An Adaptive Exposure Control Method and System for Unmanned Aerial Vehicle (UAV) Road Inspection

Abstract

This application discloses an adaptive exposure control method and system for UAV road inspection, relating to the field of image processing technology. It integrates ambient brightness and texture features to determine exposure parameters, avoiding overexposure caused by low grayscale on the road surface while preserving details in both highlights and shadows. Optimized filtering achieves smooth parameter switching, and closed-loop feedback improves brightness stability under sudden lighting changes, ensuring consistency across continuous images. It adapts to the core requirements of inspection tasks, improves road image quality, and provides reliable support for subsequent defect detection and 3D reconstruction, significantly improving inspection accuracy and efficiency. It effectively addresses the adaptability deficiencies of traditional automatic exposure algorithms in UAV road inspection by establishing an upper limit for exposure time through flight parameter constraints, avoiding image blurring caused by high-speed movement, and preserving key details such as road texture and cracks.

Description

Technical Field

[0001] This application relates to the field of image processing technology, specifically to an adaptive exposure control method and system for unmanned aerial vehicle (UAV) road inspection. Background Technology

[0002] With the rapid development of drone technology, its application in road inspection, infrastructure maintenance, and traffic safety monitoring is becoming increasingly widespread, making it one of the core technologies for road condition monitoring. In actual inspections, drones, equipped with cameras, collect road images, providing data support for subsequent visual tasks such as crack identification, defect detection, texture analysis, and 3D reconstruction. Image quality directly determines the accuracy and efficiency of these subsequent tasks. The AE (Auto Exposure) algorithm, as a key technology in the camera imaging process, has a decisive impact on the clarity and detail retention of road images. Currently, the automatic exposure algorithms used in drone cameras largely follow the design principles of consumer-grade photography, focusing on achieving image brightness that is visually comfortable for the human eye. While this meets basic imaging needs in conventional photography scenarios, it has gradually revealed numerous compatibility issues in the specific scenario of drone road inspection.

[0003] Drone road inspection scenarios are characterized by high-speed movement, unique target characteristics, complex lighting environments, and clear mission objectives. These characteristics differ fundamentally from consumer-grade photography scenarios, making traditional automatic exposure algorithms difficult to adapt. Firstly, drones typically maintain high-speed flight during road inspections. Traditional AE algorithms often compensate for image brightness by extending exposure time, which easily leads to motion blur, causing the loss of crucial details such as road surface texture and crack edges, affecting the accuracy of subsequent defect detection. Secondly, road surfaces are mostly dark asphalt materials with a generally low grayscale distribution. Traditional AE algorithms, using average luminance metering or center-weighted metering, are prone to misjudging road areas as underexposed, triggering overexposure adjustments and compromising the integrity of road markings, highlight areas, and crack details. In addition, the inspection route often faces complex and changing lighting environments, including tree shadows, dark areas under bridges, direct sunlight, road surface high-gloss reflections, and sudden changes in brightness at tunnel entrances. Traditional AE algorithms are slow to respond to sudden changes in lighting and are prone to image brightness jitter, resulting in inconsistent brightness of continuously acquired road surface images, which increases the difficulty of subsequent image stitching and feature extraction.

[0004] More importantly, the design goals of traditional automatic exposure algorithms are severely mismatched with the visual task requirements of UAV road inspection. The core purpose of road inspection images is to support technical processes such as crack detection, texture analysis, feature extraction, and 3D reconstruction. The core requirements are to preserve key road surface details and ensure image brightness balance, enabling subsequent algorithms to effectively extract target features. However, existing automatic exposure (AE) algorithms prioritize visual comfort and fail to consider the specific characteristics of the inspection task. In summary, influenced by factors such as high-speed motion blur, low road surface reflectivity, dynamic lighting changes, and task mismatch, existing automatic exposure methods cannot simultaneously achieve anti-blurring, brightness balance, and detail preservation in UAV road inspection scenarios. This makes it difficult to meet the high precision and high reliability requirements of road inspection. Therefore, an automatic exposure optimization technology adapted to UAV road inspection scenarios is urgently needed to address the shortcomings of existing technologies. Summary of the Invention

