinction of the contributions of different vehicles; when the speed distribution is relatively dispersed, the sensitivity of the weight is correspondingly reduced, avoiding the problem of unbalanced weight allocation caused by accidental speed fluctuations. Thus, this technical solution not only improves the accuracy of grid average vehicle speed estimation but also enhances the adaptability and stability of the algorithm in complex traffic scenarios. This provides more reliable input parameters for subsequent tunnel lighting control based on vehicle speed, contributing to more precise and efficient intelligent lighting control.

[0060] In some embodiments, step A5 includes: A501. For each grid, calculate the visible light illumination value of each grid according to the average pixel value of the visible light channel image within the grid after spatio-temporal alignment, the linear conversion coefficient from visible light pixel value to illumination, and the preset ambient background illumination compensation value; A502. For each grid, perform an affine transformation on the infrared channel image after spatio-temporal alignment using the time-compensated affine transformation matrix, and calculate the infrared illumination value corresponding to the average pixel value of the affine transformation result within the grid to obtain the initial infrared illumination value; A503. For each grid, compare the average pixel value of the visible light channel image within the grid after spatio-temporal alignment with the preset pixel value threshold, determine the dynamic infrared compensation weight, and combine it with the initial infrared illumination value to calculate the final infrared illumination value of each grid.

[0061] Among them, in step A501, the visible light illumination value of each grid can be calculated according to the following formula: ; Among them, is the visible light illumination value of the j-th grid, is the linear conversion coefficient from visible light pixel value to illumination (usually 0.02 lx / pixel, but not limited to this), is the average pixel value of the visible light channel image within the j-th grid after spatio-temporal alignment, is a preset ambient background illuminance compensation value, which is used to eliminate the influence of dark current or ambient stray light (it can be set according to actual needs, for example, it is 5 lx).

[0062] Among them, in step A502, after the affine transformation matrix compensated by time is used to perform affine transformation on the infrared channel image after spatio-temporal alignment, the average pixel value of the affine transformation result in this grid can be calculated, and then the mapping function from the pre-calibrated infrared pixel value to illuminance is used to convert this average pixel value into the initial infrared illuminance value.

[0063] Among them, in step A503, the final infrared illuminance value of each grid can be calculated through the following formula: ; Among them, is the final infrared illuminance value of the j-th grid, is the initial infrared illuminance value of the j-th grid, is a preset infrared compensation weight coefficient, which is used to control the infrared contribution intensity in the overexposed area of visible light (it can be set according to actual needs, for example, it is 0.7), is a preset pixel value threshold (it can be set according to actual needs, for example, it is 200), is the standard deviation of the Gaussian distribution, which is used to control the transition smoothness of the overexposed area (usually 30, but not limited to this).

[0064] Through the above technical solutions, the present application realizes adaptive illuminance perception based on a dual-channel camera. By introducing the ambient background illuminance compensation value, the non-linear response problem of visible light pixel values in low illuminance or overexposed areas is overcome, and the accuracy of visible light illuminance values is improved. Using the affine transformation matrix compensated by time to perform affine transformation on the infrared image reduces the spatio-temporal matching error caused by vehicle movement and enhances the spatial consistency of the initial infrared illuminance value. Adopting a dynamic near-infrared compensation weight mechanism adaptively adjusts the credibility of the infrared illuminance value according to the real-time state of the visible light channel image, reduces the infrared compensation weight when the visible light image is normal to avoid spectral redundancy, and increases the infrared compensation weight when the visible light image is overexposed to utilize the all-weather perception advantage of the infrared channel. This dynamic weight allocation strategy based on the state feedback of the visible light channel realizes the adaptive optimization fusion of the dual-channel illuminance perception results, effectively overcomes the risk of perception failure of a single channel under extreme lighting conditions, and improves the reliability and stability of illuminance perception.

[0065] In some embodiments, step A6 includes: For each grid, calculate the dimming intensity of each grid according to the preset reference dimming intensity, the ratio of the average vehicle speed of this grid to the maximum designed vehicle speed of the tunnel, the preset speed influence coefficient, the preset tunnel lighting required illuminance, and the maximum value among the visible light illuminance value and the infrared illuminance value of this grid; A602. Adjust the light intensity of the lighting fixtures corresponding to each grid according to the dimming intensity of each grid.