[0005] In order to solve the above-mentioned technical problems, this application proposes the following technical solution: In a first aspect, embodiments of this application provide an adaptive exposure control method for unmanned aerial vehicle (UAV) road inspection, comprising: Multidimensional perception and preprocessing are performed on drone inspection images to obtain the environmental brightness and texture features of the drone inspection scene. Dynamic motion constraint calculations are performed based on the flight parameters of the UAV to establish an upper limit for exposure time based on the flight state. The exposure parameter decision-making and planning are performed by integrating the ambient brightness, texture features, and upper limit of exposure time. The determined exposure parameters are optimized and filtered to obtain the optimal exposure parameters, so as to achieve smooth parameter switching and stable image effect; The exposure parameters are used to control the acquisition of data by the UAV camera and to achieve closed-loop feedback of exposure control.

[0006] In one possible implementation, the step of performing multi-dimensional perception and preprocessing on the UAV inspection image to obtain the environmental brightness and texture features of the UAV inspection scene includes: First, the original image frames captured by the camera are preprocessed, converted into grayscale images, and then noisy scenes and invalid pixels are removed using ROI masks; The grayscale histogram and cumulative distribution function are calculated for the preprocessed effective area to extract the quantile brightness to characterize the road surface brightness, thus avoiding the judgment bias caused by the low reflectivity of asphalt pavement. By using high-pass filtering or local gradient statistics to calculate the texture index of the image, a true texture measure reflecting the clarity of road surface details can be obtained. .

[0007] In one possible implementation, the preprocessing of the original image frames captured by the camera, converting them into grayscale images and removing clutter and invalid pixels using a ROI mask, includes: The raw data captured by the camera The formula for converting an image frame to a grayscale image Y is as follows:

[0008] A preset region of interest (ROI) mask is applied to the grayscale image Y to mask the sky and non-road background areas; By setting a saturation threshold and a noise threshold, and removing overexposed pixels with gray values ​​greater than the saturation threshold, as well as extremely dark noise pixels with gray values ​​less than or equal to the noise threshold, an effective road surface area is obtained.

[0009] In one possible implementation, the step of calculating a grayscale histogram and extracting a quantile brightness characterizing the pavement brightness from the preprocessed effective area using a cumulative distribution function, thereby avoiding judgment bias caused by the low reflectivity of asphalt pavement, includes: Assuming that after preprocessing, the effective road surface area contains a total of N pixels, the grayscale values ​​within the effective area are counted. The number of pixels, denoted as ,in ; Calculate the probability density function of image gray levels : ; Based on the cumulative distribution function, determine whether the gray value is less than or equal to... The cumulative proportion of pixels:

[0010] Extract representative quantile brightness and set a high quantile target value. ,exist Search from left to right in the middle to find the first one that satisfies the condition. grayscale value This is used as the representative quantile brightness of the current image. .

[0011] In one possible implementation, the texture index of the image is calculated using high-pass filtering or local gradient statistics to obtain a true texture measure reflecting the clarity of road surface details. ,include: Discrete differential operators are used to perform convolution operations on the effective road surface area to extract edge information in the horizontal and vertical directions, and the horizontal gradient is defined. and vertical gradient The convolution kernels are respectively and :

[0012] Calculate the Y coordinate of the grayscale image gradient magnitude at :

[0013] The set of valid road surface pixels filtered by ROI mask In this process, the average gradient of the global or block-level texture is calculated and used as the original texture metric. :

[0014] Camera sensor noise construction ISO By using an empirical model of gain variation or reading the factory calibration curve, the predicted noise baseline value for the current frame can be obtained. :

[0015] in: This is the noise gain coefficient. The dark current noise constant; Subtracting the pseudo-texture contribution from the noise baseline from the original texture measure yields the true texture measure that reflects the clarity of road surface details. :

[0016] in: To adjust the weighting coefficients, To ensure accurate texture measurement Non-negative.

[0017] In one possible implementation, the dynamic motion constraint calculation based on the flight parameters of the UAV to establish an upper limit for exposure time based on the flight state includes: Real-time reading of the drone's GPS or IMU data to obtain its current flight speed relative to the ground. and flight altitude ; Based on the focal length of the drone camera lens and sensor pixel size Calculate ground sampling distance :

[0018] Finally, according to the above and Determine the maximum permissible exposure time Adjust and define subsequent exposure parameters to ensure that pixel displacement does not exceed one pixel during exposure.

[0019] In one possible implementation, the step of fusing the ambient brightness, texture features, and exposure time upper limit for exposure parameter decision planning includes: The correction factor is calculated based on texture metrics to obtain the corrected target brightness. The target brightness is then dynamically fine-tuned to preserve highlight details, including:

[0020] in: To correct the target brightness, To correct the target brightness, For real texture measurement, This is the texture weight coefficient. The target texture baseline; Then calculate the representative quantile brightness of the current image. Based on the difference in magnification between the target brightness and the corrected brightness, prioritize adjusting the exposure time; if the brightness requirement is still not met, then make supplementary adjustments. ISO Gain-balanced brightness and noise control, including: Calculate the representative quantile brightness of the current image. and Difference ratio ; exist Within the range, the magnification M can be satisfied by increasing the exposure time; like If the brightness requirement is still not met, then the remaining magnification will be allocated to the camera's ISO. ISO And through the preset maximum sensitivity To control image noise.

[0021] In one possible implementation, the step of optimizing and filtering the determined exposure parameters to obtain the optimal exposure parameters, so as to achieve smooth parameter switching and stable image quality, includes: The hysteresis comparator is used to determine whether the parameter change exceeds the threshold, and time-domain smoothing filtering is performed in combination with historical parameters from the preceding frame. Simultaneously detect any sudden changes in ambient light; if a sudden change occurs, skip the smoothing process.

[0022] In one possible implementation, the step of determining whether the parameter change exceeds a threshold using a hysteresis comparator and performing time-domain smoothing filtering in conjunction with historical parameters from previous frames includes: Let the candidate exposure parameters for the current frame calculated by the exposure decision model be... The exposure parameters actually written to the camera in the previous frame were ; Calculate the relative rate of change between the two. ; Set a minimum change threshold ,when When the current disturbance is determined to be minor, the comparator does not flip and directly locks the final exposure parameters of the current frame to the specified value. This creates a control dead zone and suppresses parameter jitter; when At this point, it is determined that exposure adjustment is needed:

[0023] in: The preset smoothing filter coefficients and , This refers to the final exposure parameters after smoothing.

[0024] In one possible implementation, the step of using the exposure parameters to control the acquisition of the UAV camera and to achieve closed-loop feedback of exposure control includes: writing the smoothed final exposure parameters into the UAV camera register; after the camera acquires the next frame image, re-perceiving and pre-processing to form adaptive closed-loop control.

[0025] Secondly, embodiments of this application provide an adaptive exposure control system for unmanned aerial vehicle (UAV) road inspection, comprising: The multi-dimensional state perception and preprocessing module is used to perform multi-dimensional state perception and preprocessing on UAV inspection images to obtain the environmental brightness and texture features of the UAV inspection scene. The dynamic motion constraint calculation module is used to perform dynamic motion constraint calculations based on the flight parameters of the UAV and to establish an upper limit for exposure time based on the flight state. The exposure parameter decision planning module integrates the ambient brightness, texture features, and upper limit of exposure time to make exposure parameter decisions and plans. The stability smoothing module is used to optimize and filter the determined exposure parameters to obtain the optimal exposure parameters, so as to achieve smooth parameter switching and stable image effect. The closed-loop feedback module is used to control the acquisition of data by the UAV camera using the exposure parameters, and to realize closed-loop feedback of exposure control.