[0066] Among them, in step A601, the dimming intensity of each grid can be calculated according to the following formula: ; Among them, is the dimming intensity of the jth grid, is the reference dimming intensity (determined by the rated power of the lighting fixture, for example, set as the light intensity corresponding to the rated power of the lighting fixture), is the average vehicle speed of the jth grid, is the maximum designed vehicle speed of the tunnel, is the speed influence coefficient (usually set according to the length range of the tunnel. For example, if the tunnel length is greater than 1 km, it is set to 0.9 to give priority to ensuring driving safety; otherwise, it is set to 0.6 to focus on energy conservation), is the tunnel lighting required illuminance (which can be set according to actual needs, for example, 100 lx), is the visible light illuminance value of the jth grid, is the infrared illuminance value of the jth grid. In the normal light area, is relatively high, and the dimming is dominated by the visible light illuminance. In the area where visible light fails (such as overexposure or low illuminance), the dimming is dominated by the infrared illuminance.

[0067] In step A602, according to the calculated dimming intensity, convert it to the power of the corresponding lighting fixture, and adjust the working power of the corresponding lighting fixture according to the calculation result. For example, adjust the power through a PWM control signal.

[0068] Through the above technical solution, the present application effectively solves the problem of insufficient adaptability of traditional dimming strategies to changes in traffic flow states. By introducing the ratio of the average vehicle speed to the maximum designed vehicle speed and combining it with the speed influence coefficient, a quantitative relationship between the dynamic change of vehicle speed and the dimming intensity is established, enabling the dimming intensity to adaptively increase as the vehicle speed increases and avoiding potential safety hazards caused by dimming delays in high-speed states. At the same time, by selecting the maximum value of visible light and infrared illuminance as input parameters and making use of the complementary advantages of different spectral data in strong light overexposure or low-light failure scenarios, the stability and reliability of illuminance perception are ensured, and the influence of local monitoring blind spots on dimming accuracy is eliminated. In addition, by calculating the dimming intensity based on grid-based zoning, fine-tuning of lamp control is achieved, reducing energy waste caused by differences in lighting requirements between regions.

[0069] Preferably, in step A501, the visible light illuminance value of the grid containing the vehicle with its headlights on is set to invalid. In step A601, if the visible light illuminance value of a grid is invalid, the maximum value between the visible light illuminance value and the infrared illuminance value of this grid is the infrared illuminance value.

[0070] This solution solves the problem of distorted visible light illuminance perception caused by strong light pollution from vehicle headlights through the visible light illuminance value invalidation mechanism and infrared illuminance value substitution strategy under specific conditions. Specifically, in step A501, the visible light illuminance value of the grid containing the vehicle with its headlights on is set to invalid, avoiding the abnormal increase in the pixel values of the visible light image caused by the active light source of the headlights and preventing the overestimation of illuminance evaluation due to data distortion in the visible light channel in such areas; in step A601, when it is detected that the visible light illuminance value is invalid, the infrared illuminance value is forced to be used as the maximum value to participate in the dimming calculation, taking advantage of the characteristic that the near-infrared channel is insensitive to light sources outside the visible spectrum to ensure that light intensity adjustment decisions can still be made based on stable infrared illuminance data in an environment with strong headlight interference. This dual-channel data dynamic selection mechanism not only retains the high-precision advantage of the visible light channel under natural lighting conditions but also compensates for the perception reliability in special scenarios through the infrared channel, achieving robust illuminance perception and dimming control under all-weather complex lighting conditions.

[0071] Reference Figure 2 , the present application provides a dynamic dimming system based on dual-channel camera illuminance sensing, including a dual-channel camera 1 with a visible light channel and an infrared channel, multiple lighting fixtures 2 arranged above the tunnel pavement, and a control terminal 3; The dual-channel camera 1 is used to collect dual-channel images of the tunnel pavement and upload them to the control terminal 3; the dual-channel images include visible light channel images and infrared channel images; The control terminal 3 is used to execute: Obtain the tunnel road surface grid division result determined in advance according to the coverage range of each lighting fixture 2 in the tunnel; each grid of the grid division result corresponds to at least one lighting fixture 2 (for the specific process, refer to step A1 in the previous text); Perform preprocessing and cross-spectral spatio-temporal alignment on the dual-channel image to obtain the affine transformation matrix for spatio-temporal alignment and the dual-channel image after spatio-temporal alignment; the dual-channel image includes a visible light channel image and an infrared channel image (for the specific process, refer to step A2 in the previous text); Based on the dual-channel image after spatio-temporal alignment, perform dynamic target detection on the vehicles in the tunnel and perform time compensation on the affine transformation matrix (for the specific process, refer to step A3 in the previous text); Evaluate the average vehicle speed of each grid according to the dynamic target detection result (for the specific process, refer to step A4 in the previous text); Calculate the real-time illuminance value of each grid according to the dual-channel image after spatio-temporal alignment and the affine transformation matrix after time compensation; the real-time illuminance value includes the visible light illuminance value and the infrared illuminance value (for the specific process, refer to step A5 in the previous text); Adjust the light intensity of the lighting fixture 2 corresponding to each grid according to the real-time illuminance value and the average vehicle speed evaluation result (for the specific process, refer to step A6 in the previous text).