[0026] In this embodiment, exposure parameters are determined by integrating ambient brightness and texture features to avoid overexposure caused by low grayscale on the road surface, while balancing highlight and shadow details. Optimized filtering enables smooth parameter switching, and closed-loop feedback improves brightness stability under sudden lighting changes, ensuring consistent images. This aligns with the core requirements of inspection tasks, improving road image quality and providing reliable support for subsequent defect detection and 3D reconstruction, significantly enhancing inspection accuracy and efficiency. It effectively addresses the adaptability limitations of traditional automatic exposure algorithms in UAV road inspection by establishing an upper limit for exposure time through flight parameter constraints, avoiding image blurring caused by high-speed movement, and preserving key details such as road texture and cracks. Attached Figure Description

[0027] Figure 1 A flowchart illustrating an adaptive exposure control method for UAV road inspection provided in this application embodiment; Figure 2 A logic diagram of an adaptive exposure control method for UAV road inspection provided in an embodiment of this application; Figure 3 A schematic diagram of a typical automatic exposure road surface provided in the embodiments of this application; Figure 4 A schematic diagram of adaptive exposure control for road surface blurring provided for an embodiment of this application; Figure 5 This is a schematic diagram of standard exposed road surface crack texture provided in an embodiment of this application; Figure 6 A schematic diagram of the adaptive exposure control road surface crack texture provided for an embodiment of this application; Figure 7 This is a schematic diagram of an adaptive exposure control system for UAV road inspection provided in an embodiment of this application. Detailed Implementation

[0028] The present solution will now be described in conjunction with the accompanying drawings and specific embodiments.

[0029] See Figure 1 and Figure 2 The adaptive exposure control method for UAV road inspection provided in this embodiment includes: S101 performs multi-dimensional perception and preprocessing on drone inspection images to obtain the environmental brightness and texture features of the drone inspection scene.

[0030] In this embodiment, the original image frame I captured by the drone camera is obtained and converted into a grayscale image Y. The ROI mask is used to remove clutter from the sky and road edges, and invalid pixels with excessively high (saturation) or excessively low (noise) grayscale values ​​are removed.

[0031] Specifically, the raw image frames captured by the camera are first preprocessed, converted to grayscale, and then clutter and invalid pixels are removed using ROI masks. The formula for converting an image frame to a grayscale image Y is as follows:

[0032] A preset region of interest (ROI) mask is applied to the grayscale image Y to mask the sky and non-road background areas; and a saturation threshold and a noise threshold are set to remove overexposed highlight pixels with grayscale values ​​greater than the saturation threshold and extremely dark noise pixels with grayscale values ​​less than or equal to the noise threshold, thereby obtaining the effective road surface area. Preset Region of Interest (ROI) Mask: Based on the fixed gimbal pitch angle during UAV inspection, construct a binary mask matrix M with the same resolution as the grayscale image Y. ROI Set the mask value for pixels in areas that may contain sky, distant scenery, and the edge of the drone to 0, and set the mask value for the target road surface area to 1.

[0033] The grayscale histogram and cumulative distribution function are calculated for the preprocessed effective area, and the quantile brightness between P85 and P95 (preferably P90) is extracted. The pavement characteristics are characterized by the quantile brightness rather than the average brightness to eliminate the exposure judgment bias caused by the low reflectivity of asphalt pavement.

[0034] In this embodiment, it is assumed that after preprocessing, the total number of pixels contained in the effective road surface area is N, and the grayscale values ​​within the effective area are counted. The number of pixels, denoted as ,in Calculate the probability density function of image gray levels. : The grayscale value is determined to be less than or equal to the cumulative distribution function. The cumulative proportion of pixels:

[0035] Extract representative quantile brightness and set a high quantile target value. ,exist Search from left to right in the middle to find the first one that satisfies the condition. grayscale value This is used as the representative quantile brightness of the current image. .

[0036] Finally, high-pass filtering or local gradient statistics are used to calculate the texture index of the image, and combined with current... ISO Gain is used for noise correction to obtain a true texture measure that reflects the clarity of road surface details (such as cracks and potholes). T .

[0037] Specifically, the system first uses local gradient operators such as Sobel to calculate the gradient magnitude matrix of the effective road surface region, and then obtains the average gradient as the original texture measure. However, when using drones for inspections in low-light conditions, cameras typically increase brightness to ensure adequate illumination. ISO Gain inevitably introduces significant sensor thermal noise and readout noise. These high-frequency noises are mathematically very similar to the characteristics of tiny cracks in the road surface, causing traditional gradient algorithms to output artificially high texture scores. If not removed, the exposure decision model will be fooled by this pseudo-texture, thus suppressing the increase in exposure compensation, resulting in a dull final image with an extremely low signal-to-noise ratio.

[0038] To address this issue, the present invention introduces the current camera state. ISO The parameters estimate the noise baseline of the current image using a preset noise model, and then forcibly remove the noise contribution from the original gradient statistics.

[0039] The output true texture measure after this correction T It can accurately reflect the clarity of real physical details in the image (such as the asphalt skeleton, cracks, and pothole edges). When T When the value is low, the system can be certain that the image lacks true detail, thus safely instructing the camera to further increase or decrease the exposure. Limitations to improve signal-to-noise ratio; when T When the value is high, the system determines that the current exposure has sufficiently captured the road surface details and will switch to a highlight-preserving strategy to prevent further exposure from causing the texture to be saturated and smoothed out. Specifically, discrete differential operators (such as the Sobel operator) are used to perform convolution operations on the effective road surface area to extract edge information in the horizontal and vertical directions.

[0040] Define horizontal gradient and vertical gradient The convolution kernel is:

[0041] Calculate the Y coordinate of the grayscale image gradient magnitude at To reduce the computing power consumption of the UAV's onboard computing platform, the absolute value summation method is preferred as an approximation for square root calculation.

[0042] Then, the set of valid road surface pixels is filtered by the ROI mask. In the mean (assuming the total number of effective pixels is N), calculate the average gradient of the global or block-level data as the original texture metric. :

[0043] Camera sensor noise construction ISO By using an empirical model of gain variation or reading the factory calibration curve, the predicted noise baseline value for the current frame can be obtained. Typically, noise variance has an approximately linear relationship with analog gain:

[0044] in: This is the noise gain coefficient. This is the dark current noise constant.

[0045] Subtracting the pseudo-texture contribution from the noise baseline from the original texture measure yields the true texture measure that reflects the clarity of road surface details. :

[0046] in: To adjust the weighting coefficients, To ensure accurate texture measurement Non-negative.

[0047] S102, Perform dynamic motion constraint calculations based on the flight parameters of the UAV, and establish an upper limit for exposure time based on the flight state.

[0048] Real-time reading of the drone's GPS or IMU data to obtain its current flight speed relative to the ground. and flight altitude .

[0049] Based on the focal length of the drone camera lens and sensor pixel size Calculate ground sampling distance :

[0050] Finally, according to the above and Determine the maximum permissible exposure time The subsequent exposure parameters are adjusted and defined to ensure that the pixel displacement during the exposure does not exceed one pixel. This step provides a physical constraint boundary for subsequent decisions.

[0051] S103, integrate the ambient brightness, texture features and exposure time limit to make exposure parameter decision planning; In this embodiment, a correction factor is calculated based on texture indices to obtain the corrected target brightness, and the target brightness is dynamically fine-tuned to preserve highlight details. Then, the representative quantile brightness of the current image is calculated. Based on the difference in magnification between the target brightness and the corrected brightness, prioritize adjusting the exposure time; if the brightness requirement is still not met, then make supplementary adjustments. ISO Gain balance, brightness, and noise control.