[0072] In summary, the present application has at least the following advantages: 1. Solve the dual-channel data asynchrony problem through high-precision data synchronization to avoid "ghosting" of moving targets; 2. Combine global similarity and local geometric constraints to construct a spatially aligned objective function, optimize it by the gradient ascent method, and perform global and local two-time time dynamic compensation to achieve pixel-level spatio-temporal alignment and improve the subsequent detection and illuminance inversion accuracy; 3. Adaptively adjust the contributions of different spectra according to the lighting conditions, relying on near-infrared for low illuminance and visible light for normal lighting, improving the accuracy of target detection and anti-interference ability; 4. Overcome the failure problem of a single channel under extreme lighting by excluding vehicle interference and dynamically compensating overexposed areas, and improve the illuminance detection accuracy.

[0073] 5. The dynamic dimming strategy that fuses speed and dual-channel illuminance values reduces the dimming response delay, increases the lighting intensity when the vehicle speed is high, realizes a more intelligent and faster-response dimming control, and takes into account both safety and energy conservation; 6. Divide the tunnel road surface into grids for management, which can accurately locate local illuminance anomalies and realize local dimming repair.

[0074] The above are only the embodiments of the present application and are not intended to limit the protection scope of the present application. For those skilled in the art, the present application may have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A dynamic dimming method based on dual-channel camera illuminance sensing, which is used to perform dimming control on the lighting fixtures in a tunnel based on a dual-channel camera with a visible light channel and an infrared channel, and is characterized in that, Including the steps: A1. Obtain the tunnel road surface grid division result determined in advance according to the coverage range of each lighting fixture in the tunnel; each grid of the grid division result corresponds to at least one lighting fixture; A2. Collect a dual-channel image of the tunnel road surface through a dual-channel camera, and perform preprocessing and cross-spectrum spatio-temporal alignment on the dual-channel image to obtain an affine transformation matrix for spatio-temporal alignment and the dual-channel image after spatio-temporal alignment; the dual-channel image includes a visible light channel image and an infrared channel image; A3. Based on the dual-channel image after spatio-temporal alignment, perform dynamic target detection on the vehicles in the tunnel and perform time compensation on the affine transformation matrix; A4. Evaluate the average vehicle speed of each grid according to the dynamic target detection result; A5. Calculate the real-time illuminance value of each grid according to the dual-channel image after spatio-temporal alignment and the affine transformation matrix after time compensation; the real-time illuminance value includes visible light illuminance value and infrared illuminance value; A6. Adjust the light intensity of the lighting fixtures corresponding to each grid according to the real-time illuminance value and the average vehicle speed evaluation result.

2. The dynamic dimming method based on dual-channel camera illuminance sensing according to claim 1, wherein In step A2, performing cross-spectrum spatio-temporal alignment on the dual-channel image includes: Taking the preprocessed visible light channel image and infrared channel image as the target visible light image and target infrared image, performing SIFT feature point matching on the target visible light image and target infrared image to obtain a preliminary affine transformation matrix; Estimating the global average motion vector of the target visible light image based on the optical flow method to correct the translation parameter of the preliminary affine transformation matrix to obtain a corrected affine transformation matrix; Taking maximizing the normalized mutual information and minimizing the SIFT feature point matching error as the goal, establishing an objective function related to the affine transformation matrix; the normalized mutual information is related to the entropy and mutual information of the target visible light image and target infrared image; Taking the corrected affine transformation matrix as the initial value of the affine transformation matrix, and iteratively solving the objective function by the gradient ascent method to obtain the final affine transformation matrix; Performing an affine transformation on the target infrared image by using the final affine transformation matrix, and splicing the affine result with the target visible light image to obtain the dual-channel image after spatio-temporal alignment.

3. The dynamic dimming method based on dual-channel camera illuminance sensing according to claim 1, characterized in that Step A3 includes: A301. Use the improved YOLOv11 model to perform target detection on the dual-channel image after spatio-temporal alignment to obtain the target detection box of the vehicle and the ReID feature vector; A302. Estimate the average motion vector of the target area enclosed by the target detection box based on the optical flow method to correct the translation parameter of the affine transformation matrix to obtain the affine transformation matrix after time compensation.