[0052] Specifically, the corrected target brightness is obtained by calculating the correction factor based on the texture index, including:

[0053] in: To correct the target brightness, To correct the target brightness, For real texture measurement, This is the texture weight coefficient. The target texture baseline.

[0054] Calculate the representative quantile brightness of the current image. and Difference ratio ,exist Within the range, the magnification M can be satisfied by increasing the exposure time. If If the brightness requirement is still not met, then the remaining magnification will be allocated to the camera's ISO. ISO And through the preset maximum sensitivity To control image noise.

[0055] S104 optimizes and filters the determined exposure parameters to obtain the optimal exposure parameters, so as to achieve smooth parameter switching and stable image effect.

[0056] Recommended exposure combination calculated using S103 The data is fed into the stability control module, which uses a hysteresis comparator to determine whether the parameter change exceeds a preset threshold. The recommended exposure time is calculated by the model based on the difference between the current ambient brightness and the target brightness, and is limited by the theoretical shutter speed of the maximum anti-blur exposure time. The recommended ISO sensitivity is the theoretical sensor gain value calculated and allocated by the system when the recommended exposure time, even at its maximum, still cannot meet the brightness requirements. Together, these two values ​​constitute the target exposure requirement for the current frame under ideal conditions. A temporal smoothing filter is applied using historical parameters from previous frames, and the system determines whether a sudden change in lighting has occurred in the current environment (such as entering a bridge shadow). If a sudden change occurs, the smoothing process is skipped, and a fast response strategy is adopted.

[0057] In this embodiment, the candidate exposure parameters (exposure time or gain) of the current frame calculated by the exposure decision model are assumed to be: The exposure parameters actually written to the camera in the previous frame were The system first calculates the relative rate of change between the two. Set a minimum change threshold. (In this embodiment, 5% is preferred), when When the system determines that the current disturbance is minor, the comparator does not flip and directly locks the final exposure parameters of the current frame to [value missing]. This creates a control dead zone and suppresses parameter jitter.

[0058] when At this point, the system determines that exposure adjustment is needed. To achieve a smooth brightness transition, the system does not directly adjust the exposure. Instead of writing to the camera, a first-order low-pass filter formula is used for historical weighting:

[0059] in: The preset smoothing filter coefficients and Calculated This is the final exposure parameter after smoothing, which the system then writes into the camera register to complete the adaptive exposure control for this frame.

[0060] S105, the exposure parameters are used to control the acquisition of data by the UAV camera and to achieve closed-loop feedback of exposure control.

[0061] The optimal exposure parameters, filtered by S104, are written to the camera register. The camera acquires the next frame and returns it to S101, forming an adaptive closed-loop control that ensures the inspected image maintains clear texture and stable brightness even in motion.

[0062] This embodiment utilizes flight speed to constrain the maximum exposure time. When photographing a road surface using a drone at a speed of 6 m / s, see [reference needed]. Figure 3 and Figure 4 The image edge blur rate (blur ratio of more than 2 pixels) is significantly reduced compared to traditional automatic exposure, improving image sharpness and usability.

[0063] Texture gradient, as one of the criteria for exposure adjustment, improves detail clarity while maintaining overall brightness. (See also...) Figure 5 and Figure 6 In typical asphalt pavement scenarios, the average gradient value of the output image of the method in this embodiment is improved compared with the standard exposure, effectively preserving fine cracks, joint lines and surface roughness features.

[0064] Corresponding to the adaptive exposure control method for UAV road inspection provided in the above embodiments, this application also provides an embodiment of an adaptive exposure control system for UAV road inspection.

[0065] See Figure 7 The adaptive exposure control system 20 for UAV road inspection in this embodiment includes: The multi-dimensional state perception and preprocessing module 201 is used to perform multi-dimensional state perception and preprocessing on UAV inspection images to obtain the environmental brightness and texture features of the UAV inspection scene. The dynamic motion constraint calculation module 202 is used to perform dynamic motion constraint calculation based on the flight parameters of the UAV and establish an upper limit of exposure time based on the flight state. The exposure parameter decision planning module 203 integrates the ambient brightness, texture features and the upper limit of exposure time to make exposure parameter decision planning. The stability smoothing processing module 204 is used to optimize and filter the determined exposure parameters to obtain the optimal exposure parameters, so as to achieve smooth parameter switching and stable image effect. The closed-loop feedback module 205 is used to control the acquisition of data by the UAV camera using the exposure parameters and to realize closed-loop feedback of exposure control.

[0066] The above description is merely a specific embodiment of this application. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application. The protection scope of this application should be determined by the protection scope of the claims.

Claims

1. An adaptive exposure control method for UAV road inspection, characterized in that, include: Multidimensional perception and preprocessing are performed on drone inspection images to obtain the environmental brightness and texture features of the drone inspection scene. Dynamic motion constraint calculations are performed based on the flight parameters of the UAV to establish an upper limit for exposure time based on the flight state. The exposure parameter decision-making and planning are performed by integrating the ambient brightness, texture features, and upper limit of exposure time. The determined exposure parameters are optimized and filtered to obtain the optimal exposure parameters, so as to achieve smooth parameter switching and stable image effect; The exposure parameters are used to control the acquisition of data by the UAV camera and to achieve closed-loop feedback of exposure control. The step of performing dynamic motion constraint calculations based on the flight parameters of the UAV to establish an upper limit for exposure time based on the flight state includes: Real-time reading of the drone's GPS or IMU data to obtain its current flight speed relative to the ground. and flight altitude ; According to the focal length of the drone camera lens and sensor pixel size Calculate ground sampling distance : ； Finally, according to the above and Determine the maximum permissible exposure time Adjust and define subsequent exposure parameters to ensure that pixel displacement does not exceed one pixel during exposure; The process of integrating ambient brightness, texture features, and upper limit of exposure time to make exposure parameter decisions and plans includes: The correction factor is calculated based on texture metrics to obtain the corrected target brightness. The target brightness is then dynamically fine-tuned to preserve highlight details, including: ； in: To correct the target brightness, To correct the target brightness, For real texture measurement, This refers to the texture weight coefficient. The target texture baseline; Then calculate the representative quantile brightness of the current image. Based on the difference in magnification between the target brightness and the corrected brightness, prioritize adjusting the exposure time; if the brightness requirement is still not met, then make supplementary adjustments. ISO Gain-balanced brightness and noise control, including: Calculate the representative quantile brightness of the current image. and Difference ratio ; exist Within the range, the magnification M can be satisfied by increasing the exposure time; like If the brightness requirement is still not met, then the remaining magnification will be allocated to the camera's ISO. ISO And through the preset maximum sensitivity To control image noise.

2. The adaptive exposure control method for UAV road surface inspection according to claim 1, characterized in that, The process of performing multi-dimensional perception and preprocessing on UAV inspection images to obtain environmental brightness and texture features of the UAV inspection scene includes: First, the original image frames captured by the camera are preprocessed, converted into grayscale images, and then noisy scenes and invalid pixels are removed using ROI masks; The grayscale histogram and cumulative distribution function are calculated for the preprocessed effective area to extract the quantile brightness to characterize the road surface brightness, thus avoiding the judgment bias caused by the low reflectivity of asphalt pavement. By using high-pass filtering or local gradient statistics to calculate the texture index of the image, a true texture measure reflecting the clarity of road surface details can be obtained. .