4. The dynamic dimming method based on dual-channel camera illuminance sensing according to claim 3, wherein, Step A4 includes: A401. According to the target detection box and the ReID feature vector, use the Deepsort algorithm to perform target tracking on each vehicle to obtain the tracking trajectory; A402. Calculate the speed of each vehicle according to the tracking trajectory; A403. For each grid, perform a weighted average operation on the speeds of the vehicles on all lanes passing through this grid to obtain the average vehicle speed of this grid.

5. The dynamic dimming method based on dual-channel camera illuminance sensing according to claim 4, wherein In step A403, the average vehicle speed of the grid is calculated by the following formula: ； ； Among them, is the average vehicle speed of the grid, is the speed of the i-th vehicle on all lanes of the grid, is the weight of, N is the total number of vehicles on all lanes of the grid, is the average speed of the lanes on all lanes of the grid, is the standard deviation of the speeds of the vehicles on all lanes of the grid.

6. The dynamic dimming method based on dual-channel camera illuminance sensing according to claim 1, wherein, Step A5 includes: A501. For each grid, calculate the visible light illumination value of each grid according to the average pixel value of the visible light channel image within the grid after spatio-temporal alignment, the linear conversion coefficient from the visible light pixel value to the illuminance, and the preset environmental background illuminance compensation value; A502. For each grid, perform an affine transformation on the infrared channel image after spatio-temporal alignment using the affine transformation matrix after time compensation, and calculate the infrared illuminance value corresponding to the average pixel value of the affine transformation result within the grid to obtain the initial infrared illuminance value; A503. For each grid, compare the average pixel value of the visible light channel image within the grid after spatio-temporal alignment with the preset pixel value threshold, determine the dynamic infrared compensation weight, and combine it with the initial infrared illuminance value to calculate the final infrared illuminance value of each grid.

7. The dynamic dimming method based on dual-channel camera illuminance sensing according to claim 6, wherein Step A6 includes: A601. For each grid, calculate the dimming intensity of each grid according to the preset reference dimming intensity, the ratio of the average vehicle speed of the grid to the maximum designed vehicle speed of the tunnel, the preset speed influence coefficient, the preset tunnel lighting required illuminance, and the maximum value of the visible light illumination value and the infrared illumination value of the grid; A602. Adjust the light intensity of the lighting fixtures corresponding to each grid according to the dimming intensity of each grid.

8. The dynamic dimming method based on dual-channel camera illuminance sensing according to claim 7, wherein, In step A601, calculate the dimming intensity of each grid according to the following formula: ； Among them, is the dimming intensity of the j-th grid, is the reference dimming intensity, is the average vehicle speed of the j-th grid, is the maximum designed vehicle speed of the tunnel, is the speed influence coefficient, is the required illuminance of the tunnel lighting, is the visible light illuminance value of the j-th grid, is the infrared illuminance value of the j-th grid.

9. The dynamic dimming method based on dual-channel camera illuminance sensing according to claim 7, characterized in that, In step A501, set the visible light illumination value of the grid containing the vehicle with the headlights on to be invalid; In step A601, if the visible light illumination value of a grid is invalid, then the maximum value of the visible light illumination value and the infrared illumination value of the grid is the infrared illumination value.

10. A dynamic dimming system based on illuminance sensing of a dual-channel camera, characterized in that, It includes a dual-channel camera with a visible light channel and an infrared channel, a plurality of lighting fixtures arranged above the tunnel road surface, and a control terminal; The dual-channel camera is used to collect the dual-channel image of the tunnel road surface and upload it to the control terminal; the dual-channel image includes a visible light channel image and an infrared channel image; The control terminal is used to execute: Obtain the tunnel road surface grid division result determined in advance according to the coverage range of each lighting fixture in the tunnel; each grid of the grid division result corresponds to at least one lighting fixture; Perform preprocessing and cross-spectrum spatio-temporal alignment on the dual-channel image to obtain the affine transformation matrix for spatio-temporal alignment and the dual-channel image after spatio-temporal alignment; the dual-channel image includes a visible light channel image and an infrared channel image; Based on the dual-channel image after spatio-temporal alignment, perform dynamic target detection on the vehicles in the tunnel and perform time compensation on the affine transformation matrix; Evaluate the average vehicle speed of each grid according to the dynamic target detection result; Calculate the real-time illuminance value of each grid according to the dual-channel image after spatio-temporal alignment and the affine transformation matrix after time compensation; the real-time illuminance value includes the visible light illumination value and the infrared illumination value; Adjust the light intensity of the lighting fixtures corresponding to each grid according to the real-time illuminance value and the average vehicle speed evaluation result.

Images