3. The adaptive exposure control method for UAV road surface inspection according to claim 2, characterized in that, The first step involves preprocessing the raw image frames captured by the camera, converting them into grayscale images, and removing clutter and invalid pixels using a ROI mask. This includes: The raw data captured by the camera The formula for converting an image frame to a grayscale image Y is as follows: ； A preset region of interest (ROI) mask is applied to the grayscale image Y to mask the sky and non-road background areas; By setting a saturation threshold and a noise threshold, and removing overexposed pixels with gray values ​​greater than the saturation threshold, as well as extremely dark noise pixels with gray values ​​less than or equal to the noise threshold, an effective road surface area is obtained.

4. The adaptive exposure control method for UAV road surface inspection according to claim 2 or 3, characterized in that, The calculation of grayscale histograms and cumulative distribution functions of the preprocessed effective areas to extract quantile brightness characterizes the road surface brightness, avoiding judgment bias caused by the low reflectivity of asphalt pavements, includes: Assuming that after preprocessing, the effective road surface area contains a total of N pixels, the grayscale values ​​within the effective area are counted. The number of pixels, denoted as ,in ; Calculate the probability density function of image gray levels : ; Based on the cumulative distribution function, determine whether the gray value is less than or equal to... The cumulative proportion of pixels: ； Extract representative quantile brightness and set a high quantile target value. ,exist Search from left to right in the middle to find the first one that satisfies the condition. grayscale value This is used as the representative quantile brightness of the current image. .

5. The adaptive exposure control method for UAV road inspection according to claim 4, characterized in that, The textural indices of the image are calculated using high-pass filtering or local gradient statistics to obtain a true texture measure that reflects the clarity of road surface details. ,include: Discrete differential operators are used to perform convolution operations on the effective road surface area to extract edge information in the horizontal and vertical directions, and the horizontal gradient is defined. and vertical gradient The convolution kernels are respectively and : ； Calculate the Y coordinate of the grayscale image gradient magnitude at : ； The set of valid road surface pixels filtered by ROI mask In this process, the average gradient of the global or block-level texture is calculated and used as the original texture metric. : ； Camera sensor noise construction ISO By using an empirical model of gain variation or reading the factory calibration curve, the predicted noise baseline value for the current frame can be obtained. : ； in: This is the noise gain coefficient. The dark current noise constant; Subtracting the pseudo-texture contribution from the noise baseline from the original texture measure yields the true texture measure that reflects the clarity of road surface details. : ； in: To adjust the weighting coefficients, To ensure accurate texture measurement Non-negative.

6. The adaptive exposure control method for UAV road inspection according to claim 1, characterized in that, The process of optimizing and filtering the determined exposure parameters to obtain the optimal exposure parameters, in order to achieve smooth parameter switching and stable image quality, includes: The hysteresis comparator is used to determine whether the parameter change exceeds the threshold, and time-domain smoothing filtering is performed in combination with historical parameters from the preceding frame. Simultaneously detect any sudden changes in ambient light; if a sudden change occurs, skip the smoothing process.

7. The adaptive exposure control method for UAV road inspection according to claim 6, characterized in that, The step of determining whether the parameter change exceeds the threshold using a hysteresis comparator and performing time-domain smoothing filtering in conjunction with historical parameters from previous frames includes: Let the candidate exposure parameters for the current frame calculated by the exposure decision model be... The exposure parameters actually written to the camera in the previous frame were ; Calculate the relative rate of change between the two. ; Set a minimum change threshold ,when When the current disturbance is determined to be minor, the comparator does not flip and directly locks the final exposure parameters of the current frame to the specified value. This creates a control dead zone and suppresses parameter jitter; when At this point, it is determined that exposure adjustment is needed: ; in: The preset smoothing filter coefficients and , This refers to the final exposure parameters after smoothing.

8. The adaptive exposure control method for UAV road surface inspection according to claim 6 or 7, characterized in that, The method of using the exposure parameters to control the acquisition of the UAV camera and realize closed-loop feedback of exposure control includes: writing the smoothed final exposure parameters into the UAV camera register; after the camera acquires the next frame image, it re-perceives and preprocesses the image to form adaptive closed-loop control.

Images